Re: [glob2-devel] Another speedup (computeColors)

[Kieran P]

Wow. You just keep churning them out :P

I've applied all changes in the diff files to the files and committed. Unfortunately I must have committed wrongly, and its in master instead of the optimized branch I was aiming for, but never mind. clone the code base, clone the revision master of that clone, hg update -C tip, and you should be set to go.

Any more files, I'll try to update (probably to master if it compiles fine). 

Question: You are thoroughly testing these changes? All I'm going to do is make sure they compile.

Regards
Kieran

On 9/7/07, Erik Søe Sørensen <address@hidden> wrote:

Hi again,
I said I'd look at algorithmic improvements to the gradient updating
operations... well, I did, but I chose the wrong starting point. I've
got some ideas to try out (most of which concern avoiding to update all
of the map), but that'll wait a bit.

For now, I've got some other spectacular result for ya :-D - a twofold
increase in the speed of Minimap::computeColors() (on my box at least) :-)

I tried several different things, but what turned out to be the real
problem was the use of floats (and perhaps especially the
float-to-integer conversion which was done for each relevant map square).

So the attached patch just replaces this with fixed-point computations,
and then moves some variable declarations to their 'proper' places
(because that's nicer, for programmers and compilers alike :-).