Sending Lower Triangular Portion of a Matrix

double L[n][n];

int block_lengths[n];

int displacements[n];

MPI_Datatype index_t;

for (i=0;i<n;i++)

{

block_lengths[i] = i+1;

displacements[i] = n*i;

}

MPI_Type_indexed(n, block_lengths, displacements, MPI_DOUBLE, &index_t);

MPI_Type_commit(&index_t);

if (!my_id)

{MPI_Send(A, 1, index_t, 1, 0, MPI_COMM_WORLD);}

else if (my_id==1)

{MPI_Recv(L, 1, index_t, 0, 0, MPI_COMM_WORLD, &status);}

Note that even though the blocks are uniformly spaced in memory we couldn’t

use MPI_Type_vector because the block lengths are different for each row