The author says that in userspace memcpy, AVX is used, but > The implementation of copy_user_enhanced_fast_string, on the other hand, is much more modest. Why

The author says that in userspace memcpy, AVX is used, but > The implementation of copy_user_enhanced_fast_string, on the other hand, is much more modest. Why is that? I mean, if you compiled your kernel for a wide range of machines, then fine, but if you compiled targeting your actual CPU, why would the kernel functions not use AVX?