I've noticed that amdlibm was removed from ACML. So, I'm hoping this is still the right place to ask this question.
First, amdlibm provides optimzations of core math functions. Is this needed to get the same level of optimization using openf90, or does the fortran compiler have its own optimized functions?
Second, does the open64 compiler choose optimal function versions (scalar vs vector), or does the author need to specify these manually?