Merge #268: cmake: Port PR28893 from the master branch

07e7fdc fixup! cmake: Build `bitcoin_crypto` library (Hennadii Stepanov)
6e9c0c6 crypto: Guard code with `ENABLE_SSE41` macro (Hennadii Stepanov)
7db01e1 build: Fix test for SSE4.1 intrinsics (Hennadii Stepanov)

Pull request description:

  This PR ports bitcoin#28893. All commits from there are picked here to avoid conflicts and CI failures.

ACKs for top commit:
  theuni:
    utACK 07e7fdc

Tree-SHA512: 58b5167d644e47f4e5d499f21a11acc66f9c8c2d23ee76ec1886d701d4f34965069af25a0a86d5c40ccbc3e064346bbe68f1e7451e7653419ad68a13ccc1492b

configure.ac

@@ -482,11 +482,12 @@ TEMP_CXXFLAGS="$CXXFLAGS"
 CXXFLAGS="$SSE41_CXXFLAGS $CXXFLAGS"
 AC_MSG_CHECKING([for SSE4.1 intrinsics])
 AC_COMPILE_IFELSE([AC_LANG_PROGRAM([[
-    #include <stdint.h>
     #include <immintrin.h>
   ]],[[
-    __m128i l = _mm_set1_epi32(0);
-    return _mm_extract_epi32(l, 3);
+    __m128i a = _mm_set1_epi32(0);
+    __m128i b = _mm_set1_epi32(1);
+    __m128i r = _mm_blend_epi16(a, b, 0xFF);
+    return _mm_extract_epi32(r, 3);
   ]])],
  [ AC_MSG_RESULT([yes]); enable_sse41=yes; AC_DEFINE([ENABLE_SSE41], [1], [Define this symbol to build code that uses SSE4.1 intrinsics]) ],
  [ AC_MSG_RESULT([no])]

src/Makefile.am

@@ -51,15 +51,15 @@ LIBBITCOIN_CRYPTO = $(LIBBITCOIN_CRYPTO_BASE)
 if ENABLE_SSE41
 LIBBITCOIN_CRYPTO_SSE41 = crypto/libbitcoin_crypto_sse41.la
 LIBBITCOIN_CRYPTO += $(LIBBITCOIN_CRYPTO_SSE41)
+if ENABLE_X86_SHANI
+LIBBITCOIN_CRYPTO_X86_SHANI = crypto/libbitcoin_crypto_x86_shani.la
+LIBBITCOIN_CRYPTO += $(LIBBITCOIN_CRYPTO_X86_SHANI)
+endif
 endif
 if ENABLE_AVX2
 LIBBITCOIN_CRYPTO_AVX2 = crypto/libbitcoin_crypto_avx2.la
 LIBBITCOIN_CRYPTO += $(LIBBITCOIN_CRYPTO_AVX2)
 endif
-if ENABLE_X86_SHANI
-LIBBITCOIN_CRYPTO_X86_SHANI = crypto/libbitcoin_crypto_x86_shani.la
-LIBBITCOIN_CRYPTO += $(LIBBITCOIN_CRYPTO_X86_SHANI)
-endif
 if ENABLE_ARM_SHANI
 LIBBITCOIN_CRYPTO_ARM_SHANI = crypto/libbitcoin_crypto_arm_shani.la
 LIBBITCOIN_CRYPTO += $(LIBBITCOIN_CRYPTO_ARM_SHANI)
@@ -622,7 +622,7 @@ crypto_libbitcoin_crypto_x86_shani_la_LDFLAGS = $(AM_LDFLAGS) -static
 crypto_libbitcoin_crypto_x86_shani_la_CXXFLAGS = $(AM_CXXFLAGS) $(PIE_FLAGS) -static
 crypto_libbitcoin_crypto_x86_shani_la_CPPFLAGS = $(AM_CPPFLAGS)
 crypto_libbitcoin_crypto_x86_shani_la_CXXFLAGS += $(X86_SHANI_CXXFLAGS)
-crypto_libbitcoin_crypto_x86_shani_la_CPPFLAGS += -DENABLE_X86_SHANI
+crypto_libbitcoin_crypto_x86_shani_la_CPPFLAGS += -DENABLE_SSE41 -DENABLE_X86_SHANI
 crypto_libbitcoin_crypto_x86_shani_la_SOURCES = crypto/sha256_x86_shani.cpp
 
 # See explanation for -static in crypto_libbitcoin_crypto_base_la's LDFLAGS and

src/crypto/CMakeLists.txt

@@ -12,8 +12,10 @@ if(NOT MSVC)
 
     int main()
     {
-      __m128i l = _mm_set1_epi32(0);
-      return _mm_extract_epi32(l, 3);
+      __m128i a = _mm_set1_epi32(0);
+      __m128i b = _mm_set1_epi32(1);
+      __m128i r = _mm_blend_epi16(a, b, 0xFF);
+      return _mm_extract_epi32(r, 3);
     }
     " HAVE_SSE41
   )
@@ -108,11 +110,11 @@ if(HAVE_AVX2)
   target_link_libraries(bitcoin_crypto PRIVATE bitcoin_crypto_avx2)
 endif()
 
-if(HAVE_X86_SHANI)
+if(HAVE_SSE41 AND HAVE_X86_SHANI)
   add_library(bitcoin_crypto_x86_shani STATIC EXCLUDE_FROM_ALL
     sha256_x86_shani.cpp
   )
-  target_compile_definitions(bitcoin_crypto_x86_shani PUBLIC ENABLE_X86_SHANI)
+  target_compile_definitions(bitcoin_crypto_x86_shani PUBLIC ENABLE_SSE41 ENABLE_X86_SHANI)
   target_compile_options(bitcoin_crypto_x86_shani PRIVATE ${X86_SHANI_CXXFLAGS})
   target_link_libraries(bitcoin_crypto_x86_shani PRIVATE core_interface)
   target_link_libraries(bitcoin_crypto PRIVATE bitcoin_crypto_x86_shani)

src/crypto/sha256.cpp

@@ -619,7 +619,7 @@ std::string SHA256AutoDetect(sha256_implementation::UseImplementation use_implem
         }
     }
 
-#if defined(ENABLE_X86_SHANI)
+#if defined(ENABLE_SSE41) && defined(ENABLE_X86_SHANI)
     if (have_x86_shani) {
         Transform = sha256_x86_shani::Transform;
         TransformD64 = TransformD64Wrapper<sha256_x86_shani::Transform>;

src/crypto/sha256_x86_shani.cpp

@@ -6,7 +6,7 @@
 // Written and placed in public domain by Jeffrey Walton.
 // Based on code from Intel, and by Sean Gulley for the miTLS project.
 
-#ifdef ENABLE_X86_SHANI
+#if defined(ENABLE_SSE41) && defined(ENABLE_X86_SHANI)
 
 #include <stdint.h>
 #include <immintrin.h>