Re: Anti-aliasing via GPU

Hi Kohei,

On Thu, 2012-12-13 at 20:21 -0500, Kohei Yoshida wrote:
> Since proper profiling of a drawing framework takes at least a few days
> (running various benchmarks under various scenarios, and interpreting &
> writing about the results)

        If you experience slow performance with a real-world document; it takes
only 10 minutes to load the thing in callgrind, reset the counters
before doing the slow thing, call callgrind_control --dump afterwards
and bingo: there is a profile if the real-world use-case we want to
speed up :-) that might have general utility etc. - but ... constructing
arbitrary artificial benchmarks can also be a problem in this way.

> I would be happy to spend a few days to profile each drawing frameworks 
> if you want me to.  But alas, right now I'm soooo loaded with Calc work 
> that I may have to wait a year or so to be able to do that.

        Getting the latest version of your document would be really helpful; is
it available somewhere ? last I looked there was only an out-of-date
version in some obscure git repo somewhere ;-)

> I'd rather not.  That document is my personal note.  I put that up on my 
> site in the hope that someone finds it useful.  But I'm not ready to 
> subject that to multiple people editing it, nor am I looking for 
> increasing its visibility.

        Why not increase it's visibility ? last I looked it was beautiful and
helpful :-) I'd love to have the latest version at least -linked- into
the code. Would you accept patches / new bits to it ?

        Anyhow :-) in general I agree: it is a good idea to re-use someone
else's faster/better backend rendering infrastructure - I'm completely
sold on that. AFAIR there are two good options - cairo or the new
cross-platform rendering thing that Mozilla is producing. On the other
hand - switching those technologies will almost certainly bring
performance regressions rather than benefits: as a different set of
performance mis-matches between caller and callee arrive :-) So we
shouldn't do that switch for an unknown performance win :-) Of course,
it makes complete sense for improved maintenance and if we could unify
chunks of rendering code that would be lovely - what we have is broadly
a big/horrible mess of half-finished duplicate functionality
(AFAICS) :-)

        HTH,

                Michael.

-- 
michael.meeks@suse.com  <><, Pseudo Engineer, itinerant idiot