AMD AOCC 1.1 Shows Compiler Improvements vs. GCC vs. Clang

AOCC 1.1 was trying to run faster than the others with Hackbench,

AOCC 1.1 was also trying its best to outperform GCC and Clang with the Minion benchmarks.

But in some common and already thoroughly optimized applications like Apache and OpenSSL, AOCC 1.1 didn't make any difference over the mature GCC and Clang compilers.

These AOCC 1.1 benchmarks are certainly more interesting than what we saw out of AOCC 1.1 last spring when its performance was barely any different than upstream LLVM Clang 5.0. We were told by AMD at the time they had mostly been optimizing AOCC for SPEC CPU performance benchmarks. Thus it's pleasant to see that 2018 is starting with several performance improvements out of the AOCC compiler stack compared to what's currently available in upstream GCC or Clang. But we hope AMD is working on getting more of their Zen/znver1 improvements upstream into these dominant open-source compilers soon so there will be more widespread usage and adoption of their znver1 tuning. I'm currently running some AOCC 1.1 benchmarks on Ryzen hardware and will be posting more benchmarks if making any other interesting performance discoveries on that front.