Add xxh3 hash function (#43411)

* impl

* try fix

* add docs

* add test

* rm unused file

* excellent

.gitmodules

@@ -290,3 +290,6 @@
 [submodule "contrib/morton-nd"]
 	path = contrib/morton-nd
 	url = https://github.com/morton-nd/morton-nd
+[submodule "contrib/xxHash"]
+	path = contrib/xxHash
+	url = https://github.com/Cyan4973/xxHash.git

contrib/CMakeLists.txt

@@ -167,7 +167,9 @@ add_contrib (c-ares-cmake c-ares)
 add_contrib (qpl-cmake qpl)
 add_contrib (morton-nd-cmake morton-nd)
 
-add_contrib(annoy-cmake annoy)
+add_contrib (annoy-cmake annoy)
+
+add_contrib (xxHash-cmake xxHash)
 
 # Put all targets defined here and in subdirectories under "contrib/<immediate-subdir>" folders in GUI-based IDEs.
 # Some of third-party projects may override CMAKE_FOLDER or FOLDER property of their targets, so they would not appear

contrib/xxHash

@@ -0,0 +1 @@
+Subproject commit 3078dc6039f8c0bffcb1904f81cfe6b2c3209435

contrib/xxHash-cmake/CMakeLists.txt

@@ -0,0 +1,13 @@
+set (LIBRARY_DIR "${ClickHouse_SOURCE_DIR}/contrib/xxHash")
+set (SRCS
+    "${LIBRARY_DIR}/xxhash.c"
+)
+
+add_library(xxHash ${SRCS})
+target_include_directories(xxHash SYSTEM BEFORE INTERFACE "${LIBRARY_DIR}")
+
+# XXH_INLINE_ALL - Make all functions inline, with implementations being directly included within xxhash.h. Inlining functions is beneficial for speed on small keys.
+# https://github.com/Cyan4973/xxHash/tree/v0.8.1#build-modifiers
+target_compile_definitions(xxHash PUBLIC XXH_INLINE_ALL)
+
+add_library(ch_contrib::xxHash ALIAS xxHash)

docker/test/fasttest/run.sh

@@ -137,6 +137,7 @@ function clone_submodules
             contrib/hashidsxx
             contrib/c-ares
             contrib/morton-nd
+            contrib/xxHash
         )
 
         git submodule sync

src/Functions/CMakeLists.txt

@@ -29,9 +29,9 @@ list (APPEND PRIVATE_LIBS
         ch_contrib::zlib
         boost::filesystem
         divide_impl
+        ch_contrib::xxHash
 )
 
-
 if (TARGET ch_rust::blake3)
     list (APPEND PUBLIC_LIBS
         ch_rust::blake3
@@ -66,8 +66,6 @@ if (TARGET ch_contrib::base64)
     list (APPEND PRIVATE_LIBS ch_contrib::base64)
 endif()
 
-list (APPEND PRIVATE_LIBS ch_contrib::lz4)
-
 if (ENABLE_NLP)
     list (APPEND PRIVATE_LIBS ch_contrib::cld2)
 endif()

src/Functions/FunctionsHashing.cpp

@@ -39,6 +39,13 @@ REGISTER_FUNCTION(Hashing)
 
     factory.registerFunction<FunctionXxHash32>();
     factory.registerFunction<FunctionXxHash64>();
+    factory.registerFunction<FunctionXXH3>(
+        {
+            "Calculates value of XXH3 64-bit hash function. Refer to https://github.com/Cyan4973/xxHash for detailed documentation.",
+            Documentation::Examples{{"hash", "SELECT xxh3('ClickHouse')"}},
+            Documentation::Categories{"Hash"}
+        },
+        FunctionFactory::CaseSensitive);
 
     factory.registerFunction<FunctionWyHash64>();
 

src/Functions/FunctionsHashing.h

@@ -3,12 +3,18 @@
 #include <city.h>
 #include <farmhash.h>
 #include <metrohash.h>
+#include <wyhash.h>
 #include <MurmurHash2.h>
 #include <MurmurHash3.h>
-#include <wyhash.h>
 
 #include "config.h"
 
+#ifdef __clang__
+#    pragma clang diagnostic push
+#    pragma clang diagnostic ignored "-Wused-but-marked-unused"
+#endif
+#include <xxhash.h>
+
 #if USE_BLAKE3
 #    include <blake3.h>
 #endif
@@ -17,7 +23,6 @@
 #include <Common/typeid_cast.h>
 #include <Common/safe_cast.h>
 #include <Common/HashTable/Hash.h>
-#include <xxhash.h>
 
 #if USE_SSL
 #    include <openssl/md4.h>
@@ -588,7 +593,7 @@ struct ImplXxHash32
     static constexpr auto name = "xxHash32";
     using ReturnType = UInt32;
 
-    static auto apply(const char * s, const size_t len) { return XXH32(s, len, 0); }
+    static auto apply(const char * s, const size_t len) { return XXH_INLINE_XXH32(s, len, 0); }
     /**
       *  With current implementation with more than 1 arguments it will give the results
       *  non-reproducible from outside of CH.
@@ -609,7 +614,24 @@ struct ImplXxHash64
     using ReturnType = UInt64;
     using uint128_t = CityHash_v1_0_2::uint128;
 
-    static auto apply(const char * s, const size_t len) { return XXH64(s, len, 0); }
+    static auto apply(const char * s, const size_t len) { return XXH_INLINE_XXH64(s, len, 0); }
+
+    /*
+       With current implementation with more than 1 arguments it will give the results
+       non-reproducible from outside of CH. (see comment on ImplXxHash32).
+     */
+    static auto combineHashes(UInt64 h1, UInt64 h2) { return CityHash_v1_0_2::Hash128to64(uint128_t(h1, h2)); }
+
+    static constexpr bool use_int_hash_for_pods = false;
+};
+
+struct ImplXXH3
+{
+    static constexpr auto name = "xxh3";
+    using ReturnType = UInt64;
+    using uint128_t = CityHash_v1_0_2::uint128;
+
+    static auto apply(const char * s, const size_t len) { return XXH_INLINE_XXH3_64bits(s, len); }
 
     /*
        With current implementation with more than 1 arguments it will give the results
@@ -1508,7 +1530,12 @@ using FunctionHiveHash = FunctionAnyHash<HiveHashImpl>;
 
 using FunctionXxHash32 = FunctionAnyHash<ImplXxHash32>;
 using FunctionXxHash64 = FunctionAnyHash<ImplXxHash64>;
+using FunctionXXH3 = FunctionAnyHash<ImplXXH3>;
 
 using FunctionWyHash64 = FunctionAnyHash<ImplWyHash64>;
 using FunctionBLAKE3 = FunctionStringHashFixedString<ImplBLAKE3>;
 }
+
+#ifdef __clang__
+#    pragma clang diagnostic pop
+#endif

tests/performance/general_purpose_hashes.xml

@@ -15,6 +15,7 @@
                <value>hiveHash</value>
                <value>xxHash32</value>
                <value>xxHash64</value>
+               <value>xxh3</value>
                <value>CRC32</value>
            </values>
         </substitution>

tests/queries/0_stateless/02481_xxh3_hash_function.reference

@@ -0,0 +1 @@
+18009318874338624809

tests/queries/0_stateless/02481_xxh3_hash_function.sql

@@ -0,0 +1 @@
+SELECT xxh3('ClickHouse');