The course includes a competition, in which you should optimize median filtering as far as possible, under certain conditions, and with a given interface.

The GPU part is based on lab 5, which is written in a way so that you can write a kernel, which I can easily insert in a program from which I insert whatever input I like and measure the performance.

It is important that you do not alter the interface. If you do, I can’t measure performance in a fair way.

More details should be added here.

THE COMPETITION IS OFF! The text below is moved from lab 5 and is not intended to be visible any more.

Almost new lab - your opinions are welcome and important!

Also note: The competition rules are sketchy at this time. If you wish to submit a contribution for the GPU part, it must be based on part 4 of this lab, median filter. The judgement will be given with a balance between precision, performance and flexibility.
Precision: Will it compute an exact median filter or does it use a separable solution, separate passes?
Performance: Your standard timings.
Flexibility: What are its limits for filter sizes?

This is basically what I will be looking at. Your solution should be structured like the lab so I can plug in any additional test code that I wish to use without having to rewrite the whole thing.
This page is maintained by Ingemar Ragnemalm