Hmm, been a while since I updated this.
Submited a few patches for gtk+ 2.0. Mostly UI stuff. I don't mean making the widgets look prettier which is what many people think of when you mention the UI word. I mean making the users interactions with the widgets richer and more predictable.
I've been testing the performance characteristics of different coding styles on pixel manipulation lately. Results from these tests will be used to make sure that GEGL pushes pixels faster than oil companys destroy wildlife refuges.
Things I've learned so far:
- On the sgi good code beats bad code by about 50%
- On a P3 or Athlon good code beats bad code by about 600%
- On the G4 good code beats bad code by about 375%
- Altivec is a well designed, well documented, well suported and easy to use instruction set.
- MMX is a half-assed, less-well documented, unsuported pain in the ass.
- Altivec can with only a very small investment of time speed up your code by 350% (even without using the memory prefech instructions)
- MMX takes a signifigant amount of time and effort to code with, but eventualy you will have a faster less understandable program
MMX has been out for years now. Why can't compilers generate MMX instructions automatically. Why isn't there a simple to use interface to the SIMD instructions on the pc platform? Why can't you write programs that take advantage of MMX without dropping down to asembly language? Why the heck hasn't intel submited patches to gcc to make it easy for programmers to take advantage of their SIMD instructions? It would sell a heck of a lot more chips than paying those blue apple using freaks to dance around on their television commercials.