GCC & LLVM Clang Compiler Benchmarks On AMD's EPYC 7601

In many benchmarks these days, LLVM Clang and GCC are producing binaries that perform at close to the same speed... Certainly that couldn't be said just a few years ago when Clang was still catching up to the performance levels and optimizations of GCC.

With the BLAKE2 crypto test, LLVM Clang was more efficient than the three most recent GCC branches tested.

With the C-written micro-benchmarks of SciMark2, GCC 6.3 as found on Debian Stretch was the compiler yielding the fastest binaries. GCC 8.0 actually has regressed in performance more than GCC 7.2 and GCC 6.3.

The Monte Carlo sub-test is where GCC has regressed in comparison to GCC 6.3 or Clang.

While in some tests like with FFT and Sparse Matrix Multiply, Clang is slightly faster than GCC.

GCC 8.0 is also currently regressed when it comes to the dense LU matrix factorization.