如何在多线程矩阵乘法中使用线程池和消息队列？

Question

I'm trying to learn multithreading by doing a multithreaded matrix multiplication program.I'm computing one row at a time. I am facing problem when I am using fewer threads than the rows. I read lot of similar posts but could not understand clearly how can I reuse them. But there are two possible methods.

1. Using a thread pool and making a task queue- I did not understand after completion of task , how the next task is assigned to a particular thread among pool of threads

2. message queues.

How do I use mutex lock on shared variable sum?

Please suggest me possible changes I should add in following program.

#include <pthread.h>
#include <stdio.h>
#include <stdlib.h>

#define M 6
#define K 7
#define N 8
#define NUM_THREADS 4

int A[M][K] = { { 1, 4, 8, 4, 5, 6, 2 }, { 7, 3, 2, 4, 1, 4, 5 }, { 2, 3, 9, 4,
        7, 1, 5 }, { 4, 3, 9, 4, 7, 2, 5 }, { 1, 3, 9, 9, 7, 1, 3 }, { 2, 4, 9,
        3, 7, 1, 5 } };
int B[K][N] = { { 8, 3, 8, 4, 5, 6, 2, 3 }, { 1, 3, 2, 2, 3, 4, 8, 1 }, { 8, 3,
        9, 1, 7, 1, 5, 2 }, { 1, 3, 9, 2, 7, 2, 5, 2 },
        { 1, 3, 9, 2, 7, 1, 3, 3 }, { 2, 4, 9, 3, 7, 1, 5, 2 }, { 2, 4, 9, 3, 7,
                1, 5, 2 } };
int C[M][N];

struct v {
    int i; /* row */
    int j; /* column */
};

void *runner(void *param); /* the thread */

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

    int i, j, count = 0;
    for (i = 0; i < NUM_THREADS; i++) {

        //Assign a row and column for each thread
        struct v *data = (struct v *) malloc(sizeof(struct v));
        data->i = i;
        data->j = j;
        /* Now create the thread passing it data as a parameter */
        pthread_t tid[NUM_THREADS];       //Thread ID
        pthread_attr_t attr; //Set of thread attributes
        //Get the default attributes
        pthread_attr_init(&attr);
        //Create the thread
        pthread_create(&tid, &attr, runner, data);
        //Make sure the parent waits for all thread to complete
        pthread_join(tid, NULL );
        count++;

    }

    //Print out the resulting matrix
    for (i = 0; i < M; i++) {
        for (j = 0; j < N; j++) {
            printf("%d ", C[i][j]);
        }
        printf("\n");
    }
}

//The thread will begin control in this function
void *runner(void *param) {
    struct v *data = param; // the structure that holds our data
    int n, x, sum = 0; //the counter and sum

    //Calculating one row

    for (x = 0; x < N; x++) {
        for (n = 0; n < K; n++)

        {
            sum += A[data->i][n] * B[n][x];
        }
        //assign the sum to its coordinate
        C[data->i][data->j] = sum;
    }
    //Exit the thread
    pthread_exit(0); // How can I reuse this thread for another row instead of exiting it ?
}

Any help will be appreciated!

Answer 1

How do I use mutex lock on shared variable sum?

You need to declare a mutex and use it to lock the sum before use sum ,and unlock it after use sum. Do that every time you use sum to protect it.

Here is a example :

pthread_mutex_t lock = PTHREAD_MUTEX_INITIALIZER;

pthread_mutex_lock(&lock);

use sum ....

pthread_mutex_unlock(&lock);

Answer 2

You can reuse a pthread attribute to create more then one thread. Also it needs to be destroyed to free any resources used.

Also to allow the threads to run in parallel first create them all and then join them:

  ... 

  pthread_attr_t attr; //Set of thread attributes
  //Get the default attributes
  pthread_attr_init(&attr);
  pthread_t tids[NUM_THREADS] = {0};
  for (i = 0; i < NUM_THREADS; i++) {

    //Assign a row and column for each thread
    struct v *data = (struct v *) malloc(sizeof(struct v));
    data->i = i;
    data->j = j;

    //Create the thread
    pthread_create(&tids[i], &attr, runner, data);
    //Make sure the parent waits for all thread to complete
    count++;
  }
  pthread_attr_destroy(&attr);

  for (i = 0; i < NUM_THREADS; i++) {
    pthread_join(tids[i], NULL);
  }

  ...

As a final note: Error checking is ommitted in the example lines above. But for any "real" code always check the outcome of a system call.