AMD Llano Compiler Performance

There were not any major differences between GCC and LLVM/Clang for the OpenSSL benchmark, which is already highly optimized for both compilers.

The DES encryption performance with John The Ripper did not favor a particular compiler of those tested.

While the DES workload was not affected in a significant manner by any of the three compilers, the Blowfish performance with John The Ripper did the best under GCC 4.6.1 and was the slowest with the Low-Level Virtual Machine.

These results are not terribly surprising. For the most part, the LLVM/Clang performance is on par with GCC when both compilers see the same compiler flags. To some dismay, neither compiler is yet optimized for AMD's Lynx/Llano platform, but the architecture isn't too different from past AMD platforms. Only with GCC 4.6 are there now optimizations for AMD Bobcat (Brazos platform; e.g. AMD E-350) hardware. Hopefully with GCC 4.7 and LLVM 3.0 the performance out of Llano will become more interesting for these leading open-source compilers.

The performance of the Low-Level Virtual Machine on AMD's Fusion APUs will be particularly important since LLVM is also plugged into other areas of the stack like graphics drivers, Mono, and with Gallium3D for its upcoming OpenCL support.

As said in the introduction, a larger compiler round up is planned for the coming months with more hardware and more compilers. For now if you wish to compare these GCC and LLVM results to your own configurations and hardware, run phoronix-test-suite benchmark 1108133-LI-AMDLLANOC78 (you can also view its OpenBenchmarking.org result file).