GCC vs. LLVM Clang vs. AOCC Compilers On AMD Threadripper
First up was FFTW with a small size at which case the performance was very close between compilers and AMD's LLVM/Clang-branched AOCC 1.2 compiler release was slightly faster.
When increasing the problem size for FFTW, GCC 8.1/7.3 were now faster than LLVM Clang, which for the Clang 5.0 and 6.0 releases tended to be much slower. Fortunately, LLVM Clang 7.0 SVN was much faster than the current stable Clang 6.0. Of the LLVM-based compilers, the AOCC 1.2 release was slightly faster than what is currently offered via Subversion/Git with LLVM/Clang.
In the HMMer scientific test, the compiler performance was about the same except for GCC 8.1 regressing on this Threadripper 1950X system.
Under the SciMark 2 benchmark, LLVM Clang 7.0 again showed a performance improvement compared to the current 6.0 stable release and came out slightly ahead of GCC.
With the TSCP chess benchmark, AMD AOCC 1.2 came in at around the same speed of Clang 6.0.0 that was slightly faster than the other compilers benchmarked.
While in the TTSIOD 3D Renderer it still heavily favors GCC with GCC 8.1 offering significantly better performance.