cmake: replace -fopenmp-target-fast with -O3 for LLVMFlang AMD GPU builds

Author: Spencer Bryngelson

CMakeLists.txt

@@ -643,7 +643,7 @@ exit 0
                     target_compile_options(${a_target} PRIVATE -fopenmp)
                     target_link_options(${a_target} PRIVATE -fopenmp)
                 elseif(CMAKE_Fortran_COMPILER_ID STREQUAL "LLVMFlang")
-                    target_compile_options(${a_target} PRIVATE -fopenmp --offload-arch=gfx90a -fopenmp-target-fast -fopenmp-assume-threads-oversubscription -fopenmp-assume-teams-oversubscription)
+                    target_compile_options(${a_target} PRIVATE -fopenmp --offload-arch=gfx90a -O3 -fopenmp-assume-threads-oversubscription -fopenmp-assume-teams-oversubscription)
                     target_link_options(${a_target} PRIVATE -fopenmp --offload-arch=gfx90a -flto-partitions=${MFC_BUILD_JOBS})
                 endif()
             endif()