Merge pull request #1549 from evoskuil/master

Mild hash library refactoring.

Makefile.am

@@ -306,7 +306,7 @@ test_libbitcoin_system_test_SOURCES = \
     test/hash/sha/sha160.cpp \
     test/hash/sha/sha256.cpp \
     test/hash/sha/sha512.cpp \
-    test/hash/sha/vectorization.cpp \
+    test/hash/sha/vector.cpp \
     test/intrinsics/haves.cpp \
     test/intrinsics/intrinsics.cpp \
     test/intrinsics/xcpu/cpuid.cpp \
@@ -619,8 +619,8 @@ include_bitcoin_system_impl_hash_rmd_HEADERS = \
 include_bitcoin_system_impl_hash_shadir = ${includedir}/bitcoin/system/impl/hash/sha
 include_bitcoin_system_impl_hash_sha_HEADERS = \
     include/bitcoin/system/impl/hash/sha/algorithm.ipp \
-    include/bitcoin/system/impl/hash/sha/algorithm_compression.ipp \
-    include/bitcoin/system/impl/hash/sha/algorithm_vectorization.ipp
+    include/bitcoin/system/impl/hash/sha/algorithm_native.ipp \
+    include/bitcoin/system/impl/hash/sha/algorithm_vector.ipp
 
 include_bitcoin_system_impl_machinedir = ${includedir}/bitcoin/system/impl/machine
 include_bitcoin_system_impl_machine_HEADERS = \

builds/cmake/CMakeLists.txt

@@ -793,7 +793,7 @@ if (with-tests)
         "../../test/hash/sha/sha160.cpp"
         "../../test/hash/sha/sha256.cpp"
         "../../test/hash/sha/sha512.cpp"
-        "../../test/hash/sha/vectorization.cpp"
+        "../../test/hash/sha/vector.cpp"
         "../../test/intrinsics/haves.cpp"
         "../../test/intrinsics/intrinsics.cpp"
         "../../test/intrinsics/xcpu/cpuid.cpp"

builds/msvc/vs2022/libbitcoin-system-test/libbitcoin-system-test.vcxproj

@@ -174,7 +174,7 @@
       <ObjectFileName>$(IntDir)test_hash_sha_sha256.obj</ObjectFileName>
     </ClCompile>
     <ClCompile Include="..\..\..\..\test\hash\sha\sha512.cpp" />
-    <ClCompile Include="..\..\..\..\test\hash\sha\vectorization.cpp" />
+    <ClCompile Include="..\..\..\..\test\hash\sha\vector.cpp" />
     <ClCompile Include="..\..\..\..\test\hash\siphash.cpp" />
     <ClCompile Include="..\..\..\..\test\intrinsics\haves.cpp" />
     <ClCompile Include="..\..\..\..\test\intrinsics\intrinsics.cpp" />

builds/msvc/vs2022/libbitcoin-system-test/libbitcoin-system-test.vcxproj.filters

@@ -354,7 +354,7 @@
     <ClCompile Include="..\..\..\..\test\hash\sha\sha512.cpp">
       <Filter>src\hash\sha</Filter>
     </ClCompile>
-    <ClCompile Include="..\..\..\..\test\hash\sha\vectorization.cpp">
+    <ClCompile Include="..\..\..\..\test\hash\sha\vector.cpp">
       <Filter>src\hash\sha</Filter>
     </ClCompile>
     <ClCompile Include="..\..\..\..\test\hash\siphash.cpp">

builds/msvc/vs2022/libbitcoin-system/libbitcoin-system.vcxproj

@@ -545,8 +545,8 @@
     <None Include="..\..\..\..\include\bitcoin\system\impl\hash\rmd\algorithm.ipp" />
     <None Include="..\..\..\..\include\bitcoin\system\impl\hash\scrypt.ipp" />
     <None Include="..\..\..\..\include\bitcoin\system\impl\hash\sha\algorithm.ipp" />
-    <None Include="..\..\..\..\include\bitcoin\system\impl\hash\sha\algorithm_compression.ipp" />
-    <None Include="..\..\..\..\include\bitcoin\system\impl\hash\sha\algorithm_vectorization.ipp" />
+    <None Include="..\..\..\..\include\bitcoin\system\impl\hash\sha\algorithm_native.ipp" />
+    <None Include="..\..\..\..\include\bitcoin\system\impl\hash\sha\algorithm_vector.ipp" />
     <None Include="..\..\..\..\include\bitcoin\system\impl\machine\interpreter.ipp" />
     <None Include="..\..\..\..\include\bitcoin\system\impl\machine\number.ipp" />
     <None Include="..\..\..\..\include\bitcoin\system\impl\machine\program.ipp" />

builds/msvc/vs2022/libbitcoin-system/libbitcoin-system.vcxproj.filters

@@ -1510,10 +1510,10 @@
     <None Include="..\..\..\..\include\bitcoin\system\impl\hash\sha\algorithm.ipp">
       <Filter>include\bitcoin\system\impl\hash\sha</Filter>
     </None>
-    <None Include="..\..\..\..\include\bitcoin\system\impl\hash\sha\algorithm_compression.ipp">
+    <None Include="..\..\..\..\include\bitcoin\system\impl\hash\sha\algorithm_native.ipp">
       <Filter>include\bitcoin\system\impl\hash\sha</Filter>
     </None>
-    <None Include="..\..\..\..\include\bitcoin\system\impl\hash\sha\algorithm_vectorization.ipp">
+    <None Include="..\..\..\..\include\bitcoin\system\impl\hash\sha\algorithm_vector.ipp">
       <Filter>include\bitcoin\system\impl\hash\sha</Filter>
     </None>
     <None Include="..\..\..\..\include\bitcoin\system\impl\machine\interpreter.ipp">

include/bitcoin/system/constraints.hpp

@@ -167,6 +167,10 @@ template <typename Type>
 using if_integer = bool_if<
     is_integer<Type>>;
 
+template <typename Type>
+using if_non_integer = bool_if<
+    !is_integer<Type>>;
+
 template <typename Type>
 using if_signed_integer = bool_if<
     is_integer<Type> &&
@@ -198,15 +202,25 @@ using if_not_uintx = bool_if<
     !is_uintx<Type>>;
 
 /// Integral integer types (native, non-floating point, non-bool).
+/// All integrals are integers but not all integers are integral.
+
+template <typename Type>
+using if_integral = bool_if<
+    is_integral<Type>>;
+
+template <typename Type>
+using if_non_integral = bool_if<
+    !is_integral<Type>>;
 
+// Alias for if_integral.
 template <typename Type>
 using if_integral_integer = bool_if<
     is_integral_integer<Type>>;
 
 template <typename Type>
 using if_non_integral_integer = bool_if<
-    is_integer<Type> &&
-    !is_integral<Type>>;
+    !is_integral<Type>  &&
+    is_integer<Type>>;
 
 template <typename Type>
 using if_signed_integral_integer = bool_if<

include/bitcoin/system/hash/sha/algorithm.hpp

@@ -37,10 +37,10 @@ namespace system {
 namespace sha {
 
 /// SHA hashing algorithm.
-/// Compression not yet implemented.
+/// Native not yet implemented.
 /// Vectorization of message schedules and merkle hashes.
-template <typename SHA, bool Compressed = true, bool Vectorized = true,
-    bool Cached = true, if_same<typename SHA::T, sha::shah_t> = true>
+template <typename SHA, bool Native = true, bool Vector = true, bool Cached = true,
+    if_same<typename SHA::T, sha::shah_t> = true>
 class algorithm
   : algorithm_t
 {
@@ -163,13 +163,14 @@ class algorithm
     INLINE static constexpr void schedule_(auto& buffer) NOEXCEPT;
     static constexpr void schedule(auto& buffer) NOEXCEPT;
 
-    INLINE static constexpr void input(auto& buffer, const auto& state) NOEXCEPT;
+    /// A double hash optimization (hashing the output of the first hash).
+    INLINE static constexpr void reinput(auto& buffer, const auto& state) NOEXCEPT;
 
-    /// Parsing
+    /// Parsing (endian sensitive)
     /// -----------------------------------------------------------------------
     INLINE static constexpr void input(buffer_t& buffer, const block_t& block) NOEXCEPT;
-    INLINE static constexpr void input1(buffer_t& buffer, const half_t& half) NOEXCEPT;
-    INLINE static constexpr void input2(buffer_t& buffer, const half_t& half) NOEXCEPT;
+    INLINE static constexpr void input_left(buffer_t& buffer, const half_t& half) NOEXCEPT;
+    INLINE static constexpr void input_right(buffer_t& buffer, const half_t& half) NOEXCEPT;
     INLINE static constexpr digest_t output(const state_t& state) NOEXCEPT;
 
     /// Padding
@@ -197,7 +198,7 @@ class algorithm
     /// Merkle iteration.
     /// -----------------------------------------------------------------------
     VCONSTEXPR static void merkle_hash_(digests_t& digests,
-        size_t offset = zero) NOEXCEPT;
+        size_t offset=zero) NOEXCEPT;
 
 private:
     using pad_t = std_array<word_t, subtract(SHA::block_words,
@@ -208,24 +209,16 @@ class algorithm
     static CONSTEVAL chunk_t chunk_pad() NOEXCEPT;
     static CONSTEVAL pad_t stream_pad() NOEXCEPT;
 
-/// Compression.
-/// -----------------------------------------------------------------------
-protected:
-public:
-    static constexpr auto have_shani = Compressed && system::with_shani;
-    static constexpr auto have_neon = Compressed && system::with_neon;
-    static constexpr auto compression = have_shani || have_neon;
-
 /// Vectorization.
-/// -----------------------------------------------------------------------
+/// ---------------------------------------------------------------------------
 protected:
     /// Extended integer capacity for uint32_t/uint64_t is 2/4/8/16 only.
     template <size_t Lanes>
     static constexpr auto is_valid_lanes =
         (Lanes == 16u || Lanes == 8u || Lanes == 4u || Lanes == 2u);
 
     template <size_t Lanes, bool_if<is_valid_lanes<Lanes>> = true>
-    using wblock_t = std_array<words_t, Lanes>;
+    using xblock_t = std_array<words_t, Lanes>;
     template <typename xWord, if_extended<xWord> = true>
     using xbuffer_t = std_array<xWord, SHA::rounds>;
     template <typename xWord, if_extended<xWord> = true>
@@ -238,7 +231,7 @@ class algorithm
     /// -----------------------------------------------------------------------
 
     template <size_t Word, size_t Lanes>
-    INLINE static auto pack(const wblock_t<Lanes>& wblock) NOEXCEPT;
+    INLINE static auto pack(const xblock_t<Lanes>& xblock) NOEXCEPT;
 
     template <typename xWord>
     INLINE static void input(xbuffer_t<xWord>& xbuffer,
@@ -286,7 +279,7 @@ class algorithm
     INLINE static Word extract(xWord a) NOEXCEPT;
 
     template <typename xWord>
-    INLINE static void compress_dispatch(state_t& state,
+    INLINE static void compress_invoke(state_t& state,
         const xbuffer_t<xWord>& xbuffer) NOEXCEPT;
 
     template <typename xWord, if_extended<xWord> = true>
@@ -305,34 +298,56 @@ class algorithm
     INLINE static auto sigma0_8(auto x1, auto x2, auto x3, auto x4, auto x5,
         auto x6, auto x7, auto x8) NOEXCEPT;
 
+    template<size_t Round, size_t Offset>
+    INLINE static void prepare1(buffer_t& buffer, const auto& xsigma0) NOEXCEPT;
     template<size_t Round>
-    INLINE static void prepare_dispatch(buffer_t& buffer) NOEXCEPT;
-    INLINE static void schedule_invoke(buffer_t& buffer) NOEXCEPT;
-    INLINE static void schedule_dispatch(auto& buffer) NOEXCEPT;
+    INLINE static void prepare8(buffer_t& buffer) NOEXCEPT;
+
+    template <typename xWord>
+    INLINE static void schedule_vector(xbuffer_t<xWord>& xbuffer) NOEXCEPT;
+    INLINE static void schedule_vector(buffer_t& buffer) NOEXCEPT;
+
+    /// Native.
+    /// ---------------------------------------------------------------------------
+protected:
+    using cword_t = xint128_t;
+    static constexpr auto cratio = sizeof(cword_t) / SHA::word_bytes;
+    static constexpr auto crounds = SHA::rounds / cratio;
+    using cbuffer_t = std_array<cword_t, crounds>;
+    using cstate_t = std_array<xint128_t, two>;
+
+    template <typename xWord>
+    INLINE static void schedule_native(xbuffer_t<xWord>& xbuffer) NOEXCEPT;
+    INLINE static void schedule_native(buffer_t& buffer) NOEXCEPT;
 
 public:
-    static constexpr auto have_x128     = Vectorized && system::with_sse41;
-    static constexpr auto have_x256     = Vectorized && system::with_avx2;
-    static constexpr auto have_x512     = Vectorized && system::with_avx512;
-    static constexpr auto vectorization = (have_x128 || have_x256 || have_x512)
+    static constexpr auto use_neon = Native && system::with_neon;
+    static constexpr auto use_shani = Native && system::with_shani;
+    static constexpr auto native = use_shani || use_neon;
+
+    static constexpr auto use_x128 = Vector && system::with_sse41;
+    static constexpr auto use_x256 = Vector && system::with_avx2;
+    static constexpr auto use_x512 = Vector && system::with_avx512;
+    static constexpr auto vector = (use_x128 || use_x256 || use_x512)
         && !(build_x32 && is_same_size<word_t, uint64_t>);
+
     static constexpr auto min_lanes =
-        (have_x128 ? bytes<128> :
-            (have_x256 ? bytes<256> :
-                (have_x512 ? bytes<512> : 0))) / SHA::word_bytes;
+        (use_x128 ? bytes<128> :
+            (use_x256 ? bytes<256> :
+                (use_x512 ? bytes<512> : 0))) / SHA::word_bytes;
 };
 
 } // namespace sha
 } // namespace system
 } // namespace libbitcoin
 
-#define TEMPLATE template <typename SHA, bool Compressed, bool Vectorized, \
-    bool Cached, if_same<typename SHA::T, sha::shah_t> If>
-#define CLASS algorithm<SHA, Compressed, Vectorized, Cached, If>
+#define TEMPLATE template <typename SHA, bool Native, bool Vector, bool Cached, \
+    if_same<typename SHA::T, sha::shah_t> If>
+#define CLASS algorithm<SHA, Native, Vector, Cached, If>
 
 #include <bitcoin/system/impl/hash/sha/algorithm.ipp>
-#include <bitcoin/system/impl/hash/sha/algorithm_compression.ipp>
-#include <bitcoin/system/impl/hash/sha/algorithm_vectorization.ipp>
+#include <bitcoin/system/impl/hash/sha/algorithm_native.ipp>
+#include <bitcoin/system/impl/hash/sha/algorithm_vector.ipp>
 
 #undef CLASS
 #undef TEMPLATE

include/bitcoin/system/have.hpp

@@ -128,21 +128,22 @@
 #endif
 
 /// XCPU architecture intrinsics sse41, avx2, avx512f, sha-ni.
-/// All require runtime evaluation, as the binary is portable across XCPUs.
+/// This assumes that avx512 implies avx2 and that all imply sse41.
+/// All require runtime evaluation for the binary is portable across XCPUs.
 #if defined(HAVE_XCPU)
     // TODO: CLANG/GCC compile test and set -msse4 -mavx2 -mavx512f -msha.
-    #if defined (WITH_AVX512)
+    #if defined(WITH_SHANI)
+        #define HAVE_SHANI
+    #endif
+    #if defined(WITH_AVX512)
         #define HAVE_AVX512
     #endif
-    #if defined (WITH_AVX2)
+    #if defined(WITH_AVX2) || defined(WITH_AVX512)
         #define HAVE_AVX2
     #endif
-    #if defined (WITH_SSE41)
+    #if defined(WITH_SSE41) || defined(WITH_AVX2) || defined(WITH_SHANI)
         #define HAVE_SSE41
     #endif
-    #if defined (WITH_SHANI)
-        #define HAVE_SHANI
-    #endif
 #endif
 
 /// MSC predefined constant for Visual Studio version (exclusive).