AMD AOCC 1.2 Code Compiler Offers Some Performance Benefits For EPYC
The LLVM-based compilers tend to not perform as well as GCC still with OpenMP workloads, but at least there is this OMP support compared to when it was absent just a few years ago.
The Himeno pressure solver ended up being the fastest on this EPYC server with LLVM Clang 6.0 but surprisingly was much slower with AOCC 1.2.
The additional optimization work with AOCC 1.2 will lead to slower build times albeit that's not bad when it means greater run-time performance of the resulting binaries.
AOCC 1.2 was the fastest with the Stockfish chess engine benchmark.
While GCC still reigned in tests like the AOBench renderer.