LLVM Clang 3.4 Already Has Some Performance Changes

While the composite result was a big improvement on LLVM/Clang 3.4, other test cases within SciMark2 indicate the current compiler SVN code being slightly slower on the core-avx2 processor.

The Dense LU Matrix Factorization process is where the LLVM/Clang 3.4 performance jumped by an incredible 36% over the 3.3 stable release, which puts it well ahead of the GCC performance on the same hardware. It's this test likely responsible for much of the boost in the SciMark2 composite number.

Jacobi Successive Over-Relaxation is still much faster with LLVM/Clang over GCC on the Core i7 4770K Linux system, but with the 3.4 SVN code it appears at least temporarily regressed.

There isn't much to see out of the x264 video encoder in this compiler benchmark since most of their hot-paths are written in hand-tuned Assembly.