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

When looking at the John The Ripper Blowfish performance, the unreleased GCC 4.6 is again performing the best and well ahead of previous GCC releases on both Intel and AMD systems. DragonEgg and Clang both lagged behind in performance miserably compared to GCC.

BYTE could only be successfully built on GCC 4.2.1/4.3.0/4.4.0. BYTE could be built on GCC 4.5/4.6, LLVM-GCC, and DragonEgg, but with these three compilers, it produced incorrect results. LLVM/Clang 2.8 could not successfully build this small open-source benchmark.

C-Ray is an always-interesting test as a ray-tracing benchmark and when testing compilers it produced interesting results. Across the Penryn, Gulftown, and Shanghai systems, the C-Ray performance was worse under GCC 4.4.0, but this problem as fixed in GCC 4.5. Unlike some other tests where a performance boost could be found in the most recent GCC 4.6 snapshot, this was not the case for C-Ray. Interestingly though, we finally see the LLVM-based compilers pull out ahead of GCC with this ray-tracing test. On both multi-core Intel computers, the LLVM-GCC and LLVM DragonEgg performance was much faster with the LLVM code generators and optimizers rather than those from GCC. However, the LLVM Clang performance was not any better than GCC, except with the AMD workstation setup where the latest release of Clang produced the best results.