From ae3dc3f13413e55a0fe1dcf4fba7e82a5445f2cf Mon Sep 17 00:00:00 2001 From: Pieter Wuille Date: Wed, 27 Sep 2017 01:45:12 -0700 Subject: [PATCH] 8-way AVX2 implementation for double SHA256 on 64-byte inputs (cherry picked from commit bitcoin/bitcoin@4437d6e1f3107a20a8c7b66be8b4b972a82e3b28) --- configure.ac | 22 +++ src/Makefile.am | 13 ++ src/Makefile.bench.include | 1 + src/Makefile.gtest.include | 1 + src/Makefile.test.include | 2 +- src/crypto/sha256.cpp | 33 +++- src/crypto/sha256_avx2.cpp | 329 +++++++++++++++++++++++++++++++++++++ 7 files changed, 399 insertions(+), 2 deletions(-) create mode 100644 src/crypto/sha256_avx2.cpp diff --git a/configure.ac b/configure.ac index 293bb87bd..cbff02929 100644 --- a/configure.ac +++ b/configure.ac @@ -335,6 +335,7 @@ fi # compatibility. AX_CHECK_COMPILE_FLAG([-msse4.2],[[SSE42_CXXFLAGS="-msse4.2"]],,[[$CXXFLAG_WERROR]]) AX_CHECK_COMPILE_FLAG([-msse4.1],[[SSE41_CXXFLAGS="-msse4.1"]],,[[$CXXFLAG_WERROR]]) +AX_CHECK_COMPILE_FLAG([-mavx -mavx2],[[AVX2_CXXFLAGS="-mavx -mavx2"]],,[[$CXXFLAG_WERROR]]) TEMP_CXXFLAGS="$CXXFLAGS" CXXFLAGS="$CXXFLAGS $SSE42_CXXFLAGS" @@ -377,6 +378,25 @@ AC_COMPILE_IFELSE([AC_LANG_PROGRAM([[ ) CXXFLAGS="$TEMP_CXXFLAGS" +TEMP_CXXFLAGS="$CXXFLAGS" +CXXFLAGS="$CXXFLAGS $AVX2_CXXFLAGS" +AC_MSG_CHECKING(for AVX2 intrinsics) +AC_COMPILE_IFELSE([AC_LANG_PROGRAM([[ + #include + #if defined(_MSC_VER) + #include + #elif defined(__GNUC__) && defined(__AVX2__) + #include + #endif + ]],[[ + __m256i l = _mm256_set1_epi32(0); + return _mm256_extract_epi32(l, 7); + ]])], + [ AC_MSG_RESULT(yes); enable_avx2=yes; AC_DEFINE(ENABLE_AVX2, 1, [Define this symbol to build code that uses AVX2 intrinsics]) ], + [ AC_MSG_RESULT(no)] +) +CXXFLAGS="$TEMP_CXXFLAGS" + CPPFLAGS="$CPPFLAGS -DHAVE_BUILD_INFO -D__STDC_FORMAT_MACROS" AC_ARG_WITH([utils], @@ -1014,6 +1034,7 @@ AM_CONDITIONAL([GLIBC_BACK_COMPAT],[test x$use_glibc_compat = xyes]) AM_CONDITIONAL([HARDEN],[test x$use_hardening = xyes]) AM_CONDITIONAL([ENABLE_HWCRC32],[test x$enable_hwcrc32 = xyes]) AM_CONDITIONAL([ENABLE_SSE41],[test x$enable_sse41 = xyes]) +AM_CONDITIONAL([ENABLE_AVX2],[test x$enable_avx2 = xyes]) AM_CONDITIONAL([USE_ASM],[test x$use_asm = xyes]) AC_DEFINE(CLIENT_VERSION_MAJOR, _CLIENT_VERSION_MAJOR, [Major version]) @@ -1049,6 +1070,7 @@ AC_SUBST(SANITIZER_CXXFLAGS) AC_SUBST(SANITIZER_LDFLAGS) AC_SUBST(SSE42_CXXFLAGS) AC_SUBST(SSE41_CXXFLAGS) +AC_SUBST(AVX2_CXXFLAGS) AC_SUBST(LIBTOOL_APP_LDFLAGS) AC_SUBST(BOOST_LIBS) AC_SUBST(TESTDEFS) diff --git a/src/Makefile.am b/src/Makefile.am index 1daf6756a..21e6cf527 100644 --- a/src/Makefile.am +++ b/src/Makefile.am @@ -27,6 +27,7 @@ LIBBITCOIN_CLI=libbitcoin_cli.a LIBBITCOIN_UTIL=libbitcoin_util.a LIBBITCOIN_CRYPTO=crypto/libbitcoin_crypto.a LIBBITCOIN_CRYPTO_SSE41=crypto/libbitcoin_crypto_sse41.a +LIBBITCOIN_CRYPTO_AVX2=crypto/libbitcoin_crypto_avx2.a LIBRUSTZCASH=$(top_builddir)/target/$(RUST_TARGET)/release/librustzcash.a LIBSECP256K1=secp256k1/libsecp256k1.la LIBUNIVALUE=univalue/libunivalue.la @@ -143,6 +144,7 @@ $(LIBUNIVALUE): $(wildcard univalue/lib/*) $(wildcard univalue/include/*) EXTRA_LIBRARIES += \ $(LIBBITCOIN_CRYPTO) \ $(LIBBITCOIN_CRYPTO_SSE41) \ + $(LIBBITCOIN_CRYPTO_AVX2) \ $(LIBBITCOIN_UTIL) \ $(LIBBITCOIN_COMMON) \ $(LIBBITCOIN_SERVER) \ @@ -456,6 +458,14 @@ crypto_libbitcoin_crypto_sse41_a_CPPFLAGS += -DENABLE_SSE41 endif crypto_libbitcoin_crypto_sse41_a_SOURCES = crypto/sha256_sse41.cpp +crypto_libbitcoin_crypto_avx2_a_CXXFLAGS = $(AM_CXXFLAGS) $(PIE_FLAGS) +crypto_libbitcoin_crypto_avx2_a_CPPFLAGS = $(AM_CPPFLAGS) +if ENABLE_AVX2 +crypto_libbitcoin_crypto_avx2_a_CXXFLAGS += $(AVX2_CXXFLAGS) +crypto_libbitcoin_crypto_avx2_a_CPPFLAGS += -DENABLE_AVX2 +endif +crypto_libbitcoin_crypto_avx2_a_SOURCES = crypto/sha256_avx2.cpp + # common: shared between zcashd and non-server tools libbitcoin_common_a_CPPFLAGS = $(AM_CPPFLAGS) $(BITCOIN_INCLUDES) libbitcoin_common_a_CXXFLAGS = $(AM_CXXFLAGS) $(PIE_FLAGS) @@ -556,6 +566,7 @@ zcashd_LDADD = \ $(LIBBITCOIN_ZMQ) \ $(LIBBITCOIN_CRYPTO) \ $(LIBBITCOIN_CRYPTO_SSE41) \ + $(LIBBITCOIN_CRYPTO_AVX2) \ $(LIBZCASH) \ $(LIBRUSTZCASH) \ $(LIBLEVELDB) \ @@ -592,6 +603,7 @@ zcash_cli_LDADD = \ $(LIBRUSTZCASH) \ $(LIBBITCOIN_CRYPTO) \ $(LIBBITCOIN_CRYPTO_SSE41) \ + $(LIBBITCOIN_CRYPTO_AVX2) \ $(LIBSECP256K1) \ $(LIBZCASH_LIBS) # @@ -616,6 +628,7 @@ zcash_tx_LDADD = \ $(LIBRUSTZCASH) \ $(LIBBITCOIN_CRYPTO) \ $(LIBBITCOIN_CRYPTO_SSE41) \ + $(LIBBITCOIN_CRYPTO_AVX2) \ $(LIBZCASH_LIBS) zcash_tx_LDADD += $(BOOST_LIBS) diff --git a/src/Makefile.bench.include b/src/Makefile.bench.include index 0465ce9a9..d177f7b51 100644 --- a/src/Makefile.bench.include +++ b/src/Makefile.bench.include @@ -33,6 +33,7 @@ bench_bench_bitcoin_LDADD = \ $(LIBBITCOIN_UTIL) \ $(LIBBITCOIN_CRYPTO) \ $(LIBBITCOIN_CRYPTO_SSE41) \ + $(LIBBITCOIN_CRYPTO_AVX2) \ $(LIBLEVELDB) \ $(LIBLEVELDB_SSE42) \ $(LIBMEMENV) \ diff --git a/src/Makefile.gtest.include b/src/Makefile.gtest.include index c176e449e..b0c92fd6d 100644 --- a/src/Makefile.gtest.include +++ b/src/Makefile.gtest.include @@ -78,6 +78,7 @@ zcash_gtest_LDADD = \ $(LIBBITCOIN_PROTON) \ $(LIBBITCOIN_CRYPTO) \ $(LIBBITCOIN_CRYPTO_SSE41) \ + $(LIBBITCOIN_CRYPTO_AVX2) \ $(LIBUNIVALUE) \ $(LIBLEVELDB) \ $(LIBLEVELDB_SSE42) \ diff --git a/src/Makefile.test.include b/src/Makefile.test.include index 6a8bcbb17..1be490b1f 100644 --- a/src/Makefile.test.include +++ b/src/Makefile.test.include @@ -138,7 +138,7 @@ if ENABLE_WALLET test_test_bitcoin_LDADD += $(LIBBITCOIN_WALLET) endif -test_test_bitcoin_LDADD += $(LIBBITCOIN_SERVER) $(LIBBITCOIN_CLI) $(LIBBITCOIN_COMMON) $(LIBBITCOIN_UTIL) $(LIBBITCOIN_CRYPTO) $(LIBBITCOIN_CRYPTO_SSE41) $(LIBUNIVALUE) \ +test_test_bitcoin_LDADD += $(LIBBITCOIN_SERVER) $(LIBBITCOIN_CLI) $(LIBBITCOIN_COMMON) $(LIBBITCOIN_UTIL) $(LIBBITCOIN_CRYPTO) $(LIBBITCOIN_CRYPTO_SSE41) $(LIBBITCOIN_CRYPTO_AVX2) $(LIBUNIVALUE) \ $(LIBLEVELDB) $(LIBLEVELDB_SSE42) $(LIBMEMENV) $(BOOST_LIBS) $(BOOST_UNIT_TEST_FRAMEWORK_LIB) $(LIBSECP256K1) $(EVENT_LIBS) $(EVENT_PTHREADS_LIBS) test_test_bitcoin_CXXFLAGS = $(AM_CXXFLAGS) $(PIE_FLAGS) diff --git a/src/crypto/sha256.cpp b/src/crypto/sha256.cpp index 5baedb51a..5d59905f3 100644 --- a/src/crypto/sha256.cpp +++ b/src/crypto/sha256.cpp @@ -26,6 +26,11 @@ namespace sha256d64_sse41 void Transform_4way(unsigned char* out, const unsigned char* in); } +namespace sha256d64_avx2 +{ +void Transform_8way(unsigned char* out, const unsigned char* in); +} + // Internal implementation code. namespace { @@ -473,19 +478,37 @@ bool SelfTest(TransformType tr) { TransformType Transform = sha256::Transform; TransformD64Type TransformD64 = sha256::TransformD64; TransformD64Type TransformD64_4way = nullptr; +TransformD64Type TransformD64_8way = nullptr; + +#if defined(USE_ASM) && (defined(__x86_64__) || defined(__amd64__)) +// We can't use cpuid.h's __get_cpuid as it does not support subleafs. +void inline cpuid(uint32_t leaf, uint32_t subleaf, uint32_t& a, uint32_t& b, uint32_t& c, uint32_t& d) +{ + __asm__ ("cpuid" : "=a"(a), "=b"(b), "=c"(c), "=d"(d) : "0"(leaf), "2"(subleaf)); +} +#endif } // namespace + std::string SHA256AutoDetect() { std::string ret = "standard"; #if defined(USE_ASM) && (defined(__x86_64__) || defined(__amd64__)) uint32_t eax, ebx, ecx, edx; - if (__get_cpuid(1, &eax, &ebx, &ecx, &edx) && (ecx >> 19) & 1) { + cpuid(1, 0, eax, ebx, ecx, edx); + if ((ecx >> 19) & 1) { Transform = sha256_sse4::Transform; TransformD64 = TransformD64Wrapper; #if defined(ENABLE_SSE41) && !defined(BUILD_BITCOIN_INTERNAL) TransformD64_4way = sha256d64_sse41::Transform_4way; ret = "sse4(1way+4way)"; +#if defined(ENABLE_AVX2) && !defined(BUILD_BITCOIN_INTERNAL) + cpuid(7, 0, eax, ebx, ecx, edx); + if ((ebx >> 5) & 1) { + TransformD64_8way = sha256d64_avx2::Transform_8way; + ret += ",avx2(8way)"; + } +#endif #else ret = "sse4"; #endif @@ -564,6 +587,14 @@ CSHA256& CSHA256::Reset() void SHA256D64(unsigned char* out, const unsigned char* in, size_t blocks) { + if (TransformD64_8way) { + while (blocks >= 8) { + TransformD64_8way(out, in); + out += 256; + in += 512; + blocks -= 8; + } + } if (TransformD64_4way) { while (blocks >= 4) { TransformD64_4way(out, in); diff --git a/src/crypto/sha256_avx2.cpp b/src/crypto/sha256_avx2.cpp new file mode 100644 index 000000000..f45c1d4ab --- /dev/null +++ b/src/crypto/sha256_avx2.cpp @@ -0,0 +1,329 @@ +#ifdef ENABLE_AVX2 + +#include +#if defined(_MSC_VER) +#include +#elif defined(__GNUC__) +#include +#endif + +#include "crypto/sha256.h" +#include "crypto/common.h" + +namespace sha256d64_avx2 { +namespace { + +__m256i inline K(uint32_t x) { return _mm256_set1_epi32(x); } + +__m256i inline Add(__m256i x, __m256i y) { return _mm256_add_epi32(x, y); } +__m256i inline Add(__m256i x, __m256i y, __m256i z) { return Add(Add(x, y), z); } +__m256i inline Add(__m256i x, __m256i y, __m256i z, __m256i w) { return Add(Add(x, y), Add(z, w)); } +__m256i inline Add(__m256i x, __m256i y, __m256i z, __m256i w, __m256i v) { return Add(Add(x, y, z), Add(w, v)); } +__m256i inline Inc(__m256i& x, __m256i y) { x = Add(x, y); return x; } +__m256i inline Inc(__m256i& x, __m256i y, __m256i z) { x = Add(x, y, z); return x; } +__m256i inline Inc(__m256i& x, __m256i y, __m256i z, __m256i w) { x = Add(x, y, z, w); return x; } +__m256i inline Xor(__m256i x, __m256i y) { return _mm256_xor_si256(x, y); } +__m256i inline Xor(__m256i x, __m256i y, __m256i z) { return Xor(Xor(x, y), z); } +__m256i inline Or(__m256i x, __m256i y) { return _mm256_or_si256(x, y); } +__m256i inline And(__m256i x, __m256i y) { return _mm256_and_si256(x, y); } +__m256i inline ShR(__m256i x, int n) { return _mm256_srli_epi32(x, n); } +__m256i inline ShL(__m256i x, int n) { return _mm256_slli_epi32(x, n); } + +__m256i inline Ch(__m256i x, __m256i y, __m256i z) { return Xor(z, And(x, Xor(y, z))); } +__m256i inline Maj(__m256i x, __m256i y, __m256i z) { return Or(And(x, y), And(z, Or(x, y))); } +__m256i inline Sigma0(__m256i x) { return Xor(Or(ShR(x, 2), ShL(x, 30)), Or(ShR(x, 13), ShL(x, 19)), Or(ShR(x, 22), ShL(x, 10))); } +__m256i inline Sigma1(__m256i x) { return Xor(Or(ShR(x, 6), ShL(x, 26)), Or(ShR(x, 11), ShL(x, 21)), Or(ShR(x, 25), ShL(x, 7))); } +__m256i inline sigma0(__m256i x) { return Xor(Or(ShR(x, 7), ShL(x, 25)), Or(ShR(x, 18), ShL(x, 14)), ShR(x, 3)); } +__m256i inline sigma1(__m256i x) { return Xor(Or(ShR(x, 17), ShL(x, 15)), Or(ShR(x, 19), ShL(x, 13)), ShR(x, 10)); } + +/** One round of SHA-256. */ +void inline __attribute__((always_inline)) Round(__m256i a, __m256i b, __m256i c, __m256i& d, __m256i e, __m256i f, __m256i g, __m256i& h, __m256i k) +{ + __m256i t1 = Add(h, Sigma1(e), Ch(e, f, g), k); + __m256i t2 = Add(Sigma0(a), Maj(a, b, c)); + d = Add(d, t1); + h = Add(t1, t2); +} + +__m256i inline Read8(const unsigned char* chunk, int offset) { + __m256i ret = _mm256_set_epi32( + ReadLE32(chunk + 0 + offset), + ReadLE32(chunk + 64 + offset), + ReadLE32(chunk + 128 + offset), + ReadLE32(chunk + 192 + offset), + ReadLE32(chunk + 256 + offset), + ReadLE32(chunk + 320 + offset), + ReadLE32(chunk + 384 + offset), + ReadLE32(chunk + 448 + offset) + ); + return _mm256_shuffle_epi8(ret, _mm256_set_epi32(0x0C0D0E0FUL, 0x08090A0BUL, 0x04050607UL, 0x00010203UL, 0x0C0D0E0FUL, 0x08090A0BUL, 0x04050607UL, 0x00010203UL)); +} + +void inline Write8(unsigned char* out, int offset, __m256i v) { + v = _mm256_shuffle_epi8(v, _mm256_set_epi32(0x0C0D0E0FUL, 0x08090A0BUL, 0x04050607UL, 0x00010203UL, 0x0C0D0E0FUL, 0x08090A0BUL, 0x04050607UL, 0x00010203UL)); + WriteLE32(out + 0 + offset, _mm256_extract_epi32(v, 7)); + WriteLE32(out + 32 + offset, _mm256_extract_epi32(v, 6)); + WriteLE32(out + 64 + offset, _mm256_extract_epi32(v, 5)); + WriteLE32(out + 96 + offset, _mm256_extract_epi32(v, 4)); + WriteLE32(out + 128 + offset, _mm256_extract_epi32(v, 3)); + WriteLE32(out + 160 + offset, _mm256_extract_epi32(v, 2)); + WriteLE32(out + 192 + offset, _mm256_extract_epi32(v, 1)); + WriteLE32(out + 224 + offset, _mm256_extract_epi32(v, 0)); +} + +} + +void Transform_8way(unsigned char* out, const unsigned char* in) +{ + // Transform 1 + __m256i a = K(0x6a09e667ul); + __m256i b = K(0xbb67ae85ul); + __m256i c = K(0x3c6ef372ul); + __m256i d = K(0xa54ff53aul); + __m256i e = K(0x510e527ful); + __m256i f = K(0x9b05688cul); + __m256i g = K(0x1f83d9abul); + __m256i h = K(0x5be0cd19ul); + + __m256i w0, w1, w2, w3, w4, w5, w6, w7, w8, w9, w10, w11, w12, w13, w14, w15; + + Round(a, b, c, d, e, f, g, h, Add(K(0x428a2f98ul), w0 = Read8(in, 0))); + Round(h, a, b, c, d, e, f, g, Add(K(0x71374491ul), w1 = Read8(in, 4))); + Round(g, h, a, b, c, d, e, f, Add(K(0xb5c0fbcful), w2 = Read8(in, 8))); + Round(f, g, h, a, b, c, d, e, Add(K(0xe9b5dba5ul), w3 = Read8(in, 12))); + Round(e, f, g, h, a, b, c, d, Add(K(0x3956c25bul), w4 = Read8(in, 16))); + Round(d, e, f, g, h, a, b, c, Add(K(0x59f111f1ul), w5 = Read8(in, 20))); + Round(c, d, e, f, g, h, a, b, Add(K(0x923f82a4ul), w6 = Read8(in, 24))); + Round(b, c, d, e, f, g, h, a, Add(K(0xab1c5ed5ul), w7 = Read8(in, 28))); + Round(a, b, c, d, e, f, g, h, Add(K(0xd807aa98ul), w8 = Read8(in, 32))); + Round(h, a, b, c, d, e, f, g, Add(K(0x12835b01ul), w9 = Read8(in, 36))); + Round(g, h, a, b, c, d, e, f, Add(K(0x243185beul), w10 = Read8(in, 40))); + Round(f, g, h, a, b, c, d, e, Add(K(0x550c7dc3ul), w11 = Read8(in, 44))); + Round(e, f, g, h, a, b, c, d, Add(K(0x72be5d74ul), w12 = Read8(in, 48))); + Round(d, e, f, g, h, a, b, c, Add(K(0x80deb1feul), w13 = Read8(in, 52))); + Round(c, d, e, f, g, h, a, b, Add(K(0x9bdc06a7ul), w14 = Read8(in, 56))); + Round(b, c, d, e, f, g, h, a, Add(K(0xc19bf174ul), w15 = Read8(in, 60))); + Round(a, b, c, d, e, f, g, h, Add(K(0xe49b69c1ul), Inc(w0, sigma1(w14), w9, sigma0(w1)))); + Round(h, a, b, c, d, e, f, g, Add(K(0xefbe4786ul), Inc(w1, sigma1(w15), w10, sigma0(w2)))); + Round(g, h, a, b, c, d, e, f, Add(K(0x0fc19dc6ul), Inc(w2, sigma1(w0), w11, sigma0(w3)))); + Round(f, g, h, a, b, c, d, e, Add(K(0x240ca1ccul), Inc(w3, sigma1(w1), w12, sigma0(w4)))); + Round(e, f, g, h, a, b, c, d, Add(K(0x2de92c6ful), Inc(w4, sigma1(w2), w13, sigma0(w5)))); + Round(d, e, f, g, h, a, b, c, Add(K(0x4a7484aaul), Inc(w5, sigma1(w3), w14, sigma0(w6)))); + Round(c, d, e, f, g, h, a, b, Add(K(0x5cb0a9dcul), Inc(w6, sigma1(w4), w15, sigma0(w7)))); + Round(b, c, d, e, f, g, h, a, Add(K(0x76f988daul), Inc(w7, sigma1(w5), w0, sigma0(w8)))); + Round(a, b, c, d, e, f, g, h, Add(K(0x983e5152ul), Inc(w8, sigma1(w6), w1, sigma0(w9)))); + Round(h, a, b, c, d, e, f, g, Add(K(0xa831c66dul), Inc(w9, sigma1(w7), w2, sigma0(w10)))); + Round(g, h, a, b, c, d, e, f, Add(K(0xb00327c8ul), Inc(w10, sigma1(w8), w3, sigma0(w11)))); + Round(f, g, h, a, b, c, d, e, Add(K(0xbf597fc7ul), Inc(w11, sigma1(w9), w4, sigma0(w12)))); + Round(e, f, g, h, a, b, c, d, Add(K(0xc6e00bf3ul), Inc(w12, sigma1(w10), w5, sigma0(w13)))); + Round(d, e, f, g, h, a, b, c, Add(K(0xd5a79147ul), Inc(w13, sigma1(w11), w6, sigma0(w14)))); + Round(c, d, e, f, g, h, a, b, Add(K(0x06ca6351ul), Inc(w14, sigma1(w12), w7, sigma0(w15)))); + Round(b, c, d, e, f, g, h, a, Add(K(0x14292967ul), Inc(w15, sigma1(w13), w8, sigma0(w0)))); + Round(a, b, c, d, e, f, g, h, Add(K(0x27b70a85ul), Inc(w0, sigma1(w14), w9, sigma0(w1)))); + Round(h, a, b, c, d, e, f, g, Add(K(0x2e1b2138ul), Inc(w1, sigma1(w15), w10, sigma0(w2)))); + Round(g, h, a, b, c, d, e, f, Add(K(0x4d2c6dfcul), Inc(w2, sigma1(w0), w11, sigma0(w3)))); + Round(f, g, h, a, b, c, d, e, Add(K(0x53380d13ul), Inc(w3, sigma1(w1), w12, sigma0(w4)))); + Round(e, f, g, h, a, b, c, d, Add(K(0x650a7354ul), Inc(w4, sigma1(w2), w13, sigma0(w5)))); + Round(d, e, f, g, h, a, b, c, Add(K(0x766a0abbul), Inc(w5, sigma1(w3), w14, sigma0(w6)))); + Round(c, d, e, f, g, h, a, b, Add(K(0x81c2c92eul), Inc(w6, sigma1(w4), w15, sigma0(w7)))); + Round(b, c, d, e, f, g, h, a, Add(K(0x92722c85ul), Inc(w7, sigma1(w5), w0, sigma0(w8)))); + Round(a, b, c, d, e, f, g, h, Add(K(0xa2bfe8a1ul), Inc(w8, sigma1(w6), w1, sigma0(w9)))); + Round(h, a, b, c, d, e, f, g, Add(K(0xa81a664bul), Inc(w9, sigma1(w7), w2, sigma0(w10)))); + Round(g, h, a, b, c, d, e, f, Add(K(0xc24b8b70ul), Inc(w10, sigma1(w8), w3, sigma0(w11)))); + Round(f, g, h, a, b, c, d, e, Add(K(0xc76c51a3ul), Inc(w11, sigma1(w9), w4, sigma0(w12)))); + Round(e, f, g, h, a, b, c, d, Add(K(0xd192e819ul), Inc(w12, sigma1(w10), w5, sigma0(w13)))); + Round(d, e, f, g, h, a, b, c, Add(K(0xd6990624ul), Inc(w13, sigma1(w11), w6, sigma0(w14)))); + Round(c, d, e, f, g, h, a, b, Add(K(0xf40e3585ul), Inc(w14, sigma1(w12), w7, sigma0(w15)))); + Round(b, c, d, e, f, g, h, a, Add(K(0x106aa070ul), Inc(w15, sigma1(w13), w8, sigma0(w0)))); + Round(a, b, c, d, e, f, g, h, Add(K(0x19a4c116ul), Inc(w0, sigma1(w14), w9, sigma0(w1)))); + Round(h, a, b, c, d, e, f, g, Add(K(0x1e376c08ul), Inc(w1, sigma1(w15), w10, sigma0(w2)))); + Round(g, h, a, b, c, d, e, f, Add(K(0x2748774cul), Inc(w2, sigma1(w0), w11, sigma0(w3)))); + Round(f, g, h, a, b, c, d, e, Add(K(0x34b0bcb5ul), Inc(w3, sigma1(w1), w12, sigma0(w4)))); + Round(e, f, g, h, a, b, c, d, Add(K(0x391c0cb3ul), Inc(w4, sigma1(w2), w13, sigma0(w5)))); + Round(d, e, f, g, h, a, b, c, Add(K(0x4ed8aa4aul), Inc(w5, sigma1(w3), w14, sigma0(w6)))); + Round(c, d, e, f, g, h, a, b, Add(K(0x5b9cca4ful), Inc(w6, sigma1(w4), w15, sigma0(w7)))); + Round(b, c, d, e, f, g, h, a, Add(K(0x682e6ff3ul), Inc(w7, sigma1(w5), w0, sigma0(w8)))); + Round(a, b, c, d, e, f, g, h, Add(K(0x748f82eeul), Inc(w8, sigma1(w6), w1, sigma0(w9)))); + Round(h, a, b, c, d, e, f, g, Add(K(0x78a5636ful), Inc(w9, sigma1(w7), w2, sigma0(w10)))); + Round(g, h, a, b, c, d, e, f, Add(K(0x84c87814ul), Inc(w10, sigma1(w8), w3, sigma0(w11)))); + Round(f, g, h, a, b, c, d, e, Add(K(0x8cc70208ul), Inc(w11, sigma1(w9), w4, sigma0(w12)))); + Round(e, f, g, h, a, b, c, d, Add(K(0x90befffaul), Inc(w12, sigma1(w10), w5, sigma0(w13)))); + Round(d, e, f, g, h, a, b, c, Add(K(0xa4506cebul), Inc(w13, sigma1(w11), w6, sigma0(w14)))); + Round(c, d, e, f, g, h, a, b, Add(K(0xbef9a3f7ul), Inc(w14, sigma1(w12), w7, sigma0(w15)))); + Round(b, c, d, e, f, g, h, a, Add(K(0xc67178f2ul), Inc(w15, sigma1(w13), w8, sigma0(w0)))); + + a = Add(a, K(0x6a09e667ul)); + b = Add(b, K(0xbb67ae85ul)); + c = Add(c, K(0x3c6ef372ul)); + d = Add(d, K(0xa54ff53aul)); + e = Add(e, K(0x510e527ful)); + f = Add(f, K(0x9b05688cul)); + g = Add(g, K(0x1f83d9abul)); + h = Add(h, K(0x5be0cd19ul)); + + __m256i t0 = a, t1 = b, t2 = c, t3 = d, t4 = e, t5 = f, t6 = g, t7 = h; + + // Transform 2 + Round(a, b, c, d, e, f, g, h, K(0xc28a2f98ul)); + Round(h, a, b, c, d, e, f, g, K(0x71374491ul)); + Round(g, h, a, b, c, d, e, f, K(0xb5c0fbcful)); + Round(f, g, h, a, b, c, d, e, K(0xe9b5dba5ul)); + Round(e, f, g, h, a, b, c, d, K(0x3956c25bul)); + Round(d, e, f, g, h, a, b, c, K(0x59f111f1ul)); + Round(c, d, e, f, g, h, a, b, K(0x923f82a4ul)); + Round(b, c, d, e, f, g, h, a, K(0xab1c5ed5ul)); + Round(a, b, c, d, e, f, g, h, K(0xd807aa98ul)); + Round(h, a, b, c, d, e, f, g, K(0x12835b01ul)); + Round(g, h, a, b, c, d, e, f, K(0x243185beul)); + Round(f, g, h, a, b, c, d, e, K(0x550c7dc3ul)); + Round(e, f, g, h, a, b, c, d, K(0x72be5d74ul)); + Round(d, e, f, g, h, a, b, c, K(0x80deb1feul)); + Round(c, d, e, f, g, h, a, b, K(0x9bdc06a7ul)); + Round(b, c, d, e, f, g, h, a, K(0xc19bf374ul)); + Round(a, b, c, d, e, f, g, h, K(0x649b69c1ul)); + Round(h, a, b, c, d, e, f, g, K(0xf0fe4786ul)); + Round(g, h, a, b, c, d, e, f, K(0x0fe1edc6ul)); + Round(f, g, h, a, b, c, d, e, K(0x240cf254ul)); + Round(e, f, g, h, a, b, c, d, K(0x4fe9346ful)); + Round(d, e, f, g, h, a, b, c, K(0x6cc984beul)); + Round(c, d, e, f, g, h, a, b, K(0x61b9411eul)); + Round(b, c, d, e, f, g, h, a, K(0x16f988faul)); + Round(a, b, c, d, e, f, g, h, K(0xf2c65152ul)); + Round(h, a, b, c, d, e, f, g, K(0xa88e5a6dul)); + Round(g, h, a, b, c, d, e, f, K(0xb019fc65ul)); + Round(f, g, h, a, b, c, d, e, K(0xb9d99ec7ul)); + Round(e, f, g, h, a, b, c, d, K(0x9a1231c3ul)); + Round(d, e, f, g, h, a, b, c, K(0xe70eeaa0ul)); + Round(c, d, e, f, g, h, a, b, K(0xfdb1232bul)); + Round(b, c, d, e, f, g, h, a, K(0xc7353eb0ul)); + Round(a, b, c, d, e, f, g, h, K(0x3069bad5ul)); + Round(h, a, b, c, d, e, f, g, K(0xcb976d5ful)); + Round(g, h, a, b, c, d, e, f, K(0x5a0f118ful)); + Round(f, g, h, a, b, c, d, e, K(0xdc1eeefdul)); + Round(e, f, g, h, a, b, c, d, K(0x0a35b689ul)); + Round(d, e, f, g, h, a, b, c, K(0xde0b7a04ul)); + Round(c, d, e, f, g, h, a, b, K(0x58f4ca9dul)); + Round(b, c, d, e, f, g, h, a, K(0xe15d5b16ul)); + Round(a, b, c, d, e, f, g, h, K(0x007f3e86ul)); + Round(h, a, b, c, d, e, f, g, K(0x37088980ul)); + Round(g, h, a, b, c, d, e, f, K(0xa507ea32ul)); + Round(f, g, h, a, b, c, d, e, K(0x6fab9537ul)); + Round(e, f, g, h, a, b, c, d, K(0x17406110ul)); + Round(d, e, f, g, h, a, b, c, K(0x0d8cd6f1ul)); + Round(c, d, e, f, g, h, a, b, K(0xcdaa3b6dul)); + Round(b, c, d, e, f, g, h, a, K(0xc0bbbe37ul)); + Round(a, b, c, d, e, f, g, h, K(0x83613bdaul)); + Round(h, a, b, c, d, e, f, g, K(0xdb48a363ul)); + Round(g, h, a, b, c, d, e, f, K(0x0b02e931ul)); + Round(f, g, h, a, b, c, d, e, K(0x6fd15ca7ul)); + Round(e, f, g, h, a, b, c, d, K(0x521afacaul)); + Round(d, e, f, g, h, a, b, c, K(0x31338431ul)); + Round(c, d, e, f, g, h, a, b, K(0x6ed41a95ul)); + Round(b, c, d, e, f, g, h, a, K(0x6d437890ul)); + Round(a, b, c, d, e, f, g, h, K(0xc39c91f2ul)); + Round(h, a, b, c, d, e, f, g, K(0x9eccabbdul)); + Round(g, h, a, b, c, d, e, f, K(0xb5c9a0e6ul)); + Round(f, g, h, a, b, c, d, e, K(0x532fb63cul)); + Round(e, f, g, h, a, b, c, d, K(0xd2c741c6ul)); + Round(d, e, f, g, h, a, b, c, K(0x07237ea3ul)); + Round(c, d, e, f, g, h, a, b, K(0xa4954b68ul)); + Round(b, c, d, e, f, g, h, a, K(0x4c191d76ul)); + + w0 = Add(t0, a); + w1 = Add(t1, b); + w2 = Add(t2, c); + w3 = Add(t3, d); + w4 = Add(t4, e); + w5 = Add(t5, f); + w6 = Add(t6, g); + w7 = Add(t7, h); + + // Transform 3 + a = K(0x6a09e667ul); + b = K(0xbb67ae85ul); + c = K(0x3c6ef372ul); + d = K(0xa54ff53aul); + e = K(0x510e527ful); + f = K(0x9b05688cul); + g = K(0x1f83d9abul); + h = K(0x5be0cd19ul); + + Round(a, b, c, d, e, f, g, h, Add(K(0x428a2f98ul), w0)); + Round(h, a, b, c, d, e, f, g, Add(K(0x71374491ul), w1)); + Round(g, h, a, b, c, d, e, f, Add(K(0xb5c0fbcful), w2)); + Round(f, g, h, a, b, c, d, e, Add(K(0xe9b5dba5ul), w3)); + Round(e, f, g, h, a, b, c, d, Add(K(0x3956c25bul), w4)); + Round(d, e, f, g, h, a, b, c, Add(K(0x59f111f1ul), w5)); + Round(c, d, e, f, g, h, a, b, Add(K(0x923f82a4ul), w6)); + Round(b, c, d, e, f, g, h, a, Add(K(0xab1c5ed5ul), w7)); + Round(a, b, c, d, e, f, g, h, K(0x5807aa98ul)); + Round(h, a, b, c, d, e, f, g, K(0x12835b01ul)); + Round(g, h, a, b, c, d, e, f, K(0x243185beul)); + Round(f, g, h, a, b, c, d, e, K(0x550c7dc3ul)); + Round(e, f, g, h, a, b, c, d, K(0x72be5d74ul)); + Round(d, e, f, g, h, a, b, c, K(0x80deb1feul)); + Round(c, d, e, f, g, h, a, b, K(0x9bdc06a7ul)); + Round(b, c, d, e, f, g, h, a, K(0xc19bf274ul)); + Round(a, b, c, d, e, f, g, h, Add(K(0xe49b69c1ul), Inc(w0, sigma0(w1)))); + Round(h, a, b, c, d, e, f, g, Add(K(0xefbe4786ul), Inc(w1, K(0xa00000ul), sigma0(w2)))); + Round(g, h, a, b, c, d, e, f, Add(K(0x0fc19dc6ul), Inc(w2, sigma1(w0), sigma0(w3)))); + Round(f, g, h, a, b, c, d, e, Add(K(0x240ca1ccul), Inc(w3, sigma1(w1), sigma0(w4)))); + Round(e, f, g, h, a, b, c, d, Add(K(0x2de92c6ful), Inc(w4, sigma1(w2), sigma0(w5)))); + Round(d, e, f, g, h, a, b, c, Add(K(0x4a7484aaul), Inc(w5, sigma1(w3), sigma0(w6)))); + Round(c, d, e, f, g, h, a, b, Add(K(0x5cb0a9dcul), Inc(w6, sigma1(w4), K(0x100ul), sigma0(w7)))); + Round(b, c, d, e, f, g, h, a, Add(K(0x76f988daul), Inc(w7, sigma1(w5), w0, K(0x11002000ul)))); + Round(a, b, c, d, e, f, g, h, Add(K(0x983e5152ul), w8 = Add(K(0x80000000ul), sigma1(w6), w1))); + Round(h, a, b, c, d, e, f, g, Add(K(0xa831c66dul), w9 = Add(sigma1(w7), w2))); + Round(g, h, a, b, c, d, e, f, Add(K(0xb00327c8ul), w10 = Add(sigma1(w8), w3))); + Round(f, g, h, a, b, c, d, e, Add(K(0xbf597fc7ul), w11 = Add(sigma1(w9), w4))); + Round(e, f, g, h, a, b, c, d, Add(K(0xc6e00bf3ul), w12 = Add(sigma1(w10), w5))); + Round(d, e, f, g, h, a, b, c, Add(K(0xd5a79147ul), w13 = Add(sigma1(w11), w6))); + Round(c, d, e, f, g, h, a, b, Add(K(0x06ca6351ul), w14 = Add(sigma1(w12), w7, K(0x400022ul)))); + Round(b, c, d, e, f, g, h, a, Add(K(0x14292967ul), w15 = Add(K(0x100ul), sigma1(w13), w8, sigma0(w0)))); + Round(a, b, c, d, e, f, g, h, Add(K(0x27b70a85ul), Inc(w0, sigma1(w14), w9, sigma0(w1)))); + Round(h, a, b, c, d, e, f, g, Add(K(0x2e1b2138ul), Inc(w1, sigma1(w15), w10, sigma0(w2)))); + Round(g, h, a, b, c, d, e, f, Add(K(0x4d2c6dfcul), Inc(w2, sigma1(w0), w11, sigma0(w3)))); + Round(f, g, h, a, b, c, d, e, Add(K(0x53380d13ul), Inc(w3, sigma1(w1), w12, sigma0(w4)))); + Round(e, f, g, h, a, b, c, d, Add(K(0x650a7354ul), Inc(w4, sigma1(w2), w13, sigma0(w5)))); + Round(d, e, f, g, h, a, b, c, Add(K(0x766a0abbul), Inc(w5, sigma1(w3), w14, sigma0(w6)))); + Round(c, d, e, f, g, h, a, b, Add(K(0x81c2c92eul), Inc(w6, sigma1(w4), w15, sigma0(w7)))); + Round(b, c, d, e, f, g, h, a, Add(K(0x92722c85ul), Inc(w7, sigma1(w5), w0, sigma0(w8)))); + Round(a, b, c, d, e, f, g, h, Add(K(0xa2bfe8a1ul), Inc(w8, sigma1(w6), w1, sigma0(w9)))); + Round(h, a, b, c, d, e, f, g, Add(K(0xa81a664bul), Inc(w9, sigma1(w7), w2, sigma0(w10)))); + Round(g, h, a, b, c, d, e, f, Add(K(0xc24b8b70ul), Inc(w10, sigma1(w8), w3, sigma0(w11)))); + Round(f, g, h, a, b, c, d, e, Add(K(0xc76c51a3ul), Inc(w11, sigma1(w9), w4, sigma0(w12)))); + Round(e, f, g, h, a, b, c, d, Add(K(0xd192e819ul), Inc(w12, sigma1(w10), w5, sigma0(w13)))); + Round(d, e, f, g, h, a, b, c, Add(K(0xd6990624ul), Inc(w13, sigma1(w11), w6, sigma0(w14)))); + Round(c, d, e, f, g, h, a, b, Add(K(0xf40e3585ul), Inc(w14, sigma1(w12), w7, sigma0(w15)))); + Round(b, c, d, e, f, g, h, a, Add(K(0x106aa070ul), Inc(w15, sigma1(w13), w8, sigma0(w0)))); + Round(a, b, c, d, e, f, g, h, Add(K(0x19a4c116ul), Inc(w0, sigma1(w14), w9, sigma0(w1)))); + Round(h, a, b, c, d, e, f, g, Add(K(0x1e376c08ul), Inc(w1, sigma1(w15), w10, sigma0(w2)))); + Round(g, h, a, b, c, d, e, f, Add(K(0x2748774cul), Inc(w2, sigma1(w0), w11, sigma0(w3)))); + Round(f, g, h, a, b, c, d, e, Add(K(0x34b0bcb5ul), Inc(w3, sigma1(w1), w12, sigma0(w4)))); + Round(e, f, g, h, a, b, c, d, Add(K(0x391c0cb3ul), Inc(w4, sigma1(w2), w13, sigma0(w5)))); + Round(d, e, f, g, h, a, b, c, Add(K(0x4ed8aa4aul), Inc(w5, sigma1(w3), w14, sigma0(w6)))); + Round(c, d, e, f, g, h, a, b, Add(K(0x5b9cca4ful), Inc(w6, sigma1(w4), w15, sigma0(w7)))); + Round(b, c, d, e, f, g, h, a, Add(K(0x682e6ff3ul), Inc(w7, sigma1(w5), w0, sigma0(w8)))); + Round(a, b, c, d, e, f, g, h, Add(K(0x748f82eeul), Inc(w8, sigma1(w6), w1, sigma0(w9)))); + Round(h, a, b, c, d, e, f, g, Add(K(0x78a5636ful), Inc(w9, sigma1(w7), w2, sigma0(w10)))); + Round(g, h, a, b, c, d, e, f, Add(K(0x84c87814ul), Inc(w10, sigma1(w8), w3, sigma0(w11)))); + Round(f, g, h, a, b, c, d, e, Add(K(0x8cc70208ul), Inc(w11, sigma1(w9), w4, sigma0(w12)))); + Round(e, f, g, h, a, b, c, d, Add(K(0x90befffaul), Inc(w12, sigma1(w10), w5, sigma0(w13)))); + Round(d, e, f, g, h, a, b, c, Add(K(0xa4506cebul), Inc(w13, sigma1(w11), w6, sigma0(w14)))); + Round(c, d, e, f, g, h, a, b, Add(K(0xbef9a3f7ul), w14, sigma1(w12), w7, sigma0(w15))); + Round(b, c, d, e, f, g, h, a, Add(K(0xc67178f2ul), w15, sigma1(w13), w8, sigma0(w0))); + + // Output + Write8(out, 0, Add(a, K(0x6a09e667ul))); + Write8(out, 4, Add(b, K(0xbb67ae85ul))); + Write8(out, 8, Add(c, K(0x3c6ef372ul))); + Write8(out, 12, Add(d, K(0xa54ff53aul))); + Write8(out, 16, Add(e, K(0x510e527ful))); + Write8(out, 20, Add(f, K(0x9b05688cul))); + Write8(out, 24, Add(g, K(0x1f83d9abul))); + Write8(out, 28, Add(h, K(0x5be0cd19ul))); +} + +} + +#endif