FastFloat library integration

.gitmodules

@@ -209,3 +209,6 @@
 [submodule "contrib/dragonbox"]
 	path = contrib/dragonbox
 	url = https://github.com/ClickHouse-Extras/dragonbox.git
+[submodule "contrib/fast_float"]
+	path = contrib/fast_float
+	url = https://github.com/lemire/fast_float

CMakeLists.txt

@@ -457,6 +457,7 @@ include (cmake/find/s3.cmake)
 include (cmake/find/base64.cmake)
 include (cmake/find/parquet.cmake)
 include (cmake/find/simdjson.cmake)
+include (cmake/find/fast_float.cmake)
 include (cmake/find/rapidjson.cmake)
 include (cmake/find/fastops.cmake)
 include (cmake/find/odbc.cmake)

cmake/find/fast_float.cmake

@@ -0,0 +1,15 @@
+option (USE_FAST_FLOAT "Use fast_float" ${ENABLE_LIBRARIES})
+
+if (NOT EXISTS "${ClickHouse_SOURCE_DIR}/contrib/fast_float/include/fast_float/fast_float.h")
+    message (WARNING "submodule contrib/fast_float is missing. to fix try run: \n git submodule update --init --recursive")
+    if (USE_FAST_FLOAT)
+        message (${RECONFIGURE_MESSAGE_LEVEL} "Can't find internal fast_float library")
+    endif()
+    return()
+endif ()
+
+if (USE_FAST_FLOAT)
+    set(FAST_FLOAT_INCLUDE_DIR "${ClickHouse_SOURCE_DIR}/contrib/fast_float/include/")
+endif ()
+
+message(STATUS "Using fast_float=${USE_FAST_FLOAT}")

contrib/CMakeLists.txt

@@ -286,6 +286,10 @@ if (USE_SIMDJSON)
     add_subdirectory (simdjson-cmake)
 endif()
 
+if (USE_FAST_FLOAT)
+    add_subdirectory (fast_float)
+endif()
+
 if (USE_FASTOPS)
     add_subdirectory (fastops-cmake)
 endif()

src/CMakeLists.txt

@@ -403,6 +403,11 @@ if (USE_MSGPACK)
     target_include_directories (clickhouse_common_io SYSTEM BEFORE PUBLIC ${MSGPACK_INCLUDE_DIR})
 endif()
 
+if (USE_FAST_FLOAT)
+    target_link_libraries (clickhouse_common_io PRIVATE fast_float)
+    target_include_directories (clickhouse_common_io SYSTEM BEFORE PRIVATE ${FAST_FLOAT_INCLUDE_DIR})
+endif()
+
 if (USE_ORC)
     dbms_target_link_libraries(PUBLIC ${ORC_LIBRARIES})
     dbms_target_include_directories(SYSTEM BEFORE PUBLIC ${ORC_INCLUDE_DIR} ${CMAKE_BINARY_DIR}/contrib/orc/c++/include)

src/Common/config.h.in

@@ -14,3 +14,4 @@
 #cmakedefine01 USE_GRPC
 #cmakedefine01 USE_STATS
 #cmakedefine01 CLICKHOUSE_SPLIT_BINARY
+#cmakedefine01 USE_FAST_FLOAT

src/IO/tests/CMakeLists.txt

@@ -7,6 +7,11 @@ target_link_libraries (read_buffer_perf PRIVATE clickhouse_common_io)
 add_executable (read_float_perf read_float_perf.cpp)
 target_link_libraries (read_float_perf PRIVATE clickhouse_common_io)
 
+if (USE_FAST_FLOAT)
+    target_link_libraries (read_float_perf PRIVATE fast_float)
+    target_include_directories (read_float_perf SYSTEM BEFORE PRIVATE ${FAST_FLOAT_INCLUDE_DIR})
+endif()
+
 add_executable (write_buffer write_buffer.cpp)
 target_link_libraries (write_buffer PRIVATE clickhouse_common_io)
 

src/IO/tests/read_float_perf.cpp

@@ -13,6 +13,9 @@
 #include <IO/WriteBufferFromFileDescriptor.h>
 #include <Compression/CompressedReadBuffer.h>
 
+#if !defined(ARCADIA_BUILD)
+#    include <Common/config.h>
+#endif
 
 /** How to test:
 
@@ -39,6 +42,52 @@ $ for i in {1..10}; do echo $i; time ./read_float_perf 2 < numbers$i.tsv; done
 
 using namespace DB;
 
+#ifdef USE_FAST_FLOAT
+#include <fast_float/fast_float.h>
+
+template <typename T, typename ReturnType>
+ReturnType readFloatTextFastFloatImpl(T & x, ReadBuffer & in)
+{
+    static_assert(std::is_same_v<T, double> || std::is_same_v<T, float>, "Argument for readFloatTextImpl must be float or double");
+    static_assert('a' > '.' && 'A' > '.' && '\n' < '.' && '\t' < '.' && '\'' < '.' && '"' < '.', "Layout of char is not like ASCII"); //-V590
+
+    static constexpr bool throw_exception = std::is_same_v<ReturnType, void>;
+
+    String buff;
+    
+    /// TODO: Optimize 
+    /// Currently fast_float interface need begin and end
+    /// ReadBuffers current begin end can have only part of data
+    while (!in.eof() && isAlphaNumericASCII(*in.position()))  {
+        buff += *in.position();
+        ++in.position();
+    }
+
+    if (buff.empty())
+    {
+        
+        if constexpr (throw_exception)
+            throw Exception("Cannot read floating point value: no digits read", ErrorCodes::CANNOT_PARSE_NUMBER);
+        else
+            return ReturnType(false);
+  
+    }
+
+    auto res = fast_float::from_chars(buff.data(), buff.data() + buff.size(), x);
+
+    if (res.ec != std::errc())
+    {
+        if constexpr (throw_exception)
+            throw Exception("Cannot read floating point value", ErrorCodes::CANNOT_PARSE_NUMBER);
+        else
+            return ReturnType(false);
+    }
+
+    return ReturnType(true);
+}
+
+#endif
+
 
 template <typename T, void F(T&, ReadBuffer&)>
 void NO_INLINE loop(ReadBuffer & in, WriteBuffer & out)
@@ -76,6 +125,9 @@ try
     if (method == 1) loop<T, readFloatTextPrecise>(in, out);
     if (method == 2) loop<T, readFloatTextFast>(in, out);
     if (method == 3) loop<T, readFloatTextSimple>(in, out);
+    #ifdef USE_FAST_FLOAT
+    if (method == 4) loop<T, readFloatTextFastFloatImpl>(in, out);
+    #endif
 
     return 0;
 }