GCC vs. LLVM Clang vs. AOCC Compiler Benchmarks On The AMD EPYC 7742 2P Linux Server

The CppPerformanceBenchmarks synthetic compiler tests were mixed between LLVM and Clang based upon the particular functionality being stressed.

For the Apache web server performance, the results were too close to call but that the Clang and GCC compilers were leading to similar performance.

In several of the tests the AOCC 2.0 compiler did deliver better performance than the upstream LLVM Clang compiler. Hopefully AMD's compiler team will be able to get the relevant patches upstreamed in a timely manner for LLVM Clang 10.0 in order to ensure more broad Znver2 compiler optimizations to benefit their customers. As for the Clang performance against GCC, like with Intel's recent microarchitectures, it remains an extremely competitive race. Unlike years ago where LLVM Clang generally trailed GCC on x86_64, these days it's a brutal race and largely depends upon which workload(s)/applications(s) are most important to your needs for determining which compiler is more appropriate. Granted, there is also other factors to consider when choosing a code compiler beyond just the performance of the resulting binary.