Effective SIMD Vectorization for Intel Xeon Phi Coprocessors

Figure 3

Performance gain with “less-than-full-vector” loop vectorization.