Testing LLVM Clang 3.5's Code Generation Optimizations

With GraphicsMagick image resizing we were back to experiencing a performance drop when passing the -march=native flag. Between -O1 and -Ofast wasn't much of a change in performance for this test.

The optimizations look good for the HWB Color Space conversion test along with Local Adaptive Thresholding.

Only basic compiler optimizations were of benefit to Himeno.