------=_20040914011153_23758
Content-Type: text/plain; charset="iso-8859-1"
Content-Transfer-Encoding: 8bit


>> You're getting ahead of yourself, but no problem w/the move itself.
>> However, plastik needs some performance work.
> To be honest, I didn't notice these general performance problems.

I am not sure they're noticeable, they're easily measurable.

> So if you could help me identifying them, I would love trying to improve
> this situation.

Please see the attached .cpp file; apologies for the uglyness, not meant
to be any sort of production code. This is basically what I used to tune
keramik --- it just hammers on PE_ButtonCommand which for Ker hits the
core rendering path -- and I think Plastik is similar with its bevel
rendering accessible from here and occuring all over the place (please
correct me if I am wrong)

Here, when I run this with Keramik, I get it finishing in 26.5 seconds, of
which 15.5 are in-process (that's the "clock:" line, the portions is -way-
too high, likely due to all the temporary TilePainter objects; I hope
nextgen designs I do can eliminate that; hard to do w/the current
codebase).  Light3 comes in at much quicker 9.2 seconds... Plastik takes
200 seconds to run this app here, with 88 in-process, which is 7.5 times
slower than keramik, and 21.7 times slower than Light3. A candidate for a
drag on speed are all the save()/restore() calls, but I am not sure: for
some reason I recall these functions as being slow, but I don't have a
rationale.


-Maks

------=_20040914011153_23758
Content-Type: application/octet-stream; name="TestBtnSpd.cpp"
Content-Transfer-Encoding: base64
Content-Disposition: attachment; filename="TestBtnSpd.cpp"
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------=_20040914011153_23758--