LLVM 2.9 Clang Performance On Intel's Sandy Bridge

With the Local Adaptive Thresholding test, the LLVM 2.9 performance was disturbingly slow.

Even with a simple image resizing operation on GraphicsMagick, LLVM with Clang and GCC 4.6 with the corei7-avx support had lagged.

The OpenSSL performance was largely unaffected by the choice of compiler or compiler options.

There we have it, our first tests of the latest LLVM/Clang 2.9 code on an Intel Sandy Bridge processor. The Low-Level Virtual Machine with Clang C/C++ compiler front-end was competitive with GCC 4.5/4.6 -- and this Apple-sponsored compiler won in some benchmarks too -- but it should be more interesting once LLVM is better optimized for this latest Intel hardware. We will continue to benchmark the latest LLVM/Clang code to see how the performance evolves, since LLVM continues to grow increasingly important not just as an alternative compiler within the Linux, Mac OS X, and BSD worlds in particular, but also for its use in a growing number of open-source projects such as Gallium3D and Mono for its IR and optimizations.