diff --git a/CHANGELOG.md b/CHANGELOG.md index 24bba74..f97df81 100644 --- a/CHANGELOG.md +++ b/CHANGELOG.md @@ -5,6 +5,7 @@ * Add documentation for ChaChaPoly1305 * Add missing documentation for various modules * Add a way to create Poly1305 Auth tag. +* Added support for the BLAKE2 family of hash algorithms ## 0.8 diff --git a/Crypto/Hash/Algorithms.hs b/Crypto/Hash/Algorithms.hs index a05a8e9..7c46a37 100644 --- a/Crypto/Hash/Algorithms.hs +++ b/Crypto/Hash/Algorithms.hs @@ -10,6 +10,10 @@ module Crypto.Hash.Algorithms ( HashAlgorithm -- * hash algorithms + , BLAKE2s(..) + , BLAKE2sp(..) + , BLAKE2b(..) + , BLAKE2bp(..) , MD2(..) , MD4(..) , MD5(..) @@ -40,6 +44,10 @@ module Crypto.Hash.Algorithms ) where import Crypto.Hash.Types (HashAlgorithm) +import Crypto.Hash.BLAKE2s +import Crypto.Hash.BLAKE2sp +import Crypto.Hash.BLAKE2b +import Crypto.Hash.BLAKE2bp import Crypto.Hash.MD2 import Crypto.Hash.MD4 import Crypto.Hash.MD5 diff --git a/Crypto/Hash/BLAKE2b.hs b/Crypto/Hash/BLAKE2b.hs new file mode 100644 index 0000000..9521bfa --- /dev/null +++ b/Crypto/Hash/BLAKE2b.hs @@ -0,0 +1,37 @@ +-- | +-- Module : Crypto.Hash.BLAKE2b +-- License : BSD-style +-- Maintainer : Vincent Hanquez +-- Stability : experimental +-- Portability : unknown +-- +-- module containing the binding functions to work with the +-- BLAKE2b cryptographic hash. +-- +{-# LANGUAGE ForeignFunctionInterface #-} +module Crypto.Hash.BLAKE2b ( BLAKE2b (..) ) where + +import Crypto.Hash.Types +import Foreign.Ptr (Ptr) +import Data.Word (Word8, Word32) + +-- | BLAKE2b cryptographic hash algorithm +data BLAKE2b = BLAKE2b + deriving (Show) + +instance HashAlgorithm BLAKE2b where + hashBlockSize _ = 128 + hashDigestSize _ = 64 + hashInternalContextSize _ = 361 + hashInternalInit = c_blake2b_init + hashInternalUpdate = c_blake2b_update + hashInternalFinalize = c_blake2b_finalize + +foreign import ccall unsafe "cryptonite_blake2b_init" + c_blake2b_init :: Ptr (Context a)-> IO () + +foreign import ccall "cryptonite_blake2b_update" + c_blake2b_update :: Ptr (Context a) -> Ptr Word8 -> Word32 -> IO () + +foreign import ccall unsafe "cryptonite_blake2b_finalize" + c_blake2b_finalize :: Ptr (Context a) -> Ptr (Digest a) -> IO () diff --git a/Crypto/Hash/BLAKE2bp.hs b/Crypto/Hash/BLAKE2bp.hs new file mode 100644 index 0000000..1e71ba2 --- /dev/null +++ b/Crypto/Hash/BLAKE2bp.hs @@ -0,0 +1,37 @@ +-- | +-- Module : Crypto.Hash.BLAKE2bp +-- License : BSD-style +-- Maintainer : Vincent Hanquez +-- Stability : experimental +-- Portability : unknown +-- +-- module containing the binding functions to work with the +-- BLAKE2bp cryptographic hash. +-- +{-# LANGUAGE ForeignFunctionInterface #-} +module Crypto.Hash.BLAKE2bp ( BLAKE2bp (..) ) where + +import Crypto.Hash.Types +import Foreign.Ptr (Ptr) +import Data.Word (Word8, Word32) + +-- | BLAKE2bp cryptographic hash algorithm +data BLAKE2bp = BLAKE2bp + deriving (Show) + +instance HashAlgorithm BLAKE2bp where + hashBlockSize _ = 128 + hashDigestSize _ = 64 + hashInternalContextSize _ = 2325 + hashInternalInit = c_blake2sp_init + hashInternalUpdate = c_blake2sp_update + hashInternalFinalize = c_blake2sp_finalize + +foreign import ccall unsafe "cryptonite_blake2sp_init" + c_blake2sp_init :: Ptr (Context a)-> IO () + +foreign import ccall "cryptonite_blake2sp_update" + c_blake2sp_update :: Ptr (Context a) -> Ptr Word8 -> Word32 -> IO () + +foreign import ccall unsafe "cryptonite_blake2sp_finalize" + c_blake2sp_finalize :: Ptr (Context a) -> Ptr (Digest a) -> IO () diff --git a/Crypto/Hash/BLAKE2s.hs b/Crypto/Hash/BLAKE2s.hs new file mode 100644 index 0000000..ab7ea67 --- /dev/null +++ b/Crypto/Hash/BLAKE2s.hs @@ -0,0 +1,37 @@ +-- | +-- Module : Crypto.Hash.BLAKE2s +-- License : BSD-style +-- Maintainer : Vincent Hanquez +-- Stability : experimental +-- Portability : unknown +-- +-- module containing the binding functions to work with the +-- BLAKE2s cryptographic hash. +-- +{-# LANGUAGE ForeignFunctionInterface #-} +module Crypto.Hash.BLAKE2s ( BLAKE2s (..) ) where + +import Crypto.Hash.Types +import Foreign.Ptr (Ptr) +import Data.Word (Word8, Word32) + +-- | BLAKE2s cryptographic hash algorithm +data BLAKE2s = BLAKE2s + deriving (Show) + +instance HashAlgorithm BLAKE2s where + hashBlockSize _ = 64 + hashDigestSize _ = 32 + hashInternalContextSize _ = 185 + hashInternalInit = c_blake2s_init + hashInternalUpdate = c_blake2s_update + hashInternalFinalize = c_blake2s_finalize + +foreign import ccall unsafe "cryptonite_blake2s_init" + c_blake2s_init :: Ptr (Context a)-> IO () + +foreign import ccall "cryptonite_blake2s_update" + c_blake2s_update :: Ptr (Context a) -> Ptr Word8 -> Word32 -> IO () + +foreign import ccall unsafe "cryptonite_blake2s_finalize" + c_blake2s_finalize :: Ptr (Context a) -> Ptr (Digest a) -> IO () diff --git a/Crypto/Hash/BLAKE2sp.hs b/Crypto/Hash/BLAKE2sp.hs new file mode 100644 index 0000000..5061722 --- /dev/null +++ b/Crypto/Hash/BLAKE2sp.hs @@ -0,0 +1,37 @@ +-- | +-- Module : Crypto.Hash.BLAKE2sp +-- License : BSD-style +-- Maintainer : Vincent Hanquez +-- Stability : experimental +-- Portability : unknown +-- +-- module containing the binding functions to work with the +-- BLAKE2sp cryptographic hash. +-- +{-# LANGUAGE ForeignFunctionInterface #-} +module Crypto.Hash.BLAKE2sp ( BLAKE2sp (..) ) where + +import Crypto.Hash.Types +import Foreign.Ptr (Ptr) +import Data.Word (Word8, Word32) + +-- | BLAKE2sp cryptographic hash algorithm +data BLAKE2sp = BLAKE2sp + deriving (Show) + +instance HashAlgorithm BLAKE2sp where + hashBlockSize _ = 64 + hashDigestSize _ = 32 + hashInternalContextSize _ = 2185 + hashInternalInit = c_blake2sp_init + hashInternalUpdate = c_blake2sp_update + hashInternalFinalize = c_blake2sp_finalize + +foreign import ccall unsafe "cryptonite_blake2sp_init" + c_blake2sp_init :: Ptr (Context a)-> IO () + +foreign import ccall "cryptonite_blake2sp_update" + c_blake2sp_update :: Ptr (Context a) -> Ptr Word8 -> Word32 -> IO () + +foreign import ccall unsafe "cryptonite_blake2sp_finalize" + c_blake2sp_finalize :: Ptr (Context a) -> Ptr (Digest a) -> IO () diff --git a/cbits/blake2/blake2-config.h b/cbits/blake2/blake2-config.h new file mode 100644 index 0000000..70d61f1 --- /dev/null +++ b/cbits/blake2/blake2-config.h @@ -0,0 +1,72 @@ +/* + BLAKE2 reference source code package - optimized C implementations + + Written in 2012 by Samuel Neves + + To the extent possible under law, the author(s) have dedicated all copyright + and related and neighboring rights to this software to the public domain + worldwide. This software is distributed without any warranty. + + You should have received a copy of the CC0 Public Domain Dedication along with + this software. If not, see . +*/ +#pragma once +#ifndef __BLAKE2_CONFIG_H__ +#define __BLAKE2_CONFIG_H__ + +// These don't work everywhere +#if defined(__SSE2__) +#define HAVE_SSE2 +#endif + +#if defined(__SSSE3__) +#define HAVE_SSSE3 +#endif + +#if defined(__SSE4_1__) +#define HAVE_SSE41 +#endif + +#if defined(__AVX__) +#define HAVE_AVX +#endif + +#if defined(__XOP__) +#define HAVE_XOP +#endif + + +#ifdef HAVE_AVX2 +#ifndef HAVE_AVX +#define HAVE_AVX +#endif +#endif + +#ifdef HAVE_XOP +#ifndef HAVE_AVX +#define HAVE_AVX +#endif +#endif + +#ifdef HAVE_AVX +#ifndef HAVE_SSE41 +#define HAVE_SSE41 +#endif +#endif + +#ifdef HAVE_SSE41 +#ifndef HAVE_SSSE3 +#define HAVE_SSSE3 +#endif +#endif + +#ifdef HAVE_SSSE3 +#define HAVE_SSE2 +#endif + +#if !defined(HAVE_SSE2) +#error "This code requires at least SSE2." +#endif + +#endif + diff --git a/cbits/blake2/blake2-impl.h b/cbits/blake2/blake2-impl.h new file mode 100644 index 0000000..16219db --- /dev/null +++ b/cbits/blake2/blake2-impl.h @@ -0,0 +1,136 @@ +/* + BLAKE2 reference source code package - optimized C implementations + + Written in 2012 by Samuel Neves + + To the extent possible under law, the author(s) have dedicated all copyright + and related and neighboring rights to this software to the public domain + worldwide. This software is distributed without any warranty. + + You should have received a copy of the CC0 Public Domain Dedication along with + this software. If not, see . +*/ +#pragma once +#ifndef __BLAKE2_IMPL_H__ +#define __BLAKE2_IMPL_H__ + +#include + +static inline uint32_t load32( const void *src ) +{ +#if defined(NATIVE_LITTLE_ENDIAN) + uint32_t w; + memcpy(&w, src, sizeof w); + return w; +#else + const uint8_t *p = ( const uint8_t * )src; + uint32_t w = *p++; + w |= ( uint32_t )( *p++ ) << 8; + w |= ( uint32_t )( *p++ ) << 16; + w |= ( uint32_t )( *p++ ) << 24; + return w; +#endif +} + +static inline uint64_t load64( const void *src ) +{ +#if defined(NATIVE_LITTLE_ENDIAN) + uint64_t w; + memcpy(&w, src, sizeof w); + return w; +#else + const uint8_t *p = ( const uint8_t * )src; + uint64_t w = *p++; + w |= ( uint64_t )( *p++ ) << 8; + w |= ( uint64_t )( *p++ ) << 16; + w |= ( uint64_t )( *p++ ) << 24; + w |= ( uint64_t )( *p++ ) << 32; + w |= ( uint64_t )( *p++ ) << 40; + w |= ( uint64_t )( *p++ ) << 48; + w |= ( uint64_t )( *p++ ) << 56; + return w; +#endif +} + +static inline void store32( void *dst, uint32_t w ) +{ +#if defined(NATIVE_LITTLE_ENDIAN) + memcpy(dst, &w, sizeof w); +#else + uint8_t *p = ( uint8_t * )dst; + *p++ = ( uint8_t )w; w >>= 8; + *p++ = ( uint8_t )w; w >>= 8; + *p++ = ( uint8_t )w; w >>= 8; + *p++ = ( uint8_t )w; +#endif +} + +static inline void store64( void *dst, uint64_t w ) +{ +#if defined(NATIVE_LITTLE_ENDIAN) + memcpy(dst, &w, sizeof w); +#else + uint8_t *p = ( uint8_t * )dst; + *p++ = ( uint8_t )w; w >>= 8; + *p++ = ( uint8_t )w; w >>= 8; + *p++ = ( uint8_t )w; w >>= 8; + *p++ = ( uint8_t )w; w >>= 8; + *p++ = ( uint8_t )w; w >>= 8; + *p++ = ( uint8_t )w; w >>= 8; + *p++ = ( uint8_t )w; w >>= 8; + *p++ = ( uint8_t )w; +#endif +} + +static inline uint64_t load48( const void *src ) +{ + const uint8_t *p = ( const uint8_t * )src; + uint64_t w = *p++; + w |= ( uint64_t )( *p++ ) << 8; + w |= ( uint64_t )( *p++ ) << 16; + w |= ( uint64_t )( *p++ ) << 24; + w |= ( uint64_t )( *p++ ) << 32; + w |= ( uint64_t )( *p++ ) << 40; + return w; +} + +static inline void store48( void *dst, uint64_t w ) +{ + uint8_t *p = ( uint8_t * )dst; + *p++ = ( uint8_t )w; w >>= 8; + *p++ = ( uint8_t )w; w >>= 8; + *p++ = ( uint8_t )w; w >>= 8; + *p++ = ( uint8_t )w; w >>= 8; + *p++ = ( uint8_t )w; w >>= 8; + *p++ = ( uint8_t )w; +} + +static inline uint32_t rotl32( const uint32_t w, const unsigned c ) +{ + return ( w << c ) | ( w >> ( 32 - c ) ); +} + +static inline uint64_t rotl64( const uint64_t w, const unsigned c ) +{ + return ( w << c ) | ( w >> ( 64 - c ) ); +} + +static inline uint32_t rotr32( const uint32_t w, const unsigned c ) +{ + return ( w >> c ) | ( w << ( 32 - c ) ); +} + +static inline uint64_t rotr64( const uint64_t w, const unsigned c ) +{ + return ( w >> c ) | ( w << ( 64 - c ) ); +} + +/* prevents compiler optimizing out memset() */ +static inline void secure_zero_memory( void *v, size_t n ) +{ + volatile uint8_t *p = ( volatile uint8_t * )v; + while( n-- ) *p++ = 0; +} + +#endif + diff --git a/cbits/blake2/blake2.h b/cbits/blake2/blake2.h new file mode 100644 index 0000000..20bc22d --- /dev/null +++ b/cbits/blake2/blake2.h @@ -0,0 +1,150 @@ +/* + BLAKE2 reference source code package - optimized C implementations + + Written in 2012 by Samuel Neves + + To the extent possible under law, the author(s) have dedicated all copyright + and related and neighboring rights to this software to the public domain + worldwide. This software is distributed without any warranty. + + You should have received a copy of the CC0 Public Domain Dedication along with + this software. If not, see . +*/ +#pragma once +#ifndef __BLAKE2_H__ +#define __BLAKE2_H__ + +#include +#include + +#if defined(__cplusplus) +extern "C" { +#endif + + enum blake2s_constant + { + BLAKE2S_BLOCKBYTES = 64, + BLAKE2S_OUTBYTES = 32, + BLAKE2S_KEYBYTES = 32, + BLAKE2S_SALTBYTES = 8, + BLAKE2S_PERSONALBYTES = 8 + }; + + enum blake2b_constant + { + BLAKE2B_BLOCKBYTES = 128, + BLAKE2B_OUTBYTES = 64, + BLAKE2B_KEYBYTES = 64, + BLAKE2B_SALTBYTES = 16, + BLAKE2B_PERSONALBYTES = 16 + }; + +#pragma pack(push, 1) + typedef struct __blake2s_param + { + uint8_t digest_length; // 1 + uint8_t key_length; // 2 + uint8_t fanout; // 3 + uint8_t depth; // 4 + uint32_t leaf_length; // 8 + uint8_t node_offset[6];// 14 + uint8_t node_depth; // 15 + uint8_t inner_length; // 16 + // uint8_t reserved[0]; + uint8_t salt[BLAKE2S_SALTBYTES]; // 24 + uint8_t personal[BLAKE2S_PERSONALBYTES]; // 32 + } blake2s_param; + + typedef struct __blake2s_state + { + uint32_t h[8]; + uint32_t t[2]; + uint32_t f[2]; + uint8_t buf[2 * BLAKE2S_BLOCKBYTES]; + size_t buflen; + uint8_t last_node; + } blake2s_state; + + typedef struct __blake2b_param + { + uint8_t digest_length; // 1 + uint8_t key_length; // 2 + uint8_t fanout; // 3 + uint8_t depth; // 4 + uint32_t leaf_length; // 8 + uint64_t node_offset; // 16 + uint8_t node_depth; // 17 + uint8_t inner_length; // 18 + uint8_t reserved[14]; // 32 + uint8_t salt[BLAKE2B_SALTBYTES]; // 48 + uint8_t personal[BLAKE2B_PERSONALBYTES]; // 64 + } blake2b_param; + + typedef struct __blake2b_state + { + uint64_t h[8]; + uint64_t t[2]; + uint64_t f[2]; + uint8_t buf[2 * BLAKE2B_BLOCKBYTES]; + size_t buflen; + uint8_t last_node; + } blake2b_state; + + typedef struct __blake2sp_state + { + blake2s_state S[8][1]; + blake2s_state R[1]; + uint8_t buf[8 * BLAKE2S_BLOCKBYTES]; + size_t buflen; + } blake2sp_state; + + typedef struct __blake2bp_state + { + blake2b_state S[4][1]; + blake2b_state R[1]; + uint8_t buf[4 * BLAKE2B_BLOCKBYTES]; + size_t buflen; + } blake2bp_state; +#pragma pack(pop) + + // Streaming API + int blake2s_init( blake2s_state *S, const uint8_t outlen ); + int blake2s_init_key( blake2s_state *S, const uint8_t outlen, const void *key, const uint8_t keylen ); + int blake2s_init_param( blake2s_state *S, const blake2s_param *P ); + int blake2s_update( blake2s_state *S, const uint8_t *in, uint64_t inlen ); + int blake2s_final( blake2s_state *S, uint8_t *out, uint8_t outlen ); + + int blake2b_init( blake2b_state *S, const uint8_t outlen ); + int blake2b_init_key( blake2b_state *S, const uint8_t outlen, const void *key, const uint8_t keylen ); + int blake2b_init_param( blake2b_state *S, const blake2b_param *P ); + int blake2b_update( blake2b_state *S, const uint8_t *in, uint64_t inlen ); + int blake2b_final( blake2b_state *S, uint8_t *out, uint8_t outlen ); + + int blake2sp_init( blake2sp_state *S, const uint8_t outlen ); + int blake2sp_init_key( blake2sp_state *S, const uint8_t outlen, const void *key, const uint8_t keylen ); + int blake2sp_update( blake2sp_state *S, const uint8_t *in, uint64_t inlen ); + int blake2sp_final( blake2sp_state *S, uint8_t *out, uint8_t outlen ); + + int blake2bp_init( blake2bp_state *S, const uint8_t outlen ); + int blake2bp_init_key( blake2bp_state *S, const uint8_t outlen, const void *key, const uint8_t keylen ); + int blake2bp_update( blake2bp_state *S, const uint8_t *in, uint64_t inlen ); + int blake2bp_final( blake2bp_state *S, uint8_t *out, uint8_t outlen ); + + // Simple API + int blake2s( uint8_t *out, const void *in, const void *key, const uint8_t outlen, const uint64_t inlen, uint8_t keylen ); + int blake2b( uint8_t *out, const void *in, const void *key, const uint8_t outlen, const uint64_t inlen, uint8_t keylen ); + + int blake2sp( uint8_t *out, const void *in, const void *key, const uint8_t outlen, const uint64_t inlen, uint8_t keylen ); + int blake2bp( uint8_t *out, const void *in, const void *key, const uint8_t outlen, const uint64_t inlen, uint8_t keylen ); + + static inline int blake2( uint8_t *out, const void *in, const void *key, const uint8_t outlen, const uint64_t inlen, uint8_t keylen ) + { + return blake2b( out, in, key, outlen, inlen, keylen ); + } + +#if defined(__cplusplus) +} +#endif + +#endif + diff --git a/cbits/blake2/blake2b-load-sse2.h b/cbits/blake2/blake2b-load-sse2.h new file mode 100644 index 0000000..1ba153c --- /dev/null +++ b/cbits/blake2/blake2b-load-sse2.h @@ -0,0 +1,68 @@ +/* + BLAKE2 reference source code package - optimized C implementations + + Written in 2012 by Samuel Neves + + To the extent possible under law, the author(s) have dedicated all copyright + and related and neighboring rights to this software to the public domain + worldwide. This software is distributed without any warranty. + + You should have received a copy of the CC0 Public Domain Dedication along with + this software. If not, see . +*/ +#pragma once +#ifndef __BLAKE2B_LOAD_SSE2_H__ +#define __BLAKE2B_LOAD_SSE2_H__ + +#define LOAD_MSG_0_1(b0, b1) b0 = _mm_set_epi64x(m2, m0); b1 = _mm_set_epi64x(m6, m4) +#define LOAD_MSG_0_2(b0, b1) b0 = _mm_set_epi64x(m3, m1); b1 = _mm_set_epi64x(m7, m5) +#define LOAD_MSG_0_3(b0, b1) b0 = _mm_set_epi64x(m10, m8); b1 = _mm_set_epi64x(m14, m12) +#define LOAD_MSG_0_4(b0, b1) b0 = _mm_set_epi64x(m11, m9); b1 = _mm_set_epi64x(m15, m13) +#define LOAD_MSG_1_1(b0, b1) b0 = _mm_set_epi64x(m4, m14); b1 = _mm_set_epi64x(m13, m9) +#define LOAD_MSG_1_2(b0, b1) b0 = _mm_set_epi64x(m8, m10); b1 = _mm_set_epi64x(m6, m15) +#define LOAD_MSG_1_3(b0, b1) b0 = _mm_set_epi64x(m0, m1); b1 = _mm_set_epi64x(m5, m11) +#define LOAD_MSG_1_4(b0, b1) b0 = _mm_set_epi64x(m2, m12); b1 = _mm_set_epi64x(m3, m7) +#define LOAD_MSG_2_1(b0, b1) b0 = _mm_set_epi64x(m12, m11); b1 = _mm_set_epi64x(m15, m5) +#define LOAD_MSG_2_2(b0, b1) b0 = _mm_set_epi64x(m0, m8); b1 = _mm_set_epi64x(m13, m2) +#define LOAD_MSG_2_3(b0, b1) b0 = _mm_set_epi64x(m3, m10); b1 = _mm_set_epi64x(m9, m7) +#define LOAD_MSG_2_4(b0, b1) b0 = _mm_set_epi64x(m6, m14); b1 = _mm_set_epi64x(m4, m1) +#define LOAD_MSG_3_1(b0, b1) b0 = _mm_set_epi64x(m3, m7); b1 = _mm_set_epi64x(m11, m13) +#define LOAD_MSG_3_2(b0, b1) b0 = _mm_set_epi64x(m1, m9); b1 = _mm_set_epi64x(m14, m12) +#define LOAD_MSG_3_3(b0, b1) b0 = _mm_set_epi64x(m5, m2); b1 = _mm_set_epi64x(m15, m4) +#define LOAD_MSG_3_4(b0, b1) b0 = _mm_set_epi64x(m10, m6); b1 = _mm_set_epi64x(m8, m0) +#define LOAD_MSG_4_1(b0, b1) b0 = _mm_set_epi64x(m5, m9); b1 = _mm_set_epi64x(m10, m2) +#define LOAD_MSG_4_2(b0, b1) b0 = _mm_set_epi64x(m7, m0); b1 = _mm_set_epi64x(m15, m4) +#define LOAD_MSG_4_3(b0, b1) b0 = _mm_set_epi64x(m11, m14); b1 = _mm_set_epi64x(m3, m6) +#define LOAD_MSG_4_4(b0, b1) b0 = _mm_set_epi64x(m12, m1); b1 = _mm_set_epi64x(m13, m8) +#define LOAD_MSG_5_1(b0, b1) b0 = _mm_set_epi64x(m6, m2); b1 = _mm_set_epi64x(m8, m0) +#define LOAD_MSG_5_2(b0, b1) b0 = _mm_set_epi64x(m10, m12); b1 = _mm_set_epi64x(m3, m11) +#define LOAD_MSG_5_3(b0, b1) b0 = _mm_set_epi64x(m7, m4); b1 = _mm_set_epi64x(m1, m15) +#define LOAD_MSG_5_4(b0, b1) b0 = _mm_set_epi64x(m5, m13); b1 = _mm_set_epi64x(m9, m14) +#define LOAD_MSG_6_1(b0, b1) b0 = _mm_set_epi64x(m1, m12); b1 = _mm_set_epi64x(m4, m14) +#define LOAD_MSG_6_2(b0, b1) b0 = _mm_set_epi64x(m15, m5); b1 = _mm_set_epi64x(m10, m13) +#define LOAD_MSG_6_3(b0, b1) b0 = _mm_set_epi64x(m6, m0); b1 = _mm_set_epi64x(m8, m9) +#define LOAD_MSG_6_4(b0, b1) b0 = _mm_set_epi64x(m3, m7); b1 = _mm_set_epi64x(m11, m2) +#define LOAD_MSG_7_1(b0, b1) b0 = _mm_set_epi64x(m7, m13); b1 = _mm_set_epi64x(m3, m12) +#define LOAD_MSG_7_2(b0, b1) b0 = _mm_set_epi64x(m14, m11); b1 = _mm_set_epi64x(m9, m1) +#define LOAD_MSG_7_3(b0, b1) b0 = _mm_set_epi64x(m15, m5); b1 = _mm_set_epi64x(m2, m8) +#define LOAD_MSG_7_4(b0, b1) b0 = _mm_set_epi64x(m4, m0); b1 = _mm_set_epi64x(m10, m6) +#define LOAD_MSG_8_1(b0, b1) b0 = _mm_set_epi64x(m14, m6); b1 = _mm_set_epi64x(m0, m11) +#define LOAD_MSG_8_2(b0, b1) b0 = _mm_set_epi64x(m9, m15); b1 = _mm_set_epi64x(m8, m3) +#define LOAD_MSG_8_3(b0, b1) b0 = _mm_set_epi64x(m13, m12); b1 = _mm_set_epi64x(m10, m1) +#define LOAD_MSG_8_4(b0, b1) b0 = _mm_set_epi64x(m7, m2); b1 = _mm_set_epi64x(m5, m4) +#define LOAD_MSG_9_1(b0, b1) b0 = _mm_set_epi64x(m8, m10); b1 = _mm_set_epi64x(m1, m7) +#define LOAD_MSG_9_2(b0, b1) b0 = _mm_set_epi64x(m4, m2); b1 = _mm_set_epi64x(m5, m6) +#define LOAD_MSG_9_3(b0, b1) b0 = _mm_set_epi64x(m9, m15); b1 = _mm_set_epi64x(m13, m3) +#define LOAD_MSG_9_4(b0, b1) b0 = _mm_set_epi64x(m14, m11); b1 = _mm_set_epi64x(m0, m12) +#define LOAD_MSG_10_1(b0, b1) b0 = _mm_set_epi64x(m2, m0); b1 = _mm_set_epi64x(m6, m4) +#define LOAD_MSG_10_2(b0, b1) b0 = _mm_set_epi64x(m3, m1); b1 = _mm_set_epi64x(m7, m5) +#define LOAD_MSG_10_3(b0, b1) b0 = _mm_set_epi64x(m10, m8); b1 = _mm_set_epi64x(m14, m12) +#define LOAD_MSG_10_4(b0, b1) b0 = _mm_set_epi64x(m11, m9); b1 = _mm_set_epi64x(m15, m13) +#define LOAD_MSG_11_1(b0, b1) b0 = _mm_set_epi64x(m4, m14); b1 = _mm_set_epi64x(m13, m9) +#define LOAD_MSG_11_2(b0, b1) b0 = _mm_set_epi64x(m8, m10); b1 = _mm_set_epi64x(m6, m15) +#define LOAD_MSG_11_3(b0, b1) b0 = _mm_set_epi64x(m0, m1); b1 = _mm_set_epi64x(m5, m11) +#define LOAD_MSG_11_4(b0, b1) b0 = _mm_set_epi64x(m2, m12); b1 = _mm_set_epi64x(m3, m7) + + +#endif + diff --git a/cbits/blake2/blake2b-load-sse41.h b/cbits/blake2/blake2b-load-sse41.h new file mode 100644 index 0000000..f6c1bc8 --- /dev/null +++ b/cbits/blake2/blake2b-load-sse41.h @@ -0,0 +1,402 @@ +/* + BLAKE2 reference source code package - optimized C implementations + + Written in 2012 by Samuel Neves + + To the extent possible under law, the author(s) have dedicated all copyright + and related and neighboring rights to this software to the public domain + worldwide. This software is distributed without any warranty. + + You should have received a copy of the CC0 Public Domain Dedication along with + this software. If not, see . +*/ +#pragma once +#ifndef __BLAKE2B_LOAD_SSE41_H__ +#define __BLAKE2B_LOAD_SSE41_H__ + +#define LOAD_MSG_0_1(b0, b1) \ +do \ +{ \ +b0 = _mm_unpacklo_epi64(m0, m1); \ +b1 = _mm_unpacklo_epi64(m2, m3); \ +} while(0) + + +#define LOAD_MSG_0_2(b0, b1) \ +do \ +{ \ +b0 = _mm_unpackhi_epi64(m0, m1); \ +b1 = _mm_unpackhi_epi64(m2, m3); \ +} while(0) + + +#define LOAD_MSG_0_3(b0, b1) \ +do \ +{ \ +b0 = _mm_unpacklo_epi64(m4, m5); \ +b1 = _mm_unpacklo_epi64(m6, m7); \ +} while(0) + + +#define LOAD_MSG_0_4(b0, b1) \ +do \ +{ \ +b0 = _mm_unpackhi_epi64(m4, m5); \ +b1 = _mm_unpackhi_epi64(m6, m7); \ +} while(0) + + +#define LOAD_MSG_1_1(b0, b1) \ +do \ +{ \ +b0 = _mm_unpacklo_epi64(m7, m2); \ +b1 = _mm_unpackhi_epi64(m4, m6); \ +} while(0) + + +#define LOAD_MSG_1_2(b0, b1) \ +do \ +{ \ +b0 = _mm_unpacklo_epi64(m5, m4); \ +b1 = _mm_alignr_epi8(m3, m7, 8); \ +} while(0) + + +#define LOAD_MSG_1_3(b0, b1) \ +do \ +{ \ +b0 = _mm_shuffle_epi32(m0, _MM_SHUFFLE(1,0,3,2)); \ +b1 = _mm_unpackhi_epi64(m5, m2); \ +} while(0) + + +#define LOAD_MSG_1_4(b0, b1) \ +do \ +{ \ +b0 = _mm_unpacklo_epi64(m6, m1); \ +b1 = _mm_unpackhi_epi64(m3, m1); \ +} while(0) + + +#define LOAD_MSG_2_1(b0, b1) \ +do \ +{ \ +b0 = _mm_alignr_epi8(m6, m5, 8); \ +b1 = _mm_unpackhi_epi64(m2, m7); \ +} while(0) + + +#define LOAD_MSG_2_2(b0, b1) \ +do \ +{ \ +b0 = _mm_unpacklo_epi64(m4, m0); \ +b1 = _mm_blend_epi16(m1, m6, 0xF0); \ +} while(0) + + +#define LOAD_MSG_2_3(b0, b1) \ +do \ +{ \ +b0 = _mm_blend_epi16(m5, m1, 0xF0); \ +b1 = _mm_unpackhi_epi64(m3, m4); \ +} while(0) + + +#define LOAD_MSG_2_4(b0, b1) \ +do \ +{ \ +b0 = _mm_unpacklo_epi64(m7, m3); \ +b1 = _mm_alignr_epi8(m2, m0, 8); \ +} while(0) + + +#define LOAD_MSG_3_1(b0, b1) \ +do \ +{ \ +b0 = _mm_unpackhi_epi64(m3, m1); \ +b1 = _mm_unpackhi_epi64(m6, m5); \ +} while(0) + + +#define LOAD_MSG_3_2(b0, b1) \ +do \ +{ \ +b0 = _mm_unpackhi_epi64(m4, m0); \ +b1 = _mm_unpacklo_epi64(m6, m7); \ +} while(0) + + +#define LOAD_MSG_3_3(b0, b1) \ +do \ +{ \ +b0 = _mm_blend_epi16(m1, m2, 0xF0); \ +b1 = _mm_blend_epi16(m2, m7, 0xF0); \ +} while(0) + + +#define LOAD_MSG_3_4(b0, b1) \ +do \ +{ \ +b0 = _mm_unpacklo_epi64(m3, m5); \ +b1 = _mm_unpacklo_epi64(m0, m4); \ +} while(0) + + +#define LOAD_MSG_4_1(b0, b1) \ +do \ +{ \ +b0 = _mm_unpackhi_epi64(m4, m2); \ +b1 = _mm_unpacklo_epi64(m1, m5); \ +} while(0) + + +#define LOAD_MSG_4_2(b0, b1) \ +do \ +{ \ +b0 = _mm_blend_epi16(m0, m3, 0xF0); \ +b1 = _mm_blend_epi16(m2, m7, 0xF0); \ +} while(0) + + +#define LOAD_MSG_4_3(b0, b1) \ +do \ +{ \ +b0 = _mm_blend_epi16(m7, m5, 0xF0); \ +b1 = _mm_blend_epi16(m3, m1, 0xF0); \ +} while(0) + + +#define LOAD_MSG_4_4(b0, b1) \ +do \ +{ \ +b0 = _mm_alignr_epi8(m6, m0, 8); \ +b1 = _mm_blend_epi16(m4, m6, 0xF0); \ +} while(0) + + +#define LOAD_MSG_5_1(b0, b1) \ +do \ +{ \ +b0 = _mm_unpacklo_epi64(m1, m3); \ +b1 = _mm_unpacklo_epi64(m0, m4); \ +} while(0) + + +#define LOAD_MSG_5_2(b0, b1) \ +do \ +{ \ +b0 = _mm_unpacklo_epi64(m6, m5); \ +b1 = _mm_unpackhi_epi64(m5, m1); \ +} while(0) + + +#define LOAD_MSG_5_3(b0, b1) \ +do \ +{ \ +b0 = _mm_blend_epi16(m2, m3, 0xF0); \ +b1 = _mm_unpackhi_epi64(m7, m0); \ +} while(0) + + +#define LOAD_MSG_5_4(b0, b1) \ +do \ +{ \ +b0 = _mm_unpackhi_epi64(m6, m2); \ +b1 = _mm_blend_epi16(m7, m4, 0xF0); \ +} while(0) + + +#define LOAD_MSG_6_1(b0, b1) \ +do \ +{ \ +b0 = _mm_blend_epi16(m6, m0, 0xF0); \ +b1 = _mm_unpacklo_epi64(m7, m2); \ +} while(0) + + +#define LOAD_MSG_6_2(b0, b1) \ +do \ +{ \ +b0 = _mm_unpackhi_epi64(m2, m7); \ +b1 = _mm_alignr_epi8(m5, m6, 8); \ +} while(0) + + +#define LOAD_MSG_6_3(b0, b1) \ +do \ +{ \ +b0 = _mm_unpacklo_epi64(m0, m3); \ +b1 = _mm_shuffle_epi32(m4, _MM_SHUFFLE(1,0,3,2)); \ +} while(0) + + +#define LOAD_MSG_6_4(b0, b1) \ +do \ +{ \ +b0 = _mm_unpackhi_epi64(m3, m1); \ +b1 = _mm_blend_epi16(m1, m5, 0xF0); \ +} while(0) + + +#define LOAD_MSG_7_1(b0, b1) \ +do \ +{ \ +b0 = _mm_unpackhi_epi64(m6, m3); \ +b1 = _mm_blend_epi16(m6, m1, 0xF0); \ +} while(0) + + +#define LOAD_MSG_7_2(b0, b1) \ +do \ +{ \ +b0 = _mm_alignr_epi8(m7, m5, 8); \ +b1 = _mm_unpackhi_epi64(m0, m4); \ +} while(0) + + +#define LOAD_MSG_7_3(b0, b1) \ +do \ +{ \ +b0 = _mm_unpackhi_epi64(m2, m7); \ +b1 = _mm_unpacklo_epi64(m4, m1); \ +} while(0) + + +#define LOAD_MSG_7_4(b0, b1) \ +do \ +{ \ +b0 = _mm_unpacklo_epi64(m0, m2); \ +b1 = _mm_unpacklo_epi64(m3, m5); \ +} while(0) + + +#define LOAD_MSG_8_1(b0, b1) \ +do \ +{ \ +b0 = _mm_unpacklo_epi64(m3, m7); \ +b1 = _mm_alignr_epi8(m0, m5, 8); \ +} while(0) + + +#define LOAD_MSG_8_2(b0, b1) \ +do \ +{ \ +b0 = _mm_unpackhi_epi64(m7, m4); \ +b1 = _mm_alignr_epi8(m4, m1, 8); \ +} while(0) + + +#define LOAD_MSG_8_3(b0, b1) \ +do \ +{ \ +b0 = m6; \ +b1 = _mm_alignr_epi8(m5, m0, 8); \ +} while(0) + + +#define LOAD_MSG_8_4(b0, b1) \ +do \ +{ \ +b0 = _mm_blend_epi16(m1, m3, 0xF0); \ +b1 = m2; \ +} while(0) + + +#define LOAD_MSG_9_1(b0, b1) \ +do \ +{ \ +b0 = _mm_unpacklo_epi64(m5, m4); \ +b1 = _mm_unpackhi_epi64(m3, m0); \ +} while(0) + + +#define LOAD_MSG_9_2(b0, b1) \ +do \ +{ \ +b0 = _mm_unpacklo_epi64(m1, m2); \ +b1 = _mm_blend_epi16(m3, m2, 0xF0); \ +} while(0) + + +#define LOAD_MSG_9_3(b0, b1) \ +do \ +{ \ +b0 = _mm_unpackhi_epi64(m7, m4); \ +b1 = _mm_unpackhi_epi64(m1, m6); \ +} while(0) + + +#define LOAD_MSG_9_4(b0, b1) \ +do \ +{ \ +b0 = _mm_alignr_epi8(m7, m5, 8); \ +b1 = _mm_unpacklo_epi64(m6, m0); \ +} while(0) + + +#define LOAD_MSG_10_1(b0, b1) \ +do \ +{ \ +b0 = _mm_unpacklo_epi64(m0, m1); \ +b1 = _mm_unpacklo_epi64(m2, m3); \ +} while(0) + + +#define LOAD_MSG_10_2(b0, b1) \ +do \ +{ \ +b0 = _mm_unpackhi_epi64(m0, m1); \ +b1 = _mm_unpackhi_epi64(m2, m3); \ +} while(0) + + +#define LOAD_MSG_10_3(b0, b1) \ +do \ +{ \ +b0 = _mm_unpacklo_epi64(m4, m5); \ +b1 = _mm_unpacklo_epi64(m6, m7); \ +} while(0) + + +#define LOAD_MSG_10_4(b0, b1) \ +do \ +{ \ +b0 = _mm_unpackhi_epi64(m4, m5); \ +b1 = _mm_unpackhi_epi64(m6, m7); \ +} while(0) + + +#define LOAD_MSG_11_1(b0, b1) \ +do \ +{ \ +b0 = _mm_unpacklo_epi64(m7, m2); \ +b1 = _mm_unpackhi_epi64(m4, m6); \ +} while(0) + + +#define LOAD_MSG_11_2(b0, b1) \ +do \ +{ \ +b0 = _mm_unpacklo_epi64(m5, m4); \ +b1 = _mm_alignr_epi8(m3, m7, 8); \ +} while(0) + + +#define LOAD_MSG_11_3(b0, b1) \ +do \ +{ \ +b0 = _mm_shuffle_epi32(m0, _MM_SHUFFLE(1,0,3,2)); \ +b1 = _mm_unpackhi_epi64(m5, m2); \ +} while(0) + + +#define LOAD_MSG_11_4(b0, b1) \ +do \ +{ \ +b0 = _mm_unpacklo_epi64(m6, m1); \ +b1 = _mm_unpackhi_epi64(m3, m1); \ +} while(0) + + +#endif + diff --git a/cbits/blake2/blake2b-round.h b/cbits/blake2/blake2b-round.h new file mode 100644 index 0000000..200b066 --- /dev/null +++ b/cbits/blake2/blake2b-round.h @@ -0,0 +1,157 @@ +/* + BLAKE2 reference source code package - optimized C implementations + + Written in 2012 by Samuel Neves + + To the extent possible under law, the author(s) have dedicated all copyright + and related and neighboring rights to this software to the public domain + worldwide. This software is distributed without any warranty. + + You should have received a copy of the CC0 Public Domain Dedication along with + this software. If not, see . +*/ +#pragma once +#ifndef __BLAKE2B_ROUND_H__ +#define __BLAKE2B_ROUND_H__ + +#define LOADU(p) _mm_loadu_si128( (const __m128i *)(p) ) +#define STOREU(p,r) _mm_storeu_si128((__m128i *)(p), r) + +#define TOF(reg) _mm_castsi128_ps((reg)) +#define TOI(reg) _mm_castps_si128((reg)) + +#define LIKELY(x) __builtin_expect((x),1) + + +/* Microarchitecture-specific macros */ +#ifndef HAVE_XOP +#ifdef HAVE_SSSE3 +#define _mm_roti_epi64(x, c) \ + (-(c) == 32) ? _mm_shuffle_epi32((x), _MM_SHUFFLE(2,3,0,1)) \ + : (-(c) == 24) ? _mm_shuffle_epi8((x), r24) \ + : (-(c) == 16) ? _mm_shuffle_epi8((x), r16) \ + : (-(c) == 63) ? _mm_xor_si128(_mm_srli_epi64((x), -(c)), _mm_add_epi64((x), (x))) \ + : _mm_xor_si128(_mm_srli_epi64((x), -(c)), _mm_slli_epi64((x), 64-(-(c)))) +#else +#define _mm_roti_epi64(r, c) _mm_xor_si128(_mm_srli_epi64( (r), -(c) ),_mm_slli_epi64( (r), 64-(-(c)) )) +#endif +#else +/* ... */ +#endif + + + +#define G1(row1l,row2l,row3l,row4l,row1h,row2h,row3h,row4h,b0,b1) \ + row1l = _mm_add_epi64(_mm_add_epi64(row1l, b0), row2l); \ + row1h = _mm_add_epi64(_mm_add_epi64(row1h, b1), row2h); \ + \ + row4l = _mm_xor_si128(row4l, row1l); \ + row4h = _mm_xor_si128(row4h, row1h); \ + \ + row4l = _mm_roti_epi64(row4l, -32); \ + row4h = _mm_roti_epi64(row4h, -32); \ + \ + row3l = _mm_add_epi64(row3l, row4l); \ + row3h = _mm_add_epi64(row3h, row4h); \ + \ + row2l = _mm_xor_si128(row2l, row3l); \ + row2h = _mm_xor_si128(row2h, row3h); \ + \ + row2l = _mm_roti_epi64(row2l, -24); \ + row2h = _mm_roti_epi64(row2h, -24); \ + +#define G2(row1l,row2l,row3l,row4l,row1h,row2h,row3h,row4h,b0,b1) \ + row1l = _mm_add_epi64(_mm_add_epi64(row1l, b0), row2l); \ + row1h = _mm_add_epi64(_mm_add_epi64(row1h, b1), row2h); \ + \ + row4l = _mm_xor_si128(row4l, row1l); \ + row4h = _mm_xor_si128(row4h, row1h); \ + \ + row4l = _mm_roti_epi64(row4l, -16); \ + row4h = _mm_roti_epi64(row4h, -16); \ + \ + row3l = _mm_add_epi64(row3l, row4l); \ + row3h = _mm_add_epi64(row3h, row4h); \ + \ + row2l = _mm_xor_si128(row2l, row3l); \ + row2h = _mm_xor_si128(row2h, row3h); \ + \ + row2l = _mm_roti_epi64(row2l, -63); \ + row2h = _mm_roti_epi64(row2h, -63); \ + +#if defined(HAVE_SSSE3) +#define DIAGONALIZE(row1l,row2l,row3l,row4l,row1h,row2h,row3h,row4h) \ + t0 = _mm_alignr_epi8(row2h, row2l, 8); \ + t1 = _mm_alignr_epi8(row2l, row2h, 8); \ + row2l = t0; \ + row2h = t1; \ + \ + t0 = row3l; \ + row3l = row3h; \ + row3h = t0; \ + \ + t0 = _mm_alignr_epi8(row4h, row4l, 8); \ + t1 = _mm_alignr_epi8(row4l, row4h, 8); \ + row4l = t1; \ + row4h = t0; + +#define UNDIAGONALIZE(row1l,row2l,row3l,row4l,row1h,row2h,row3h,row4h) \ + t0 = _mm_alignr_epi8(row2l, row2h, 8); \ + t1 = _mm_alignr_epi8(row2h, row2l, 8); \ + row2l = t0; \ + row2h = t1; \ + \ + t0 = row3l; \ + row3l = row3h; \ + row3h = t0; \ + \ + t0 = _mm_alignr_epi8(row4l, row4h, 8); \ + t1 = _mm_alignr_epi8(row4h, row4l, 8); \ + row4l = t1; \ + row4h = t0; +#else + +#define DIAGONALIZE(row1l,row2l,row3l,row4l,row1h,row2h,row3h,row4h) \ + t0 = row4l;\ + t1 = row2l;\ + row4l = row3l;\ + row3l = row3h;\ + row3h = row4l;\ + row4l = _mm_unpackhi_epi64(row4h, _mm_unpacklo_epi64(t0, t0)); \ + row4h = _mm_unpackhi_epi64(t0, _mm_unpacklo_epi64(row4h, row4h)); \ + row2l = _mm_unpackhi_epi64(row2l, _mm_unpacklo_epi64(row2h, row2h)); \ + row2h = _mm_unpackhi_epi64(row2h, _mm_unpacklo_epi64(t1, t1)) + +#define UNDIAGONALIZE(row1l,row2l,row3l,row4l,row1h,row2h,row3h,row4h) \ + t0 = row3l;\ + row3l = row3h;\ + row3h = t0;\ + t0 = row2l;\ + t1 = row4l;\ + row2l = _mm_unpackhi_epi64(row2h, _mm_unpacklo_epi64(row2l, row2l)); \ + row2h = _mm_unpackhi_epi64(t0, _mm_unpacklo_epi64(row2h, row2h)); \ + row4l = _mm_unpackhi_epi64(row4l, _mm_unpacklo_epi64(row4h, row4h)); \ + row4h = _mm_unpackhi_epi64(row4h, _mm_unpacklo_epi64(t1, t1)) + +#endif + +#if defined(HAVE_SSE41) +#include "blake2b-load-sse41.h" +#else +#include "blake2b-load-sse2.h" +#endif + +#define ROUND(r) \ + LOAD_MSG_ ##r ##_1(b0, b1); \ + G1(row1l,row2l,row3l,row4l,row1h,row2h,row3h,row4h,b0,b1); \ + LOAD_MSG_ ##r ##_2(b0, b1); \ + G2(row1l,row2l,row3l,row4l,row1h,row2h,row3h,row4h,b0,b1); \ + DIAGONALIZE(row1l,row2l,row3l,row4l,row1h,row2h,row3h,row4h); \ + LOAD_MSG_ ##r ##_3(b0, b1); \ + G1(row1l,row2l,row3l,row4l,row1h,row2h,row3h,row4h,b0,b1); \ + LOAD_MSG_ ##r ##_4(b0, b1); \ + G2(row1l,row2l,row3l,row4l,row1h,row2h,row3h,row4h,b0,b1); \ + UNDIAGONALIZE(row1l,row2l,row3l,row4l,row1h,row2h,row3h,row4h); + +#endif + diff --git a/cbits/blake2/blake2b.c b/cbits/blake2/blake2b.c new file mode 100644 index 0000000..7026d62 --- /dev/null +++ b/cbits/blake2/blake2b.c @@ -0,0 +1,403 @@ +/* + BLAKE2 reference source code package - optimized C implementations + + Written in 2012 by Samuel Neves + + To the extent possible under law, the author(s) have dedicated all copyright + and related and neighboring rights to this software to the public domain + worldwide. This software is distributed without any warranty. + + You should have received a copy of the CC0 Public Domain Dedication along with + this software. If not, see . +*/ + +#include +#include +#include + +#include "blake2.h" +#include "blake2-impl.h" + +#include "blake2-config.h" + +#ifdef _MSC_VER +#include /* for _mm_set_epi64x */ +#endif +#include +#if defined(HAVE_SSSE3) +#include +#endif +#if defined(HAVE_SSE41) +#include +#endif +#if defined(HAVE_AVX) +#include +#endif +#if defined(HAVE_XOP) +#include +#endif + +#include "blake2b-round.h" + +static const uint64_t blake2b_IV[8] = +{ + 0x6a09e667f3bcc908ULL, 0xbb67ae8584caa73bULL, + 0x3c6ef372fe94f82bULL, 0xa54ff53a5f1d36f1ULL, + 0x510e527fade682d1ULL, 0x9b05688c2b3e6c1fULL, + 0x1f83d9abfb41bd6bULL, 0x5be0cd19137e2179ULL +}; + +static const uint8_t blake2b_sigma[12][16] = +{ + { 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15 } , + { 14, 10, 4, 8, 9, 15, 13, 6, 1, 12, 0, 2, 11, 7, 5, 3 } , + { 11, 8, 12, 0, 5, 2, 15, 13, 10, 14, 3, 6, 7, 1, 9, 4 } , + { 7, 9, 3, 1, 13, 12, 11, 14, 2, 6, 5, 10, 4, 0, 15, 8 } , + { 9, 0, 5, 7, 2, 4, 10, 15, 14, 1, 11, 12, 6, 8, 3, 13 } , + { 2, 12, 6, 10, 0, 11, 8, 3, 4, 13, 7, 5, 15, 14, 1, 9 } , + { 12, 5, 1, 15, 14, 13, 4, 10, 0, 7, 6, 3, 9, 2, 8, 11 } , + { 13, 11, 7, 14, 12, 1, 3, 9, 5, 0, 15, 4, 8, 6, 2, 10 } , + { 6, 15, 14, 9, 11, 3, 0, 8, 12, 2, 13, 7, 1, 4, 10, 5 } , + { 10, 2, 8, 4, 7, 6, 1, 5, 15, 11, 9, 14, 3, 12, 13 , 0 } , + { 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15 } , + { 14, 10, 4, 8, 9, 15, 13, 6, 1, 12, 0, 2, 11, 7, 5, 3 } +}; + + +/* Some helper functions, not necessarily useful */ +static inline int blake2b_set_lastnode( blake2b_state *S ) +{ + S->f[1] = -1; + return 0; +} + +static inline int blake2b_clear_lastnode( blake2b_state *S ) +{ + S->f[1] = 0; + return 0; +} + +static inline int blake2b_set_lastblock( blake2b_state *S ) +{ + if( S->last_node ) blake2b_set_lastnode( S ); + + S->f[0] = -1; + return 0; +} + +static inline int blake2b_clear_lastblock( blake2b_state *S ) +{ + if( S->last_node ) blake2b_clear_lastnode( S ); + + S->f[0] = 0; + return 0; +} + + +static inline int blake2b_increment_counter( blake2b_state *S, const uint64_t inc ) +{ +#if __x86_64__ + // ADD/ADC chain + __uint128_t t = ( ( __uint128_t )S->t[1] << 64 ) | S->t[0]; + t += inc; + S->t[0] = ( uint64_t )( t >> 0 ); + S->t[1] = ( uint64_t )( t >> 64 ); +#else + S->t[0] += inc; + S->t[1] += ( S->t[0] < inc ); +#endif + return 0; +} + + +// Parameter-related functions +static inline int blake2b_param_set_digest_length( blake2b_param *P, const uint8_t digest_length ) +{ + P->digest_length = digest_length; + return 0; +} + +static inline int blake2b_param_set_fanout( blake2b_param *P, const uint8_t fanout ) +{ + P->fanout = fanout; + return 0; +} + +static inline int blake2b_param_set_max_depth( blake2b_param *P, const uint8_t depth ) +{ + P->depth = depth; + return 0; +} + +static inline int blake2b_param_set_leaf_length( blake2b_param *P, const uint32_t leaf_length ) +{ + P->leaf_length = leaf_length; + return 0; +} + +static inline int blake2b_param_set_node_offset( blake2b_param *P, const uint64_t node_offset ) +{ + P->node_offset = node_offset; + return 0; +} + +static inline int blake2b_param_set_node_depth( blake2b_param *P, const uint8_t node_depth ) +{ + P->node_depth = node_depth; + return 0; +} + +static inline int blake2b_param_set_inner_length( blake2b_param *P, const uint8_t inner_length ) +{ + P->inner_length = inner_length; + return 0; +} + +static inline int blake2b_param_set_salt( blake2b_param *P, const uint8_t salt[BLAKE2B_SALTBYTES] ) +{ + memcpy( P->salt, salt, BLAKE2B_SALTBYTES ); + return 0; +} + +static inline int blake2b_param_set_personal( blake2b_param *P, const uint8_t personal[BLAKE2B_PERSONALBYTES] ) +{ + memcpy( P->personal, personal, BLAKE2B_PERSONALBYTES ); + return 0; +} + +static inline int blake2b_init0( blake2b_state *S ) +{ + memset( S, 0, sizeof( blake2b_state ) ); + + for( int i = 0; i < 8; ++i ) S->h[i] = blake2b_IV[i]; + + return 0; +} + +/* init xors IV with input parameter block */ +__attribute__((visibility ("hidden"))) +int blake2b_init_param( blake2b_state *S, const blake2b_param *P ) +{ + //blake2b_init0( S ); + const uint8_t * v = ( const uint8_t * )( blake2b_IV ); + const uint8_t * p = ( const uint8_t * )( P ); + uint8_t * h = ( uint8_t * )( S->h ); + /* IV XOR ParamBlock */ + memset( S, 0, sizeof( blake2b_state ) ); + + for( int i = 0; i < BLAKE2B_OUTBYTES; ++i ) h[i] = v[i] ^ p[i]; + + return 0; +} + + +/* Some sort of default parameter block initialization, for sequential blake2b */ +__attribute__((visibility ("hidden"))) +int blake2b_init( blake2b_state *S, const uint8_t outlen ) +{ + if ( ( !outlen ) || ( outlen > BLAKE2B_OUTBYTES ) ) return -1; + + const blake2b_param P = + { + outlen, + 0, + 1, + 1, + 0, + 0, + 0, + 0, + {0}, + {0}, + {0} + }; + return blake2b_init_param( S, &P ); +} + +__attribute__((visibility ("hidden"))) +int blake2b_init_key( blake2b_state *S, const uint8_t outlen, const void *key, const uint8_t keylen ) +{ + if ( ( !outlen ) || ( outlen > BLAKE2B_OUTBYTES ) ) return -1; + + if ( ( !keylen ) || keylen > BLAKE2B_KEYBYTES ) return -1; + + const blake2b_param P = + { + outlen, + keylen, + 1, + 1, + 0, + 0, + 0, + 0, + {0}, + {0}, + {0} + }; + + if( blake2b_init_param( S, &P ) < 0 ) + return 0; + + { + uint8_t block[BLAKE2B_BLOCKBYTES]; + memset( block, 0, BLAKE2B_BLOCKBYTES ); + memcpy( block, key, keylen ); + blake2b_update( S, block, BLAKE2B_BLOCKBYTES ); + secure_zero_memory( block, BLAKE2B_BLOCKBYTES ); /* Burn the key from stack */ + } + return 0; +} + +static inline int blake2b_compress( blake2b_state *S, const uint8_t block[BLAKE2B_BLOCKBYTES] ) +{ + __m128i row1l, row1h; + __m128i row2l, row2h; + __m128i row3l, row3h; + __m128i row4l, row4h; + __m128i b0, b1; + __m128i t0, t1; +#if defined(HAVE_SSSE3) && !defined(HAVE_XOP) + const __m128i r16 = _mm_setr_epi8( 2, 3, 4, 5, 6, 7, 0, 1, 10, 11, 12, 13, 14, 15, 8, 9 ); + const __m128i r24 = _mm_setr_epi8( 3, 4, 5, 6, 7, 0, 1, 2, 11, 12, 13, 14, 15, 8, 9, 10 ); +#endif +#if defined(HAVE_SSE41) + const __m128i m0 = LOADU( block + 00 ); + const __m128i m1 = LOADU( block + 16 ); + const __m128i m2 = LOADU( block + 32 ); + const __m128i m3 = LOADU( block + 48 ); + const __m128i m4 = LOADU( block + 64 ); + const __m128i m5 = LOADU( block + 80 ); + const __m128i m6 = LOADU( block + 96 ); + const __m128i m7 = LOADU( block + 112 ); +#else + const uint64_t m0 = ( ( uint64_t * )block )[ 0]; + const uint64_t m1 = ( ( uint64_t * )block )[ 1]; + const uint64_t m2 = ( ( uint64_t * )block )[ 2]; + const uint64_t m3 = ( ( uint64_t * )block )[ 3]; + const uint64_t m4 = ( ( uint64_t * )block )[ 4]; + const uint64_t m5 = ( ( uint64_t * )block )[ 5]; + const uint64_t m6 = ( ( uint64_t * )block )[ 6]; + const uint64_t m7 = ( ( uint64_t * )block )[ 7]; + const uint64_t m8 = ( ( uint64_t * )block )[ 8]; + const uint64_t m9 = ( ( uint64_t * )block )[ 9]; + const uint64_t m10 = ( ( uint64_t * )block )[10]; + const uint64_t m11 = ( ( uint64_t * )block )[11]; + const uint64_t m12 = ( ( uint64_t * )block )[12]; + const uint64_t m13 = ( ( uint64_t * )block )[13]; + const uint64_t m14 = ( ( uint64_t * )block )[14]; + const uint64_t m15 = ( ( uint64_t * )block )[15]; +#endif + row1l = LOADU( &S->h[0] ); + row1h = LOADU( &S->h[2] ); + row2l = LOADU( &S->h[4] ); + row2h = LOADU( &S->h[6] ); + row3l = LOADU( &blake2b_IV[0] ); + row3h = LOADU( &blake2b_IV[2] ); + row4l = _mm_xor_si128( LOADU( &blake2b_IV[4] ), LOADU( &S->t[0] ) ); + row4h = _mm_xor_si128( LOADU( &blake2b_IV[6] ), LOADU( &S->f[0] ) ); + ROUND( 0 ); + ROUND( 1 ); + ROUND( 2 ); + ROUND( 3 ); + ROUND( 4 ); + ROUND( 5 ); + ROUND( 6 ); + ROUND( 7 ); + ROUND( 8 ); + ROUND( 9 ); + ROUND( 10 ); + ROUND( 11 ); + row1l = _mm_xor_si128( row3l, row1l ); + row1h = _mm_xor_si128( row3h, row1h ); + STOREU( &S->h[0], _mm_xor_si128( LOADU( &S->h[0] ), row1l ) ); + STOREU( &S->h[2], _mm_xor_si128( LOADU( &S->h[2] ), row1h ) ); + row2l = _mm_xor_si128( row4l, row2l ); + row2h = _mm_xor_si128( row4h, row2h ); + STOREU( &S->h[4], _mm_xor_si128( LOADU( &S->h[4] ), row2l ) ); + STOREU( &S->h[6], _mm_xor_si128( LOADU( &S->h[6] ), row2h ) ); + return 0; +} + +__attribute__((visibility ("hidden"))) +int blake2b_update( blake2b_state *S, const uint8_t *in, uint64_t inlen ) +{ + while( inlen > 0 ) + { + size_t left = S->buflen; + size_t fill = 2 * BLAKE2B_BLOCKBYTES - left; + + if( inlen > fill ) + { + memcpy( S->buf + left, in, fill ); // Fill buffer + S->buflen += fill; + blake2b_increment_counter( S, BLAKE2B_BLOCKBYTES ); + blake2b_compress( S, S->buf ); // Compress + memcpy( S->buf, S->buf + BLAKE2B_BLOCKBYTES, BLAKE2B_BLOCKBYTES ); // Shift buffer left + S->buflen -= BLAKE2B_BLOCKBYTES; + in += fill; + inlen -= fill; + } + else // inlen <= fill + { + memcpy( S->buf + left, in, inlen ); + S->buflen += inlen; // Be lazy, do not compress + in += inlen; + inlen -= inlen; + } + } + + return 0; +} + +__attribute__((visibility ("hidden"))) +int blake2b_final( blake2b_state *S, uint8_t *out, uint8_t outlen ) +{ + if( outlen > BLAKE2B_OUTBYTES ) + return -1; + + if( S->buflen > BLAKE2B_BLOCKBYTES ) + { + blake2b_increment_counter( S, BLAKE2B_BLOCKBYTES ); + blake2b_compress( S, S->buf ); + S->buflen -= BLAKE2B_BLOCKBYTES; + memcpy( S->buf, S->buf + BLAKE2B_BLOCKBYTES, S->buflen ); + } + + blake2b_increment_counter( S, S->buflen ); + blake2b_set_lastblock( S ); + memset( S->buf + S->buflen, 0, 2 * BLAKE2B_BLOCKBYTES - S->buflen ); /* Padding */ + blake2b_compress( S, S->buf ); + memcpy( out, &S->h[0], outlen ); + return 0; +} + +__attribute__((visibility ("hidden"))) +int blake2b( uint8_t *out, const void *in, const void *key, const uint8_t outlen, const uint64_t inlen, uint8_t keylen ) +{ + blake2b_state S[1]; + + /* Verify parameters */ + if ( NULL == in && inlen > 0 ) return -1; + + if ( NULL == out ) return -1; + + if( NULL == key && keylen > 0 ) return -1; + + if( !outlen || outlen > BLAKE2B_OUTBYTES ) return -1; + + if( keylen > BLAKE2B_KEYBYTES ) return -1; + + if( keylen ) + { + if( blake2b_init_key( S, outlen, key, keylen ) < 0 ) return -1; + } + else + { + if( blake2b_init( S, outlen ) < 0 ) return -1; + } + + blake2b_update( S, ( const uint8_t * )in, inlen ); + blake2b_final( S, out, outlen ); + return 0; +} diff --git a/cbits/blake2/blake2bp.c b/cbits/blake2/blake2bp.c new file mode 100644 index 0000000..a6a4f4d --- /dev/null +++ b/cbits/blake2/blake2bp.c @@ -0,0 +1,263 @@ +/* + BLAKE2 reference source code package - optimized C implementations + + Written in 2012 by Samuel Neves + + To the extent possible under law, the author(s) have dedicated all copyright + and related and neighboring rights to this software to the public domain + worldwide. This software is distributed without any warranty. + + You should have received a copy of the CC0 Public Domain Dedication along with + this software. If not, see . +*/ + +#include +#include +#include +#include + +#if defined(_OPENMP) +#include +#endif + +#include "blake2.h" +#include "blake2-impl.h" + +#define PARALLELISM_DEGREE 4 + +static inline int blake2bp_init_leaf( blake2b_state *S, uint8_t outlen, uint8_t keylen, uint64_t offset ) +{ + blake2b_param P[1]; + P->digest_length = outlen; + P->key_length = keylen; + P->fanout = PARALLELISM_DEGREE; + P->depth = 2; + P->leaf_length = 0; + P->node_offset = offset; + P->node_depth = 0; + P->inner_length = BLAKE2B_OUTBYTES; + memset( P->reserved, 0, sizeof( P->reserved ) ); + memset( P->salt, 0, sizeof( P->salt ) ); + memset( P->personal, 0, sizeof( P->personal ) ); + return blake2b_init_param( S, P ); +} + +static inline int blake2bp_init_root( blake2b_state *S, uint8_t outlen, uint8_t keylen ) +{ + blake2b_param P[1]; + P->digest_length = outlen; + P->key_length = keylen; + P->fanout = PARALLELISM_DEGREE; + P->depth = 2; + P->leaf_length = 0; + P->node_offset = 0; + P->node_depth = 1; + P->inner_length = BLAKE2B_OUTBYTES; + memset( P->reserved, 0, sizeof( P->reserved ) ); + memset( P->salt, 0, sizeof( P->salt ) ); + memset( P->personal, 0, sizeof( P->personal ) ); + return blake2b_init_param( S, P ); +} + +__attribute__((visibility ("hidden"))) +int blake2bp_init( blake2bp_state *S, const uint8_t outlen ) +{ + if( !outlen || outlen > BLAKE2B_OUTBYTES ) return -1; + + memset( S->buf, 0, sizeof( S->buf ) ); + S->buflen = 0; + + if( blake2bp_init_root( S->R, outlen, 0 ) < 0 ) + return -1; + + for( size_t i = 0; i < PARALLELISM_DEGREE; ++i ) + if( blake2bp_init_leaf( S->S[i], outlen, 0, i ) < 0 ) return -1; + + S->R->last_node = 1; + S->S[PARALLELISM_DEGREE - 1]->last_node = 1; + return 0; +} + +__attribute__((visibility ("hidden"))) +int blake2bp_init_key( blake2bp_state *S, const uint8_t outlen, const void *key, const uint8_t keylen ) +{ + if( !outlen || outlen > BLAKE2B_OUTBYTES ) return -1; + + if( !key || !keylen || keylen > BLAKE2B_KEYBYTES ) return -1; + + memset( S->buf, 0, sizeof( S->buf ) ); + S->buflen = 0; + + if( blake2bp_init_root( S->R, outlen, keylen ) < 0 ) + return -1; + + for( size_t i = 0; i < PARALLELISM_DEGREE; ++i ) + if( blake2bp_init_leaf( S->S[i], outlen, keylen, i ) < 0 ) return -1; + + S->R->last_node = 1; + S->S[PARALLELISM_DEGREE - 1]->last_node = 1; + { + uint8_t block[BLAKE2B_BLOCKBYTES]; + memset( block, 0, BLAKE2B_BLOCKBYTES ); + memcpy( block, key, keylen ); + + for( size_t i = 0; i < PARALLELISM_DEGREE; ++i ) + blake2b_update( S->S[i], block, BLAKE2B_BLOCKBYTES ); + + secure_zero_memory( block, BLAKE2B_BLOCKBYTES ); /* Burn the key from stack */ + } + return 0; +} + +__attribute__((visibility ("hidden"))) +int blake2bp_update( blake2bp_state *S, const uint8_t *in, uint64_t inlen ) +{ + size_t left = S->buflen; + size_t fill = sizeof( S->buf ) - left; + + if( left && inlen >= fill ) + { + memcpy( S->buf + left, in, fill ); + + for( size_t i = 0; i < PARALLELISM_DEGREE; ++i ) + blake2b_update( S->S[i], S->buf + i * BLAKE2B_BLOCKBYTES, BLAKE2B_BLOCKBYTES ); + + in += fill; + inlen -= fill; + left = 0; + } + +#if defined(_OPENMP) + #pragma omp parallel shared(S), num_threads(PARALLELISM_DEGREE) +#else + + for( size_t id__ = 0; id__ < PARALLELISM_DEGREE; ++id__ ) +#endif + { +#if defined(_OPENMP) + size_t id__ = omp_get_thread_num(); +#endif + uint64_t inlen__ = inlen; + const uint8_t *in__ = ( const uint8_t * )in; + in__ += id__ * BLAKE2B_BLOCKBYTES; + + while( inlen__ >= PARALLELISM_DEGREE * BLAKE2B_BLOCKBYTES ) + { + blake2b_update( S->S[id__], in__, BLAKE2B_BLOCKBYTES ); + in__ += PARALLELISM_DEGREE * BLAKE2B_BLOCKBYTES; + inlen__ -= PARALLELISM_DEGREE * BLAKE2B_BLOCKBYTES; + } + } + + in += inlen - inlen % ( PARALLELISM_DEGREE * BLAKE2B_BLOCKBYTES ); + inlen %= PARALLELISM_DEGREE * BLAKE2B_BLOCKBYTES; + + if( inlen > 0 ) + memcpy( S->buf + left, in, inlen ); + + S->buflen = left + inlen; + return 0; +} + + +__attribute__((visibility ("hidden"))) +int blake2bp_final( blake2bp_state *S, uint8_t *out, const uint8_t outlen ) +{ + uint8_t hash[PARALLELISM_DEGREE][BLAKE2B_OUTBYTES]; + + for( size_t i = 0; i < PARALLELISM_DEGREE; ++i ) + { + if( S->buflen > i * BLAKE2B_BLOCKBYTES ) + { + size_t left = S->buflen - i * BLAKE2B_BLOCKBYTES; + + if( left > BLAKE2B_BLOCKBYTES ) left = BLAKE2B_BLOCKBYTES; + + blake2b_update( S->S[i], S->buf + i * BLAKE2B_BLOCKBYTES, left ); + } + + blake2b_final( S->S[i], hash[i], BLAKE2B_OUTBYTES ); + } + + for( size_t i = 0; i < PARALLELISM_DEGREE; ++i ) + blake2b_update( S->R, hash[i], BLAKE2B_OUTBYTES ); + + return blake2b_final( S->R, out, outlen ); +} + +__attribute__((visibility ("hidden"))) +int blake2bp( uint8_t *out, const void *in, const void *key, uint8_t outlen, uint64_t inlen, uint8_t keylen ) +{ + uint8_t hash[PARALLELISM_DEGREE][BLAKE2B_OUTBYTES]; + blake2b_state S[PARALLELISM_DEGREE][1]; + blake2b_state FS[1]; + + /* Verify parameters */ + if ( NULL == in && inlen > 0 ) return -1; + + if ( NULL == out ) return -1; + + if( NULL == key && keylen > 0 ) return -1; + + if( !outlen || outlen > BLAKE2B_OUTBYTES ) return -1; + + if( keylen > BLAKE2B_KEYBYTES ) return -1; + + for( size_t i = 0; i < PARALLELISM_DEGREE; ++i ) + if( blake2bp_init_leaf( S[i], outlen, keylen, i ) < 0 ) return -1; + + S[PARALLELISM_DEGREE - 1]->last_node = 1; // mark last node + + if( keylen > 0 ) + { + uint8_t block[BLAKE2B_BLOCKBYTES]; + memset( block, 0, BLAKE2B_BLOCKBYTES ); + memcpy( block, key, keylen ); + + for( size_t i = 0; i < PARALLELISM_DEGREE; ++i ) + blake2b_update( S[i], block, BLAKE2B_BLOCKBYTES ); + + secure_zero_memory( block, BLAKE2B_BLOCKBYTES ); /* Burn the key from stack */ + } + +#if defined(_OPENMP) + #pragma omp parallel shared(S,hash), num_threads(PARALLELISM_DEGREE) +#else + + for( size_t id__ = 0; id__ < PARALLELISM_DEGREE; ++id__ ) +#endif + { +#if defined(_OPENMP) + size_t id__ = omp_get_thread_num(); +#endif + uint64_t inlen__ = inlen; + const uint8_t *in__ = ( const uint8_t * )in; + in__ += id__ * BLAKE2B_BLOCKBYTES; + + while( inlen__ >= PARALLELISM_DEGREE * BLAKE2B_BLOCKBYTES ) + { + blake2b_update( S[id__], in__, BLAKE2B_BLOCKBYTES ); + in__ += PARALLELISM_DEGREE * BLAKE2B_BLOCKBYTES; + inlen__ -= PARALLELISM_DEGREE * BLAKE2B_BLOCKBYTES; + } + + if( inlen__ > id__ * BLAKE2B_BLOCKBYTES ) + { + const size_t left = inlen__ - id__ * BLAKE2B_BLOCKBYTES; + const size_t len = left <= BLAKE2B_BLOCKBYTES ? left : BLAKE2B_BLOCKBYTES; + blake2b_update( S[id__], in__, len ); + } + + blake2b_final( S[id__], hash[id__], BLAKE2B_OUTBYTES ); + } + + if( blake2bp_init_root( FS, outlen, keylen ) < 0 ) + return -1; + + FS->last_node = 1; // Mark as last node + + for( size_t i = 0; i < PARALLELISM_DEGREE; ++i ) + blake2b_update( FS, hash[i], BLAKE2B_OUTBYTES ); + + return blake2b_final( FS, out, outlen ); +} diff --git a/cbits/blake2/blake2s-load-sse2.h b/cbits/blake2/blake2s-load-sse2.h new file mode 100644 index 0000000..b24483c --- /dev/null +++ b/cbits/blake2/blake2s-load-sse2.h @@ -0,0 +1,59 @@ +/* + BLAKE2 reference source code package - optimized C implementations + + Written in 2012 by Samuel Neves + + To the extent possible under law, the author(s) have dedicated all copyright + and related and neighboring rights to this software to the public domain + worldwide. This software is distributed without any warranty. + + You should have received a copy of the CC0 Public Domain Dedication along with + this software. If not, see . +*/ +#pragma once +#ifndef __BLAKE2S_LOAD_SSE2_H__ +#define __BLAKE2S_LOAD_SSE2_H__ + +#define LOAD_MSG_0_1(buf) buf = _mm_set_epi32(m6,m4,m2,m0) +#define LOAD_MSG_0_2(buf) buf = _mm_set_epi32(m7,m5,m3,m1) +#define LOAD_MSG_0_3(buf) buf = _mm_set_epi32(m14,m12,m10,m8) +#define LOAD_MSG_0_4(buf) buf = _mm_set_epi32(m15,m13,m11,m9) +#define LOAD_MSG_1_1(buf) buf = _mm_set_epi32(m13,m9,m4,m14) +#define LOAD_MSG_1_2(buf) buf = _mm_set_epi32(m6,m15,m8,m10) +#define LOAD_MSG_1_3(buf) buf = _mm_set_epi32(m5,m11,m0,m1) +#define LOAD_MSG_1_4(buf) buf = _mm_set_epi32(m3,m7,m2,m12) +#define LOAD_MSG_2_1(buf) buf = _mm_set_epi32(m15,m5,m12,m11) +#define LOAD_MSG_2_2(buf) buf = _mm_set_epi32(m13,m2,m0,m8) +#define LOAD_MSG_2_3(buf) buf = _mm_set_epi32(m9,m7,m3,m10) +#define LOAD_MSG_2_4(buf) buf = _mm_set_epi32(m4,m1,m6,m14) +#define LOAD_MSG_3_1(buf) buf = _mm_set_epi32(m11,m13,m3,m7) +#define LOAD_MSG_3_2(buf) buf = _mm_set_epi32(m14,m12,m1,m9) +#define LOAD_MSG_3_3(buf) buf = _mm_set_epi32(m15,m4,m5,m2) +#define LOAD_MSG_3_4(buf) buf = _mm_set_epi32(m8,m0,m10,m6) +#define LOAD_MSG_4_1(buf) buf = _mm_set_epi32(m10,m2,m5,m9) +#define LOAD_MSG_4_2(buf) buf = _mm_set_epi32(m15,m4,m7,m0) +#define LOAD_MSG_4_3(buf) buf = _mm_set_epi32(m3,m6,m11,m14) +#define LOAD_MSG_4_4(buf) buf = _mm_set_epi32(m13,m8,m12,m1) +#define LOAD_MSG_5_1(buf) buf = _mm_set_epi32(m8,m0,m6,m2) +#define LOAD_MSG_5_2(buf) buf = _mm_set_epi32(m3,m11,m10,m12) +#define LOAD_MSG_5_3(buf) buf = _mm_set_epi32(m1,m15,m7,m4) +#define LOAD_MSG_5_4(buf) buf = _mm_set_epi32(m9,m14,m5,m13) +#define LOAD_MSG_6_1(buf) buf = _mm_set_epi32(m4,m14,m1,m12) +#define LOAD_MSG_6_2(buf) buf = _mm_set_epi32(m10,m13,m15,m5) +#define LOAD_MSG_6_3(buf) buf = _mm_set_epi32(m8,m9,m6,m0) +#define LOAD_MSG_6_4(buf) buf = _mm_set_epi32(m11,m2,m3,m7) +#define LOAD_MSG_7_1(buf) buf = _mm_set_epi32(m3,m12,m7,m13) +#define LOAD_MSG_7_2(buf) buf = _mm_set_epi32(m9,m1,m14,m11) +#define LOAD_MSG_7_3(buf) buf = _mm_set_epi32(m2,m8,m15,m5) +#define LOAD_MSG_7_4(buf) buf = _mm_set_epi32(m10,m6,m4,m0) +#define LOAD_MSG_8_1(buf) buf = _mm_set_epi32(m0,m11,m14,m6) +#define LOAD_MSG_8_2(buf) buf = _mm_set_epi32(m8,m3,m9,m15) +#define LOAD_MSG_8_3(buf) buf = _mm_set_epi32(m10,m1,m13,m12) +#define LOAD_MSG_8_4(buf) buf = _mm_set_epi32(m5,m4,m7,m2) +#define LOAD_MSG_9_1(buf) buf = _mm_set_epi32(m1,m7,m8,m10) +#define LOAD_MSG_9_2(buf) buf = _mm_set_epi32(m5,m6,m4,m2) +#define LOAD_MSG_9_3(buf) buf = _mm_set_epi32(m13,m3,m9,m15) +#define LOAD_MSG_9_4(buf) buf = _mm_set_epi32(m0,m12,m14,m11) + + +#endif diff --git a/cbits/blake2/blake2s-load-sse41.h b/cbits/blake2/blake2s-load-sse41.h new file mode 100644 index 0000000..3ac12eb --- /dev/null +++ b/cbits/blake2/blake2s-load-sse41.h @@ -0,0 +1,229 @@ +/* + BLAKE2 reference source code package - optimized C implementations + + Written in 2012 by Samuel Neves + + To the extent possible under law, the author(s) have dedicated all copyright + and related and neighboring rights to this software to the public domain + worldwide. This software is distributed without any warranty. + + You should have received a copy of the CC0 Public Domain Dedication along with + this software. If not, see . +*/ +#pragma once +#ifndef __BLAKE2S_LOAD_SSE41_H__ +#define __BLAKE2S_LOAD_SSE41_H__ + +#define LOAD_MSG_0_1(buf) \ +buf = TOI(_mm_shuffle_ps(TOF(m0), TOF(m1), _MM_SHUFFLE(2,0,2,0))); + +#define LOAD_MSG_0_2(buf) \ +buf = TOI(_mm_shuffle_ps(TOF(m0), TOF(m1), _MM_SHUFFLE(3,1,3,1))); + +#define LOAD_MSG_0_3(buf) \ +buf = TOI(_mm_shuffle_ps(TOF(m2), TOF(m3), _MM_SHUFFLE(2,0,2,0))); + +#define LOAD_MSG_0_4(buf) \ +buf = TOI(_mm_shuffle_ps(TOF(m2), TOF(m3), _MM_SHUFFLE(3,1,3,1))); + +#define LOAD_MSG_1_1(buf) \ +t0 = _mm_blend_epi16(m1, m2, 0x0C); \ +t1 = _mm_slli_si128(m3, 4); \ +t2 = _mm_blend_epi16(t0, t1, 0xF0); \ +buf = _mm_shuffle_epi32(t2, _MM_SHUFFLE(2,1,0,3)); + +#define LOAD_MSG_1_2(buf) \ +t0 = _mm_shuffle_epi32(m2,_MM_SHUFFLE(0,0,2,0)); \ +t1 = _mm_blend_epi16(m1,m3,0xC0); \ +t2 = _mm_blend_epi16(t0, t1, 0xF0); \ +buf = _mm_shuffle_epi32(t2, _MM_SHUFFLE(2,3,0,1)); + +#define LOAD_MSG_1_3(buf) \ +t0 = _mm_slli_si128(m1, 4); \ +t1 = _mm_blend_epi16(m2, t0, 0x30); \ +t2 = _mm_blend_epi16(m0, t1, 0xF0); \ +buf = _mm_shuffle_epi32(t2, _MM_SHUFFLE(2,3,0,1)); + +#define LOAD_MSG_1_4(buf) \ +t0 = _mm_unpackhi_epi32(m0,m1); \ +t1 = _mm_slli_si128(m3, 4); \ +t2 = _mm_blend_epi16(t0, t1, 0x0C); \ +buf = _mm_shuffle_epi32(t2, _MM_SHUFFLE(2,3,0,1)); + +#define LOAD_MSG_2_1(buf) \ +t0 = _mm_unpackhi_epi32(m2,m3); \ +t1 = _mm_blend_epi16(m3,m1,0x0C); \ +t2 = _mm_blend_epi16(t0, t1, 0x0F); \ +buf = _mm_shuffle_epi32(t2, _MM_SHUFFLE(3,1,0,2)); + +#define LOAD_MSG_2_2(buf) \ +t0 = _mm_unpacklo_epi32(m2,m0); \ +t1 = _mm_blend_epi16(t0, m0, 0xF0); \ +t2 = _mm_slli_si128(m3, 8); \ +buf = _mm_blend_epi16(t1, t2, 0xC0); + +#define LOAD_MSG_2_3(buf) \ +t0 = _mm_blend_epi16(m0, m2, 0x3C); \ +t1 = _mm_srli_si128(m1, 12); \ +t2 = _mm_blend_epi16(t0,t1,0x03); \ +buf = _mm_shuffle_epi32(t2, _MM_SHUFFLE(1,0,3,2)); + +#define LOAD_MSG_2_4(buf) \ +t0 = _mm_slli_si128(m3, 4); \ +t1 = _mm_blend_epi16(m0, m1, 0x33); \ +t2 = _mm_blend_epi16(t1, t0, 0xC0); \ +buf = _mm_shuffle_epi32(t2, _MM_SHUFFLE(0,1,2,3)); + +#define LOAD_MSG_3_1(buf) \ +t0 = _mm_unpackhi_epi32(m0,m1); \ +t1 = _mm_unpackhi_epi32(t0, m2); \ +t2 = _mm_blend_epi16(t1, m3, 0x0C); \ +buf = _mm_shuffle_epi32(t2, _MM_SHUFFLE(3,1,0,2)); + +#define LOAD_MSG_3_2(buf) \ +t0 = _mm_slli_si128(m2, 8); \ +t1 = _mm_blend_epi16(m3,m0,0x0C); \ +t2 = _mm_blend_epi16(t1, t0, 0xC0); \ +buf = _mm_shuffle_epi32(t2, _MM_SHUFFLE(2,0,1,3)); + +#define LOAD_MSG_3_3(buf) \ +t0 = _mm_blend_epi16(m0,m1,0x0F); \ +t1 = _mm_blend_epi16(t0, m3, 0xC0); \ +buf = _mm_shuffle_epi32(t1, _MM_SHUFFLE(3,0,1,2)); + +#define LOAD_MSG_3_4(buf) \ +t0 = _mm_unpacklo_epi32(m0,m2); \ +t1 = _mm_unpackhi_epi32(m1,m2); \ +buf = _mm_unpacklo_epi64(t1,t0); + +#define LOAD_MSG_4_1(buf) \ +t0 = _mm_unpacklo_epi64(m1,m2); \ +t1 = _mm_unpackhi_epi64(m0,m2); \ +t2 = _mm_blend_epi16(t0,t1,0x33); \ +buf = _mm_shuffle_epi32(t2, _MM_SHUFFLE(2,0,1,3)); + +#define LOAD_MSG_4_2(buf) \ +t0 = _mm_unpackhi_epi64(m1,m3); \ +t1 = _mm_unpacklo_epi64(m0,m1); \ +buf = _mm_blend_epi16(t0,t1,0x33); + +#define LOAD_MSG_4_3(buf) \ +t0 = _mm_unpackhi_epi64(m3,m1); \ +t1 = _mm_unpackhi_epi64(m2,m0); \ +buf = _mm_blend_epi16(t1,t0,0x33); + +#define LOAD_MSG_4_4(buf) \ +t0 = _mm_blend_epi16(m0,m2,0x03); \ +t1 = _mm_slli_si128(t0, 8); \ +t2 = _mm_blend_epi16(t1,m3,0x0F); \ +buf = _mm_shuffle_epi32(t2, _MM_SHUFFLE(1,2,0,3)); + +#define LOAD_MSG_5_1(buf) \ +t0 = _mm_unpackhi_epi32(m0,m1); \ +t1 = _mm_unpacklo_epi32(m0,m2); \ +buf = _mm_unpacklo_epi64(t0,t1); + +#define LOAD_MSG_5_2(buf) \ +t0 = _mm_srli_si128(m2, 4); \ +t1 = _mm_blend_epi16(m0,m3,0x03); \ +buf = _mm_blend_epi16(t1,t0,0x3C); + +#define LOAD_MSG_5_3(buf) \ +t0 = _mm_blend_epi16(m1,m0,0x0C); \ +t1 = _mm_srli_si128(m3, 4); \ +t2 = _mm_blend_epi16(t0,t1,0x30); \ +buf = _mm_shuffle_epi32(t2, _MM_SHUFFLE(1,2,3,0)); + +#define LOAD_MSG_5_4(buf) \ +t0 = _mm_unpacklo_epi64(m1,m2); \ +t1= _mm_shuffle_epi32(m3, _MM_SHUFFLE(0,2,0,1)); \ +buf = _mm_blend_epi16(t0,t1,0x33); + +#define LOAD_MSG_6_1(buf) \ +t0 = _mm_slli_si128(m1, 12); \ +t1 = _mm_blend_epi16(m0,m3,0x33); \ +buf = _mm_blend_epi16(t1,t0,0xC0); + +#define LOAD_MSG_6_2(buf) \ +t0 = _mm_blend_epi16(m3,m2,0x30); \ +t1 = _mm_srli_si128(m1, 4); \ +t2 = _mm_blend_epi16(t0,t1,0x03); \ +buf = _mm_shuffle_epi32(t2, _MM_SHUFFLE(2,1,3,0)); + +#define LOAD_MSG_6_3(buf) \ +t0 = _mm_unpacklo_epi64(m0,m2); \ +t1 = _mm_srli_si128(m1, 4); \ +buf = _mm_shuffle_epi32(_mm_blend_epi16(t0,t1,0x0C), _MM_SHUFFLE(2,3,1,0)); + +#define LOAD_MSG_6_4(buf) \ +t0 = _mm_unpackhi_epi32(m1,m2); \ +t1 = _mm_unpackhi_epi64(m0,t0); \ +buf = _mm_shuffle_epi32(t1, _MM_SHUFFLE(3,0,1,2)); + +#define LOAD_MSG_7_1(buf) \ +t0 = _mm_unpackhi_epi32(m0,m1); \ +t1 = _mm_blend_epi16(t0,m3,0x0F); \ +buf = _mm_shuffle_epi32(t1,_MM_SHUFFLE(2,0,3,1)); + +#define LOAD_MSG_7_2(buf) \ +t0 = _mm_blend_epi16(m2,m3,0x30); \ +t1 = _mm_srli_si128(m0,4); \ +t2 = _mm_blend_epi16(t0,t1,0x03); \ +buf = _mm_shuffle_epi32(t2, _MM_SHUFFLE(1,0,2,3)); + +#define LOAD_MSG_7_3(buf) \ +t0 = _mm_unpackhi_epi64(m0,m3); \ +t1 = _mm_unpacklo_epi64(m1,m2); \ +t2 = _mm_blend_epi16(t0,t1,0x3C); \ +buf = _mm_shuffle_epi32(t2,_MM_SHUFFLE(0,2,3,1)); + +#define LOAD_MSG_7_4(buf) \ +t0 = _mm_unpacklo_epi32(m0,m1); \ +t1 = _mm_unpackhi_epi32(m1,m2); \ +buf = _mm_unpacklo_epi64(t0,t1); + +#define LOAD_MSG_8_1(buf) \ +t0 = _mm_unpackhi_epi32(m1,m3); \ +t1 = _mm_unpacklo_epi64(t0,m0); \ +t2 = _mm_blend_epi16(t1,m2,0xC0); \ +buf = _mm_shufflehi_epi16(t2,_MM_SHUFFLE(1,0,3,2)); + +#define LOAD_MSG_8_2(buf) \ +t0 = _mm_unpackhi_epi32(m0,m3); \ +t1 = _mm_blend_epi16(m2,t0,0xF0); \ +buf = _mm_shuffle_epi32(t1,_MM_SHUFFLE(0,2,1,3)); + +#define LOAD_MSG_8_3(buf) \ +t0 = _mm_blend_epi16(m2,m0,0x0C); \ +t1 = _mm_slli_si128(t0,4); \ +buf = _mm_blend_epi16(t1,m3,0x0F); + +#define LOAD_MSG_8_4(buf) \ +t0 = _mm_blend_epi16(m1,m0,0x30); \ +buf = _mm_shuffle_epi32(t0,_MM_SHUFFLE(1,0,3,2)); + +#define LOAD_MSG_9_1(buf) \ +t0 = _mm_blend_epi16(m0,m2,0x03); \ +t1 = _mm_blend_epi16(m1,m2,0x30); \ +t2 = _mm_blend_epi16(t1,t0,0x0F); \ +buf = _mm_shuffle_epi32(t2,_MM_SHUFFLE(1,3,0,2)); + +#define LOAD_MSG_9_2(buf) \ +t0 = _mm_slli_si128(m0,4); \ +t1 = _mm_blend_epi16(m1,t0,0xC0); \ +buf = _mm_shuffle_epi32(t1,_MM_SHUFFLE(1,2,0,3)); + +#define LOAD_MSG_9_3(buf) \ +t0 = _mm_unpackhi_epi32(m0,m3); \ +t1 = _mm_unpacklo_epi32(m2,m3); \ +t2 = _mm_unpackhi_epi64(t0,t1); \ +buf = _mm_shuffle_epi32(t2,_MM_SHUFFLE(3,0,2,1)); + +#define LOAD_MSG_9_4(buf) \ +t0 = _mm_blend_epi16(m3,m2,0xC0); \ +t1 = _mm_unpacklo_epi32(m0,m3); \ +t2 = _mm_blend_epi16(t0,t1,0x0F); \ +buf = _mm_shuffle_epi32(t2,_MM_SHUFFLE(0,1,2,3)); + +#endif + diff --git a/cbits/blake2/blake2s-load-xop.h b/cbits/blake2/blake2s-load-xop.h new file mode 100644 index 0000000..ac591a7 --- /dev/null +++ b/cbits/blake2/blake2s-load-xop.h @@ -0,0 +1,189 @@ +/* + BLAKE2 reference source code package - optimized C implementations + + Written in 2012 by Samuel Neves + + To the extent possible under law, the author(s) have dedicated all copyright + and related and neighboring rights to this software to the public domain + worldwide. This software is distributed without any warranty. + + You should have received a copy of the CC0 Public Domain Dedication along with + this software. If not, see . +*/ +#pragma once +#ifndef __BLAKE2S_LOAD_XOP_H__ +#define __BLAKE2S_LOAD_XOP_H__ + +#define TOB(x) ((x)*4*0x01010101 + 0x03020100) // ..or not TOB + +/* Basic VPPERM emulation, for testing purposes */ +/*static __m128i _mm_perm_epi8(const __m128i src1, const __m128i src2, const __m128i sel) +{ + const __m128i sixteen = _mm_set1_epi8(16); + const __m128i t0 = _mm_shuffle_epi8(src1, sel); + const __m128i s1 = _mm_shuffle_epi8(src2, _mm_sub_epi8(sel, sixteen)); + const __m128i mask = _mm_or_si128(_mm_cmpeq_epi8(sel, sixteen), + _mm_cmpgt_epi8(sel, sixteen)); // (>=16) = 0xff : 00 + return _mm_blendv_epi8(t0, s1, mask); +}*/ + +#define LOAD_MSG_0_1(buf) \ +buf = _mm_perm_epi8(m0, m1, _mm_set_epi32(TOB(6),TOB(4),TOB(2),TOB(0)) ); + +#define LOAD_MSG_0_2(buf) \ +buf = _mm_perm_epi8(m0, m1, _mm_set_epi32(TOB(7),TOB(5),TOB(3),TOB(1)) ); + +#define LOAD_MSG_0_3(buf) \ +buf = _mm_perm_epi8(m2, m3, _mm_set_epi32(TOB(6),TOB(4),TOB(2),TOB(0)) ); + +#define LOAD_MSG_0_4(buf) \ +buf = _mm_perm_epi8(m2, m3, _mm_set_epi32(TOB(7),TOB(5),TOB(3),TOB(1)) ); + +#define LOAD_MSG_1_1(buf) \ +t0 = _mm_perm_epi8(m1, m2, _mm_set_epi32(TOB(0),TOB(5),TOB(0),TOB(0)) ); \ +buf = _mm_perm_epi8(t0, m3, _mm_set_epi32(TOB(5),TOB(2),TOB(1),TOB(6)) ); + +#define LOAD_MSG_1_2(buf) \ +t1 = _mm_perm_epi8(m1, m2, _mm_set_epi32(TOB(2),TOB(0),TOB(4),TOB(6)) ); \ +buf = _mm_perm_epi8(t1, m3, _mm_set_epi32(TOB(3),TOB(7),TOB(1),TOB(0)) ); + +#define LOAD_MSG_1_3(buf) \ +t0 = _mm_perm_epi8(m0, m1, _mm_set_epi32(TOB(5),TOB(0),TOB(0),TOB(1)) ); \ +buf = _mm_perm_epi8(t0, m2, _mm_set_epi32(TOB(3),TOB(7),TOB(1),TOB(0)) ); + +#define LOAD_MSG_1_4(buf) \ +t1 = _mm_perm_epi8(m0, m1, _mm_set_epi32(TOB(3),TOB(7),TOB(2),TOB(0)) ); \ +buf = _mm_perm_epi8(t1, m3, _mm_set_epi32(TOB(3),TOB(2),TOB(1),TOB(4)) ); + +#define LOAD_MSG_2_1(buf) \ +t0 = _mm_perm_epi8(m1, m2, _mm_set_epi32(TOB(0),TOB(1),TOB(0),TOB(7)) ); \ +buf = _mm_perm_epi8(t0, m3, _mm_set_epi32(TOB(7),TOB(2),TOB(4),TOB(0)) ); + +#define LOAD_MSG_2_2(buf) \ +t1 = _mm_perm_epi8(m0, m2, _mm_set_epi32(TOB(0),TOB(2),TOB(0),TOB(4)) ); \ +buf = _mm_perm_epi8(t1, m3, _mm_set_epi32(TOB(5),TOB(2),TOB(1),TOB(0)) ); + +#define LOAD_MSG_2_3(buf) \ +t0 = _mm_perm_epi8(m0, m1, _mm_set_epi32(TOB(0),TOB(7),TOB(3),TOB(0)) ); \ +buf = _mm_perm_epi8(t0, m2, _mm_set_epi32(TOB(5),TOB(2),TOB(1),TOB(6)) ); + +#define LOAD_MSG_2_4(buf) \ +t1 = _mm_perm_epi8(m0, m1, _mm_set_epi32(TOB(4),TOB(1),TOB(6),TOB(0)) ); \ +buf = _mm_perm_epi8(t1, m3, _mm_set_epi32(TOB(3),TOB(2),TOB(1),TOB(6)) ); + +#define LOAD_MSG_3_1(buf) \ +t0 = _mm_perm_epi8(m0, m1, _mm_set_epi32(TOB(0),TOB(0),TOB(3),TOB(7)) ); \ +t0 = _mm_perm_epi8(t0, m2, _mm_set_epi32(TOB(7),TOB(2),TOB(1),TOB(0)) ); \ +buf = _mm_perm_epi8(t0, m3, _mm_set_epi32(TOB(3),TOB(5),TOB(1),TOB(0)) ); + +#define LOAD_MSG_3_2(buf) \ +t1 = _mm_perm_epi8(m0, m2, _mm_set_epi32(TOB(0),TOB(0),TOB(1),TOB(5)) ); \ +buf = _mm_perm_epi8(t1, m3, _mm_set_epi32(TOB(6),TOB(4),TOB(1),TOB(0)) ); + +#define LOAD_MSG_3_3(buf) \ +t0 = _mm_perm_epi8(m0, m1, _mm_set_epi32(TOB(0),TOB(4),TOB(5),TOB(2)) ); \ +buf = _mm_perm_epi8(t0, m3, _mm_set_epi32(TOB(7),TOB(2),TOB(1),TOB(0)) ); + +#define LOAD_MSG_3_4(buf) \ +t1 = _mm_perm_epi8(m0, m1, _mm_set_epi32(TOB(0),TOB(0),TOB(0),TOB(6)) ); \ +buf = _mm_perm_epi8(t1, m2, _mm_set_epi32(TOB(4),TOB(2),TOB(6),TOB(0)) ); + +#define LOAD_MSG_4_1(buf) \ +t0 = _mm_perm_epi8(m0, m1, _mm_set_epi32(TOB(0),TOB(2),TOB(5),TOB(0)) ); \ +buf = _mm_perm_epi8(t0, m2, _mm_set_epi32(TOB(6),TOB(2),TOB(1),TOB(5)) ); + +#define LOAD_MSG_4_2(buf) \ +t1 = _mm_perm_epi8(m0, m1, _mm_set_epi32(TOB(0),TOB(4),TOB(7),TOB(0)) ); \ +buf = _mm_perm_epi8(t1, m3, _mm_set_epi32(TOB(7),TOB(2),TOB(1),TOB(0)) ); + +#define LOAD_MSG_4_3(buf) \ +t0 = _mm_perm_epi8(m0, m1, _mm_set_epi32(TOB(3),TOB(6),TOB(0),TOB(0)) ); \ +t0 = _mm_perm_epi8(t0, m2, _mm_set_epi32(TOB(3),TOB(2),TOB(7),TOB(0)) ); \ +buf = _mm_perm_epi8(t0, m3, _mm_set_epi32(TOB(3),TOB(2),TOB(1),TOB(6)) ); + +#define LOAD_MSG_4_4(buf) \ +t1 = _mm_perm_epi8(m0, m2, _mm_set_epi32(TOB(0),TOB(4),TOB(0),TOB(1)) ); \ +buf = _mm_perm_epi8(t1, m3, _mm_set_epi32(TOB(5),TOB(2),TOB(4),TOB(0)) ); + +#define LOAD_MSG_5_1(buf) \ +t0 = _mm_perm_epi8(m0, m1, _mm_set_epi32(TOB(0),TOB(0),TOB(6),TOB(2)) ); \ +buf = _mm_perm_epi8(t0, m2, _mm_set_epi32(TOB(4),TOB(2),TOB(1),TOB(0)) ); + +#define LOAD_MSG_5_2(buf) \ +t1 = _mm_perm_epi8(m0, m2, _mm_set_epi32(TOB(3),TOB(7),TOB(6),TOB(0)) ); \ +buf = _mm_perm_epi8(t1, m3, _mm_set_epi32(TOB(3),TOB(2),TOB(1),TOB(4)) ); + +#define LOAD_MSG_5_3(buf) \ +t0 = _mm_perm_epi8(m0, m1, _mm_set_epi32(TOB(1),TOB(0),TOB(7),TOB(4)) ); \ +buf = _mm_perm_epi8(t0, m3, _mm_set_epi32(TOB(3),TOB(7),TOB(1),TOB(0)) ); + +#define LOAD_MSG_5_4(buf) \ +t1 = _mm_perm_epi8(m1, m2, _mm_set_epi32(TOB(5),TOB(0),TOB(1),TOB(0)) ); \ +buf = _mm_perm_epi8(t1, m3, _mm_set_epi32(TOB(3),TOB(6),TOB(1),TOB(5)) ); + +#define LOAD_MSG_6_1(buf) \ +t0 = _mm_perm_epi8(m0, m1, _mm_set_epi32(TOB(4),TOB(0),TOB(1),TOB(0)) ); \ +buf = _mm_perm_epi8(t0, m3, _mm_set_epi32(TOB(3),TOB(6),TOB(1),TOB(4)) ); + +#define LOAD_MSG_6_2(buf) \ +t1 = _mm_perm_epi8(m1, m2, _mm_set_epi32(TOB(6),TOB(0),TOB(0),TOB(1)) ); \ +buf = _mm_perm_epi8(t1, m3, _mm_set_epi32(TOB(3),TOB(5),TOB(7),TOB(0)) ); + +#define LOAD_MSG_6_3(buf) \ +t0 = _mm_perm_epi8(m0, m1, _mm_set_epi32(TOB(0),TOB(0),TOB(6),TOB(0)) ); \ +buf = _mm_perm_epi8(t0, m2, _mm_set_epi32(TOB(4),TOB(5),TOB(1),TOB(0)) ); + +#define LOAD_MSG_6_4(buf) \ +t1 = _mm_perm_epi8(m0, m1, _mm_set_epi32(TOB(0),TOB(2),TOB(3),TOB(7)) ); \ +buf = _mm_perm_epi8(t1, m2, _mm_set_epi32(TOB(7),TOB(2),TOB(1),TOB(0)) ); + +#define LOAD_MSG_7_1(buf) \ +t0 = _mm_perm_epi8(m0, m1, _mm_set_epi32(TOB(3),TOB(0),TOB(7),TOB(0)) ); \ +buf = _mm_perm_epi8(t0, m3, _mm_set_epi32(TOB(3),TOB(4),TOB(1),TOB(5)) ); + +#define LOAD_MSG_7_2(buf) \ +t1 = _mm_perm_epi8(m0, m2, _mm_set_epi32(TOB(5),TOB(1),TOB(0),TOB(7)) ); \ +buf = _mm_perm_epi8(t1, m3, _mm_set_epi32(TOB(3),TOB(2),TOB(6),TOB(0)) ); + +#define LOAD_MSG_7_3(buf) \ +t0 = _mm_perm_epi8(m0, m1, _mm_set_epi32(TOB(2),TOB(0),TOB(0),TOB(5)) ); \ +t0 = _mm_perm_epi8(t0, m2, _mm_set_epi32(TOB(3),TOB(4),TOB(1),TOB(0)) ); \ +buf = _mm_perm_epi8(t0, m3, _mm_set_epi32(TOB(3),TOB(2),TOB(7),TOB(0)) ); + +#define LOAD_MSG_7_4(buf) \ +t1 = _mm_perm_epi8(m0, m1, _mm_set_epi32(TOB(0),TOB(6),TOB(4),TOB(0)) ); \ +buf = _mm_perm_epi8(t1, m2, _mm_set_epi32(TOB(6),TOB(2),TOB(1),TOB(0)) ); + +#define LOAD_MSG_8_1(buf) \ +t0 = _mm_perm_epi8(m0, m1, _mm_set_epi32(TOB(0),TOB(0),TOB(0),TOB(6)) ); \ +t0 = _mm_perm_epi8(t0, m2, _mm_set_epi32(TOB(3),TOB(7),TOB(1),TOB(0)) ); \ +buf = _mm_perm_epi8(t0, m3, _mm_set_epi32(TOB(3),TOB(2),TOB(6),TOB(0)) ); + +#define LOAD_MSG_8_2(buf) \ +t1 = _mm_perm_epi8(m0, m2, _mm_set_epi32(TOB(4),TOB(3),TOB(5),TOB(0)) ); \ +buf = _mm_perm_epi8(t1, m3, _mm_set_epi32(TOB(3),TOB(2),TOB(1),TOB(7)) ); + +#define LOAD_MSG_8_3(buf) \ +t0 = _mm_perm_epi8(m0, m2, _mm_set_epi32(TOB(6),TOB(1),TOB(0),TOB(0)) ); \ +buf = _mm_perm_epi8(t0, m3, _mm_set_epi32(TOB(3),TOB(2),TOB(5),TOB(4)) ); \ + +#define LOAD_MSG_8_4(buf) \ +buf = _mm_perm_epi8(m0, m1, _mm_set_epi32(TOB(5),TOB(4),TOB(7),TOB(2)) ); + +#define LOAD_MSG_9_1(buf) \ +t0 = _mm_perm_epi8(m0, m1, _mm_set_epi32(TOB(1),TOB(7),TOB(0),TOB(0)) ); \ +buf = _mm_perm_epi8(t0, m2, _mm_set_epi32(TOB(3),TOB(2),TOB(4),TOB(6)) ); + +#define LOAD_MSG_9_2(buf) \ +buf = _mm_perm_epi8(m0, m1, _mm_set_epi32(TOB(5),TOB(6),TOB(4),TOB(2)) ); + +#define LOAD_MSG_9_3(buf) \ +t0 = _mm_perm_epi8(m0, m2, _mm_set_epi32(TOB(0),TOB(3),TOB(5),TOB(0)) ); \ +buf = _mm_perm_epi8(t0, m3, _mm_set_epi32(TOB(5),TOB(2),TOB(1),TOB(7)) ); + +#define LOAD_MSG_9_4(buf) \ +t1 = _mm_perm_epi8(m0, m2, _mm_set_epi32(TOB(0),TOB(0),TOB(0),TOB(7)) ); \ +buf = _mm_perm_epi8(t1, m3, _mm_set_epi32(TOB(3),TOB(4),TOB(6),TOB(0)) ); + +#endif + diff --git a/cbits/blake2/blake2s-round.h b/cbits/blake2/blake2s-round.h new file mode 100644 index 0000000..9c58b52 --- /dev/null +++ b/cbits/blake2/blake2s-round.h @@ -0,0 +1,88 @@ +/* + BLAKE2 reference source code package - optimized C implementations + + Written in 2012 by Samuel Neves + + To the extent possible under law, the author(s) have dedicated all copyright + and related and neighboring rights to this software to the public domain + worldwide. This software is distributed without any warranty. + + You should have received a copy of the CC0 Public Domain Dedication along with + this software. If not, see . +*/ +#pragma once +#ifndef __BLAKE2S_ROUND_H__ +#define __BLAKE2S_ROUND_H__ + +#define LOADU(p) _mm_loadu_si128( (const __m128i *)(p) ) +#define STOREU(p,r) _mm_storeu_si128((__m128i *)(p), r) + +#define TOF(reg) _mm_castsi128_ps((reg)) +#define TOI(reg) _mm_castps_si128((reg)) + +#define LIKELY(x) __builtin_expect((x),1) + + +/* Microarchitecture-specific macros */ +#ifndef HAVE_XOP +#ifdef HAVE_SSSE3 +#define _mm_roti_epi32(r, c) ( \ + (8==-(c)) ? _mm_shuffle_epi8(r,r8) \ + : (16==-(c)) ? _mm_shuffle_epi8(r,r16) \ + : _mm_xor_si128(_mm_srli_epi32( (r), -(c) ),_mm_slli_epi32( (r), 32-(-(c)) )) ) +#else +#define _mm_roti_epi32(r, c) _mm_xor_si128(_mm_srli_epi32( (r), -(c) ),_mm_slli_epi32( (r), 32-(-(c)) )) +#endif +#else +/* ... */ +#endif + + +#define G1(row1,row2,row3,row4,buf) \ + row1 = _mm_add_epi32( _mm_add_epi32( row1, buf), row2 ); \ + row4 = _mm_xor_si128( row4, row1 ); \ + row4 = _mm_roti_epi32(row4, -16); \ + row3 = _mm_add_epi32( row3, row4 ); \ + row2 = _mm_xor_si128( row2, row3 ); \ + row2 = _mm_roti_epi32(row2, -12); + +#define G2(row1,row2,row3,row4,buf) \ + row1 = _mm_add_epi32( _mm_add_epi32( row1, buf), row2 ); \ + row4 = _mm_xor_si128( row4, row1 ); \ + row4 = _mm_roti_epi32(row4, -8); \ + row3 = _mm_add_epi32( row3, row4 ); \ + row2 = _mm_xor_si128( row2, row3 ); \ + row2 = _mm_roti_epi32(row2, -7); + +#define DIAGONALIZE(row1,row2,row3,row4) \ + row4 = _mm_shuffle_epi32( row4, _MM_SHUFFLE(2,1,0,3) ); \ + row3 = _mm_shuffle_epi32( row3, _MM_SHUFFLE(1,0,3,2) ); \ + row2 = _mm_shuffle_epi32( row2, _MM_SHUFFLE(0,3,2,1) ); + +#define UNDIAGONALIZE(row1,row2,row3,row4) \ + row4 = _mm_shuffle_epi32( row4, _MM_SHUFFLE(0,3,2,1) ); \ + row3 = _mm_shuffle_epi32( row3, _MM_SHUFFLE(1,0,3,2) ); \ + row2 = _mm_shuffle_epi32( row2, _MM_SHUFFLE(2,1,0,3) ); + +#if defined(HAVE_XOP) +#include "blake2s-load-xop.h" +#elif defined(HAVE_SSE41) +#include "blake2s-load-sse41.h" +#else +#include "blake2s-load-sse2.h" +#endif + +#define ROUND(r) \ + LOAD_MSG_ ##r ##_1(buf1); \ + G1(row1,row2,row3,row4,buf1); \ + LOAD_MSG_ ##r ##_2(buf2); \ + G2(row1,row2,row3,row4,buf2); \ + DIAGONALIZE(row1,row2,row3,row4); \ + LOAD_MSG_ ##r ##_3(buf3); \ + G1(row1,row2,row3,row4,buf3); \ + LOAD_MSG_ ##r ##_4(buf4); \ + G2(row1,row2,row3,row4,buf4); \ + UNDIAGONALIZE(row1,row2,row3,row4); \ + +#endif + diff --git a/cbits/blake2/blake2s.c b/cbits/blake2/blake2s.c new file mode 100644 index 0000000..67bfcc7 --- /dev/null +++ b/cbits/blake2/blake2s.c @@ -0,0 +1,387 @@ +/* + BLAKE2 reference source code package - optimized C implementations + + Written in 2012 by Samuel Neves + + To the extent possible under law, the author(s) have dedicated all copyright + and related and neighboring rights to this software to the public domain + worldwide. This software is distributed without any warranty. + + You should have received a copy of the CC0 Public Domain Dedication along with + this software. If not, see . +*/ + +#include +#include +#include + +#include "blake2.h" +#include "blake2-impl.h" + +#include "blake2-config.h" + + +#include +#if defined(HAVE_SSSE3) +#include +#endif +#if defined(HAVE_SSE41) +#include +#endif +#if defined(HAVE_AVX) +#include +#endif +#if defined(HAVE_XOP) +#include +#endif + +#include "blake2s-round.h" + +static const uint32_t blake2s_IV[8] = +{ + 0x6A09E667UL, 0xBB67AE85UL, 0x3C6EF372UL, 0xA54FF53AUL, + 0x510E527FUL, 0x9B05688CUL, 0x1F83D9ABUL, 0x5BE0CD19UL +}; + +static const uint8_t blake2s_sigma[10][16] = +{ + { 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15 } , + { 14, 10, 4, 8, 9, 15, 13, 6, 1, 12, 0, 2, 11, 7, 5, 3 } , + { 11, 8, 12, 0, 5, 2, 15, 13, 10, 14, 3, 6, 7, 1, 9, 4 } , + { 7, 9, 3, 1, 13, 12, 11, 14, 2, 6, 5, 10, 4, 0, 15, 8 } , + { 9, 0, 5, 7, 2, 4, 10, 15, 14, 1, 11, 12, 6, 8, 3, 13 } , + { 2, 12, 6, 10, 0, 11, 8, 3, 4, 13, 7, 5, 15, 14, 1, 9 } , + { 12, 5, 1, 15, 14, 13, 4, 10, 0, 7, 6, 3, 9, 2, 8, 11 } , + { 13, 11, 7, 14, 12, 1, 3, 9, 5, 0, 15, 4, 8, 6, 2, 10 } , + { 6, 15, 14, 9, 11, 3, 0, 8, 12, 2, 13, 7, 1, 4, 10, 5 } , + { 10, 2, 8, 4, 7, 6, 1, 5, 15, 11, 9, 14, 3, 12, 13 , 0 } , +}; + + +/* Some helper functions, not necessarily useful */ +static inline int blake2s_set_lastnode( blake2s_state *S ) +{ + S->f[1] = -1; + return 0; +} + +static inline int blake2s_clear_lastnode( blake2s_state *S ) +{ + S->f[1] = 0; + return 0; +} + +static inline int blake2s_set_lastblock( blake2s_state *S ) +{ + if( S->last_node ) blake2s_set_lastnode( S ); + + S->f[0] = -1; + return 0; +} + +static inline int blake2s_clear_lastblock( blake2s_state *S ) +{ + if( S->last_node ) blake2s_clear_lastnode( S ); + + S->f[0] = 0; + return 0; +} + +static inline int blake2s_increment_counter( blake2s_state *S, const uint32_t inc ) +{ + uint64_t t = ( ( uint64_t )S->t[1] << 32 ) | S->t[0]; + t += inc; + S->t[0] = ( uint32_t )( t >> 0 ); + S->t[1] = ( uint32_t )( t >> 32 ); + return 0; +} + + +// Parameter-related functions +static inline int blake2s_param_set_digest_length( blake2s_param *P, const uint8_t digest_length ) +{ + P->digest_length = digest_length; + return 0; +} + +static inline int blake2s_param_set_fanout( blake2s_param *P, const uint8_t fanout ) +{ + P->fanout = fanout; + return 0; +} + +static inline int blake2s_param_set_max_depth( blake2s_param *P, const uint8_t depth ) +{ + P->depth = depth; + return 0; +} + +static inline int blake2s_param_set_leaf_length( blake2s_param *P, const uint32_t leaf_length ) +{ + P->leaf_length = leaf_length; + return 0; +} + +static inline int blake2s_param_set_node_offset( blake2s_param *P, const uint64_t node_offset ) +{ + store48( P->node_offset, node_offset ); + return 0; +} + +static inline int blake2s_param_set_node_depth( blake2s_param *P, const uint8_t node_depth ) +{ + P->node_depth = node_depth; + return 0; +} + +static inline int blake2s_param_set_inner_length( blake2s_param *P, const uint8_t inner_length ) +{ + P->inner_length = inner_length; + return 0; +} + +static inline int blake2s_param_set_salt( blake2s_param *P, const uint8_t salt[BLAKE2S_SALTBYTES] ) +{ + memcpy( P->salt, salt, BLAKE2S_SALTBYTES ); + return 0; +} + +static inline int blake2s_param_set_personal( blake2s_param *P, const uint8_t personal[BLAKE2S_PERSONALBYTES] ) +{ + memcpy( P->personal, personal, BLAKE2S_PERSONALBYTES ); + return 0; +} + +static inline int blake2s_init0( blake2s_state *S ) +{ + memset( S, 0, sizeof( blake2s_state ) ); + + for( int i = 0; i < 8; ++i ) S->h[i] = blake2s_IV[i]; + + return 0; +} + +/* init2 xors IV with input parameter block */ +__attribute__((visibility ("hidden"))) +int blake2s_init_param( blake2s_state *S, const blake2s_param *P ) +{ + //blake2s_init0( S ); + const uint8_t * v = ( const uint8_t * )( blake2s_IV ); + const uint8_t * p = ( const uint8_t * )( P ); + uint8_t * h = ( uint8_t * )( S->h ); + /* IV XOR ParamBlock */ + memset( S, 0, sizeof( blake2s_state ) ); + + for( int i = 0; i < BLAKE2S_OUTBYTES; ++i ) h[i] = v[i] ^ p[i]; + + return 0; +} + + +/* Some sort of default parameter block initialization, for sequential blake2s */ +__attribute__((visibility ("hidden"))) +int blake2s_init( blake2s_state *S, const uint8_t outlen ) +{ + /* Move interval verification here? */ + if ( ( !outlen ) || ( outlen > BLAKE2S_OUTBYTES ) ) return -1; + + const blake2s_param P = + { + outlen, + 0, + 1, + 1, + 0, + {0}, + 0, + 0, + {0}, + {0} + }; + return blake2s_init_param( S, &P ); +} + +__attribute__((visibility ("hidden"))) +int blake2s_init_key( blake2s_state *S, const uint8_t outlen, const void *key, const uint8_t keylen ) +{ + /* Move interval verification here? */ + if ( ( !outlen ) || ( outlen > BLAKE2S_OUTBYTES ) ) return -1; + + if ( ( !key ) || ( !keylen ) || keylen > BLAKE2S_KEYBYTES ) return -1; + + const blake2s_param P = + { + outlen, + keylen, + 1, + 1, + 0, + {0}, + 0, + 0, + {0}, + {0} + }; + + if( blake2s_init_param( S, &P ) < 0 ) + return -1; + + { + uint8_t block[BLAKE2S_BLOCKBYTES]; + memset( block, 0, BLAKE2S_BLOCKBYTES ); + memcpy( block, key, keylen ); + blake2s_update( S, block, BLAKE2S_BLOCKBYTES ); + secure_zero_memory( block, BLAKE2S_BLOCKBYTES ); /* Burn the key from stack */ + } + return 0; +} + + +static inline int blake2s_compress( blake2s_state *S, const uint8_t block[BLAKE2S_BLOCKBYTES] ) +{ + __m128i row1, row2, row3, row4; + __m128i buf1, buf2, buf3, buf4; +#if defined(HAVE_SSE41) + __m128i t0, t1; +#if !defined(HAVE_XOP) + __m128i t2; +#endif +#endif + __m128i ff0, ff1; +#if defined(HAVE_SSSE3) && !defined(HAVE_XOP) + const __m128i r8 = _mm_set_epi8( 12, 15, 14, 13, 8, 11, 10, 9, 4, 7, 6, 5, 0, 3, 2, 1 ); + const __m128i r16 = _mm_set_epi8( 13, 12, 15, 14, 9, 8, 11, 10, 5, 4, 7, 6, 1, 0, 3, 2 ); +#endif +#if defined(HAVE_SSE41) + const __m128i m0 = LOADU( block + 00 ); + const __m128i m1 = LOADU( block + 16 ); + const __m128i m2 = LOADU( block + 32 ); + const __m128i m3 = LOADU( block + 48 ); +#else + const uint32_t m0 = ( ( uint32_t * )block )[ 0]; + const uint32_t m1 = ( ( uint32_t * )block )[ 1]; + const uint32_t m2 = ( ( uint32_t * )block )[ 2]; + const uint32_t m3 = ( ( uint32_t * )block )[ 3]; + const uint32_t m4 = ( ( uint32_t * )block )[ 4]; + const uint32_t m5 = ( ( uint32_t * )block )[ 5]; + const uint32_t m6 = ( ( uint32_t * )block )[ 6]; + const uint32_t m7 = ( ( uint32_t * )block )[ 7]; + const uint32_t m8 = ( ( uint32_t * )block )[ 8]; + const uint32_t m9 = ( ( uint32_t * )block )[ 9]; + const uint32_t m10 = ( ( uint32_t * )block )[10]; + const uint32_t m11 = ( ( uint32_t * )block )[11]; + const uint32_t m12 = ( ( uint32_t * )block )[12]; + const uint32_t m13 = ( ( uint32_t * )block )[13]; + const uint32_t m14 = ( ( uint32_t * )block )[14]; + const uint32_t m15 = ( ( uint32_t * )block )[15]; +#endif + row1 = ff0 = LOADU( &S->h[0] ); + row2 = ff1 = LOADU( &S->h[4] ); + row3 = _mm_setr_epi32( 0x6A09E667, 0xBB67AE85, 0x3C6EF372, 0xA54FF53A ); + row4 = _mm_xor_si128( _mm_setr_epi32( 0x510E527F, 0x9B05688C, 0x1F83D9AB, 0x5BE0CD19 ), LOADU( &S->t[0] ) ); + ROUND( 0 ); + ROUND( 1 ); + ROUND( 2 ); + ROUND( 3 ); + ROUND( 4 ); + ROUND( 5 ); + ROUND( 6 ); + ROUND( 7 ); + ROUND( 8 ); + ROUND( 9 ); + STOREU( &S->h[0], _mm_xor_si128( ff0, _mm_xor_si128( row1, row3 ) ) ); + STOREU( &S->h[4], _mm_xor_si128( ff1, _mm_xor_si128( row2, row4 ) ) ); + return 0; +} + +/* inlen now in bytes */ +__attribute__((visibility ("hidden"))) +int blake2s_update( blake2s_state *S, const uint8_t *in, uint64_t inlen ) +{ + while( inlen > 0 ) + { + size_t left = S->buflen; + size_t fill = 2 * BLAKE2S_BLOCKBYTES - left; + + if( inlen > fill ) + { + memcpy( S->buf + left, in, fill ); // Fill buffer + S->buflen += fill; + blake2s_increment_counter( S, BLAKE2S_BLOCKBYTES ); + blake2s_compress( S, S->buf ); // Compress + memcpy( S->buf, S->buf + BLAKE2S_BLOCKBYTES, BLAKE2S_BLOCKBYTES ); // Shift buffer left + S->buflen -= BLAKE2S_BLOCKBYTES; + in += fill; + inlen -= fill; + } + else // inlen <= fill + { + memcpy( S->buf + left, in, inlen ); + S->buflen += inlen; // Be lazy, do not compress + in += inlen; + inlen -= inlen; + } + } + + return 0; +} + +/* Is this correct? */ +__attribute__((visibility ("hidden"))) +int blake2s_final( blake2s_state *S, uint8_t *out, uint8_t outlen ) +{ + uint8_t buffer[BLAKE2S_OUTBYTES] = {0}; + + if( outlen > BLAKE2S_OUTBYTES ) + return -1; + + if( S->buflen > BLAKE2S_BLOCKBYTES ) + { + blake2s_increment_counter( S, BLAKE2S_BLOCKBYTES ); + blake2s_compress( S, S->buf ); + S->buflen -= BLAKE2S_BLOCKBYTES; + memcpy( S->buf, S->buf + BLAKE2S_BLOCKBYTES, S->buflen ); + } + + blake2s_increment_counter( S, ( uint32_t )S->buflen ); + blake2s_set_lastblock( S ); + memset( S->buf + S->buflen, 0, 2 * BLAKE2S_BLOCKBYTES - S->buflen ); /* Padding */ + blake2s_compress( S, S->buf ); + + for( int i = 0; i < 8; ++i ) /* Output full hash to temp buffer */ + store32( buffer + sizeof( S->h[i] ) * i, S->h[i] ); + + memcpy( out, buffer, outlen ); + return 0; +} + +/* inlen, at least, should be uint64_t. Others can be size_t. */ +__attribute__((visibility ("hidden"))) +int blake2s( uint8_t *out, const void *in, const void *key, const uint8_t outlen, const uint64_t inlen, uint8_t keylen ) +{ + blake2s_state S[1]; + + /* Verify parameters */ + if ( NULL == in && inlen > 0 ) return -1; + + if ( NULL == out ) return -1; + + if ( NULL == key && keylen > 0) return -1; + + if( !outlen || outlen > BLAKE2S_OUTBYTES ) return -1; + + if( keylen > BLAKE2S_KEYBYTES ) return -1; + + if( keylen > 0 ) + { + if( blake2s_init_key( S, outlen, key, keylen ) < 0 ) return -1; + } + else + { + if( blake2s_init( S, outlen ) < 0 ) return -1; + } + + blake2s_update( S, ( const uint8_t * )in, inlen ); + blake2s_final( S, out, outlen ); + return 0; +} diff --git a/cbits/blake2/blake2sp.c b/cbits/blake2/blake2sp.c new file mode 100644 index 0000000..3daf712 --- /dev/null +++ b/cbits/blake2/blake2sp.c @@ -0,0 +1,259 @@ +/* + BLAKE2 reference source code package - optimized C implementations + + Written in 2012 by Samuel Neves + + To the extent possible under law, the author(s) have dedicated all copyright + and related and neighboring rights to this software to the public domain + worldwide. This software is distributed without any warranty. + + You should have received a copy of the CC0 Public Domain Dedication along with + this software. If not, see . +*/ + +#include +#include +#include + +#if defined(_OPENMP) +#include +#endif + +#include "blake2.h" +#include "blake2-impl.h" + +#define PARALLELISM_DEGREE 8 + +static inline int blake2sp_init_leaf( blake2s_state *S, uint8_t outlen, uint8_t keylen, uint64_t offset ) +{ + blake2s_param P[1]; + P->digest_length = outlen; + P->key_length = keylen; + P->fanout = PARALLELISM_DEGREE; + P->depth = 2; + P->leaf_length = 0; + store48( P->node_offset, offset ); + P->node_depth = 0; + P->inner_length = BLAKE2S_OUTBYTES; + memset( P->salt, 0, sizeof( P->salt ) ); + memset( P->personal, 0, sizeof( P->personal ) ); + return blake2s_init_param( S, P ); +} + +static inline int blake2sp_init_root( blake2s_state *S, uint8_t outlen, uint8_t keylen ) +{ + blake2s_param P[1]; + P->digest_length = outlen; + P->key_length = keylen; + P->fanout = PARALLELISM_DEGREE; + P->depth = 2; + P->leaf_length = 0; + store48( P->node_offset, 0ULL ); + P->node_depth = 1; + P->inner_length = BLAKE2S_OUTBYTES; + memset( P->salt, 0, sizeof( P->salt ) ); + memset( P->personal, 0, sizeof( P->personal ) ); + return blake2s_init_param( S, P ); +} + +__attribute__((visibility ("hidden"))) +int blake2sp_init( blake2sp_state *S, const uint8_t outlen ) +{ + if( !outlen || outlen > BLAKE2S_OUTBYTES ) return -1; + + memset( S->buf, 0, sizeof( S->buf ) ); + S->buflen = 0; + + if( blake2sp_init_root( S->R, outlen, 0 ) < 0 ) + return -1; + + for( size_t i = 0; i < PARALLELISM_DEGREE; ++i ) + if( blake2sp_init_leaf( S->S[i], outlen, 0, i ) < 0 ) return -1; + + S->R->last_node = 1; + S->S[PARALLELISM_DEGREE - 1]->last_node = 1; + return 0; +} + +__attribute__((visibility ("hidden"))) +int blake2sp_init_key( blake2sp_state *S, const uint8_t outlen, const void *key, const uint8_t keylen ) +{ + if( !outlen || outlen > BLAKE2S_OUTBYTES ) return -1; + + if( !key || !keylen || keylen > BLAKE2S_KEYBYTES ) return -1; + + memset( S->buf, 0, sizeof( S->buf ) ); + S->buflen = 0; + + if( blake2sp_init_root( S->R, outlen, keylen ) < 0 ) + return -1; + + for( size_t i = 0; i < PARALLELISM_DEGREE; ++i ) + if( blake2sp_init_leaf( S->S[i], outlen, keylen, i ) < 0 ) return -1; + + S->R->last_node = 1; + S->S[PARALLELISM_DEGREE - 1]->last_node = 1; + { + uint8_t block[BLAKE2S_BLOCKBYTES]; + memset( block, 0, BLAKE2S_BLOCKBYTES ); + memcpy( block, key, keylen ); + + for( size_t i = 0; i < PARALLELISM_DEGREE; ++i ) + blake2s_update( S->S[i], block, BLAKE2S_BLOCKBYTES ); + + secure_zero_memory( block, BLAKE2S_BLOCKBYTES ); /* Burn the key from stack */ + } + return 0; +} + +__attribute__((visibility ("hidden"))) +int blake2sp_update( blake2sp_state *S, const uint8_t *in, uint64_t inlen ) +{ + size_t left = S->buflen; + size_t fill = sizeof( S->buf ) - left; + + if( left && inlen >= fill ) + { + memcpy( S->buf + left, in, fill ); + + for( size_t i = 0; i < PARALLELISM_DEGREE; ++i ) + blake2s_update( S->S[i], S->buf + i * BLAKE2S_BLOCKBYTES, BLAKE2S_BLOCKBYTES ); + + in += fill; + inlen -= fill; + left = 0; + } + +#if defined(_OPENMP) + #pragma omp parallel shared(S), num_threads(PARALLELISM_DEGREE) +#else + + for( size_t id__ = 0; id__ < PARALLELISM_DEGREE; ++id__ ) +#endif + { +#if defined(_OPENMP) + size_t id__ = omp_get_thread_num(); +#endif + uint64_t inlen__ = inlen; + const uint8_t *in__ = ( const uint8_t * )in; + in__ += id__ * BLAKE2S_BLOCKBYTES; + + while( inlen__ >= PARALLELISM_DEGREE * BLAKE2S_BLOCKBYTES ) + { + blake2s_update( S->S[id__], in__, BLAKE2S_BLOCKBYTES ); + in__ += PARALLELISM_DEGREE * BLAKE2S_BLOCKBYTES; + inlen__ -= PARALLELISM_DEGREE * BLAKE2S_BLOCKBYTES; + } + } + + in += inlen - inlen % ( PARALLELISM_DEGREE * BLAKE2S_BLOCKBYTES ); + inlen %= PARALLELISM_DEGREE * BLAKE2S_BLOCKBYTES; + + if( inlen > 0 ) + memcpy( S->buf + left, in, inlen ); + + S->buflen = left + inlen; + return 0; +} + +__attribute__((visibility ("hidden"))) +int blake2sp_final( blake2sp_state *S, uint8_t *out, const uint8_t outlen ) +{ + uint8_t hash[PARALLELISM_DEGREE][BLAKE2S_OUTBYTES]; + + for( size_t i = 0; i < PARALLELISM_DEGREE; ++i ) + { + if( S->buflen > i * BLAKE2S_BLOCKBYTES ) + { + size_t left = S->buflen - i * BLAKE2S_BLOCKBYTES; + + if( left > BLAKE2S_BLOCKBYTES ) left = BLAKE2S_BLOCKBYTES; + + blake2s_update( S->S[i], S->buf + i * BLAKE2S_BLOCKBYTES, left ); + } + + blake2s_final( S->S[i], hash[i], BLAKE2S_OUTBYTES ); + } + + for( size_t i = 0; i < PARALLELISM_DEGREE; ++i ) + blake2s_update( S->R, hash[i], BLAKE2S_OUTBYTES ); + + return blake2s_final( S->R, out, outlen ); +} + +__attribute__((visibility ("hidden"))) +int blake2sp( uint8_t *out, const void *in, const void *key, uint8_t outlen, uint64_t inlen, uint8_t keylen ) +{ + uint8_t hash[PARALLELISM_DEGREE][BLAKE2S_OUTBYTES]; + blake2s_state S[PARALLELISM_DEGREE][1]; + blake2s_state FS[1]; + + /* Verify parameters */ + if ( NULL == in && inlen > 0 ) return -1; + + if ( NULL == out ) return -1; + + if ( NULL == key && keylen > 0) return -1; + + if( !outlen || outlen > BLAKE2S_OUTBYTES ) return -1; + + if( keylen > BLAKE2S_KEYBYTES ) return -1; + + for( size_t i = 0; i < PARALLELISM_DEGREE; ++i ) + if( blake2sp_init_leaf( S[i], outlen, keylen, i ) < 0 ) return -1; + + S[PARALLELISM_DEGREE - 1]->last_node = 1; // mark last node + + if( keylen > 0 ) + { + uint8_t block[BLAKE2S_BLOCKBYTES]; + memset( block, 0, BLAKE2S_BLOCKBYTES ); + memcpy( block, key, keylen ); + + for( size_t i = 0; i < PARALLELISM_DEGREE; ++i ) + blake2s_update( S[i], block, BLAKE2S_BLOCKBYTES ); + + secure_zero_memory( block, BLAKE2S_BLOCKBYTES ); /* Burn the key from stack */ + } + +#if defined(_OPENMP) + #pragma omp parallel shared(S,hash), num_threads(PARALLELISM_DEGREE) +#else + + for( size_t id__ = 0; id__ < PARALLELISM_DEGREE; ++id__ ) +#endif + { +#if defined(_OPENMP) + size_t id__ = omp_get_thread_num(); +#endif + uint64_t inlen__ = inlen; + const uint8_t *in__ = ( const uint8_t * )in; + in__ += id__ * BLAKE2S_BLOCKBYTES; + + while( inlen__ >= PARALLELISM_DEGREE * BLAKE2S_BLOCKBYTES ) + { + blake2s_update( S[id__], in__, BLAKE2S_BLOCKBYTES ); + in__ += PARALLELISM_DEGREE * BLAKE2S_BLOCKBYTES; + inlen__ -= PARALLELISM_DEGREE * BLAKE2S_BLOCKBYTES; + } + + if( inlen__ > id__ * BLAKE2S_BLOCKBYTES ) + { + const size_t left = inlen__ - id__ * BLAKE2S_BLOCKBYTES; + const size_t len = left <= BLAKE2S_BLOCKBYTES ? left : BLAKE2S_BLOCKBYTES; + blake2s_update( S[id__], in__, len ); + } + + blake2s_final( S[id__], hash[id__], BLAKE2S_OUTBYTES ); + } + + if( blake2sp_init_root( FS, outlen, keylen ) < 0 ) + return -1; + + FS->last_node = 1; + + for( size_t i = 0; i < PARALLELISM_DEGREE; ++i ) + blake2s_update( FS, hash[i], BLAKE2S_OUTBYTES ); + + return blake2s_final( FS, out, outlen ); +} diff --git a/cbits/cryptonite_blake2b.c b/cbits/cryptonite_blake2b.c new file mode 100644 index 0000000..1f2dd8d --- /dev/null +++ b/cbits/cryptonite_blake2b.c @@ -0,0 +1,16 @@ +#include "cryptonite_blake2b.h" + +void cryptonite_blake2b_init(blake2b_ctx *ctx) +{ + blake2b_init(ctx, 64); +} + +void cryptonite_blake2b_update(blake2b_ctx *ctx, const uint8_t *data, uint32_t len) +{ + blake2b_update(ctx, data, len); +} + +void cryptonite_blake2b_finalize(blake2b_ctx *ctx, uint8_t *out) +{ + blake2b_final(ctx, out, 64); +} diff --git a/cbits/cryptonite_blake2b.h b/cbits/cryptonite_blake2b.h new file mode 100644 index 0000000..8edc8ac --- /dev/null +++ b/cbits/cryptonite_blake2b.h @@ -0,0 +1,12 @@ +#ifndef CRYPTOHASH_BLAKE2B_H +#define CRYPTOHASH_BLAKE2B_H + +#include "blake2.h" + +typedef blake2b_state blake2b_ctx; + +void cryptonite_blake2b_init(blake2b_ctx *ctx); +void cryptonite_blake2b_update(blake2b_ctx *ctx, const uint8_t *data, uint32_t len); +void cryptonite_blake2b_finalize(blake2b_ctx *ctx, uint8_t *out); + +#endif diff --git a/cbits/cryptonite_blake2bp.c b/cbits/cryptonite_blake2bp.c new file mode 100644 index 0000000..318833d --- /dev/null +++ b/cbits/cryptonite_blake2bp.c @@ -0,0 +1,16 @@ +#include "cryptonite_blake2bp.h" + +void cryptonite_blake2bp_init(blake2bp_ctx *ctx) +{ + blake2bp_init(ctx, 64); +} + +void cryptonite_blake2bp_update(blake2bp_ctx *ctx, const uint8_t *data, uint32_t len) +{ + blake2bp_update(ctx, data, len); +} + +void cryptonite_blake2bp_finalize(blake2bp_ctx *ctx, uint8_t *out) +{ + blake2bp_final(ctx, out, 64); +} diff --git a/cbits/cryptonite_blake2bp.h b/cbits/cryptonite_blake2bp.h new file mode 100644 index 0000000..4ccf54c --- /dev/null +++ b/cbits/cryptonite_blake2bp.h @@ -0,0 +1,12 @@ +#ifndef CRYPTOHASH_BLAKE2BP_H +#define CRYPTOHASH_BLAKE2BP_H + +#include "blake2.h" + +typedef blake2bp_state blake2bp_ctx; + +void cryptonite_blake2bp_init(blake2bp_ctx *ctx); +void cryptonite_blake2bp_update(blake2bp_ctx *ctx, const uint8_t *data, uint32_t len); +void cryptonite_blake2bp_finalize(blake2bp_ctx *ctx, uint8_t *out); + +#endif diff --git a/cbits/cryptonite_blake2s.c b/cbits/cryptonite_blake2s.c new file mode 100644 index 0000000..07d9dac --- /dev/null +++ b/cbits/cryptonite_blake2s.c @@ -0,0 +1,16 @@ +#include "cryptonite_blake2s.h" + +void cryptonite_blake2s_init(blake2s_ctx *ctx) +{ + blake2s_init(ctx, 32); +} + +void cryptonite_blake2s_update(blake2s_ctx *ctx, const uint8_t *data, uint32_t len) +{ + blake2s_update(ctx, data, len); +} + +void cryptonite_blake2s_finalize(blake2s_ctx *ctx, uint8_t *out) +{ + blake2s_final(ctx, out, 32); +} diff --git a/cbits/cryptonite_blake2s.h b/cbits/cryptonite_blake2s.h new file mode 100644 index 0000000..d847a17 --- /dev/null +++ b/cbits/cryptonite_blake2s.h @@ -0,0 +1,12 @@ +#ifndef CRYPTOHASH_BLAKE2S_H +#define CRYPTOHASH_BLAKE2S_H + +#include "blake2.h" + +typedef blake2s_state blake2s_ctx; + +void cryptonite_blake2s_init(blake2s_ctx *ctx); +void cryptonite_blake2s_update(blake2s_ctx *ctx, const uint8_t *data, uint32_t len); +void cryptonite_blake2s_finalize(blake2s_ctx *ctx, uint8_t *out); + +#endif diff --git a/cbits/cryptonite_blake2sp.c b/cbits/cryptonite_blake2sp.c new file mode 100644 index 0000000..9a8580a --- /dev/null +++ b/cbits/cryptonite_blake2sp.c @@ -0,0 +1,16 @@ +#include "cryptonite_blake2sp.h" + +void cryptonite_blake2sp_init(blake2sp_ctx *ctx) +{ + blake2sp_init(ctx, 32); +} + +void cryptonite_blake2sp_update(blake2sp_ctx *ctx, const uint8_t *data, uint32_t len) +{ + blake2sp_update(ctx, data, len); +} + +void cryptonite_blake2sp_finalize(blake2sp_ctx *ctx, uint8_t *out) +{ + blake2sp_final(ctx, out, 32); +} diff --git a/cbits/cryptonite_blake2sp.h b/cbits/cryptonite_blake2sp.h new file mode 100644 index 0000000..7d040a4 --- /dev/null +++ b/cbits/cryptonite_blake2sp.h @@ -0,0 +1,12 @@ +#ifndef CRYPTOHASH_BLAKE2SP_H +#define CRYPTOHASH_BLAKE2SP_H + +#include "blake2.h" + +typedef blake2sp_state blake2sp_ctx; + +void cryptonite_blake2sp_init(blake2sp_ctx *ctx); +void cryptonite_blake2sp_update(blake2sp_ctx *ctx, const uint8_t *data, uint32_t len); +void cryptonite_blake2sp_finalize(blake2sp_ctx *ctx, uint8_t *out); + +#endif diff --git a/cryptonite.cabal b/cryptonite.cabal index 41a017a..5359b07 100644 --- a/cryptonite.cabal +++ b/cryptonite.cabal @@ -39,6 +39,8 @@ extra-source-files: cbits/*.h cbits/aes/*.h cbits/ed25519/*.h cbits/p256/*.h + cbits/blake2/*.h + cbits/blake2/*.c cbits/aes/x86ni_impl.c tests/*.hs @@ -147,6 +149,10 @@ Library Crypto.Hash.MD2 Crypto.Hash.MD4 Crypto.Hash.MD5 + Crypto.Hash.BLAKE2s + Crypto.Hash.BLAKE2sp + Crypto.Hash.BLAKE2b + Crypto.Hash.BLAKE2bp Crypto.Hash.RIPEMD160 Crypto.Hash.Skein256 Crypto.Hash.Skein512 @@ -172,6 +178,7 @@ Library , ghc-prim ghc-options: -Wall -fwarn-tabs -optc-O3 -fno-warn-unused-imports default-language: Haskell2010 + cc-options: -std=gnu99 C-sources: cbits/cryptonite_chacha.c , cbits/cryptonite_salsa.c , cbits/cryptonite_rc4.c @@ -192,6 +199,10 @@ Library , cbits/cryptonite_md2.c , cbits/cryptonite_md4.c , cbits/cryptonite_md5.c + , cbits/cryptonite_blake2s.c + , cbits/cryptonite_blake2sp.c + , cbits/cryptonite_blake2b.c + , cbits/cryptonite_blake2bp.c , cbits/cryptonite_ripemd.c , cbits/cryptonite_skein256.c , cbits/cryptonite_skein512.c @@ -199,7 +210,11 @@ Library , cbits/cryptonite_whirlpool.c , cbits/cryptonite_scrypt.c , cbits/cryptonite_sysrand.c - include-dirs: cbits cbits/ed25519 + , cbits/blake2/blake2s.c + , cbits/blake2/blake2sp.c + , cbits/blake2/blake2b.c + , cbits/blake2/blake2bp.c + include-dirs: cbits cbits/ed25519 cbits/blake2 -- FIXME armel or mispel is also little endian. -- might be a good idea to also add a runtime autodetect mode. diff --git a/gen/Gen.hs b/gen/Gen.hs index 0e15191..8ea7d17 100644 --- a/gen/Gen.hs +++ b/gen/Gen.hs @@ -21,7 +21,11 @@ data GenHashModule = GenHashModule hashModules = -- module header hash ctx dg blk - [ GenHashModule "MD2" "md2.h" "md2" 96 16 16 [] + [ GenHashModule "BLAKE2s" "blake2.h" "blake2s" 185 32 64 [] + , GenHashModule "BLAKE2sp" "blake2.h" "blake2sp" 2185 32 64 [] + , GenHashModule "BLAKE2b" "blake2.h" "blake2b" 361 64 128 [] + , GenHashModule "BLAKE2bp" "blake2.h" "blake2sp" 2325 64 128 [] + , GenHashModule "MD2" "md2.h" "md2" 96 16 16 [] , GenHashModule "MD4" "md4.h" "md4" 96 16 64 [] , GenHashModule "MD5" "md5.h" "md5" 96 16 64 [] , GenHashModule "SHA1" "sha1.h" "sha1" 96 20 64 []