Andre Poenitz wrote: > On Mon, Jun 12, 2006 at 04:32:46PM +0200, Peter Kümmel wrote: >> Peter Kümmel wrote: >>> I don't think the event handling counts, because it shows >>> the uptime time of lyx, even when nothing is painted. >> This makes more sense (I've disabled all std::* calls): >> >> Name, Self visit time >> >> QRasterPaintEngine::drawTextItem, 5.622828 >> qBlue, 5.467107 >> QLFontInfo::width, 4.111963 >> InsetList::insetIterator, 3.755528 >> QRasterBuffer::resetBuffer, 2.911848 >> ExtTextOutW, 1.485099 -> THIS is the Windows text painting function >> comp_func_solid_Source, 0.763000 >> Paragraph::getChar, 0.617311 >> >> >> Here the time which is spend in the different dlls/exe: >> >> QtGuid4.dll , 31.600339 > ^---
I can't profile the release version of QtGui, because lyx crashes then (because of mixing debug and release versions), and I'm not sure if its possible to profiling a release build. >> lyx-qt4.exe , 21.862434 >> QtCored4.dll, 3.386272 > ^---- >> GDI32.dll , 2.411191 > > Well, profiling the non-optimized version is close to useless. > >> This means when you code lyx in plain C by only >> using the GDI functions directly you could get >> a boost of factor 14 = (31.6+2.4)/2.4 - the upper >> limit for Trolltech's Qt/win32 optimizations. :) >> (You could also say Qt's overhead) > > I like TT bashing as much as anybody else, but this is unfair. I don't wanna bash them, I just wanna know what's going on, and how expensive the Qt overhead is. And it's always good to know a upper limit, so you don't have to think about claims like "I could do it 30 times faster without Qt". Peter
