Adaptation of MPDATA Heterogeneous Stencil Computation to Intel Xeon Phi Coprocessor

Algorithm 1

Part of MPDATA implementation.

//stage 1

for i-dim

 for j-dim

  for k-dim

   f1[i][j][k] =  xIn[i][j][k] * max(0.0, u1[i][j][k])

-  xIn[i-1][j][k] * (-min(0.0, u1[i][j][k]));

//stage 2

for i-dim

 for j-dim

  for k-dim

   f2[i][j][k] =  xIn[i][j][k] * max(0.0, u2[i][j][k])

 -  xIn[i][j-1][k] * (-min(0.0, u2[i][j][k]));

//stage 3

for i-dim

 for j-dim

  for k-dim

   f3[i][j][k] =  xIn[i][j][k] * max(0.0, u3[i][j][k])

-  xIn[i][j][k-1] * (-min(0.0, u3[i][j][k]));

//stage 4

for i-dim

 for j-dim

  for k-dim

   xOut[i][j][k] =  xIn[i][j][k] - (f1[i+1][j][k]

-  f1[i][j][k] + f2[i][j+1][k] - f2[i][j][k]

+  f3[i][j][k+1] - f3[i][j][k])/h[i][j][k];

/*⋯*/