LLVM/Clang 3.2 SVN Compiler Performance

Kicking things off with the HMMer testing, there was no change in performance from this Linux system with the Core i7 3960X EE processor for either the nocona or corei7-avx targets. There was also no performance difference between nocona and corei7-avx optimizations for this particular workload.

There's several different workloads within GraphicsMagick, but for the image blur task we see a small performance improvement now out of using corei7-avx optimizations under LLVM/Clang 3.2. (Note: LLVM/Clang as of the latest SVN code still doesn't support OpenMP, although it's actively being worked on.)

The sharpening task shows a large degredation in performance for the LLVM/Clang 3.2 SVN code when using building with the -march=corei7-avx flag.

Resizing with GraphicsMagick when using the corei7-avx binary from LLVM/Clang 3.2 also led to a performance drop.