Skip to content

Commit

Permalink
8-way AVX2 implementation for double SHA256 on 64-byte inputs
Browse files Browse the repository at this point in the history
  • Loading branch information
sipa committed May 29, 2018
1 parent 230294b commit 4437d6e
Show file tree
Hide file tree
Showing 8 changed files with 402 additions and 5 deletions.
22 changes: 22 additions & 0 deletions configure.ac
Original file line number Diff line number Diff line change
Expand Up @@ -305,6 +305,7 @@ 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 $SSE42_CXXFLAGS"
Expand Down Expand Up @@ -347,6 +348,25 @@ AC_COMPILE_IFELSE([AC_LANG_PROGRAM([[
)
CXXFLAGS="$TEMP_CXXFLAGS"

TEMP_CXXFLAGS="$CXXFLAGS"
CXXFLAGS="$CXXFLAGS $AVX2_CXXFLAGS"
AC_MSG_CHECKING(for AVX2 intrinsics)
AC_COMPILE_IFELSE([AC_LANG_PROGRAM([[
#include <stdint.h>
#if defined(_MSC_VER)
#include <immintrin.h>
#elif defined(__GNUC__) && defined(__AVX2__)
#include <x86intrin.h>
#endif
]],[[
__m256i l = _mm256_set1_epi32(0);
return _mm256_extract_epi32(l, 7);
]])],
[ AC_MSG_RESULT(yes); enable_avx2=yes; AC_DEFINE(ENABLE_AVX2, 1, [Define this symbol to build code that uses AVX2 intrinsics]) ],
[ AC_MSG_RESULT(no)]
)
CXXFLAGS="$TEMP_CXXFLAGS"

CPPFLAGS="$CPPFLAGS -DHAVE_BUILD_INFO -D__STDC_FORMAT_MACROS"

AC_ARG_WITH([utils],
Expand Down Expand Up @@ -1266,6 +1286,7 @@ 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])
Expand Down Expand Up @@ -1305,6 +1326,7 @@ 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(USE_UPNP)
AC_SUBST(USE_QRCODE)
Expand Down
15 changes: 14 additions & 1 deletion src/Makefile.am
Original file line number Diff line number Diff line change
Expand Up @@ -31,6 +31,7 @@ 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
LIBBITCOINQT=qt/libbitcoinqt.a
LIBSECP256K1=secp256k1/libsecp256k1.la

Expand All @@ -52,6 +53,7 @@ $(LIBSECP256K1): $(wildcard secp256k1/src/*) $(wildcard secp256k1/include/*)
EXTRA_LIBRARIES += \
$(LIBBITCOIN_CRYPTO) \
$(LIBBITCOIN_CRYPTO_SSE41) \
$(LIBBITCOIN_CRYPTO_AVX2) \
$(LIBBITCOIN_UTIL) \
$(LIBBITCOIN_COMMON) \
$(LIBBITCOIN_CONSENSUS) \
Expand Down Expand Up @@ -299,6 +301,14 @@ 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

# consensus: shared between all executables that validate any consensus rules.
libbitcoin_consensus_a_CPPFLAGS = $(AM_CPPFLAGS) $(BITCOIN_INCLUDES)
libbitcoin_consensus_a_CXXFLAGS = $(AM_CXXFLAGS) $(PIE_FLAGS)
Expand Down Expand Up @@ -422,6 +432,7 @@ bitcoind_LDADD = \
$(LIBBITCOIN_CONSENSUS) \
$(LIBBITCOIN_CRYPTO) \
$(LIBBITCOIN_CRYPTO_SSE41) \
$(LIBBITCOIN_CRYPTO_AVX2) \
$(LIBLEVELDB) \
$(LIBLEVELDB_SSE42) \
$(LIBMEMENV) \
Expand All @@ -444,7 +455,8 @@ bitcoin_cli_LDADD = \
$(LIBUNIVALUE) \
$(LIBBITCOIN_UTIL) \
$(LIBBITCOIN_CRYPTO) \
$(LIBBITCOIN_CRYPTO_SSE41)
$(LIBBITCOIN_CRYPTO_SSE41) \
$(LIBBITCOIN_CRYPTO_AVX2)

bitcoin_cli_LDADD += $(BOOST_LIBS) $(SSL_LIBS) $(CRYPTO_LIBS) $(EVENT_LIBS)
#
Expand All @@ -466,6 +478,7 @@ bitcoin_tx_LDADD = \
$(LIBBITCOIN_CONSENSUS) \
$(LIBBITCOIN_CRYPTO) \
$(LIBBITCOIN_CRYPTO_SSE41) \
$(LIBBITCOIN_CRYPTO_AVX2) \
$(LIBSECP256K1)

bitcoin_tx_LDADD += $(BOOST_LIBS) $(CRYPTO_LIBS)
Expand Down
1 change: 1 addition & 0 deletions src/Makefile.bench.include
Original file line number Diff line number Diff line change
Expand Up @@ -40,6 +40,7 @@ bench_bench_bitcoin_LDADD = \
$(LIBBITCOIN_CONSENSUS) \
$(LIBBITCOIN_CRYPTO) \
$(LIBBITCOIN_CRYPTO_SSE41) \
$(LIBBITCOIN_CRYPTO_AVX2) \
$(LIBLEVELDB) \
$(LIBLEVELDB_SSE42) \
$(LIBMEMENV) \
Expand Down
2 changes: 1 addition & 1 deletion src/Makefile.qt.include
Original file line number Diff line number Diff line change
Expand Up @@ -407,7 +407,7 @@ endif
if ENABLE_ZMQ
qt_bitcoin_qt_LDADD += $(LIBBITCOIN_ZMQ) $(ZMQ_LIBS)
endif
qt_bitcoin_qt_LDADD += $(LIBBITCOIN_CLI) $(LIBBITCOIN_COMMON) $(LIBBITCOIN_UTIL) $(LIBBITCOIN_CONSENSUS) $(LIBBITCOIN_CRYPTO) $(LIBBITCOIN_CRYPTO_SSE41) $(LIBUNIVALUE) $(LIBLEVELDB) $(LIBLEVELDB_SSE42) $(LIBMEMENV) \
qt_bitcoin_qt_LDADD += $(LIBBITCOIN_CLI) $(LIBBITCOIN_COMMON) $(LIBBITCOIN_UTIL) $(LIBBITCOIN_CONSENSUS) $(LIBBITCOIN_CRYPTO) $(LIBBITCOIN_CRYPTO_SSE41) $(LIBBITCOIN_CRYPTO_AVX2) $(LIBUNIVALUE) $(LIBLEVELDB) $(LIBLEVELDB_SSE42) $(LIBMEMENV) \
$(BOOST_LIBS) $(QT_LIBS) $(QT_DBUS_LIBS) $(QR_LIBS) $(PROTOBUF_LIBS) $(BDB_LIBS) $(SSL_LIBS) $(CRYPTO_LIBS) $(MINIUPNPC_LIBS) $(LIBSECP256K1) \
$(EVENT_PTHREADS_LIBS) $(EVENT_LIBS)
qt_bitcoin_qt_LDFLAGS = $(RELDFLAGS) $(AM_LDFLAGS) $(QT_LDFLAGS) $(LIBTOOL_APP_LDFLAGS)
Expand Down
2 changes: 1 addition & 1 deletion src/Makefile.qttest.include
Original file line number Diff line number Diff line change
Expand Up @@ -62,7 +62,7 @@ endif
if ENABLE_ZMQ
qt_test_test_bitcoin_qt_LDADD += $(LIBBITCOIN_ZMQ) $(ZMQ_LIBS)
endif
qt_test_test_bitcoin_qt_LDADD += $(LIBBITCOIN_CLI) $(LIBBITCOIN_COMMON) $(LIBBITCOIN_UTIL) $(LIBBITCOIN_CONSENSUS) $(LIBBITCOIN_CRYPTO) $(LIBBITCOIN_CRYPTO_SSE41) $(LIBUNIVALUE) $(LIBLEVELDB) \
qt_test_test_bitcoin_qt_LDADD += $(LIBBITCOIN_CLI) $(LIBBITCOIN_COMMON) $(LIBBITCOIN_UTIL) $(LIBBITCOIN_CONSENSUS) $(LIBBITCOIN_CRYPTO) $(LIBBITCOIN_CRYPTO_SSE41) $(LIBBITCOIN_CRYPTO_AVX2) $(LIBUNIVALUE) $(LIBLEVELDB) \
$(LIBLEVELDB_SSE42) $(LIBMEMENV) $(BOOST_LIBS) $(QT_DBUS_LIBS) $(QT_TEST_LIBS) $(QT_LIBS) \
$(QR_LIBS) $(PROTOBUF_LIBS) $(BDB_LIBS) $(SSL_LIBS) $(CRYPTO_LIBS) $(MINIUPNPC_LIBS) $(LIBSECP256K1) \
$(EVENT_PTHREADS_LIBS) $(EVENT_LIBS)
Expand Down
3 changes: 2 additions & 1 deletion src/Makefile.test.include
Original file line number Diff line number Diff line change
Expand Up @@ -109,7 +109,7 @@ if ENABLE_WALLET
test_test_bitcoin_LDADD += $(LIBBITCOIN_WALLET)
endif

test_test_bitcoin_LDADD += $(LIBBITCOIN_SERVER) $(LIBBITCOIN_CLI) $(LIBBITCOIN_COMMON) $(LIBBITCOIN_UTIL) $(LIBBITCOIN_CONSENSUS) $(LIBBITCOIN_CRYPTO) $(LIBBITCOIN_CRYPTO_SSE41) $(LIBUNIVALUE) \
test_test_bitcoin_LDADD += $(LIBBITCOIN_SERVER) $(LIBBITCOIN_CLI) $(LIBBITCOIN_COMMON) $(LIBBITCOIN_UTIL) $(LIBBITCOIN_CONSENSUS) $(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)

Expand All @@ -135,6 +135,7 @@ test_test_bitcoin_fuzzy_LDADD = \
$(LIBBITCOIN_CONSENSUS) \
$(LIBBITCOIN_CRYPTO) \
$(LIBBITCOIN_CRYPTO_SSE41) \
$(LIBBITCOIN_CRYPTO_AVX2) \
$(LIBSECP256K1)

test_test_bitcoin_fuzzy_LDADD += $(BOOST_LIBS) $(CRYPTO_LIBS)
Expand Down
33 changes: 32 additions & 1 deletion src/crypto/sha256.cpp
Original file line number Diff line number Diff line change
Expand Up @@ -24,6 +24,11 @@ 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
{
Expand Down Expand Up @@ -471,19 +476,37 @@ 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<sha256_sse4::Transform>;
#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
Expand Down Expand Up @@ -553,6 +576,14 @@ 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);
Expand Down
Loading

0 comments on commit 4437d6e

Please sign in to comment.