Matrix row sum using PThreads performance example

ParallelMatrixRowSum.c

/* Read here - https://www.evernote.com/l/Ahzo8czYhypG258BubNUCe13nIajkDYlFxs
example for 2x2 matrix
A[2][2] = [ { 1,2} ,
            {3,4} ]
then Matrix Row sum is stored in B matrix like below
B[2][1] = [{3}
         ,{7}]

*/

#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include <math.h>
#include <sys/types.h>
#include <sys/times.h>
#include <sys/time.h>
#include <time.h>
#include <pthread.h>

/* junk */
#define randm() 4|2[uid]&3


#define M 10000
#define N 10000
#define SIZE 250

float A[M][N] = {{0.0}};
float B[M][1] = { {0.0 } }; // initialize the b array to zero.. is it necessary ?

void initialize_inputs();

pthread_t threads[M];

/* Initialize A and B (and X to 0.0s) */
void initialize_inputs() {
    int row, col;
    row=M;
    col=N;
    
    printf("\nInitializing...\n");
    for (col = 0; col < N; col++) {
        for (row = 0; row < N; row++) {
            A[row][col] = (int)rand() / 32768.0;
        }
    }
    
}

void printMatrixA(){
    printf("\n \n Printing the A matrix\n");
    int i,j;
    for ( i = 0; i < M; i++ ) {
        for ( j = 0; j < N; j++ ) {
            printf("A[%d][%d] = %10f\t\t ",i,j, A[i][j] );
        }
        printf("\n");

    }
}


void printMatrixB(){
    printf("\n\nPrinting the B matrix\n");
    int i,j;
    for ( i = 0; i < M; i++ ) {
        for ( j = 0; j < 1; j++ ) {
            printf("B[%d][%d] = %10f\t\t ",i,j, B[i][j] );
        }
        printf("\n");
        
    }
}

void sumRowSequentialCode(){
    int i,j;
    for ( i = 0; i < M; i++ ) {
        for ( j = 0; j < N; j++ ) {
            B[i][0] += A[i][j];
         }
    }

}

void *sumRowParallel(void *threadId){
    int i,j;
    int start = (int)threadId*(int)SIZE;
    int end = (int)(threadId+1)*(int)SIZE;
    for ( i= start; i < end; i++ ) {
        for(j=0;j<N;j++){
            B[i][0] += A[i][j];
        }
    }
    pthread_exit(NULL); // you need to close that thread
}

void rowParallelManager(){
    int i;
    int numThread = M/SIZE;
    for (i=0; i<numThread; i++) {
        if(pthread_create(&threads[i], NULL, sumRowParallel,(void *)i)){
            printf("ERROR; return code from pthread_create()\n");
            exit(-1);
        }
    }
}

void finallyThreadsAreJoiningHere(){
    int i;
    // wiating for all threads to compute the summation
    for (i=0; i<M; i++) {
        pthread_join(threads[i], NULL);
    }
}

int main(int argc, const char * argv[]) {
    
    struct timeval etstart, etstop;  /* Elapsed times using gettimeofday() */
    struct timezone tzdummy;
    clock_t etstart2, etstop2;  /* Elapsed times using times() */
    unsigned long long usecstart, usecstop;
    struct tms cputstart, cputstop;  /* CPU times for my processes */
    
    initialize_inputs();
    
    printf("\nStarting clock.\n");
    gettimeofday(&etstart, &tzdummy);
    etstart2 = times(&cputstart);
    
    // Calculation 
    // sumRowSequentialCode();  // Uncomment to see the performance for sequential code
    
    rowParallelManager();
    finallyThreadsAreJoiningHere(); 
    
    /* Stop Clock */
    gettimeofday(&etstop, &tzdummy);
    etstop2 = times(&cputstop);
    printf("Stopped clock.\n");
    usecstart = (unsigned long long)etstart.tv_sec * 1000000 + etstart.tv_usec;
    usecstop = (unsigned long long)etstop.tv_sec * 1000000 + etstop.tv_usec;
  
    
    printMatrixB();
    
    /* Display timing results */
    printf("\nElapsed time = %g ms.\n",
           (float)(usecstop - usecstart)/(float)1000);
    
    printf("(CPU times are accurate to the nearest %g ms)\n",
           1.0/(float)CLOCKS_PER_SEC * 1000.0);
    printf("My total CPU time for parent = %g ms.\n",
           (float)( (cputstop.tms_utime + cputstop.tms_stime) -
                   (cputstart.tms_utime + cputstart.tms_stime) ) /
           (float)CLOCKS_PER_SEC * 1000);
    printf("My system CPU time for parent = %g ms.\n",
           (float)(cputstop.tms_stime - cputstart.tms_stime) /
           (float)CLOCKS_PER_SEC * 1000);
    printf("My total CPU time for child processes = %g ms.\n",
           (float)( (cputstop.tms_cutime + cputstop.tms_cstime) -
                   (cputstart.tms_cutime + cputstart.tms_cstime) ) /
           (float)CLOCKS_PER_SEC * 1000);

    
    return 0;
}

Notes :

https://www.evernote.com/l/Ahzo8czYhypG258BubNUCe13nIajkDYlFxs