/*
//	====================== mpi_scatter_reduce01.c ======================
//	
//	 Master distributes (scatters) an array across proccesses.
//	 Processes add their elements, then combine sum in master
//	 through a reduction operation
//	====================================================================
*/
#include<stdio.h>
#include<mpi.h>

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

	float	A[16],		//Global array
			MyA[4],		//My portion of A
			MySum,		//My partial sum
			Sum;		//Total sum

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

	//Master fill array (analogous to a read operation)
	if (MyPE == 0)
	{
		for (i = 0; i < 16; i ++)
		{
			A[i] = i + 1;
		}
	}

	//Scatter array among all processes - each process gets
	//4 contiguous entries of A
	MPI_Scatter(A, 4, MPI_FLOAT, MyA, 4, MPI_FLOAT, 0, MPI_COMM_WORLD);
	printf("PE %d: MyA is - ", MyPE);
	for (i = 0; i < 4; i ++)
	{
		printf("%f ", MyA[i]);
	}
	printf("\n");

	//Calculate sums, then reduce to master process

	MySum = 0;
	for (i = 0; i < 4; i ++)
	{
		MySum += MyA[i];
	}

	MPI_Reduce(&MySum, &Sum, 1, MPI_FLOAT, MPI_SUM, 0, MPI_COMM_WORLD);
	if (MyPE == 0)
	{
		printf("Total Sum is: %f\n", Sum);
	}
	MPI_Finalize();
	return 0;
}