Ryzen Compiler Performance: Clang 4/5 vs. GCC 6/7/8 Benchmarks

With the FFTW benchmark there is clear benefits from znver1 over generic x86-64 tuning on GCC. GCC with the znver1 tuning ended up being right in line with the LLVM Clang 5 SVN performance.

With SciMark2, overall, Clang remains faster than GCC. Interestingly, the znver1 tuning had regressed the performance on GCC 7.1 and GCC 8.0 compared to just the x86-64 target.

It appears Monte Carlo is the particular SciMark2 test where znver1 pulls back hard with GCC 7 and newer.

GCC is much faster than Clang with the FFT test.

While with the dense LU matrix factorization and Jacobi are where Clang did very well to help boost its composite score.