/*====================== mpi_gather01.c ==============================

 Calculates the matrix-vector product of a 4x4 matrix A and
 4x1 vector x, using 4 processes. Originally, each process
 stores a row of A and a single entry of x.
 This method uses all_gather operations to place a full
 copy of x in each process, which then performs the row-column
 inner-product operation on its data.
====================================================================*/

#include<stdio.h>
#include<mpi.h>

main( int argc, char** argv)
{
//	MyPE: My logical process number
//	NumPE: Number of processes
//	i: loop index variable
	int MyPE, NumPE, i;

//	MyA: Single 1x4 row of 4x4 array A
//	MyX: Row entry of 4x1 column vector x
//	FullX: Full copy of x
//	ip: Inner product (result for row i)
//	product: Final vector result
	double MyA[4], MyX, FullX[4], ip, product[4];

	MPI_Init(&argc, &argv);
	MPI_Comm_rank(MPI_COMM_WORLD, &MyPE);
	MPI_Comm_size(MPI_COMM_WORLD, &NumPE);

//	Assign myself a row of A and x
	for (i=0; i<4; i++)
	{
		MyA[i] = MyPE + 1 + 4 * i;
	}

	MyX = MyPE + 17;

//	We want every process to have a copy of x, so
//	we perform all_gather operations
	MPI_Allgather(&MyX, 1, MPI_DOUBLE, FullX, 1, MPI_DOUBLE, MPI_COMM_WORLD);

//	Now, each process creates inner product for ith row
	ip = 0.0;
	for (i=0; i<4; i++)
	{
		ip = ip + MyA[i]*FullX[i];
	}

//	Finally, we gather all process' ip into master process
	MPI_Gather(&ip, 1, MPI_DOUBLE, product, 1, MPI_DOUBLE, 0, MPI_COMM_WORLD);
	if (MyPE == 0)
	{
		printf("Matrix-vector product is:\n");
		for (i=0; i<4; i++)
		{
			printf("%lf\n", product[i]);
		}
	}
	MPI_Finalize();
	return 0;
}