Benchmarking LLVM/Clang's New AMD Zen Scheduler Model

FFTW is slightly faster with Clang SVN over Clang 4.0, but using the znver1 target rather than generic x86-64 had actually yielded a slight regression in performance, but at least both these Clang SVN runs are faster than the current stable Clang release.

The SciMark2 results are basically unchanged with this Clang 5.0 code, although znver1 does continue to yield some minor performance advantages over generic x86-64 march.

The Monte Carlo sub-test is one that benefits from znver1 tuning as well as Dense LU Matrix Factorization.