Initial Benchmarks Of The AMD AOCC 5.0 Compiler On 5th Gen EPYC
For some of the crypto workloads like SMHasher was slightly better performance on AOCC 5.0 compared to upstream LLVM/Clang... LLVM/Clang 17 and 19 upstream versions in turn were already delivering some nice gains compared to upstream GCC. Though for clarity in the case of SMHasher itself it does use slightly different compiler flags in its build system detection between GCC and Clang around the LTO options.
Granted in some workloads there is virtually no difference between GCC and Clang and AOCC performance.
In the case of the LZ4 compressor code there was just ever so slightly better performance out of AOCC 5.0 but while very narrow margins was seen consistently seen across the various sub-tests.
Similarly, AOCC 5.0 maintained an ever so slight lead over the other compilers with the widely-used Zstandard compression code that is being used these days by countless projects for data compression/decompression needs.