Higher precision

[ewx]

I’ve continued hacking on my Mandelbrot set program.

Firstly, it’s gained the ability to draw Julia sets (which are associated with the Mandelbrot set).

Secondly, the use of double limits you to around 50 binary digits after the point (all the interesting stuff has a magnitude of at least 0.25 or so, so the “floating” nature of floating point doesn’t really do you much good). Given how fast it is, it doesn’t take much frantic zooming before things start to go blocky:

So I’ve been working away on using fixed point arithmetic to achieve the same goal. To do this at any reasonable speed required assembly language, and now it’s working. It’s somewhat slower than the floating point version, but gives 96 bits after the point, i.e. it allows you to zoom in almost twice as far.

You can browser the source, or git clone it, via gitweb.

Comments

[gerald_duck]

I assume long double wasn't any use to you? On x86 it should be almost as fast as double, but gets you an extra eleven bits of significand.

[ewx.livejournal.com]

I haven’t tried, but it wouldn’t be hard now that the infrastructure for using different types is in place.