Thread Sincronization PRoblem (Mandlebrot)

This is a discussion on Thread Sincronization PRoblem (Mandlebrot) within the C Programming forums, part of the General Programming Boards category; Hello i am trying to create mandlebrot images using threads ... i have managed to create them using only one ...

  1. #1
    Registered User
    Join Date
    Mar 2012
    Posts
    20

    Thread Sincronization PRoblem (Mandlebrot)

    Hello i am trying to create mandlebrot images using threads ... i have managed to create them using only one process now i want to create them using threads... i have managed to do this .. but i have not managed to synchronize the threads. I need some help in making each new thread wait for the thread before to put something in the buffer before running.

    Here is the code:

    Code:
    #include "recursos.h"
    double *b;
    pthread_mutex_t mut;
    int max_iterations = 256;
    
    
    /* This function (compute_point) is the algorithmic core of this program,
       computing what could be called the Mandelbrot function of the complex
       number (cr, ci).
    */
    
    double compute_point(double ci, double cr) {
      int iterations = 0;
      double zi = 0;
      double zr = 0;
      while ((zr*zr + zi*zi < 4) && (iterations < max_iterations)) {
        double nr, ni; 
    
        /* Z <-- Z^2 + C */
       
        nr = zr*zr - zi*zi + cr;
        ni = 2*zr*zi + ci;
        
        zi = ni;
        zr = nr;
    
        iterations ++;
      }  
      return iterations;  
    }
    
    /* The "compute" function computes the Mandelbrot function over every
       point on a grid that is "nx" points wide by "ny" points tall, where
       (xmin,ymin) and (xmax,ymax) give two corners of the region the 
       complex plane.
    */
    
    void compute(double *buffer, int nx, int ny, double xmin, double xmax, 
             double ymin, double ymax) {
      double delta_x, delta_y;
      int x, y;
    
      delta_x = (xmax - xmin)/nx;
      delta_y = (ymax - ymin)/ny;
    
      for (y=0;  y<ny; y++) {
        double y_value = ymin + delta_y * y;
        for (x=0; x<nx; x++) {
          double x_value = xmin + delta_x * x;
          buffer[y*nx + x] = compute_point(x_value, y_value);
        }
      }
    }
    
    /* Output the data contained in the buffer to a Portable Greymap format
       image file.  The parameter "max" should be an upper bound for the
       data values in the buffer. 
    */
    
    void output_pgm(char *filename,double *buffer, int nx, int ny, double max) {
      int i;
      FILE *file;
      file = fopen(filename,"w");
      fprintf(file,"P2\n");
      fprintf(file,"%d %d\n",nx,ny);
      fprintf(file,"%d\n",(int)max);
      for (i=0; i<nx*ny; i++) {
        if (!(i%nx)) fprintf(file,"\n");
        fprintf(file,"%d ",(int)buffer[i]);
      }
      fclose(file);
    }
     
    void *computeFunc(void *args) {
       //  printf("TestChegar\n");
         struct threadArgs *targs;
         targs = (struct threadArgs *)args;
         int n = targs->nFatias;
         double yM = targs->yMin;
         printf("%f\n",yM);
         double multiplier = ((1.0/n) * 2);
         pthread_mutex_lock(&mut);
         compute(b,1000,(int)1000/n,-1,1,yM,(yM+multiplier));
         char str[80]; 
         char c[10]; 
         sprintf(c,"%d",(targs->indice+1)); 
         strcpy(str,"mandel-"); 
         strcat(str,c); 
         strcat(str,".pgm"); 
         output_pgm(str,b,1000,(1000/n),255);
         b = b + (1000/n)*1000;
         pthread_mutex_unlock(&mut); 
    }
    
    int main(int argc, char *argv[]) {
        if(argc != 2) {
        
        printf("Introduziu o formato errado para o programa \n O formato correcto  : ./nomeprog <nrworkers>\n");
    
        } else {    
    
        pthread_mutex_init(&mut,NULL);
        int nWorkers = atoi(argv[1]);        
        int i = 0;
        pthread_t threads[nWorkers];
    
        struct threadArgs targs[nWorkers];
        
        double *ptr;
        b= ptr = malloc(1000*1000*sizeof(double));
    
        double aux = -1.0;
        double multiplier = ((1.0/nWorkers) * 2);
        for(i = 0; i < nWorkers;i++) {
             targs[i].nFatias = nWorkers;
            targs[i].yMin = aux;
            targs[i].b = b;
            targs[i].indice = i;    
            pthread_create(&threads[i],NULL,computeFunc,(void *)&targs[i]);
            aux = aux + multiplier;
        }
        pthread_join(threads[nWorkers - 1],NULL);
        output_pgm( "mandel.pgm", ptr, 1000, 1000, 255);
    
        }
    
        
    }
    And the header file:

    Code:
    #include <stdio.h>
    #include <pthread.h>
    #include <string.h>
    #include <stdlib.h>
    
    struct threadArgs{
       int nFatias;
       double yMin;
       int indice;
       double *b;
    };

  2. #2
    - - - - - - - - oogabooga's Avatar
    Join Date
    Jan 2008
    Posts
    2,808
    What's the point of the mutexes here? Don't you want these threads all working at once?
    The cost of software maintenance increases with the square of the programmer's creativity. - Robert D. Bliss

  3. #3
    and the hat of int overfl Salem's Avatar
    Join Date
    Aug 2001
    Location
    The edge of the known universe
    Posts
    32,851
    > b= ptr = malloc(1000*1000*sizeof(double));
    Perhaps you should allocate separate memory for each thread, for the image size they work on.

    Because at the moment, all the threads fight over the SAME block of global memory.

    If you allocate what you need for each thread, AND make this bit of code use the targs->b member (and NOT the global), then you might get somewhere.
    Code:
         compute(b,1000,(int)1000/n,-1,1,yM,(yM+multiplier));
         char str[80];
         char c[10];
         sprintf(c,"%d",(targs->indice+1));
         strcpy(str,"mandel-");
         strcat(str,c);
         strcat(str,".pgm");
         output_pgm(str,b,1000,(1000/n),255);
         b = b + (1000/n)*1000;  //!! WTF!!!
    Delete the global variable, then compile and fix what it wrong.
    If you dance barefoot on the broken glass of undefined behaviour, you've got to expect the occasional cut.
    If at first you don't succeed, try writing your phone number on the exam paper.
    I support http://www.ukip.org/ as the first necessary step to a free Europe.

  4. #4
    Registered User
    Join Date
    Mar 2012
    Posts
    20
    Oh yes .. i understand ..
    the problem is that .. i would need that global variable because the main picture would have to contain the whole drawing
    is there no way of synchronizing the threads so that all of them finish what they have to do then wait for the thread before to write to the buffer before writing themselves
    the mutex are just to ensure mutual exclusion , i don't know if they can be ommited.

  5. #5
    and the hat of int overfl Salem's Avatar
    Join Date
    Aug 2001
    Location
    The edge of the known universe
    Posts
    32,851
    So do you want one global "picture" and each thread works on a part of the picture?

    Then after ALL the threads have passed join, you output the image from main?
    pthread_join(threads[nWorkers - 1],NULL); //!! join ALL of them
    output_pgm( "mandel.pgm", ptr, 1000, 1000, 255);

    Sure, that would work, but you need to make sure each thread ONLY operates on the part of the image it's supposed to.
    If you dance barefoot on the broken glass of undefined behaviour, you've got to expect the occasional cut.
    If at first you don't succeed, try writing your phone number on the exam paper.
    I support http://www.ukip.org/ as the first necessary step to a free Europe.

  6. #6
    Registered User
    Join Date
    Mar 2012
    Posts
    20
    ohhh thank you .. you helped me get to the solution ..
    it's exactly what you said ..
    I have to join each thread..
    and yes each thread works on a part of the image and then we output evertything on the main ...
    the outputs on the threadfunction are just tests ...


    Thank you!

  7. #7
    Registered User
    Join Date
    Mar 2012
    Posts
    20
    Here is the Working code

    Code:
    #include "recursos.h"
    double *b;
    pthread_mutex_t mut;
    int max_iterations = 256;
    
    
    /* This function (compute_point) is the algorithmic core of this program,
       computing what could be called the Mandelbrot function of the complex
       number (cr, ci).
    */
    
    double compute_point(double ci, double cr) {
      int iterations = 0;
      double zi = 0;
      double zr = 0;
      while ((zr*zr + zi*zi < 4) && (iterations < max_iterations)) {
        double nr, ni; 
    
        /* Z <-- Z^2 + C */
       
        nr = zr*zr - zi*zi + cr;
        ni = 2*zr*zi + ci;
        
        zi = ni;
        zr = nr;
    
        iterations ++;
      }  
      return iterations;  
    }
    
    /* The "compute" function computes the Mandelbrot function over every
       point on a grid that is "nx" points wide by "ny" points tall, where
       (xmin,ymin) and (xmax,ymax) give two corners of the region the 
       complex plane.
    */
    
    void compute(double *buffer, int nx, int ny, double xmin, double xmax, 
             double ymin, double ymax) {
      double delta_x, delta_y;
      int x, y;
    
      delta_x = (xmax - xmin)/nx;
      delta_y = (ymax - ymin)/ny;
    
      for (y=0;  y<ny; y++) {
        double y_value = ymin + delta_y * y;
        for (x=0; x<nx; x++) {
          double x_value = xmin + delta_x * x;
          buffer[y*nx + x] = compute_point(x_value, y_value);
        }
      }
    }
    
    /* Output the data contained in the buffer to a Portable Greymap format
       image file.  The parameter "max" should be an upper bound for the
       data values in the buffer. 
    */
    
    void output_pgm(char *filename,double *buffer, int nx, int ny, double max) {
      int i;
      FILE *file;
      file = fopen(filename,"w");
      fprintf(file,"P2\n");
      fprintf(file,"%d %d\n",nx,ny);
      fprintf(file,"%d\n",(int)max);
      for (i=0; i<nx*ny; i++) {
        if (!(i%nx)) fprintf(file,"\n");
        fprintf(file,"%d ",(int)buffer[i]);
      }
      fclose(file);
    }
     
    void *computeFunc(void *args) {
       //  printf("TestChegar\n");
         struct threadArgs *targs;
         targs = (struct threadArgs *)args;
         int n = targs->nFatias;
         double yM = targs->yMin;
         printf("%f\n",yM);
         double multiplier = ((1.0/n) * 2);
         pthread_mutex_lock(&mut);
         compute(b,1000,(int)1000/n,-1,1,yM,(yM+multiplier));
     /*  
         char str[80]; 
         char c[10]; 
         sprintf(c,"%d",(targs->indice+1)); 
         strcpy(str,"mandel-"); 
         strcat(str,c); 
         strcat(str,".pgm"); 
         output_pgm(str,b,1000,(1000/n),255);
    */
         b = b + (1000/n)*1000;
         pthread_mutex_unlock(&mut); 
    }
    
    int main(int argc, char *argv[]) {
        if(argc != 2) {
        
        printf("Introduziu o formato errado para o programa \n O formato correcto  : ./nomeprog <nrworkers>\n");
    
        } else {    
    
        pthread_mutex_init(&mut,NULL);
        int nWorkers = atoi(argv[1]);        
        int i = 0;
        pthread_t threads[nWorkers];
    
        struct threadArgs targs[nWorkers];
        
        double *ptr;
        b= ptr = malloc(1000*1000*sizeof(double));
    
        double aux = -1.0;
        double multiplier = ((1.0/nWorkers) * 2);
        for(i = 0; i < nWorkers;i++) {
             targs[i].nFatias = nWorkers;
            targs[i].yMin = aux;
            targs[i].b = b;
            targs[i].indice = i;    
            pthread_create(&threads[i],NULL,computeFunc,(void *)&targs[i]);
            aux = aux + multiplier;
            pthread_join(threads[i],NULL);
        }
        output_pgm( "mandel.pgm", ptr, 1000, 1000, 255);
    
        }
    
        
    }

  8. #8
    and the hat of int overfl Salem's Avatar
    Join Date
    Aug 2001
    Location
    The edge of the known universe
    Posts
    32,851
    You're still running threads serially with that mutex in place.

    Plus you're still messing with a global variable.

    Each thread should have just ONE pointer to one bit of exclusive memory, and then you can run the threads in parallel.
    If you dance barefoot on the broken glass of undefined behaviour, you've got to expect the occasional cut.
    If at first you don't succeed, try writing your phone number on the exam paper.
    I support http://www.ukip.org/ as the first necessary step to a free Europe.

  9. #9
    - - - - - - - - oogabooga's Avatar
    Join Date
    Jan 2008
    Posts
    2,808
    Quote Originally Posted by Salem View Post
    You're still running threads serially with that mutex in place.
    And even if you removed them from the above code, the way you use pthread_join ALSO defeats the whole purpose of the exercise. Move it out of that loop into a loop of its own. And be sure to inline compute_point to avoid the call overhead of one million calls.
    The cost of software maintenance increases with the square of the programmer's creativity. - Robert D. Bliss

Popular pages Recent additions subscribe to a feed

Similar Threads

  1. Thread problem?
    By bungkai in forum C Programming
    Replies: 10
    Last Post: 02-12-2012, 08:19 AM
  2. Replies: 3
    Last Post: 11-20-2011, 12:01 AM
  3. Yet another newbie problem thread
    By High Overlord in forum C++ Programming
    Replies: 2
    Last Post: 03-05-2011, 09:39 AM
  4. problem with creating a thread
    By neeharikach in forum C Programming
    Replies: 1
    Last Post: 11-28-2010, 03:46 AM
  5. WINAPI thread problem
    By samguddy in forum Windows Programming
    Replies: 1
    Last Post: 11-13-2009, 03:02 PM

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21