// Sorting using a generalized pattern pipeline B. Wilkinson Dec 19, 2015.

#include <stdio.h>
#include <string.h>
#include <stdlib.h>
#include "suzaku.h"  	// Basic Suzaku macros

#define D 1	// # of data items in slave data.
#define P 4	// Number of processes -- this code must be run only with this number of processes, each process only handles one number

int main(int argc, char *argv[]) {

	int i, j, p,pid;		// p is actual number of processes when executing program
	int T = 3 * P;			// number of time steps
	double A[D];			// data to send (D = 1). Needs to be array as generalized send is a routine not a macro
	double B[P][D];			// received data, from each source
	static double largest = 0;	

	for (i = 0; i < D; i++) A[i] = 0;	// initialize to zero
	for (i = 0; i < P; i++) 		// initialize receive so can see what received
		for (j = 0; j < D; j++) 
			B[i][j] = -999;

	srand(1);				// initialize rand()

	SZ_Init(p);				// initialize MPI message-passing environment
	if (p != P) 				// number of processes hardcoded
		printf("ERROR number of processes must be %d\n",P); 
	
	SZ_Parallel_begin			// parallel section, all processes do this

	   SZ_Pattern_init("pipeline",D);	// set up slave interconnections
	   SZ_Print_connection_graph();		// for checking
	   
	   for (i = 0; i < T; i++) {

   		pid = SZ_Get_process_num();	// identify process

		if (pid ==0)  {			// master generates next number to sort, ends with a terminator

			if (i < P) A[0] = rand()% 100; 	// P numbers, a number between 0 and 99
			else A[0] = 999;		// otherwise terminator
			printf("Master sends %3.0f and receives %3.0f\n",A[0],B[0][0]); 

		} else  { 				// slaves execute compute, using B to create A.

			if (B[0][0] > largest) { 
				A[0] = largest; 	// copy current largest into send array
				largest = B[0][0];	// replace largest with received number
			} else {
				A[0] = B[0][0];		// copy received number into send array
			}

		} 

		SZ_Generalized_send(A,B);	// sent compute results to connected slaves, includes master to slave, slave to master
		SZ_Barrier();			// wait for every process to complete

	   }

	SZ_Parallel_end;			// end of parallel

	SZ_Finalize(); 

	return 0;
}