Compiler Benchmarks Of GCC, LLVM-GCC, DragonEgg, Clang

The open-source Crafty chess program also produced interesting results. Crafty could not be built with GCC-DragonEgg, LLVM-GCC, or Clang, but when looking at the GCC versions the best numbers were found with GCC 4.4. The performance with GCC 4.5/4.6 is slightly pushed back compared to the 4.4 release, albeit these three releases are all much faster than using GCC 4.2/4.3.

The Himeno benchmarks were rather uninteresting on the two Intel systems where the performance was dead even across the eight tested 64-bit compiler configurations. With the AMD Opteron, however, there was a difference when it came to deploying the Low-Level Virtual Machine. LLVM-GCC, DragonEgg, and Clang all did significantly better than any of the tested GCC releases! GCC 4.5 with DragonEgg 2.8 was the best on this quad-core AMD processor to the point that it did even better than the 12-thread Core i7 970 processor. Simply plugging-in DragonEgg to GCC 4.5 caused its performance to increase for this pressure solver by 51%!

The MAFFT test also produced interesting results. Unlike with Himeno where the LLVM-based x86_64 compilers came out ahead on the AMD system, this molecular biology test did measurably worse with LLVM-GCC / DragonEgg / Clang. This happened just not on the AMD system, but the Intel notebook and desktop too. The best compiler for MAFFT was actually GCC 4.3.0 for the Core 2 Duo and Opteron systems while it was GCC 4.6 with the expensive Intel Gulftown.