thsum.c sum vector in threads

// avp 2016 for http://ru.stackoverflow.com/questions/521709/%D0%9A%D0%B0%D0%BA-%D0%BC%D0%BE%D0%B6%D0%BD%D0%BE-%D1%83%D1%81%D0%BA%D0%BE%D1%80%D0%B8%D1%82%D1%8C-%D1%84%D1%83%D0%BD%D0%BA%D1%86%D0%B8%D1%8E-%D0%9F%D0%BE%D1%87%D0%B5%D0%BC%D1%83-%D1%82%D0%BE-%D1%80%D0%B0%D0%B1%D0%BE%D1%82%D0%B0%D0%B5%D1%82-%D1%82%D0%BE%D0%BB%D1%8C%D0%BA%D0%BE-%D0%BE%D0%B4%D0%BD%D0%BE-%D1%8F%D0%B4%D1%80%D0%BE

// gcc thsum.c -pthread -O3
// ./a.out [N-threads]

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

int sum (int *v, int s, int e) {
  int su = 0;

  for (; s < e; s++)
    su += v[s];

  return su;
}

struct req {
  int *v, s, e;
};

void *
call_sum (void *a)
{
  struct req *r = (__typeof__(r))a;
  int s = sum(r->v, r->s, r->e);
  free(a);
  return (void *)(long)s;
}


int handle (int *v, int sz, void * f(void *), int n)
{
  pthread_t th[n];
  int answers[n];
  int mid = sz / n,
    i, start = 0, end = start + mid;

  for (i = 0; i < n; i++) {
    start = mid * i;
    end = start + mid;
    if (end != sz && sz - end < mid)
      end = sz;
    struct req  *req = malloc(sizeof(*req));
    req->v = v; req->s = start; req->e = end;
    pthread_create(th + i, 0, f, req);
  }
  for (i = 0; i < n; i++) {
    void *r;
    pthread_join(th[i], &r);
    answers[i] = (long)r;
  }

  return sum(answers, 0, n);
}

void print (const char *nm, int s, clock_t t, int n)
{
  printf("%s: %d %ld clocks (%g msec)  (%d threads)\n",
     nm, s, (long)t, ((double)t / CLOCKS_PER_SEC) * 1000, n);
}

int main (int ac, char *av[])
{
  int N = 1000021, i;
  int *v = malloc(sizeof(int) * N);
  for (i = 0; i < N; i++)
    v[i] = random() % 10;
  int s1, s2, nt = atoi(av[1] ? : "2"),
    l = 10000; // number of outer iterations
  long long ls = 0;

  if (nt < 2)
    nt = 2;
  clock_t start, t1, t2;
  start = clock();
  for (int i = 0; i < l; i++)
    s2 = handle(v, N, call_sum, nt), ls += s2;
  t2 = clock() - start;
  start = clock();
  puts("one thread test\n");
  for (int i = 0; i < l; i++)
    s1 = sum(v, 0, N), ls += s1;
  t1 = clock() - start;

  print("s1", s1, t1, 1);
  print("s2", s2, t2, nt);
  printf("ls = %lld\n", ls);

  return puts("End") == EOF;
}