Re: About scrolling speed
Andre == Andre Poenitz [EMAIL PROTECTED] writes: Andre On Tue, Oct 24, 2006 at 10:01:43AM +0200, Juergen Vigna wrote: Hi Abdel, as much as I know before we started on fixing bugs there I was not able to scroll trough the UserGuide without hitting an Assert ;) I think that this is due to the more metrics calculations we have to do in order to fix all that scrolling bugs and with it the removal of the nullpainter which was not calculating important stuff as *normal* text, and therefore with it in place the x positions of the insets where all wrong! Andre Hm, so maybe we should resurrect nullpainter as a Andre non-drawing-but-metrics-computing version of the real painter. Andre Of course, this could be as simple as having a bool flag in Andre PainterInfo stating whether we actually want to output anything Andre or not. This looks like a good idea (and simple to implement!) JMarc
Re: About scrolling speed
Andre Poenitz wrote: On Mon, Oct 23, 2006 at 06:33:16PM +0200, Abdelrazak Younes wrote: Hi Andre, Before Denmark, the UserGuide PageDown test was at 18 seconds. Now it is at 25 seconds. I hope you have some more code in store for speed ;-) ? The decrease in speed was due to the new black square cursor. The speed is now the same as before, not slower but not faster either. Abdel.
Re: About scrolling speed
Abdelrazak Younes wrote: Andre Poenitz wrote: On Mon, Oct 23, 2006 at 06:33:16PM +0200, Abdelrazak Younes wrote: Hi Andre, Before Denmark, the UserGuide PageDown test was at 18 seconds. Now it is at 25 seconds. I hope you have some more code in store for speed ;-) ? The decrease in speed was due to the new black square cursor. The speed is now the same as before, not slower but not faster either. I've disables assertions and 'checked iterators' for msvc release builds, this gives us maybe a little bit. -- Peter Kümmel
Re: About scrolling speed
Peter Kümmel wrote: Abdelrazak Younes wrote: Andre Poenitz wrote: On Mon, Oct 23, 2006 at 06:33:16PM +0200, Abdelrazak Younes wrote: Hi Andre, Before Denmark, the UserGuide PageDown test was at 18 seconds. Now it is at 25 seconds. I hope you have some more code in store for speed ;-) ? The decrease in speed was due to the new black square cursor. The speed is now the same as before, not slower but not faster either. I've disables assertions and 'checked iterators' for msvc release builds, this gives us maybe a little bit. Good idea. I'll test that. Thanks, Abdel.
Re: About scrolling speed
> "Andre" == Andre Poenitz <[EMAIL PROTECTED]> writes: Andre> On Tue, Oct 24, 2006 at 10:01:43AM +0200, Juergen Vigna wrote: >> Hi Abdel, >> >> as much as I know before we started on fixing bugs there I was not >> able to scroll trough the UserGuide without hitting an Assert ;) >> >> I think that this is due to the "more" metrics calculations we have >> to do in order to fix all that scrolling bugs and with it the >> removal of the nullpainter which was not calculating important >> stuff as *normal* text, and therefore with it in place the x >> positions of the insets where all wrong! Andre> Hm, so maybe we should resurrect nullpainter as a Andre> non-drawing-but-metrics-computing version of the real painter. Andre> Of course, this could be as simple as having a bool flag in Andre> PainterInfo stating whether we actually want to output anything Andre> or not. This looks like a good idea (and simple to implement!) JMarc
Re: About scrolling speed
Andre Poenitz wrote: On Mon, Oct 23, 2006 at 06:33:16PM +0200, Abdelrazak Younes wrote: Hi Andre, Before Denmark, the UserGuide PageDown test was at 18 seconds. Now it is at 25 seconds. I hope you have some more code in store for speed ;-) ? The decrease in speed was due to the new black square cursor. The speed is now the same as before, not slower but not faster either. Abdel.
Re: About scrolling speed
Abdelrazak Younes wrote: > Andre Poenitz wrote: >> On Mon, Oct 23, 2006 at 06:33:16PM +0200, Abdelrazak Younes wrote: >>> Hi Andre, >>> >>> Before Denmark, the UserGuide PageDown test was at 18 seconds. Now it >>> is at 25 seconds. I hope you have some more code in store for speed ;-) >> >> ? > > The decrease in speed was due to the new black square cursor. The speed > is now the same as before, not slower but not faster either. I've disables assertions and 'checked iterators' for msvc release builds, this gives us maybe a little bit. -- Peter Kümmel
Re: About scrolling speed
Peter Kümmel wrote: Abdelrazak Younes wrote: Andre Poenitz wrote: On Mon, Oct 23, 2006 at 06:33:16PM +0200, Abdelrazak Younes wrote: Hi Andre, Before Denmark, the UserGuide PageDown test was at 18 seconds. Now it is at 25 seconds. I hope you have some more code in store for speed ;-) ? The decrease in speed was due to the new black square cursor. The speed is now the same as before, not slower but not faster either. I've disables assertions and 'checked iterators' for msvc release builds, this gives us maybe a little bit. Good idea. I'll test that. Thanks, Abdel.
Re: About scrolling speed
On Mon, Oct 23, 2006 at 07:12:13PM +0200, Asger Ottar Alstrup wrote: Finally, it could be André's new rendering scheme which is slower on your setup. Unlikely. Maybe not much of an inprovemnt, but certanily not slower. The wrong cursor size is my fault though.. Andre'
Re: About scrolling speed
On Mon, Oct 23, 2006 at 06:33:16PM +0200, Abdelrazak Younes wrote: Hi Andre, Before Denmark, the UserGuide PageDown test was at 18 seconds. Now it is at 25 seconds. I hope you have some more code in store for speed ;-) ? Andre'
Re: About scrolling speed
On Tue, Oct 24, 2006 at 10:01:43AM +0200, Juergen Vigna wrote: Hi Abdel, as much as I know before we started on fixing bugs there I was not able to scroll trough the UserGuide without hitting an Assert ;) I think that this is due to the more metrics calculations we have to do in order to fix all that scrolling bugs and with it the removal of the nullpainter which was not calculating important stuff as *normal* text, and therefore with it in place the x positions of the insets where all wrong! Hm, so maybe we should resurrect nullpainter as a non-drawing-but-metrics-computing version of the real painter. Of course, this could be as simple as having a bool flag in PainterInfo stating whether we actually want to output anything or not. Andre'
Re: About scrolling speed
On Mon, Oct 23, 2006 at 07:12:13PM +0200, Asger Ottar Alstrup wrote: > Finally, it could be André's new rendering scheme which is slower on > your setup. Unlikely. Maybe not much of an inprovemnt, but certanily not slower. The wrong cursor size is my fault though.. Andre'
Re: About scrolling speed
On Mon, Oct 23, 2006 at 06:33:16PM +0200, Abdelrazak Younes wrote: > Hi Andre, > > Before Denmark, the UserGuide PageDown test was at 18 seconds. Now it is > at 25 seconds. I hope you have some more code in store for speed ;-) ? Andre'
Re: About scrolling speed
On Tue, Oct 24, 2006 at 10:01:43AM +0200, Juergen Vigna wrote: > Hi Abdel, > > as much as I know before we started on fixing bugs there I was not able > to scroll trough the UserGuide without hitting an Assert ;) > > I think that this is due to the "more" metrics calculations we have to do > in order to fix all that scrolling bugs and with it the removal of the > nullpainter which was not calculating important stuff as *normal* text, and > therefore with it in place the x positions of the insets where all wrong! Hm, so maybe we should resurrect nullpainter as a non-drawing-but-metrics-computing version of the real painter. Of course, this could be as simple as having a bool flag in PainterInfo stating whether we actually want to output anything or not. Andre'
Re: About scrolling speed
Asger Ottar Alstrup wrote: Martin Vermeer wrote: By the way, I cannot see any text rendering difference between the two version. Looks to me that the text is anti-aliased in both cases. Looks to me that this setRenderHint() is just a helper for something else. In windows, you can turn off anti-aliasing. The Qt setRenderHint overrides that setting. When you turned anti-aliasing off, you should not comment out the setRenderHint, but rather call it with a ,false as second parameter. I notice now that one of the machines we were using had anti-aliasing turned off in windows, so that's the explanation. I guess you can remove those setRenderHint lines, so that we respect the windows setting people might have. The wikipedia link that Martin's provided indicated that this is not about anti-aliasing on/off but about different methods for anti-aliasing (sharpness versus contrast). Could you please check that before we settle on any solution? Thanks idvance, Abdel.
Re: About scrolling speed
Abdelrazak Younes wrote: The wikipedia link that Martin's provided indicated that this is not about anti-aliasing on/off but about different methods for anti-aliasing (sharpness versus contrast). Could you please check that before we settle on any solution? It is about anti-aliasing. You can see for yourself by changing the setRenderHint calls to take false as the second parameter to disable anti-aliasing. Regards, Asger
Re: About scrolling speed
On Wed, Oct 25, 2006 at 10:19:12AM +0200, Asger Ottar Alstrup wrote: Abdelrazak Younes wrote: The wikipedia link that Martin's provided indicated that this is not about anti-aliasing on/off but about different methods for anti-aliasing (sharpness versus contrast). Could you please check that before we settle on any solution? It is about anti-aliasing. You can see for yourself by changing the setRenderHint calls to take false as the second parameter to disable anti-aliasing. On Win2k, I get antialiased text only when setting it in Windows, irrespective of the setRenderHint call. -- Enrico
Re: About scrolling speed
On Oct 24, 2006, at 3:07 AM, Jean-Marc Lasgouttes wrote: Bennett == Bennett Helm [EMAIL PROTECTED] writes: Bennett Scrolling through the User's Guide (with math preview turned Bennett off) takes about 70 seconds. And does math preview improve things? Math preview doesn't seem to work (though I don't have time right now to check why). I've now updated to Qt-4.2.1. Scrolling through User's Guide takes 67 seconds -- i.e., about the same. Bennett
Re: About scrolling speed
Bennett Helm wrote: On Oct 24, 2006, at 3:07 AM, Jean-Marc Lasgouttes wrote: Bennett == Bennett Helm [EMAIL PROTECTED] writes: Bennett Scrolling through the User's Guide (with math preview turned Bennett off) takes about 70 seconds. And does math preview improve things? Math preview doesn't seem to work (though I don't have time right now to check why). I've now updated to Qt-4.2.1. Scrolling through User's Guide takes 67 seconds -- i.e., about the same. Hello Bennett, Now is a good time I think to do your useful profiles ;-) We'll try to work this out. Abdel.
Re: About scrolling speed
On Oct 25, 2006, at 11:01 AM, Abdelrazak Younes wrote: Bennett Helm wrote: On Oct 24, 2006, at 3:07 AM, Jean-Marc Lasgouttes wrote: Bennett == Bennett Helm [EMAIL PROTECTED] writes: Bennett Scrolling through the User's Guide (with math preview turned Bennett off) takes about 70 seconds. And does math preview improve things? Math preview doesn't seem to work (though I don't have time right now to check why). I've now updated to Qt-4.2.1. Scrolling through User's Guide takes 67 seconds -- i.e., about the same. Hello Bennett, Now is a good time I think to do your useful profiles ;-) We'll try to work this out. I'm compiled for profiling, but I'm swamped with other work now, so I'm not sure when I'll be able to get to it. Hopefully this week Bennett
Re: About scrolling speed
Asger Ottar Alstrup wrote: Martin Vermeer wrote: By the way, I cannot see any text rendering difference between the two version. Looks to me that the text is anti-aliased in both cases. Looks to me that this setRenderHint() is just a helper for something else. In windows, you can turn off anti-aliasing. The Qt setRenderHint overrides that setting. When you turned anti-aliasing off, you should not comment out the setRenderHint, but rather call it with a ",false" as second parameter. I notice now that one of the machines we were using had anti-aliasing turned off in windows, so that's the explanation. I guess you can remove those setRenderHint lines, so that we respect the windows setting people might have. The wikipedia link that Martin's provided indicated that this is not about anti-aliasing on/off but about different methods for anti-aliasing (sharpness versus contrast). Could you please check that before we settle on any solution? Thanks idvance, Abdel.
Re: About scrolling speed
Abdelrazak Younes wrote: The wikipedia link that Martin's provided indicated that this is not about anti-aliasing on/off but about different methods for anti-aliasing (sharpness versus contrast). Could you please check that before we settle on any solution? It is about anti-aliasing. You can see for yourself by changing the setRenderHint calls to take "false" as the second parameter to disable anti-aliasing. Regards, Asger
Re: About scrolling speed
On Wed, Oct 25, 2006 at 10:19:12AM +0200, Asger Ottar Alstrup wrote: > Abdelrazak Younes wrote: > >The wikipedia link that Martin's provided indicated that this is not > >about anti-aliasing on/off but about different methods for anti-aliasing > >(sharpness versus contrast). Could you please check that before we > >settle on any solution? > > It is about anti-aliasing. You can see for yourself by changing the > setRenderHint calls to take "false" as the second parameter to disable > anti-aliasing. On Win2k, I get antialiased text only when setting it in Windows, irrespective of the setRenderHint call. -- Enrico
Re: About scrolling speed
On Oct 24, 2006, at 3:07 AM, Jean-Marc Lasgouttes wrote: "Bennett" == Bennett Helm <[EMAIL PROTECTED]> writes: Bennett> Scrolling through the User's Guide (with math preview turned Bennett> off) takes about 70 seconds. And does math preview improve things? Math preview doesn't seem to work (though I don't have time right now to check why). I've now updated to Qt-4.2.1. Scrolling through User's Guide takes 67 seconds -- i.e., about the same. Bennett
Re: About scrolling speed
Bennett Helm wrote: On Oct 24, 2006, at 3:07 AM, Jean-Marc Lasgouttes wrote: "Bennett" == Bennett Helm <[EMAIL PROTECTED]> writes: Bennett> Scrolling through the User's Guide (with math preview turned Bennett> off) takes about 70 seconds. And does math preview improve things? Math preview doesn't seem to work (though I don't have time right now to check why). I've now updated to Qt-4.2.1. Scrolling through User's Guide takes 67 seconds -- i.e., about the same. Hello Bennett, Now is a good time I think to do your useful profiles ;-) We'll try to work this out. Abdel.
Re: About scrolling speed
On Oct 25, 2006, at 11:01 AM, Abdelrazak Younes wrote: Bennett Helm wrote: On Oct 24, 2006, at 3:07 AM, Jean-Marc Lasgouttes wrote: "Bennett" == Bennett Helm <[EMAIL PROTECTED]> writes: Bennett> Scrolling through the User's Guide (with math preview turned Bennett> off) takes about 70 seconds. And does math preview improve things? Math preview doesn't seem to work (though I don't have time right now to check why). I've now updated to Qt-4.2.1. Scrolling through User's Guide takes 67 seconds -- i.e., about the same. Hello Bennett, Now is a good time I think to do your useful profiles ;-) We'll try to work this out. I'm compiled for profiling, but I'm swamped with other work now, so I'm not sure when I'll be able to get to it. Hopefully this week Bennett
Re: About scrolling speed
Bennett == Bennett Helm [EMAIL PROTECTED] writes: Bennett Scrolling through the User's Guide (with math preview turned Bennett off) takes about 70 seconds. And does math preview improve things? JMarc
Re: About scrolling speed
Hi Abdel, as much as I know before we started on fixing bugs there I was not able to scroll trough the UserGuide without hitting an Assert ;) I think that this is due to the more metrics calculations we have to do in order to fix all that scrolling bugs and with it the removal of the nullpainter which was not calculating important stuff as *normal* text, and therefore with it in place the x positions of the insets where all wrong! Greets, Jürgen Abdelrazak Younes wrote: Hi Andre, Before Denmark, the UserGuide PageDown test was at 18 seconds. Now it is at 25 seconds. I hope you have some more code in store for speed ;-) Abdel. -- ** Würth Phoenix S.r.l. Dr. Jürgen Vigna System Integration Via Kravogl 4, I-39100 Bolzano Phone: +39 0471 564111 Direct: +39 0471 564172 Fax:+39 0471 564122 mailto:[EMAIL PROTECTED] http://www.wuerth-phoenix.com
Re: About scrolling speed
Asger Ottar Alstrup wrote: One conclusion is clear: The only sure way to get substantially faster rendering is to draw less on the screen, as discussed earlier today, either by introducing the old singlePar optimisation, or do some drawing-caching scheme. I'm of the opinion that in 1.6 we should work in be able to decide if we have to redraw just *one* row, one paragraph or the whole of the screen. Especially the *one* row stuff would increase typing performance a lot (and make it usefull again ;) Greets, Jürgen -- ** Würth Phoenix S.r.l. Dr. Jürgen Vigna System Integration Via Kravogl 4, I-39100 Bolzano Phone: +39 0471 564111 Direct: +39 0471 564172 Fax:+39 0471 564122 mailto:[EMAIL PROTECTED] http://www.wuerth-phoenix.com
Re: About scrolling speed
Juergen Vigna wrote: Hi Abdel, as much as I know before we started on fixing bugs there I was not able to scroll trough the UserGuide without hitting an Assert ;) But this has nothing to do with speed ;-) More seriously, I was of course speaking about a built that was before the scrolling breakage (which was very recent contrarily to what you might think). Please note that I am not criticizing your work. I think Andre's solution for the painter and the workarea is elegant and I would have done the same if I were not limited by the multiple frontends support. I think that this is due to the more metrics calculations we have to do in order to fix all that scrolling bugs and with it the removal of the nullpainter which was not calculating important stuff as *normal* text, and therefore with it in place the x positions of the insets where all wrong! Year, that and maybe also the font stuff that Asger mentioned yesterday. Abdel. PS: I didn't know that there was two Jurgen in this list. I was confused by the fact Jurgen (S) criticized the Denmark people while he was part of them :-)
Re: About scrolling speed
On Tue, 2006-10-24 at 10:03 +0200, Juergen Vigna wrote: Asger Ottar Alstrup wrote: One conclusion is clear: The only sure way to get substantially faster rendering is to draw less on the screen, as discussed earlier today, either by introducing the old singlePar optimisation, or do some drawing-caching scheme. I'm of the opinion that in 1.6 we should work in be able to decide if we have to redraw just *one* row, one paragraph or the whole of the screen. Especially the *one* row stuff would increase typing performance a lot (and make it usefull again ;) Greets, Jürgen But that is happening right now... see rowpainter.C, variable row_has_changed etc. You can test it by turning on debugging for Debug::PAINTING. Don't ask me what's consuming all the drawing time. Of course if you scroll page wise up or down, you're going to repaint pretty much every screenful that you see. - Martin signature.asc Description: This is a digitally signed message part
Re: About scrolling speed
Abdelrazak Younes wrote: PS: I didn't know that there was two Jurgen in this list. I was confused by the fact Jurgen (S) criticized the Denmark people while he was part of them :-) To clarify, Jürgen V. is the real Jürgen, I'm just a parvenu. Jürgen
Re: About scrolling speed
Juergen == Juergen Spitzmueller [EMAIL PROTECTED] writes: Juergen Abdelrazak Younes wrote: PS: I didn't know that there was two Jurgen in this list. I was confused by the fact Jurgen (S) criticized the Denmark people while he was part of them :-) Juergen To clarify, Jürgen V. is the real Jürgen, I'm just a Juergen parvenu. Well, these days you are the real Juergen, and he is a has been :) JMarc
Re: About scrolling speed
On Tuesday 24 October 2006 9:53 am, Juergen Spitzmueller wrote: To clarify, Jürgen V. is the real Jürgen, I'm just a parvenu. It will be fun to have at one of the meetings and say Hello, my name is Jürgen S. and I am a lyx developer. That will make the meeting look like an AA meeting, I can only laugh with such analogy. ;-) Jürgen -- José Abílio
Re: About scrolling speed
Asger == Asger Ottar Alstrup [EMAIL PROTECTED] writes: Asger It might very well be the anti-aliasing we enabled. Can you try Asger to revert that? Just search for setRenderHint and comment those Asger guys out. If this is the case, I guess anti-aliasing should be Asger optional through some preference setting. Are you sure this is anti aliasing? We always had this on... Hinting is something different. Asger The profilings were conflicting depending on the tools used, Asger but on Windows using Glowcode, my results showed that rendering Asger alone took around 70% of all time. updateMetrics was next in Asger line with 7%. The rest was scattered all over the place. Did you try documents with a lot of math in windows, like http://bugzilla.lyx.org/show_bug.cgi?id=2900 JMarc PS: thanks for the technical details. They _are_ appreciated.
Re: About scrolling speed
José Matos wrote: It will be fun to have at one of the meetings and say Hello, my name is Jürgen S. and I am a lyx developer. I'll noted that sentence for the case I'll enter reality ;-) That will make the meeting look like an AA meeting, I can only laugh with such analogy. ;-) What's AA? I only know the AAA meetings. Jürgen
Re: About scrolling speed
Juergen == Juergen Spitzmueller [EMAIL PROTECTED] writes: Juergen What's AA? I only know the AAA meetings. http://www.aa.org/en_information_aa.cfm JMarc
Re: About scrolling speed
Jean-Marc Lasgouttes wrote: http://www.aa.org/en_information_aa.cfm I see, thanks. That's indeed something different than AAA (and the LyX meetings, as far as I can see). Jürgen
Re: About scrolling speed
Asger Ottar Alstrup wrote: Abdelrazak Younes wrote: Before Denmark, the UserGuide PageDown test was at 18 seconds. Now it is at 25 seconds. I hope you have some more code in store for speed ;-) It might very well be the anti-aliasing we enabled. Can you try to revert that? Just search for setRenderHint and comment those guys out. If this is the case, I guess anti-aliasing should be optional through some preference setting. The difference is minimal but it is there, I now have better results with a different window size: - 21 s with setRenderHint(QPainter::TextAntialiasing) - 22-23 s without. So I guess this setRenderHint() is not guilty of anything. If anything, I would say the contrary. One thing that is maybe impacting the scrolling speed is the bogus cursor. I guess painting a big square square in addition does not help. By the way, I cannot see any text rendering difference between the two version. Looks to me that the text is anti-aliased in both cases. Looks to me that this setRenderHint() is just a helper for something else. Abdel.
Re: About scrolling speed
Abdelrazak Younes wrote: Asger Ottar Alstrup wrote: Abdelrazak Younes wrote: Before Denmark, the UserGuide PageDown test was at 18 seconds. Now it is at 25 seconds. I hope you have some more code in store for speed ;-) It might very well be the anti-aliasing we enabled. Can you try to revert that? Just search for setRenderHint and comment those guys out. If this is the case, I guess anti-aliasing should be optional through some preference setting. The difference is minimal but it is there, I now have better results with a different window size: - 21 s with setRenderHint(QPainter::TextAntialiasing) - 22-23 s without. So I guess this setRenderHint() is not guilty of anything. If anything, I would say the contrary. One thing that is maybe impacting the scrolling speed is the bogus cursor. I guess painting a big square square in addition does not help. Indeed... Without it, the UserGuide test is now at 15 seconds! That's better than ever! There's some bright light at the end of the tunnel :-) Abdel.
Re: About scrolling speed
Abdelrazak Younes wrote: One thing that is maybe impacting the scrolling speed is the bogus cursor. I guess painting a big square square in addition does not help. Indeed... Without it, the UserGuide test is now at 15 seconds! That's better than ever! By the way, you will need the attached patch to disable the cursor. Abdel. Index: frontends/WorkArea.C === --- frontends/WorkArea.C(revision 15520) +++ frontends/WorkArea.C(working copy) @@ -288,7 +288,7 @@ return; cursor_visible_ = true; - showCursor(x, y, h, shape); + //showCursor(x, y, h, shape); }
Re: About scrolling speed
On Tue, Oct 24, 2006 at 05:53:05PM +0200, Abdelrazak Younes wrote: Asger Ottar Alstrup wrote: ... By the way, I cannot see any text rendering difference between the two version. Looks to me that the text is anti-aliased in both cases. Looks to me that this setRenderHint() is just a helper for something else. Yes. http://en.wikipedia.org/wiki/Font_hinting - Martin pgpxT3KrYkB1o.pgp Description: PGP signature
Re: About scrolling speed
Martin Vermeer wrote: By the way, I cannot see any text rendering difference between the two version. Looks to me that the text is anti-aliased in both cases. Looks to me that this setRenderHint() is just a helper for something else. In windows, you can turn off anti-aliasing. The Qt setRenderHint overrides that setting. When you turned anti-aliasing off, you should not comment out the setRenderHint, but rather call it with a ,false as second parameter. I notice now that one of the machines we were using had anti-aliasing turned off in windows, so that's the explanation. I guess you can remove those setRenderHint lines, so that we respect the windows setting people might have. Regards, Asger
Re: About scrolling speed
> "Bennett" == Bennett Helm <[EMAIL PROTECTED]> writes: Bennett> Scrolling through the User's Guide (with math preview turned Bennett> off) takes about 70 seconds. And does math preview improve things? JMarc
Re: About scrolling speed
Hi Abdel, as much as I know before we started on fixing bugs there I was not able to scroll trough the UserGuide without hitting an Assert ;) I think that this is due to the "more" metrics calculations we have to do in order to fix all that scrolling bugs and with it the removal of the nullpainter which was not calculating important stuff as *normal* text, and therefore with it in place the x positions of the insets where all wrong! Greets, Jürgen Abdelrazak Younes wrote: Hi Andre, Before Denmark, the UserGuide PageDown test was at 18 seconds. Now it is at 25 seconds. I hope you have some more code in store for speed ;-) Abdel. -- ** Würth Phoenix S.r.l. Dr. Jürgen Vigna System Integration Via Kravogl 4, I-39100 Bolzano Phone: +39 0471 564111 Direct: +39 0471 564172 Fax:+39 0471 564122 mailto:[EMAIL PROTECTED] http://www.wuerth-phoenix.com
Re: About scrolling speed
Asger Ottar Alstrup wrote: One conclusion is clear: The only sure way to get substantially faster rendering is to draw less on the screen, as discussed earlier today, either by introducing the old singlePar optimisation, or do some drawing-caching scheme. I'm of the opinion that in 1.6 we should work in be able to decide if we have to redraw just *one* row, one paragraph or the whole of the screen. Especially the *one* row stuff would increase "typing" performance a lot (and make it usefull again ;) Greets, Jürgen -- ** Würth Phoenix S.r.l. Dr. Jürgen Vigna System Integration Via Kravogl 4, I-39100 Bolzano Phone: +39 0471 564111 Direct: +39 0471 564172 Fax:+39 0471 564122 mailto:[EMAIL PROTECTED] http://www.wuerth-phoenix.com
Re: About scrolling speed
Juergen Vigna wrote: Hi Abdel, as much as I know before we started on fixing bugs there I was not able to scroll trough the UserGuide without hitting an Assert ;) But this has nothing to do with speed ;-) More seriously, I was of course speaking about a built that was before the scrolling breakage (which was very recent contrarily to what you might think). Please note that I am not criticizing your work. I think Andre's solution for the painter and the workarea is elegant and I would have done the same if I were not limited by the multiple frontends support. I think that this is due to the "more" metrics calculations we have to do in order to fix all that scrolling bugs and with it the removal of the nullpainter which was not calculating important stuff as *normal* text, and therefore with it in place the x positions of the insets where all wrong! Year, that and maybe also the font stuff that Asger mentioned yesterday. Abdel. PS: I didn't know that there was two Jurgen in this list. I was confused by the fact Jurgen (S) criticized the Denmark people while he was part of them :-)
Re: About scrolling speed
On Tue, 2006-10-24 at 10:03 +0200, Juergen Vigna wrote: > > Asger Ottar Alstrup wrote: > > One conclusion is clear: The only sure way to get substantially faster > > rendering is to draw less on the screen, as discussed earlier today, > > either by introducing the old singlePar optimisation, or do some > > drawing-caching scheme. > > I'm of the opinion that in 1.6 we should work in be able to decide if > we have to redraw just *one* row, one paragraph or the whole of the > screen. Especially the *one* row stuff would increase "typing" performance > a lot (and make it usefull again ;) > > Greets, > > Jürgen But that is happening right now... see rowpainter.C, variable row_has_changed etc. You can test it by turning on debugging for Debug::PAINTING. Don't ask me what's consuming all the drawing time. Of course if you scroll page wise up or down, you're going to repaint pretty much every screenful that you see. - Martin signature.asc Description: This is a digitally signed message part
Re: About scrolling speed
Abdelrazak Younes wrote: > PS: I didn't know that there was two Jurgen in this list. I was confused > by the fact Jurgen (S) criticized the Denmark people while he was part > of them :-) To clarify, Jürgen V. is the "real" Jürgen, I'm just a parvenu. Jürgen
Re: About scrolling speed
> "Juergen" == Juergen Spitzmueller <[EMAIL PROTECTED]> writes: Juergen> Abdelrazak Younes wrote: >> PS: I didn't know that there was two Jurgen in this list. I was >> confused by the fact Jurgen (S) criticized the Denmark people while >> he was part of them :-) Juergen> To clarify, Jürgen V. is the "real" Jürgen, I'm just a Juergen> parvenu. Well, these days you are the real Juergen, and he is a has been :) JMarc
Re: About scrolling speed
On Tuesday 24 October 2006 9:53 am, Juergen Spitzmueller wrote: > To clarify, Jürgen V. is the "real" Jürgen, I'm just a parvenu. It will be fun to have at one of the meetings and say "Hello, my name is Jürgen S. and I am a lyx developer". That will make the meeting look like an AA meeting, I can only laugh with such analogy. ;-) > Jürgen -- José Abílio
Re: About scrolling speed
> "Asger" == Asger Ottar Alstrup <[EMAIL PROTECTED]> writes: Asger> It might very well be the anti-aliasing we enabled. Can you try Asger> to revert that? Just search for setRenderHint and comment those Asger> guys out. If this is the case, I guess anti-aliasing should be Asger> optional through some preference setting. Are you sure this is anti aliasing? We always had this on... Hinting is something different. Asger> The profilings were conflicting depending on the tools used, Asger> but on Windows using Glowcode, my results showed that rendering Asger> alone took around 70% of all time. updateMetrics was next in Asger> line with 7%. The rest was scattered all over the place. Did you try documents with a lot of math in windows, like http://bugzilla.lyx.org/show_bug.cgi?id=2900 JMarc PS: thanks for the technical details. They _are_ appreciated.
Re: About scrolling speed
José Matos wrote: > It will be fun to have at one of the meetings and say "Hello, my name is > Jürgen S. and I am a lyx developer". I'll noted that sentence for the case I'll enter reality ;-) > That will make the meeting look like an AA meeting, I can only laugh with > such analogy. ;-) What's AA? I only know the AAA meetings. Jürgen
Re: About scrolling speed
> "Juergen" == Juergen Spitzmueller <[EMAIL PROTECTED]> writes: Juergen> What's AA? I only know the AAA meetings. http://www.aa.org/en_information_aa.cfm JMarc
Re: About scrolling speed
Jean-Marc Lasgouttes wrote: > http://www.aa.org/en_information_aa.cfm I see, thanks. That's indeed something different than AAA (and the LyX meetings, as far as I can see). Jürgen
Re: About scrolling speed
Asger Ottar Alstrup wrote: Abdelrazak Younes wrote: Before Denmark, the UserGuide PageDown test was at 18 seconds. Now it is at 25 seconds. I hope you have some more code in store for speed ;-) It might very well be the anti-aliasing we enabled. Can you try to revert that? Just search for setRenderHint and comment those guys out. If this is the case, I guess anti-aliasing should be optional through some preference setting. The difference is minimal but it is there, I now have better results with a different window size: - 21 s with setRenderHint(QPainter::TextAntialiasing) - 22-23 s without. So I guess this setRenderHint() is not guilty of anything. If anything, I would say the contrary. One thing that is maybe impacting the scrolling speed is the bogus cursor. I guess painting a big square square in addition does not help. By the way, I cannot see any text rendering difference between the two version. Looks to me that the text is anti-aliased in both cases. Looks to me that this setRenderHint() is just a helper for something else. Abdel.
Re: About scrolling speed
Abdelrazak Younes wrote: Asger Ottar Alstrup wrote: Abdelrazak Younes wrote: Before Denmark, the UserGuide PageDown test was at 18 seconds. Now it is at 25 seconds. I hope you have some more code in store for speed ;-) It might very well be the anti-aliasing we enabled. Can you try to revert that? Just search for setRenderHint and comment those guys out. If this is the case, I guess anti-aliasing should be optional through some preference setting. The difference is minimal but it is there, I now have better results with a different window size: - 21 s with setRenderHint(QPainter::TextAntialiasing) - 22-23 s without. So I guess this setRenderHint() is not guilty of anything. If anything, I would say the contrary. One thing that is maybe impacting the scrolling speed is the bogus cursor. I guess painting a big square square in addition does not help. Indeed... Without it, the UserGuide test is now at 15 seconds! That's better than ever! There's some bright light at the end of the tunnel :-) Abdel.
Re: About scrolling speed
Abdelrazak Younes wrote: One thing that is maybe impacting the scrolling speed is the bogus cursor. I guess painting a big square square in addition does not help. Indeed... Without it, the UserGuide test is now at 15 seconds! That's better than ever! By the way, you will need the attached patch to disable the cursor. Abdel. Index: frontends/WorkArea.C === --- frontends/WorkArea.C(revision 15520) +++ frontends/WorkArea.C(working copy) @@ -288,7 +288,7 @@ return; cursor_visible_ = true; - showCursor(x, y, h, shape); + //showCursor(x, y, h, shape); }
Re: About scrolling speed
On Tue, Oct 24, 2006 at 05:53:05PM +0200, Abdelrazak Younes wrote: > Asger Ottar Alstrup wrote: ... > By the way, I cannot see any text rendering difference between the two > version. Looks to me that the text is anti-aliased in both cases. Looks > to me that this setRenderHint() is just a helper for something else. Yes. http://en.wikipedia.org/wiki/Font_hinting - Martin pgpxT3KrYkB1o.pgp Description: PGP signature
Re: About scrolling speed
Martin Vermeer wrote: By the way, I cannot see any text rendering difference between the two version. Looks to me that the text is anti-aliased in both cases. Looks to me that this setRenderHint() is just a helper for something else. In windows, you can turn off anti-aliasing. The Qt setRenderHint overrides that setting. When you turned anti-aliasing off, you should not comment out the setRenderHint, but rather call it with a ",false" as second parameter. I notice now that one of the machines we were using had anti-aliasing turned off in windows, so that's the explanation. I guess you can remove those setRenderHint lines, so that we respect the windows setting people might have. Regards, Asger
About scrolling speed
Hi Andre, Before Denmark, the UserGuide PageDown test was at 18 seconds. Now it is at 25 seconds. I hope you have some more code in store for speed ;-) Abdel.
Re: About scrolling speed
Abdelrazak Younes wrote: Before Denmark, the UserGuide PageDown test was at 18 seconds. Now it is at 25 seconds. I hope you have some more code in store for speed ;-) It might very well be the anti-aliasing we enabled. Can you try to revert that? Just search for setRenderHint and comment those guys out. If this is the case, I guess anti-aliasing should be optional through some preference setting. If that doesn't help, you should try to revert the nullpainter change and test again. It might be those extra paragraphs that have to be drawn outside the screen for real now that makes the difference. Finally, it could be André's new rendering scheme which is slower on your setup. Also, we'd like some MacOSX users to try the new code before we make the final judgement. One conclusion is clear: The only sure way to get substantially faster rendering is to draw less on the screen, as discussed earlier today, either by introducing the old singlePar optimisation, or do some drawing-caching scheme. The profilings were conflicting depending on the tools used, but on Windows using Glowcode, my results showed that rendering alone took around 70% of all time. updateMetrics was next in line with 7%. The rest was scattered all over the place. Regards, Asger
Re: About scrolling speed
On Oct 23, 2006, at 1:12 PM, Asger Ottar Alstrup wrote: Abdelrazak Younes wrote: Before Denmark, the UserGuide PageDown test was at 18 seconds. Now it is at 25 seconds. I hope you have some more code in store for speed ;-) ... Also, we'd like some MacOSX users to try the new code before we make the final judgement. Not sure about relative speed, and I don't have much time now. But my *impression* is that is pretty much the same, which is to say still much too slow. (Typing in particular makes it unusable: once more than a short paragraph is displayed on screen, there's noticeable screen lag, and it reaches more than a second per character once the application window is full of text.) Scrolling through the User's Guide (with math preview turned off) takes about 70 seconds. Bennett
About scrolling speed
Hi Andre, Before Denmark, the UserGuide PageDown test was at 18 seconds. Now it is at 25 seconds. I hope you have some more code in store for speed ;-) Abdel.
Re: About scrolling speed
Abdelrazak Younes wrote: Before Denmark, the UserGuide PageDown test was at 18 seconds. Now it is at 25 seconds. I hope you have some more code in store for speed ;-) It might very well be the anti-aliasing we enabled. Can you try to revert that? Just search for setRenderHint and comment those guys out. If this is the case, I guess anti-aliasing should be optional through some preference setting. If that doesn't help, you should try to revert the nullpainter change and test again. It might be those extra paragraphs that have to be drawn outside the screen for real now that makes the difference. Finally, it could be André's new rendering scheme which is slower on your setup. Also, we'd like some MacOSX users to try the new code before we make the final judgement. One conclusion is clear: The only sure way to get substantially faster rendering is to draw less on the screen, as discussed earlier today, either by introducing the old singlePar optimisation, or do some drawing-caching scheme. The profilings were conflicting depending on the tools used, but on Windows using Glowcode, my results showed that rendering alone took around 70% of all time. updateMetrics was next in line with 7%. The rest was scattered all over the place. Regards, Asger
Re: About scrolling speed
On Oct 23, 2006, at 1:12 PM, Asger Ottar Alstrup wrote: Abdelrazak Younes wrote: Before Denmark, the UserGuide PageDown test was at 18 seconds. Now it is at 25 seconds. I hope you have some more code in store for speed ;-) ... Also, we'd like some MacOSX users to try the new code before we make the final judgement. Not sure about relative speed, and I don't have much time now. But my *impression* is that is pretty much the same, which is to say still much too slow. (Typing in particular makes it unusable: once more than a short paragraph is displayed on screen, there's noticeable screen lag, and it reaches more than a second per character once the application window is full of text.) Scrolling through the User's Guide (with math preview turned off) takes about 70 seconds. Bennett