v3.19.1

INSTALL_LINUX

@@ -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

INSTALL_WINDOWS

@@ -40,7 +40,7 @@ $ mkdir $HOME/usr/lib
    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. 
+respected download locations, copy them to $HOME/usr/lib/ and uncompress them. 
 
 openssl: https://github.com/openssl/openssl/releases
 
@@ -149,85 +149,10 @@ Copy cpuminer.exe to the release directory, compress and copy the release direct
 
 Run cpuminer
 
-In a command windows change directories to the unzipped release folder. to get a list of all options:
+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.
-
-sed -i 's/"-lpthread"/"-lpthreadGC2"/g' configure.ac
-
-
-7. Compile
-
-you can use the default compile if you intend to use cpuminer-opt on the
-same CPU and the virtual machine supports that architecture.
-
-./build.sh
-
-Otherwise you can compile manually while setting options in CFLAGS.
-
-Some common options:
-
-To compile for a specific CPU architecture:
-
-CFLAGS="-O3 -march=znver1 -Wall" ./configure --with-curl
-
-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 --with-curl
-
-This will compile for an older CPU that does not have AVX.
-
-You can find several examples in build-allarch.sh
-
-If you have a CPU with more than 64 threads and Windows 7 or higher you
-can enable the CPU Groups feature:
-
--D_WIN32_WINNT==0x0601
-
-Once you have run configure successfully run make 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.

Makefile.am

@@ -158,7 +158,9 @@ 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 \
@@ -167,7 +169,9 @@ cpuminer_SOURCES = \
   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/sha256d-4way.c \
   algo/sha/sha256t-gate.c \
   algo/sha/sha256t-4way.c \
   algo/sha/sha256t.c \

README.txt

@@ -18,14 +18,14 @@ 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.
 
-Architecture names and compile options used are only provided for Intel
-Core series. Budget CPUs like Pentium and Celeron are often missing some
-features.
+Architecture names and compile options used are only provided for 
+mainstream desktop CPUs. Budget CPUs like Pentium and Celeron are often
+missing some features. Check your CPU.
 
-AMD CPUs older than Piledriver, including Athlon x2 and Phenom II x4, are not
-supported by cpuminer-opt due to an incompatible implementation of SSE2 on
-these CPUs. Some algos may crash the miner with an invalid instruction.
-Users are recommended to use an unoptimized miner such as cpuminer-multi.
+Support for AMD CPUs older than Ryzen is incomplete and without specific 
+recommendations. Find the best fit. CPUs older than Piledriver, including
+Athlon x2 and Phenom II x4, are not supported by cpuminer-opt due to an
+incompatible implementation of SSE2 on these CPUs. 
 
 More information for Intel and AMD CPU architectures and their features
 can be found on Wikipedia.
@@ -34,26 +34,21 @@ https://en.wikipedia.org/wiki/List_of_Intel_CPU_microarchitectures
 
 https://en.wikipedia.org/wiki/List_of_AMD_CPU_microarchitectures
 
+File name                      Architecture name
 
-Exe file name                Compile flags              Arch name
+cpuminer-sse2.exe              Core2, Nehalem, generic x86_64 with SSE2   
+cpuminer-aes-sse42.exe         Westmere
+cpuminer-avx.exe               Sandybridge, Ivybridge
+cpuminer-avx2.exe              Haswell, Skylake, Kabylake, Coffeelake, Cometlake
+cpuminer-avx2-sha.exe          AMD Zen1, Zen2
+cpuminer-avx2-sha-vaes.exe     Intel Alderlake*, AMD Zen3
+cpuminer-avx512.exe            Intel HEDT Skylake-X, Cascadelake
+cpuminer-avx512-sha-vaes.exe   Icelake, Tigerlake, Rocketlake
 
-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)
-
-(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.
+* Alderlake is a hybrid architecture. With the E-cores disabled it may be
+  possible to enable AVX512 on the the P-cores and use the avx512-sha-vaes
+  build. This is not officially supported by Intel at time of writing.
+  Check for current information.
 
 Notes about included DLL files:
 
@@ -66,8 +61,7 @@ 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.
+file. 
 
 If you like this software feel free to donate:
 

RELEASE_NOTES

@@ -65,6 +65,92 @@ If not what makes it happen or not happen?
 Change Log
 ----------
 
+v3.19.1
+
+Changes to Windows binaries package:
+ - builds for CPUs with AVX or lower have CPU groups disabled,
+ - zen3 build renamed to avx2-sha-vaes to support Alderlake as well as Zen3,
+ - zen build renamed to avx2-sha, supports Zen1 & Zen2,
+ - avx512-sha build removed, Rocketlake CPUs can use avx512-sha-vaes,
+ - see README.txt for compatibility details.
+
+Fixed a few compiler warnings that are new in GCC 11.
+Other minor fixes.
+
+v3.19.0
+
+Windows binaries now built with support for CPU groups, requires Windows 7.
+
+Changes to cpu-affinity:
+  - PR#346: Fixed incorrect CPU affinity on Windows built for CPU groups,
+  - added support for CPU affinity for up to 256 threads or CPUs,
+  - streamlined code for more efficient initialization of miner threads,
+  - precise affining of each miner thread to a specific CPU,
+  - added an option to disable CPU affinity with "--cpu-affinity 0"
+
+Faster sha256t with AVX512 & AVX2.
+
+Added stratum error count to stats log, reported only when non-zero.
+
+v3.18.2
+
+Issue #342, fixed Groestl AES on Windows, broken in v3.18.0.
+
+AVX512 for sha256d.
+
+SSE42 and AVX may now be displayed as mining features at startup.
+This is hard coded for each algo, and is only implemented for scrypt
+at this time as it is the only algo with significant performance differences
+with those features.
+
+Fixed an issue where a high hashrate algo could cause excessive invalid hash
+rate log reports when starting up in benchmark mode.
+
+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.

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>
@@ -94,7 +97,6 @@ typedef  uint32_t set_t;
 #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
@@ -319,3 +321,4 @@ 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

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) \

algo/blake/decred-gate.c

@@ -8,7 +8,7 @@ uint32_t *decred_get_nonceptr( uint32_t *work_data )
    return &work_data[ DECRED_NONCE_INDEX ];
 }
 
-double decred_calc_network_diff( struct work* work )
+long double decred_calc_network_diff( struct work* work )
 {
    // sample for diff 43.281 : 1c05ea29
    // todo: endian reversed on longpoll could be zr5 specific...
@@ -16,7 +16,7 @@ double decred_calc_network_diff( struct work* work )
    uint32_t bits = ( nbits & 0xffffff );
    int16_t shift = ( swab32(nbits) & 0xff ); // 0x1c = 28
    int m;
-   double d = (double)0x0000ffff / (double)bits;
+   long double d = (long double)0x0000ffff / (long double)bits;
 
    for ( m = shift; m < 29; m++ )
        d *= 256.0;
@@ -25,7 +25,7 @@ double decred_calc_network_diff( struct work* work )
    if ( shift == 28 )
        d *= 256.0; // testnet
    if ( opt_debug_diff )
-       applog( LOG_DEBUG, "net diff: %f -> shift %u, bits %08x", d,
+       applog( LOG_DEBUG, "net diff: %f -> shift %u, bits %08x", (double)d,
                            shift, bits );
    return net_diff;
 }

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] );
@@ -1268,22 +1169,60 @@ 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 );
+
+   __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] ),

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 );
     }
 

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
 {

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 );
     }
 

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)))))

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)))))

algo/groestl/myr-groestl.c

@@ -11,7 +11,7 @@
 #else
   #include "sph_groestl.h"
 #endif
-#include "algo/sha/sph_sha2.h"
+#include "algo/sha/sha256-hash.h"
 
 typedef struct {
 #ifdef __AES__
@@ -19,7 +19,6 @@ typedef struct {
 #else
     sph_groestl512_context  groestl;
 #endif
-    sph_sha256_context      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
-     sph_sha256_init( &myrgr_ctx.sha );
 }
 
 void myriad_hash(void *output, const void *input)
@@ -49,8 +47,7 @@ void myriad_hash(void *output, const void *input)
    sph_groestl512_close(&ctx.groestl, hash);
 #endif
 
-   sph_sha256( &ctx.sha, hash, 64 );
-   sph_sha256_close( &ctx.sha, hash );
+   sha256_full( hash, hash, 64 );
 
    memcpy(output, hash, 32);
 }

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

@@ -632,26 +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 ); \
@@ -731,28 +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 \
@@ -965,26 +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 ); \
@@ -1064,28 +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 \

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__)               
 

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;
 

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

@@ -70,13 +70,13 @@ 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 XOROR(d, a, b, c) (d = mm512_xoror(a, b, c))
+#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))
 
 

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 );
@@ -255,7 +247,7 @@ 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;
@@ -269,13 +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 );
-#if defined(__VAES__)
-   groestl256_2way_init( &allium_8way_ctx.groestl, 32 );
-#else
-   init_groestl256( &allium_8way_ctx.groestl, 32 );
-#endif
    return true;
 }
 
@@ -320,21 +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 );

algo/lyra2/lyra2-gate.c

@@ -188,7 +188,7 @@ bool register_allium_algo( algo_gate_t* gate )
   gate->hash      = (void*)&allium_hash;
 #endif
   gate->optimizations = SSE2_OPT | AES_OPT | AVX2_OPT | AVX512_OPT
-	              | VAES_OPT | VAES256_OPT;
+	                   | VAES_OPT;
   opt_target_factor = 256.0;
   return true;
 };

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) \

algo/m7m/m7m.c

@@ -13,6 +13,7 @@
 #include "algo/whirlpool/sph_whirlpool.h"
 #include "algo/ripemd/sph_ripemd.h"
 #include "algo/sha/sph_sha2.h"
+#include "algo/sha/sha256-hash.h"
 
 #define EPSa DBL_EPSILON
 #define EPS1 DBL_EPSILON
@@ -104,8 +105,8 @@ uint32_t sw2_( int nnounce )
 }
 
 typedef struct {
-    sph_sha256_context      sha256;
-    sph_sha512_context      sha512;
+    sha256_context           sha256;
+    sph_sha512_context       sha512;
     sph_keccak512_context    keccak;
     sph_whirlpool_context    whirlpool;
     sph_haval256_5_context   haval;
@@ -117,7 +118,7 @@ m7m_ctx_holder m7m_ctx;
 
 void init_m7m_ctx()
 {
-    sph_sha256_init( &m7m_ctx );
+    sha256_ctx_init( &m7m_ctx.sha256 );
     sph_sha512_init( &m7m_ctx.sha512 );
     sph_keccak512_init( &m7m_ctx.keccak );
     sph_whirlpool_init( &m7m_ctx.whirlpool );
@@ -153,11 +154,10 @@ 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) );
-    sph_sha256_context ctxf_sha256;
 
     memcpy(data, pdata, 80);
 
-    sph_sha256(     &ctx1.sha256,    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 );
@@ -189,8 +189,8 @@ int scanhash_m7m_hash( struct work* work, uint64_t max_nonce,
 
         memcpy( &ctx2, &ctx1, sizeof(m7m_ctx) );
 
-        sph_sha256( &ctx2.sha256, data_p64, 80 - M7_MIDSTATE_LEN );
-        sph_sha256_close( &ctx2.sha256, bhash[0] );
+        sha256_update( &ctx2.sha256, data_p64, 80 - M7_MIDSTATE_LEN );
+        sha256_final( &ctx2.sha256, bhash[0] );
 
         sph_sha512(  &ctx2.sha512, data_p64, 80 - M7_MIDSTATE_LEN );
         sph_sha512_close( &ctx2.sha512, bhash[1] );
@@ -225,9 +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);
 
-        sph_sha256_init( &ctxf_sha256 );
-        sph_sha256( &ctxf_sha256, bdata, bytes );
-        sph_sha256_close( &ctxf_sha256, hash );
+        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);
@@ -260,10 +258,8 @@ int scanhash_m7m_hash( struct work* work, uint64_t max_nonce,
             mpzscale=bytes;
             mpz_export(bdata, NULL, -1, 1, 0, 0, product);
 
-            sph_sha256_init( &ctxf_sha256 );
-            sph_sha256( &ctxf_sha256, bdata, bytes );
-            sph_sha256_close( &ctxf_sha256, hash );
-	}
+            sha256_full( hash, bdata, bytes );
+	     }
 
         if ( unlikely( valid_hash( (uint64_t*)hash, (uint64_t*)ptarget ) 
              && !opt_benchmark ) )

algo/ripemd/lbry-gate.c

@@ -4,7 +4,7 @@
 #include <string.h>
 #include <stdio.h>
 
-double lbry_calc_network_diff( struct work *work )
+long double lbry_calc_network_diff( struct work *work )
 {
         // sample for diff 43.281 : 1c05ea29
         // todo: endian reversed on longpoll could be zr5 specific...
@@ -12,7 +12,7 @@ double lbry_calc_network_diff( struct work *work )
    uint32_t nbits = swab32( work->data[ LBRY_NBITS_INDEX ] );
    uint32_t bits = (nbits & 0xffffff);
    int16_t shift = (swab32(nbits) & 0xff); // 0x1c = 28
-   double d = (double)0x0000ffff / (double)bits;
+   long double d = (long double)0x0000ffff / (long double)bits;
 
    for (int m=shift; m < 29; m++) d *= 256.0;
    for (int m=29; m < shift; m++) d /= 256.0;

algo/ripemd/lbry.c

@@ -7,24 +7,19 @@
 #include <string.h>
 #include <stdio.h>
 #include "sph_ripemd.h"
-#include "algo/sha/sph_sha2.h"
+#include "algo/sha/sha256-hash.h"
 
 void lbry_hash(void* output, const void* input)
 {
-   sph_sha256_context    ctx_sha256 __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];
 
-   sph_sha256_init( &ctx_sha256 );
-   sph_sha256( &ctx_sha256, input, 112 );
-   sph_sha256_close( &ctx_sha256, hashA );
-
-   sph_sha256_init( &ctx_sha256 );
-   sph_sha256( &ctx_sha256, hashA, 32 );
-   sph_sha256_close( &ctx_sha256, hashA );
+   sha256_full( hashA, input, 112 );
+   sha256_full( hashA, hashA, 32 );
 
    sph_sha512_init( &ctx_sha512 );
    sph_sha512( &ctx_sha512, 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 );
 
-   sph_sha256_init( &ctx_sha256 );
-   sph_sha256( &ctx_sha256, hashB, 20 );
-   sph_sha256( &ctx_sha256, hashC, 20 );
-   sph_sha256_close( &ctx_sha256, hashA );
-
-   sph_sha256_init( &ctx_sha256 );
-   sph_sha256( &ctx_sha256, hashA, 32 );
-   sph_sha256_close( &ctx_sha256, hashA );
+   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 );
 }
 

algo/ripemd/sph_ripemd.c

@@ -479,7 +479,7 @@ sph_ripemd_comp(const sph_u32 msg[16], sph_u32 val[4])
  * One round of RIPEMD-128. The data must be aligned for 32-bit access.
  */
 static void
-ripemd128_round(const unsigned char *data, sph_u32 r[5])
+ripemd128_round(const unsigned char *data, sph_u32 r[4])
 {
 #if SPH_LITTLE_FAST
 

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) | \

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   
+

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]);
+   }
+}
+

algo/sha/hmac-sha256-hash.c

@@ -39,10 +39,10 @@
 void
 SHA256_Buf( const void * in, size_t len, uint8_t digest[32] )
 {
-   sph_sha256_context ctx;
-   sph_sha256_init( &ctx );
-   sph_sha256( &ctx, in, len );
-   sph_sha256_close( &ctx, digest );
+   sha256_context ctx;
+   sha256_ctx_init( &ctx );
+   sha256_update( &ctx, in, len );
+   sha256_final( &ctx, digest );
 }
 
 /**
@@ -64,7 +64,7 @@ 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 pad[64] __attribute__ ((aligned (64)));
    unsigned char khash[32];
    const unsigned char * K = _K;
    size_t i;
@@ -72,29 +72,28 @@ HMAC_SHA256_Init( HMAC_SHA256_CTX *ctx, const void *_K, size_t Klen )
    /* If Klen > 64, the key is really SHA256(K). */
    if ( Klen > 64 )
    {
-      sph_sha256_init( &ctx->ictx );
-      sph_sha256( &ctx->ictx, K, Klen );
-      sph_sha256_close( &ctx->ictx, khash );
-
+      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). */
-   sph_sha256_init( &ctx->ictx );
+   sha256_ctx_init( &ctx->ictx );
 
    for ( i = 0; i < Klen; i++ )  pad[i] = K[i] ^ 0x36;
 
    memset( pad + Klen, 0x36, 64 - Klen );
-   sph_sha256( &ctx->ictx, pad, 64 );
+   sha256_update( &ctx->ictx, pad, 64 );
 
    /* Outer SHA256 operation is SHA256(K xor [block of 0x5c] || hash). */
-   sph_sha256_init( &ctx->octx );
+   sha256_ctx_init( &ctx->octx );
 
    for ( i = 0; i < Klen; i++ )  pad[i] = K[i] ^ 0x5c;
 
    memset( pad + Klen, 0x5c, 64 - Klen );
-   sph_sha256( &ctx->octx, pad, 64 );
+   sha256_update( &ctx->octx, pad, 64 );
 }
 
 /* Add bytes to the HMAC-SHA256 operation. */
@@ -102,18 +101,17 @@ void
 HMAC_SHA256_Update( HMAC_SHA256_CTX *ctx, const void *in, size_t len )
 {
 	/* Feed data to the inner SHA256 operation. */
-   sph_sha256( &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];
-
-   sph_sha256_close( &ctx->ictx, ihash );
-   sph_sha256( &ctx->octx, ihash, 32 );
-   sph_sha256_close( &ctx->octx, digest );
+   uint32_t ihash[8] __attribute__ ((aligned (32)));
+   sha256_final( &ctx->ictx, ihash );
+   sha256_update( &ctx->octx, ihash, 32 );
+   sha256_final( &ctx->octx, digest );
 }
 
 /**
@@ -126,8 +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;
@@ -163,10 +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. */

algo/sha/hmac-sha256-hash.h

@@ -31,18 +31,18 @@
 
 #include <sys/types.h>
 #include <stdint.h>
-#include "sph_sha2.h"
+#include "sha256-hash.h"
 
 typedef struct HMAC_SHA256Context
 {
-   sph_sha256_context ictx;
-   sph_sha256_context 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] );
 

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 );
@@ -59,8 +58,16 @@ 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( __m128i *state_out,  const __m128i *data,
+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 );
+void sha256_4way_prehash_3rounds( __m128i *state_mid, __m128i *X,
+                                   const __m128i *W, const __m128i *state_in );
+void sha256_4way_final_rounds( __m128i *state_out, const __m128i *data,
+        const __m128i *state_in, const __m128i *state_mid, const __m128i *X );
+int sha256_4way_transform_le_short( __m128i *state_out, const __m128i *data,
+                                     const __m128i *state_in );
 
 #endif  // SSE2
 
@@ -72,15 +79,23 @@ 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( __m256i *state_out, const __m256i *data,
-                            const __m256i *state_in );
+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 );
+
+void sha256_8way_prehash_3rounds( __m256i *state_mid, __m256i *X,
+                                 const __m256i *W, const __m256i *state_in );
+void sha256_8way_final_rounds( __m256i *state_out, const __m256i *data,
+        const __m256i *state_in, const __m256i *state_mid, const __m256i *X );
+int sha256_8way_transform_le_short( __m256i *state_out, const __m256i *data,
+                                     const __m256i *state_in );
 
 #endif  // AVX2
 
@@ -92,19 +107,23 @@ 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( __m512i *state_out, const __m512i *data,
+void sha256_16way_transform_le( __m512i *state_out, const __m512i *data,
                              const __m512i *state_in );
-void sha256_16way_prehash_3rounds( __m512i *state_mid, const __m512i *W,
+void sha256_16way_transform_be( __m512i *state_out, const __m512i *data,
                              const __m512i *state_in );
+void sha256_16way_prehash_3rounds( __m512i *state_mid, __m512i *X,
+                                  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 );
+        const __m512i *state_in, const __m512i *state_mid, const __m512i *X );
+
+int sha256_16way_transform_le_short( __m512i *state_out, const __m512i *data,
+                                     const __m512i *state_in );
 
 #endif // AVX512
 

algo/sha/sha2.c

@@ -8,7 +8,7 @@
  * any later version.  See COPYING for more details.
  */
 
-#include "algo-gate-api.h"
+#include "sha256d-4way.h"
 
 #include <string.h>
 #include <inttypes.h>
@@ -180,6 +180,9 @@ static const uint32_t sha256d_hash1[16] = {
 	0x00000000, 0x00000000, 0x00000000, 0x00000100
 };
 
+// this performs the entire hash all over again, why?
+// because main function only does 56 rounds.
+
 static void sha256d_80_swap(uint32_t *hash, const uint32_t *data)
 {
 	uint32_t S[16];
@@ -195,6 +198,7 @@ static void sha256d_80_swap(uint32_t *hash, const uint32_t *data)
 		hash[i] = swab32(hash[i]);
 }
 
+/*
 #if defined (__SHA__)
 
 #include "algo/sha/sph_sha2.h"
@@ -241,6 +245,7 @@ void sha256d(unsigned char *hash, const unsigned char *data, int len)
 }
 
 #endif
+*/
 
 static inline void sha256d_preextend(uint32_t *W)
 {
@@ -489,7 +494,7 @@ static inline void sha256d_ms(uint32_t *hash, uint32_t *W,
 void sha256d_ms_4way(uint32_t *hash,  uint32_t *data,
 	const uint32_t *midstate, const uint32_t *prehash);
 
-static inline int scanhash_sha256d_4way( struct work *work,
+static inline int scanhash_sha256d_4way_pooler( struct work *work,
              uint32_t max_nonce, uint64_t *hashes_done, struct thr_info *mythr )
 {
         uint32_t *pdata = work->data;
@@ -550,7 +555,7 @@ static inline int scanhash_sha256d_4way( struct work *work,
 void sha256d_ms_8way(uint32_t *hash,  uint32_t *data,
 	const uint32_t *midstate, const uint32_t *prehash);
 
-static inline int scanhash_sha256d_8way( struct work *work,
+static inline int scanhash_sha256d_8way_pooler( struct work *work,
             uint32_t max_nonce, uint64_t *hashes_done, struct thr_info *mythr )
 {
         uint32_t *pdata = work->data;
@@ -606,11 +611,11 @@ static inline int scanhash_sha256d_8way( struct work *work,
 
 #endif /* HAVE_SHA256_8WAY */
 
-int scanhash_sha256d( struct work *work,
-	uint32_t max_nonce, uint64_t *hashes_done, struct thr_info *mythr )
+int scanhash_sha256d_pooler( 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 _ALIGN(128) data[64];
 	uint32_t _ALIGN(32) hash[8];
 	uint32_t _ALIGN(32) midstate[8];
@@ -621,12 +626,12 @@ int scanhash_sha256d( struct work *work,
    int thr_id = mythr->id;  // thr_id arg is deprecated
 
 #ifdef HAVE_SHA256_8WAY
-	if (sha256_use_8way())
-		return scanhash_sha256d_8way( work,	max_nonce, hashes_done, mythr );
+	if ( sha256_use_8way() )
+		return scanhash_sha256d_8way_pooler( work, max_nonce, hashes_done, mythr );
 #endif
 #ifdef HAVE_SHA256_4WAY
-	if (sha256_use_4way())
-		return scanhash_sha256d_4way( work,	max_nonce, hashes_done, mythr );
+	if ( sha256_use_4way() )
+		return scanhash_sha256d_4way_pooler( work, max_nonce, hashes_done, mythr );
 #endif
 	
 	memcpy(data, pdata + 16, 64);
@@ -653,6 +658,7 @@ 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 )
 {
@@ -682,13 +688,20 @@ int scanhash_SHA256d( struct work *work, const uint32_t max_nonce,
    pdata[19] = n;
    return 0;
 }
-
+*/
 
 bool register_sha256d_algo( algo_gate_t* gate )
 {
-   gate->optimizations = SSE2_OPT | AVX2_OPT;
-   gate->scanhash = (void*)&scanhash_sha256d;
-   gate->hash     = (void*)&sha256d;
+   gate->optimizations = SSE2_OPT | AVX2_OPT | AVX512_OPT;
+#if defined(SHA256D_16WAY)
+   gate->scanhash = (void*)&scanhash_sha256d_16way;
+//#elif defined(SHA256D_8WAY)
+//   gate->scanhash = (void*)&scanhash_sha256d_8way;
+#else
+   gate->scanhash = (void*)&scanhash_sha256d_pooler;
+//   gate->scanhash = (void*)&scanhash_sha256d_4way;
+#endif
+   //   gate->hash     = (void*)&sha256d;
    return true;
 };
 

algo/sha/sha256-hash-2way-ni.c

@@ -7,9 +7,9 @@
 
 #if defined(__SHA__)
 
-#include "sha256-hash-opt.h"
+#include "sha256-hash.h"
 
-void sha256_ni2way_transform( uint32_t *out_X, uint32_t*out_Y,
+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 )
 {
@@ -342,4 +342,348 @@ void sha256_ni2way_transform( uint32_t *out_X, uint32_t*out_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
+

algo/sha/sha256-hash-opt.c

@@ -7,9 +7,9 @@
 
 #if defined(__SHA__)
 
-#include "sha256-hash-opt.h"
+#include "sha256-hash.h"
 
-void sha256_opt_transform( uint32_t *state_out, const void *input,
+void sha256_opt_transform_le( uint32_t *state_out, const void *input,
                            const uint32_t *state_in )
 {
     __m128i STATE0, STATE1;
@@ -197,4 +197,192 @@ void sha256_opt_transform( uint32_t *state_out, const void *input,
     _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

algo/sha/sha256-hash-opt.h

@@ -1,18 +0,0 @@
-#ifndef SHA2_HASH_OPT_H__
-#define SHA2_HASH_OPT_H__ 1
-
-#include <stddef.h>
-#include "simd-utils.h"
-
-#if defined(__SHA__)
-
-void sha256_opt_transform( uint32_t *state_out, const void *input,
-                           const uint32_t *state_in );
-
-// 2 way with interleaved instructions
-void sha256_ni2way_transform( 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 );
-
-#endif
-#endif

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 );
+}
+

algo/sha/sha256-hash.h

@@ -0,0 +1,60 @@
+#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
+
+// SHA can't do only 3 rounds
+#define sha256_prehash_3rounds sph_sha256_prehash_3rounds
+
+#endif

algo/sha/sha256d-4way.c

@@ -1,4 +1,4 @@
-#include "sha256t-gate.h"
+#include "sha256d-4way.h"
 #include <stdlib.h>
 #include <stdint.h>
 #include <string.h>
@@ -10,12 +10,14 @@
 int scanhash_sha256d_16way( struct work *work, const uint32_t max_nonce,
                            uint64_t *hashes_done, struct thr_info *mythr )
 {
+   __m512i  vdata[32]    __attribute__ ((aligned (128)));
    __m512i  block[16]    __attribute__ ((aligned (64)));
-   __m512i  hash32[8]    __attribute__ ((aligned (32)));
-   __m512i  initstate[8] __attribute__ ((aligned (32)));
-   __m512i  midstate[8]  __attribute__ ((aligned (32)));
-   uint32_t lane_hash[8] __attribute__ ((aligned (32)));
-   __m512i  vdata[20]    __attribute__ ((aligned (32)));
+   __m512i  hash32[8]    __attribute__ ((aligned (64)));
+   __m512i  initstate[8] __attribute__ ((aligned (64)));
+   __m512i  midstate1[8] __attribute__ ((aligned (64)));
+   __m512i  midstate2[8] __attribute__ ((aligned (64)));
+   __m512i  mexp_pre[16] __attribute__ ((aligned (64)));
+   uint32_t lane_hash[8] __attribute__ ((aligned (64)));
    uint32_t *hash32_d7 =  (uint32_t*)&( hash32[7] );
    uint32_t *pdata = work->data;
    const uint32_t *ptarget = work->target;
@@ -23,7 +25,7 @@ int scanhash_sha256d_16way( struct work *work, const uint32_t max_nonce,
    const uint32_t first_nonce = pdata[19];
    const uint32_t last_nonce = max_nonce - 16;
    uint32_t n = first_nonce;
-   __m512i *noncev = vdata + 19; 
+   __m512i *noncev = vdata + 19;
    const int thr_id = mythr->id;
    const bool bench = opt_benchmark;
    const __m512i last_byte = m512_const1_32( 0x80000000 );
@@ -35,6 +37,14 @@ int scanhash_sha256d_16way( struct work *work, const uint32_t max_nonce,
    *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 );
 
+   vdata[16+4] = last_byte;
+   memset_zero_512( vdata+16 + 5, 10 );
+   vdata[16+15] = m512_const1_32( 80*8 ); // bit count
+
+   block[ 8] = last_byte;
+   memset_zero_512( block + 9, 6 );
+   block[15] = m512_const1_32( 32*8 ); // bit count
+   
    // initialize state
    initstate[0] = m512_const1_64( 0x6A09E6676A09E667 );
    initstate[1] = m512_const1_64( 0xBB67AE85BB67AE85 );
@@ -45,47 +55,42 @@ int scanhash_sha256d_16way( struct work *work, const uint32_t max_nonce,
    initstate[6] = m512_const1_64( 0x1F83D9AB1F83D9AB );
    initstate[7] = m512_const1_64( 0x5BE0CD195BE0CD19 );
 
-   // hash first 64 bytes of data
-   sha256_16way_transform( midstate, vdata, initstate );
+   sha256_16way_transform_le( midstate1, vdata, initstate );
+
+   // Do 3 rounds on the first 12 bytes of the next block
+   sha256_16way_prehash_3rounds( midstate2, mexp_pre, vdata+16, midstate1 );
 
    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_transform( hash32, block, midstate );
+      sha256_16way_final_rounds( block, vdata+16, midstate1, midstate2,
+                                 mexp_pre );
 
       // 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( 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 ) )
+      if ( sha256_16way_transform_le_short( hash32, block, initstate ) )
       {
-         extr_lane_16x32( lane_hash, hash32, lane, 256 );
-         if ( likely( valid_hash( lane_hash, ptarget ) && !bench ) )
+         // 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 ) )
          {
-            pdata[19] = n + lane;
-            submit_solution( work, lane_hash, mythr );
+            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;
+      }
+      *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)
@@ -93,12 +98,14 @@ int scanhash_sha256d_16way( struct work *work, const uint32_t max_nonce,
 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  vdata[32]    __attribute__ ((aligned (64)));
+   __m256i  block[16]    __attribute__ ((aligned (32)));
    __m256i  hash32[8]    __attribute__ ((aligned (32)));
    __m256i  initstate[8] __attribute__ ((aligned (32)));
-   __m256i  midstate[8]  __attribute__ ((aligned (32)));
+   __m256i  midstate1[8] __attribute__ ((aligned (32)));
+   __m256i  midstate2[8] __attribute__ ((aligned (32)));
+   __m256i  mexp_pre[16] __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;
@@ -113,10 +120,18 @@ int scanhash_sha256d_8way( struct work *work, const uint32_t max_nonce,
    const __m256i eight = m256_const1_32( 8 );
 
    for ( int i = 0; i < 19; i++ )
-       vdata[i] = m256_const1_32( pdata[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 );
 
+   vdata[16+4] = last_byte;
+   memset_zero_256( vdata+16 + 5, 10 );
+   vdata[16+15] = m256_const1_32( 80*8 ); // bit count
+
+   block[ 8] = last_byte;
+   memset_zero_256( block + 9, 6 );
+   block[15] = m256_const1_32( 32*8 ); // bit count
+   
    // initialize state
    initstate[0] = m256_const1_64( 0x6A09E6676A09E667 );
    initstate[1] = m256_const1_64( 0xBB67AE85BB67AE85 );
@@ -127,36 +142,33 @@ int scanhash_sha256d_8way( struct work *work, const uint32_t max_nonce,
    initstate[6] = m256_const1_64( 0x1F83D9AB1F83D9AB );
    initstate[7] = m256_const1_64( 0x5BE0CD195BE0CD19 );
 
-   // hash first 64 bytes of data
-   sha256_8way_transform( midstate, vdata, initstate );
+   sha256_8way_transform_le( midstate1, vdata, initstate );
+   
+   // Do 3 rounds on the first 12 bytes of the next block
+   sha256_8way_prehash_3rounds( midstate2, mexp_pre, vdata + 16, midstate1 );
 
    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( hash32, block, midstate );
+      sha256_8way_final_rounds( block, vdata+16, midstate1, midstate2,
+                                mexp_pre );
 
       // 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( 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 ) )
+      if ( unlikely(
+               sha256_8way_transform_le_short( hash32, block, initstate ) ) )
       {
-         extr_lane_8x32( lane_hash, hash32, lane, 256 );
-         if ( likely( valid_hash( lane_hash, ptarget ) && !bench ) )
+         // 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 ) )
          {
-            pdata[19] = n + lane;
-            submit_solution( work, lane_hash, mythr );
+            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 );
@@ -174,12 +186,14 @@ int scanhash_sha256d_8way( struct work *work, const uint32_t max_nonce,
 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)));
+   __m128i  vdata[32]     __attribute__ ((aligned (64)));
+   __m128i  block[16]     __attribute__ ((aligned (32)));
+   __m128i  hash32[8]     __attribute__ ((aligned (32)));
+   __m128i  initstate[8]  __attribute__ ((aligned (32)));
+   __m128i  midstate1[8]   __attribute__ ((aligned (32)));
+   __m128i  midstate2[8]  __attribute__ ((aligned (32)));
+   __m128i  mexp_pre[16]  __attribute__ ((aligned (32)));
+   uint32_t lane_hash[8]  __attribute__ ((aligned (32)));
    uint32_t *hash32_d7 =  (uint32_t*)&( hash32[7] );
    uint32_t *pdata = work->data;
    const uint32_t *ptarget = work->target;
@@ -198,6 +212,14 @@ int scanhash_sha256d_4way( struct work *work, const uint32_t max_nonce,
 
    *noncev = _mm_set_epi32( n+ 3, n+ 2, n+1, n );
 
+   vdata[16+4] = last_byte;
+   memset_zero_128( vdata+16 + 5, 10 );
+   vdata[16+15] = m128_const1_32( 80*8 ); // bit count
+
+   block[ 8] = last_byte;
+   memset_zero_128( block + 9, 6 );
+   block[15] = m128_const1_32( 32*8 ); // bit count
+
    // initialize state
    initstate[0] = m128_const1_64( 0x6A09E6676A09E667 );
    initstate[1] = m128_const1_64( 0xBB67AE85BB67AE85 );
@@ -209,39 +231,36 @@ int scanhash_sha256d_4way( struct work *work, const uint32_t max_nonce,
    initstate[7] = m128_const1_64( 0x5BE0CD195BE0CD19 );
 
    // hash first 64 bytes of data
-   sha256_4way_transform( midstate, vdata, initstate );
+   sha256_4way_transform_le( midstate1, vdata, initstate );
+   // Do 3 rounds on the first 12 bytes of the next block
+   sha256_4way_prehash_3rounds( midstate2, mexp_pre, vdata + 16, midstate1 );
 
    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( hash32, block, midstate );
+      sha256_4way_final_rounds( block, vdata+16, midstate1, midstate2,
+                                mexp_pre );
 
       // 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( 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 ) )
+      if ( unlikely(
+              sha256_4way_transform_le_short( hash32, block, initstate ) ) )
       {
-         extr_lane_4x32( lane_hash, hash32, lane, 256 );
-         if ( likely( valid_hash( lane_hash, ptarget ) && !bench ) )
+         // 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 ) )
          {
-            pdata[19] = n + lane;
-            submit_solution( work, lane_hash, mythr );
+            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;
+      }
+      *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;
@@ -250,3 +269,20 @@ int scanhash_sha256d_4way( struct work *work, const uint32_t max_nonce,
 
 #endif
 
+/*
+bool register_sha256d_algo( algo_gate_t* gate )
+{
+   gate->optimizations = SSE2_OPT | AVX2_OPT | AVX512_OPT;
+#if defined(SHA256D_16WAY)
+   gate->scanhash = (void*)&scanhash_sha256d_16way;
+#elif defined(SHA256D_8WAY)
+   gate->scanhash = (void*)&scanhash_sha256d_8way;
+#elif defined(SHA256D_4WAY)
+   gate->scanhash = (void*)&scanhash_sha256d_4way;
+#endif
+   
+//   gate->hash     = (void*)&sha256d;
+   return true;
+};
+*/
+

algo/sha/sha256d-4way.h

@@ -0,0 +1,46 @@
+#ifndef __SHA256D_4WAY_H__
+#define __SHA256D_4WAY_H__ 1
+
+#include <stdint.h>
+#include "algo-gate-api.h"
+
+#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+  #define SHA256D_16WAY 1
+#elif defined(__AVX2__)
+  #define SHA256D_8WAY 1
+#else
+  #define SHA256D_4WAY 1
+#endif
+
+bool register_sha256d_algo( algo_gate_t* gate );
+
+#if defined(SHA256D_16WAY)
+
+int scanhash_sha256d_16way( struct work *work, uint32_t max_nonce,
+                           uint64_t *hashes_done, struct thr_info *mythr );
+#endif
+
+#if defined(SHA256D_8WAY)
+
+int scanhash_sha256d_8way( struct work *work, uint32_t max_nonce,
+                           uint64_t *hashes_done, struct thr_info *mythr );
+#endif
+
+#if defined(SHA256D_4WAY)
+
+int scanhash_sha256d_4way( struct work *work, uint32_t max_nonce,
+                           uint64_t *hashes_done, struct thr_info *mythr );
+#endif
+
+
+/*
+#if defined(__SHA__)
+
+int scanhash_sha256d( struct work *work, uint32_t max_nonce,
+                      uint64_t *hashes_done, struct thr_info *mythr );
+
+#endif
+*/
+
+#endif
+

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 );
+}
+

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 );
+

algo/sha/sha256q.c

@@ -3,14 +3,14 @@
 #include <stdint.h>
 #include <string.h>
 #include <stdio.h>
-#include "algo/sha/sph_sha2.h"
+#include "algo/sha/sha256-hash.h"
 
-static __thread sph_sha256_context sha256q_ctx __attribute__ ((aligned (64)));
+static __thread sha256_context sha256q_ctx __attribute__ ((aligned (64)));
 
 void sha256q_midstate( const void* input )
 {
-   sph_sha256_init( &sha256q_ctx );
-   sph_sha256( &sha256q_ctx, input, 64 );
+   sha256_ctx_init( &sha256q_ctx );
+   sha256_update( &sha256q_ctx, input, 64 );
 }
 
 int sha256q_hash( void* output, const void* input )
@@ -19,24 +19,16 @@ int sha256q_hash( void* output, const void* input )
    const int midlen = 64;            // bytes
    const int tail   = 80 - midlen;   // 16
 
-   sph_sha256_context ctx __attribute__ ((aligned (64)));
+   sha256_context ctx __attribute__ ((aligned (64)));
    memcpy( &ctx, &sha256q_ctx, sizeof sha256q_ctx );
 
-   sph_sha256( &ctx, input + midlen, tail );
-   sph_sha256_close( &ctx, hash );
-
-   sph_sha256_init( &ctx );
-   sph_sha256( &ctx, hash, 32 );
-   sph_sha256_close( &ctx, hash );
-
-   sph_sha256_init( &ctx );
-   sph_sha256( &ctx, hash, 32 );
-   sph_sha256_close( &ctx, hash );
-
-   sph_sha256_init( &ctx );
-   sph_sha256( &ctx, hash, 32 );
-   sph_sha256_close( &ctx, output );
+   sha256_update( &ctx, input + midlen, tail );
+   sha256_final( &ctx, hash );
 
+   sha256_full( hash,   hash, 32 );
+   sha256_full( hash,   hash, 32 );
+   sha256_full( output, hash, 32 );
+   
    return 1;
 }
 

algo/sha/sha256t-4way.c

@@ -10,13 +10,14 @@
 int scanhash_sha256t_16way( struct work *work, const uint32_t max_nonce,
                            uint64_t *hashes_done, struct thr_info *mythr )
 {
+   __m512i  vdata[32]    __attribute__ ((aligned (128)));
    __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)));
+   __m512i  hash32[8]    __attribute__ ((aligned (64)));
+   __m512i  initstate[8] __attribute__ ((aligned (64)));
+   __m512i  midstate1[8] __attribute__ ((aligned (64)));
+   __m512i  midstate2[8] __attribute__ ((aligned (64)));
+   __m512i  mexp_pre[16] __attribute__ ((aligned (64)));
+   uint32_t lane_hash[8] __attribute__ ((aligned (64)));
    uint32_t *hash32_d7 =  (uint32_t*)&( hash32[7] );
    uint32_t *pdata = work->data;
    const uint32_t *ptarget = work->target;
@@ -31,12 +32,19 @@ int scanhash_sha256t_16way( struct work *work, const uint32_t max_nonce,
    const __m512i sixteen = m512_const1_32( 16 );
 
    for ( int i = 0; i < 19; i++ )
-       vdata[i] = m512_const1_32( pdata[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
+   vdata[16+4] = last_byte;
+   memset_zero_512( vdata+16 + 5, 10 );
+   vdata[16+15] = m512_const1_32( 80*8 ); // bit count
+   
+   block[ 8] = last_byte;
+   memset_zero_512( block + 9, 6 );
+   block[15] = m512_const1_32( 32*8 ); // bit count
+   
    initstate[0] = m512_const1_64( 0x6A09E6676A09E667 );
    initstate[1] = m512_const1_64( 0xBB67AE85BB67AE85 );
    initstate[2] = m512_const1_64( 0x3C6EF3723C6EF372 );
@@ -46,48 +54,40 @@ int scanhash_sha256t_16way( struct work *work, const uint32_t max_nonce,
    initstate[6] = m512_const1_64( 0x1F83D9AB1F83D9AB );
    initstate[7] = m512_const1_64( 0x5BE0CD195BE0CD19 );
 
-   // hash first 64 byte block of data
-   sha256_16way_transform( midstate, vdata, initstate );
-
+   sha256_16way_transform_le( midstate1, vdata, initstate );
+   
    // Do 3 rounds on the first 12 bytes of the next block
-   sha256_16way_prehash_3rounds( midstate2, vdata + 16, midstate );
+   sha256_16way_prehash_3rounds( midstate2, mexp_pre, vdata+16, midstate1 );
 
    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 );
-//      sha256_16way_transform( hash32, block, midstate );
+      // 1. final 16 bytes of data, pre-padded
+      sha256_16way_final_rounds( block, vdata+16, midstate1, midstate2,
+                                 mexp_pre );
 
       // 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( hash32, block, initstate );
+      sha256_16way_transform_le( block, block, initstate );
 
       // 3. 32 byte hash from 2.
-      memcpy_512( block, hash32, 8 );
-      sha256_16way_transform( 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 ) )
+      if ( unlikely(
+               sha256_16way_transform_le_short( hash32, block, initstate ) ) )
       {
-         extr_lane_16x32( lane_hash, hash32, lane, 256 );
-         if ( likely( valid_hash( lane_hash, ptarget ) && !bench ) )
+         // byte swap final hash for testing
+         mm512_block_bswap_32( hash32, hash32 );    
+
+         for ( int lane = 0; lane < 16; lane++ )
+         if ( hash32_d7[ lane ] <= targ32_d7 )
          {
-            pdata[19] = n + lane;
-            submit_solution( work, lane_hash, mythr );
+            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;
+      }
+      *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;
@@ -102,12 +102,14 @@ int scanhash_sha256t_16way( struct work *work, const uint32_t max_nonce,
 int scanhash_sha256t_8way( struct work *work, const uint32_t max_nonce,
                            uint64_t *hashes_done, struct thr_info *mythr )
 {
-   __m256i  block[16]    __attribute__ ((aligned (64)));
+   __m256i  vdata[32]    __attribute__ ((aligned (64)));
+   __m256i  block[16]    __attribute__ ((aligned (32)));
    __m256i  hash32[8]    __attribute__ ((aligned (32)));
    __m256i  initstate[8] __attribute__ ((aligned (32)));
-   __m256i  midstate[8]  __attribute__ ((aligned (32)));
+   __m256i  midstate1[8] __attribute__ ((aligned (32)));
+   __m256i  midstate2[8] __attribute__ ((aligned (32)));
+   __m256i  mexp_pre[16] __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;
@@ -122,10 +124,18 @@ int scanhash_sha256t_8way( struct work *work, const uint32_t max_nonce,
    const __m256i eight = m256_const1_32( 8 );
 
    for ( int i = 0; i < 19; i++ )
-       vdata[i] = m256_const1_32( pdata[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 );
 
+   vdata[16+4] = last_byte;
+   memset_zero_256( vdata+16 + 5, 10 );
+   vdata[16+15] = m256_const1_32( 80*8 ); // bit count
+
+   block[ 8] = last_byte;
+   memset_zero_256( block + 9, 6 );
+   block[15] = m256_const1_32( 32*8 ); // bit count
+   
    // initialize state
    initstate[0] = m256_const1_64( 0x6A09E6676A09E667 );
    initstate[1] = m256_const1_64( 0xBB67AE85BB67AE85 );
@@ -136,44 +146,40 @@ int scanhash_sha256t_8way( struct work *work, const uint32_t max_nonce,
    initstate[6] = m256_const1_64( 0x1F83D9AB1F83D9AB );
    initstate[7] = m256_const1_64( 0x5BE0CD195BE0CD19 );
 
-   // hash first 64 bytes of data
-   sha256_8way_transform( midstate, vdata, initstate );
+   sha256_8way_transform_le( midstate1, vdata, initstate );
 
+   // Do 3 rounds on the first 12 bytes of the next block
+   sha256_8way_prehash_3rounds( midstate2, mexp_pre, vdata + 16, midstate1 );
+   
    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( hash32, block, midstate );
+      sha256_8way_final_rounds( block, vdata+16, midstate1, midstate2,
+                                mexp_pre );
 
       // 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( hash32, block, initstate );
+      sha256_8way_transform_le( block, block, initstate );
 
       // 3. 32 byte hash from 2.
-      memcpy_256( block, hash32, 8 );
-      sha256_8way_transform( 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 ) )
+      if ( unlikely(
+               sha256_8way_transform_le_short( hash32, block, initstate ) ) )
       {
-         extr_lane_8x32( lane_hash, hash32, lane, 256 );
-         if ( likely( valid_hash( lane_hash, ptarget ) && !bench ) )
+         // byte swap final hash for testing
+         mm256_block_bswap_32( hash32, hash32 );
+
+         for ( int lane = 0; lane < 8; lane++ )
+         if ( hash32_d7[ lane ] <= targ32_d7 )
          {
-            pdata[19] = n + lane;
-            submit_solution( work, lane_hash, mythr );
+            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;
+      }
+      *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;
@@ -182,17 +188,110 @@ int scanhash_sha256t_8way( struct work *work, const uint32_t max_nonce,
 
 #endif
 
+
 #if defined(SHA256T_4WAY)
 
+// Optimizations are slower with AVX/SSE2
+// https://github.com/JayDDee/cpuminer-opt/issues/344
+/*
+int scanhash_sha256t_4way( struct work *work, const uint32_t max_nonce,
+                           uint64_t *hashes_done, struct thr_info *mythr )
+{
+   __m128i  vdata[32]     __attribute__ ((aligned (64)));
+   __m128i  block[16]     __attribute__ ((aligned (32)));
+   __m128i  hash32[8]     __attribute__ ((aligned (32)));
+   __m128i  initstate[8]  __attribute__ ((aligned (32)));
+   __m128i  midstate1[8]  __attribute__ ((aligned (32)));
+   __m128i  midstate2[8]  __attribute__ ((aligned (32)));
+   __m128i  mexp_pre[16]  __attribute__ ((aligned (32)));
+   uint32_t lane_hash[8]  __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 );
+
+   vdata[16+4] = last_byte;
+   memset_zero_128( vdata+16 + 5, 10 );
+   vdata[16+15] = m128_const1_32( 80*8 ); // bit count
+
+   block[ 8] = last_byte;
+   memset_zero_128( block + 9, 6 );
+   block[15] = m128_const1_32( 32*8 ); // bit count
+   
+   // 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( midstate1, vdata, initstate );
+
+   // Do 3 rounds on the first 12 bytes of the next block
+   sha256_4way_prehash_3rounds( midstate2, mexp_pre, vdata + 16, midstate1 );
+
+   do
+   {
+      // 1. final 16 bytes of data, with padding
+      sha256_4way_final_rounds( block, vdata+16, midstate1, midstate2,
+                                mexp_pre );
+
+      // 2. 32 byte hash from 1.
+      sha256_4way_transform_le( block, block, initstate );
+
+      // 3. 32 byte hash from 2.
+      if ( unlikely(
+              sha256_4way_transform_le_short( 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;
+}
+*/
+
 int scanhash_sha256t_4way( struct work *work, const uint32_t max_nonce,
                            uint64_t *hashes_done, struct thr_info *mythr )
 {
-   __m128i  block[16]    __attribute__ ((aligned (64)));
+   __m128i  vdata[32]    __attribute__ ((aligned (64)));
+   __m128i  block[16]    __attribute__ ((aligned (32)));
    __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;
@@ -211,6 +310,14 @@ int scanhash_sha256t_4way( struct work *work, const uint32_t max_nonce,
 
    *noncev = _mm_set_epi32( n+ 3, n+ 2, n+1, n );
 
+   vdata[16+4] = last_byte;
+   memset_zero_128( vdata+16 + 5, 10 );
+   vdata[16+15] = m128_const1_32( 80*8 ); // bit count
+
+   block[ 8] = last_byte;
+   memset_zero_128( block + 9, 6 );
+   block[15] = m128_const1_32( 32*8 ); // bit count
+   
    // initialize state
    initstate[0] = m128_const1_64( 0x6A09E6676A09E667 );
    initstate[1] = m128_const1_64( 0xBB67AE85BB67AE85 );
@@ -222,29 +329,13 @@ int scanhash_sha256t_4way( struct work *work, const uint32_t max_nonce,
    initstate[7] = m128_const1_64( 0x5BE0CD195BE0CD19 );
 
    // hash first 64 bytes of data
-   sha256_4way_transform( midstate, vdata, initstate );
+   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( 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( hash32, block, initstate );
-
-      // 3. 32 byte hash from 2.
-      memcpy_128( block, hash32, 8 );
-      sha256_4way_transform( hash32, block, initstate );
-
-      // byte swap final hash for testing
+      sha256_4way_transform_le( block,  vdata+16, midstate  );
+      sha256_4way_transform_le( block,  block,    initstate );
+      sha256_4way_transform_le( hash32, block,    initstate );
       mm128_block_bswap_32( hash32, hash32 );
 
       for ( int lane = 0; lane < 4; lane++ )
@@ -265,5 +356,6 @@ int scanhash_sha256t_4way( struct work *work, const uint32_t max_nonce,
    return 0;
 }
 
+
 #endif
 

algo/sha/sha256t.c

@@ -4,120 +4,12 @@
 #include <string.h>
 #include <stdio.h>
 //#include "algo/sha/sph_sha2.h"
-#include "sha256-hash-opt.h"
+#include "sha256-hash.h"
 
 #if defined(__SHA__)
 
 // Only used on CPUs with SHA
 
-/*
-static __thread sph_sha256_context sha256t_ctx __attribute__ ((aligned (64)));
-
-void sha256t_midstate( const void* input )
-{
-   sph_sha256_init( &sha256t_ctx );
-   sph_sha256( &sha256t_ctx, input, 64 );
-}
-
-int sha256t_hash( void* output, const void* input )
-{
-   uint32_t _ALIGN(64) hash[16];
-   const int midlen = 64;            // bytes
-   const int tail   = 80 - midlen;   // 16
-
-   sph_sha256_context ctx __attribute__ ((aligned (64)));
-   memcpy( &ctx, &sha256t_ctx, sizeof sha256t_ctx );
-
-   sph_sha256( &ctx, input + midlen, tail );
-   sph_sha256_close( &ctx, hash );
-
-   sph_sha256_init( &ctx );
-   sph_sha256( &ctx, hash, 32 );
-   sph_sha256_close( &ctx, hash );
-
-   sph_sha256_init( &ctx );
-   sph_sha256( &ctx, hash, 32 );
-   sph_sha256_close( &ctx, output );
-
-   return 1;
-}
-*/
-
-/*
-int scanhash_sha256t( struct work *work, uint32_t max_nonce,
-                      uint64_t *hashes_done, struct thr_info *mythr )
-{
-   uint32_t block[16]    __attribute__ ((aligned (64)));
-   uint32_t hash32[8]    __attribute__ ((aligned (32)));
-   uint32_t initstate[8] __attribute__ ((aligned (32)));
-   uint32_t midstate[8]  __attribute__ ((aligned (32)));
-
-
-
-//   uint32_t edata[20] __attribute__((aligned(64)));
-//   uint32_t hash[8] __attribute__((aligned(64)));
-   uint32_t *pdata = work->data;
-   uint32_t *ptarget = work->target;
-   const uint32_t first_nonce = pdata[19];
-   const uint32_t 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 );
-
-//   mm128_bswap32_80( edata, pdata );
-//   sha256t_midstate( edata );
-
-   // 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;
-
-   // hash first 64 bytes of data
-   sha256_opt_transform( midstate, pdata, initstate );
-
-   do
-   {
-      // 1. final 16 bytes of data, with padding
-      memcpy( block, pdata + 16, 16 );
-      block[ 4] = 0x80000000;
-      memset( block + 5, 0, 40 );
-      block[15] = 80*8; // bit count
-      sha256_opt_transform( hash32, block, midstate );
-
-      // 2. 32 byte hash from 1.
-      memcpy( block, hash32, 32 );
-      block[ 8] = 0x80000000;
-      memset( block + 9, 0, 24 );
-      block[15] = 32*8; // bit count
-      sha256_opt_transform( hash32, block, initstate );
-
-      // 3. 32 byte hash from 2.
-      memcpy( block, hash32, 32 );
-      sha256_opt_transform( hash32, block, initstate );
-
-      // byte swap final hash for testing
-      casti_m128i( hash32, 0 ) =
-               _mm_shuffle_epi8( casti_m128i( hash32, 0 ), shuf_bswap32 );
-      casti_m128i( hash32, 1 ) =
-               _mm_shuffle_epi8( casti_m128i( hash32, 1 ), shuf_bswap32 );
-
-      if ( unlikely( valid_hash( hash32, ptarget ) && !bench ) )
-         submit_solution( work, hash32, mythr );
-      n++;
-      pdata[19] = n;
-   } while ( (n < last_nonce) && !work_restart[thr_id].restart );
-
-   *hashes_done = n - first_nonce;
-   return 0;
-}
-*/
 
 int scanhash_sha256t( struct work *work, uint32_t max_nonce,
                       uint64_t *hashes_done, struct thr_info *mythr )
@@ -149,7 +41,7 @@ int scanhash_sha256t( struct work *work, uint32_t max_nonce,
    initstate[7] = 0x5BE0CD19;
 
    // hash first 64 bytes of data
-   sha256_opt_transform( midstate, pdata, initstate );
+   sha256_opt_transform_le( midstate, pdata, initstate );
 
    do
    {
@@ -162,7 +54,7 @@ int scanhash_sha256t( struct work *work, uint32_t max_nonce,
       memset( block0 + 5, 0, 40 );
       memset( block1 + 5, 0, 40 );
       block0[15] = block1[15] = 80*8; // bit count
-      sha256_ni2way_transform( hash0, hash1, block0, block1, midstate, midstate );
+      sha256_ni2way_transform_le( hash0, hash1, block0, block1, midstate, midstate );
 
       // 2. 32 byte hash from 1.
       memcpy( block0, hash0, 32 );
@@ -171,12 +63,12 @@ int scanhash_sha256t( struct work *work, uint32_t max_nonce,
       memset( block0 + 9, 0, 24 );
       memset( block1 + 9, 0, 24 );
       block0[15] = block1[15] = 32*8; // bit count
-      sha256_ni2way_transform( hash0, hash1, block0, block1, initstate, initstate );
+      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( hash0, hash1, block0, block1, initstate, initstate );
+      sha256_ni2way_transform_le( hash0, hash1, block0, block1, initstate, initstate );
 
       // byte swap final hash for testing
       casti_m128i( hash0, 0 ) =

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

@@ -95,32 +95,36 @@ static const uint64_t K512[80] =
 
 // SHA-512 8 way 64 bit
 
-#define CH8W(X, Y, Z) \
-   _mm512_ternarylogic_epi64( X, Y, Z, 0xca )
+#define CH8W( X, Y, Z )    _mm512_ternarylogic_epi64( X, Y, Z, 0xca )
 
-#define MAJ8W(X, Y, Z) \
-   _mm512_ternarylogic_epi64( X, Y, Z, 0xe8 )
+#define MAJ8W( X, Y, Z )   _mm512_ternarylogic_epi64( X, Y, Z, 0xe8 )
 
-#define BSG8W_5_0(x) \
-   mm512_xor3( 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_xor3( 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_xor3( 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_xor3( 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 ) )
 
-#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)
 
@@ -267,16 +271,9 @@ 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 ) )
-*/
-
 #define MAJ(X, Y, Z) \
   _mm256_xor_si256( Y, _mm256_and_si256( X_xor_Y = _mm256_xor_si256( X, Y ), \
                                          Y_xor_Z ) )
@@ -289,15 +286,6 @@ void sha512_8way_close( sha512_8way_context *sc, void *dst )
   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( \
@@ -325,94 +313,20 @@ 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 ); \
-  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) ); \
+  __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 ); \
+  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] )
 {

algo/sha/sph_sha2.c

@@ -71,198 +71,6 @@ static const sph_u32 H256[8] = {
  * of the compression function.
  */
 
-#if defined(__SHA__)
-
-#include "simd-utils.h"
-
-static void sha2_round( const uint8_t input[], uint32_t state[8] )
-{
-    __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[0]);
-    STATE1 = _mm_load_si128((__m128i*) &state[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
-    MSG = _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[0], STATE0);
-    _mm_store_si128((__m128i*) &state[4], STATE1);
-}
-
-#else   // no SHA
 
 /*
 static const sph_u32 K[64] = {
@@ -875,8 +683,54 @@ sha2_round(const unsigned char *data, sph_u32 r[8])
 #undef SHA2_IN
 }
 
-#endif   // SHA else
+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
+
+}
+
+void sph_sha256_prehash_3rounds( uint32_t *state_out, const uint32_t *data,
+                                 const uint32_t *state_in )
+{
+   uint32_t t1, t2, X_xor_Y, Y_xor_Z = state_in[1] ^ state_in[2];
+   memcpy( state_out, state_in, 32 );
+
+   t1 = state_out[7] + BSG2_1( state_out[4] )
+      + CH( state_out[4], state_out[5], state_out[6] ) + 0x428A2F98 + data[0];
+   t2 = BSG2_0( state_out[0] )
+      + MAJ( state_out[0], state_out[1], state_out[2] );
+   Y_xor_Z = X_xor_Y;
+   state_out[3] += t1;
+   state_out[7] = t1 + t2;
+
+   t1 = state_out[6] + BSG2_1( state_out[3] ) 
+      + CH( state_out[3], state_out[4], state_out[5] ) + 0x71374491 + data[1];
+   t2 = BSG2_0( state_out[7] )
+      + MAJ( state_out[7], state_out[0], state_out[1] );
+   Y_xor_Z = X_xor_Y;
+   state_out[2] += t1;
+   state_out[6] = t1 + t2;
+
+   t1 = state_out[5] + BSG2_1( state_out[2] )
+      + CH( state_out[2], state_out[3], state_out[4] ) + 0xB5C0FBCF + data[2];
+   t2 = BSG2_0( state_out[6] )
+      + MAJ( state_out[6], state_out[7], state_out[0] );
+   state_out[1] += t1;
+   state_out[5] = t1 + t2;
+}   
 
 /* see sph_sha2.h */
 void

algo/sha/sph_sha2.h

@@ -207,6 +207,16 @@ void sph_sha256_comp(const sph_u32 msg[16], sph_u32 val[8]);
 
 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 );
+
+void sph_sha256_prehash_3rounds( uint32_t *state_out, const uint32_t *data,
+                                 const uint32_t *state_in );
 
 
 #if SPH_64

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 \
@@ -314,8 +316,7 @@ do { \
             _mm256_andnot_si256( xb3, xb2 ), \
             _mm256_mullo_epi32( mm256_xor3( xa0, xc, \
                _mm256_mullo_epi32( mm256_rol_32( xa1, 15 ), \
-                                   _mm256_set1_epi32(5UL) ) ), \
-               _mm256_set1_epi32(3UL) ) ) ); \
+                                   FIVE ) ), THREE ) ) ); \
    xb0 = mm256_xnor( xa0, mm256_rol_32( xb0, 1 ) ); \
 } while (0)
 
@@ -667,7 +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 \
 { \
@@ -931,8 +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)
 

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

@@ -20,8 +20,8 @@ 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__)
 
@@ -78,7 +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 )
@@ -88,7 +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(
@@ -97,25 +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(
@@ -151,31 +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 );
 
@@ -209,35 +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 );
 

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 )
@@ -60,7 +60,7 @@ 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 )
@@ -69,33 +69,33 @@ c512_4way( shavite512_4way_context *ctx, const void *msg )
 
      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_ror128_32(
+        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_swap128_64(
@@ -130,31 +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 );
 
@@ -187,34 +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 );
 

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 )
@@ -135,7 +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 )
@@ -144,7 +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 )
@@ -153,31 +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 )
@@ -222,38 +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 );
@@ -295,39 +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 );

algo/skein/skein-4way.c

@@ -3,7 +3,7 @@
 #include <stdint.h>
 #include "skein-hash-4way.h"
 #include "algo/sha/sha-hash-4way.h"
-#include "algo/sha/sph_sha2.h"
+#include "algo/sha/sha256-hash.h"
 
 #if defined (SKEIN_8WAY)
 
@@ -87,7 +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)));
-     sph_sha256_context ctx_sha256;
 #else
      uint32_t vhash32[16*4] __attribute__ ((aligned (64)));
      sha256_4way_context ctx_sha256;
@@ -98,18 +97,12 @@ void skeinhash_4way( void *state, const void *input )
 #if defined(__SHA__)      
 
      dintrlv_4x64( hash0, hash1, hash2, hash3, vhash64, 512 );
-     sph_sha256_init( &ctx_sha256 );
-     sph_sha256( &ctx_sha256, hash0, 64 );
-     sph_sha256_close( &ctx_sha256, hash0 );
-     sph_sha256_init( &ctx_sha256 );
-     sph_sha256( &ctx_sha256, hash1, 64 );
-     sph_sha256_close( &ctx_sha256, hash1 );
-     sph_sha256_init( &ctx_sha256 );
-     sph_sha256( &ctx_sha256, hash2, 64 );
-     sph_sha256_close( &ctx_sha256, hash2 );
-     sph_sha256_init( &ctx_sha256 );
-     sph_sha256( &ctx_sha256, hash3, 64 );
-     sph_sha256_close( &ctx_sha256, hash3 );
+
+     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

algo/skein/skein.c

@@ -5,21 +5,18 @@
 #include <string.h>
 #include <stdint.h>
 #include "sph_skein.h"
-#include "algo/sha/sph_sha2.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;
-     sph_sha256_context   ctx_sha256;
 
      sph_skein512_init( &ctx_skein );
      sph_skein512( &ctx_skein, input, 80 );
      sph_skein512_close( &ctx_skein, hash );
 
-     sph_sha256_init( &ctx_sha256 );
-     sph_sha256( &ctx_sha256, hash, 64 );
-     sph_sha256_close( &ctx_sha256, hash );
+     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); 

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;

algo/swifftx/swifftx-4way.c

@@ -1,912 +0,0 @@
-///////////////////////////////////////////////////////////////////////////////////////////////
-//
-//  SWIFFTX ANSI C OPTIMIZED 32BIT IMPLEMENTATION FOR NIST SHA-3 COMPETITION
-//
-//  SWIFFTX.c
-//
-//  October 2008
-//
-//  This is the source file of the OPTIMIZED 32BIT implementation of SWIFFTX hash function.
-//  SWIFFTX is a candidate function for SHA-3 NIST competition.
-//  More details about SWIFFTX can be found in the accompanying submission documents.
-//
-///////////////////////////////////////////////////////////////////////////////////////////////
-#include "swifftx.h"
-// See the remarks concerning compatibility issues inside stdint.h.
-#include "stdint.h"
-// Remove this while using gcc:
-//#include "stdbool.h"
-#include <memory.h>
-
-///////////////////////////////////////////////////////////////////////////////////////////////
-// Constants and static tables portion.
-///////////////////////////////////////////////////////////////////////////////////////////////
-
-// In SWIFFTX we work over Z_257, so this is the modulus and the arithmetic is performed modulo
-// this number.
-#define FIELD_SIZE 257
-
-// The size of FFT we use:
-#define N 64
-
-#define LOGN 6
-
-#define EIGHTH_N (N / 8)
-
-// The number of FFTS done on the input.
-#define M (SWIFFTX_INPUT_BLOCK_SIZE / 8)   // 32
-
-// Omega is the 128th root of unity in Z_257.
-// We choose w = 42.
-#define OMEGA 42
-
-// The size of the inner FFT lookup table:
-#define W 8
-
-// Calculates the sum and the difference of two numbers.
-//
-// Parameters:
-// - A: the first operand. After the operation stores the sum of the two operands.
-// - B: the second operand. After the operation stores the difference between the first and the
-//   second operands.
-#define ADD_SUB_4WAY( A, B ) \
-{ \
-  __m128i temp = B; \
-  B = _mm_sub_epi32( A, B ); \
-  A = _mm_add_epi32( A, temp ); \
-}
-
-
-//#define ADD_SUB(A, B) {register int temp = (B); B = ((A) - (B)); A = ((A) + (temp));}
-
-// Quickly reduces an integer modulo 257.
-//
-// Parameters:
-// - A: the input.
-
-#define Q_REDUCE( A ) ( _mm_sub_epi32( \
-                               _mm_and_epi32( A, m128_const1_32( 0xff ) ), \
-                               _mm_srli_epi32( A, 8 ) ) )
-
-//#define Q_REDUCE(A) (((A) & 0xff) - ((A) >> 8))
-
-// Since we need to do the setup only once, this is the indicator variable:
-static bool wasSetupDone = false;
-
-// This array stores the powers of omegas that correspond to the indices, which are the input
-// values. Known also as the "outer FFT twiddle factors".
-swift_int16_t multipliers[N];
-
-// This array stores the powers of omegas, multiplied by the corresponding values.
-// We store this table to save computation time.
-//
-// To calculate the intermediate value of the compression function (the first out of two
-// stages), we multiply the k-th bit of x_i by w^[(2i + 1) * k]. {x_i} is the input to the
-// compression function, i is between 0 and 31, x_i is a 64-bit value.
-// One can see the formula for this (intermediate) stage in the SWIFFT FSE 2008 paper --
-// formula (2), section 3, page 6.
-swift_int16_t fftTable[256 * EIGHTH_N];
-
-// The A's we use in SWIFFTX shall be random elements of Z_257.
-// We generated these A's from the decimal expansion of PI as follows:  we converted each
-// triple of digits into a decimal number d. If d < (257 * 3) we used (d % 257) for the next A
-// element, otherwise move to the next triple of digits in the expansion. This guarntees that
-// the A's are random, provided that PI digits are.
-const swift_int16_t As[3 * M * N] =
-{141,  78, 139,  75, 238, 205, 129, 126,  22, 245, 197, 169, 142, 118, 105,  78,
-  50, 149,  29, 208, 114,  34,  85, 117,  67, 148,  86, 256,  25,  49, 133,  93,
-  95,  36,  68, 231, 211, 102, 151, 128, 224, 117, 193,  27, 102, 187,   7, 105,
-  45, 130, 108, 124, 171, 151, 189, 128, 218, 134, 233, 165,  14, 201, 145, 134,
-  52, 203,  91,  96, 197,  69, 134, 213, 136,  93,   3, 249, 141,  16, 210,  73,
-   6,  92,  58,  74, 174,   6, 254,  91, 201, 107, 110,  76, 103,  11,  73,  16,
-  34, 209,   7, 127, 146, 254,  95, 176,  57,  13, 108, 245,  77,  92, 186, 117,
- 124,  97, 105, 118,  34,  74, 205, 122, 235,  53,  94, 238, 210, 227, 183,  11,
- 129, 159, 105, 183, 142, 129,  86,  21, 137, 138, 224, 223, 190, 188, 179, 188,
- 256,  25, 217, 176,  36, 176, 238, 127, 160, 210, 155, 148, 132,   0,  54, 127,
- 145,   6,  46,  85, 243,  95, 173, 123, 178, 207, 211, 183, 224, 173, 146,  35,
-  71, 114,  50,  22, 175,   1,  28,  19, 112, 129,  21,  34, 161, 159, 115,  52,
-   4, 193, 211,  92, 115,  49,  59, 217, 218,  96,  61,  81,  24, 202, 198,  89,
-  45, 128,   8,  51, 253,  87, 171,  35,   4, 188, 171,  10,   3, 137, 238,  73,
-  19, 208, 124, 163, 103, 177, 155, 147,  46,  84, 253, 233, 171, 241, 211, 217,
- 159,  48,  96,  79, 237,  18, 171, 226,  99,   1,  97, 195, 216, 163, 198,  95,
-   0, 201,  65, 228,  21, 153, 124, 230,  44,  35,  44, 108,  85, 156, 249, 207,
-  26, 222, 131,   1,  60, 242, 197, 150, 181,  19, 116, 213,  75,  98, 124, 240,
- 123, 207,  62, 255,  60, 143, 187, 157, 139,   9,  12, 104,  89,  49, 193, 146,
- 104, 196, 181,  82, 198, 253, 192, 191, 255, 122, 212, 104,  47,  20, 132, 208,
-  46, 170,   2,  69, 234,  36,  56, 163,  28, 152, 104, 238, 162,  56,  24,  58,
-  38, 150, 193, 254, 253, 125, 173,  35,  73, 126, 247, 239, 216,   6, 199,  15,
-  90,  12,  97, 122,   9,  84, 207, 127, 219,  72,  58,  30,  29, 182,  41, 192,
- 235, 248, 237,  74,  72, 176, 210, 252,  45,  64, 165,  87, 202, 241, 236, 223,
- 151, 242, 119, 239,  52, 112, 169,  28,  13,  37, 160,  60, 158,  81, 133,  60,
-  16, 145, 249, 192, 173, 217, 214,  93, 141, 184,  54,  34, 161, 104, 157,  95,
-  38, 133, 218, 227, 211, 181,   9,  66, 137, 143,  77,  33, 248, 159,   4,  55,
- 228,  48,  99, 219, 222, 184,  15,  36, 254, 256, 157, 237,  87, 139, 209, 113,
- 232,  85, 126, 167, 197, 100, 103, 166,  64, 225, 125, 205, 117, 135,  84, 128,
- 231, 112,  90, 241,  28,  22, 210, 147, 186,  49, 230,  21, 108,  39, 194,  47,
- 123, 199, 107, 114,  30, 210, 250, 143,  59, 156, 131, 133, 221,  27,  76,  99,
- 208, 250,  78,  12, 211, 141,  95,  81, 195, 106,   8, 232, 150, 212, 205, 221,
-  11, 225,  87, 219, 126, 136, 137, 180, 198,  48,  68, 203, 239, 252, 194, 235,
- 142, 137, 174, 172, 190, 145, 250, 221, 182, 204,   1, 195, 130, 153,  83, 241,
- 161, 239, 211, 138,  11, 169, 155, 245, 174,  49,  10, 166,  16, 130, 181, 139,
- 222, 222, 112,  99, 124,  94,  51, 243, 133, 194, 244, 136,  35, 248, 201, 177,
- 178, 186, 129, 102,  89, 184, 180,  41, 149,  96, 165,  72, 225, 231, 134, 158,
- 199,  28, 249,  16, 225, 195,  10, 210, 164, 252, 138,   8,  35, 152, 213, 199,
-  82, 116,  97, 230,  63, 199, 241,  35,  79, 120,  54, 174,  67, 112,   1,  76,
-  69, 222, 194,  96,  82,  94,  25, 228, 196, 145, 155, 136, 228, 234,  46, 101,
- 246,  51, 103, 166, 246,  75,   9, 200, 161,   4, 108,  35, 129, 168, 208, 144,
-  50,  14,  13, 220,  41, 132, 122, 127, 194,   9, 232, 234, 107,  28, 187,   8,
-  51, 141,  97, 221, 225,   9, 113, 170, 166, 102, 135,  22, 231, 185, 227, 187,
- 110, 145, 251, 146,  76,  22, 146, 228,   7,  53,  64,  25,  62, 198, 130, 190,
- 221, 232, 169,  64, 188, 199, 237, 249, 173, 218, 196, 191,  48, 224,   5, 113,
- 100, 166, 160,  21, 191, 197,  61, 162, 149, 171, 240, 183, 129, 231, 123, 204,
- 192, 179, 134,  15,  47, 161, 142, 177, 239, 234, 186, 237, 231,  53, 208,  95,
- 146,  36, 225, 231,  89, 142,  93, 248, 137, 124,  83,  39,  69,  77,  89, 208,
- 182,  48,  85, 147, 244, 164, 246,  68,  38, 190, 220,  35, 202,  91, 157, 151,
- 201, 240, 185, 218,   4, 152,   2, 132, 177,  88, 190, 196, 229,  74, 220, 135,
- 137, 196,  11,  47,   5, 251, 106, 144, 163,  60, 222, 127,  52,  57, 202, 102,
-  64, 140, 110, 206,  23, 182,  39, 245,   1, 163, 157, 186, 163,  80,   7, 230,
-  44, 249, 176, 102, 164, 125, 147, 120,  18, 191, 186, 125,  64,  65, 198, 157,
- 164, 213,  95,  61,  13, 181, 208,  91, 242, 197, 158,  34,  98, 169,  91,  14,
-  17,  93, 157,  17,  65,  30, 183,   6, 139,  58, 255, 108, 100, 136, 209, 144,
- 164,   6, 237,  33, 210, 110,  57, 126, 197, 136, 125, 244, 165, 151, 168,   3,
- 143, 251, 247, 155, 136, 130,  88,  14,  74, 121, 250, 133,  21, 226, 185, 232,
- 118, 132,  89,  64, 204, 161,   2,  70, 224, 159,  35, 204, 123, 180,  13,  52,
- 231,  57,  25,  78,  66,  69,  97,  42, 198,  84, 176,  59,   8, 232, 125, 134,
- 193,   2, 232, 109, 216,  69,  90, 142,  32,  38, 249,  37,  75, 180, 184, 188,
-  19,  47, 120,  87, 146,  70, 232, 120, 191,  45,  33,  38,  19, 248, 110, 110,
-  44,  64,   2,  84, 244, 228, 252, 228, 170, 123,  38, 144, 213, 144, 171, 212,
- 243,  87, 189,  46, 128, 110,  84,  77,  65, 183,  61, 184, 101,  44, 168,  68,
-  14, 106, 105,   8, 227, 211, 166,  39, 152,  43,  52, 254, 197,  55, 119,  89,
- 168,  65,  53, 138, 177,  56, 219,   0,  58, 121, 148,  18,  44, 100, 215, 103,
- 145, 229, 117, 196,  91,  89, 113, 143, 172, 239, 249, 184, 154,  39, 112,  65,
- 204,  42,  84,  38, 155, 151, 151,  16, 100,  87, 174, 162, 145, 147, 149, 186,
- 237, 145, 134, 144, 198, 235, 213, 163,  48, 230,  24,  47,  57,  71, 127,   0,
- 150, 219,  12,  81, 197, 150, 131,  13, 169,  63, 175, 184,  48, 235,  65, 243,
- 149, 200, 163, 254, 202, 114, 247,  67, 143, 250, 126, 228,  80, 130, 216, 214,
-  36,   2, 230,  33, 119, 125,   3, 142, 237, 100,   3, 152, 197, 174, 244, 129,
- 232,  30, 206, 199,  39, 210, 220,  43, 237, 221, 201,  54, 179,  42,  28, 133,
- 246, 203, 198, 177,   0,  28, 194,  85, 223, 109, 155, 147, 221,  60, 133, 108,
- 157, 254,  26,  75, 157, 185,  49, 142,  31, 137,  71,  43,  63,  64, 237, 148,
- 237, 172, 159, 160, 155, 254, 234, 224, 140, 193, 114, 140,  62, 109, 136,  39,
- 255,   8, 158, 146, 128,  49, 222,  96,  57, 209, 180, 249, 202, 127, 113, 231,
-  78, 178,  46,  33, 228, 215, 104,  31, 207, 186,  82,  41,  42,  39, 103, 119,
- 123, 133, 243, 254, 238, 156,  90, 186,  37, 212,  33, 107, 252,  51, 177,  36,
- 237,  76, 159, 245,  93, 214,  97,  56, 190,  38, 160,  94, 105, 222, 220, 158,
-  49,  16, 191,  52, 120,  87, 179,   2,  27, 144, 223, 230, 184,   6, 129, 227,
-  69,  47, 215, 181, 162, 139,  72, 200,  45, 163, 159,  62,   2, 221, 124,  40,
- 159, 242,  35, 208, 179, 166,  98,  67, 178,  68, 143, 225, 178, 146, 187, 159,
-  57,  66, 176, 192, 236, 250, 168, 224, 122,  43, 159, 120, 133, 165, 122,  64,
-  87,  74, 161, 241,   9,  87,  90,  24, 255, 113, 203, 220,  57, 139, 197, 159,
-  31, 151,  27, 140,  77, 162,   7,  27,  84, 228, 187, 220,  53, 126, 162, 242,
-  84, 181, 223, 103,  86, 177, 207,  31, 140,  18, 207, 256, 201, 166,  96,  23,
- 233, 103, 197,  84, 161,  75,  59, 149, 138, 154, 119,  92,  16,  53, 116,  97,
- 220, 114,  35,  45,  77, 209,  40, 196,  71,  22,  81, 178, 110,  14,   3, 180,
- 110, 129, 112,  47,  18,  61, 134,  78,  73,  79, 254, 232, 125, 180, 205,  54,
- 220, 119,  63,  89, 181,  52,  77, 109, 151,  77,  80, 207, 144,  25,  20,   6,
- 208,  47, 201, 206, 192,  14,  73, 176, 256, 201, 207,  87, 216,  60,  56,  73,
-  92, 243, 179, 113,  49,  59,  55, 168, 121, 137,  69, 154,  95,  57, 187,  47,
- 129,   4,  15,  92,   6, 116,  69, 196,  48, 134,  84,  81, 111,  56,  38, 176,
- 239,   6, 128,  72, 242, 134,  36, 221,  59,  48, 242,  68, 130, 110, 171,  89,
-  13, 220,  48,  29,   5,  75, 104, 233,  91, 129, 105, 162,  44, 113, 163, 163,
-  85, 147, 190, 111, 197,  80, 213, 153,  81,  68, 203,  33, 161, 165,  10,  61,
- 120, 252,   0, 205,  28,  42, 193,  64,  39,  37,  83, 175,   5, 218, 215, 174,
- 128, 121, 231,  11, 150, 145, 135, 197, 136,  91, 193,   5, 107,  88,  82,   6,
-   4, 188, 256,  70,  40,   2, 167,  57, 169, 203, 115, 254, 215, 172,  84,  80,
- 188, 167,  34, 137,  43, 243,   2,  79, 178,  38, 188, 135, 233, 194, 208,  13,
-  11, 151, 231, 196,  12, 122, 162,  56,  17, 114, 191, 207,  90, 132,  64, 238,
- 187,   6, 198, 176, 240,  88, 118, 236,  15, 226, 166,  22, 193, 229,  82, 246,
- 213,  64,  37,  63,  31, 243, 252,  37, 156,  38, 175, 204, 138, 141, 211,  82,
- 106, 217,  97, 139, 153,  56, 129, 218, 158,   9,  83,  26,  87, 112,  71,  21,
- 250,   5,  65, 141,  68, 116, 231, 113,  10, 218,  99, 205, 201,  92, 157,   4,
-  97,  46,  49, 220,  72, 139, 103, 171, 149, 129, 193,  19,  69, 245,  43,  31,
-  58,  68,  36, 195, 159,  22,  54,  34, 233, 141, 205, 100, 226,  96,  22, 192,
-  41, 231,  24,  79, 234, 138,  30, 120, 117, 216, 172, 197, 172, 107,  86,  29,
- 181, 151,   0,   6, 146, 186,  68,  55,  54,  58, 213, 182,  60, 231,  33, 232,
-  77, 210, 216, 154,  80,  51, 141, 122,  68, 148, 219, 122, 254,  48,  64, 175,
-  41, 115,  62, 243, 141,  81, 119, 121,   5,  68, 121,  88, 239,  29, 230,  90,
- 135, 159,  35, 223, 168, 112,  49,  37, 146,  60, 126, 134,  42, 145, 115,  90,
-  73, 133, 211,  86, 120, 141, 122, 241, 127,  56, 130,  36, 174,  75,  83, 246,
- 112,  45, 136, 194, 201, 115,   1, 156, 114, 167, 208,  12, 176, 147,  32, 170,
- 251, 100, 102, 220, 122, 210,   6,  49,  75, 201,  38, 105, 132, 135, 126, 102,
-  13, 121,  76, 228, 202,  20,  61, 213, 246,  13, 207,  42, 148, 168,  37, 253,
-  34,  94, 141, 185,  18, 234, 157, 109, 104,  64, 250, 125,  49, 236,  86,  48,
- 196,  77,  75, 237, 156, 103, 225,  19, 110, 229,  22,  68, 177,  93, 221, 181,
- 152, 153,  61, 108, 101,  74, 247, 195, 127, 216,  30, 166, 168,  61,  83, 229,
- 120, 156,  96, 120, 201, 124,  43,  27, 253, 250, 120, 143,  89, 235, 189, 243,
- 150,   7, 127, 119, 149, 244,  84, 185, 134,  34, 128, 193, 236, 234, 132, 117,
- 137,  32, 145, 184,  44, 121,  51,  76,  11, 228, 142, 251,  39,  77, 228, 251,
-  41,  58, 246, 107, 125, 187,   9, 240,  35,   8,  11, 162, 242, 220, 158, 163,
-   2, 184, 163, 227, 242,   2, 100, 101,   2,  78, 129,  34,  89,  28,  26, 157,
-  79,  31, 107, 250, 194, 156, 186,  69, 212,  66,  41, 180, 139,  42, 211, 253,
- 256, 239,  29, 129, 104, 248, 182,  68,   1, 189,  48, 226,  36, 229,   3, 158,
-  41,  53, 241,  22, 115, 174,  16, 163, 224,  19, 112, 219, 177, 233,  42,  27,
- 250, 134,  18,  28, 145, 122,  68,  34, 134,  31, 147,  17,  39, 188, 150,  76,
-  45,  42, 167, 249,  12,  16,  23, 182,  13,  79, 121,   3,  70, 197, 239,  44,
-  86, 177, 255,  81,  64, 171, 138, 131,  73, 110,  44, 201, 254, 198, 146,  91,
-  48,   9, 104,  31,  29, 161, 101,  31, 138, 180, 231, 233,  79, 137,  61, 236,
- 140,  15, 249, 218, 234, 119,  99, 195, 110, 137, 237, 207,   8,  31,  45,  24,
-  90, 155, 203, 253, 192, 203,  65, 176, 210, 171, 142, 214, 220, 122, 136, 237,
- 189, 186, 147,  40,  80, 254, 173,  33, 191,  46, 192,  26, 108, 255, 228, 205,
-  61,  76,  39, 107, 225, 126, 228, 182, 140, 251, 143, 134, 252, 168, 221,   8,
- 185,  85,  60, 233, 147, 244,  87, 137,   8, 140,  96,  80,  53,  45, 175, 160,
- 124, 189, 112,  37, 144,  19,  70,  17, 170, 242,   2,   3,  28,  95, 120, 199,
- 212,  43,   9, 117,  86, 151, 101, 241, 200, 145, 241,  19, 178,  69, 204, 197,
- 227, 166,  94,   7, 193,  45, 247, 234,  19, 187, 212, 212, 236, 125,  33,  95,
- 198, 121, 122, 103,  77, 155, 235,  49,  25, 237, 249,  11, 162,   7, 238,  24,
-  16, 150, 129,  25, 152,  17,  42,  67, 247, 162,  77, 154,  31, 133,  55, 137,
-  79, 119, 153,  10,  86,  28, 244, 186,  41, 169, 106,  44,  10,  49, 110, 179,
-  32, 133, 155, 244,  61,  70, 131, 168, 170,  39, 231, 252,  32,  69,  92, 238,
- 239,  35, 132, 136, 236, 167,  90,  32, 123,  88,  69,  22,  20,  89, 145, 166,
-  30, 118,  75,   4,  49,  31, 225,  54,  11,  50,  56, 191, 246,   1, 187,  33,
- 119, 107, 139,  68,  19, 240, 131,  55,  94, 113,  31, 252,  12, 179, 121,   2,
- 120, 252,   0,  76,  41,  80, 185,  42,  62, 121, 105, 159, 121, 109, 111,  98,
-   7, 118,  86,  29, 210,  70, 231, 179, 223, 229, 164,  70,  62,  47,   0, 206,
- 204, 178, 168, 120, 224, 166,  99,  25, 103,  63, 246, 224, 117, 204,  75, 124,
- 140, 133, 110, 110, 222,  88, 151, 118,  46,  37,  22, 143, 158,  40,   2,  50,
- 153,  94, 190, 199,  13, 198, 127, 211, 180,  90, 183,  98,   0, 142, 210, 154,
- 100, 187,  67, 231, 202, 100, 198, 235, 252, 160, 247, 124, 247,  14, 121, 221,
-  57,  88, 253, 243, 185,  89,  45, 249, 221, 194, 108, 175, 193, 119,  50, 141,
- 223, 133, 136,  64, 176, 250, 129, 100, 124,  94, 181, 159,  99, 185, 177, 240,
- 135,  42, 103,  52, 202, 208, 143, 186, 193, 103, 154, 237, 102,  88, 225, 161,
-  50, 188, 191, 109,  12,  87,  19, 227, 247, 183,  13,  52, 205, 170, 205, 146,
-  89, 160,  18, 105, 192,  73, 231, 225, 184, 157, 252, 220,  61,  59, 169, 183,
- 221,  20, 141,  20, 158, 101, 245,   7, 245, 225, 118, 137,  84,  55,  19,  27,
- 164, 110,  35,  25, 202,  94, 150,  46,  91, 152, 130,   1,   7,  46,  16, 237,
- 171, 109,  19, 200,  65,  38,  10, 213,  70,  96, 126, 226, 185, 225, 181,  46,
-  10, 165,  11, 123,  53, 158,  22, 147,  64,  22, 227,  69, 182, 237, 197,  37,
-  39,  49, 186, 223, 139, 128,  55,  36, 166, 178, 220,  20,  98, 172, 166, 253,
-  45,   0, 120, 180, 189, 185, 158, 159, 196,   6, 214,  79, 141,  52, 156, 107,
-   5, 109, 142, 159,  33,  64, 190, 133,  95, 132,  95, 202, 160,  63, 186,  23,
- 231, 107, 163,  33, 234,  15, 244,  77, 108,  49,  51,   7, 164,  87, 142,  99,
- 240, 202,  47, 256, 118, 190, 196, 178, 217,  42,  39, 153,  21, 192, 232, 202,
-  14,  82, 179,  64, 233,   4, 219,  10, 133,  78,  43, 144, 146, 216, 202,  81,
-  71, 252,   8, 201,  68, 256,  85, 233, 164,  88, 176,  30,   5, 152, 126, 179,
- 249,  84, 140, 190, 159,  54, 118,  98,   2, 159,  27, 133,  74, 121, 239, 196,
-  71, 149, 119, 135, 102,  20,  87, 112,  44,  75, 221,   3, 151, 158,   5,  98,
- 152,  25,  97, 106,  63, 171, 240,  79, 234, 240, 230,  92,  76,  70, 173, 196,
-  36, 225, 218, 133,  64, 240, 150,  41, 146,  66, 133,  51, 134,  73, 170, 238,
- 140,  90,  45,  89,  46, 147,  96, 169, 174, 174, 244, 151,  90,  40,  32,  74,
-  38, 154, 246,  57,  31,  14, 189, 151,  83, 243, 197, 183, 220, 185,  53, 225,
-  51, 106, 188, 208, 222, 248,  93,  13,  93, 215, 131,  25, 142, 185, 113, 222,
- 131, 215, 149,  50, 159,  85,  32,   5, 205, 192,   2, 227,  42, 214, 197,  42,
- 126, 182,  68, 123, 109,  36, 237, 179, 170, 199,  77, 256,   5, 128, 214, 243,
- 137, 177, 170, 253, 179, 180, 153, 236, 100, 196, 216, 231, 198,  37, 192,  80,
- 121, 221, 246,   1,  16, 246,  29,  78,  64, 148, 124,  38,  96, 125,  28,  20,
-  48,  51,  73, 187, 139, 208,  98, 253, 221, 188,  84, 129,   1, 205,  95, 205,
- 117,  79,  71, 126, 134, 237,  19, 184, 137, 125, 129, 178, 223,  54, 188, 112,
-  30,   7, 225, 228, 205, 184, 233,  87, 117,  22,  58,  10,   8,  42,   2, 114,
- 254,  19,  17,  13, 150,  92, 233, 179,  63,  12,  60, 171, 127,  35,  50,   5,
- 195, 113, 241,  25, 249, 184, 166,  44, 221,  35, 151, 116,   8,  54, 195,  89,
- 218, 186, 132,   5,  41,  89, 226, 177,  11,  41,  87, 172,   5,  23,  20,  59,
- 228,  94,  76,  33, 137,  43, 151, 221,  61, 232,   4, 120,  93, 217,  80, 228,
- 228,   6,  58,  25,  62,  84,  91,  48, 209,  20, 247, 243,  55, 106,  80,  79,
- 235,  34,  20, 180, 146,   2, 236,  13, 236, 206, 243, 222, 204,  83, 148, 213,
- 214, 117, 237,  98,   0,  90, 204, 168,  32,  41, 126,  67, 191,  74,  27, 255,
-  26,  75, 240, 113, 185, 105, 167, 154, 112,  67, 151,  63, 161, 134, 239, 176,
-  42,  87, 249, 130,  45, 242,  17, 100, 107, 120, 212, 218, 237,  76, 231, 162,
- 175, 172, 118, 155,  92,  36, 124,  17, 121,  71,  13,   9,  82, 126, 147, 142,
- 218, 148, 138,  80, 163, 106, 164, 123, 140, 129,  35,  42, 186, 154, 228, 214,
-  75,  73,   8, 253,  42, 153, 232, 164,  95,  24, 110,  90, 231, 197,  90, 196,
-  57, 164, 252, 181,  31,   7,  97, 256,  35,  77, 200, 212,  99, 179,  92, 227,
-  17, 180,  49, 176,   9, 188,  13, 182,  93,  44, 128, 219, 134,  92, 151,   6,
-  23, 126, 200, 109,  66,  30, 140, 180, 146, 134,  67, 200,   7,   9, 223, 168,
- 186, 221,   3, 154, 150, 165,  43,  53, 138,  27,  86, 213, 235, 160,  70,   2,
- 240,  20,  89, 212,  84, 141, 168, 246, 183, 227,  30, 167, 138, 185, 253,  83,
-  52, 143, 236,  94,  59,  65,  89, 218, 194, 157, 164, 156, 111,  95, 202, 168,
- 245, 256, 151,  28, 222, 194,  72, 130, 217, 134, 253,  77, 246, 100,  76,  32,
- 254, 174, 182, 193,  14, 237,  74,   1,  74,  26, 135, 216, 152, 208, 112,  38,
- 181,  62,  25,  71,  61, 234, 254,  97, 191,  23,  92, 256, 190, 205,   6,  16,
- 134, 147, 210, 219, 148,  59,  73, 185,  24, 247, 174, 143, 116, 220, 128, 144,
- 111, 126, 101,  98, 130, 136, 101, 102,  69, 127,  24, 168, 146, 226, 226, 207,
- 176, 122, 149, 254, 134, 196,  22, 151, 197,  21,  50, 205, 116, 154,  65, 116,
- 177, 224, 127,  77, 177, 159, 225,  69, 176,  54, 100, 104, 140,   8,  11, 126,
-  11, 188, 185, 159, 107,  16, 254, 142,  80,  28,   5, 157, 104,  57, 109,  82,
- 102,  80, 173, 242, 238, 207,  57, 105, 237, 160,  59, 189, 189, 199,  26,  11,
- 190, 156,  97, 118,  20,  12, 254, 189, 165, 147, 142, 199,   5, 213,  64, 133,
- 108, 217, 133,  60,  94,  28, 116, 136,  47, 165, 125,  42, 183, 143,  14, 129,
- 223,  70, 212, 205, 181, 180,   3, 201, 182,  46,  57, 104, 239,  60,  99, 181,
- 220, 231,  45,  79, 156,  89, 149, 143, 190, 103, 153,  61, 235,  73, 136,  20,
-  89, 243,  16, 130, 247, 141, 134,  93,  80,  68,  85,  84,   8,  72, 194,   4,
- 242, 110,  19, 133, 199,  70, 172,  92, 132, 254,  67,  74,  36,  94,  13,  90,
- 154, 184,   9, 109, 118, 243, 214,  71,  36,  95,   0,  90, 201, 105, 112, 215,
-  69, 196, 224, 210, 236, 242, 155, 211,  37, 134,  69, 113, 157,  97,  68,  26,
- 230, 149, 219, 180,  20,  76, 172, 145, 154,  40, 129,   8,  93,  56, 162, 124,
- 207, 233, 105,  19,   3, 183, 155, 134,   8, 244, 213,  78, 139,  88, 156,  37,
-  51, 152, 111, 102, 112, 250, 114, 252, 201, 241, 133,  24, 136, 153,   5,  90,
- 210, 197, 216,  24, 131,  17, 147, 246,  13,  86,   3, 253, 179, 237, 101, 114,
- 243, 191, 207,   2, 220, 133, 244,  53,  87, 125, 154, 158, 197,  20,   8,  83,
-  32, 191,  38, 241, 204,  22, 168,  59, 217, 123, 162,  82,  21,  50, 130,  89,
- 239, 253, 195,  56, 253,  74, 147, 125, 234, 199, 250,  28,  65, 193,  22, 237,
- 193,  94,  58, 229, 139, 176,  69,  42, 179, 164, 150, 168, 246, 214,  86, 174,
-  59, 117,  15,  19,  76,  37, 214, 238, 153, 226, 154,  45, 109, 114, 198, 107,
-  45,  70, 238, 196, 142, 252, 244,  71, 123, 136, 134, 188,  99, 132,  25,  42,
- 240,   0, 196,  33,  26, 124, 256, 145,  27, 102, 153,  35,  28, 132, 221, 167,
- 138, 133,  41, 170,  95, 224,  40, 139, 239, 153,   1, 106, 255, 106, 170, 163,
- 127,  44, 155, 232, 194, 119, 232, 117, 239, 143, 108,  41,   3,   9, 180, 256,
- 144, 113, 133, 200,  79,  69, 128, 216,  31,  50, 102, 209, 249, 136, 150, 154,
- 182,  51, 228,  39, 127, 142,  87,  15,  94,  92, 187, 245,  31, 236,  64,  58,
- 114,  11,  17, 166, 189, 152, 218,  34, 123,  39,  58,  37, 153,  91,  63, 121,
-  31,  34,  12, 254, 106,  96, 171,  14, 155, 247, 214,  69,  24,  98,   3, 204,
- 202, 194, 207,  30, 253,  44, 119,  70,  14,  96,  82, 250,  63,   6, 232,  38,
-  89, 144, 102, 191,  82, 254,  20, 222,  96, 162, 110,   6, 159,  58, 200, 226,
-  98, 128,  42,  70,  84, 247, 128, 211, 136,  54, 143, 166,  60, 118,  99, 218,
-  27, 193,  85,  81, 219, 223,  46,  41,  23, 233, 152, 222,  36, 236,  54, 181,
-  56,  50,   4, 207, 129,  92,  78,  88, 197, 251, 131, 105,  31, 172,  38, 131,
-  19, 204, 129,  47, 227, 106, 202, 183,  23,   6,  77, 224, 102, 147,  11, 218,
- 131, 132,  60, 192, 208, 223, 236,  23, 103, 115,  89,  18, 185, 171,  70, 174,
- 139,   0, 100, 160, 221,  11, 228,  60,  12, 122, 114,  12, 157, 235, 148,  57,
-  83,  62, 173, 131, 169, 126,  85,  99,  93, 243,  81,  80,  29, 245, 206,  82,
- 236, 227, 166,  14, 230, 213, 144,  97,  27, 111,  99, 164, 105, 150,  89, 111,
- 252, 118, 140, 232, 120, 183, 137, 213, 232, 157, 224,  33, 134, 118, 186,  80,
- 159,   2, 186, 193,  54, 242,  25, 237, 232, 249, 226, 213,  90, 149,  90, 160,
- 118,  69,  64,  37,  10, 183, 109, 246,  30,  52, 219,  69, 189,  26, 116, 220,
-  50, 244, 243, 243, 139, 137, 232,  98,  38,  45, 256, 143, 171, 101,  73, 238,
- 123,  45, 194, 167, 250, 123,  12,  29, 136, 237, 141,  21,  89,  96, 199,  44,
-   8, 214, 208,  17, 113,  41, 137,  26, 166, 155,  89,  85,  54,  58,  97, 160,
-  50, 239,  58,  71,  21, 157, 139,  12,  37, 198, 182, 131, 149, 134,  16, 204,
- 164, 181, 248, 166,  52, 216, 136, 201,  37, 255, 187, 240,   5, 101, 147, 231,
-  14, 163, 253, 134, 146, 216,   8,  54, 224,  90, 220, 195,  75, 215, 186,  58,
-  71, 204, 124, 105, 239,  53,  16,  85,  69, 163, 195, 223,  33,  38,  69,  88,
-  88, 203,  99,  55, 176,  13, 156, 204, 236,  99, 194, 134,  75, 247, 126, 129,
- 160, 124, 233, 206, 139, 144, 154,  45, 233,  51, 206,  61,  60,  55, 205, 107,
-  84, 108,  96, 188, 203,  31,  89,  20, 115, 144, 137,  90, 237,  78, 231, 185,
- 120, 217,   1, 176, 169,  30, 155, 176, 100, 113,  53,  42, 193, 108,  14, 121,
- 176, 158, 137,  92, 178,  44, 110, 249, 108, 234,  94, 101, 128,  12, 250, 173,
-  72, 202, 232,  66, 139, 152, 189,  18,  32, 197,   9, 238, 246,  55, 119, 183,
- 196, 119, 113, 247, 191, 100, 200, 245,  46,  16, 234, 112, 136, 116, 232,  48,
- 176, 108,  11, 237,  14, 153,  93, 177, 124,  72,  67, 121, 135, 143,  45,  18,
-  97, 251, 184, 172, 136,  55, 213,   8, 103,  12, 221, 212,  13, 160, 116,  91,
- 237, 127, 218, 190, 103, 131,  77,  82,  36, 100,  22, 252,  79,  69,  54,  26,
-  65, 182, 115, 142, 247,  20,  89,  81, 188, 244,  27, 120, 240, 248,  13, 230,
-  67, 133,  32, 201, 129,  87,   9, 245,  66,  88, 166,  34,  46, 184, 119, 218,
- 144, 235, 163,  40, 138, 134, 127, 217,  64, 227, 116,  67,  55, 202, 130,  48,
- 199,  42, 251, 112, 124, 153, 123, 194, 243,  49, 250,  12,  78, 157, 167, 134,
- 210,  73, 156, 102,  21,  88, 216, 123,  45,  11, 208,  18,  47, 187,  20,  43,
-   3, 180, 124,   2, 136, 176,  77, 111, 138, 139,  91, 225, 126,   8,  74, 255,
-  88, 192, 193, 239, 138, 204, 139, 194, 166, 130, 252, 184, 140, 168,  30, 177,
- 121,  98, 131, 124,  69, 171,  75,  49, 184,  34,  76, 122, 202, 115, 184, 253,
- 120, 182,  33, 251,   1,  74, 216, 217, 243, 168,  70, 162, 119, 158, 197, 198,
-  61,  89,   7,   5,  54, 199, 211, 170,  23, 226,  44, 247, 165, 195,   7, 225,
-  91,  23,  50,  15,  51, 208, 106,  94,  12,  31,  43, 112, 146, 139, 246, 182,
- 113,   1,  97,  15,  66,   2,  51,  76, 164, 184, 237, 200, 218, 176,  72,  98,
-  33, 135,  38, 147, 140, 229,  50,  94,  81, 187, 129,  17, 238, 168, 146, 203,
- 181,  99, 164,   3, 104,  98, 255, 189, 114, 142,  86, 102, 229, 102,  80, 129,
-  64,  84,  79, 161,  81, 156, 128, 111, 164, 197,  18,  15,  55, 196, 198, 191,
-  28, 113, 117,  96, 207, 253,  19, 158, 231,  13,  53, 130, 252, 211,  58, 180,
- 212, 142,   7, 219,  38,  81,  62, 109, 167, 113,  33,  56,  97, 185, 157, 130,
- 186, 129, 119, 182, 196,  26,  54, 110,  65, 170, 166, 236,  30,  22, 162,   0,
- 106,  12, 248,  33,  48,  72, 159,  17,  76, 244, 172, 132,  89, 171, 196,  76,
- 254, 166,  76, 218, 226,   3,  52, 220, 238, 181, 179, 144, 225,  23,   3, 166,
- 158,  35, 228, 154, 204,  23, 203,  71, 134, 189,  18, 168, 236, 141, 117, 138,
-   2, 132,  78,  57, 154,  21, 250, 196, 184,  40, 161,  40,  10, 178, 134, 120,
- 132, 123, 101,  82, 205, 121,  55, 140, 231,  56, 231,  71, 206, 246, 198, 150,
- 146, 192,  45, 105, 242,   1, 125,  18, 176,  46, 222, 122,  19,  80, 113, 133,
- 131, 162,  81,  51,  98, 168, 247, 161, 139,  39,  63, 162,  22, 153, 170,  92,
-  91, 130, 174, 200,  45, 112,  99, 164, 132, 184, 191, 186, 200, 167,  86, 145,
- 167, 227, 130,  44,  12, 158, 172, 249, 204,  17,  54, 249,  16, 200,  21, 174,
-  67, 223, 105, 201,  50,  36, 133, 203, 244, 131, 228,  67,  29, 195,  91,  91,
-  55, 107, 167, 154, 170, 137, 218, 183, 169,  61,  99, 175, 128,  23, 142, 183,
-  66, 255,  59, 187,  66,  85, 212, 109, 168,  82,  16,  43,  67, 139, 114, 176,
- 216, 255, 130,  94, 152,  79, 183,  64, 100,  23, 214,  82,  34, 230,  48,  15,
- 242, 130,  50, 241,  81,  32,   5, 125, 183, 182, 184,  99, 248, 109, 159, 210,
- 226,  61, 119, 129,  39, 149,  78, 214, 107,  78, 147, 124, 228,  18, 143, 188,
-  84, 180, 233, 119,  64,  39, 158, 133, 177, 168,   6, 150,  80, 117, 150,  56,
-  49,  72,  49,  37,  30, 242,  49, 142,  33, 156,  34,  44,  44,  72,  58,  22,
- 249,  46, 168,  80,  25, 196,  64, 174,  97, 179, 244, 134, 213, 105,  63, 151,
-  21,  90, 168,  90, 245,  28, 157,  65, 250, 232, 188,  27,  99, 160, 156, 127,
-  68, 193,  10,  80, 205,  36, 138, 229,  12, 223,  70, 169, 251,  41,  48,  94,
-  41, 177,  99, 256, 158,   0,   6,  83, 231, 191, 120, 135, 157, 146, 218, 213,
- 160,   7,  47, 234,  98, 211,  79, 225, 179,  95, 175, 105, 185,  79, 115,   0,
- 104,  14,  65, 124,  15, 188,  52,   9, 253,  27, 132, 137,  13, 127,  75, 238,
- 185, 253,  33,   8,  52, 157, 164,  68, 232, 188,  69,  28, 209, 233,   5, 129,
- 216,  90, 252, 212,  33, 200, 222,   9, 112,  15,  43,  36, 226, 114,  15, 249,
- 217,   8, 148,  22, 147,  23, 143,  67, 222, 116, 235, 250, 212, 210,  39, 142,
- 108,  64, 209,  83,  73,  66,  99,  34,  17,  29,  45, 151, 244, 114,  28, 241,
- 144, 208, 146, 179, 132,  89, 217, 198, 252, 219, 205, 165,  75, 107,  11, 173,
-  76,   6, 196, 247, 152, 216, 248,  91, 209, 178,  57, 250, 174,  60,  79, 123,
-  18, 135,   9, 241, 230, 159, 184,  68, 156, 251, 215,   9, 113, 234,  75, 235,
- 103, 194, 205, 129, 230,  45,  96,  73, 157,  20, 200, 212, 212, 228, 161,   7,
- 231, 228, 108,  43, 198,  87, 140, 140,   4, 182, 164,   3,  53, 104, 250, 213,
-  85,  38,  89,  61,  52, 187,  35, 204,  86, 249, 100,  71, 248, 213, 163, 215,
-  66, 106, 252, 129,  40, 111,  47,  24, 186, 221,  85, 205, 199, 237, 122, 181,
-  32,  46, 182, 135,  33, 251, 142,  34, 208, 242, 128, 255,   4, 234,  15,  33,
- 167, 222,  32, 186, 191,  34, 255, 244,  98, 240, 228, 204,  30, 142,  32,  70,
-  69,  83, 110, 151,  10, 243, 141,  21, 223,  69,  61,  37,  59, 209, 102, 114,
- 223,  33, 129, 254, 255, 103,  86, 247, 235,  72, 126, 177, 102, 226, 102,  30,
- 149, 221,  62, 247, 251, 120, 163, 173,  57, 202, 204,  24,  39, 106, 120, 143,
- 202, 176, 191, 147,  37,  38,  51, 133,  47, 245, 157, 132, 154,  71, 183, 111,
-  30, 180,  18, 202,  82,  96, 170,  91, 157, 181, 212, 140, 256,   8, 196, 121,
- 149,  79,  66, 127, 113,  78,   4, 197,  84, 256, 111, 222, 102,  63, 228, 104,
- 136, 223,  67, 193,  93, 154, 249,  83, 204, 101, 200, 234,  84, 252, 230, 195,
-  43, 140, 120, 242,  89,  63, 166, 233, 209,  94,  43, 170, 126,   5, 205,  78,
- 112,  80, 143, 151, 146, 248, 137, 203,  45, 183,  61,   1, 155,   8, 102,  59,
-  68, 212, 230,  61, 254, 191, 128, 223, 176, 123, 229,  27, 146, 120,  96, 165,
- 213,  12, 232,  40, 186, 225,  66, 105, 200, 195, 212, 110, 237, 238, 151,  19,
-  12, 171, 150,  82,   7, 228,  79,  52,  15,  78,  62,  43,  21, 154, 114,  21,
-  12, 212, 256, 232, 125, 127,   5,  51,  37, 252, 136,  13,  47, 195, 168, 191,
- 231,  55,  57, 251, 214, 116,  15,  86, 210,  41, 249, 242, 119,  27, 250, 203,
- 107,  69,  90,  43, 206, 154, 127,  54, 100,  78, 187,  54, 244, 177, 234, 167,
- 202, 136, 209, 171,  69, 114, 133, 173,  26, 139,  78, 141, 128,  32, 124,  39,
-  45, 218,  96,  68,  90,  44,  67,  62,  83, 190, 188, 256, 103,  42, 102,  64,
- 249,   0, 141,  11,  61,  69,  70,  66, 233, 237,  29, 200, 251, 157,  71,  51,
-  64, 133, 113,  76,  35, 125,  76, 137, 217, 145,  35,  69, 226, 180,  56, 249,
- 156, 163, 176, 237,  81,  54,  85, 169, 115, 211, 129,  70, 248,  40, 252, 192,
- 194, 101, 247,   8, 181, 124, 217, 191, 194,  93,  99, 127, 117, 177, 144, 151,
- 228, 121,  32,  11,  89,  81,  26,  29, 183,  76, 249, 132, 179,  70,  34, 102,
-  20,  66,  87,  63, 124, 205, 174, 177,  87, 219,  73, 218,  91,  87, 176,  72,
-  15, 211,  47,  61, 251, 165,  39, 247, 146,  70, 150,  57,   1, 212,  36, 162,
-  39,  38,  16, 216,   3,  50, 116, 200,  32, 234,  77, 181, 155,  19,  90, 188,
-  36,   6, 254,  46,  46, 203,  25, 230, 181, 196,   4, 151, 225,  65, 122, 216,
- 168,  86, 158, 131, 136,  16,  49, 102, 233,  64, 154,  88, 228,  52, 146,  69,
-  93, 157, 243, 121,  70, 209, 126, 213,  88, 145, 236,  65,  70,  96, 204,  47,
-  10, 200,  77,   8, 103, 150,  48, 153,   5,  37,  52, 235, 209,  31, 181, 126,
-  83, 142, 224, 140,   6,  32, 200, 171, 160, 179, 115, 229,  75, 194, 208,  39,
-  59, 223,  52, 247,  38, 197, 135,   1,   6, 189, 106, 114, 168,   5, 211, 222,
-  44,  63,  90, 160, 116, 172, 170, 133, 125, 138,  39, 131,  23, 178,  10, 214,
-  36,  93,  28,  59,  68,  17, 123,  25, 255, 184, 204, 102, 194, 214, 129,  94,
- 159, 245, 112, 141,  62,  11,  61, 197, 124, 221, 205,  11,  79,  71, 201,  54,
-  58, 150,  29, 121,  87,  46, 240, 201,  68,  20, 194, 209,  47, 152, 158, 174,
- 193, 164, 120, 255, 216, 165, 247,  58,  85, 130, 220,  23, 122, 223, 188,  98,
-  21,  70,  72, 170, 150, 237,  76, 143, 112, 238, 206, 146, 215, 110,   4, 250,
-  68,  44, 174, 177,  30,  98, 143, 241, 180, 127, 113,  48,   0,   1, 179, 199,
-  59, 106, 201, 114,  29,  86, 173, 133, 217,  44, 200, 141, 107, 172,  16,  60,
-  82,  58, 239,  94, 141, 234, 186, 235, 109, 173, 249, 139, 141,  59, 100, 248,
-  84, 144,  49, 160,  51, 207, 164, 103,  74,  97, 146, 202, 193, 125, 168, 134,
- 236, 111, 135, 121,  59, 145, 168, 200, 181, 173, 109,   2, 255,   6,   9, 245,
-  90, 202, 214, 143, 121,  65,  85, 232, 132,  77, 228,  84,  26,  54, 184,  15,
- 161,  29, 177,  79,  43,   0, 156, 184, 163, 165,  62,  90, 179,  93,  45, 239,
-   1,  16, 120, 189, 127,  47,  74, 166,  20, 214, 233, 226,  89, 217, 229,  26,
- 156,  53, 162,  60,  21,   3, 192,  72, 111,  51,  53, 101, 181, 208,  88,  82,
- 179, 160, 219, 113, 240, 108,  43, 224, 162, 147,  62,  14,  95,  81, 205,   4,
- 160, 177, 225, 115,  29,  69, 235, 168, 148,  29, 128, 114, 124, 129, 172, 165,
- 215, 231, 214,  86, 160,  44, 157,  91, 248, 183,  73, 164,  56, 181, 162,  92,
- 141, 118, 127, 240, 196,  77,   0,   9, 244,  79, 250, 100, 195,  25, 255,  85,
-  94,  35, 212, 137, 107,  34, 110,  20, 200, 104,  17,  32, 231,  43, 150, 159,
- 231, 216, 223, 190, 226, 109, 162, 197,  87,  92, 224,  11, 111,  73,  60, 225,
- 238,  73, 246, 169,  19, 217, 119,  38, 121, 118,  70,  82,  99, 241, 110,  67,
-  31,  76, 146, 215, 124, 240,  31, 103, 139, 224,  75, 160,  31,  78,  93,   4,
-  64,   9, 103, 223,   6, 227, 119,  85, 116,  81,  21,  43,  46, 206, 234, 132,
-  85,  99,  22, 131, 135,  97,  86,  13, 234, 188,  21,  14,  89, 169, 207, 238,
- 219, 177, 190,  72, 157,  41, 114, 140,  92, 141, 186,   1,  63, 107, 225, 184,
- 118, 150, 153, 254, 241, 106, 120, 210, 104, 144, 151, 161,  88, 206, 125, 164,
-  15, 211, 173,  49, 146, 241,  71,  36,  58, 201,  46,  27,  33, 187,  91, 162,
- 117,  19, 210, 213, 187,  97, 193,  50, 190, 114, 217,  60,  61, 167, 207, 213,
- 213,  53, 135,  34, 156,  91, 115, 119,  46,  99, 242,   1,  90,  52, 198, 227,
- 201,  91, 216, 146, 210,  82, 121,  38,  73, 133, 182, 193, 132, 148, 246,  75,
- 109, 157, 179, 113, 176, 134, 205, 159, 148,  58, 103, 171, 132, 156, 133, 147,
- 161, 231,  39, 100, 175,  97, 125,  28, 183, 129, 135, 191, 202, 181,  29, 218,
-  43, 104, 148, 203, 189, 204,   4, 182, 169,   1, 134, 122, 141, 202,  13, 187,
- 177, 112, 162,  35, 231,   6,   8, 241,  99,   6, 191,  45, 113, 113, 101, 104};
-
-// The S-Box we use for further linearity breaking.
-// We created it by taking the digits of decimal expansion of e.
-// The code that created it can be found in 'ProduceRandomSBox.c'.
-unsigned char SBox[256] = {
-//0     1    2      3     4    5     6     7      8    9     A      B    C     D     E     F
-0x7d, 0xd1, 0x70, 0x0b, 0xfa, 0x39, 0x18, 0xc3, 0xf3, 0xbb, 0xa7, 0xd4, 0x84, 0x25, 0x3b, 0x3c,   // 0
-0x2c, 0x15, 0x69, 0x9a, 0xf9, 0x27, 0xfb, 0x02, 0x52, 0xba, 0xa8, 0x4b, 0x20, 0xb5, 0x8b, 0x3a,   // 1
-0x88, 0x8e, 0x26, 0xcb, 0x71, 0x5e, 0xaf, 0xad, 0x0c, 0xac, 0xa1, 0x93, 0xc6, 0x78, 0xce, 0xfc,   // 2
-0x2a, 0x76, 0x17, 0x1f, 0x62, 0xc2, 0x2e, 0x99, 0x11, 0x37, 0x65, 0x40, 0xfd, 0xa0, 0x03, 0xc1,   // 3
-0xca, 0x48, 0xe2, 0x9b, 0x81, 0xe4, 0x1c, 0x01, 0xec, 0x68, 0x7a, 0x5a, 0x50, 0xf8, 0x0e, 0xa3,   // 4
-0xe8, 0x61, 0x2b, 0xa2, 0xeb, 0xcf, 0x8c, 0x3d, 0xb4, 0x95, 0x13, 0x08, 0x46, 0xab, 0x91, 0x7b,   // 5
-0xea, 0x55, 0x67, 0x9d, 0xdd, 0x29, 0x6a, 0x8f, 0x9f, 0x22, 0x4e, 0xf2, 0x57, 0xd2, 0xa9, 0xbd,   // 6
-0x38, 0x16, 0x5f, 0x4c, 0xf7, 0x9e, 0x1b, 0x2f, 0x30, 0xc7, 0x41, 0x24, 0x5c, 0xbf, 0x05, 0xf6,   // 7
-0x0a, 0x31, 0xa5, 0x45, 0x21, 0x33, 0x6b, 0x6d, 0x6c, 0x86, 0xe1, 0xa4, 0xe6, 0x92, 0x9c, 0xdf,   // 8
-0xe7, 0xbe, 0x28, 0xe3, 0xfe, 0x06, 0x4d, 0x98, 0x80, 0x04, 0x96, 0x36, 0x3e, 0x14, 0x4a, 0x34,   // 9
-0xd3, 0xd5, 0xdb, 0x44, 0xcd, 0xf5, 0x54, 0xdc, 0x89, 0x09, 0x90, 0x42, 0x87, 0xff, 0x7e, 0x56,   // A
-0x5d, 0x59, 0xd7, 0x23, 0x75, 0x19, 0x97, 0x73, 0x83, 0x64, 0x53, 0xa6, 0x1e, 0xd8, 0xb0, 0x49,   // B
-0x3f, 0xef, 0xbc, 0x7f, 0x43, 0xf0, 0xc9, 0x72, 0x0f, 0x63, 0x79, 0x2d, 0xc0, 0xda, 0x66, 0xc8,   // C
-0x32, 0xde, 0x47, 0x07, 0xb8, 0xe9, 0x1d, 0xc4, 0x85, 0x74, 0x82, 0xcc, 0x60, 0x51, 0x77, 0x0d,   // D
-0xaa, 0x35, 0xed, 0x58, 0x7c, 0x5b, 0xb9, 0x94, 0x6e, 0x8d, 0xb1, 0xc5, 0xb7, 0xee, 0xb6, 0xae,   // E
-0x10, 0xe0, 0xd6, 0xd9, 0xe5, 0x4f, 0xf1, 0x12, 0x00, 0xd0, 0xf4, 0x1a, 0x6f, 0x8a, 0xb3, 0xb2 }; // F
-
-///////////////////////////////////////////////////////////////////////////////////////////////
-//
-//	Helper functions definition portion.
-//
-///////////////////////////////////////////////////////////////////////////////////////////////
-
-// Don't vectorize, move decl to header file
-
-// Translates an input array with values in base 257 to output array with values in base 256.
-// Returns the carry bit.
-//
-// Parameters:
-// - input: the input array of size EIGHTH_N. Each value in the array is a number in Z_257.
-//          The MSB is assumed to be the last one in the array.
-// - output: the input array encoded in base 256.
-//
-// Returns:
-// - The carry bit (MSB).
-swift_int16_t TranslateToBase256(swift_int32_t input[EIGHTH_N], unsigned char output[EIGHTH_N]);
-
-// Translates an input integer into the range (-FIELD_SIZE / 2) <= result <= (FIELD_SIZE / 2).
-//
-// Parameters:
-// - x: the input integer.
-//
-// Returns:
-// - The result, which equals (x MOD FIELD_SIZE), such that |result| <= (FIELD_SIZE / 2).
-int Center(int x);
-
-// Calculates bit reversal permutation.
-//
-// Parameters:
-// - input: the input to reverse.
-// - numOfBits: the number of bits in the input to reverse.
-//
-// Returns:
-// - The resulting number, which is obtained from the input by reversing its bits.
-int ReverseBits(int input, int numOfBits);
-
-// Initializes the FFT fast lookup table.
-// Shall be called only once.
-void InitializeSWIFFTX();
-
-// Calculates the FFT.
-//
-// Parameters:
-// - input: the input to the FFT.
-// - output: the resulting output.
-void FFT(const unsigned char input[EIGHTH_N], swift_int32_t *output);
-
-///////////////////////////////////////////////////////////////////////////////////////////////
-// Helper functions implementation portion.
-///////////////////////////////////////////////////////////////////////////////////////////////
-
-// Don't vectorize, delete this copy.
-
-swift_int16_t TranslateToBase256(swift_int32_t input[EIGHTH_N], unsigned char output[EIGHTH_N])
-{
-	swift_int32_t pairs[EIGHTH_N / 2];
-	int i;
-
-	for (i = 0; i < EIGHTH_N; i += 2)
-	{
-		// input[i] + 257 * input[i + 1]
-		pairs[i >> 1] = input[i] + input[i + 1] + (input[i + 1] << 8);
-	}
-
-	for (i = (EIGHTH_N / 2) - 1; i > 0; --i)
-	{
-		int j;
-
-		for (j = i - 1; j < (EIGHTH_N / 2) - 1; ++j)
-		{
-			// pairs[j + 1] * 513, because 257^2 = 513 % 256^2.
-			register swift_int32_t temp = pairs[j] + pairs[j + 1] + (pairs[j + 1] << 9);
-			pairs[j] = temp & 0xffff;
-			pairs[j + 1] += (temp >> 16);
-		}
-	}
-
-	for (i = 0; i < EIGHTH_N; i += 2)
-	{
-		output[i] = (unsigned char) (pairs[i >> 1] & 0xff);
-		output[i + 1] = (unsigned char) ((pairs[i >> 1] >> 8) & 0xff);
-	}
-
-	return (pairs[EIGHTH_N/2 - 1] >> 16);
-}
-
-int Center(int x)
-{
-	int result = x % FIELD_SIZE;
-
-	if (result > (FIELD_SIZE / 2))
-		result -= FIELD_SIZE;
-
-	if (result < (FIELD_SIZE / -2))
-		result += FIELD_SIZE;
-
-	return result;
-}
-
-int ReverseBits(int input, int numOfBits)
-{
-	register int reversed = 0;
-
-	for (input |= numOfBits; input > 1; input >>= 1)
-		reversed = (reversed << 1) | (input & 1);
-
-	return reversed;
-}
-
-void InitializeSWIFFTX()
-{
-	int i, j, k, x;
-	// The powers of OMEGA
-	int omegaPowers[2 * N];
-	omegaPowers[0] = 1;
-
-	if (wasSetupDone)
-		return;
-
-	for (i = 1; i < (2 * N); ++i)
-	{
-		omegaPowers[i] = Center(omegaPowers[i - 1] * OMEGA);
-	}
-
-	for (i = 0; i < (N / W); ++i)
-	{
-		for (j = 0; j < W; ++j)
-		{
-			multipliers[(i << 3) + j] = omegaPowers[ReverseBits(i, N / W) * (2 * j + 1)];
-		}
-	}
-
-	for (x = 0; x < 256; ++x)
-	{
-		for (j = 0; j < 8; ++j)
-		{
-			register int temp = 0;
-			for (k = 0; k < 8; ++k)
-			{
-				temp += omegaPowers[(EIGHTH_N * (2 * j + 1) * ReverseBits(k, W)) % (2 * N)]
-					  * ((x >> k) & 1);
-			}
-
-			fftTable[(x << 3) + j] = Center(temp);
-		}
-	}
-
-	wasSetupDone = true;
-}
-
-// input should be deinterleaved in contiguos memory
-// output and F are 4x32
-// multipliers & fftTable are scalar 16
-
-
-void FFT_4way(const unsigned char input[EIGHTH_N], swift_int32_t *output)
-{
-	swift_int16_t *mult = multipliers;
-   m128_swift_int32_t F[64];
-
-   for (int i = 0; i < 8; i++)
-   {
-      int j = i<<3;
-
-// Need to isolate bytes in input, 8 bytes per lane.
-// Each iteration of the loop process one input vector
-// Each lane reads a different index to ffttable.
-
-// deinterleave the input!
-
-// load table with 4 lanes from different indexes into fftTable
-// extract bytes into m128 4x16
-// mutiply by vectorized mult
-
-// input[lane][byte]
-
-      __m128i table;
-      table = _mm_set_epi32( fftTable[ input[3][i] ],
-                             fftTable[ input[2][i] ],
-                             fftTable[ input[1][i] ],
-                             fftTable[ input[0][i] ] );
-
-      F[i  ] = _mm_mullo_epi32( mm128_const1_32( mult[j+0] ), table );
-
-      table = _mm_set_epi32( fftTable[ input[3][i+1] ]
-                             fftTable[ input[2][i+1] ]
-                             fftTable[ input[1][i+1] ]
-                             fftTable[ input[0][i+1] ] );
-
-      F[i+8] = _mm_mullo_epi32( mm128_const1_32( mult[j+0] ), table );
-
-
-      m128_swift_int16_t *table = &( fftTable[input[i] << 3] );
-
-      F[i   ] = _mm_mullo_epi32( mm128_const1_32( mult[j+0] ),
-                                 mm128_const1_32( table[0] ) );
-      F[i+ 8] = _mm_mullo_epi32( mm128_const1_32( mult[j+1] ),
-                                 mm128_const1_32( table[1] ) );
-      F[i+16] = _mm_mullo_epi32( mm128_const1_32( mult[j+2] ),
-                                 mm128_const1_32( table[2] ) );
-      F[i+24] = _mm_mullo_epi32( mm128_const1_32( mult[j+3] ),
-                                 mm128_const1_32( table[3] ) );
-      F[i+32] = _mm_mullo_epi32( mm128_const1_32( mult[j+4] ),
-                                 mm128_const1_32( table[4] ) );
-      F[i+40] = _mm_mullo_epi32( mm128_const1_32( mult[j+5] ),
-                                 mm128_const1_32( table[5] ) );
-      F[i+48] = _mm_mullo_epi32( mm128_const1_32( mult[j+6] ),
-                                 mm128_const1_32( table[6] ) );
-      F[i+56] = _mm_mullo_epi32( mm128_const1_32( mult[j+7] ),
-                                 mm128_const1_32( table[7] ) );
-   }
-
-
-   for ( int i = 0; i < 8; i++ )
-   {
-      int j = i<<3;
-      ADD_SUB_4WAY( F[j  ], F[j+1] );
-      ADD_SUB_4WAY( F[j+2], F[j+3] );
-      ADD_SUB_4WAY( F[j+4], F[j+5] );
-      ADD_SUB_4WAY( F[j+6], F[j+7] );
-
-      F[j+3] = _mm_slli_epi32( F[j+3], 4 );
-      F[j+7] = _mm_slli_epi32( F[j+7], 4 );
-
-      ADD_SUB_4WAY( F[j  ], F[j+2] );
-      ADD_SUB_4WAY( F[j+1], F[j+3] );
-      ADD_SUB_4WAY( F[j+4], F[j+6] );
-      ADD_SUB_4WAY( F[j+5], F[j+7] );
-
-      F[j+5] = _mm_slli_epi32( F[j+5], 2 );
-      F[j+6] = _mm_slli_epi32( F[j+6], 4 );
-      F[j+7] = _mm_slli_epi32( F[j+7], 6 );
-
-      ADD_SUB_4WAY( F[j  ], F[j+4] );
-      ADD_SUB_4WAY( F[j+1], F[j+5] );
-      ADD_SUB_4WAY( F[j+2], F[j+6] );
-      ADD_SUB_4WAY( F[j+3], F[j+7] );
-
-      output[i   ] = Q_REDUCE_4WAY( F[j  ] );
-      output[i+ 8] = Q_REDUCE_4WAY( F[j+1] );
-      output[i+16] = Q_REDUCE_4WAY( F[j+2] );
-      output[i+24] = Q_REDUCE_4WAY( F[j+3] );
-      output[i+32] = Q_REDUCE_4WAY( F[j+4] );
-      output[i+40] = Q_REDUCE_4WAY( F[j+5] );
-      output[i+48] = Q_REDUCE_4WAY( F[j+6] );
-      output[i+56] = Q_REDUCE_4WAY( F[j+7] );
-   }
-}
-
-// Calculates the FFT part of SWIFFT.
-// We divided the SWIFFT calculation into two, because that way we could save 2 computations of
-// the FFT part, since in the first stage of SWIFFTX the difference between the first 3 SWIFFTs
-// is only the A's part.
-//
-// Parameters:
-// - input: the input to FFT.
-// - m: the input size divided by 8. The function performs m FFTs.
-// - output: will store the result.
-void SWIFFTFFT(const unsigned char *input, int m, swift_int32_t *output)
-{
-	int i;
-
-	for (i = 0;
-		 i < m;
-		 i++, input += EIGHTH_N, output += N)
-	{
-		FFT(input, output);
-	}
-}
-
-// Calculates the 'sum' part of SWIFFT, including the base change at the end.
-// We divided the SWIFFT calculation into two, because that way we could save 2 computations of
-// the FFT part, since in the first stage of SWIFFTX the difference between the first 3 SWIFFTs
-// is only the A's part.
-//
-// Parameters:
-// - input: the input. Of size 64 * m.
-// - m: the input size divided by 64.
-// - output: will store the result.
-// - a: the coefficients in the sum. Of size 64 * m.
-void SWIFFTSum(const swift_int32_t *input, int m, unsigned char *output, const swift_int16_t *a)
-{
-	int i, j;
-	swift_int32_t result[N];
-	register swift_int16_t carry = 0;
-
-	for (j = 0; j < N; ++j)
-	{
-		register swift_int32_t sum = 0;
-		const register swift_int32_t *f = input + j;
-		const register swift_int16_t *k = a + j;
-
-		for (i = 0; i < m; i++, f += N,k += N)
-		{
-			sum += (*f) * (*k);
-		}
-
-		result[j] = sum;
-	}
-
-	for (j = 0; j < N; ++j)
-	{
-		result[j] = ((FIELD_SIZE << 22) + result[j]) % FIELD_SIZE;
-	}
-
-	for (j = 0; j < 8; ++j)
-	{
-		int register carryBit = TranslateToBase256(result + (j << 3), output + (j << 3));
-		carry |= carryBit << j;
-	}
-
-	output[N] = carry;
-}
-
-
-// On entry input is interleaved 4x64. SIZE is *4 lanes / 8 bytes,
-// multiply by 2.
-
-
-void ComputeSingleSWIFFTX_4way( unsigned char input[SWIFFTX_INPUT_BLOCK_SIZE],
-                          unsigned char output[SWIFFTX_OUTPUT_BLOCK_SIZE],
-						  bool doSmooth)
-{
-	int i;
-	// Will store the result of the FFT parts:
-   m128_swift_int32_t fftOut[N * M];
-//   swift_int32_t fftOut[N * M];
-	unsigned char intermediate[N * 3 + 8];
-	unsigned char carry0,carry1,carry2;
-
-	// Do the three SWIFFTS while remembering the three carry bytes (each carry byte gets
-	// overriden by the following SWIFFT):
-
-	// 1. Compute the FFT of the input - the common part for the first 3 SWIFFTs:
-	SWIFFTFFT(input, M, fftOut);
-
-	// 2. Compute the sums of the 3 SWIFFTs, each using a different set of coefficients:
-
-	// 2a. The first SWIFFT:
-	SWIFFTSum(fftOut, M, intermediate, As);
-	// Remember the carry byte:
-	carry0 = intermediate[N];
-
-	// 2b. The second one:
-	SWIFFTSum(fftOut, M, intermediate + N, As + (M * N));
-	carry1 = intermediate[2 * N];
-
-	// 2c. The third one:
-	SWIFFTSum(fftOut, M, intermediate + (2 * N), As + 2 * (M * N));
-	carry2 = intermediate[3 * N];
-
-	//2d. Put three carry bytes in their place
-	intermediate[3 * N] = carry0;
-	intermediate[(3 * N) + 1] = carry1;
-	intermediate[(3 * N) + 2] = carry2;
-
-	// Padding  intermediate output with 5 zeroes.
-	memset(intermediate + (3 * N) + 3, 0, 5);
-
-	// Apply the S-Box:
-	for (i = 0; i < (3 * N) + 8; ++i)
-	{
-		intermediate[i] = SBox[intermediate[i]];
-	}
-
-	// 3. The final and last SWIFFT:
-	SWIFFTFFT(intermediate, 3 * (N/8) + 1, fftOut);
-	SWIFFTSum(fftOut,       3 * (N/8) + 1, output, As);
-
-	if (doSmooth)
-	{
-		unsigned char sum[N];
-		register int i, j;
-		memset(sum, 0, N);
-
-		for (i = 0; i < (N + 1) * 8; ++i)
-		{
-			register const swift_int16_t *AsRow;
-			register int AShift;
-
-			if  (!(output[i >> 3] & (1 << (i & 7))))
-			{
-				continue;
-			}
-
-			AsRow = As + N * M + (i & ~(N - 1)) ;
-			AShift = i & 63;
-
-			for (j = AShift; j < N; ++j)
-			{
-				sum[j] += AsRow[j - AShift];
-			}
-
-			for(j = 0; j < AShift; ++j)
-			{
-				sum[j] -= AsRow[N - AShift + j];
-			}
-		}
-
-		for (i = 0; i < N; ++i)
-		{
-			output[i] = sum[i];
-		}
-
-		output[N] = 0;
-	}
-}

algo/swifftx/swifftx.c

@@ -18,6 +18,8 @@
 //#include "stdbool.h"
 #include <memory.h>
 
+#include "simd-utils.h"
+
 ///////////////////////////////////////////////////////////////////////////////////////////////
 // Constants and static tables portion.
 ///////////////////////////////////////////////////////////////////////////////////////////////
@@ -49,20 +51,20 @@
 // - A: the first operand. After the operation stores the sum of the two operands.
 // - B: the second operand. After the operation stores the difference between the first and the
 //   second operands.
-#define ADD_SUB(A, B) {register int temp = (B); B = ((A) - (B)); A = ((A) + (temp));}
+//#define ADD_SUB(A, B) {register int temp = (B); B = ((A) - (B)); A = ((A) + (temp));}
 
 // Quickly reduces an integer modulo 257.
 //
 // Parameters:
 // - A: the input.
-#define Q_REDUCE(A) (((A) & 0xff) - ((A) >> 8))
+//#define Q_REDUCE(A) (((A) & 0xff) - ((A) >> 8))
 
 // Since we need to do the setup only once, this is the indicator variable:
 static bool wasSetupDone = false;
 
 // This array stores the powers of omegas that correspond to the indices, which are the input
 // values. Known also as the "outer FFT twiddle factors".
-swift_int16_t multipliers[N];
+swift_int16_t multipliers[N] __attribute__ ((aligned (64)));
 
 // This array stores the powers of omegas, multiplied by the corresponding values.
 // We store this table to save computation time.
@@ -72,14 +74,14 @@ swift_int16_t multipliers[N];
 // compression function, i is between 0 and 31, x_i is a 64-bit value.
 // One can see the formula for this (intermediate) stage in the SWIFFT FSE 2008 paper --
 // formula (2), section 3, page 6.
-swift_int16_t fftTable[256 * EIGHTH_N];
+swift_int16_t fftTable[256 * EIGHTH_N] __attribute__ ((aligned (64)));
 
 // The A's we use in SWIFFTX shall be random elements of Z_257.
 // We generated these A's from the decimal expansion of PI as follows:  we converted each
 // triple of digits into a decimal number d. If d < (257 * 3) we used (d % 257) for the next A
 // element, otherwise move to the next triple of digits in the expansion. This guarntees that
 // the A's are random, provided that PI digits are.
-const swift_int16_t As[3 * M * N] =
+const swift_int16_t As[3 * M * N] __attribute__ ((aligned (64))) =
 {141,  78, 139,  75, 238, 205, 129, 126,  22, 245, 197, 169, 142, 118, 105,  78,
   50, 149,  29, 208, 114,  34,  85, 117,  67, 148,  86, 256,  25,  49, 133,  93,
   95,  36,  68, 231, 211, 102, 151, 128, 224, 117, 193,  27, 102, 187,   7, 105,
@@ -602,21 +604,14 @@ void InitializeSWIFFTX()
 	int omegaPowers[2 * N];
 	omegaPowers[0] = 1;
 
-	if (wasSetupDone)
-		return;
+	if (wasSetupDone) return;
 
 	for (i = 1; i < (2 * N); ++i)
-	{
 		omegaPowers[i] = Center(omegaPowers[i - 1] * OMEGA);
-	}
 
 	for (i = 0; i < (N / W); ++i)
-	{
 		for (j = 0; j < W; ++j)
-		{
 			multipliers[(i << 3) + j] = omegaPowers[ReverseBits(i, N / W) * (2 * j + 1)];
-		}
-	}
 
 	for (x = 0; x < 256; ++x)
 	{
@@ -624,10 +619,8 @@ void InitializeSWIFFTX()
 		{
 			register int temp = 0;
 			for (k = 0; k < 8; ++k)
-			{
 				temp += omegaPowers[(EIGHTH_N * (2 * j + 1) * ReverseBits(k, W)) % (2 * N)]
 					  * ((x >> k) & 1);
-			}
 
 			fftTable[(x << 3) + j] = Center(temp);
 		}
@@ -636,9 +629,203 @@ void InitializeSWIFFTX()
 	wasSetupDone = true;
 }
 
+// In the original code the F matrix is rotated so it was not aranged
+// the same as all the other data. Rearanging F to match all the other
+// data made vectorizing possible, the compiler probably could have been
+// able to auto-vectorize with proper data organisation.
+// Also in the original code the custom 16 bit data types are all now 32
+// bit int32_t regardless of the type name.
+//
 void FFT(const unsigned char input[EIGHTH_N], swift_int32_t *output)
 {
-	swift_int16_t *mult = multipliers;
+#if defined(__AVX2__)
+
+   __m256i F[8] __attribute__ ((aligned (64)));
+   __m256i *mul = (__m256i*)multipliers;
+   __m256i *out = (__m256i*)output;
+   __m256i *tbl = (__m256i*)&( fftTable[ input[0] << 3 ] );
+
+   F[0] = _mm256_mullo_epi32( mul[0], *tbl );
+   tbl = (__m256i*)&( fftTable[ input[1] << 3 ] );
+   F[1] = _mm256_mullo_epi32( mul[1], *tbl );
+   tbl = (__m256i*)&( fftTable[ input[2] << 3 ] );
+   F[2] = _mm256_mullo_epi32( mul[2], *tbl );
+   tbl = (__m256i*)&( fftTable[ input[3] << 3 ] );
+   F[3] = _mm256_mullo_epi32( mul[3], *tbl );
+   tbl = (__m256i*)&( fftTable[ input[4] << 3 ] );
+   F[4] = _mm256_mullo_epi32( mul[4], *tbl );
+   tbl = (__m256i*)&( fftTable[ input[5] << 3 ] );
+   F[5] = _mm256_mullo_epi32( mul[5], *tbl );
+   tbl = (__m256i*)&( fftTable[ input[6] << 3 ] );
+   F[6] = _mm256_mullo_epi32( mul[6], *tbl );
+   tbl = (__m256i*)&( fftTable[ input[7] << 3 ] );
+   F[7] = _mm256_mullo_epi32( mul[7], *tbl );
+
+   #define ADD_SUB( a, b ) \
+   { \
+      __m256i tmp = b; \
+      b = _mm256_sub_epi32( a, b ); \
+      a = _mm256_add_epi32( a, tmp ); \
+   }
+   
+   ADD_SUB( F[0], F[1] );
+   ADD_SUB( F[2], F[3] );
+   ADD_SUB( F[4], F[5] );
+   ADD_SUB( F[6], F[7] );
+
+   F[3] = _mm256_slli_epi32( F[3], 4 );
+   F[7] = _mm256_slli_epi32( F[7], 4 );
+
+   ADD_SUB( F[0], F[2] );
+   ADD_SUB( F[1], F[3] );
+   ADD_SUB( F[4], F[6] );
+   ADD_SUB( F[5], F[7] );  
+
+   F[5] = _mm256_slli_epi32( F[5], 2 );
+   F[6] = _mm256_slli_epi32( F[6], 4 );
+   F[7] = _mm256_slli_epi32( F[7], 6 );
+
+   ADD_SUB( F[0], F[4] );
+   ADD_SUB( F[1], F[5] );
+   ADD_SUB( F[2], F[6] );
+   ADD_SUB( F[3], F[7] );
+
+   #undef ADD_SUB
+
+#if defined (__AVX512VL__) && defined(__AVX512BW__)   
+
+   const __m256i mask = _mm256_movm_epi8( 0x11111111 );
+
+#else
+
+   const __m256i mask = m256_const1_32( 0x000000ff );
+
+#endif
+
+   #define Q_REDUCE( a ) \
+       _mm256_sub_epi32( _mm256_and_si256( a, mask ), \
+                         _mm256_srai_epi32( a, 8 ) )
+
+   out[0] = Q_REDUCE( F[0] );  
+   out[1] = Q_REDUCE( F[1] );                        
+   out[2] = Q_REDUCE( F[2] );                        
+   out[3] = Q_REDUCE( F[3] );                        
+   out[4] = Q_REDUCE( F[4] );                        
+   out[5] = Q_REDUCE( F[5] );                        
+   out[6] = Q_REDUCE( F[6] );                        
+   out[7] = Q_REDUCE( F[7] );
+
+   #undef Q_REDUCE
+
+#elif defined(__SSE4_1__)
+
+   __m128i F[16] __attribute__ ((aligned (64)));
+   __m128i *mul = (__m128i*)multipliers;
+   __m128i *out = (__m128i*)output;
+   __m128i *tbl = (__m128i*)&( fftTable[ input[0] << 3 ] );
+
+   F[ 0] = _mm_mullo_epi32( mul[ 0], tbl[0] );
+   F[ 1] = _mm_mullo_epi32( mul[ 1], tbl[1] );
+   tbl = (__m128i*)&( fftTable[ input[1] << 3 ] );
+   F[ 2] = _mm_mullo_epi32( mul[ 2], tbl[0] );
+   F[ 3] = _mm_mullo_epi32( mul[ 3], tbl[1] );
+   tbl = (__m128i*)&( fftTable[ input[2] << 3 ] );
+   F[ 4] = _mm_mullo_epi32( mul[ 4], tbl[0] );
+   F[ 5] = _mm_mullo_epi32( mul[ 5], tbl[1] );
+   tbl = (__m128i*)&( fftTable[ input[3] << 3 ] );
+   F[ 6] = _mm_mullo_epi32( mul[ 6], tbl[0] );
+   F[ 7] = _mm_mullo_epi32( mul[ 7], tbl[1] );
+   tbl = (__m128i*)&( fftTable[ input[4] << 3 ] );
+   F[ 8] = _mm_mullo_epi32( mul[ 8], tbl[0] );
+   F[ 9] = _mm_mullo_epi32( mul[ 9], tbl[1] );
+   tbl = (__m128i*)&( fftTable[ input[5] << 3 ] );
+   F[10] = _mm_mullo_epi32( mul[10], tbl[0] );
+   F[11] = _mm_mullo_epi32( mul[11], tbl[1] );
+   tbl = (__m128i*)&( fftTable[ input[6] << 3 ] );
+   F[12] = _mm_mullo_epi32( mul[12], tbl[0] );
+   F[13] = _mm_mullo_epi32( mul[13], tbl[1] );
+   tbl = (__m128i*)&( fftTable[ input[7] << 3 ] );
+   F[14] = _mm_mullo_epi32( mul[14], tbl[0] );
+   F[15] = _mm_mullo_epi32( mul[15], tbl[1] );
+
+   #define ADD_SUB( a, b ) \
+   { \
+      __m128i tmp = b; \
+      b = _mm_sub_epi32( a, b ); \
+      a = _mm_add_epi32( a, tmp ); \
+   }
+
+   ADD_SUB( F[ 0], F[ 2] );
+   ADD_SUB( F[ 1], F[ 3] );
+   ADD_SUB( F[ 4], F[ 6] );
+   ADD_SUB( F[ 5], F[ 7] );
+   ADD_SUB( F[ 8], F[10] );
+   ADD_SUB( F[ 9], F[11] );
+   ADD_SUB( F[12], F[14] );
+   ADD_SUB( F[13], F[15] );
+
+   F[ 6] = _mm_slli_epi32( F[ 6], 4 );
+   F[ 7] = _mm_slli_epi32( F[ 7], 4 );
+   F[14] = _mm_slli_epi32( F[14], 4 );
+   F[15] = _mm_slli_epi32( F[15], 4 );
+
+   ADD_SUB( F[ 0], F[ 4] );
+   ADD_SUB( F[ 1], F[ 5] );
+   ADD_SUB( F[ 2], F[ 6] );
+   ADD_SUB( F[ 3], F[ 7] );
+   ADD_SUB( F[ 8], F[12] );
+   ADD_SUB( F[ 9], F[13] );
+   ADD_SUB( F[10], F[14] );
+   ADD_SUB( F[11], F[15] );
+
+   F[10] = _mm_slli_epi32( F[10], 2 );
+   F[11] = _mm_slli_epi32( F[11], 2 );
+   F[12] = _mm_slli_epi32( F[12], 4 );
+   F[13] = _mm_slli_epi32( F[13], 4 );
+   F[14] = _mm_slli_epi32( F[14], 6 );
+   F[15] = _mm_slli_epi32( F[15], 6 );
+   
+   ADD_SUB( F[ 0], F[ 8] );
+   ADD_SUB( F[ 1], F[ 9] );
+   ADD_SUB( F[ 2], F[10] );
+   ADD_SUB( F[ 3], F[11] );
+   ADD_SUB( F[ 4], F[12] );
+   ADD_SUB( F[ 5], F[13] );
+   ADD_SUB( F[ 6], F[14] );
+   ADD_SUB( F[ 7], F[15] );
+
+   #undef ADD_SUB
+
+   const __m128i mask = m128_const1_32( 0x000000ff );
+
+   #define Q_REDUCE( a ) \
+      _mm_sub_epi32( _mm_and_si128( a, mask ), _mm_srai_epi32( a, 8 ) ) 
+
+   out[ 0] = Q_REDUCE( F[ 0] );
+   out[ 1] = Q_REDUCE( F[ 1] );
+   out[ 2] = Q_REDUCE( F[ 2] );
+   out[ 3] = Q_REDUCE( F[ 3] );
+   out[ 4] = Q_REDUCE( F[ 4] );
+   out[ 5] = Q_REDUCE( F[ 5] );
+   out[ 6] = Q_REDUCE( F[ 6] );
+   out[ 7] = Q_REDUCE( F[ 7] );
+   out[ 8] = Q_REDUCE( F[ 8] );
+   out[ 9] = Q_REDUCE( F[ 9] );
+   out[10] = Q_REDUCE( F[10] );
+   out[11] = Q_REDUCE( F[11] );
+   out[12] = Q_REDUCE( F[12] );
+   out[13] = Q_REDUCE( F[13] );
+   out[14] = Q_REDUCE( F[14] );
+   out[15] = Q_REDUCE( F[15] );
+
+   #undef Q_REDUCE
+
+#else   // < SSE4.1
+   
+   swift_int16_t *mult = multipliers;
+
+   // First loop unrolling:
+	register swift_int16_t *table = &(fftTable[input[0] << 3]);
 
 /*
    swift_int32_t F[64];
@@ -666,11 +853,8 @@ void FFT(const unsigned char input[EIGHTH_N], swift_int32_t *output)
                 F50, F51, F52, F53, F54, F55, F56, F57, F58, F59,
                 F60, F61, F62, F63;
    
-   // First loop unrolling:
-	register swift_int16_t *table = &(fftTable[input[0] << 3]);
-
-	F0 = mult[0] * table[0];
-	F8 = mult[1] * table[1];
+	F0  = mult[0] * table[0];
+	F8  = mult[1] * table[1];
 	F16 = mult[2] * table[2];
 	F24 = mult[3] * table[3];
 	F32 = mult[4] * table[4];
@@ -678,90 +862,93 @@ void FFT(const unsigned char input[EIGHTH_N], swift_int32_t *output)
 	F48 = mult[6] * table[6];
 	F56 = mult[7] * table[7];
 
-	mult += 8;
 	table = &(fftTable[input[1] << 3]);
 
-	F1 = mult[0] * table[0];
-	F9 = mult[1] * table[1];
-	F17 = mult[2] * table[2];
-	F25 = mult[3] * table[3];
-	F33 = mult[4] * table[4];
-	F41 = mult[5] * table[5];
-	F49 = mult[6] * table[6];
-	F57 = mult[7] * table[7];
+	F1  = mult[ 8] * table[0];
+	F9  = mult[ 9] * table[1];
+	F17 = mult[10] * table[2];
+	F25 = mult[11] * table[3];
+	F33 = mult[12] * table[4];
+	F41 = mult[13] * table[5];
+	F49 = mult[14] * table[6];
+	F57 = mult[15] * table[7];
 
-	mult += 8;
 	table = &(fftTable[input[2] << 3]);
 
-	F2 = mult[0] * table[0];
-	F10 = mult[1] * table[1];
-	F18 = mult[2] * table[2];
-	F26 = mult[3] * table[3];
-	F34 = mult[4] * table[4];
-	F42 = mult[5] * table[5];
-	F50 = mult[6] * table[6];
-	F58 = mult[7] * table[7];
+	F2  = mult[16] * table[0];
+	F10 = mult[17] * table[1];
+	F18 = mult[18] * table[2];
+	F26 = mult[19] * table[3];
+	F34 = mult[20] * table[4];
+	F42 = mult[21] * table[5];
+	F50 = mult[22] * table[6];
+	F58 = mult[23] * table[7];
 
-	mult += 8;
 	table = &(fftTable[input[3] << 3]);
 
-	F3 = mult[0] * table[0];
-	F11 = mult[1] * table[1];
-	F19 = mult[2] * table[2];
-	F27 = mult[3] * table[3];
-	F35 = mult[4] * table[4];
-	F43 = mult[5] * table[5];
-	F51 = mult[6] * table[6];
-	F59 = mult[7] * table[7];
+	F3  = mult[24] * table[0];
+	F11 = mult[25] * table[1];
+	F19 = mult[26] * table[2];
+	F27 = mult[27] * table[3];
+	F35 = mult[28] * table[4];
+	F43 = mult[29] * table[5];
+	F51 = mult[30] * table[6];
+	F59 = mult[31] * table[7];
 
-	mult += 8;
 	table = &(fftTable[input[4] << 3]);
 
-	F4 = mult[0] * table[0];
-	F12 = mult[1] * table[1];
-	F20 = mult[2] * table[2];
-	F28 = mult[3] * table[3];
-	F36 = mult[4] * table[4];
-	F44 = mult[5] * table[5];
-	F52 = mult[6] * table[6];
-	F60 = mult[7] * table[7];
+	F4  = mult[32] * table[0];
+	F12 = mult[33] * table[1];
+	F20 = mult[34] * table[2];
+	F28 = mult[35] * table[3];
+	F36 = mult[36] * table[4];
+	F44 = mult[37] * table[5];
+	F52 = mult[38] * table[6];
+	F60 = mult[39] * table[7];
 
-	mult += 8;
 	table = &(fftTable[input[5] << 3]);
 
-	F5 = mult[0] * table[0];
-	F13 = mult[1] * table[1];
-	F21 = mult[2] * table[2];
-	F29 = mult[3] * table[3];
-	F37 = mult[4] * table[4];
-	F45 = mult[5] * table[5];
-	F53 = mult[6] * table[6];
-	F61 = mult[7] * table[7];
+	F5  = mult[40] * table[0];
+	F13 = mult[41] * table[1];
+	F21 = mult[42] * table[2];
+	F29 = mult[43] * table[3];
+	F37 = mult[44] * table[4];
+	F45 = mult[45] * table[5];
+	F53 = mult[46] * table[6];
+	F61 = mult[47] * table[7];
 
-	mult += 8;
 	table = &(fftTable[input[6] << 3]);
 
-	F6 = mult[0] * table[0];
-	F14 = mult[1] * table[1];
-	F22 = mult[2] * table[2];
-	F30 = mult[3] * table[3];
-	F38 = mult[4] * table[4];
-	F46 = mult[5] * table[5];
-	F54 = mult[6] * table[6];
-	F62 = mult[7] * table[7];
+	F6  = mult[48] * table[0];
+	F14 = mult[49] * table[1];
+	F22 = mult[50] * table[2];
+	F30 = mult[51] * table[3];
+	F38 = mult[52] * table[4];
+	F46 = mult[53] * table[5];
+	F54 = mult[54] * table[6];
+	F62 = mult[55] * table[7];
 
-	mult += 8;
 	table = &(fftTable[input[7] << 3]);
 
-	F7 = mult[0] * table[0];
-	F15 = mult[1] * table[1];
-	F23 = mult[2] * table[2];
-	F31 = mult[3] * table[3];
-	F39 = mult[4] * table[4];
-	F47 = mult[5] * table[5];
-	F55 = mult[6] * table[6];
-	F63 = mult[7] * table[7];
+	F7  = mult[56] * table[0];
+	F15 = mult[57] * table[1];
+	F23 = mult[58] * table[2];
+	F31 = mult[59] * table[3];
+	F39 = mult[60] * table[4];
+	F47 = mult[61] * table[5];
+	F55 = mult[62] * table[6];
+	F63 = mult[63] * table[7];
 
+   #define ADD_SUB( a, b ) \
+   { \
+      int temp = b; \
+      b = a - b; \
+      a = a + temp; \
+   }
+   
+   #define Q_REDUCE( a ) \
+      ( ( (a) & 0xff ) - ( (a) >> 8 ) )
+   
 /*
 
    for ( int i = 0; i < 8; i++ )
@@ -800,7 +987,6 @@ void FFT(const unsigned char input[EIGHTH_N], swift_int32_t *output)
    }
 */
 
-
 	// Second loop unrolling:
 	// Iteration 0:
 	ADD_SUB(F0, F1);
@@ -1057,6 +1243,11 @@ void FFT(const unsigned char input[EIGHTH_N], swift_int32_t *output)
 	output[47] = Q_REDUCE(F61);
 	output[55] = Q_REDUCE(F62);
 	output[63] = Q_REDUCE(F63);
+
+   #undef ADD_SUB
+   #undef Q_REDUCE
+
+#endif  // AVX2 elif SSE4.1 else
 }
 
 // Calculates the FFT part of SWIFFT.
@@ -1086,24 +1277,66 @@ void SWIFFTFFT(const unsigned char *input, int m, swift_int32_t *output)
 // - m: the input size divided by 64.
 // - output: will store the result.
 // - a: the coefficients in the sum. Of size 64 * m.
-void SWIFFTSum(const swift_int32_t *input, int m, unsigned char *output, const swift_int16_t *a)
+void SWIFFTSum( const swift_int32_t *input, int m, unsigned char *output,
+                const swift_int16_t *a )
 {
 	int i, j;
-	swift_int32_t result[N];
+	swift_int32_t result[N] __attribute__ ((aligned (64)));
 	register swift_int16_t carry = 0;
 
+#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+
+   __m512i *res = (__m512i*)result;
+   for ( j = 0; j < N/16; ++j )
+   {
+      __m512i sum = _mm512_setzero_si512();
+      const __m512i *f = (__m512i*)input + j;
+      const __m512i *k = (__m512i*)a + j;
+      for ( i = 0; i < m; i++, f += N/16, k += N/16 )
+         sum = _mm512_add_epi32( sum, _mm512_mullo_epi32( *f, *k ) );
+      res[j] = sum;
+   }
+
+#elif defined(__AVX2__)
+
+   __m256i *res = (__m256i*)result;
+   for ( j = 0; j < N/8; ++j )
+   {
+      __m256i sum = _mm256_setzero_si256();
+      const __m256i *f = (__m256i*)input + j;
+      const __m256i *k = (__m256i*)a + j;
+      for ( i = 0; i < m; i++, f += N/8, k += N/8 )
+         sum = _mm256_add_epi32( sum, _mm256_mullo_epi32( *f, *k ) );
+      res[j] = sum;
+   }
+
+#elif defined(__SSE4_1__)
+
+   __m128i *res = (__m128i*)result;
+   for ( j = 0; j < N/4; ++j )
+   {
+      __m128i sum = _mm_setzero_si128();
+      const __m128i *f = (__m128i*)input + j;
+      const __m128i *k = (__m128i*)a + j;
+      for ( i = 0; i < m; i++, f += N/4, k += N/4 )
+         sum = _mm_add_epi32( sum, _mm_mullo_epi32( *f, *k ) );
+      res[j] = sum;
+   }
+
+#else
+
 	for (j = 0; j < N; ++j)
 	{
 		register swift_int32_t sum = 0;
 		const register swift_int32_t *f = input + j;
 		const register swift_int16_t *k = a + j;
-
 		for (i = 0; i < m; i++, f += N,k += N)
 			sum += (*f) * (*k);
-
 		result[j] = sum;
 	}
 
+#endif
+
 	for (j = 0; j < N; ++j)
 		result[j] = ((FIELD_SIZE << 22) + result[j]) % FIELD_SIZE;
 
@@ -1116,14 +1349,15 @@ void SWIFFTSum(const swift_int32_t *input, int m, unsigned char *output, const s
 	output[N] = carry;
 }
 
+/*
 void ComputeSingleSWIFFTX_smooth(unsigned char input[SWIFFTX_INPUT_BLOCK_SIZE],
                           unsigned char output[SWIFFTX_OUTPUT_BLOCK_SIZE],
 						  bool doSmooth)
 {
 	int i;
 	// Will store the result of the FFT parts:
-	swift_int32_t fftOut[N * M];
-	unsigned char intermediate[N * 3 + 8];
+	swift_int32_t fftOut[N * M] __attribute__ ((aligned (64)));
+	unsigned char intermediate[N * 3 + 8] __attribute__ ((aligned (64)));
 	unsigned char carry0,carry1,carry2;
 
 	// Do the three SWIFFTS while remembering the three carry bytes (each carry byte gets
@@ -1193,51 +1427,50 @@ void ComputeSingleSWIFFTX_smooth(unsigned char input[SWIFFTX_INPUT_BLOCK_SIZE],
 		output[N] = 0;
 	}
 }
+*/
 
-void ComputeSingleSWIFFTX( unsigned char input[SWIFFTX_INPUT_BLOCK_SIZE],
-                           unsigned char output[SWIFFTX_OUTPUT_BLOCK_SIZE] )
+void ComputeSingleSWIFFTX( unsigned char *input, unsigned char *output )
 {
    int i;
    // Will store the result of the FFT parts:
-   swift_int32_t fftOut[N * M];
-   unsigned char intermediate[N * 3 + 8];
+   swift_int32_t fftOut[N * M] __attribute__ ((aligned (64)));
+   unsigned char sum[ N*3 + 8 ] __attribute__ ((aligned (64)));
    unsigned char carry0,carry1,carry2;
 
    // Do the three SWIFFTS while remembering the three carry bytes (each carry byte gets
    // overriden by the following SWIFFT):
 
    // 1. Compute the FFT of the input - the common part for the first 3 SWIFFTs:
-   SWIFFTFFT(input, M, fftOut);
+   SWIFFTFFT( input, M, fftOut );
 
    // 2. Compute the sums of the 3 SWIFFTs, each using a different set of coefficients:
 
    // 2a. The first SWIFFT:
-   SWIFFTSum(fftOut, M, intermediate, As);
-   // Remember the carry byte:
-   carry0 = intermediate[N];
+   SWIFFTSum( fftOut, M, sum,       As         );
+   carry0 = sum[N];
 
    // 2b. The second one:
-   SWIFFTSum(fftOut, M, intermediate + N, As + (M * N));
-   carry1 = intermediate[2 * N];
+   SWIFFTSum( fftOut, M, sum + N,   As +   M*N );
+   carry1 = sum[ 2*N ];
 
    // 2c. The third one:
-   SWIFFTSum(fftOut, M, intermediate + (2 * N), As + 2 * (M * N));
-   carry2 = intermediate[3 * N];
+   SWIFFTSum( fftOut, M, sum + 2*N, As + 2*M*N );
+   carry2 = sum[ 3*N ];
 
    //2d. Put three carry bytes in their place
-   intermediate[3 * N] = carry0;
-   intermediate[(3 * N) + 1] = carry1;
-   intermediate[(3 * N) + 2] = carry2;
+   sum[ 3*N     ] = carry0;
+   sum[ 3*N + 1 ] = carry1;
+   sum[ 3*N + 2 ] = carry2;
 
    // Padding  intermediate output with 5 zeroes.
-   memset(intermediate + (3 * N) + 3, 0, 5);
+   memset( sum + 3*N + 3, 0, 5 );
 
    // Apply the S-Box:
    for ( i = 0; i < (3 * N) + 8; ++i )
-      intermediate[i] = SBox[intermediate[i]];
+      sum[i] = SBox[ sum[i] ];
 
    // 3. The final and last SWIFFT:
-   SWIFFTFFT(intermediate, 3 * (N/8) + 1, fftOut);
-   SWIFFTSum(fftOut,       3 * (N/8) + 1, output, As);
-
+   SWIFFTFFT( sum, 3 * (N/8) + 1, fftOut );
+   SWIFFTSum( fftOut,       3 * (N/8) + 1, sum, As );
+   memcpy( output, sum, SWIFFTX_OUTPUT_BLOCK_SIZE - 1 );
 }

algo/swifftx/swifftx.h

@@ -61,11 +61,10 @@ void ComputeSingleSWIFFT(unsigned char *input, unsigned short m,
 //
 // Returns:
 // - Success value.
-void ComputeSingleSWIFFTX( unsigned char input[SWIFFTX_INPUT_BLOCK_SIZE],
-                           unsigned char output[SWIFFTX_OUTPUT_BLOCK_SIZE] );
+void ComputeSingleSWIFFTX( unsigned char *input, unsigned char *output );
 
-void ComputeSingleSWIFFTX_smooth( unsigned char input[SWIFFTX_INPUT_BLOCK_SIZE],
-	            unsigned char output[SWIFFTX_OUTPUT_BLOCK_SIZE], bool doSmooth);
+//void ComputeSingleSWIFFTX_smooth( unsigned char input[SWIFFTX_INPUT_BLOCK_SIZE],
+//	            unsigned char output[SWIFFTX_OUTPUT_BLOCK_SIZE], bool doSmooth);
 
 // Calculates the powers of OMEGA and generates the bit reversal permutation.
 // You must call this function before doing SWIFFT/X, otherwise you will get zeroes everywhere.

algo/verthash/Verthash.c

@@ -176,12 +176,6 @@ static void rotate_indexes( uint32_t *p )
 */
 }
 #endif
-
-static inline uint32_t rotl32( uint32_t a, size_t r )
-{
-   return ( a << r ) | ( a >> (32-r) );
-}
-
 // Vectorized and targetted version of fnv1a
 #if defined (__AVX2__)        
 
@@ -232,7 +226,7 @@ for ( size_t i = 0; i < VH_N_SUBSET / sizeof(uint32_t); i++ ) \
 for ( size_t i = 0; i < VH_N_SUBSET / sizeof(uint32_t); i++ ) \
 { \
    const uint32_t *blob_off = blob + \
-                 ( ( fnv1a( rotl32( subset[i], r ), accumulator ) % mdiv ) \
+                 ( ( fnv1a( rol32( subset[i], r ), accumulator ) % mdiv ) \
                  * ( VH_BYTE_ALIGNMENT / sizeof(uint32_t) ) ); \
    UPDATE_ACCUMULATOR; \
    MULXOR; \

algo/verthash/verthash-gate.c

@@ -1,5 +1,5 @@
 #include "algo-gate-api.h"
-#include "algo/sha/sph_sha2.h"
+#include "algo/sha/sha256-hash.h"
 #include "Verthash.h"
 #include "tiny_sha3/sha3-4way.h"
 
@@ -140,7 +140,7 @@ bool register_verthash_algo( algo_gate_t* gate )
          uint8_t vhDataFileHash[32] = { 0 };
 
          applog( LOG_NOTICE, "Verifying Verthash data" );
-         sph_sha256_full( vhDataFileHash, verthashInfo.data,
+         sha256_full( vhDataFileHash, verthashInfo.data,
                           verthashInfo.dataSize );
          if ( memcmp( vhDataFileHash, verthashDatFileHash_bytes,
                       sizeof(verthashDatFileHash_bytes) ) == 0 )

algo/whirlpool/whirlpool.c

@@ -82,7 +82,7 @@ int scanhash_whirlpool( struct work* work, uint32_t max_nonce,
 		be32enc(&endiandata[19], n );
 		whirlpool_hash(vhash, endiandata);
 
-		if (vhash[7] <= Htarg && fulltest(vhash, ptarget))
+		if (vhash[7] <= Htarg && fulltest(vhash, ptarget) & ! opt_benchmark )
              submit_solution( work, vhash, mythr );
 	} while ( n < max_nonce && !work_restart[thr_id].restart);
 

algo/x16/x16r-4way.c

@@ -52,10 +52,10 @@ void x16r_8way_prehash( void *vdata, void *pdata )
       break;
       case CUBEHASH:
          mm128_bswap32_80( edata, pdata );
-         cubehashInit( &x16r_ctx.cube, 512, 16, 32 );
-         cubehashUpdate( &x16r_ctx.cube, (const byte*)edata, 64 );
-         intrlv_8x64( vdata, edata, edata, edata, edata,
-                             edata, edata, edata, edata, 640 );
+         intrlv_4x128( vdata2, edata, edata, edata, edata, 640 );
+         cube_4way_init( &x16r_ctx.cube, 512, 16, 32 );
+         cube_4way_update( &x16r_ctx.cube, vdata2, 64 );
+         rintrlv_4x128_8x64( vdata, vdata2, vdata2, 640 );
       break;
       case HAMSI:
          mm512_bswap32_intrlv80_8x64( vdata, pdata );
@@ -207,15 +207,15 @@ int x16r_8way_hash_generic( void* output, const void* input, int thrid )
          case LUFFA:
             if ( i == 0 )
             {
-                intrlv_4x128( vhash, in0, in1, in2, in3, size<<3 );
-                luffa_4way_update_close( &ctx.luffa, vhash,
-                                                     vhash + (16<<2), 16 );
-                dintrlv_4x128_512( hash0, hash1, hash2, hash3, vhash );
-                memcpy( &ctx, &x16r_ctx, sizeof(ctx) );
-                intrlv_4x128( vhash, in4, in5, in6, in7, size<<3 );
-                luffa_4way_update_close( &ctx.luffa, vhash, 
-                                                     vhash + (16<<2), 16 );
-                dintrlv_4x128_512( hash4, hash5, hash6, hash7, vhash );
+               intrlv_4x128( vhash, in0, in1, in2, in3, size<<3 );
+               luffa_4way_update_close( &ctx.luffa, vhash,
+                                                    vhash + (16<<2), 16 );
+               dintrlv_4x128_512( hash0, hash1, hash2, hash3, vhash );
+               memcpy( &ctx, &x16r_ctx, sizeof(ctx) );
+               intrlv_4x128( vhash, in4, in5, in6, in7, size<<3 );
+               luffa_4way_update_close( &ctx.luffa, vhash, 
+                                                    vhash + (16<<2), 16 );
+               dintrlv_4x128_512( hash4, hash5, hash6, hash7, vhash );
             }
             else
             {
@@ -230,56 +230,24 @@ int x16r_8way_hash_generic( void* output, const void* input, int thrid )
          case CUBEHASH:
             if ( i == 0 )
             {
-               cubehashUpdateDigest( &ctx.cube, (byte*)hash0,
-                                            (const byte*)in0 + 64, 16 );
+               intrlv_4x128( vhash, in0, in1, in2, in3, size<<3 );
+               cube_4way_update_close( &ctx.cube, vhash,
+                                                  vhash + (16<<2), 16 );
+               dintrlv_4x128_512( hash0, hash1, hash2, hash3, vhash );
                memcpy( &ctx, &x16r_ctx, sizeof(ctx) );
-               cubehashUpdateDigest( &ctx.cube, (byte*)hash1,
-                                            (const byte*)in1 + 64, 16 );
-               memcpy( &ctx, &x16r_ctx, sizeof(ctx) );
-               cubehashUpdateDigest( &ctx.cube, (byte*)hash2,
-                                            (const byte*)in2 + 64, 16 );
-               memcpy( &ctx, &x16r_ctx, sizeof(ctx) );
-               cubehashUpdateDigest( &ctx.cube, (byte*)hash3,
-                                            (const byte*)in3 + 64, 16 );
-               memcpy( &ctx, &x16r_ctx, sizeof(ctx) );
-               cubehashUpdateDigest( &ctx.cube, (byte*)hash4,
-                                            (const byte*)in4 + 64, 16 );
-               memcpy( &ctx, &x16r_ctx, sizeof(ctx) );
-               cubehashUpdateDigest( &ctx.cube, (byte*)hash5,
-                                            (const byte*)in5 + 64, 16 );
-               memcpy( &ctx, &x16r_ctx, sizeof(ctx) );
-               cubehashUpdateDigest( &ctx.cube, (byte*)hash6,
-                                            (const byte*)in6 + 64, 16 );
-               memcpy( &ctx, &x16r_ctx, sizeof(ctx) );
-               cubehashUpdateDigest( &ctx.cube, (byte*)hash7,
-                                            (const byte*)in7 + 64, 16 );
+               intrlv_4x128( vhash, in4, in5, in6, in7, size<<3 );
+               cube_4way_update_close( &ctx.cube, vhash,
+                                                  vhash + (16<<2), 16 );
+               dintrlv_4x128_512( hash4, hash5, hash6, hash7, vhash );
             }
             else
             {
-               cubehashInit( &ctx.cube, 512, 16, 32 );
-               cubehashUpdateDigest( &ctx.cube, (byte*) hash0,
-                                             (const byte*)in0, size );
-               cubehashInit( &ctx.cube, 512, 16, 32 );
-               cubehashUpdateDigest( &ctx.cube, (byte*) hash1,
-                                             (const byte*)in1, size );
-               cubehashInit( &ctx.cube, 512, 16, 32 );
-               cubehashUpdateDigest( &ctx.cube, (byte*) hash2,
-                                             (const byte*)in2, size );
-               cubehashInit( &ctx.cube, 512, 16, 32 );
-               cubehashUpdateDigest( &ctx.cube, (byte*) hash3,
-                                             (const byte*)in3, size );
-               cubehashInit( &ctx.cube, 512, 16, 32 );
-               cubehashUpdateDigest( &ctx.cube, (byte*) hash4,
-                                             (const byte*)in4, size );
-               cubehashInit( &ctx.cube, 512, 16, 32 );
-               cubehashUpdateDigest( &ctx.cube, (byte*) hash5,
-                                             (const byte*)in5, size );
-               cubehashInit( &ctx.cube, 512, 16, 32 );
-               cubehashUpdateDigest( &ctx.cube, (byte*) hash6,
-                                             (const byte*)in6, size );
-               cubehashInit( &ctx.cube, 512, 16, 32 );
-               cubehashUpdateDigest( &ctx.cube, (byte*) hash7,
-                                             (const byte*)in7, size );
+               intrlv_4x128( vhash, in0, in1, in2, in3, size<<3 );
+               cube_4way_full( &ctx.cube, vhash, 512, vhash, size );
+               dintrlv_4x128_512( hash0, hash1, hash2, hash3, vhash );
+               intrlv_4x128( vhash, in4, in5, in6, in7, size<<3 );
+               cube_4way_full( &ctx.cube, vhash, 512, vhash, size );
+               dintrlv_4x128_512( hash4, hash5, hash6, hash7, vhash );
             }
          break;
          case SHAVITE:
@@ -556,9 +524,10 @@ void x16r_4way_prehash( void *vdata, void *pdata )
          break;
       case CUBEHASH:
          mm128_bswap32_80( edata, pdata );
-         cubehashInit( &x16r_ctx.cube, 512, 16, 32 );
-         cubehashUpdate( &x16r_ctx.cube, (const byte*)edata, 64 );
-         intrlv_4x64( vdata, edata, edata, edata, edata, 640 );
+         intrlv_2x128( vdata2, edata, edata, 640 );
+         cube_2way_init( &x16r_ctx.cube, 512, 16, 32 );
+         cube_2way_update( &x16r_ctx.cube, vdata2, 64 );
+         rintrlv_2x128_4x64( vdata, vdata2, vdata2, 640 );
       break;
       case HAMSI:
          mm256_bswap32_intrlv80_4x64( vdata, pdata );
@@ -680,13 +649,13 @@ int x16r_4way_hash_generic( void* output, const void* input, int thrid )
          case LUFFA:
             if ( i == 0 )
             {
-               intrlv_2x128( vhash, hash0, hash1, 640 );
-               luffa_2way_update_close( &ctx.luffa, vhash, vhash + (16<<1), 16 );
-               dintrlv_2x128_512( hash0, hash1, vhash );
-               intrlv_2x128( vhash, hash2, hash3, 640 );
-               memcpy( &ctx, &x16r_ctx, sizeof(ctx) );
-               luffa_2way_update_close( &ctx.luffa, vhash, vhash + (16<<1), 16 );
-               dintrlv_2x128_512( hash2, hash3, vhash );
+              intrlv_2x128( vhash, hash0, hash1, 640 );
+              luffa_2way_update_close( &ctx.luffa, vhash, vhash + (16<<1), 16 );
+              dintrlv_2x128_512( hash0, hash1, vhash );
+              intrlv_2x128( vhash, hash2, hash3, 640 );
+              memcpy( &ctx, &x16r_ctx, sizeof(ctx) );
+              luffa_2way_update_close( &ctx.luffa, vhash, vhash + (16<<1), 16 );
+              dintrlv_2x128_512( hash2, hash3, vhash );
             }
             else
             {
@@ -701,32 +670,24 @@ int x16r_4way_hash_generic( void* output, const void* input, int thrid )
          case CUBEHASH:
             if ( i == 0 )
             {
-               cubehashUpdateDigest( &ctx.cube, (byte*)hash0,
-                                          (const byte*)in0 + 64, 16 );
+               intrlv_2x128( vhash, in0, in1, size<<3 );
+               cube_2way_update_close( &ctx.cube, vhash,
+                                                  vhash + (16<<1), 16 );
+               dintrlv_2x128_512( hash0, hash1, vhash );
                memcpy( &ctx, &x16r_ctx, sizeof(ctx) );
-               cubehashUpdateDigest( &ctx.cube, (byte*) hash1,
-                                          (const byte*)in1 + 64, 16 );
-               memcpy( &ctx, &x16r_ctx, sizeof(ctx) );
-               cubehashUpdateDigest( &ctx.cube, (byte*) hash2,
-                                          (const byte*)in2 + 64, 16 );
-               memcpy( &ctx, &x16r_ctx, sizeof(ctx) );
-               cubehashUpdateDigest( &ctx.cube, (byte*) hash3,
-                                          (const byte*)in3 + 64, 16 );
+               intrlv_2x128( vhash, in2, in3, size<<3 );
+               cube_2way_update_close( &ctx.cube, vhash,
+                                                  vhash + (16<<1), 16 );
+               dintrlv_2x128_512( hash2, hash3, vhash );
             }
             else
             {
-               cubehashInit( &ctx.cube, 512, 16, 32 );
-               cubehashUpdateDigest( &ctx.cube, (byte*) hash0,
-                                     (const byte*)in0, size );
-               cubehashInit( &ctx.cube, 512, 16, 32 );
-               cubehashUpdateDigest( &ctx.cube, (byte*) hash1,
-                                     (const byte*)in1, size );
-               cubehashInit( &ctx.cube, 512, 16, 32 );
-               cubehashUpdateDigest( &ctx.cube, (byte*) hash2,
-                                     (const byte*)in2, size );
-               cubehashInit( &ctx.cube, 512, 16, 32 );
-               cubehashUpdateDigest( &ctx.cube, (byte*) hash3,
-                                     (const byte*)in3, size );
+               intrlv_2x128( vhash, in0, in1, size<<3 );
+               cube_2way_full( &ctx.cube, vhash, 512, vhash, size );
+               dintrlv_2x128_512( hash0, hash1, vhash );
+               intrlv_2x128( vhash, in2, in3, size<<3 );
+               cube_2way_full( &ctx.cube, vhash, 512, vhash, size );
+               dintrlv_2x128_512( hash2, hash3, vhash );
             }
          break;
          case SHAVITE:

algo/x16/x16r-gate.c

@@ -1,4 +1,5 @@
 #include "x16r-gate.h"
+#include "algo/sha/sha256d.h"
 
 __thread char x16r_hash_order[ X16R_HASH_FUNC_COUNT + 1 ] = { 0 };
 
@@ -61,8 +62,7 @@ bool register_x16r_algo( algo_gate_t* gate )
   gate->scanhash  = (void*)&scanhash_x16r;
   gate->hash      = (void*)&x16r_hash;
 #endif
-  gate->optimizations = SSE2_OPT | AES_OPT | AVX2_OPT | AVX512_OPT |
-	                VAES_OPT | VAES256_OPT;
+  gate->optimizations = SSE2_OPT | AES_OPT | AVX2_OPT | AVX512_OPT | VAES_OPT;
   x16_r_s_getAlgoString = (void*)&x16r_getAlgoString;
   opt_target_factor = 256.0;
   return true;
@@ -80,8 +80,7 @@ bool register_x16rv2_algo( algo_gate_t* gate )
   gate->scanhash  = (void*)&scanhash_x16rv2;
   gate->hash      = (void*)&x16rv2_hash;
 #endif
-  gate->optimizations = SSE2_OPT | AES_OPT | AVX2_OPT | AVX512_OPT |
-	                VAES_OPT | VAES256_OPT;
+  gate->optimizations = SSE2_OPT | AES_OPT | AVX2_OPT | AVX512_OPT | VAES_OPT;
   x16_r_s_getAlgoString = (void*)&x16r_getAlgoString;
   opt_target_factor = 256.0;
   return true;
@@ -99,8 +98,7 @@ bool register_x16s_algo( algo_gate_t* gate )
   gate->scanhash  = (void*)&scanhash_x16r;
   gate->hash      = (void*)&x16r_hash;
 #endif
-  gate->optimizations = SSE2_OPT | AES_OPT | AVX2_OPT | AVX512_OPT |
-	                VAES_OPT | VAES256_OPT;
+  gate->optimizations = SSE2_OPT | AES_OPT | AVX2_OPT | AVX512_OPT | VAES_OPT;
   x16_r_s_getAlgoString = (void*)&x16s_getAlgoString;
   opt_target_factor = 256.0;
   return true;
@@ -233,8 +231,7 @@ bool register_x16rt_algo( algo_gate_t* gate )
   gate->scanhash  = (void*)&scanhash_x16rt;
   gate->hash      = (void*)&x16r_hash;
 #endif
-  gate->optimizations = SSE2_OPT | AES_OPT | AVX2_OPT | AVX512_OPT |
-	                VAES_OPT | VAES256_OPT;
+  gate->optimizations = SSE2_OPT | AES_OPT | AVX2_OPT | AVX512_OPT | VAES_OPT;
   opt_target_factor = 256.0;
   return true;
 };
@@ -251,8 +248,7 @@ bool register_x16rt_veil_algo( algo_gate_t* gate )
   gate->scanhash  = (void*)&scanhash_x16rt;
   gate->hash      = (void*)&x16r_hash;
 #endif
-  gate->optimizations = SSE2_OPT | AES_OPT | AVX2_OPT | AVX512_OPT |
-	                VAES_OPT | VAES256_OPT;
+  gate->optimizations = SSE2_OPT | AES_OPT | AVX2_OPT | AVX512_OPT | VAES_OPT;
   gate->build_extraheader = (void*)&veil_build_extraheader;
   opt_target_factor = 256.0;
   return true;
@@ -291,8 +287,7 @@ bool register_x21s_algo( algo_gate_t* gate )
   gate->hash              = (void*)&x21s_hash;
   gate->miner_thread_init = (void*)&x21s_thread_init;
 #endif
-  gate->optimizations     = SSE2_OPT | AES_OPT | AVX2_OPT | AVX512_OPT |
-	                    VAES_OPT | VAES256_OPT;
+  gate->optimizations  = SSE2_OPT | AES_OPT | AVX2_OPT | AVX512_OPT | VAES_OPT;
   x16_r_s_getAlgoString   = (void*)&x16s_getAlgoString;
   opt_target_factor = 256.0;
   return true;

algo/x16/x16r-gate.h

@@ -37,6 +37,7 @@
 #include "algo/jh/jh-hash-4way.h"
 #include "algo/keccak/keccak-hash-4way.h"
 #include "algo/luffa/luffa-hash-2way.h"
+#include "algo/cubehash/cube-hash-2way.h"
 #include "algo/simd/simd-hash-2way.h"
 #include "algo/echo/aes_ni/hash_api.h"
 #include "algo/hamsi/hamsi-hash-4way.h"
@@ -115,7 +116,7 @@ union _x16r_8way_context_overlay
     jh512_8way_context      jh;
     keccak512_8way_context  keccak;
     luffa_4way_context      luffa;
-    cubehashParam           cube;
+    cube_4way_context       cube;
     simd_4way_context       simd;
     hamsi512_8way_context   hamsi;
     hashState_fugue         fugue;
@@ -164,8 +165,8 @@ union _x16r_4way_context_overlay
     jh512_4way_context      jh;
     keccak512_4way_context  keccak;
     luffa_2way_context      luffa;
+    cube_2way_context       cube;
     hashState_luffa         luffa1;
-    cubehashParam           cube;
     simd_2way_context       simd;
     hamsi512_4way_context   hamsi;
     hashState_fugue         fugue;

algo/x16/x21s-4way.c

@@ -13,7 +13,7 @@
 #include "algo/gost/sph_gost.h"
 #include "algo/lyra2/lyra2.h"
 #if defined(__SHA__)
-  #include "algo/sha/sph_sha2.h"
+  #include "algo/sha/sha256-hash.h"
 #endif
 
 #if defined (X21S_8WAY)
@@ -208,9 +208,7 @@ union _x21s_4way_context_overlay
     haval256_5_4way_context haval;
     sph_tiger_context       tiger;
     sph_gost512_context     gost;
-#if defined(__SHA__)
-    sph_sha256_context      sha256;
-#else
+#if !defined(__SHA__)
     sha256_4way_context     sha256;
 #endif
 } __attribute__ ((aligned (64)));
@@ -275,18 +273,10 @@ int x21s_4way_hash( void* output, const void* input, int thrid )
 
 #if defined(__SHA__)
 
-   sph_sha256_init( &ctx.sha256 );
-   sph_sha256( &ctx.sha256, hash0, 64 );
-   sph_sha256_close( &ctx.sha256, output );
-   sph_sha256_init( &ctx.sha256 );
-   sph_sha256( &ctx.sha256, hash1, 64 );
-   sph_sha256_close( &ctx.sha256, output+32 );
-   sph_sha256_init( &ctx.sha256 );
-   sph_sha256( &ctx.sha256, hash2, 64 );
-   sph_sha256_close( &ctx.sha256, output+64 );
-   sph_sha256_init( &ctx.sha256 );
-   sph_sha256( &ctx.sha256, hash3, 64 );
-   sph_sha256_close( &ctx.sha256, output+96 );
+   sha256_full( output,    hash0, 64 );
+   sha256_full( output+32, hash1, 64 );
+   sha256_full( output+64, hash2, 64 );
+   sha256_full( output+96, hash3, 64 );
 
 #else
 

algo/x16/x21s.c

@@ -8,7 +8,7 @@
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
-#include "algo/sha/sph_sha2.h"
+#include "algo/sha/sha256-hash.h"
 #include "algo/haval/sph-haval.h"
 #include "algo/tiger/sph_tiger.h"
 #include "algo/gost/sph_gost.h"
@@ -23,7 +23,7 @@ union _x21s_context_overlay
         sph_haval256_5_context  haval;
         sph_tiger_context       tiger;
         sph_gost512_context     gost;
-        sph_sha256_context      sha256;
+        sha256_context      sha256;
 };
 typedef union _x21s_context_overlay x21s_context_overlay;
 
@@ -50,9 +50,7 @@ int x21s_hash( void* output, const void* input, int thrid )
    sph_gost512 ( &ctx.gost, (const void*) hash, 64 );
    sph_gost512_close( &ctx.gost, (void*) hash );
 
-   sph_sha256_init( &ctx.sha256 );
-   sph_sha256( &ctx.sha256, hash, 64 );
-   sph_sha256_close( &ctx.sha256, hash );
+   sha256_full( hash, hash, 64 );
 
    memcpy( output, hash, 32 );
 

algo/x17/sonoa-gate.c

@@ -12,7 +12,7 @@ bool register_sonoa_algo( algo_gate_t* gate )
   init_sonoa_ctx();
   gate->hash      = (void*)&sonoa_hash;
 #endif
-  gate->optimizations = SSE2_OPT | AES_OPT | AVX2_OPT | AVX512_OPT | VAES_OPT | VAES256_OPT;
+  gate->optimizations = SSE2_OPT | AES_OPT | AVX2_OPT | AVX512_OPT | VAES_OPT;
   return true;
 };
 

algo/x17/x17-4way.c

@@ -37,7 +37,8 @@ union _x17_8way_context_overlay
     jh512_8way_context      jh;
     keccak512_8way_context  keccak;
     luffa_4way_context      luffa;
-    cube_4way_context       cube;
+//    cube_4way_context       cube;
+    cube_4way_2buf_context   cube;
 #if defined(__VAES__)
     groestl512_4way_context groestl;
     shavite512_4way_context shavite;
@@ -119,8 +120,10 @@ int x17_8way_hash( void *state, const void *input, int thr_id )
      luffa512_4way_full( &ctx.luffa, vhashA, vhashA, 64 );
      luffa512_4way_full( &ctx.luffa, vhashB, vhashB, 64 );
 
-     cube_4way_full( &ctx.cube, vhashA, 512, vhashA, 64 );
-     cube_4way_full( &ctx.cube, vhashB, 512, vhashB, 64 );
+     cube_4way_2buf_full( &ctx.cube, vhashA, vhashB, 512, vhashA, vhashB, 64 );
+     
+//     cube_4way_full( &ctx.cube, vhashA, 512, vhashA, 64 );
+//     cube_4way_full( &ctx.cube, vhashB, 512, vhashB, 64 );
 
 #if defined(__VAES__)
 

algo/x17/x17-gate.c

@@ -11,7 +11,7 @@ bool register_x17_algo( algo_gate_t* gate )
 #else
   gate->hash      = (void*)&x17_hash;
 #endif
-  gate->optimizations = SSE2_OPT | AES_OPT | AVX2_OPT | AVX512_OPT | VAES_OPT | VAES256_OPT;
+  gate->optimizations = SSE2_OPT | AES_OPT | AVX2_OPT | AVX512_OPT | VAES_OPT;
   return true;
 };
 

algo/x17/xevan-gate.c

@@ -12,7 +12,7 @@ bool register_xevan_algo( algo_gate_t* gate )
   init_xevan_ctx();
   gate->hash      = (void*)&xevan_hash;
 #endif
-  gate->optimizations = SSE2_OPT | AES_OPT | AVX2_OPT | AVX512_OPT | VAES_OPT | VAES256_OPT;
+  gate->optimizations = SSE2_OPT | AES_OPT | AVX2_OPT | AVX512_OPT | VAES_OPT;
   opt_target_factor = 256.0;
   return true;
 };

algo/x22/x22i-4way.c

@@ -28,7 +28,7 @@
   #include "algo/echo/echo-hash-4way.h"
 #endif
 #if defined(__SHA__)
-  #include "algo/sha/sph_sha2.h"
+  #include "algo/sha/sha256-hash.h"
 #endif
 
 #if defined(X22I_8WAY)
@@ -51,9 +51,7 @@ union _x22i_8way_ctx_overlay
     haval256_5_8way_context haval;
     sph_tiger_context       tiger;
     sph_gost512_context     gost;
-#if defined(X22I_8WAY_SHA)
-    sph_sha256_context      sha256;
-#else
+#if !defined(X22I_8WAY_SHA)
     sha256_8way_context     sha256;
 #endif
 #if defined(__VAES__)
@@ -391,30 +389,14 @@ int x22i_8way_hash( void *output, const void *input, int thrid )
 
 #if defined(X22I_8WAY_SHA)
 
-   sph_sha256_init( &ctx.sha256 );
-   sph_sha256( &ctx.sha256, hash0, 64 );
-   sph_sha256_close( &ctx.sha256, output );
-   sph_sha256_init( &ctx.sha256 );
-   sph_sha256( &ctx.sha256, hash1, 64 );
-   sph_sha256_close( &ctx.sha256, output+32 );
-   sph_sha256_init( &ctx.sha256 );
-   sph_sha256( &ctx.sha256, hash2, 64 );
-   sph_sha256_close( &ctx.sha256, output+64 );
-   sph_sha256_init( &ctx.sha256 );
-   sph_sha256( &ctx.sha256, hash3, 64 );
-   sph_sha256_close( &ctx.sha256, output+96 );
-   sph_sha256_init( &ctx.sha256 );
-   sph_sha256( &ctx.sha256, hash4, 64 );
-   sph_sha256_close( &ctx.sha256, output+128 );
-   sph_sha256_init( &ctx.sha256 );
-   sph_sha256( &ctx.sha256, hash5, 64 );
-   sph_sha256_close( &ctx.sha256, output+160 );
-   sph_sha256_init( &ctx.sha256 );
-   sph_sha256( &ctx.sha256, hash6, 64 );
-   sph_sha256_close( &ctx.sha256, output+192 );
-   sph_sha256_init( &ctx.sha256 );
-   sph_sha256( &ctx.sha256, hash7, 64 );
-   sph_sha256_close( &ctx.sha256, output+224 );
+   sha256_full( hash0, hash0, 64 );
+   sha256_full( hash1, hash1, 64 );
+   sha256_full( hash2, hash2, 64 );
+   sha256_full( hash3, hash3, 64 );
+   sha256_full( hash4, hash4, 64 );
+   sha256_full( hash5, hash5, 64 );
+   sha256_full( hash6, hash6, 64 );
+   sha256_full( hash7, hash7, 64 );
    
 #else
 
@@ -551,9 +533,7 @@ union _x22i_4way_ctx_overlay
     haval256_5_4way_context haval;
     sph_tiger_context       tiger;
     sph_gost512_context     gost;
-#if defined(X22I_4WAY_SHA)
-    sph_sha256_context      sha256;
-#else
+#if !defined(X22I_4WAY_SHA)
     sha256_4way_context     sha256;
 #endif
 };
@@ -757,18 +737,10 @@ int x22i_4way_hash( void *output, const void *input, int thrid )
 
 #if defined(X22I_4WAY_SHA)
 
-   sph_sha256_init( &ctx.sha256 );
-   sph_sha256( &ctx.sha256, hash0, 64 );
-   sph_sha256_close( &ctx.sha256, output );
-   sph_sha256_init( &ctx.sha256 );
-   sph_sha256( &ctx.sha256, hash1, 64 );
-   sph_sha256_close( &ctx.sha256, output+32 );
-   sph_sha256_init( &ctx.sha256 );
-   sph_sha256( &ctx.sha256, hash2, 64 );
-   sph_sha256_close( &ctx.sha256, output+64 );
-   sph_sha256_init( &ctx.sha256 );
-   sph_sha256( &ctx.sha256, hash3, 64 );
-   sph_sha256_close( &ctx.sha256, output+96 );
+   sha256_full( hash0, hash0, 64 );
+   sha256_full( hash1, hash1, 64 );
+   sha256_full( hash2, hash2, 64 );
+   sha256_full( hash3, hash3, 64 );
 
 #else
 

algo/x22/x22i-gate.c

@@ -31,8 +31,8 @@ bool register_x22i_algo( algo_gate_t* gate )
 
 #endif
 
-  gate->optimizations = SSE2_OPT | AES_OPT | AVX2_OPT | SHA_OPT
-                      | AVX512_OPT | VAES_OPT | VAES256_OPT;
+  gate->optimizations = SSE2_OPT | SSE42_OPT | AES_OPT | AVX2_OPT | SHA_OPT
+                      | AVX512_OPT | VAES_OPT;
   return true;
 };
 
@@ -48,8 +48,8 @@ bool register_x25x_algo( algo_gate_t* gate )
   gate->scanhash  = (void*)&scanhash_x25x;
   gate->hash      = (void*)&x25x_hash;
 #endif
-  gate->optimizations = SSE2_OPT | AES_OPT | AVX2_OPT | SHA_OPT |
-                        AVX512_OPT | VAES_OPT | VAES256_OPT;
+  gate->optimizations = SSE2_OPT | SSE42_OPT | AES_OPT | AVX2_OPT | SHA_OPT |
+                        AVX512_OPT | VAES_OPT;
   return true;
 };
 

algo/x22/x22i.c

@@ -24,6 +24,7 @@
 #include "algo/shabal/sph_shabal.h"
 #include "algo/whirlpool/sph_whirlpool.h"
 #include "algo/sha/sph_sha2.h"
+#include "algo/sha/sha256-hash.h"
 #include "algo/haval/sph-haval.h"
 #include "algo/tiger/sph_tiger.h"
 #include "algo/lyra2/lyra2.h"
@@ -57,7 +58,6 @@ union _x22i_context_overlay
         sph_haval256_5_context  haval;
         sph_tiger_context       tiger;
         sph_gost512_context     gost;
-        sph_sha256_context      sha256;
 };
 typedef union _x22i_context_overlay x22i_context_overlay;
 
@@ -172,9 +172,7 @@ int x22i_hash( void *output, const void *input, int thrid )
    sph_gost512 (&ctx.gost, (const void*) hash, 64);
    sph_gost512_close(&ctx.gost, (void*) hash);
 
-   sph_sha256_init( &ctx.sha256 );
-   sph_sha256( &ctx.sha256, hash, 64 );
-   sph_sha256_close( &ctx.sha256, hash );
+   sha256_full( hash, hash, 64 );
 
    memcpy(output, hash, 32);
 

algo/x22/x25x-4way.c

@@ -33,7 +33,7 @@
   #include "algo/echo/echo-hash-4way.h"
 #endif
 #if defined(__SHA__)
-  #include "algo/sha/sph_sha2.h"
+  #include "algo/sha/sha256-hash.h"
 #endif
 
 void x25x_shuffle( void *hash )
@@ -84,7 +84,7 @@ union _x25x_8way_ctx_overlay
     sph_tiger_context       tiger;
     sph_gost512_context     gost;
 #if defined(X25X_8WAY_SHA)
-    sph_sha256_context      sha256;
+    sha256_context          sha256;
 #else
     sha256_8way_context     sha256;
 #endif
@@ -447,31 +447,15 @@ int x25x_8way_hash( void *output, const void *input, int thrid )
 
 #if defined(X25X_8WAY_SHA)
 
-   sph_sha256_init( &ctx.sha256 );
-   sph_sha256( &ctx.sha256, hash0[20], 64 );
-   sph_sha256_close( &ctx.sha256, hash0[21] );
-   sph_sha256_init( &ctx.sha256 );
-   sph_sha256( &ctx.sha256, hash1[20], 64 );
-   sph_sha256_close( &ctx.sha256, hash1[21] );
-   sph_sha256_init( &ctx.sha256 );
-   sph_sha256( &ctx.sha256, hash2[20], 64 );
-   sph_sha256_close( &ctx.sha256, hash2[21] );
-   sph_sha256_init( &ctx.sha256 );
-   sph_sha256( &ctx.sha256, hash3[20], 64 );
-   sph_sha256_close( &ctx.sha256, hash3[21] );
-   sph_sha256_init( &ctx.sha256 );
-   sph_sha256( &ctx.sha256, hash4[20], 64 );
-   sph_sha256_close( &ctx.sha256, hash4[21] );
-   sph_sha256_init( &ctx.sha256 );
-   sph_sha256( &ctx.sha256, hash5[20], 64 );
-   sph_sha256_close( &ctx.sha256, hash5[21] );
-   sph_sha256_init( &ctx.sha256 );
-   sph_sha256( &ctx.sha256, hash6[20], 64 );
-   sph_sha256_close( &ctx.sha256, hash6[21] );
-   sph_sha256_init( &ctx.sha256 );
-   sph_sha256( &ctx.sha256, hash7[20], 64 );
-   sph_sha256_close( &ctx.sha256, hash7[21] );
-
+   sha256_full( hash0[21], hash0[20], 64 );
+   sha256_full( hash1[21], hash1[20], 64 );
+   sha256_full( hash2[21], hash2[20], 64 );
+   sha256_full( hash3[21], hash3[20], 64 );
+   sha256_full( hash4[21], hash4[20], 64 );
+   sha256_full( hash5[21], hash5[20], 64 );
+   sha256_full( hash6[21], hash6[20], 64 );
+   sha256_full( hash7[21], hash7[20], 64 );
+   
    intrlv_8x32_512( vhash, hash0[21], hash1[21], hash2[21], hash3[21],
                            hash4[21], hash5[21], hash6[21], hash7[21] );
    
@@ -646,7 +630,7 @@ union _x25x_4way_ctx_overlay
     sph_tiger_context       tiger;
     sph_gost512_context     gost;
 #if defined(X25X_4WAY_SHA)
-    sph_sha256_context      sha256;
+    sha256_context          sha256;
 #else
     sha256_4way_context     sha256;
 #endif
@@ -848,18 +832,10 @@ int x25x_4way_hash( void *output, const void *input, int thrid )
 
 #if defined(X25X_4WAY_SHA)
 
-   sph_sha256_init( &ctx.sha256 );
-   sph_sha256( &ctx.sha256, hash0[20], 64 );
-   sph_sha256_close( &ctx.sha256, hash0[21] );
-   sph_sha256_init( &ctx.sha256 );
-   sph_sha256( &ctx.sha256, hash1[20], 64 );
-   sph_sha256_close( &ctx.sha256, hash1[21] );
-   sph_sha256_init( &ctx.sha256 );
-   sph_sha256( &ctx.sha256, hash2[20], 64 );
-   sph_sha256_close( &ctx.sha256, hash2[21] );
-   sph_sha256_init( &ctx.sha256 );
-   sph_sha256( &ctx.sha256, hash3[20], 64 );
-   sph_sha256_close( &ctx.sha256, hash3[21] );
+   sha256_full( hash0[21], hash0[20], 64 );
+   sha256_full( hash1[21], hash1[20], 64 );
+   sha256_full( hash2[21], hash2[20], 64 );
+   sha256_full( hash3[21], hash3[20], 64 );
 
    intrlv_4x32_512( vhash, hash0[21], hash1[21], hash2[21], hash3[21] );
 

algo/x22/x25x.c

@@ -23,7 +23,7 @@
 #include "algo/hamsi/sph_hamsi.h"
 #include "algo/shabal/sph_shabal.h"
 #include "algo/whirlpool/sph_whirlpool.h"
-#include "algo/sha/sph_sha2.h"
+#include "algo/sha/sha256-hash.h"
 #include "algo/haval/sph-haval.h"
 #include "algo/tiger/sph_tiger.h"
 #include "algo/lyra2/lyra2.h"
@@ -60,7 +60,7 @@ union _x25x_context_overlay
         sph_haval256_5_context  haval;
         sph_tiger_context       tiger;
         sph_gost512_context     gost;
-        sph_sha256_context      sha256;
+        sha256_context          sha256;
         sph_panama_context      panama;
         blake2s_state           blake2s;
 };
@@ -174,9 +174,7 @@ int x25x_hash( void *output, const void *input, int thrid )
    sph_gost512 (&ctx.gost, (const void*) &hash[19], 64);
    sph_gost512_close(&ctx.gost, (void*) &hash[20]);
 
-   sph_sha256_init( &ctx.sha256 );
-   sph_sha256( &ctx.sha256, &hash[20], 64 );
-   sph_sha256_close( &ctx.sha256, &hash[21] );
+   sha256_full( &hash[21], &hash[20], 64 );
 
    sph_panama_init(&ctx.panama);
    sph_panama (&ctx.panama, (const void*) &hash[21], 64 );

algo/yespower/crypto/blake2b-yp.c

@@ -35,9 +35,11 @@
 #include "blake2b-yp.h"
 
 // Cyclic right rotation.
-#ifndef ROTR64
-#define ROTR64(x, y)  (((x) >> (y)) ^ ((x) << (64 - (y))))
-#endif
+//#ifndef ROTR64
+//#define ROTR64(x, y)  (((x) >> (y)) ^ ((x) << (64 - (y))))
+//#endif
+
+#define ROTR64(x, y) ror64( x, y )
 
 // Little-endian byte access.
 #define B2B_GET64(p)                            \

algo/yespower/yescrypt-r8g.c

@@ -52,8 +52,8 @@ int scanhash_yespower_r8g( struct work *work, uint32_t max_nonce,
     endiandata[19] = n;
 
 // do sha256 prehash
-   sph_sha256_init( &sha256_prehash_ctx );
-   sph_sha256( &sha256_prehash_ctx, endiandata, 64 );
+   sha256_ctx_init( &sha256_prehash_ctx );
+   sha256_update( &sha256_prehash_ctx, endiandata, 64 );
     
     do {
        yespower_tls( (unsigned char *)endiandata, params.perslen,

algo/yespower/yespower-gate.c

@@ -27,14 +27,11 @@
  * coin.
  */
 #include "yespower.h"
-
 #include "algo-gate-api.h"
 
 yespower_params_t yespower_params;
 
-//SHA256_CTX sha256_prehash_ctx;
-__thread sph_sha256_context sha256_prehash_ctx;
-//__thread SHA256_CTX sha256_prehash_ctx;
+__thread sha256_context sha256_prehash_ctx;
 
 // YESPOWER
 
@@ -61,8 +58,8 @@ int scanhash_yespower( struct work *work, uint32_t max_nonce,
    endiandata[19] = n;
 
    // do sha256 prehash
-   sph_sha256_init( &sha256_prehash_ctx );
-   sph_sha256( &sha256_prehash_ctx, endiandata, 64 );
+   sha256_ctx_init( &sha256_prehash_ctx );
+   sha256_update( &sha256_prehash_ctx, endiandata, 64 );
 
    do {
       if ( yespower_hash( (char*)endiandata, (char*)vhash, 80, thr_id ) )
@@ -101,10 +98,6 @@ int scanhash_yespower_b2b( struct work *work, uint32_t max_nonce,
       be32enc( &endiandata[k], pdata[k] );
    endiandata[19] = n;
 
-   // do sha256 prehash
-   sph_sha256_init( &sha256_prehash_ctx );
-   sph_sha256( &sha256_prehash_ctx, endiandata, 64 );
-
    do {
       if (yespower_b2b_hash( (char*) endiandata, (char*) vhash, 80, thr_id ) )
       if unlikely( valid_hash( vhash, ptarget ) && !opt_benchmark )

algo/yespower/yespower-opt.c

@@ -203,17 +203,17 @@ static inline void salsa20_simd_unshuffle(const salsa20_blk_t *Bin,
 	ARX(X0, X3, X2, 18) \
 	/* Rearrange data */ \
 	X1 = _mm_shuffle_epi32(X1, 0x93); \
+   X3 = _mm_shuffle_epi32(X3, 0x39); \
 	X2 = _mm_shuffle_epi32(X2, 0x4E); \
-	X3 = _mm_shuffle_epi32(X3, 0x39); \
 	/* Operate on "rows" */ \
 	ARX(X3, X0, X1, 7) \
 	ARX(X2, X3, X0, 9) \
 	ARX(X1, X2, X3, 13) \
 	ARX(X0, X1, X2, 18) \
 	/* Rearrange data */ \
+   X3 = _mm_shuffle_epi32(X3, 0x93); \
 	X1 = _mm_shuffle_epi32(X1, 0x39); \
-	X2 = _mm_shuffle_epi32(X2, 0x4E); \
-	X3 = _mm_shuffle_epi32(X3, 0x93);
+	X2 = _mm_shuffle_epi32(X2, 0x4E);
 
 /**
  * Apply the Salsa20 core to the block provided in (X0 ... X3).
@@ -1095,7 +1095,7 @@ int yespower(yespower_local_t *local,
    salsa20_blk_t *V, *XY;
    pwxform_ctx_t ctx;
    uint8_t sha256[32];
-   sph_sha256_context sha256_ctx;
+   sha256_context sha256_ctx;
 
    /* Sanity-check parameters */
    if ( (version != YESPOWER_0_5 && version != YESPOWER_1_0)
@@ -1138,10 +1138,9 @@ int yespower(yespower_local_t *local,
 
    // copy prehash, do tail   
    memcpy( &sha256_ctx, &sha256_prehash_ctx, sizeof sha256_ctx );
-
-   sph_sha256( &sha256_ctx, src+64, srclen-64 );
-   sph_sha256_close( &sha256_ctx, sha256 );
-
+   sha256_update( &sha256_ctx, src+64, srclen-64 );
+   sha256_final( &sha256_ctx, sha256 );
+   
    if ( version == YESPOWER_0_5 )
    {
       PBKDF2_SHA256( sha256, sizeof(sha256), src, srclen, 1, B, B_size );
@@ -1186,7 +1185,9 @@ int yespower(yespower_local_t *local,
       if ( work_restart[thrid].restart ) return 0;
 
       smix_1_0( B, r, N, V, XY, &ctx );
-      
+
+      if ( work_restart[thrid].restart ) return 0;
+
       HMAC_SHA256_Buf( B + B_size - 64, 64, sha256, sizeof(sha256),
                        (uint8_t *)dst );
    }

algo/yespower/yespower.h

@@ -34,7 +34,7 @@
 #include <stdlib.h> /* for size_t */
 #include "miner.h"
 #include "simd-utils.h"
-#include "algo/sha/sph_sha2.h"
+#include "algo/sha/sha256-hash.h"
 
 #ifdef __cplusplus
 extern "C" {
@@ -78,9 +78,7 @@ typedef struct {
 
 extern yespower_params_t yespower_params;
 
-//SHA256_CTX sha256_prehash_ctx;
-extern __thread sph_sha256_context sha256_prehash_ctx;
-//extern __thread SHA256_CTX sha256_prehash_ctx;
+extern __thread sha256_context sha256_prehash_ctx;
 
 /**
  * yespower_init_local(local):

build-allarch.sh

@@ -4,128 +4,97 @@
 # during develpment. However the information contained may provide compilation
 # tips to users.
 
-rm cpuminer-avx512-sha-vaes cpuminer-avx512-sha cpuminer-avx512 cpuminer-avx2 cpuminer-aes-avx cpuminer-aes-sse42 cpuminer-sse42 cpuminer-ssse3 cpuminer-sse2 cpuminer-zen cpuminer-zen3  > /dev/null
+rm cpuminer-avx512-sha-vaes cpuminer-avx512 cpuminer-avx2 cpuminer-avx cpuminer-aes-sse42 cpuminer-sse42 cpuminer-ssse3 cpuminer-sse2 cpuminer-zen cpuminer-zen3  > /dev/null
 
-# Icelake AVX512 SHA VAES
+# AVX512 SHA VAES: Intel Core Icelake, Rocketlake
 make distclean || echo clean
 rm -f config.status
 ./autogen.sh || echo done
 CFLAGS="-O3 -march=icelake-client -Wall -fno-common" ./configure --with-curl
+#CFLAGS="-O3 -march=rocketlake -Wall -fno-common" ./configure --with-curl
 make -j 8
-strip -s cpuminer.exe
-mv cpuminer.exe cpuminer-avx512-sha-vaes.exe
 strip -s cpuminer
 mv cpuminer cpuminer-avx512-sha-vaes
 
-# Rocketlake AVX512 SHA AES
+# AVX512 AES: Intel Core HEDT Sylake-X, Cascadelake
 make clean || echo clean
 rm -f config.status
-CFLAGS="-O3 -march=cascadelake -msha -Wall -fno-common" ./configure --with-curl
-#CFLAGS="-O3 -march=skylake-avx512 -msha -Wall -fno-common" ./configure --with-curl
-# CFLAGS="-O3 -march=rocketlake -Wall -fno-common" ./configure --with-curl
+CFLAGS="-O3 -march=skylake-avx512 -maes -Wall -fno-common" ./configure --with-curl
 make -j 8
-strip -s cpuminer.exe
-mv cpuminer.exe cpuminer-avx512-sha.exe
-strip -s cpuminer
-mv cpuminer cpuminer-avx512-sha
-
-# Slylake-X AVX512 AES
-make clean || echo clean
-rm -f config.status
-CFLAGS="-O3 -march=skylake-avx512 -Wall -fno-common" ./configure --with-curl
-make -j 8
-strip -s cpuminer.exe
-mv cpuminer.exe cpuminer-avx512.exe
 strip -s cpuminer
 mv cpuminer cpuminer-avx512
 
-# Haswell AVX2 AES
+# AVX2 SHA VAES: Intel Alderlake, AMD Zen3
+make clean || echo done
+rm -f config.status
+# vaes doesn't include aes
+CFLAGS="-O3 -maes -mavx2 -msha -mvaes -Wall -fno-common" ./configure --with-curl
+make -j 8
+strip -s cpuminer
+mv cpuminer cpuminer-avx2-sha-vaes
+
+# AVX2 SHA AES: AMD Zen1
+make clean || echo done
+rm -f config.status
+CFLAGS="-O3 -march=znver1 -maes -Wall -fno-common" ./configure --with-curl
+#CFLAGS="-O3 -maes -mavx2 -msha -Wall -fno-common" ./configure --with-curl
+make -j 8
+strip -s cpuminer
+mv cpuminer cpuminer-avx2-sha
+
+# AVX2 AES: Intel Haswell..Cometlake
 make clean || echo clean
 rm -f config.status
 # GCC 9 doesn't include AES with core-avx2
 CFLAGS="-O3 -march=core-avx2 -maes -Wall -fno-common" ./configure --with-curl
 make -j 8
-strip -s cpuminer.exe
-mv cpuminer.exe cpuminer-avx2.exe
 strip -s cpuminer
 mv cpuminer cpuminer-avx2
 
-# Sandybridge AVX AES
+# AVX AES: Intel Sandybridge, Ivybridge
 make clean || echo clean
 rm -f config.status
 CFLAGS="-O3 -march=corei7-avx -maes -Wall -fno-common" ./configure --with-curl
 make -j 8
-strip -s cpuminer.exe
-mv cpuminer.exe cpuminer-avx.exe
 strip -s cpuminer
 mv cpuminer cpuminer-avx
 
-# Westmere SSE4.2 AES
+# SSE4.2 AES: Intel Westmere
 make clean || echo clean
 rm -f config.status
 CFLAGS="-O3 -march=westmere -maes -Wall -fno-common" ./configure --with-curl
 make -j 8
-strip -s cpuminer.exe
-mv cpuminer.exe cpuminer-aes-sse42.exe
 strip -s cpuminer
 mv cpuminer cpuminer-aes-sse42
 
-# Nehalem SSE4.2
+# SSE4.2: Intel Nehalem
 make clean || echo clean
 rm -f config.status
 CFLAGS="-O3 -march=corei7 -Wall -fno-common" ./configure --with-curl
 make -j 8
-strip -s cpuminer.exe
-mv cpuminer.exe cpuminer-sse42.exe
 strip -s cpuminer
 mv cpuminer cpuminer-sse42
 
-# Core2 SSSE3
+# SSSE3: Intel Core2
 make clean || echo clean
 rm -f config.status
 CFLAGS="-O3 -march=core2 -Wall -fno-common" ./configure --with-curl
 make -j 8
-strip -s cpuminer.exe
-mv cpuminer.exe cpuminer-ssse3.exe
 strip -s cpuminer
 mv cpuminer cpuminer-ssse3
 
-# Generic SSE2
+# SSE2
 make clean || echo clean
 rm -f config.status
 CFLAGS="-O3 -msse2 -Wall -fno-common" ./configure --with-curl
 make -j 8
-strip -s cpuminer.exe
-mv cpuminer.exe cpuminer-sse2.exe
 strip -s cpuminer
 mv cpuminer cpuminer-sse2
 
-# AMD Zen1 AVX2 SHA
-make clean || echo done
-rm -f config.status
-CFLAGS="-O3 -march=znver1 -Wall -fno-common" ./configure --with-curl
-make -j 8
-strip -s cpuminer.exe
-mv cpuminer.exe cpuminer-zen.exe
-strip -s cpuminer
-mv cpuminer cpuminer-zen
-
-# AMD Zen3 AVX2 SHA VAES
-make clean || echo done
-rm -f config.status
-CFLAGS="-O3 -march=znver2 -mvaes -Wall -fno-common" ./configure --with-curl
-# CFLAGS="-O3 -march=znver3 -Wall -fno-common" ./configure --with-curl
-make -j 8
-strip -s cpuminer.exe
-mv cpuminer.exe cpuminer-zen3.exe
-strip -s cpuminer
-mv cpuminer cpuminer-zen3
-
-# Native to current CPU
+# Native to host CPU
 make clean || echo done
 rm -f config.status
 CFLAGS="-O3 -march=native -Wall -fno-common" ./configure --with-curl
 make -j 8
-strip -s cpuminer.exe
 strip -s cpuminer
 

clean-all.sh

@@ -2,8 +2,8 @@
 #
 # make clean and rm all the targetted executables.
 
-rm cpuminer-avx512-sha-vaes cpuminer-avx512-sha cpuminer-avx512 cpuminer-avx2 cpuminer-avx cpuminer-aes-sse42 cpuminer-sse2 cpuminer-zen cpuminer-sse42 cpuminer-ssse3 cpuminer-zen3 > /dev/null
+rm cpuminer-avx512-sha-vaes cpuminer-avx512 cpuminer-avx2 cpuminer-avx cpuminer-aes-sse42 cpuminer-sse2 cpuminer-avx2-sha cpuminer-sse42 cpuminer-ssse3 cpuminer-avx2-sha-vaes > /dev/null
 
-rm cpuminer-avx512-sha-vaes.exe cpuminer-avx512-sha.exe cpuminer-avx512.exe cpuminer-avx2.exe cpuminer-avx.exe cpuminer-aes-sse42.exe cpuminer-sse2.exe cpuminer-zen.exe  cpuminer-sse42.exe cpuminer-ssse3.exe cpuminer-zen3.exe > /dev/null
+rm cpuminer-avx512-sha-vaes.exe cpuminer-avx512-sha.exe cpuminer-avx512.exe cpuminer-avx2.exe cpuminer-avx.exe cpuminer-aes-sse42.exe cpuminer-sse2.exe cpuminer-avx2-sha.exe cpuminer-sse42.exe cpuminer-ssse3.exe cpuminer-avx2-sha-vaes.exe > /dev/null
 
 make distclean > /dev/null

compat.h

@@ -3,6 +3,10 @@
 
 #ifdef WIN32
 
+#if _WIN32_WINNT==0x0601    // Windows 7
+ #define WINDOWS_CPU_GROUPS_ENABLED 1
+#endif
+
 #include <windows.h>
 #include <time.h>
 

configure

@@ -1,6 +1,6 @@
 #! /bin/sh
 # Guess values for system-dependent variables and create Makefiles.
-# Generated by GNU Autoconf 2.69 for cpuminer-opt 3.17.1.
+# Generated by GNU Autoconf 2.69 for cpuminer-opt 3.19.1.
 #
 #
 # Copyright (C) 1992-1996, 1998-2012 Free Software Foundation, Inc.
@@ -577,8 +577,8 @@ MAKEFLAGS=
 # Identity of this package.
 PACKAGE_NAME='cpuminer-opt'
 PACKAGE_TARNAME='cpuminer-opt'
-PACKAGE_VERSION='3.17.1'
-PACKAGE_STRING='cpuminer-opt 3.17.1'
+PACKAGE_VERSION='3.19.1'
+PACKAGE_STRING='cpuminer-opt 3.19.1'
 PACKAGE_BUGREPORT=''
 PACKAGE_URL=''
 
@@ -1332,7 +1332,7 @@ if test "$ac_init_help" = "long"; then
   # Omit some internal or obsolete options to make the list less imposing.
   # This message is too long to be a string in the A/UX 3.1 sh.
   cat <<_ACEOF
-\`configure' configures cpuminer-opt 3.17.1 to adapt to many kinds of systems.
+\`configure' configures cpuminer-opt 3.19.1 to adapt to many kinds of systems.
 
 Usage: $0 [OPTION]... [VAR=VALUE]...
 
@@ -1404,7 +1404,7 @@ fi
 
 if test -n "$ac_init_help"; then
   case $ac_init_help in
-     short | recursive ) echo "Configuration of cpuminer-opt 3.17.1:";;
+     short | recursive ) echo "Configuration of cpuminer-opt 3.19.1:";;
    esac
   cat <<\_ACEOF
 
@@ -1509,7 +1509,7 @@ fi
 test -n "$ac_init_help" && exit $ac_status
 if $ac_init_version; then
   cat <<\_ACEOF
-cpuminer-opt configure 3.17.1
+cpuminer-opt configure 3.19.1
 generated by GNU Autoconf 2.69
 
 Copyright (C) 2012 Free Software Foundation, Inc.
@@ -2012,7 +2012,7 @@ cat >config.log <<_ACEOF
 This file contains any messages produced by compilers while
 running configure, to aid debugging if configure makes a mistake.
 
-It was created by cpuminer-opt $as_me 3.17.1, which was
+It was created by cpuminer-opt $as_me 3.19.1, which was
 generated by GNU Autoconf 2.69.  Invocation command line was
 
   $ $0 $@
@@ -2993,7 +2993,7 @@ fi
 
 # Define the identity of the package.
  PACKAGE='cpuminer-opt'
- VERSION='3.17.1'
+ VERSION='3.19.1'
 
 
 cat >>confdefs.h <<_ACEOF
@@ -6690,7 +6690,7 @@ cat >>$CONFIG_STATUS <<\_ACEOF || ac_write_fail=1
 # report actual input values of CONFIG_FILES etc. instead of their
 # values after options handling.
 ac_log="
-This file was extended by cpuminer-opt $as_me 3.17.1, which was
+This file was extended by cpuminer-opt $as_me 3.19.1, which was
 generated by GNU Autoconf 2.69.  Invocation command line was
 
   CONFIG_FILES    = $CONFIG_FILES
@@ -6756,7 +6756,7 @@ _ACEOF
 cat >>$CONFIG_STATUS <<_ACEOF || ac_write_fail=1
 ac_cs_config="`$as_echo "$ac_configure_args" | sed 's/^ //; s/[\\""\`\$]/\\\\&/g'`"
 ac_cs_version="\\
-cpuminer-opt config.status 3.17.1
+cpuminer-opt config.status 3.19.1
 configured by $0, generated by GNU Autoconf 2.69,
   with options \\"\$ac_cs_config\\"
 

configure.ac

@@ -1,4 +1,4 @@
-AC_INIT([cpuminer-opt], [3.17.1])
+AC_INIT([cpuminer-opt], [3.19.1])
 
 AC_PREREQ([2.59c])
 AC_CANONICAL_SYSTEM

miner.h

@@ -70,17 +70,25 @@ void *alloca (size_t);
 
 #ifdef HAVE_SYSLOG_H
 #include <syslog.h>
-#define LOG_BLUE 0x10 /* unique value */
+#define LOG_BLUE  0x10 /* unique value */
+#define LOG_MAJR  0x11 /* unique value */
+#define LOG_MINR  0x12 /* unique value */
+#define LOG_GREEN 0x13 /* unique value */
+#define LOG_PINK  0x14 /* unique value */
 #else
 enum {
-	LOG_ERR,
+   LOG_CRIT,
+   LOG_ERR,
 	LOG_WARNING,
 	LOG_NOTICE,
 	LOG_INFO,
 	LOG_DEBUG,
-	/* custom notices */
-	LOG_BLUE = 0x10,
-};
+   /* custom notices */
+	LOG_BLUE  = 0x10,
+   LOG_MAJR  = 0x11,
+   LOG_MINR  = 0x12,
+   LOG_GREEN = 0x13,
+   LOG_PINK  = 0x14 };
 #endif
 
 extern bool is_power_of_2( int n );
@@ -216,7 +224,7 @@ json_t* json_load_url(char* cfg_url, json_error_t *err);
 
 void sha256_init(uint32_t *state);
 void sha256_transform(uint32_t *state, const uint32_t *block, int swap);
-void sha256d(unsigned char *hash, const unsigned char *data, int len);
+//void sha256d(unsigned char *hash, const unsigned char *data, int len);
 
 #ifdef USE_ASM
 #if defined(__ARM_NEON__) || defined(__i386__) || defined(__x86_64__)
@@ -225,7 +233,8 @@ int sha256_use_4way();
 void sha256_init_4way(uint32_t *state);
 void sha256_transform_4way(uint32_t *state, const uint32_t *block, int swap);
 #endif
-#if defined(__x86_64__) && defined(USE_AVX2)
+//#if defined(__x86_64__) && defined(USE_AVX2)
+#if defined(__x86_64__) && defined(__AVX2__)
 #define HAVE_SHA256_8WAY 1
 int sha256_use_8way();
 void sha256_init_8way(uint32_t *state);
@@ -271,9 +280,9 @@ struct thr_api {
 #define CL_N    "\x1B[0m"
 #define CL_RED  "\x1B[31m"
 #define CL_GRN  "\x1B[32m"
-#define CL_YLW  "\x1B[33m"
+#define CL_YLW  "\x1B[33m"  // dark yellow
 #define CL_BLU  "\x1B[34m"
-#define CL_MAG  "\x1B[35m"
+#define CL_MAG  "\x1B[35m"  // purple
 #define CL_CYN  "\x1B[36m"
 
 #define CL_BLK  "\x1B[22;30m" /* black */
@@ -281,7 +290,7 @@ struct thr_api {
 #define CL_GR2  "\x1B[22;32m" /* green */
 #define CL_BRW  "\x1B[22;33m" /* brown */
 #define CL_BL2  "\x1B[22;34m" /* blue */
-#define CL_MA2  "\x1B[22;35m" /* magenta */
+#define CL_MA2  "\x1B[22;35m" /* purple */
 #define CL_CY2  "\x1B[22;36m" /* cyan */
 #define CL_SIL  "\x1B[22;37m" /* gray */
 
@@ -290,9 +299,9 @@ struct thr_api {
 #else
 #define CL_GRY  "\x1B[90m"    /* dark gray selectable in putty */
 #endif
-#define CL_LRD  "\x1B[01;31m" /* light red */
-#define CL_LGR  "\x1B[01;32m" /* light green */
-#define CL_YL2  "\x1B[01;33m" /* yellow */
+#define CL_LRD  "\x1B[01;31m" /* bright red */
+#define CL_LGR  "\x1B[01;32m" /* bright green */
+#define CL_YL2  "\x1B[01;33m" /* bright yellow */
 #define CL_LBL  "\x1B[01;34m" /* light blue */
 #define CL_LMA  "\x1B[01;35m" /* light magenta */
 #define CL_LCY  "\x1B[01;36m" /* light cyan */
@@ -481,7 +490,7 @@ void format_hashrate(double hashrate, char *output);
 void print_hash_tests(void);
 
 void scale_hash_for_display ( double* hashrate, char* units );
-
+void format_number_si( double* hashrate, char* si_units );
 void report_summary_log( bool force );
 
 /*
@@ -859,9 +868,9 @@ Options:\n\
                           yespowerr16   Yenten (YTN)\n\
                           yespower-b2b  generic yespower + blake2b\n\
                           zr5           Ziftr\n\
-  -N, --param-n         N parameter for scrypt based algos\n\
-  -R, --param-r         R parameter for scrypt based algos\n\
-  -K, --param-key       Key (pers) parameter for algos that use it\n\
+  -N, --param-n=N       N parameter for scrypt based algos\n\
+  -R, --param-r=N       R parameter for scrypt based algos\n\
+  -K, --param-key=STRING  Key (pers) parameter for algos that use it\n\
   -o, --url=URL         URL of mining server\n\
   -O, --userpass=U:P    username:password pair for mining server\n\
   -u, --user=USERNAME   username for mining server\n\
@@ -877,8 +886,8 @@ Options:\n\
   -s, --scantime=N      upper bound on time spent scanning current work when\n\
                           long polling is unavailable, in seconds (default: 5)\n\
       --randomize       Randomize scan range start to reduce duplicates\n\
-  -f, --diff-factor     Divide req. difficulty by this factor (std is 1.0)\n\
-  -m, --diff-multiplier Multiply difficulty by this factor (std is 1.0)\n\
+  -f, --diff-factor=N   Divide req. difficulty by this factor (std is 1.0)\n\
+  -m, --diff-multiplier=N Multiply difficulty by this factor (std is 1.0)\n\
       --hash-meter      Display thread hash rates\n\
       --coinbase-addr=ADDR  payout address for solo mining\n\
       --coinbase-sig=TEXT  data to insert in the coinbase when possible\n\
@@ -886,9 +895,9 @@ Options:\n\
       --no-getwork      disable getwork support\n\
       --no-gbt          disable getblocktemplate support\n\
       --no-stratum      disable X-Stratum support\n\
-      --no-extranonce   disable Stratum extranonce support\n\
+      --no-extranonce   disable Stratum extranonce subscribe\n\
       --no-redirect     ignore requests to change the URL of the mining server\n\
-  -q, --quiet           disable per-thread hashmeter output\n\
+  -q, --quiet           reduce log verbosity\n\
       --no-color        disable colored output\n\
   -D, --debug           enable debug output\n\
   -P, --protocol-dump   verbose dump of protocol-level activities\n"
@@ -907,9 +916,9 @@ Options:\n\
       --max-rate=N[KMG] Only mine if net hashrate is less than specified value\n\
       --max-diff=N      Only mine if net difficulty is less than specified value\n\
   -c, --config=FILE     load a JSON-format configuration file\n\
-      --data-file       path and name of data file\n\
+      --data-file=FILE  path and name of data file\n\
       --verify          enable additional time consuming start up tests\n\
-  -V, --version         display version information and exit\n\
+  -V, --version         display version and CPU information and exit\n\
   -h, --help            display this help text and exit\n\
 ";
 

simd-utils.h

@@ -78,6 +78,8 @@
 //   - specialized shift and rotate functions that move elements around
 //     use the notation "1x32" to indicate the distance moved as units of
 //     the element size.
+//     Vector shuffle rotations are being renamed to "vrol" and "vror"
+//     to avoid confusion with bit rotations.
 //   - there is a subset of some functions for scalar data. They may have
 //     no prefix nor vec-size, just one size, the size of the data.
 //   - Some integer functions are also defined which use a similar notation.

simd-utils/intrlv.h

@@ -65,7 +65,7 @@ static inline void dintrlv_2x32( void *dst0, void *dst1,
    d0[24] = s[48];   d1[24] = s[49];   d0[25] = s[50];   d1[25] = s[51];
    d0[26] = s[52];   d1[26] = s[53];   d0[27] = s[54];   d1[27] = s[55];
    d0[28] = s[56];   d1[28] = s[57];   d0[29] = s[58];   d1[29] = s[59];
-   d0[30] = s[60];   d1[30] = s[61];   d0[31] = s[61];   d1[31] = s[63];
+   d0[30] = s[60];   d1[30] = s[61];   d0[31] = s[62];   d1[31] = s[63];
 }
 
 static inline void extr_lane_2x32( void *dst, const void *src,

simd-utils/simd-128.h

@@ -35,6 +35,13 @@
 ///////////////////////////////////////////////////////////////////////////
 
 
+// Used instead if casting.
+typedef union
+{
+   __m128i m128;
+   uint32_t u32[4];
+} __attribute__ ((aligned (16))) m128_ovly;
+
 // Efficient and convenient moving between GP & low bits of XMM.
 // Use VEX when available to give access to xmm8-15 and zero extend for
 // larger vectors.
@@ -61,7 +68,10 @@ static inline __m128i mm128_mov32_128( const uint32_t n )
   return  a;
 }
 
-static inline uint64_t mm128_mov128_64( const __m128i a )
+// Inconstant naming, prefix should reflect return value:
+// u64_mov128_64
+
+static inline uint64_t u64_mov128_64( const __m128i a )
 {
   uint64_t n;
 #if defined(__AVX__)
@@ -72,7 +82,7 @@ static inline uint64_t mm128_mov128_64( const __m128i a )
   return  n;
 }
 
-static inline uint32_t mm128_mov128_32( const __m128i a )
+static inline uint32_t u32_mov128_32( const __m128i a )
 {
   uint32_t n;
 #if defined(__AVX__)
@@ -166,12 +176,17 @@ static inline __m128i mm128_insert_32( const __m128i v, const uint32_t i,
 
 // Extract 32 bit element c from v and return as integer.
 static inline uint32_t mm128_extract_32( const __m128i v, const int c )
-{   return mm128_mov128_32( mm128_xim_32( v, v, c<<6 ) ); }
+{   return u32_mov128_32( mm128_xim_32( v, v, c<<6 ) ); }
 
 // Clear (zero) 32 bit elements based on bits set in 4 bit mask.
 static inline __m128i mm128_mask_32( const __m128i v, const int m ) 
 {   return mm128_xim_32( v, v, m ); }
 
+// Move element i2 of v2 to element i1 of v1. For reference and convenience,
+// it's faster to precalculate the index.
+#define mm128_shuflmov_32( v1, i1, v2, i2 ) \
+  mm128_xim_32( v1, v2, ( (i1)<<4 ) | ( (i2)<<6 ) )
+
 #endif  // SSE4_1
 
 //
@@ -257,12 +272,37 @@ static inline void memcpy_128( __m128i *dst, const __m128i *src, const int n )
 
 #endif
 
+
+
+// Diagonal blend: d = s3[3], s2[2], s1[1], s0[0] ||
+
+// Blend 4 32 bit elements from 4 vectors
+
+#if defined (__AVX2__)
+
+#define mm128_diagonal_32( v3, v2, v1, v0 ) \
+  mm_blend_epi32( _mm_blend_epi32( s3, s2, 0x4 ), \
+                  _mm_blend_epi32( s1, s0, 0x1 ), 0x3 )
+
+#elif defined(__SSE4_1)
+
+#define mm128_diagonal_32( v3, v2, v1, v0 ) \
+  mm_blend_epi16( _mm_blend_epi16( s3, s2, 0x30 ), \
+                  _mm_blend_epi16( s1, s0, 0x03 ), 0x0f )
+
+#endif
+
+
 //
 // Bit rotations
 
 // AVX512VL has implemented bit rotation for 128 bit vectors with
 // 64 and 32 bit elements.
 
+// x2 rotates elements in 2 individual vectors in a double buffered
+// optimization for SSE2, does nothing for AVX512 but is there for
+// transparency.
+
 // compiler doesn't like when a variable is used for the last arg of
 // _mm_rol_epi32, must be "8 bit immediate". Oddly _mm_slli has the same
 // specification but works with a variable. Therefore use rol_var where
@@ -290,6 +330,22 @@ static inline void memcpy_128( __m128i *dst, const __m128i *src, const int n )
 #define mm128_ror_32    _mm_ror_epi32
 #define mm128_rol_32    _mm_rol_epi32
 
+#define mm128_rorx2_64( v1, v0, c ) \
+   _mm_ror_epi64( v0, c ); \
+   _mm_ror_epi64( v1, c )
+
+#define mm128_rolx2_64( v1, v0, c ) \
+   _mm_rol_epi64( v0, c ); \
+   _mm_rol_epi64( v1, c )
+
+#define mm128_rorx2_32( v1, v0, c ) \
+   _mm_ror_epi32( v0, c ); \
+   _mm_ror_epi32( v1, c )
+
+#define mm128_rolx2_32( v1, v0, c ) \
+   _mm_rol_epi32( v0, c ); \
+   _mm_rol_epi32( v1, c )
+
 #else  // SSE2
 
 #define mm128_ror_64   mm128_ror_var_64
@@ -297,6 +353,46 @@ static inline void memcpy_128( __m128i *dst, const __m128i *src, const int n )
 #define mm128_ror_32   mm128_ror_var_32
 #define mm128_rol_32   mm128_rol_var_32
 
+#define mm128_rorx2_64( v1, v0, c ) \
+{ \
+ __m128i t0 = _mm_srli_epi64( v0, c ); \
+ __m128i t1 = _mm_srli_epi64( v1, c ); \
+ v0 = _mm_slli_epi64( v0, 64-(c) ); \
+ v1 = _mm_slli_epi64( v1, 64-(c) ); \
+ v0 = _mm_or_si256( v0, t0 ); \
+ v1 = _mm_or_si256( v1, t1 ); \
+}
+
+#define mm128_rolx2_64( v1, v0, c ) \
+{ \
+ __m128i t0 = _mm_slli_epi64( v0, c ); \
+ __m128i t1 = _mm_slli_epi64( v1, c ); \
+ v0 = _mm_srli_epi64( v0, 64-(c) ); \
+ v1 = _mm_srli_epi64( v1, 64-(c) ); \
+ v0 = _mm_or_si256( v0, t0 ); \
+ v1 = _mm_or_si256( v1, t1 ); \
+}
+
+#define mm128_rorx2_32( v1, v0, c ) \
+{ \
+ __m128i t0 = _mm_srli_epi32( v0, c ); \
+ __m128i t1 = _mm_srli_epi32( v1, c ); \
+ v0 = _mm_slli_epi32( v0, 32-(c) ); \
+ v1 = _mm_slli_epi32( v1, 32-(c) ); \
+ v0 = _mm_or_si256( v0, t0 ); \
+ v1 = _mm_or_si256( v1, t1 ); \
+}
+
+#define mm128_rolx2_32( v1, v0, c ) \
+{ \
+ __m128i t0 = _mm_slli_epi32( v0, c ); \
+ __m128i t1 = _mm_slli_epi32( v1, c ); \
+ v0 = _mm_srli_epi32( v0, 32-(c) ); \
+ v1 = _mm_srli_epi32( v1, 32-(c) ); \
+ v0 = _mm_or_si256( v0, t0 ); \
+ v1 = _mm_or_si256( v1, t1 ); \
+}
+
 #endif   // AVX512 else SSE2
 
 #define mm128_ror_16( v, c ) \
@@ -309,16 +405,22 @@ static inline void memcpy_128( __m128i *dst, const __m128i *src, const int n )
 // Rotate vector elements accross all lanes
 
 #define mm128_swap_64( v )    _mm_shuffle_epi32( v, 0x4e )
-#define mm128_ror_1x32( v )   _mm_shuffle_epi32( v, 0x39 )
-#define mm128_rol_1x32( v )   _mm_shuffle_epi32( v, 0x93 )
+#define mm128_shuflr_64       mm128_swap_64
+#define mm128_shufll_64       mm128_swap_64
+
+#define mm128_shuflr_32( v )   _mm_shuffle_epi32( v, 0x39 )
+#define mm128_shufll_32( v )   _mm_shuffle_epi32( v, 0x93 )
+
 
 // Swap 32 bit elements in 64 bit lanes
 #define mm128_swap64_32( v )  _mm_shuffle_epi32( v, 0xb1 )
+#define mm128_shuflr64_32 mm128_swap64_32
+#define mm128_shufll64_32 mm128_swap64_32
 
 #if defined(__SSSE3__)
 
 // Rotate right by c bytes, no SSE2 equivalent.
-static inline __m128i mm128_ror_x8( const __m128i v, const int c )
+static inline __m128i mm128_shuflr_x8( const __m128i v, const int c )
 { return _mm_alignr_epi8( v, v, c ); }
 
 //
@@ -422,59 +524,88 @@ static inline void mm128_block_bswap_32( __m128i *d, const __m128i *s )
    v1 = _mm_xor_si128( v1, v2 );
 
 
+// Two input shuffle-rotate.
 // Concatenate v1 & v2 and rotate as one 256 bit vector.
-#if defined(__SSE4_1__)
+// Continue to use vror/vrol for now to avoid confusion with
+// shufl2r/shufl2l function macros available with AVX512.
 
-#define mm128_ror256_64( v1, v2 ) \
+#if defined(__SSSE3__)
+
+// Function macro with two inputs and one output, inputs are preserved.
+// Returns modified first arg.
+// Two input functions are not available without SSSE3. Use procedure
+// belowe instead.
+
+#define mm128_shufl2r_64( v1, v2 )     _mm_alignr_epi8( v2, v1, 8 )
+#define mm128_shufl2l_64( v1, v2 )     _mm_alignr_epi8( v1, v2, 8 )
+
+#define mm128_shufl2r_32( v1, v2 )     _mm_alignr_epi8( v2, v1, 4 )
+#define mm128_shufl2l_32( v1, v2 )     _mm_alignr_epi8( v1, v2, 4 )
+
+#define mm128_shufl2r_16( v1, v2 )     _mm_alignr_epi8( v2, v1, 2 )
+#define mm128_shufl2l_16( v1, v2 )     _mm_alignr_epi8( v1, v2, 2 )
+
+#define mm128_shufl2r_8( v1, v2 )      _mm_alignr_epi8( v2, v1, 8 )
+#define mm128_shufl2l_8( v1, v2 )      _mm_alignr_epi8( v1, v2, 8 )
+
+// Procedure macroswith 2 inputs and 2 outputs, inputs are destroyed.
+// Returns both modified args in place.
+
+// These macros retain the vrol/vror name for now to avoid
+// confusion with the shufl2r/shuffle2l function macros above.
+// These may be renamed to something like shufl2r2 for 2 1nputs and
+// 2 outputs, ie SHUFfLe 2 inputs Right with 2 outputs.
+
+#define mm128_vror256_64( v1, v2 ) \
 do { \
    __m128i t  = _mm_alignr_epi8( v1, v2, 8 ); \
            v1 = _mm_alignr_epi8( v2, v1, 8 ); \
            v2 = t; \
 } while(0)
 
-#define mm128_rol256_64( v1, v2 ) \
+#define mm128_vrol256_64( v1, v2 ) \
 do { \
    __m128i t  = _mm_alignr_epi8( v1, v2, 8 ); \
            v2 = _mm_alignr_epi8( v2, v1, 8 ); \
            v1 = t; \
 } while(0)
 
-#define mm128_ror256_32( v1, v2 ) \
+#define mm128_vror256_32( v1, v2 ) \
 do { \
    __m128i t  = _mm_alignr_epi8( v1, v2, 4 ); \
            v1 = _mm_alignr_epi8( v2, v1, 4 ); \
            v2 = t; \
 } while(0)
 
-#define mm128_rol256_32( v1, v2 ) \
+#define mm128_vrol256_32( v1, v2 ) \
 do { \
    __m128i t  = _mm_alignr_epi8( v1, v2, 12 ); \
            v2 = _mm_alignr_epi8( v2, v1, 12 ); \
            v1 = t; \
 } while(0)
 
-#define mm128_ror256_16( v1, v2 ) \
+#define mm128_vror256_16( v1, v2 ) \
 do { \
    __m128i t  = _mm_alignr_epi8( v1, v2, 2 ); \
            v1 = _mm_alignr_epi8( v2, v1, 2 ); \
            v2 = t; \
 } while(0)
 
-#define mm128_rol256_16( v1, v2 ) \
+#define mm128_vrol256_16( v1, v2 ) \
 do { \
    __m128i t  = _mm_alignr_epi8( v1, v2, 14 ); \
            v2 = _mm_alignr_epi8( v2, v1, 14 ); \
            v1 = t; \
 } while(0)
 
-#define mm128_ror256_8( v1, v2 ) \
+#define mm128_vror256_8( v1, v2 ) \
 do { \
    __m128i t  = _mm_alignr_epi8( v1, v2, 1 ); \
            v1 = _mm_alignr_epi8( v2, v1, 1 ); \
            v2 = t; \
 } while(0)
 
-#define mm128_rol256_8( v1, v2 ) \
+#define mm128_vrol256_8( v1, v2 ) \
 do { \
    __m128i t  = _mm_alignr_epi8( v1, v2, 15 ); \
            v2 = _mm_alignr_epi8( v2, v1, 15 ); \
@@ -483,7 +614,7 @@ do { \
 
 #else  // SSE2
 
-#define mm128_ror256_64( v1, v2 ) \
+#define mm128_vror256_64( v1, v2 ) \
 do { \
    __m128i t  = _mm_or_si128( _mm_srli_si128( v1, 8 ), \
                               _mm_slli_si128( v2, 8 ) ); \
@@ -492,7 +623,7 @@ do { \
            v1 = t; \
 } while(0)
 
-#define mm128_rol256_64( v1, v2 ) \
+#define mm128_vrol256_64( v1, v2 ) \
 do { \
    __m128i t  = _mm_or_si128( _mm_slli_si128( v1, 8 ), \
                               _mm_srli_si128( v2, 8 ) ); \
@@ -501,7 +632,7 @@ do { \
            v1 = t; \
 } while(0)
 
-#define mm128_ror256_32( v1, v2 ) \
+#define mm128_vror256_32( v1, v2 ) \
 do { \
    __m128i t  = _mm_or_si128( _mm_srli_si128( v1, 4 ), \
                               _mm_slli_si128( v2, 12 ) ); \
@@ -510,7 +641,7 @@ do { \
            v1 = t; \
 } while(0)
 
-#define mm128_rol256_32( v1, v2 ) \
+#define mm128_vrol256_32( v1, v2 ) \
 do { \
    __m128i t  = _mm_or_si128( _mm_slli_si128( v1, 4 ), \
                               _mm_srli_si128( v2, 12 ) ); \
@@ -519,7 +650,7 @@ do { \
            v1 = t; \
 } while(0)
 
-#define mm128_ror256_16( v1, v2 ) \
+#define mm128_vror256_16( v1, v2 ) \
 do { \
    __m128i t  = _mm_or_si128( _mm_srli_si128( v1, 2 ), \
                               _mm_slli_si128( v2, 14 ) ); \
@@ -528,7 +659,7 @@ do { \
            v1 = t; \
 } while(0)
 
-#define mm128_rol256_16( v1, v2 ) \
+#define mm128_vrol256_16( v1, v2 ) \
 do { \
    __m128i t  = _mm_or_si128( _mm_slli_si128( v1, 2 ), \
                               _mm_srli_si128( v2, 14 ) ); \
@@ -537,7 +668,7 @@ do { \
            v1 = t; \
 } while(0)
 
-#define mm128_ror256_8( v1, v2 ) \
+#define mm128_vror256_8( v1, v2 ) \
 do { \
    __m128i t  = _mm_or_si128( _mm_srli_si128( v1, 1 ), \
                               _mm_slli_si128( v2, 15 ) ); \
@@ -546,7 +677,7 @@ do { \
            v1 = t; \
 } while(0)
 
-#define mm128_rol256_8( v1, v2 ) \
+#define mm128_vrol256_8( v1, v2 ) \
 do { \
    __m128i t  = _mm_or_si128( _mm_slli_si128( v1, 1 ), \
                               _mm_srli_si128( v2, 15 ) ); \

simd-utils/simd-256.h

@@ -1,7 +1,7 @@
 #if !defined(SIMD_256_H__)
 #define SIMD_256_H__ 1
 
-#if defined(__AVX2__)
+//#if defined(__AVX2__)
 
 /////////////////////////////////////////////////////////////////////
 //
@@ -14,13 +14,52 @@
 // is limited because 256 bit vectors are less likely to be used when 512
 // is available.
 
+#if defined(__AVX__)
+
+// Used instead of casting.
+typedef union
+{
+   __m256i m256;
+   __m128i m128[2];
+   uint64_t u64[4];
+   uint32_t u32[8];
+} __attribute__ ((aligned (32))) m256_ovly;
+
+//
+// Pointer casting
+
+// p = any aligned pointer
+// returns p as pointer to vector type, not very useful
+#define castp_m256i(p) ((__m256i*)(p))
+
+// p = any aligned pointer
+// returns *p, watch your pointer arithmetic
+#define cast_m256i(p) (*((__m256i*)(p)))
+
+// p = any aligned pointer, i = scaled array index
+// returns value p[i]
+#define casti_m256i(p,i) (((__m256i*)(p))[(i)])
+
+// p = any aligned pointer, o = scaled offset
+// returns pointer p+o
+#define casto_m256i(p,o) (((__m256i*)(p))+(o))
+
+#endif
+#if defined(__AVX2__)
+
+
 // Move integer to low element of vector, other elements are set to zero.
 #define mm256_mov64_256( i ) _mm256_castsi128_si256( mm128_mov64_128( i ) )
 #define mm256_mov32_256( i ) _mm256_castsi128_si256( mm128_mov32_128( i ) )
 
 // Move low element of vector to integer.
-#define mm256_mov256_64( v ) mm128_mov128_64( _mm256_castsi256_si128( v ) )
-#define mm256_mov256_32( v ) mm128_mov128_32( _mm256_castsi256_si128( v ) )
+#define u64_mov256_64( v ) u64_mov128_64( _mm256_castsi256_si128( v ) )
+#define u32_mov256_32( v ) u32_mov128_32( _mm256_castsi256_si128( v ) )
+
+// deprecated
+//#define mm256_mov256_64 u64_mov256_64 
+//#define mm256_mov256_32 u32_mov256_32
+
 
 // concatenate two 128 bit vectors into one 256 bit vector: { hi, lo }
 #define mm256_concat_128( hi, lo ) \
@@ -76,26 +115,6 @@ static inline __m256i mm256_neg1_fn()
 #define mm128_extr_lo128_256( v ) _mm256_castsi256_si128( v )
 #define mm128_extr_hi128_256( v ) _mm256_extracti128_si256( v, 1 )
 
-//
-// Pointer casting
-
-// p = any aligned pointer
-// returns p as pointer to vector type, not very useful
-#define castp_m256i(p) ((__m256i*)(p))
-
-// p = any aligned pointer
-// returns *p, watch your pointer arithmetic
-#define cast_m256i(p) (*((__m256i*)(p)))
-
-// p = any aligned pointer, i = scaled array index
-// returns value p[i]
-#define casti_m256i(p,i) (((__m256i*)(p))[(i)])
-
-// p = any aligned pointer, o = scaled offset
-// returns pointer p+o
-#define casto_m256i(p,o) (((__m256i*)(p))+(o))
-
-
 //
 // Memory functions
 // n = number of 256 bit (32 byte) vectors
@@ -214,12 +233,41 @@ static inline void memcpy_256( __m256i *dst, const __m256i *src, const int n )
 
 #endif
 
+// Diagonal blending
+
+// Blend 4 64 bit elements from 4 vectors
+#define mm256_diagonal_64( v3, v2, v1, v0 ) \
+  mm256_blend_epi32( _mm256_blend_epi32( v3, v2, 0x30 ), \
+                     _mm256_blend_epi32( v1, v0, 0x03 ), 0x0f )
+
+// Blend 8 32 bit elements from 8 vectors
+#define mm256_diagonal_32( v7, v6, v5, v4, v3, v2, v1, v0 ) \
+  _mm256_blend_epi32( \
+        _mm256_blend_epi32( \
+               _mm256_blend_epi32( v7, v6, 0x40 ), \
+               _mm256_blend_epi32( v5, v4, 0x10 ) 0x30 ), \
+        _mm256_blend_epi32( \
+               _mm256_blend_epi32( v3, v2, 0x04) \
+               _mm256_blend_epi32( v1, v0, 0x01 ), 0x03 ), 0x0f )  
+
+
+// Blend 4 32 bit elements from each 128 bit lane.
+#define mm256_diagonal128_32( v3, v2, v1, v0 ) \
+    _mm256_blend_epi32( \
+           _mm256_blend_epi32( v3, v2, 0x44) \
+           _mm256_blend_epi32( v1, v0, 0x11 ) )  
+
+
 //
 //           Bit rotations.
 //
-// The only bit shift for more than 64 bits is with __int128.
+// The only bit shift for more than 64 bits is with __int128 which is slow.
 //
 // AVX512 has bit rotate for 256 bit vectors with 64 or 32 bit elements
+//
+// x2 rotates elements in 2 individual vectors in a double buffered
+// optimization for SSE2, does nothing for AVX512 but is there for
+// transparency.
 
 
 // compiler doesn't like when a variable is used for the last arg of
@@ -255,6 +303,22 @@ static inline void memcpy_256( __m256i *dst, const __m256i *src, const int n )
 #define mm256_ror_32    _mm256_ror_epi32
 #define mm256_rol_32    _mm256_rol_epi32
 
+#define mm256_rorx2_64( v1, v0, c ) \
+   _mm256_ror_epi64( v0, c ); \
+   _mm256_ror_epi64( v1, c )
+
+#define mm256_rolx2_64( v1, v0, c ) \
+   _mm256_rol_epi64( v0, c ); \
+   _mm256_rol_epi64( v1, c )
+
+#define mm256_rorx2_32( v1, v0, c ) \
+   _mm256_ror_epi32( v0, c ); \
+   _mm256_ror_epi32( v1, c )
+
+#define mm256_rolx2_32( v1, v0, c ) \
+   _mm256_rol_epi32( v0, c ); \
+   _mm256_rol_epi32( v1, c )
+
 #else   // AVX2
 
 #define mm256_ror_64    mm256_ror_var_64 
@@ -262,6 +326,46 @@ static inline void memcpy_256( __m256i *dst, const __m256i *src, const int n )
 #define mm256_ror_32    mm256_ror_var_32
 #define mm256_rol_32    mm256_rol_var_32
 
+#define mm256_rorx2_64( v1, v0, c ) \
+{ \
+ __m256i t0 = _mm256_srli_epi64( v0, c ); \
+ __m256i t1 = _mm256_srli_epi64( v1, c ); \
+ v0 = _mm256_slli_epi64( v0, 64-(c) ); \
+ v1 = _mm256_slli_epi64( v1, 64-(c) ); \
+ v0 = _mm256_or_si256( v0, t0 ); \
+ v1 = _mm256_or_si256( v1, t1 ); \
+}
+
+#define mm256_rolx2_64( v1, v0, c ) \
+{ \
+ __m256i t0 = _mm256_slli_epi64( v0, c ); \
+ __m256i t1 = _mm256_slli_epi64( v1, c ); \
+ v0 = _mm256_srli_epi64( v0, 64-(c) ); \
+ v1 = _mm256_srli_epi64( v1, 64-(c) ); \
+ v0 = _mm256_or_si256( v0, t0 ); \
+ v1 = _mm256_or_si256( v1, t1 ); \
+}
+
+#define mm256_rorx2_32( v1, v0, c ) \
+{ \
+ __m256i t0 = _mm256_srli_epi32( v0, c ); \
+ __m256i t1 = _mm256_srli_epi32( v1, c ); \
+ v0 = _mm256_slli_epi32( v0, 32-(c) ); \
+ v1 = _mm256_slli_epi32( v1, 32-(c) ); \
+ v0 = _mm256_or_si256( v0, t0 ); \
+ v1 = _mm256_or_si256( v1, t1 ); \
+}
+
+#define mm256_rolx2_32( v1, v0, c ) \
+{ \
+ __m256i t0 = _mm256_slli_epi32( v0, c ); \
+ __m256i t1 = _mm256_slli_epi32( v1, c ); \
+ v0 = _mm256_srli_epi32( v0, 32-(c) ); \
+ v1 = _mm256_srli_epi32( v1, 32-(c) ); \
+ v0 = _mm256_or_si256( v0, t0 ); \
+ v1 = _mm256_or_si256( v1, t1 ); \
+}
+
 #endif     // AVX512 else AVX2
 
 #define  mm256_ror_16( v, c ) \
@@ -276,58 +380,45 @@ static inline void memcpy_256( __m256i *dst, const __m256i *src, const int n )
 //
 // Rotate elements accross all lanes.
 
-#if defined(__AVX512VL__)
-
-static inline __m256i mm256_swap_128( const __m256i v )
-{ return _mm256_alignr_epi64( v, v, 2 ); }
-
-static inline __m256i mm256_ror_1x64( const __m256i v )
-{ return _mm256_alignr_epi64( v, v, 1 ); }
-
-static inline __m256i mm256_rol_1x64( const __m256i v )
-{ return _mm256_alignr_epi64( v, v, 3 ); }
-
-static inline __m256i mm256_ror_1x32( const __m256i v )
-{ return _mm256_alignr_epi32( v, v, 1 ); }
-
-static inline __m256i mm256_rol_1x32( const __m256i v )
-{ return _mm256_alignr_epi32( v, v, 7 ); }
-
-#else   // AVX2
-
 // Swap 128 bit elements in 256 bit vector.
 #define mm256_swap_128( v )     _mm256_permute4x64_epi64( v, 0x4e )
+#define mm256_shuflr_128 mm256_swap_128
+#define mm256_shufll_128 mm256_swap_128
 
 // Rotate 256 bit vector by one 64 bit element
-#define mm256_ror_1x64( v )     _mm256_permute4x64_epi64( v, 0x39 )
-#define mm256_rol_1x64( v )     _mm256_permute4x64_epi64( v, 0x93 )
+#define mm256_shuflr_64( v )    _mm256_permute4x64_epi64( v, 0x39 )
+
+#define mm256_shufll_64( v )    _mm256_permute4x64_epi64( v, 0x93 )
 
 // Rotate 256 bit vector by one 32 bit element.
-#define mm256_ror_1x32( v ) \
+#define mm256_shuflr_32( v ) \
     _mm256_permutevar8x32_epi32( v, \
                      m256_const_64( 0x0000000000000007, 0x0000000600000005, \
-                                    0x0000000400000003, 0x0000000200000001 )
+                                    0x0000000400000003, 0x0000000200000001 ) )
 
-#define mm256_rol_1x32( v ) \
+#define mm256_shufll_32( v ) \
     _mm256_permutevar8x32_epi32( v, \
                      m256_const_64( 0x0000000600000005,  0x0000000400000003, \
-                                    0x0000000200000001,  0x0000000000000007 )
+                                    0x0000000200000001,  0x0000000000000007 ) )
 
        
-#endif    // AVX512 else AVX2
-
 //
 // Rotate elements within each 128 bit lane of 256 bit vector.
 
 #define mm256_swap128_64( v )  _mm256_shuffle_epi32( v, 0x4e )
-#define mm256_ror128_32( v )   _mm256_shuffle_epi32( v, 0x39 )
-#define mm256_rol128_32( v )   _mm256_shuffle_epi32( v, 0x93 )
+#define mm256_shuflr128_64 mm256_swap128_64
+#define mm256_shufll128_64 mm256_swap128_64
 
-static inline __m256i mm256_ror128_x8( const __m256i v, const int c )
+#define mm256_shuflr128_32( v )   _mm256_shuffle_epi32( v, 0x39 )
+#define mm256_shufll128_32( v )   _mm256_shuffle_epi32( v, 0x93 )
+
+static inline __m256i mm256_shuflr128_x8( const __m256i v, const int c )
 { return _mm256_alignr_epi8( v, v, c ); }
 
 // Swap 32 bit elements in each 64 bit lane.
 #define mm256_swap64_32( v )   _mm256_shuffle_epi32( v, 0xb1 )
+#define mm256_shuflr64_32 mm256_swap64_32
+#define mm256_shufll64_32 mm256_swap64_32
 
 //
 // Swap bytes in vector elements, endian bswap.
@@ -387,19 +478,21 @@ static inline __m256i mm256_ror128_x8( const __m256i v, const int c )
 //  _mm256_alignr_epi 64/32 are only available with AVX512 but AVX512 also
 //  makes these macros unnecessary.
 
+// continue using vror/vrol notation for now to avoid confusion with
+// shufl2r/shufl2l macro functions available with AVX512.
 #define mm256_swap512_256( v1, v2 ) \
    v1 = _mm256_xor_si256( v1, v2 ); \
    v2 = _mm256_xor_si256( v1, v2 ); \
    v1 = _mm256_xor_si256( v1, v2 );
 
-#define mm256_ror512_128( v1, v2 ) \
+#define mm256_vror512_128( v1, v2 ) \
 do { \
    __m256i t = _mm256_permute2x128( v1, v2, 0x03 ); \
    v1 = _mm256_permute2x128( v2, v1, 0x21 ); \
    v2 = t; \
 } while(0)
 
-#define mm256_rol512_128( v1, v2 ) \
+#define mm256_vrol512_128( v1, v2 ) \
 do { \
    __m256i t = _mm256_permute2x128( v1, v2, 0x03 ); \
    v2 = _mm256_permute2x128( v2, v1, 0x21 ); \

simd-utils/simd-512.h

@@ -74,13 +74,22 @@
 // __AVX512VBMI__  __AVX512VAES__
 //
 
+// Used instead if casting.
+typedef union
+{
+   __m512i m512;
+   __m128i m128[4];
+   uint32_t u32[16];
+   uint64_t u64[8];
+} __attribute__ ((aligned (64))) m512_ovly;
+
 // Move integer to/from element 0 of vector.
 
 #define mm512_mov64_512( n ) _mm512_castsi128_si512( mm128_mov64_128( n ) )
 #define mm512_mov32_512( n ) _mm512_castsi128_si512( mm128_mov32_128( n ) )
 
-#define mm512_mov256_64( a ) mm128_mov128_64( _mm256_castsi512_si128( a ) )
-#define mm512_mov256_32( a ) mm128_mov128_32( _mm256_castsi512_si128( a ) )
+#define u64_mov512_64( a ) u64_mov128_64( _mm256_castsi512_si128( a ) )
+#define u32_mov512_32( a ) u32_mov128_32( _mm256_castsi512_si128( a ) )
 
 // A simple 128 bit permute, using function instead of macro avoids
 // problems if the v arg passed as an expression.
@@ -91,6 +100,10 @@ static inline __m512i mm512_perm_128( const __m512i v, const int c )
 #define mm512_concat_256( hi, lo ) \
    _mm512_inserti64x4( _mm512_castsi256_si512( lo ), hi, 1 )
 
+#define m512_const_128( v3, v2, v1, v0 ) \
+   mm512_concat_256( mm256_concat_128( v3, v2 ), \
+                     mm256_concat_128( v1, v0 ) )
+
 // Equivalent of set, assign 64 bit integers to respective 64 bit elements.
 // Use stack memory overlay
 static inline __m512i m512_const_64( const uint64_t i7, const uint64_t i6,
@@ -225,7 +238,7 @@ static inline void memcpy_512( __m512i *dst, const __m512i *src, const int n )
 
 //
 // Ternary logic uses 8 bit truth table to define any 3 input logical
-// operation using any number or combinations of AND, OR XOR, NOT.
+// expression using any number or combinations of AND, OR, XOR, NOT.
 
 // a ^ b ^ c
 #define mm512_xor3( a, b, c ) \
@@ -251,11 +264,11 @@ static inline void memcpy_512( __m512i *dst, const __m512i *src, const int n )
 #define mm512_andxor( a, b, c ) \
    _mm512_ternarylogic_epi64( a, b, c, 0x60 )
 
-// a ^ ( b & c )
+// a ^ ( b | c )
 #define mm512_xoror( a, b, c ) \
    _mm512_ternarylogic_epi64( a, b, c, 0x1e )
 
-// a ^ ( ~b & c )     [ xor( a, andnot( b, c ) ]
+// a ^ ( ~b & c )     xor( a, andnot( b, c ) )
 #define mm512_xorandnot( a, b, c ) \
   _mm512_ternarylogic_epi64( a, b, c, 0xd2 ) 
 
@@ -265,11 +278,11 @@ static inline void memcpy_512( __m512i *dst, const __m512i *src, const int n )
 
 // Some 2 input operations that don't have their own instruction mnemonic.
 
-// ~( a | b )
+// ~( a | b ),  (~a) & (~b)
 #define mm512_nor( a, b ) \
    _mm512_ternarylogic_epi64( a, b, b, 0x01  )
 
-// ~( a ^ b ), same as (~a) ^ b
+// ~( a ^ b ),  (~a) ^ b
 #define mm512_xnor( a, b ) \
    _mm512_ternarylogic_epi64( a, b, b, 0x81  )
 
@@ -278,6 +291,27 @@ static inline void memcpy_512( __m512i *dst, const __m512i *src, const int n )
    _mm512_ternarylogic_epi64( a, b, b, 0xef  )
 
 
+// Diagonal blending
+// Blend 8 64 bit elements from 8 vectors
+#define mm512_diagonal_64( v7, v6, v5, v4, v3, v2, v1, v0 ) \
+  _mm512_mask_blend_epi64( 0x0f, \
+        _mm512_mask_blend_epi64( 0x30, \
+               _mm512_mask_blend_epi64( 0x40, v7, v6 ), \
+               _mm512_mask_blend_epi64( 0x40, v5, v4 ) ), \
+        _mm512_mask_blend_epi64( 0x03, \
+               _mm512_mask_blend_epi64( 0x04, v3, v2 ) \
+               _mm512_mask_blend_epi64( 0x01, v1, v0 ) ) )  
+
+
+// Blend 4 32 bit elements from each 128 bit lane.
+#define mm512_diagonal128_32( v3, v2, v1, v0 ) \
+    _mm512_mask_blend_epi32( 0x3333, \
+           _mm512_mask_blend_epi32( 0x4444, v3, v2 ), \
+           _mm512_mask_blend_epi32( 0x1111, v1, v0 ) )  
+
+
+
+
 // Bit rotations.
 
 // AVX512F has built-in fixed and variable bit rotation for 64 & 32 bit
@@ -395,59 +429,95 @@ static inline __m512i mm512_rol_16( const __m512i v, const int c )
   casti_m512i( d, 7 ) = _mm512_shuffle_epi8( casti_m512i( s, 7 ), ctl ); \
 } while(0)
 
-
 //
-// Rotate elements in 512 bit vector.
+// Shift with zero fill & shuffle-rotate elements in 512 bit vector.
+//
+
+// rename plan change ror to vror for Vector ROtate Right,
+// and vrol for Vector ROtate Left, not to be confused with
+//variable rotate rorv, rolv,
+// Plan changed, use shuflr & shufll instead symbolizing a shuffle-rotate
+// operation. 1xNN notaion ia also removed and replaced with simpler NN.
+// Swap will still have its own mnemonic and will be aliased as both
+// left and right shuffles.
+
+// Shift elements right or left in 512 bit vector, filling with zeros.
+// Multiple element shifts can be combined into a single larger
+// element shift.
+
+#define mm512_shiftr_256( v ) \
+  _mm512_alignr_epi64( _mm512_setzero, v, 4 )
+#define mm512_shiftl_256( v ) mm512_shifr_256
+
+#define mm512_shiftr_128( v ) \
+  _mm512_alignr_epi64( _mm512_setzero, v, 2 )
+#define mm512_shiftl_128( v ) \
+  _mm512_alignr_epi64( v,  _mm512_setzero, 6 )
+
+#define mm512_shiftr_64( v ) \
+  _mm512_alignr_epi64( _mm512_setzero, v, 1 )
+#define mm512_shiftl_64( v ) \
+  _mm512_alignr_epi64( v, _mm512_setzero, 7 )
+
+#define mm512_shiftr_32( v ) \
+  _mm512_alignr_epi32( _mm512_setzero, v, 1 )
+#define mm512_shiftl_32( v ) \
+  _mm512_alignr_epi32( v, _mm512_setzero, 15 )
+
+// Shuffle-rotate elements left or right in 512 bit vector.
 
 static inline __m512i mm512_swap_256( const __m512i v )
 { return _mm512_alignr_epi64( v, v, 4 ); }
+#define mm512_shuflr_256( v ) mm512_swap_256
+#define mm512_shufll_256( v ) mm512_swap_256
 
-static inline __m512i mm512_ror_1x128( const __m512i v )
+static inline __m512i mm512_shuflr_128( const __m512i v )
 { return _mm512_alignr_epi64( v, v, 2 ); }
 
-static inline __m512i mm512_rol_1x128( const __m512i v )
+static inline __m512i mm512_shufll_128( const __m512i v )
 { return _mm512_alignr_epi64( v, v, 6 ); }
 
-static inline __m512i mm512_ror_1x64( const __m512i v )
+static inline __m512i mm512_shuflr_64( const __m512i v )
 { return _mm512_alignr_epi64( v, v, 1 ); }
 
-static inline __m512i mm512_rol_1x64( const __m512i v )
+static inline __m512i mm512_shufll_64( const __m512i v )
 { return _mm512_alignr_epi64( v, v, 7 ); }
 
-static inline __m512i mm512_ror_1x32( const __m512i v )
+static inline __m512i mm512_shuflr_32( const __m512i v )
 { return _mm512_alignr_epi32( v, v, 1 ); }
 
-static inline __m512i mm512_rol_1x32( const __m512i v )
+static inline __m512i mm512_shufll_32( const __m512i v )
 { return _mm512_alignr_epi32( v, v, 15 ); }
 
-static inline __m512i mm512_ror_x64( const __m512i v, const int n )
+// Generic
+static inline __m512i mm512_shuflr_x64( const __m512i v, const int n )
 { return _mm512_alignr_epi64( v, v, n ); }
 
-static inline __m512i mm512_ror_x32( const __m512i v, const int n )
+static inline __m512i mm512_shufll_x32( const __m512i v, const int n )
 { return _mm512_alignr_epi32( v, v, n ); }
 
-#define mm512_ror_1x16( v ) \
+#define mm512_shuflr_16( v ) \
    _mm512_permutexvar_epi16( m512_const_64( \
                        0x0000001F001E001D, 0x001C001B001A0019, \
                        0X0018001700160015, 0X0014001300120011, \
                        0X0010000F000E000D, 0X000C000B000A0009, \
                        0X0008000700060005, 0X0004000300020001 ), v )
 
-#define mm512_rol_1x16( v ) \
+#define mm512_shufll_16( v ) \
    _mm512_permutexvar_epi16( m512_const_64( \
                        0x001E001D001C001B, 0x001A001900180017, \
                        0X0016001500140013, 0X001200110010000F, \
                        0X000E000D000C000B, 0X000A000900080007, \
                        0X0006000500040003, 0X000200010000001F ), v )
 
-#define mm512_ror_1x8( v ) \
+#define mm512_shuflr_8( v ) \
    _mm512_shuffle_epi8( v, m512_const_64( \
                        0x003F3E3D3C3B3A39, 0x3837363534333231, \
                        0x302F2E2D2C2B2A29, 0x2827262524232221, \
                        0x201F1E1D1C1B1A19. 0x1817161514131211, \
                        0x100F0E0D0C0B0A09, 0x0807060504030201 ) )
 
-#define mm512_rol_1x8( v ) \
+#define mm512_shufll_8( v ) \
    _mm512_shuffle_epi8( v, m512_const_64( \
                        0x3E3D3C3B3A393837, 0x363534333231302F. \
                        0x2E2D2C2B2A292827, 0x262524232221201F, \
@@ -456,51 +526,54 @@ static inline __m512i mm512_ror_x32( const __m512i v, const int n )
 
 //
 // Rotate elements within 256 bit lanes of 512 bit vector.
+// 128 bit lane shift is handled by bslli bsrli.
 
 // Swap hi & lo 128 bits in each 256 bit lane
 #define mm512_swap256_128( v )   _mm512_permutex_epi64( v, 0x4e )
+#define mm512_shuflr256_128 mm512_swap256_128
+#define mm512_shufll256_128 mm512_swap256_128
 
 // Rotate 256 bit lanes by one 64 bit element
-#define mm512_ror256_64( v )     _mm512_permutex_epi64( v, 0x39 )
-#define mm512_rol256_64( v )     _mm512_permutex_epi64( v, 0x93 )
+#define mm512_shuflr256_64( v )     _mm512_permutex_epi64( v, 0x39 )
+#define mm512_shufll256_64( v )     _mm512_permutex_epi64( v, 0x93 )
 
 // Rotate 256 bit lanes by one 32 bit element
-#define mm512_ror256_32( v ) \
+#define mm512_shuflr256_32( v ) \
    _mm512_permutexvar_epi32( m512_const_64( \
                       0x000000080000000f, 0x0000000e0000000d, \
                       0x0000000c0000000b, 0x0000000a00000009, \
                       0x0000000000000007, 0x0000000600000005, \
                       0x0000000400000003, 0x0000000200000001 ), v )
 
-#define mm512_rol256_32( v ) \
+#define mm512_shufll256_32( v ) \
    _mm512_permutexvar_epi32( m512_const_64( \
                       0x0000000e0000000d, 0x0000000c0000000b, \
                       0x0000000a00000009, 0x000000080000000f, \
                       0x0000000600000005, 0x0000000400000003, \
                       0x0000000200000001, 0x0000000000000007 ), v )
 
-#define mm512_ror256_16( v ) \
+#define mm512_shuflr256_16( v ) \
    _mm512_permutexvar_epi16( m512_const_64( \
                      0x00100001001e001d, 0x001c001b001a0019, \
                      0x0018001700160015, 0x0014001300120011, \
                      0x0000000f000e000d, 0x000c000b000a0009, \
                      0x0008000700060005, 0x0004000300020001 ), v )
 
-#define mm512_rol256_16( v ) \
+#define mm512_shufll256_16( v ) \
    _mm512_permutexvar_epi16( m512_const_64( \
                      0x001e001d001c001b, 0x001a001900180017, \
                      0x0016001500140013, 0x001200110010001f, \
                      0x000e000d000c000b, 0x000a000900080007, \
                      0x0006000500040003, 0x000200010000000f ), v )
 
-#define mm512_ror256_8( v ) \
+#define mm512_shuflr256_8( v ) \
    _mm512_shuffle_epi8( v, m512_const_64( \
                      0x203f3e3d3c3b3a39, 0x3837363534333231, \
                      0x302f2e2d2c2b2a29, 0x2827262524232221, \
                      0x001f1e1d1c1b1a19, 0x1817161514131211, \
                      0x100f0e0d0c0b0a09, 0x0807060504030201 ) )
 
-#define mm512_rol256_8( v ) \
+#define mm512_shufll256_8( v ) \
    _mm512_shuffle_epi8( v, m512_const_64( \
                      0x3e3d3c3b3a393837, 0x363534333231302f, \
                      0x2e2d2c2b2a292827, 0x262524232221203f, \
@@ -508,82 +581,117 @@ static inline __m512i mm512_ror_x32( const __m512i v, const int n )
                      0x0e0d0c0b0a090807, 0x060504030201001f ) )
 
 //
-// Rotate elements within 128 bit lanes of 512 bit vector.
-
+// Shuffle-roate elements within 128 bit lanes of 512 bit vector.
+ 
 // Swap 64 bits in each 128 bit lane
 #define mm512_swap128_64( v )   _mm512_shuffle_epi32( v, 0x4e )
+#define mm512_shuflr128_64  mm512_swap128_64
+#define mm512_shufll128_64  mm512_swap128_64
 
 // Rotate 128 bit lanes by one 32 bit element
-#define mm512_ror128_32( v )    _mm512_shuffle_epi32( v, 0x39 )
-#define mm512_rol128_32( v )    _mm512_shuffle_epi32( v, 0x93 )
+#define mm512_shuflr128_32( v )    _mm512_shuffle_epi32( v, 0x39 )
+#define mm512_shufll128_32( v )    _mm512_shuffle_epi32( v, 0x93 )
 
-// Rotate right 128 bit lanes by c bytes
-static inline __m512i mm512_ror128_x8( const __m512i v, const int c )
+// Rotate right 128 bit lanes by c bytes, versatile and just as fast
+static inline __m512i mm512_shuflr128_8( const __m512i v, const int c )
 {  return _mm512_alignr_epi8( v, v, c ); }
 
-// Swap 32 bits in each 64 bit lane.
+// Swap 32 bits in each 64 bit lane. Can be done with rotate instruction
+// but only with AVX512. Shuffle is just as fast and availble with AVX2
+// & SSE2.
 #define mm512_swap64_32( v )    _mm512_shuffle_epi32( v, 0xb1 )
+#define mm512_shuflr64_32 mm512_swap64_32
+#define mm512_shufll64_32 mm512_swap64_32
 
-
+// Need good way to distinguish 1 input shuffles, 2 input shuffle functions,
+// and 2 input 2 output shuffle macros.
 //
-//  Rotate elements from 2 512 bit vectors in place, source arguments
+// shuflr is 1 input
+// shufl2r is 2 input ...
+// Drop macros? They can easilly be rebuilt using shufl2 functions
+
+// Shuffle concatenated { v1, v2 ) right or left by 256 bits and return
+// rotated v1 
+// visually confusing for shif2r because of arg order. First arg is always
+// the target for modification, either update by reference or by function
+// return.
+#define mm512_shufl2r_256( v1, v2 )    _mm512_alignr_epi64( v2, v1, 4 )
+#define mm512_shufl2l_256( v1, v2 )    _mm512_alignr_epi64( v1, v2, 4 )
+
+#define mm512_shufl2r_128( v1, v2 )    _mm512_alignr_epi64( v2, v1, 2 )
+#define mm512_shufl2l_128( v1, v2 )    _mm512_alignr_epi64( v1, v2, 2 )
+
+#define mm512_shufl2r_64( v1, v2 )     _mm512_alignr_epi64( v2, v1, 1 )
+#define mm512_shufl2l_64( v1, v2 )     _mm512_alignr_epi64( v1, v2, 1 )
+
+#define mm512_shufl2r_32( v1, v2 )     _mm512_alignr_epi32( v2, v1, 1 )
+#define mm512_shufl2l_32( v1, v2 )     _mm512_alignr_epi32( v1, v2, 1 )
+
+// Rotate elements from 2 512 bit vectors in place, source arguments
 //  are overwritten.
 
 #define mm512_swap1024_512( v1, v2 ) \
    v1 = _mm512_xor_si512( v1, v2 ); \
    v2 = _mm512_xor_si512( v1, v2 ); \
    v1 = _mm512_xor_si512( v1, v2 );
+#define mm512_shufl2l_512 mm512_swap1024_512 \
+#define mm512_shufl2r_512 mm512_swap1024_512 \
 
-#define mm512_ror1024_256( v1, v2 ) \
+// Deprecated, will be removed. Use shufl2 functions instead. Leave them as is
+// for now.
+//  Rotate elements from 2 512 bit vectors in place, both source arguments
+//  are updated.
+
+#define mm512_vror1024_256( v1, v2 ) \
 do { \
    __m512i t = _mm512_alignr_epi64( v1, v2, 4 ); \
    v1 = _mm512_alignr_epi64( v2, v1, 4 ); \
    v2 = t; \
 } while(0)
 
-#define mm512_rol1024_256( v1, v2 ) \
+#define mm512_vrol1024_256( v1, v2 ) \
 do { \
    __m512i t = _mm512_alignr_epi64( v1, v2, 4 ); \
    v2 = _mm512_alignr_epi64( v2, v1, 4 ); \
    v1 = t; \
 } while(0)
 
-#define mm512_ror1024_128( v1, v2 ) \
+#define mm512_vror1024_128( v1, v2 ) \
 do { \
    __m512i t = _mm512_alignr_epi64( v1, v2, 2 ); \
    v1 = _mm512_alignr_epi64( v2, v1, 2 ); \
    v2 = t; \
 } while(0)
 
-#define mm512_rol1024_128( v1, v2 ) \
+#define mm512_vrol1024_128( v1, v2 ) \
 do { \
    __m512i t = _mm512_alignr_epi64( v1, v2, 6 ); \
    v2 = _mm512_alignr_epi64( v2, v1, 6 ); \
    v1 = t; \
 } while(0)
 
-#define mm512_ror1024_64( v1, v2 ) \
+#define mm512_vror1024_64( v1, v2 ) \
 do { \
    __m512i t = _mm512_alignr_epi64( v1, v2, 1 ); \
    v1 = _mm512_alignr_epi64( v2, v1, 1 ); \
    v2 = t; \
 } while(0)
 
-#define mm512_rol1024_64( v1, v2 ) \
+#define mm512_vrol1024_64( v1, v2 ) \
 do { \
    __m512i t = _mm512_alignr_epi64( v1, v2, 7 ); \
    v2 = _mm512_alignr_epi64( v2, v1, 7 ); \
    v1 = t; \
 } while(0)
 
-#define mm512_ror1024_32( v1, v2 ) \
+#define mm512_vror1024_32( v1, v2 ) \
 do { \
    __m512i t = _mm512_alignr_epi32( v1, v2, 1 ); \
    v1 = _mm512_alignr_epi32( v2, v1, 1 ); \
    v2 = t; \
 } while(0)
 
-#define mm512_rol1024_32( v1, v2 ) \
+#define mm512_vrol1024_32( v1, v2 ) \
 do { \
    __m512i t = _mm512_alignr_epi32( v1, v2, 15 ); \
    v2 = _mm512_alignr_epi32( v2, v1, 15 ); \

simd-utils/simd-64.h

@@ -68,13 +68,13 @@
 // rotation.
 
 // Swap hi & lo 32 bits.
-#define mm64_swap_32( a )     _mm_shuffle_pi16( a, 0x4e )
+#define mm64_swap_32( a )      _mm_shuffle_pi16( a, 0x4e )
 
-#define mm64_ror64_1x16( a )  _mm_shuffle_pi16( a, 0x39 ) 
-#define mm64_rol64_1x16( a )  _mm_shuffle_pi16( a, 0x93 ) 
+#define mm64_shulfr_16( a )  _mm_shuffle_pi16( a, 0x39 ) 
+#define mm64_shufll_16( a )  _mm_shuffle_pi16( a, 0x93 ) 
 
 // Swap hi & lo 16 bits of each 32 bit element
-#define mm64_swap32_16( a )  _mm_shuffle_pi16( a, 0xb1 )
+#define mm64_swap32_16( a )    _mm_shuffle_pi16( a, 0xb1 )
 
 #if defined(__SSSE3__)
 
@@ -86,7 +86,7 @@
     _mm_shuffle_pi8( v, (__m64)0x0607040502030001 );
 
 // Rotate right by c bytes
-static inline __m64 mm64_ror_x8( __m64 v, const int c )
+static inline __m64 mm64_vror_x8( __m64 v, const int c )
 { return _mm_alignr_pi8( v, v, c ); }
 
 #else

simd-utils/simd-int.h

@@ -2,10 +2,18 @@
 #define SIMD_INT_H__ 1
 
 // Endian byte swap
-#define bswap_64( a ) __builtin_bswap64( a )
-#define bswap_32( a ) __builtin_bswap32( a )
+#define bswap_64    __builtin_bswap64
+#define bswap_32    __builtin_bswap32
 
-// safe division, integer or floating point
+// Bit rotation
+#define rol64       __rolq
+#define ror64       __rorq
+#define rol32       __rold
+#define ror32       __rord
+
+// Safe division, integer or floating point. For floating point it's as  
+// safe as 0 is precisely zero.
+// Returns safe_result if division by zero, typically zero.
 #define safe_div( dividend, divisor, safe_result ) \
    ( (divisor) == 0 ? safe_result : ( (dividend) / (divisor) )  )
 
@@ -29,12 +37,14 @@
 // __m256i v256 = _mm256_set_m128i( (__m128i)my_int128, (__m128i)my_int128 );
 // my_int128 = (uint128_t)_mm256_extracti128_si256( v256, 1 );
 
+// obsolete test
 // Compiler check for __int128 support
 // Configure also has a test for int128.
 #if ( __GNUC__ > 4 ) || ( ( __GNUC__ == 4 ) && ( __GNUC_MINOR__ >= 8 ) )
   #define GCC_INT128 1
 #endif
 
+// obsolte test
 #if !defined(GCC_INT128)
   #warning "__int128 not supported, requires GCC-4.8 or newer."
 #endif

sysinfos.c

@@ -218,7 +218,7 @@ static inline void cpu_getname(char *outbuf, size_t maxsz)
       for (int i = 2; i <= (ext & 0xF); i++)
       {
          cpuid(0x80000000+i, output);
-	 memcpy(&brand[(i-2) * 4*sizeof(int)], output, 4*sizeof(int));
+         memcpy(&brand[(i-2) * 4*sizeof(int)], output, 4*sizeof(int));
       }
       snprintf(outbuf, maxsz, "%s", brand);
    }

util.c

@@ -47,6 +47,7 @@
 //#include "miner.h"
 #include "elist.h"
 #include "algo-gate-api.h"
+#include "algo/sha/sha256d.h"
 
 //extern pthread_mutex_t stats_lock;
 
@@ -129,17 +130,19 @@ void applog2( int prio, const char *fmt, ... )
 
 //    localtime_r(&now, &tm);
 
-      switch (prio) {
+      switch ( prio )
+      {
+         case LOG_CRIT:    color = CL_LRD; break;
          case LOG_ERR:     color = CL_RED; break;
-         case LOG_WARNING: color = CL_YLW; break;
+         case LOG_WARNING: color = CL_YL2; break;
+         case LOG_MAJR:    color = CL_YL2; break;
          case LOG_NOTICE:  color = CL_WHT; break;
          case LOG_INFO:    color = ""; break;
          case LOG_DEBUG:   color = CL_GRY; break;
-
-         case LOG_BLUE:
-            prio = LOG_NOTICE;
-            color = CL_CYN;
-            break;
+         case LOG_MINR:    color = CL_YLW; break;
+         case LOG_GREEN:   color = CL_GRN; prio = LOG_INFO; break;
+         case LOG_BLUE:    color = CL_CYN; prio = LOG_NOTICE; break;
+         case LOG_PINK:    color = CL_LMA; prio = LOG_NOTICE; break;
       }
       if (!use_colors)
          color = "";
@@ -206,17 +209,19 @@ void applog(int prio, const char *fmt, ...)
 
 		localtime_r(&now, &tm);
 
-		switch (prio) {
-			case LOG_ERR:     color = CL_RED; break;
-			case LOG_WARNING: color = CL_YLW; break;
+		switch ( prio )
+      {
+         case LOG_CRIT:    color = CL_LRD; break;
+         case LOG_ERR:     color = CL_RED; break;
+			case LOG_WARNING: color = CL_YL2; break;
+         case LOG_MAJR:    color = CL_YL2; break;
 			case LOG_NOTICE:  color = CL_WHT; break;
-			case LOG_INFO:    color = ""; break;
+			case LOG_INFO:    color = "";     break;
 			case LOG_DEBUG:   color = CL_GRY; break;
-
-			case LOG_BLUE:
-				prio = LOG_NOTICE;
-				color = CL_CYN;
-				break;
+         case LOG_MINR:    color = CL_YLW; break;
+         case LOG_GREEN:   color = CL_GRN; prio = LOG_INFO;  break;
+			case LOG_BLUE:    color = CL_CYN; prio = LOG_NOTICE; break;
+         case LOG_PINK:    color = CL_LMA; prio = LOG_NOTICE; break;
 		}
 		if (!use_colors)
 			color = "";
@@ -303,6 +308,29 @@ void format_hashrate(double hashrate, char *output)
 	);
 }
 
+// For use with MiB etc
+void format_number_si( double* n, char* si_units )
+{
+  if ( *n < 1024*10 )  {  *si_units = 0;   return;  }
+  *n /= 1024;
+  if ( *n < 1024*10 )  {  *si_units = 'k'; return;  }
+  *n /= 1024;
+  if ( *n < 1024*10 )  {  *si_units = 'M'; return;  }
+  *n /= 1024;
+  if ( *n < 1024*10 )  {  *si_units = 'G'; return;  }
+  *n /= 1024;
+  if ( *n < 1024*10 )  {  *si_units = 'T'; return;  }
+  *n /= 1024;
+  if ( *n < 1024*10 )  {  *si_units = 'P'; return;  }
+  *n /= 1024;
+  if ( *n < 1024*10 )  {  *si_units = 'E'; return;  }
+  *n /= 1024;
+  if ( *n < 1024*10 )  {  *si_units = 'Z'; return;  }
+  *n /= 1024;
+  *si_units = 'Y';
+}
+
+
 /* Modify the representation of integer numbers which would cause an overflow
  * so that they are treated as floating-point numbers.
  * This is a hack to overcome the limitations of some versions of Jansson. */

winbuild-cross.sh

@@ -16,13 +16,13 @@ export MINGW_LIB="/usr/x86_64-w64-mingw32/lib"
 export GCC_MINGW_LIB="/usr/lib/gcc/x86_64-w64-mingw32/9.3-win32"
 # used by GCC
 export LDFLAGS="-L$LOCAL_LIB/curl/lib/.libs -L$LOCAL_LIB/gmp/.libs -L$LOCAL_LIB/openssl"
+# support for Windows CPU groups, AES sometimes not included in -march
+export DEFAULT_CFLAGS="-O3 -maes -Wall -D_WIN32_WINNT=0x0601"
+export DEFAULT_CFLAGS_OLD="-O3 -Wall"
 
 # make link to local gmp header file.
 ln -s $LOCAL_LIB/gmp/gmp.h ./gmp.h
 
-# edit configure to fix pthread lib name for Windows.
-#sed -i 's/"-lpthread"/"-lpthreadGC2"/g' configure.ac
-
 # make release directory and copy selected DLLs.
 
 rm -rf release > /dev/null
@@ -41,74 +41,59 @@ cp $LOCAL_LIB/curl/lib/.libs/libcurl-4.dll release/
 
 # Start building...
 
-# Icelake AVX512 SHA VAES
+# AVX512 SHA VAES: Intel Core Icelake, Rocketlake
 ./clean-all.sh || echo clean
 rm -f config.status
 ./autogen.sh || echo done
-CFLAGS="-O3 -march=icelake-client -Wall" ./configure $CONFIGURE_ARGS
+CFLAGS="$DEFAULT_CFLAGS -march=icelake-client" ./configure $CONFIGURE_ARGS
 make -j 8
 strip -s cpuminer.exe
 mv cpuminer.exe release/cpuminer-avx512-sha-vaes.exe
 
-# Rocketlake AVX512 SHA AES
+# AVX512 AES: Intel Core HEDT Slylake-X, Cascadelake 
 make clean || echo clean
 rm -f config.status
-CFLAGS="-O3 -march=cascadelake -msha -Wall" ./configure $CONFIGURE_ARGS
-#CFLAGS="-O3 -march=rocketlake -Wall" ./configure $CONFIGURE_ARGS
-make -j 8
-strip -s cpuminer.exe
-mv cpuminer.exe release/cpuminer-avx512-sha.exe
-
-# Zen1 AVX2 AES SHA
-make clean || echo clean
-rm -f config.status
-CFLAGS="-O3 -march=znver1 -Wall" ./configure $CONFIGURE_ARGS
-make -j 8
-strip -s cpuminer.exe
-mv cpuminer.exe release/cpuminer-zen.exe
-
-# Zen3 AVX2 SHA VAES
-make clean || echo clean
-rm -f config.status
-CFLAGS="-O3 -march=znver2 -mvaes -Wall" ./configure $CONFIGURE_ARGS
-# CFLAGS="-O3 -march=znver3 -Wall" ./configure $CONFIGURE_ARGS
-make -j 8
-strip -s cpuminer.exe
-mv cpuminer.exe release/cpuminer-zen3.exe
-
-# Slylake-X AVX512 AES
-# mingw won't compile avx512 without -fno-asynchronous-unwind-tables
-make clean || echo clean
-rm -f config.status
-CFLAGS="-O3 -march=skylake-avx512 -Wall" ./configure $CONFIGURE_ARGS
-#CFLAGS="-O3 -march=skylake-avx512 -Wall -fno-asynchronous-unwind-tables" ./configure $CONFIGURE_ARGS
+CFLAGS="$DEFAULT_CFLAGS -march=skylake-avx512" ./configure $CONFIGURE_ARGS
 make -j 8
 strip -s cpuminer.exe
 mv cpuminer.exe release/cpuminer-avx512.exe
 
-# Haswell AVX2 AES
+# AVX2 SHA VAES: Intel Alderlake, AMD Zen3
+make clean || echo done
+rm -f config.status
+CFLAGS="$DEFAULT_CFLAGS -mavx2 -msha -mvaes" ./configure $CONFIGURE_ARGS
+make -j 8
+strip -s cpuminer.exe
+mv cpuminer.exe release/cpuminer-avx2-sha-vaes.exe
+
+# AVX2 AES SHA: AMD Zen1
 make clean || echo clean
 rm -f config.status
-# GCC 9 doesn't include AES in -march=core-avx2
-CFLAGS="-O3 -march=core-avx2 -maes -Wall" ./configure $CONFIGURE_ARGS
+CFLAGS="$DEFAULT_CFLAGS -march=znver1" ./configure $CONFIGURE_ARGS
+make -j 8
+strip -s cpuminer.exe
+mv cpuminer.exe release/cpuminer-avx2-sha.exe
+
+# AVX2 AES: Intel Core Haswell, Skylake, Kabylake, Coffeelake, Cometlake
+make clean || echo clean
+rm -f config.status
+CFLAGS="$DEFAULT_CFLAGS -march=core-avx2" ./configure $CONFIGURE_ARGS
 make -j 8
 strip -s cpuminer.exe
 mv cpuminer.exe release/cpuminer-avx2.exe
 
-# Sandybridge AVX AES
+# AVX AES: Intel Sandybridge, Ivybridge
 make clean || echo clean
 rm -f config.status
-# -march=corei7-avx still includes aes, but just in case
-CFLAGS="-O3 -march=corei7-avx -maes -Wall" ./configure $CONFIGURE_ARGS 
+CFLAGS="$DEFAULT_CFLAGS_OLD -march=corei7-avx -maes" ./configure $CONFIGURE_ARGS 
 make -j 8
 strip -s cpuminer.exe
 mv cpuminer.exe release/cpuminer-avx.exe
 
-# Westmere SSE4.2 AES
+# SSE4.2 AES: Intel Westmere
 make clean || echo clean
 rm -f config.status
-CFLAGS="-O3 -march=westmere -maes -Wall" ./configure $CONFIGURE_ARGS
-#CFLAGS="-O3 -maes -msse4.2 -Wall" ./configure $CONFIGURE_ARGS
+CFLAGS="$DEFAULT_CFLAGS_OLD -march=westmere -maes" ./configure $CONFIGURE_ARGS
 make -j 8
 strip -s cpuminer.exe
 mv cpuminer.exe release/cpuminer-aes-sse42.exe
@@ -116,7 +101,7 @@ mv cpuminer.exe release/cpuminer-aes-sse42.exe
 # Nehalem SSE4.2
 #make clean || echo clean
 #rm -f config.status
-#CFLAGS="-O3 -march=corei7 -Wall" ./configure $CONFIGURE_ARGS
+#CFLAGS="$DEFAULT_CFLAGS_OLD -march=corei7" ./configure $CONFIGURE_ARGS
 #make 
 #strip -s cpuminer.exe
 #mv cpuminer.exe release/cpuminer-sse42.exe
@@ -124,7 +109,7 @@ mv cpuminer.exe release/cpuminer-aes-sse42.exe
 # Core2 SSSE3
 #make clean || echo clean
 #rm -f config.status
-#CFLAGS="-O3 -march=core2 -Wall" ./configure $CONFIGURE_ARGS
+#CFLAGS="$DEFAULT_CFLAGS_OLD -march=core2" ./configure $CONFIGURE_ARGS
 #make 
 #strip -s cpuminer.exe
 #mv cpuminer.exe release/cpuminer-ssse3.exe
@@ -133,7 +118,7 @@ mv cpuminer.exe release/cpuminer-aes-sse42.exe
 # Generic SSE2
 make clean || echo clean
 rm -f config.status
-CFLAGS="-O3 -msse2 -Wall" ./configure $CONFIGURE_ARGS
+CFLAGS="$DEFAULT_CFLAGS_OLD -msse2" ./configure $CONFIGURE_ARGS
 make -j 8
 strip -s cpuminer.exe
 mv cpuminer.exe release/cpuminer-sse2.exe