The interpolate module uses either point sampling or bi/trilinear sampling to compute intermediate values to change the size of a field. The input consists of any 2D or 3D scalar field of any type. The output is a proportionally resized field matching the type and size of the input.
In the field math module, the operation performed on each element of the field involves only a single element and the MPI Scatter/Gather functions can be used to partition the input fields into non-overlapping blocks, which are distributed over the available PEs. In the interpolate module, the interpolation operators involve several neighboring elements in the field. For this reason, each PE must contain one or more extra rows and columns of data surrounding the block of the field that is owned by the PE.
As in field math, the input field is transferred from the Y-MP to a master PE on the T3D. The master PE then uses the MPI_Send function to distribute blocks of the field with the necessary overlap to all the processors. These blocks are received by the MPI_Recv function in the slave processors. All processors perform interpolation on their own blocks in parallel and then return the results to the master PE using another pair of MPI_Send and MPI_Recv calls.