perf(hex8): prepare guarded private four-column paired tangent packet

src/fesa/CMakeLists.txt

@@ -44,6 +44,8 @@ add_library(
     elements/element_factory.cpp
     elements/hex8_element.cpp
     elements/hex8_high_low.cpp
+    elements/hex8_pair_simd_dispatch.cpp
+    elements/hex8_pair_simd_avx2.cpp
     elements/hex8_geometry.cpp
     elements/euler_beam_3d.cpp
     elements/finite_rotation.cpp
@@ -122,6 +124,14 @@ target_compile_options(
         $<$<CXX_COMPILER_ID:MSVC>:/W4 /WX /EHsc>
 )
 
+# Keep the CPU/OS dispatch on the baseline ISA; only its private kernel uses AVX2.
+if(CMAKE_CXX_COMPILER_ID STREQUAL "MSVC" AND
+   CMAKE_CXX_COMPILER_ARCHITECTURE_ID STREQUAL "x64")
+    target_compile_definitions(fesa_solver PRIVATE FESA_HEX8_PAIR_AVX2=1)
+    set_source_files_properties(elements/hex8_pair_simd_avx2.cpp
+        PROPERTIES COMPILE_OPTIONS /arch:AVX2)
+endif()
+
 add_executable(fesa_cli app/main.cpp)
 set_target_properties(fesa_cli PROPERTIES OUTPUT_NAME fesa)
 target_link_libraries(fesa_cli PRIVATE fesa_solver)