Deepest - e10000

[Kalles Fraktaler]

Code:

$ m-box-period 53 -0.8691524744 0.2556487868 1.25e-5 1000
48

It would be really nice if you could make executables for Windows available on your site

[stardust4ever]

So basically through some kind of mathematical voodoos I won't pretend to understand, you can basically find a minibrot of any period, anywhere in the set, at any depth, without the manual work of zooming through the infinite cosmos. The limiting factor to depth is now the horsepower of your CPU and patience of the mandelbrot explorer.

Don't get me wrong, while technically impressive to find eternally deep minibrots, the automated process strips nearly all creativity out of the zoom sequence.

@KF: I watched the first 6 and last 6 minutes of your YT vid. Let's say I am thankful not to suffer from epilepy.  

@claude: I would be most interested in using this automated brot finding process to cut out the long hours of tedious manual zooming in between scenes when creating sophisticated patterns...

I discussed this a bit in the superfractalthing reboot thread.

[claude]

I rerendered this location in my own software using distance estimation, still not recommending to sit down and watch it like you would a film, so I picked some nice ambient drones as soundtrack...

https://archive.org/details/e10000

I think I started rendering it at the start of January, had to stop and fix software bugs a few times before continuing the render.. at least my program has fewer bugs now

[quaz0r]

thanks for the share      what method do you use for making zooms?  there was a post on here about mercator maps or something but the thread died without much further discussion

[claude]

I blend between successive 2x zoom level images, using the blending factors here https://mathr.co.uk/blog/2010-08-31_optimizing_zoom_animations.html (and also motion-blur with the shutter open for 1/2 the frame time)  source code here: https://code.mathr.co.uk/mightymandel/blob/8f45039d55408902d0d0c2723ed27fce65e10f14:/extra/zoom.c

I have used mercator map / exponential strip method in the past (my "emndl" project circa 2011), but I haven't added it to my pertubration renderer(s) yet - it's a bit fiddly to get right and it isn't suitable for interactive uses...

[quaz0r]

that audio track is pretty epic too.  the last ten minutes i felt like i was flying through the wormhole in 2001 a space odyssey   

[Kalles Fraktaler]

Very cool!
It shouldn't be hard for you to make even deeper zooms, despite the time consuming render...

[claude]

...
I had to update my floatexp class (used when deeper than e4920) so that it is based on 2 instead of 10 and masks off the exponent and handle it separately. That made it much faster than before, which was necessary to be able to complete this render in reasonable time.
...

I've been working on accelerating my own edouble variant of the floatexp idea today.  Made it almost three times as fast as before by using your masking code (thanks!) instead of ldexp/frexp libm calls, and annotating the branches with __builtin_expect (gcc-specific, maybe) to help CPU branch prediction.

My edouble code is still ~2x slower than your floatexp (compiled with the same compiler), but it does correctly handle 0 (with underflow), infinities (with overflow), and NaN.

But I worry that your floatexp might sometimes interpret 0.0 as 1x2^-1024 or so - I don't know how likely that is to occur in practice...

My code is here: https://code.mathr.co.uk/mandelbrot-perturbator/blob/HEAD:/lib/edouble.cc
Your code on my hard disk is dated May 22 2014, not sure which version that corresponds to.

[Kalles Fraktaler]

I've been working on accelerating my own edouble variant of the floatexp idea today.  Made it almost three times as fast as before by using your masking code (thanks!) instead of ldexp/frexp libm calls, and annotating the branches with __builtin_expect (gcc-specific, maybe) to help CPU branch prediction.

My edouble code is still ~2x slower than your floatexp (compiled with the same compiler), but it does correctly handle 0 (with underflow), infinities (with overflow), and NaN.

But I worry that your floatexp might sometimes interpret 0.0 as 1x2^-1024 or so - I don't know how likely that is to occur in practice...

My code is here: https://code.mathr.co.uk/mandelbrot-perturbator/blob/HEAD:/lib/edouble.cc
Your code on my hard disk is dated May 22 2014, not sure which version that corresponds to.

Cool, I haven't updated it since that date.
I haven't encoutered any problems that 0 is not actually 0...

[quaz0r]

ldexp/frexp is epically, ridiculously, unusably slow indeed.  i use union casting for the masking code though, which is at least explicitly supported by compiler vendors (while still technically undefined behavior by the C++ standard), whereas kalle's C-style casting is entirely undefined by anyone anywhere anytime.  i forget if modern compilers just complain very loudly about it if you try to compile it, or if they now flat-out refuse to compile it at all.

[Kalles Fraktaler]

ldexp/frexp is epically, ridiculously, unusably slow indeed.  i use union casting for the masking code though, which is at least explicitly supported by compiler vendors (while still technically undefined behavior by the C++ standard), whereas kalle's C-style casting is entirely undefined by anyone anywhere anytime.  i forget if modern compilers just complain very loudly about it if you try to compile it, or if they now flat-out refuse to compile it at all.

Yes you are right, I will use your strategy from now and not share anything ever again

[quaz0r]

relax bro, i was just commenting on C++ issues