GCC 8/9 vs. LLVM Clang 7/8 Compiler Performance On POWER9 With The Raptor Talos II
The first benchmark yielding an interesting difference in performance on the Raptor Talos II was CacheBench where the read performance did see a small uplift in performance using the GCC9 development compiler.
But interestingly in the write case for CacheBench, LLVM Clang was much faster than GCC.
In the case of the common SciMark2 C micro-benchmarks, Clang was also faster than GCC9 on this high-end POWER9 Linux system.
Clang's wins did carry through to the TSCP chess engine where Clang was faster than GCC. Fortunately, the latest development versions of GCC and Clang each show minor performance improvements over their predecessors.
x265 was having problems compiling with Clang on POWER9.
GraphicsMagick was faster on GCC than Clang, even with having LLVM's libomp support available for this OpenMP-threaded image manipulation program.
The Himeno pressure solver meanwhile saw nice performance improvements with the latest versions of both GCC and Clang, but GCC 9.0.1 ended up being about 5% faster than Clang for this scientific program.