Benchmarking AMD's New AOCC Compiler For Ryzen

First up was FFTW and here the AOCC 1.0 performance is right in line with the mainline Clang 4.0 performance numbers, which in turn is around the same level of performance as GCC 7.1/8.0. The fastest compiler in this case was the LLVM Clang 5.0 SVN snapshot. Clang 5.0 has yet to land the new Ryzen scheduler model or other Zen-specific improvements.

With the MAFFT scientific test, AOCC 1.0 continued performing at about the same speed as upstream Clang 4.0. GCC 6.3 produced the fastest binaries in this case.

With SciMark 2.0, AOCC was tied with LLVM Clang 4.0 and Clang 5.0 SVN for delivering the greatest performance....

We haven't seen any real advantages of the Advanced Optimizing C/C++ Compiler yet over mainline Clang.