Deviation Actions

phresnel's avatar

picogen 0.3 - preview: early

By phresnel
Another rendering using the new volume-integrator of picogen. Reminds me of a nice boat-tour in the morning through some (small) canyons.

To find out more about picogen, see [link] and [link] :)

--

edit: Fixed a typo, thanks to *lyc (see below)
Image details
Image size
1680x600px 1.56 MB
Published:
© 2010 - 2021 phresnel
Comments5
Join the community to add your comment. Already a deviant? Log In
lyc's avatar
the volume integration looks purdy indeed, i really wanna submit one of your recent works to gfxcoders! how about a computer-maximising render with lots of detail and aa? :D also watch the typos (samples per pixels) :giggle:
phresnel's avatar
on another note: Do you think the render time is okay for a quad-core cpu? (quick not to self: prolly need to fix some races in prng)

memory okay?
lyc's avatar
[soz for the once-again-late-reply! i'm now back in wellington after another stressful move.]

it's difficult to say if it's "good" without knowing really what's going on in the code, how fast the cpu is at doing xyz, etc... but it's easily "not bad at all" from what i can tell, it seems to be quite a general rendering system and volumetrics aren't cheap to compute. so i wouldn't worry too much about that; make sure you're happy with the code, and if you want more performance, then optimise it a bit :)

PRNG: rather use some deterministic scheme :#1: better convergence, no shared state.
phresnel's avatar
[and soz for my late reply reply :P
you move quite often, do you? anyways, glad that it succeeded :)]

That is true, without volumetrics, it usually only takes a few minutes to compute those images at relatively good quality (yay, no more waiting for hours to create a quadtree that is anyways only partially used). picogen 0.4 will probably include pruning of unused tree-nodes, so that even less memory is used :)

PRNG: I've written a portable implementation of KISS RNG, where I took care to not enter undefined behaviour, nor fragile behaviour of the assumimg-'sizeof(int)==4'-kind (straight reimplemted from
* where he relies on long==sizeof(4 and on that some_char+some_int will be taken mod 256, but the latter is a false assumption as is, but he refines it in the next link)
* [link] )

It is cheap to construct, cheap enough so I can do for each sample "KissRNG (do_seed(x,y,user_seed,sample_number))", yielding equal results amongst platforms.
phresnel's avatar
What an honour :jedi:

'Til now, picogen didn't have a Transform/Matrix (see newest commit log at [link]), but now I am thinking about making a top-down render, in the spirit of [link] ), with lots of parallel shafts, hmm, hmm, yummy. Wait some more two or three deviations :)


typo: dammit, I never preserve the gimp working copy, luckily I can black it out :)