Apple's LLVM 3.1 Clanging On Intel Sandy Bridge

Like going from GCC 4.6.3 to GCC 4.7.0, there is a noticeable performance improvement in the multi-threaded C-Ray ray-tracer when moving from LLVM-Clang 3.0 to LLVM-Clang 3.1 SVN. However, with GCC 4.7.0 it is a much bigger improvement than the LLVM optimizations, which now gives it a larger advantage over the Apple-sponsored compiler.

The LZMA compression performance on LLVM-Clang was pulling in right behind GCC.

The OpenMP-using Smallpt regresses with LLVM/Clang 3.1.

You can compare your system's compiler performance by simply running phoronix-test-suite benchmark 1203235-SU-GCC47LLVM92 to these Sandy Bridge Extreme numbers. Coming up are some more GCC and LLVM-Clang results when using an eight-core AMD Bulldozer as well as trying out other compiler tuning options. LLVM/Clang 3.1 should see its official release in May.