Randomly, I think I may have been the first person to do (or at least write about) 2D GI. I taught a bunch of people about it (and rendering / path tracing in general), some of whom became a lot more popular.
Fun seeing that people were doing monte carlo integration for 2d lighting back in the 90s :) I assume this wasn't GPU accelerated at all at that point?
I did realtime 2D lighting with per pixel sprite normals and stuff like that, but I never got to the point of doing GI bounces - sophisticated direct lighting and shadows were already demanding enough without the use of programmable shader units. https://vimeo.com/1514095
I wrote the article around the time of the Nvidia Riva TNT1/TNT2, many years before the first decently programmable GPUs (and the introduction of the term "GPU", they were still called graphics accelerators). So yeah, CPU only :)
Unfortunately I can't view your video link without a Vimeo account to "prove my age" (hoover my data).
>>> possible to calculate the direct lighting analytically ?
I don't think that is the trend the industry is following. I think all investment research is poured into neural methods and ai-enhanced sampling. Joy to be breakfasting over such illuminating (pun intended) discourse & fractal art. My interest in 2D texture synth is more toward creating the stunning graphics found in classic sega style cell shaded titles like jet set radio. So I'm into the oldskool deterministic fractal geometry techniques, and adding dynamic luminaires and guassian blur passes is enough to create cool tron glow type effects ;)
Here's the oooold and very cringy article I wrote as a teen in the 90s, there are some old crap images in the gallery: https://web.archive.org/web/20031207162649fw_/http://www.cfx...