Preview: EKOPath Benchmarks On AMD's Bulldozer

With BYTE's Dhrystone 2 test, the EKOPath compiler performance came in right around the speed of GCC 4.7. Using "bdver1" rather than "k8" as the microarchitecture value led to a small drop in performance with the GNU Compiler Collection while the speed was the same with PathScale's EKOPath compiler. LLVM/Clang 3.1 SVN could not successfully run this test profile.

When running the Himeno Poisson Pressure Solver, the GCC 4.7 performance improved by 12% when using "bdver1" over "k8" while the Open64 compiler performance improved by 5% and the EKOPath compiler was up by 11% with the Bulldozer optimizations. LLVM/Clang was the only compiler where the performance actually went down when using the bdver1 flag while its K8 numbers were the best of any compiler configuration. Between GCC 4.7.0 and PathScale's EKOPath nightly build, the Free Software Foundation's compiler did better (+18% for K8 and +20% for BDVER1), but in turn its performance was better than the Open64 compiler spin from AMD.