On Fri, Jun 09, 2006 at 10:49:02PM +0200, Abdelrazak Younes wrote:
> Note that my primary target for optimisation is X11 and Mac. For me Qt4
> on windows is fast enough.
>
> [interesting stuff]
>
> >Well... when I think about it, it might be possible that QPicture
> >compresses some operations a
Andre Poenitz wrote:
On Thu, Jun 08, 2006 at 10:29:39AM +0200, Abdelrazak Younes wrote:
Abdelrazak> The QPicture class "records" painting operations, it is
Abdelrazak> similar to vector based painting (SVG or windows Meta
Abdelrazak> file). So when you paint onto a QPicture it is not a real
Abde
On Thu, Jun 08, 2006 at 10:29:39AM +0200, Abdelrazak Younes wrote:
> >>>Abdelrazak> The QPicture class "records" painting operations, it is
> >>>Abdelrazak> similar to vector based painting (SVG or windows Meta
> >>>Abdelrazak> file). So when you paint onto a QPicture it is not a real
> >>>Abdelraz
Andre Poenitz wrote:
On Tue, Jun 06, 2006 at 06:10:08PM +0200, Abdelrazak Younes wrote:
Jean-Marc Lasgouttes wrote:
"Abdelrazak" == Abdelrazak Younes <[EMAIL PROTECTED]> writes:
Abdelrazak> The QPicture class "records" painting operations, it is
Abdelrazak> similar to vector based painting (SV
Andre Poenitz wrote:
>> complete QLPainter member functions:
>> Scope 12,ms : 1236 -> setQPainterPen
>
> Hm, switching pens is known to be extraordinarily expensive in Qt 4.
> Things have a bit improved in 4.1.3, though.
>
Yes, there is a place where we could optimize a bit.
>> Scope 14,
On Sun, Jun 04, 2006 at 07:52:42PM +0200, Peter Kümmel wrote:
> Here some statistics of QLPainter functions:
>
> I've measured the time which the program is
> in an scope, see attached patch.
Is there a reason why you can't use a profiler?
> The results are:
>
> complete QLPainter member functi
On Tue, Jun 06, 2006 at 06:10:08PM +0200, Abdelrazak Younes wrote:
> Jean-Marc Lasgouttes wrote:
> >>"Abdelrazak" == Abdelrazak Younes <[EMAIL PROTECTED]> writes:
> >
> >Abdelrazak> The QPicture class "records" painting operations, it is
> >Abdelrazak> similar to vector based painting (SVG or w
On Sun, Jun 04, 2006 at 11:27:46PM +0200, Peter Kümmel wrote:
> How often have I heard that the painting performance of Qt4 is so bad,
> bad, bad...
>
> And all what we can do is to hope it becomes better, seems Trolltech
> has forgotten it's clients.
Not really. They do work a bit on speed issue
Lars Gullik Bjønnes wrote:
> | Note that there is room for improvement in our fontcache which uses a
> | std::map. A vector will probably not use too much memory (for 8bit
> | fonts) and be much faster.
>
> It will not be 8bit fonts.
>
> We should use a unordered_map and fallback to map if that i
Lars Gullik Bjønnes wrote:
> Jean-Marc Lasgouttes <[EMAIL PROTECTED]> writes:
>
> | > "Peter" == Peter Kümmel <[EMAIL PROTECTED]> writes:
> |
> | Peter> With enabled USE_LYX_FONTCACHE it is a little bit faster
> | Peter> (because the value of QLFontInfo::width becomes smaller
> | Peter> compa
Jean-Marc Lasgouttes <[EMAIL PROTECTED]> writes:
| > "Peter" == Peter Kümmel <[EMAIL PROTECTED]> writes:
|
| Peter> With enabled USE_LYX_FONTCACHE it is a little bit faster
| Peter> (because the value of QLFontInfo::width becomes smaller
| Peter> compared to the others: enabled 100<->48, disa
Jean-Marc Lasgouttes wrote:
>> "Peter" == Peter Kümmel <[EMAIL PROTECTED]> writes:
>
> Peter> With enabled USE_LYX_FONTCACHE it is a little bit faster
> Peter> (because the value of QLFontInfo::width becomes smaller
> Peter> compared to the others: enabled 100<->48, disabled 100<->47)
>
> I'd
> "Abdelrazak" == Abdelrazak Younes <[EMAIL PROTECTED]> writes:
Abdelrazak> Just for clarity, the map doesn't cache fonts, only font
Abdelrazak> widths (with Uchar as the key). The font cache is done
Abdelrazak> elsewhere in 4-dimensional C-array (fontinfo_).
Indeed.
JMarc
Jean-Marc Lasgouttes wrote:
"Peter" == Peter Kümmel <[EMAIL PROTECTED]> writes:
Peter> With enabled USE_LYX_FONTCACHE it is a little bit faster
Peter> (because the value of QLFontInfo::width becomes smaller
Peter> compared to the others: enabled 100<->48, disabled 100<->47)
I'd say it is the s
> "Peter" == Peter Kümmel <[EMAIL PROTECTED]> writes:
Peter> With enabled USE_LYX_FONTCACHE it is a little bit faster
Peter> (because the value of QLFontInfo::width becomes smaller
Peter> compared to the others: enabled 100<->48, disabled 100<->47)
I'd say it is the same (the margin error is
Jean-Marc Lasgouttes wrote:
>> "Peter" == Peter Kümmel <[EMAIL PROTECTED]> writes:
>
> Peter> So optimizing the text drawing functions couldn't more than
> Peter> doubling the speed.
>
> On LyX/Mac, the other half is taken by QLFontInfo::width. You may want
> to experiment with USE_LYX_FONTCA
Jean-Marc Lasgouttes wrote:
> No, but Matthias says that Qt slowness is just a rumor. We must be
> dreaming.
Maybe we live in different Matrix. :)
Peter
> "Angus" == Angus Leeming <[EMAIL PROTECTED]> writes:
Angus> These two threads provide some interesting reading on the
Angus> subject of slow drawings:
Angus> http://lists.trolltech.com/qt-interest/2005-10/thread00915-0.html
Angus> http://dot.kde.org/1135084395/1135109572/
Angus> They don't
Jean-Marc Lasgouttes wrote:
"Abdelrazak" == Abdelrazak Younes <[EMAIL PROTECTED]> writes:
Where does it explain that it is faster? I did not find it.
Abdelrazak> _I_ explained to you above ;-)
So you just tried to see what happens and saw it is faster, right? But
does trolltech recommend t
Abdelrazak Younes <[EMAIL PROTECTED]> writes:
> > Abdelrazak> The QPicture class "records" painting operations, it is
> > Abdelrazak> similar to vector based painting (SVG or windows Meta
> > Abdelrazak> file). So when you paint onto a QPicture it is not a real
> > Abdelrazak> painting. The real pa
> "Abdelrazak" == Abdelrazak Younes <[EMAIL PROTECTED]> writes:
>> Where does it explain that it is faster? I did not find it.
Abdelrazak> _I_ explained to you above ;-)
So you just tried to see what happens and saw it is faster, right? But
does trolltech recommend to use this interface for
Jean-Marc Lasgouttes wrote:
"Abdelrazak" == Abdelrazak Younes <[EMAIL PROTECTED]> writes:
Abdelrazak> The QPicture class "records" painting operations, it is
Abdelrazak> similar to vector based painting (SVG or windows Meta
Abdelrazak> file). So when you paint onto a QPicture it is not a real
A
> "Abdelrazak" == Abdelrazak Younes <[EMAIL PROTECTED]> writes:
Abdelrazak> The QPicture class "records" painting operations, it is
Abdelrazak> similar to vector based painting (SVG or windows Meta
Abdelrazak> file). So when you paint onto a QPicture it is not a real
Abdelrazak> painting. The
Jean-Marc Lasgouttes wrote:
"Abdelrazak" == Abdelrazak Younes <[EMAIL PROTECTED]> writes:
Abdelrazak> Please find attached the patch. The main problem with it
Abdelrazak> is that everything that is drawn in screen stays. So it
Abdelrazak> conflict with the background drawing that the kernel doe
> "Abdelrazak" == Abdelrazak Younes <[EMAIL PROTECTED]> writes:
Abdelrazak> Please find attached the patch. The main problem with it
Abdelrazak> is that everything that is drawn in screen stays. So it
Abdelrazak> conflict with the background drawing that the kernel does
Abdelrazak> to erase so
Abdelrazak Younes wrote:
> Peter Kümmel wrote:
>> Here more numbers, with this setup:
>>
>> MSVC8
>> Qt 4.1.2 release
>
> Please update to 4.1.3. There are some nice drawing optimization there.
> Most noticeably for Mac and Linux but I've seen improvment on WinXP also.
>
Is planned.
>> So optimi
Peter Kümmel wrote:
Here more numbers, with this setup:
MSVC8
Qt 4.1.2 release
Please update to 4.1.3. There are some nice drawing optimization there.
Most noticeably for Mac and Linux but I've seen improvment on WinXP also.
So optimizing the text drawing functions couldn't
more than doubl
Jean-Marc Lasgouttes wrote:
"Peter" == Peter Kümmel <[EMAIL PROTECTED]> writes:
Peter> So optimizing the text drawing functions couldn't more than
Peter> doubling the speed.
On LyX/Mac, the other half is taken by QLFontInfo::width. You may want
to experiment with USE_LYX_FONTCACHE in qfont_loa
Michael Abshoff wrote:
>> Here more numbers, with this setup:
>>
>> MSVC8
>> Qt 4.1.2 release
>> lyx-qt4 release -O2
>> AMD XP 1.1Ghz, 100 Mhz Ram
>>
>
>
> Where do I get one of those? :)
>
It's down clocked :)
>
> Nope, valgrind doesn't help here much. There is a valgrin
>
> Here more numbers, with this setup:
>
> MSVC8
> Qt 4.1.2 release
> lyx-qt4 release -O2
> AMD XP 1.1Ghz, 100 Mhz Ram
>
Where do I get one of those? :)
>
> Page down-key scrolling time of the User's Guide:
>
> 55s
>
> but when I disable all functions of QLPainter
> (just r
> "Peter" == Peter Kümmel <[EMAIL PROTECTED]> writes:
Peter> So optimizing the text drawing functions couldn't more than
Peter> doubling the speed.
On LyX/Mac, the other half is taken by QLFontInfo::width. You may want
to experiment with USE_LYX_FONTCACHE in qfont_loader.h to see whether
our
Here more numbers, with this setup:
MSVC8
Qt 4.1.2 release
lyx-qt4 release -O2
AMD XP 1.1Ghz, 100 Mhz Ram
Page down-key scrolling time of the User's Guide:
55s
but when I disable all functions of QLPainter
(just returning immediately) I still get
25s
So optimizing the text drawing function
Peter Kümmel wrote:
Abdelrazak Younes wrote:
Yes. I have been experimenting with a QPicture cache between
QLPainter::start() and QLPainter::end() which would replay the QPicture
into the QWorkArea painting_device_ (which is a QPixmap). The speedup
is awesome. Unfortunately the core is not prep
Peter Kümmel wrote:
Abdelrazak Younes wrote:
Yes. I have been experimenting with a QPicture cache between
QLPainter::start() and QLPainter::end() which would replay the QPicture
into the QWorkArea painting_device_ (which is a QPixmap). The speedup
is awesome. Unfortunately the core is not prep
Abdelrazak Younes wrote:
> Yes. I have been experimenting with a QPicture cache between
> QLPainter::start() and QLPainter::end() which would replay the QPicture
> into the QWorkArea painting_device_ (which is a QPixmap). The speedup
> is awesome. Unfortunately the core is not prepared for that k
Peter Kümmel wrote:
Abdelrazak Younes wrote:
Peter Kümmel wrote:
The most time is spend in the Qt function QPainter::drawText.
Could this time be reduced?
I came to the same conclusion with the Mac profiling done by Bennett. I
guess we can't do much about QPainter::drawText beca
Abdelrazak Younes wrote:
> Peter Kümmel wrote:
>> The most time is spend in the Qt function QPainter::drawText.
>> Could this time be reduced?
>
> I came to the same conclusion with the Mac profiling done by Bennett. I
> guess we can't do much about QPainter::drawText because this is part of
> the
Peter Kümmel wrote:
The most time is spend in the Qt function QPainter::drawText.
Could this time be reduced?
I came to the same conclusion with the Mac profiling done by Bennett. I
guess we can't do much about QPainter::drawText because this is part of
the Qt API.
Abdel.
Peter Kümmel wrote:
> complete QLPainter member functions:
> Scope 12,ms : 1236 -> setQPainterPen
> Scope 14,ms : 812 -> line
> Scope 15,ms : 172
> Scope 16,ms : 46
> Scope 17,ms : 3683 -> fillRectangle
> Scope 20,ms : 16
> Scope 22,ms : 514
> Scope 23,ms : 201
>
39 matches
Mail list logo
