#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <malloc.h>
#include <math.h>

int main(int argc, char **argv) 
{
  float *block, maxval = 0;
  int blockSize=8388608*8, numBlocks=200, i, j, iblock;

  block = (float *)memalign(32, blockSize*sizeof(float));
  memset(block, 0, blockSize*sizeof(float));

  for(iblock = 0; iblock < numBlocks; iblock++) {
    //#pragma simd
    for(j = 0; j < blockSize; j++) {
      block[j] += sqrtf((float)(j));
    }
  }

  for(i = 0; i < blockSize; i++)
    if(block[i] > maxval)
      maxval = block[i];

  printf("maximum element is %f\n", maxval);
  return 0;
}