Daily Pathtracer, Part 2. Where I make it 10x faster (or 10x less stupidly slow) by typing one word! http://aras-p.info/blog/2018/03/28/Daily-Pathtracer-Part-2-Fix-Stupid/
Next up: let's do the same path tracer in C#, because why not.
@aras Nice catch! - it might be worth mentioning how you identified the problem. whether it's by re-reading code until it was "obvious", by instrumentation assistance, a hunch based on thread scaling, or smth else.
@jakob I actually forgot how exactly I identified it. Tried some profiling runs without clear indicaton; might have been an "a-ha" moment.
@aras sometimes it's just the smell of a 'static' i guess ..
@aras is it on your list to do cosine weighted hemisphere samples? you don't get more rays per second but the rays you get converge faster (you also remove the multiply by cos theta).
@demofox dunno, I'm thinking maybe I should ditch the explicit path tracing and just do importance sampling instead. but first I have to at least pretend to understand the math involved! :)
@aras @demofox don't ditch the explicit light sampling, do both! (they're not mutually exclusive 😄)