I used Agner Fog's vector class but met a serious performance reduction problem when the code was compiled by MSVC

Question

I tried to test original SSE/AVX functions compared with Agner Fog's vector class(My CPU supports both SSE and AVX instructions). Everything was in line with forecast on GNU compiler. Then I switched to MSVC and suddenly found that all vector class options were about 10 times or even more slower than original SSE/AVX functions. My IDE is VS2019/Clion, and the GNU version is 12.2. In addition, I used CMake to conduct compiling. The specific configurations of compilers are as follows:

cmake_minimum_required(VERSION 3.21)
project(testSIMD)

set(CMAKE_MODULE_PATH ${CMAKE_MODULE_PATH} "${CMAKE_SOURCE_DIR}/external_modules/cmakeFind/")

set(CMAKE_CXX_STANDARD 17)

set(USE_VECTORIZE ON)  

if (NOT VECTOR_TYPE)        
    set(VECTOR_TYPE full_vectorize)
endif (NOT VECTOR_TYPE)
if (VECTOR_TYPE STREQUAL "none")
    set(VECTOR_TYPE none)
endif (VECTOR_TYPE STREQUAL "none")
if (VECTOR_TYPE STREQUAL "default")
    set(VECTOR_TYPE default)
endif (VECTOR_TYPE STREQUAL "default")


# Set vectorization flags for a few compilers
set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS}")
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS}")


find_package(VectorFOG REQUIRED)

#Pushing vector flags onto compiler flags for try compiles
set(CMAKE_C_FLAGS_SAVE ${CMAKE_C_FLAGS})
set(CMAKE_CXX_FLAGS_SAVE ${CMAKE_CXX_FLAGS})
set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} ${VECTOR_C_FLAGS}")
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} ${VECTOR_CXX_FLAGS}")

try_compile(HAVE_FOG_VECTOR_CLASS "${CMAKE_BINARY_DIR}" "${CMAKE_SOURCE_DIR}/fogvectorclasstest.cpp")

if(HAVE_FOG_VECTOR_CLASS)
    message(STATUS "Trying Fog Vector Class -- works")
    add_definitions(-DHAVE_FOG_VECTOR_CLASS)
    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -DMAX_VECTOR_SIZE=512")
    set (FILE_LIST ${FILE_LIST}
            kahan_fog_vector.cpp
            kahan_fog_vector8.cpp)
else()
    message(STATUS "Trying Fog Vector Class -- fails")
endif()


# Set vectorization flags for GNU compiler
if (CMAKE_CXX_COMPILER_ID STREQUAL "GNU") # using GCC
    if (USE_VECTORIZE)
        set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -g -O3 -fstrict-aliasing -fopenmp-simd -march=native -mtune=native -ffast-math -ftree-vectorize -fopt-info-vec-optimized")
    else ()
        set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -g -O3 -fstrict-aliasing -march=native -mtune=native -ffast-math -fno-tree-vectorize -fopt-info-vec-optimized")
    endif ()
    if ("${CMAKE_CXX_COMPILER_VERSION}" VERSION_GREATER "7.4.0")    
        set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -mprefer-vector-width=512")
    endif ("${CMAKE_CXX_COMPILER_VERSION}" VERSION_GREATER "7.4.0")
elseif ("${CMAKE_CXX_COMPILER_ID}" STREQUAL "Clang") # using Clang
    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -g -O3 -fstrict-aliasing -fvectorize -march=native -mtune=native -ffast-math -Rpass-analysis=loop-vectorize")
elseif ("${CMAKE_CXX_COMPILER_ID}" STREQUAL "MSVC")
    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /O3")
    add_compile_options(/arch:AVX2)    
    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /Qvec-report:2")
endif ()

I was quite confused about this problem. If anyone knows why it happened, please let me know.