The fact that @eklein claims it is his code, then say both are based on Lukas work (wrong too) while he just uses my solution with some slightly changed values. I would like to post the code parts in E-Cycles, K-Cycles and then compare to Lukas Patch, but @eklein still didn’t provide code since my request 3 days ago.
It’s not necessary anyway, he already gave that information here. He is mapping samples values to a distance value like E-Cycles. To show how much work went in the 2 years since he took part to my making-of course, E-Cycles uses a distance value of 0.2 at 256 samples, while Erik uses 0.2 at 250 samples. I’ll let you judge of the value of that change. The benefit of that change is that it will render faster at 250 samples, thanks to a lower value, at the cost of image quality. But his benchmarks look better and some people fall in. Of course he will claim it’s a simple Idea, like the full research team who spend months of work to just change a variable. Yet neither the research team, nor Lukas, nor the BF found that solution which as an Artstation search can show ensure high quality renderings in any case.
Contrary to what he says, Lukas patch only replaced a fixed 1.0 factor in the sobol code by a variable, which he exposed to the UI. The only important change in that patch is line 56 you can enter your value manually here and it will render faster, the smaller you set it like writing shift *= 0.1 instead of shift *= kernel_data.integrator.scrambling_distance;. The rest of the patch is UI code and things to bring the user-entered value to Cycles. It was not even Lukas work. Lukas just made a quick patch based on a paper from a French research team. Both Lukas variant and the research paper had issues. The original paper if implemented as exposed is 30% to 50% slower than my solution, Lukas solution as said by himself many times can brake your renders and was thus never committed.
2 following patches were needed later due to some changes in Cycles and to make it work with OptiX. Both changes are in the code form E-Cycles he requested and got.
If you still want to fund someone claiming to have made a break through after 2 years of intensive research, not mentioning original authors anywhere, you can pay $49 to encourage him to continue on that path!
Or you can get the work from it’s original author and available for free ($1 on the Blender Market) like hundreds of people are doing each day!