Fine-Tuning Vectorization and Memory Traffic on Intel Xeon Phi Coprocessors: LU Decomposition of Small Matrices