Litecoin: Enable support for SSE2 instructions in scrypt via configure --enable-sse2 (#362)

.travis.yml

@@ -32,13 +32,13 @@ env:
 # 32-bit + dash
     - HOST=i686-pc-linux-gnu PACKAGES="g++-multilib python3-zmq" DEP_OPTS="NO_QT=1" RUN_TESTS=true GOAL="install" BITCOIN_CONFIG="--enable-zmq --enable-glibc-back-compat --enable-reduce-exports LDFLAGS=-static-libstdc++" USE_SHELL="/bin/dash" LITECOIN_SCRYPT=1
 # Win64
-    - HOST=x86_64-w64-mingw32 DPKG_ADD_ARCH="i386" DEP_OPTS="NO_QT=1" PACKAGES="python3 nsis g++-mingw-w64-x86-64 wine1.6" RUN_TESTS=true GOAL="install" BITCOIN_CONFIG="--enable-reduce-exports" LITECOIN_SCRYPT=1
+    - HOST=x86_64-w64-mingw32 DPKG_ADD_ARCH="i386" DEP_OPTS="NO_QT=1" PACKAGES="python3 nsis g++-mingw-w64-x86-64 wine1.6" RUN_TESTS=true GOAL="install" BITCOIN_CONFIG="--enable-reduce-exports --enable-sse2" LITECOIN_SCRYPT=1
 # x86_64 Linux (uses qt5 dev package instead of depends Qt to speed up build and avoid timeout)
-    - HOST=x86_64-unknown-linux-gnu PACKAGES="python3-zmq qtbase5-dev qttools5-dev-tools protobuf-compiler libdbus-1-dev libharfbuzz-dev" DEP_OPTS="NO_QT=1 NO_UPNP=1 DEBUG=1 ALLOW_HOST_PACKAGES=1" RUN_TESTS=true GOAL="install" BITCOIN_CONFIG="--enable-zmq --with-gui=qt5 --enable-glibc-back-compat --enable-reduce-exports CPPFLAGS=-DDEBUG_LOCKORDER" LITECOIN_SCRYPT=1
+    - HOST=x86_64-unknown-linux-gnu PACKAGES="python3-zmq qtbase5-dev qttools5-dev-tools protobuf-compiler libdbus-1-dev libharfbuzz-dev" DEP_OPTS="NO_QT=1 NO_UPNP=1 DEBUG=1 ALLOW_HOST_PACKAGES=1" RUN_TESTS=true GOAL="install" BITCOIN_CONFIG="--enable-zmq --with-gui=qt5 --enable-glibc-back-compat --enable-reduce-exports --enable-sse2 CPPFLAGS=-DDEBUG_LOCKORDER" LITECOIN_SCRYPT=1
 # x86_64 Linux, No wallet
-    - HOST=x86_64-unknown-linux-gnu PACKAGES="python3" DEP_OPTS="NO_WALLET=1" RUN_TESTS=true GOAL="install" BITCOIN_CONFIG="--enable-glibc-back-compat --enable-reduce-exports" LITECOIN_SCRYPT=1
+    - HOST=x86_64-unknown-linux-gnu PACKAGES="python3" DEP_OPTS="NO_WALLET=1" RUN_TESTS=true GOAL="install" BITCOIN_CONFIG="--enable-glibc-back-compat --enable-reduce-exports --enable-sse2" LITECOIN_SCRYPT=1
 # Cross-Mac
-    - HOST=x86_64-apple-darwin11 PACKAGES="cmake imagemagick libcap-dev librsvg2-bin libz-dev libbz2-dev libtiff-tools python-dev" BITCOIN_CONFIG="--enable-gui --enable-reduce-exports --enable-werror" OSX_SDK=10.11 GOAL="deploy"
+    - HOST=x86_64-apple-darwin11 PACKAGES="cmake imagemagick libcap-dev librsvg2-bin libz-dev libbz2-dev libtiff-tools python-dev" BITCOIN_CONFIG="--enable-gui --enable-reduce-exports --enable-werror --enable-sse2" OSX_SDK=10.11 GOAL="deploy"
 
 before_install:
     - export PATH=$(echo $PATH | tr ':' "\n" | sed '/\/opt\/python/d' | tr "\n" ":" | sed "s|::|:|g")
@@ -76,7 +76,7 @@ script:
     - mkdir build && cd build
     - ../configure --cache-file=config.cache $BITCOIN_CONFIG_ALL $BITCOIN_CONFIG || ( cat config.log && false)
     - make distdir VERSION=$HOST
-    - cd bitcoin-$HOST
+    - cd litecoin-$HOST
     - ./configure --cache-file=../config.cache $BITCOIN_CONFIG_ALL $BITCOIN_CONFIG || ( cat config.log && false)
     - make $MAKEJOBS $GOAL || ( echo "Build failure. Verbose build follows." && make $GOAL V=1 ; false )
     - export LD_LIBRARY_PATH=$TRAVIS_BUILD_DIR/depends/$HOST/lib

configure.ac

@@ -199,6 +199,12 @@ AC_ARG_ENABLE([zmq],
   [use_zmq=$enableval],
   [use_zmq=yes])
 
+AC_ARG_ENABLE([sse2],
+    [AS_HELP_STRING([--enable-sse2],
+    [enable SSE2 instructions in the scrypt library. (default is disabled)])],
+    [use_sse2=$enableval],
+    [use_sse2=no])
+
 AC_ARG_WITH([protoc-bindir],[AS_HELP_STRING([--with-protoc-bindir=BIN_DIR],[specify protoc bin path])], [protoc_bin_path=$withval], [])
 
 AC_ARG_ENABLE(man,
@@ -1185,6 +1191,16 @@ else
   BUILD_TEST=""
 fi
 
+AC_MSG_CHECKING([whether to enable sse2 instructions])
+if test x$use_sse2 != xno; then
+  AC_MSG_RESULT(yes)
+  AC_DEFINE([USE_SSE2],[1],[Define if SSE2 support should be compiled in])
+  use_sse2=yes
+  CPPFLAGS="$CPPFLAGS -DUSE_SSE2=1"
+else
+  AC_MSG_RESULT(no)
+fi
+
 AC_MSG_CHECKING([whether to reduce exports])
 if test x$use_reduce_exports = xyes; then
   AC_MSG_RESULT([yes])
@@ -1205,6 +1221,7 @@ AM_CONDITIONAL([ENABLE_QT],[test x$bitcoin_enable_qt = xyes])
 AM_CONDITIONAL([ENABLE_QT_TESTS],[test x$BUILD_TEST_QT = xyes])
 AM_CONDITIONAL([ENABLE_BENCH],[test x$use_bench = xyes])
 AM_CONDITIONAL([USE_QRCODE], [test x$use_qr = xyes])
+AM_CONDITIONAL([USE_SSE2], [test x$use_sse2 = xyes])
 AM_CONDITIONAL([USE_LCOV],[test x$use_lcov = xyes])
 AM_CONDITIONAL([GLIBC_BACK_COMPAT],[test x$use_glibc_compat = xyes])
 AM_CONDITIONAL([HARDEN],[test x$use_hardening = xyes])
@@ -1246,6 +1263,7 @@ AC_SUBST(SSE42_CXXFLAGS)
 AC_SUBST(LIBTOOL_APP_LDFLAGS)
 AC_SUBST(USE_UPNP)
 AC_SUBST(USE_QRCODE)
+AC_SUBST(USE_SSE2)
 AC_SUBST(BOOST_LIBS)
 AC_SUBST(TESTDEFS)
 AC_SUBST(LEVELDB_TARGET_FLAGS)

src/Makefile.am

@@ -270,6 +270,7 @@ crypto_libbitcoin_crypto_a_SOURCES = \
   crypto/ripemd160.cpp \
   crypto/ripemd160.h \
   crypto/scrypt.cpp \
+  crypto/scrypt-sse2.cpp \
   crypto/scrypt.h \
   crypto/sha1.cpp \
   crypto/sha1.h \

src/crypto/scrypt-sse2.cpp

@@ -27,6 +27,8 @@
  * online backup system.
  */
 
+#if defined(USE_SSE2)
+
 #include "crypto/scrypt.h"
 #include <stdlib.h>
 #include <stdint.h>
@@ -134,3 +136,5 @@ void scrypt_1024_1_1_256_sp_sse2(const char *input, char *output, char *scratchp
 
 	PBKDF2_SHA256((const uint8_t *)input, 80, B, 128, 1, (uint8_t *)output, 32);
 }
+
+#endif // USE_SSE2

src/crypto/scrypt.cpp

@@ -291,10 +291,11 @@ void scrypt_1024_1_1_256_sp_generic(const char *input, char *output, char *scrat
 // By default, set to generic scrypt function. This will prevent crash in case when scrypt_detect_sse2() wasn't called
 void (*scrypt_1024_1_1_256_sp_detected)(const char *input, char *output, char *scratchpad) = &scrypt_1024_1_1_256_sp_generic;
 
-void scrypt_detect_sse2()
+std::string scrypt_detect_sse2()
 {
+    std::string ret;
 #if defined(USE_SSE2_ALWAYS)
-    printf("scrypt: using scrypt-sse2 as built.\n");
+    ret = "scrypt: using scrypt-sse2 as built.";
 #else // USE_SSE2_ALWAYS
     // 32bit x86 Linux or Windows, detect cpuid features
     unsigned int cpuid_edx=0;
@@ -312,14 +313,15 @@ void scrypt_detect_sse2()
     if (cpuid_edx & 1<<26)
     {
         scrypt_1024_1_1_256_sp_detected = &scrypt_1024_1_1_256_sp_sse2;
-        printf("scrypt: using scrypt-sse2 as detected.\n");
+        ret = "scrypt: using scrypt-sse2 as detected");
     }
     else
     {
         scrypt_1024_1_1_256_sp_detected = &scrypt_1024_1_1_256_sp_generic;
-        printf("scrypt: using scrypt-generic, SSE2 unavailable.\n");
+        ret = "scrypt: using scrypt-generic, SSE2 unavailable";
     }
 #endif // USE_SSE2_ALWAYS
+    return ret;
 }
 #endif
 

src/crypto/scrypt.h

@@ -9,6 +9,7 @@ void scrypt_1024_1_1_256(const char *input, char *output);
 void scrypt_1024_1_1_256_sp_generic(const char *input, char *output, char *scratchpad);
 
 #if defined(USE_SSE2)
+#include <string>
 #if defined(_M_X64) || defined(__x86_64__) || defined(_M_AMD64) || (defined(MAC_OSX) && defined(__i386__))
 #define USE_SSE2_ALWAYS 1
 #define scrypt_1024_1_1_256_sp(input, output, scratchpad) scrypt_1024_1_1_256_sp_sse2((input), (output), (scratchpad))
@@ -16,7 +17,7 @@ void scrypt_1024_1_1_256_sp_generic(const char *input, char *output, char *scrat
 #define scrypt_1024_1_1_256_sp(input, output, scratchpad) scrypt_1024_1_1_256_sp_detected((input), (output), (scratchpad))
 #endif
 
-void scrypt_detect_sse2();
+std::string scrypt_detect_sse2();
 void scrypt_1024_1_1_256_sp_sse2(const char *input, char *output, char *scratchpad);
 extern void (*scrypt_1024_1_1_256_sp_detected)(const char *input, char *output, char *scratchpad);
 #else

src/init.cpp

@@ -67,6 +67,10 @@
 #include <zmq/zmqnotificationinterface.h>
 #endif
 
+#ifdef USE_SSE2
+#include "crypto/scrypt.h"
+#endif
+
 bool fFeeEstimatesInitialized = false;
 static const bool DEFAULT_PROXYRANDOMIZE = true;
 static const bool DEFAULT_REST_ENABLE = false;
@@ -1274,7 +1278,8 @@ bool AppInitMain()
     int64_t nStart;
 
 #if defined(USE_SSE2)
-    scrypt_detect_sse2();
+    std::string sse2detect = scrypt_detect_sse2();
+    LogPrintf("%s\n", sse2detect);
 #endif
 
     // ********************************************************* Step 5: verify wallet database integrity

src/test/scrypt_tests.cpp

@@ -14,7 +14,7 @@ BOOST_AUTO_TEST_CASE(scrypt_hashtest)
     const char* inputhex[HASHCOUNT] = { "020000004c1271c211717198227392b029a64a7971931d351b387bb80db027f270411e398a07046f7d4a08dd815412a8712f874a7ebf0507e3878bd24e20a3b73fd750a667d2f451eac7471b00de6659", "0200000011503ee6a855e900c00cfdd98f5f55fffeaee9b6bf55bea9b852d9de2ce35828e204eef76acfd36949ae56d1fbe81c1ac9c0209e6331ad56414f9072506a77f8c6faf551eac7471b00389d01", "02000000a72c8a177f523946f42f22c3e86b8023221b4105e8007e59e81f6beb013e29aaf635295cb9ac966213fb56e046dc71df5b3f7f67ceaeab24038e743f883aff1aaafaf551eac7471b0166249b", "010000007824bc3a8a1b4628485eee3024abd8626721f7f870f8ad4d2f33a27155167f6a4009d1285049603888fe85a84b6c803a53305a8d497965a5e896e1a00568359589faf551eac7471b0065434e", "0200000050bfd4e4a307a8cb6ef4aef69abc5c0f2d579648bd80d7733e1ccc3fbc90ed664a7f74006cb11bde87785f229ecd366c2d4e44432832580e0608c579e4cb76f383f7f551eac7471b00c36982" };
     const char* expected[HASHCOUNT] = { "00000000002bef4107f882f6115e0b01f348d21195dacd3582aa2dabd7985806" , "00000000003a0d11bdd5eb634e08b7feddcfbbf228ed35d250daf19f1c88fc94", "00000000000b40f895f288e13244728a6c2d9d59d8aff29c65f8dd5114a8ca81", "00000000003007005891cd4923031e99d8e8d72f6e8e7edc6a86181897e105fe", "000000000018f0b426a4afc7130ccb47fa02af730d345b4fe7c7724d3800ec8c" };
 #if defined(USE_SSE2)
-    scrypt_detect_sse2();
+    (void) scrypt_detect_sse2();
 #endif
     uint256 scrypthash;
     std::vector<unsigned char> inputbytes;