How does clang pick the memcpy optimized for the floating point co-processor on my ARM target?

I am working on project that is currently build with gcc and run-times based on newlib 4.3.0.