multithreading - tbb matrix mulitiplication stack overflow error c++ -

i'm trying matrix multiplication using task in intel tbb, algorithm i'm using strassen's algorithm...

here code main():

#include "matrix.h" #include "tbb/tick_count.h"  using namespace tbb; using namespace std;  //here how call mattask class matrica callparallel(matrx& a, matrix& b, matrix& c, int n){      mattask& t = *new (task::allocate_root ()) mattask (a, b, &c, n);      task::spawn_root_and_wait (t);      return c; }   int main(){     int rows, columns;     matrix serialc;      cout << "*******************\n" << "if rows , columns < 6 enter matric manualy\n" << "********************\n" <<endl;       cout << "enter rows matrix a: ";     cin >> rows;     cout << "enter columns matrix a: ";     cin >> columns;      matrix a(rows, columns);      if(rows > 5 && columns > 5){         a.creatematrixautomatic();     }else {         a.creatematricmanualy();     }      cout << "enter rows matrix b: ";     cin >> rows;     cout << "enter columns matrix b: ";     cin >> columns;      matrix b(rows, columns);     if(rows > 5 && columns > 5){         b.creatematrixautomatic();     }else {         b.creatematricmanualy();     }      cout << "matrix a: " << endl;     a.printmatrix();     cout << "matrix b: " << endl;     b.printmatrix();      cout << "matrix c: " << endl;      tick_count start_time = tick_count::now();     serialc.multserial(a, b);     tick_count end_time = tick_count::now();      cout << "\ntime serial: " << (end_time - start_time).seconds() * 1000 << " ms" << endl;       serialc.printmatrix();      //creating matrix result , calling parallel algorithm     matrix parallelc(rows, columns);     parallelc = callparallel(a, b, parallelc, rows);      //this here prints result matrix     parallelc.printmatrix();      system("pause"); } 

here matrix.cpp code:

#include "matrix.h"   matrix::matrix(){}  matrix::matrix(int rows, int columns){     vr = rows;     kol = columns; }  void matrix::creatematrixautomatic(){     for(int = 0; < vr; i++){         (int j = 0; j < kol; j++){             int number = rand() % 5 + 1;             matr[i][j] = number;         }     } }  void matrix::creatematricmanualy(){     cout << "enter elements: " << endl;     for(int = 0; < vr; i++){         (int j = 0; j < kol; j++){             cout << "enter [" << << "]" << "[" << j << "] element: ";             cin >> matr[i][j];         }     } }  void matrix::printmatrix(){     (int = 0; < vr; i++){         (int j = 0; j < kol; j++){             cout << matr[i][j] << "  ";         }         cout << endl << endl;     } }  void matrix::multserial(matrix& a, matrix& b){     for(int = 0; < a.vr; i++){         for(int j = 0; j < b.kol; j++){             matr[i][j] = 0;             for(int k = 0; k < b.vr; k++){                 matr[i][j] += (a.matr[i][k] * b.matr[k][j]);                 vr = a.vr;                 kol = b.kol;             }         }     } }  void matrix::substract(matrix& a, matrix& b, int dim){     (int = 0; < dim; i++) {         (int j = 0; j < dim; j++) {             matr[i][j] = a.matr[i][j] - b.matr[i][j];         }     }   }  void matrix::add(matrix& a, matrix& b, int dim){     (int = 0; < dim; i++) {         (int j = 0; j < dim; j++) {             matr[i][j] = a.matr[i][j] + b.matr[i][j];         }     } } 

and here mattask class , matrica.h class

    #pragma once      #include <iostream>     #include <tbb/task.h>      using namespace tbb;     using namespace std;      class matrix{  public:       int vr, kol;      int matr[100][100];     matrix();     matrix(int rows, int columns);      void creatematrixautomatic();     void creatematricmanualy();     void printmatrix();      void multserial(matrix&, matrix&);      void add(matrix& a, matrix& b, int dim);      void substract(matrix& a, matrix& b, int dim); };  class mattask: public task{ public:     matrix a, b;     matrix* c;     int dimension;      mattask(matrix& _a, matrix& _b, matrix* _c, int dim):     a(_a), b(_b), c(_c), dimension(dim){}      task* execute(){         if(dimension == 1){             c->multserial(a, b);         } else {              int newdimension = dimension/2;              task_list list;              int count = 1;             matrica a11(newdimension, newdimension), a12(newdimension, newdimension), a21(newdimension, newdimension), a22(newdimension, newdimension),                     b11(newdimension, newdimension), b12(newdimension, newdimension), b21(newdimension, newdimension), b22(newdimension, newdimension),                     *c11, *c12, *c21, *c22,                     p1(newdimension, newdimension), *p2, *p3, *p4, *p5, *p6, *p7,                     aresult(newdimension, newdimension), bresult(newdimension, newdimension);              //delimo matrice u 4 podmatrice             for(int = 0; < newdimension; i++){                 for(int j = 0; j < newdimension; j++){                     (a11).matr[i][j] = a.matr[i][j];                     (a12).matr[i][j] = a.matr[i][j + newdimension];                     (a21).matr[i][j] = a.matr[i + newdimension][j];                     (a22).matr[i][j] = a.matr[i + newdimension][j + newdimension];                      (b11).matr[i][j] = b.matr[i][j];                     (b12).matr[i][j] = b.matr[i][j + newdimension];                     (b21).matr[i][j] = b.matr[i + newdimension][j];                     (b22).matr[i][j] = b.matr[i + newdimension][j + newdimension];                 }             }              //racunamo p1...p7              //p1 = (a11 + a22) * (b11 + b22)             aresult.add(a11, a22, newdimension); //a11 + a22             bresult.add(b11, b22, newdimension); //b11 + b22             count++;             //mattask& = *new( allocate_child() ) mattask(aresult, bresult, &p1, newdimension);              //lista.push_back(a);             lista.push_back(*new (allocate_child()) mattask(aresult, bresult, &p1, newdimension));              //p2 = (a21 + a22) * b11             //aresult.add(a21, a22, newdimension); //a21 + a22             //count++;             ////lista.push_back(*new (allocate_child()) mattask(aresult, b11, p2, newdimension));              ////p3 = a11 * (b12 - b22)             //bresult.substract(b12, b22, newdimension); // b12 - b22             //count++;             ////lista.push_back(*new (allocate_child()) mattask(a11, bresult, p3, newdimension));              ////p4 = a22 * (b21 - b11)             //bresult.substract(b21, b11, newdimension); // b21 - b11             //count++;             ////lista.push_back(*new (allocate_child()) mattask(a22, bresult, p4, newdimension));              ////p5 = (a11 + a12) * b22             //aresult.add(a11, a12, newdimension); // a11 + a12             //count++;             ////lista.push_back(*new (allocate_child()) mattask(aresult, b22, p5, newdimension));              ////p6 = (a21 - a11) * (b11 + b12)             //bresult.add(b11, b12, newdimension); //b11 + b12             //aresult.substract(a21, a11, newdimension); //a21 - a11             //count++;             ////lista.push_back(*new (allocate_child()) mattask(aresult, bresult, p6, newdimension));              ////p7 = (a12 - a22) * (b21 + b22)             //bresult.add(b21, b22, newdimension); //b21 + b22             //aresult.substract(a12, a22, newdimension); //a12 - a22             //count++;             ////lista.push_back(*new (allocate_child()) mattask(aresult, bresult, p7, newdimension));               set_ref_count(count);             //spawn(a);             spawn_and_wait_for_all(list);             //spawn_and_wait_for_all(a);              //racunamo d11, d12, d21, d22               //c11 = p1 + p4 - p5 + p7             //aresult.add(p1, p4, newdimension); // p1 + p4             //bresult.add(aresult, p7, newdimension); // p1 + p4 + p7              //c11.oduzmi(bresult, p5, newdimension); // c11 = p1 + p4 + p7 - p5              //// c12 = p3 + p5             //c12.add(p3, p5, newdimension);             //             //// c21 = p2 + p4             //c21.add(p2, p4, newdimension);              //// c22 = p1 + p3 - p2 + p6             //aresult.add(p1, p3, newdimension); //p1 + p3             //bresult.add(aresult, p6, newdimension); //p1 + p3 + p6             //c22.substract(bresult, p2, newdimension); // c22 = p1 + p3 + p6 - p2              //grouping results obtained in single matrix:             //for (int = 0; < novadimenzija ; i++) {             //  (int j = 0 ; j < novadimenzija ; j++) {             //      c.matr[i][j] = c11.matr[i][j];             //      c.matr[i][j + newdimension] = c12.matr[i][j];             //      c.matr[i + newdimension][j] = c21.matr[i][j];             //      c.matr[i + newdimension][j + newdimension] = c22.matr[i][j];             //  }             //}         }         return null;     } }; 

as see names of functions , classes not in english, don't think problem, because code straightforward.

i error:

unhandled exception @ 0x01193787 in mnozenjematrica.exe: 0xc00000fd: stack overflow.

i think error occurres in line spawn_and_wait_for_all(lista), nut i'm not sure.

can please take @ code , me solve problem. maybe i'm not calling correctly functions, not know realy, please help. thank you

it blocking-style parallelism plus heavy use of stack matrices result in stack overflow. so, each task reserve stack data , calls spawn_root_and_wait_for_all in turn executes instance of same task recursively keeps growing stack.

use continuation-style programming , avoid allocating huge data on stack (and inside task if possible - reduces task allocator efficiency).