Kinda
So basically, a console command of "screenshot 1024 768 128" will grab a screenshot of 1024x768 with 128 samples (or, alternatively like John suggested, you can issue a "screenshot 4096 3064 128" command with the expected result). It is interesting to see the screenshot grabbing process (if you use smaller samples, like 8, you see jittering... if you use larger values at super high rez, you see what John refers to as "tiling"). However, I still can't quite grasp the r_lightSourceRadius cvar in connection with this. I still get funky disco lights when this cvar is enabled (with various values). John hasn't replied to my additional queries about this and I don't really wish to bug him so soon after the birth of his child... perhaps some of you guys know.
PS. John's mention of shadow buffer seems out of place but this is due to the fact that my email to him was about his QuakeCon04 talk, where he talked a great deal about shadow buffers.
John Carmack said:Reverend said:...
Anyway, how should I experiment with the cvar r_lightSourceRadius? It is described as "for soft shadow sampling" but using various values (positive and negative values) for this cvar results in "moving" shadows and lights instead of "soft shadows". This is on a NV40.
That is unrelated to the shadow buffer technology. If you take screenshots with large sample numbers while r_lightSourceRadius is enabled, you will get accumulated soft shadows, but because they are jittered light source positions, you need much larger values to get good looking results. For screenshots, you might as well use values like 64 or more samples, especially if you have larger values of r_lightSourceRadius set.
The screenshot command optionally takes width / height / samples parameters, and does multiple renders to tile large images if necessary. You are still limited to the hardware viewport limit, but a 4096 x 3064 x 128 sample shot with jittered light positions looks pretty good...
John Carmack
So basically, a console command of "screenshot 1024 768 128" will grab a screenshot of 1024x768 with 128 samples (or, alternatively like John suggested, you can issue a "screenshot 4096 3064 128" command with the expected result). It is interesting to see the screenshot grabbing process (if you use smaller samples, like 8, you see jittering... if you use larger values at super high rez, you see what John refers to as "tiling"). However, I still can't quite grasp the r_lightSourceRadius cvar in connection with this. I still get funky disco lights when this cvar is enabled (with various values). John hasn't replied to my additional queries about this and I don't really wish to bug him so soon after the birth of his child... perhaps some of you guys know.
PS. John's mention of shadow buffer seems out of place but this is due to the fact that my email to him was about his QuakeCon04 talk, where he talked a great deal about shadow buffers.