I need to program for parallel block matrix calculation (Element by Element wise) using MPI. I programmed before row wise allocation to nodes for calculation. In current problem i need to divide tasks according to elements of resultant matrix.
I keep trying how to distribute the data to nodes but have not reached the solution. Could you please help me in understanding the logic about how to do that?
Let's work backwards through this. My notation will come from A*B=C. Each element of C is built from the corresponding row of A and column of B. A(i,0:n)*B(0:n,j)=C(i,j). So, for each element of C, you will need a row of A and a column of B. If you want each rank to solve for a block of C, then you need all corresponding rows of A and all corresponding columns of B. So, break C into blocks, get the ranges needed for each block. C(i1:i2,j1:j2) would require A(i1:i2,0:n) and B(0:n,j1:j2). To distribute this data, I'd recommend using MPI_Scatter and/or MPI_Scatterv (depends on the memory layout which you'll need to use).
