Consider swapping atomic for reduction for incrementing counters

Using 

```
#pragma omp parallel for reduction(+:varName)
```

instead of using 

```
#pragma atomic
varName++
```

may be beneficial from a performance point of view. It's unclear what happens to `varName` if the for loop is executed multiple times (e.g. for different chunks) - will it be reinitialized to 0 each time? 