Adaptation of MPDATA Heterogeneous Stencil Computation to Intel Xeon Phi Coprocessor

Figure 6

Preliminary performance results: (a) performance gain for improved version of (3 + 1)D decomposition; (b) advantages of applying the loop fusion on tile management level; (c) performance for different block sizes; (d) performance for different configurations of teams; (e) advantages of using vectorization; (f) performance for different numbers of threads per core.

(a)

(b)

(c)

(d)

(e)

(f)