#recursive mandelbrot

34 messages ยท Page 1 of 1 (latest)

normal glen
#

very simple single threaded mandelbrot renderer, that when it reaches a member of the set, recurses until it finds one that isnt, with some blurring to look cool. will definitely get around to multithreading (and potentially SIMD) at some point because performance is lacking

gilded glacier
#

super cool ๐Ÿ™‚

#

I'm a huge fan of mandelbrot renders that provide unique visuals for the interior of the set

normal glen
#

yeah, i didnt manage to come up with much when i was gpu rendering, but cpu rendering allows for this sort of thing

gilded glacier
normal glen
#

also, its resolution agnostic, so it looks the same (albeit noisier) at lower res

#

picture there is 16k and took ~11 mins

#

on my i5-10400f

normal glen
gilded glacier
#

it's not too bad, it's just recursive ๐Ÿ™‚

normal glen
#

oh sort of divide and conquer style?

gilded glacier
#

there are some tricky aspects, easy off by one errors

normal glen
#

and still pretty paralelisable

gilded glacier
#

yep, still very parallelizable

normal glen
#

im just wondering how itd be implemented alongside the inside recursion

#

because you have to worry about things outside the squares, not just inside

gilded glacier
#

you would probably only use this for the exterior

#

everything in the inside you'd need to do individually

normal glen
#

yeah that makes more sense

#

though the outside only takes seconds

#

i should probably send the code

#

meh i need to polish it first

#

eg output jxl rather than ppm (just raw pixels)

gilded glacier
#

it's understandable that the outside is much faster

normal glen
#

you can see the difference in size, even with lossless compression

normal glen
gilded glacier
#

indeed, I was doing some mandelbrot rendering a while back and also outputting to ppms and it was a lot ๐Ÿ™‚

normal glen
#

yeah

#

anyway ill get to making a threadpool, i keep putting off multithreading and its coming back to bite me

normal glen
#

its multithreaded

#

and works a lot faster which is nice

#

lossless jxl version