Re: Fwd: render.c:GetPixelOpacity() optimization ideas?

Bob -

(I think those are excellent tips so I'm replying through the list)

Bob wrote:
> Even if you don't really understand the algorithm at all, you can
> study how it operates.  You can learn the range of values.  Once you
> know the range of values, if they are always positive, negative, or
> either way, etc., you can define a scaled integer space.  For example,
> for a floating value which varies between 0.0 and 1.0, you could
> decide to use an integer range of 0 to 10000 (a power of 2 may be a
> better choice).  The computations would be done using this boosted
> range. Whatever range you choose needs to be high enough to avoid
> inaccuracy, and low enough that values don't go past the range of
> 'long' (or 'long long' if your CPU supports it efficiently). Input
> parameters would be converted to this range before use.  If anything
> needs to be saved, then it needs to be scaled back to a double to keep
> the rest of the code happy.
>
> I believe that the polygon agorithms were influenced strongly by
> libart (it used to be credited in render.c), which is used by GNOME.
>
> Since floating point is bad for you, you should look for floating
> point which can be eliminated.  For example, the many "0.5"s and
> "0.25"s are usually there to avoid round-off errors and each of these
> is consuming a floating point operation.  Maybe you don't care about
> round-off errors since they usually can't be seen.  They are there in
> order to help our tests pass (and to be pedantic).  It is important
> that the results be accurate enough that flood fills don't start going
> where they shouldn't.
>
> Bob

Thank you very much. This gives me confidence to get started, in case I cannot 
find any outside help ;-)
If you have any other references to books, articles, URLs, please let me know.
Once I start to do it, I will certainly report back in case there is anything 
of value for you or at least I have a patch to post somewhere...

Wolfgang



-------------------------------------------------------
The SF.Net email is sponsored by EclipseCon 2004
Premiere Conference on Open Tools Development and Integration
See the breadth of Eclipse activity. February 3-5 in Anaheim, CA.
http://www.eclipsecon.org/osdn