#include <iostream>
#include <random>
#include <chrono>
#include <mpi.h>

using namespace std;

int main(int argc, char **argv) {
    int pid, nprocs;
    MPI_Init(&argc, &argv);
    MPI_Comm_rank(MPI_COMM_WORLD, &pid);
    MPI_Comm_size(MPI_COMM_WORLD, &nprocs);

    chrono::time_point<chrono::system_clock> start, end;		
    
    
    int n = atoi(argv[1]); // la taille du tableau global
    int root = atoi(argv[2]); // le processus root
    
    double *vecteur;
    double *vecres;
    if (pid == root) {
        vecteur = new double[n];
        vecres = new double[n];
        random_device rd;
        mt19937 gen(rd());
        uniform_real_distribution<> dis(0, 10.0);
        for (int i = 0; i < n; i++)
            vecteur[i] = dis(gen);
        cout << "le tableau initial est ";
        for (int i = 0; i < n; i++)
            cout << vecteur[i] << " ";
        cout << endl;
    }
 if (pid==root) {
      start = chrono::system_clock::now();
    }
    int nlocal = n / nprocs;
    int reste = n%nprocs;

    int* sendcounts;
    int* displs;
    if (pid==root) {
        sendcounts = new int[nprocs];
        displs = new int[nprocs];
        int ptr = 0;
        for (int i=0; i<reste; i++) {
            sendcounts[i] = nlocal + 1;
            displs[i] = ptr;
            ptr += nlocal+1;
        }
        for (int i=reste; i<nprocs; i++) {
            sendcounts[i] = nlocal;
            displs[i] = ptr;
            ptr+= nlocal;
        }
    }
    if (pid<reste)
        nlocal++;
    double* veclocal = new double[nlocal];
    MPI_Scatterv(vecteur,sendcounts,displs,MPI_DOUBLE,veclocal,nlocal, MPI_DOUBLE,root,MPI_COMM_WORLD);
    double norm = 0;
    for (int i=0; i<nlocal; i++)
        norm += veclocal[i]*veclocal[i];
    double res;
    MPI_Reduce(&norm,&res,1,MPI_DOUBLE,MPI_SUM,root,MPI_COMM_WORLD);
    if (pid==root)
        res = sqrt(res);
    MPI_Bcast(&res,1,MPI_DOUBLE,root,MPI_COMM_WORLD);
    for (int i=0; i<nlocal; i++)
        veclocal[i] /= res;
    MPI_Gatherv(veclocal,nlocal,MPI_DOUBLE,vecres,sendcounts,displs,MPI_DOUBLE,root,MPI_COMM_WORLD);
    if (pid==root) {
      end = chrono::system_clock::now();				
      chrono::duration<double> elapsed_seconds = end-start;		
      cout << "tps=" << elapsed_seconds.count() << endl;
      for (int i=0; i<n; i++)
          cout << vecres[i] << " ";
      cout << endl;
    }
    if (pid==root) {
        delete[] vecteur;
        delete[] vecres;
	delete[] displs;
	delete[] sendcounts;
    }
    delete[] veclocal;
    MPI_Finalize();
    return 0;
}