AMD AOCC 4.0 vs. GCC vs. LLVM Clang Compiler Benchmarks On Zen 4
With the demanding LeelaChessZero that is a chess engine leveraging neural networks, both the Eigen and BLAS back-ends were showing the fastest performance when built under AOCC 4.0. (GCC 13 failed to compile this test due to increased warnings/errors.) Upstream LLVM/Clang performance was looking good for LC0 over GCC but the AOCC 4.0 numbers show there can be real benefits from the Znver4 tuning by AMD.
For other tests like CLOMP in measuring the OpenMP efficiency were also showing the best results when built under AOCC 4.0 for this Ryzen 9 7950X desktop.
Of course, for many C/C++ programs out there the compiler doesn't make all that much difference on the resulting performance... Years ago when Clang was still playing catch-up to GCC, the compiler benchmark scene was much more lively, while these days Clang has consistently proven it can often generate at least comparable code to GCC while often times now leads the race.
LLVM/Clang and AOCC 4.0 were faster than GCC 12.2 at the Crypto++ cryptography benchmarks. Crypto++ 8.2 failed to build with GCC 13 due to increased compiler warnings/errors.