// ==================================================================
// Author: Jean-Michel Richer
// Email: jean-michel.richer@univ-angers.fr
// Date: Aug 2016
// Purpose: Demonstrate basic send functionality, send array of float
// from master (rank=0) to slave (rank=1)
// ==================================================================
#include <unistd.h>
#include <iostream>
#include <cstdlib>
#include <unistd.h> // for sleep
using namespace std;
#include <mpi.h>

// ==============================================================
// C++ version 
// ==============================================================
 
int main(int argc, char ** argv) {
	// maximum number of CPUs
	int max_cpus;
	// cpu identifier (called rank)
	int cpu_rank;
	// C-string to store name of the host
	char cpu_name[MPI::MAX_PROCESSOR_NAME];
	// length of the C-string
	int length;

	// initialization also using command line parameters
	MPI::Init(argc, argv);

	// get number of programs running
	max_cpus = MPI::COMM_WORLD.Get_size();
	
	// get program identifier
	cpu_rank = MPI::COMM_WORLD.Get_rank();
	
	// get cpu name
	memset(cpu_name, 0, MPI::MAX_PROCESSOR_NAME);
	MPI::Get_processor_name(cpu_name, length);
 
	
	cerr << cpu_rank << "/" << max_cpus << " on machine " << cpu_name << endl;
	
	// number of elements of the array 'data' that will be allocated
	int data_size;
	// array of float created by the master and sent to the slave(s)
	float *data;
	// identifier (rank) of remove processor (master or slave)
	int remote_cpu;
	
	// Status for communication
	MPI::Status status;

	// MASTER
	if (cpu_rank == 0) {
		// MASTER
		// processor of id 0 (master) fills and sends the array
		data_size = 100 + rand() % 100;
		data = new float [data_size];
		for (int i=0; i<data_size; ++i) data[i] = i+1;
		
		int remote_cpu = 1;
		
		// 1- send length
		cerr << cpu_rank << " [send]  data_size=" << data_size << endl; 
		MPI::COMM_WORLD.Send(&data_size, 1, MPI::INT, remote_cpu, 0);
		// 2- send data
		MPI::COMM_WORLD.Send(&data[0], data_size, MPI::FLOAT, remote_cpu, 0);		
		// wait for other processor to compute and send back sum
		float result = 0;

		MPI::COMM_WORLD.Recv(&result, 1, MPI::FLOAT, remote_cpu, MPI::ANY_TAG, status);
		cerr << cpu_rank << " [recv] result=" << result << endl;
		cerr << "result is " << result << " for length=" << data_size;
		cerr << ", expected=" << (data_size * (data_size+1))/ 2  << endl;
		delete [] data;
		
	} else if (cpu_rank == 1) {
		// SLAVE
		// processor of id 1 will receive the array and compute the sum
		// 1- we receive the array length and allocate space
		remote_cpu = 0; // master
		MPI::COMM_WORLD.Recv(&data_size, 1, MPI::INT, remote_cpu, MPI::ANY_TAG, status);
		cerr << cpu_rank << " [recv] data_size=" << data_size << endl;
		data = new float[data_size];
		// 2- we receive the data
		MPI::COMM_WORLD.Recv(&data[0], data_size, MPI::FLOAT, remote_cpu, MPI::ANY_TAG, status);
		// 3- compute sum
		float sum = 0;
		for (int i=0; i<data_size; ++i) {
			sum += data[i];
		}
		// 4- send back result
		cerr << cpu_rank << " [send]  sum=" << sum << endl; 
		MPI::COMM_WORLD.Send(&sum, 1, MPI::FLOAT, remote_cpu, 0);
		delete [] data;
		
	} else {
		// other processors (if any) won't do anything
	}
   
	MPI::Finalize();
 
	exit(EXIT_SUCCESS);
}