gerben-s / quicksort-blog-post Goto Github PK

At the end of your blog post you've even tried implementing CPU-pipeline-efficient MergeSort. What about enhancing Quadsort (which is MergeSort, but with a tiny sorting network built-in) by your DistributeBackward and DistributeForward approach?

I pulled this into my code to see how it compared to a sort implementation I was using and found that the tests were hanging. After a little investigation, it wasn't hanging but was just performing very badly for the cases where all elements are equal. Looks like it partiions into subarrays of size 1 and n-1 and continues that all the way down or until it blows up due to too much recursion.

I haven't looked at any fixes, but wanted to know whether that was a known issue with the experimental code or whether that behavior had been overlooked.

Although the test is easy to reproduce, here is the code for completeness:

#include <hybrid_qsort.h>
#include <vector>
int main()
{
  //  std::vector<int> v(131071,1);
  std::vector<int> v(1024,1);
  exp_gerbens::QuickSort(v.data(), v.size());

}

Hi,

Since you've tried to make this algorithm general, in BubbleSort2, the line

if (y < x) std::swap(x, y);

should be

if (comp(y, x)) std::swap(x, y);

Thanks.