c++ - Segmentation Fault calling MPI_Finalize() in c -

i running problem mpi_finalize. segmentation fault @ end of program when can mpi_finalize();, output:

a matrix = 48.3962 65.3245 15.0385  72.383 25.8898 46.0265   b matrix =  15.4881 50.6507 6.74602 71.0055  12.2209 77.5441 61.5452 31.5127  46.8515 89.4849 70.0342 57.3195   gather returned: 0 c matrix = 2252.46 8862.55 5400.1 6356.94  gather returned: 0 3593.88 9792.53 5305.12 8593.66   time=7.6e-05 seconds [dmc:13852] *** process received signal *** [dmc:13852] signal: segmentation fault (11) [dmc:13852] signal code:  (128) [dmc:13852] failing @ address: (nil) [dmc:13852] [ 0] /lib64/libpthread.so.0(+0xf850)[0x7fcba45e8850] [dmc:13852] [ 1] /opt/asn/apps/openmpi_1.8.3_intel/lib/libopen-pal.so.6(opal_memory_ptmalloc2_int_free+0x1a9)[0x7fcba3b74a79] [dmc:13852] [ 2] /opt/asn/apps/openmpi_1.8.3_intel/lib/libopen-pal.so.6(opal_memory_ptmalloc2_free+0xad)[0x7fcba3b765fd] [dmc:13852] [ 3] /opt/asn/apps/openmpi_1.8.3_intel/lib/openmpi/mca_btl_openib.so(+0x27e14)[0x7fcb8f5cbe14] [dmc:13852] [ 4] /opt/asn/apps/openmpi_1.8.3_intel/lib/openmpi/mca_btl_openib.so(mca_btl_openib_finalize+0x484)[0x7fcb8f5b0724] [dmc:13852] [ 5] /opt/asn/apps/openmpi_1.8.3_intel/lib/libmpi.so.1(+0x8c246)[0x7fcba4d9e246] [dmc:13852] [ 6] /opt/asn/apps/openmpi_1.8.3_intel/lib/libopen-pal.so.6(mca_base_framework_close+0x63)[0x7fcba3b14ac3] [dmc:13852] [ 7] /opt/asn/apps/openmpi_1.8.3_intel/lib/libopen-pal.so.6(mca_base_framework_close+0x63)[0x7fcba3b14ac3] [dmc:13852] [ 8] /opt/asn/apps/openmpi_1.8.3_intel/lib/libmpi.so.1(ompi_mpi_finalize+0x60e)[0x7fcba4d5e21e] [dmc:13852] [ 9] ./mm_mult_mpi_gnu[0x4036ba] [dmc:13852] [10] /lib64/libc.so.6(__libc_start_main+0xe6)[0x7fcba427bc36] [dmc:13852] [11] ./mm_mult_mpi_gnu[0x402859] [dmc:13852] *** end of error message *** 

and here program:

using namespace std; #include <iostream> #include <iomanip> #include <sstream> #include <stdlib.h> #include <string.h> #include <sys/time.h> #include "mpi.h" #include <cmath>  #define mx_sz 320 #define seed 2397           /* random number seed */ #define max_value  100.0    /* maximum size of array elements a, , b */  /* copied mpbench */ #define timer_clear     (tv1.tv_sec = tv1.tv_usec = tv2.tv_sec = tv2.tv_usec = 0) #define timer_start     gettimeofday(&tv1, (struct timezone*)0) #define timer_elapsed   ((tv2.tv_usec-tv1.tv_usec)+((tv2.tv_sec-tv1.tv_sec)*1000000)) #define timer_stop      gettimeofday(&tv2, (struct timezone*)0) struct timeval tv1,tv2;  /* declaration facilitates creation of 2 dimensional  dynamically allocated arrays (i.e. lxm array, mxn b array, , lxn c array).  allows pointer arithmetic  applied single data stream can dynamically allocated. address element @ row x, , column y use following notation:  a(x,y),b(x,y), or c(x,y), respectively. note differs normal c notation if 2 dimensional array of a[x][y] still descriptive of data structure. */ float *a,*b,*c,*rbuf,*sbuf; #define a(i,j) *(a+i*dim_m+j) #define b(i,j) *(b+i*dim_n+j) #define c(i,j) *(c+i*dim_n+j) #define rbuf(i,j) *(rbuf+i*dim_m+j) #define sbuf(i,j) *(sbuf+i*dim_n+j)   /*    routine retrieve data size of numbers array     command line or prompting user information */ void get_index_size(int argc,char *argv[],int *dim_l,int *dim_m,int *dim_n) {    if(argc!=2 && argc!=4) {       cout<<"usage:  mm_mult_serial [l_dimension] <m_dimension n_dimmension>"            << endl;       exit(1);    }    else {       if (argc == 2) {          *dim_l = *dim_n = *dim_m = atoi(argv[1]);       }       else {          *dim_l = atoi(argv[1]);          *dim_m = atoi(argv[2]);          *dim_n = atoi(argv[3]);       }    }    if (*dim_l<=0 || *dim_n<=0 || *dim_m<=0) {       cout<<"error: number of rows and/or columns must greater 0"           << endl;       exit(1);    } }  /*    routine fills number matrix random data values    between 0 , max_value    simulates in way might happen if there     single sequential data acquisition source such single file */ void fill_matrix(float *array,int dim_m,int dim_n) {    int i,j;    for(i=0;i<dim_m;i++) {       (j=0;j<dim_n;j++) {          array[i*dim_n+j]=drand48()*max_value;       }    } }  /*    routine outputs matrices screen  */ void print_matrix(float *array,int dim_m,int dim_n) {    int i,j;    for(i=0;i<dim_m;i++) {       (j=0;j<dim_n;j++) {          cout << array[i*dim_n+j] << " ";       }       cout << endl;    } }  /*    main routine: summation of number list */  int main( int argc, char *argv[]) {    float dot_prod;    int dim_l,dim_n,dim_m;    int i,j,k;    int id,p;    int row_per_proc;    int remain_rows;    int error;    mpi::status status;    mpi::init(argc, argv);     id = mpi::comm_world.get_rank();    p = mpi::comm_world.get_size();     if ( id == 0 ) {       /*        matrix sizes       */       get_index_size(argc,argv,&dim_l,&dim_m,&dim_n);        row_per_proc = ceil(float(dim_l)/p);       if (dim_l%p > 0) {          remain_rows = dim_l - (row_per_proc*(p-1));       } else {          remain_rows = 0;       }    }     // allocate space b in processes since shared    b = new (nothrow) float[dim_m*dim_n];    if (b==0) {       cout << "error: insufficient memory" << endl;       mpi_finalize();       exit(1);    }     if (id == 0) {       // allocate heap space , c       = new (nothrow) float[dim_l*dim_m];       c = new (nothrow) float[row_per_proc*p*dim_n];       if (a==0 || c==0) {          cout << "error: insufficient memory" << endl;          mpi_finalize();          exit(1);       }        /*          initialize numbers matrix random data       */        srand48(seed);       fill_matrix(a,dim_l,dim_m);       fill_matrix(b,dim_m,dim_n);        /*         output numbers matrix       */       cout << "a matrix =" << endl;       print_matrix(a,dim_l,dim_m);       cout << endl;        cout << "b matrix =" << endl;       print_matrix(b,dim_m,dim_n);       cout << endl;    }    /*      broadcast variables    */    mpi_bcast(&p, 1, mpi_int, 0, mpi_comm_world);    mpi_bcast(&dim_l, 1, mpi_int, 0, mpi_comm_world);    mpi_bcast(&dim_m, 1, mpi_int, 0, mpi_comm_world);    mpi_bcast(&dim_n, 1, mpi_int, 0, mpi_comm_world);    mpi_bcast(&row_per_proc, 1, mpi_int, 0, mpi_comm_world);    mpi_bcast(&remain_rows, 1, mpi_int, 0, mpi_comm_world);     /*       process 0 sends rows workers    */    rbuf = new float[row_per_proc*dim_m];    mpi_scatter(a, row_per_proc*dim_m, mpi_float, rbuf, row_per_proc*dim_m, mpi_float, 0, mpi_comm_world);     /*      send matrix b other processes    */    mpi_bcast(b, dim_m*dim_n, mpi_float, 0, mpi_comm_world);     /*    start recording execution time    */    if (id == 0) {       timer_clear;       timer_start;    }     // multiply local part of matrix    sbuf = new float[row_per_proc*dim_n];    (i=0; i<row_per_proc; i++) {       (j=0; j<dim_n; j++) {          dot_prod = 0.0;          (k=0; k<dim_m; k++) {             dot_prod += rbuf(i,k)*b(k,j);          }          sbuf(i,j) = dot_prod;       }    }     // send calculated values process 0    error = mpi_gather(sbuf, row_per_proc*dim_n, mpi_float, c, row_per_proc*dim_n, mpi_float, 0, mpi_comm_world);    cout << "gather returned: " << error << endl;     if (id == 0) {       /*          stop recording execution time       */        timer_stop;        cout << "c matrix =" << endl;       print_matrix(c,dim_l,dim_n);       cout << endl;       cout << "time=" << setprecision(8) <<  timer_elapsed/1000000.0  << " seconds" << endl;    }     // return allocated memory    delete a; delete b; delete c; delete rbuf; delete sbuf;    error = mpi_finalize();    cout << "finalize returned: " << error << endl; } 

you creating a , c on rank 0:

   if (id == 0) {       // allocate heap space , c       = new (nothrow) float[dim_l*dim_m];       c = new (nothrow) float[row_per_proc*p*dim_n]; 

therefore, should delete them on rank 0:

   if (id == 0) {       /*          stop recording execution time       */        timer_stop;        cout << "c matrix =" << endl;       print_matrix(c,dim_l,dim_n);       cout << endl;       cout << "time=" << setprecision(8) <<  timer_elapsed/1000000.0  << " seconds" << endl;       delete a; delete c;    }     // return allocated memory    delete b; delete rbuf; delete sbuf; 

additional note: calling mpi_finalize if allocation on rank 0 fails, better call mpi_abort in case, or broadcast result of check, notify processes of failure , allow proper termination. code right now, program might hang infinitely when running error.