Confused on Photon count

[S L]

Hello,

I am trying to do a GPU implementation of SPPM, then I found out it's quite difficult reading the SPPM chapter on pbrt-book. Haven't looking into pbrt-v4 code.

So, I went back to the original papers pf PPM & SPPM by Hachisuka and Jensen. They are mostly self-explained, but there are also some confusing parts.

They use Ne as the total photon number. 

- Does the equation also work for multiple light sources without any extra work?

- Do I need to do it for each light source then add them together with some weight term?

We know some photons will not hit anything, some of them will be killed by Russian roulette. I explored some CPU photon code from other people, many of them don't count the invalid photon into the number.

I also see a GPU version from Hachisuka and Jensen. They are actually adding a fixed photon number for each photon pass, so they are also counting the invalid photons. 

- Should I also count them into Ne term?

I am considering tear down multi-bounce photon test into different photon passes. Would this solution introduce any error into my final result?

The flow chart below simply explained my idea, though it's not 100% correct as a flowchart.

 

--

Best Regards