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

//put numbers 0 to N into an array
//sum numers from 2 arrs into 3rd
#define N 12//note this is 12 and not 10
long thread_elems = N/4;

/* A structure we'll use to pass arguments to our thread function. */
struct thread_arg {
//sum numbers from 2 arrs into 3rd
   long arr1[N];
   long arr2[N];
   long arr3[N];
   int id;
};

void *sum(void *varg){
    struct thread_arg chunk= *((struct thread_arg*)varg);
    long start = (chunk.id)*thread_elems;
    long end = start + thread_elems;

    long i,sum=0;
    for (i=start; i<end;i++){
        //store the int to our retArr
        //if end < N{
                printf("%d %d ", chunk.arr1[i], chunk.arr2[i]);
        chunk.arr3[i] = chunk.arr1[i] + chunk.arr2[i];
                (*((struct thread_arg*)varg)).arr3[i] = (*((struct thread_arg*)varg)).arr1[i] + (*((struct thread_arg*)varg)).arr2[i];
        printf("  %d  ", chunk.arr3[i]);
                //}
    }
        printf("\n");
}

int main(){
    pthread_t tid[4];
//  ids[4];

    long ret_arr[N];

    struct thread_arg *arguments = malloc(4 * sizeof(struct thread_arg));
    //build arrs in struct from arrays in HOST memory
    int i, j;
    for (i=0; i<4; i++){
        for(j=0;j<N;j++){
            (arguments[i]).arr1[j]= j;
            (arguments[i]).arr2[j]= j;
                        printf("%d %d ", (arguments[i]).arr1[j], (arguments[i]).arr2[j]);
        }
    }

       printf("\n");

    for (i=0; i<4; i++){
        (arguments[i]).id = i;
        pthread_create(&tid[i], NULL, sum, (void*)&arguments[i]);
    }
    //don't destroy threads until they are all done!
    for (i=0;i<4;i++){
        pthread_join(tid[i], NULL);
    }


                for(j=0;j<N;j++){
                        printf("  %d  ", (arguments[0]).arr3[j] );
                }


    //get the results into a single arr
    //arg arrays are disjoint and don't share mem!
    long t, start, end;
    for (i=0;i<4;i++){
        start = i*thread_elems;
        end = start + thread_elems;
        for(t=start; t<end; t++){
            ret_arr[t] = (arguments[i]).arr3[t];
        }
    }

    for (i=0;i<N;i++){
        printf("%d ", ret_arr[i]);
    }
    printf("\n");
    return 0;
}