Welcome, Guest. Please login or register.

Author Topic: @Bernd_afa: OWB (68k) optimizations  (Read 16133 times)

Description:

0 Members and 1 Guest are viewing this topic.

Offline unusedunused

  • Sr. Member
  • ****
  • Join Date: Nov 2005
  • Posts: 479
    • Show only replies by unusedunused
Re: @Bernd_afa: OWB (68k) optimizations
« Reply #44 from previous page: December 23, 2009, 06:55:28 PM »
>There's a difference between speed and responsivity, but that must be too subtle for you,

I know that, but when you read th second post, he write that OWB on OS4 is too responsive.

""""
http://amigaworld.net/modules/newbb/viewtopic.php?topic_id=30292&forum=32#526731

I have to say that isn't true. I can click on links, scroll, etc while it is still loading some images (and maybe other stuff). But it does seem to need to load a lot of the page first, before rendering
"""

As soon OWB on 68k or AROS show the page the first time it is responsive, at least with a delay of 200 ms.but until OWB show a page it need much more time than other browsers.

when i try 5 times to reload cnn on safari and get times to first show of 2-3 sec and then try with OWB and  AROS and then i need 7-8sec then its a huge noticable diffrence.
 

Offline 0amigan0Topic starter

  • Full Member
  • ***
  • Join Date: Dec 2006
  • Posts: 109
    • Show only replies by 0amigan0
Re: @Bernd_afa: OWB (68k) optimizations
« Reply #45 on: December 29, 2009, 11:32:42 AM »
@Fab:

You have a Linux cross-compile environment ready to go.
YOU might try your hand at it, if Bernd doesn't want to hear.

You surely will be idolized by us 68k freaks. :)
 

Offline Piru

  • \' union select name,pwd--
  • Hero Member
  • *****
  • Join Date: Aug 2002
  • Posts: 6946
    • Show only replies by Piru
    • http://www.iki.fi/sintonen/
Re: @Bernd_afa: OWB (68k) optimizations
« Reply #46 on: December 29, 2009, 03:34:51 PM »
Quote from: 0amigan0;535334
@Fab:

You have a Linux cross-compile environment ready to go.
YOU might try your hand at it

You forget that the target for that cross compiler is MorphOS PPC.
 

Offline 0amigan0Topic starter

  • Full Member
  • ***
  • Join Date: Dec 2006
  • Posts: 109
    • Show only replies by 0amigan0
Re: @Bernd_afa: OWB (68k) optimizations
« Reply #47 on: December 29, 2009, 04:27:28 PM »
Quote from: Piru;535353
You forget that the target for that cross compiler is MorphOS PPC.


I didn't know that.
well, never mind. :)
 

Offline Fab

  • Full Member
  • ***
  • Join Date: Jun 2009
  • Posts: 217
    • Show only replies by Fab
Re: @Bernd_afa: OWB (68k) optimizations
« Reply #48 on: December 29, 2009, 04:46:57 PM »
Quote from: 0amigan0;535361
I didn't know that.
well, never mind. :)

Indeed, my setup isn't ready at all for a 68k target. Not that it would be impossible of course, but sdk stuff always a tedious task (at least for me :)).

In any case, if a 3.x developer with a "correct" gcc4 + libnix setup (no ixemul, thanks :)) wants to give it a try, i'd be glad to help him. It won't be a very trivial task, so it needs a bit knowledge (i.e more than configure/make :)).

Some things would have to be adapted or "degraded" to work on OS3.x/MUI3.8 (network thread support, MUI tabs or list methods, ...), but it's just some additional work, nothing impossible (and if it's too scary, one could always start with the AROS port first, which should compile more easily on 3.x). About the speed it would have on a real amiga, i couldn't tell though, but it would certainly be substancially than current SDL version in any case, at least regarding scrolling.
 

Offline 0amigan0Topic starter

  • Full Member
  • ***
  • Join Date: Dec 2006
  • Posts: 109
    • Show only replies by 0amigan0
Re: @Bernd_afa: OWB (68k) optimizations
« Reply #49 on: December 30, 2009, 12:42:00 PM »
@Fab:

I've found this cross-compiler setup guide http://cross.zerohero.se/os3.html
It mentions gcc 3.4.0; where can I found gcc 4 ? Are the tools he use, up-to-date ??

Can I use these mui 3.8 developer files http://aminet.net/dev/mui/mui38dev.lha ? Provided of course I can manage to setup the cross-environment. :)
 

Offline unusedunused

  • Sr. Member
  • ****
  • Join Date: Nov 2005
  • Posts: 479
    • Show only replies by unusedunused
Re: @Bernd_afa: OWB (68k) optimizations
« Reply #50 on: December 31, 2009, 12:25:55 PM »
>but it would certainly be substancially than current SDL version in any case, at least >regarding scrolling.

I look at the OWb 68k 1.2 (SDL version) and test without JIT.Scrolling is very slow need ca 1,3 sec for 1 frame.netsurf scroll also when pixel format not match with at least 8 fps on my System and disable JIT.

But when use the JIT OWB 1.2 scroll with 10-15 fps and thats lots faster as AROS Version run on my System.AROS Version scroll only with 3 fps on vmware.

Now when use OWB 68k 1.4 that use a enhanced SDL Version.It scroll on my system without JIT with 10-15 fps thats fast as netsurf do when pixelformat in SDL is ok.

But!!!

The OWB 1.4 need for a page load 2* longer as the OWB 1.2 full SDL Version.

So i think best is to fix in the SDL Version the pixel Match Problem.It seem SDL is in general very slow when it must convert pixel Formats.Same Problem is maybe with Cairo.

on 68k some GFX Cards work in RGBA some in BGRA.Cairo use same as opengl intern RGBA Pixel Format.any other must convert and get speedloss.

Do you know if its possible to set the Pixelformat of the OWB render engine to any Pixelformat, or use the OWB render engine a fixed Format ?

the Netsurf Render engine work only in the ARGB Pixelformat.This Pixelformat does Cairo not support native, so Cairo get some speedloss because it need convert internaly

What Pixel Format MOS use ?
Can MOS open screens of diffrent Pixelformats ?
« Last Edit: December 31, 2009, 12:31:09 PM by bernd_afa »
 

Offline Fab

  • Full Member
  • ***
  • Join Date: Jun 2009
  • Posts: 217
    • Show only replies by Fab
Re: @Bernd_afa: OWB (68k) optimizations
« Reply #51 on: December 31, 2009, 02:28:14 PM »
Quote from: bernd_afa;535535
>but it would certainly be substancially than current SDL version in any case, at least >regarding scrolling.

I look at the OWb 68k 1.2 (SDL version) and test without JIT.Scrolling is very slow need ca 1,3 sec for 1 frame.netsurf scroll also when pixel format not match with at least 8 fps on my System and disable JIT.

But when use the JIT OWB 1.2 scroll with 10-15 fps and thats lots faster as AROS Version run on my System.AROS Version scroll only with 3 fps on vmware.

Now when use OWB 68k 1.4 that use a enhanced SDL Version.It scroll on my system without JIT with 10-15 fps thats fast as netsurf do when pixelformat in SDL is ok.



The thing is when you scroll 20 pixels or so per step, it should be much faster, when a backstore scrolling method is used, which isn't implemented in the SDL backend.
For instance, here, on peg2, on average, with a 1024x768 window, rendering+scrolling a 1024x1 line takes about 2 ms. Rendering the whole page can take about 40-100ms (depending on content). So, implementing a proper scrolling method can effectively make scrolling dozens times faster (of course, when you scroll with a very big step (page per page or more), it doesn't have any benefit anymore).

Quote

But!!!

The OWB 1.4 need for a page load 2* longer as the OWB 1.2 full SDL Version.



No idea, but like i said several times, there are many things to be done in network layer.


Quote

So i think best is to fix in the SDL Version the pixel Match Problem.It seem SDL is in general very slow when it must convert pixel Formats.Same Problem is maybe with Cairo.

on 68k some GFX Cards work in RGBA some in BGRA.Cairo use same as opengl intern RGBA Pixel Format.any other must convert and get speedloss.

Do you know if its possible to set the Pixelformat of the OWB render engine to any Pixelformat, or use the OWB render engine a fixed Format ?

the Netsurf Render engine work only in the ARGB Pixelformat.This Pixelformat does Cairo not support native, so Cairo get some speedloss because it need convert internaly

What Pixel Format MOS use ?
Can MOS open screens of diffrent Pixelformats ?


The SDL version of OWB didn't use cairo at all. It used sdlgfx to render (since a few weeks, it can also use cairo to render, blitting still happens in SDL of course).

And by the way, Cairo supports ARGB pixel format.
 

Offline unusedunused

  • Sr. Member
  • ****
  • Join Date: Nov 2005
  • Posts: 479
    • Show only replies by unusedunused
Re: @Bernd_afa: OWB (68k) optimizations
« Reply #52 on: January 02, 2010, 10:16:56 AM »
>when a backstore scrolling method is used, which isn't implemented in the SDL backend.

but its easy to do so.
see code in netsurf

http://source.netsurf-browser.org/trunk/libnsfb/src/frontend_sdl.c?r1=9719&r2=9720

and when use HW surface the System blitter is used

> No idea, but like i said several times, there are many things to be done in network layer.

when a test is done 10 times and always 10 times 1.2 work 2* faster as 1.4 and times do not differ more than 10-20% each test then can in real world 100% sure 1.2 is faster.
 

Offline 0amigan0Topic starter

  • Full Member
  • ***
  • Join Date: Dec 2006
  • Posts: 109
    • Show only replies by 0amigan0
Re: @Bernd_afa: OWB (68k) optimizations
« Reply #53 on: January 10, 2010, 11:35:26 AM »
@Fab:

Can I use this guide ( http://www.zerohero.se/cross/os3.html ) in order to setup a cross-compiler environment ??
Are the tools indicated, up-to-date ??
 

Offline samo79

  • Full Member
  • ***
  • Join Date: Jul 2002
  • Posts: 124
    • Show only replies by samo79
    • http://www.betatesting.it/backforthefuture/
Re: @Bernd_afa: OWB (68k) optimizations
« Reply #54 on: January 11, 2010, 06:07:57 AM »
Quote from: bernd_afa;534553
If OWB is faster, then wy netsurf is a google summer of Code projecrt since 2 years and there is no OWB Version in a Linux distri ?.the big disadvantage of netsurf is currently missung java script but its in Linux Distri.

netsurf is in Linux distri.Wy not OWB ?


Because OWB is a Sand-Labs project initially dedicated only to CE device and it was/is builded around WebKit, you didn't found it on Linux yet just because they are use a more complete core or complete browser around the same WebKit core

Having said that you can't compare it with NetSurf that it is actually a "spare time" project maded by a little and passionate community, mostly RiscOS developers ...

As Fab say actually OWB on OS4/MOS are MUCH faster than any NetSurf release and it is constantly updated.

What I like to see (maybe one day) is a MUI version that can cover all the Amiga like systems, expecially for AmigaOS 3.x, that because when OS4 Reaction class are constantly updated, 68k class not :-)

P.S.
I'm a NetSurf 3.x betatester on OS4 so I know very well what I say ;-)
 

Offline unusedunused

  • Sr. Member
  • ****
  • Join Date: Nov 2005
  • Posts: 479
    • Show only replies by unusedunused
Re: @Bernd_afa: OWB (68k) optimizations
« Reply #55 on: January 12, 2010, 06:28:51 PM »
Quote from: samo79;537627

P.S.
I'm a NetSurf 3.x betatester on OS4 so I know very well what I say ;-)


You cant compare the OS4 version that use another layer (Cairo) with the 68k Verison.In some post is written that 68k netsurf was faster as OS4 netsurf btw.

netsurf 68k can also show files offline when do file://dh2:test.html to show dh2:test.html.
but must be sure that all references gone to file

I have measurement enough and i notice always the same that other have written too (link in this thread).OWB show a page much later as all other browsers.I see this on a Video too.If this video is not good, then wy there is no better.

And its near impossible that this in MOS or OS4 is better

In Amiga land its known everybody want have the best system and you cant believe what cant see with own eyes.

But i not buy a MOS or OS4 system to see same as i can see on AROS.Maybe OWB scroll on MOS or OS4 faster, but how many time is need to show the page first doesnt depend on GFX Speed because also the AROS Version can render a page in 300 ms.

>What I like to see (maybe one day) is a MUI version that can cover all the Amiga like >systems, expecially for AmigaOS 3.x, that because when OS4 Reaction class are >constantly updated, 68k class not :-)

I like to see this too, but as long every system want fight for more user with a better browser its lots work when all systems do their own Version.

And i have hope in 1-2 years when the MOS/OS4 developers maybe see there cant make enough money with the OS they stop the lots work spend on browser.

and then the last existing users/dev maybe do together bring a actual browser.
I can also port in 4 years OWB to 68k, maybe then OWB is better
« Last Edit: January 12, 2010, 06:35:24 PM by bernd_afa »
 

Offline Fab

  • Full Member
  • ***
  • Join Date: Jun 2009
  • Posts: 217
    • Show only replies by Fab
Re: @Bernd_afa: OWB (68k) optimizations
« Reply #56 on: January 12, 2010, 07:33:18 PM »
Quote from: bernd_afa;537896


I have measurement enough and i notice always the same that other have written too (link in this thread).OWB show a page much later as all other browsers.I see this on a Video too.If this video is not good, then wy there is no better.

And its near impossible that this in MOS or OS4 is better

In Amiga land its known everybody want have the best system and you cant believe what cant see with own eyes.



Will you finally understand that there are some differences between all these OWB ports? In MorphOS port, I changed a few things because the default values were just not adapted for "modern" inet speeds (that small change can actually make transfers 20 times faster, going from 400kB/s to 8MB/s locally ...). And more importantly, I threaded the network to give a better responsivity.

Recently, the AROS port implemented backstore scrolling, and also the trivial (but needed) network speedup. So you might try it again, and finally understand that the plain SDL OWB port doesn't mean nothing at all, performance-wise, and that some work is needed, just like in netsurf...

Quote

I like to see this too, but as long every system want fight for more user with a better browser its lots work when all systems do their own Version.

And i have hope in 1-2 years when the MOS/OS4 developers maybe see there cant make enough money with the OS they stop the lots work spend on browser.

and then the last existing users/dev maybe do together bring a actual browser.
I can also port in 4 years OWB to 68k, maybe then OWB is better


Or maybe that in 4 years, you'll have finally understood that not everyone wants to develop on a dead OS, and that if you really want it, you could port it yourself, instead of waiting for someone to do it. And by the way, I offered my help for a 68k port, and someone contacted me about it.
 

Offline Piru

  • \' union select name,pwd--
  • Hero Member
  • *****
  • Join Date: Aug 2002
  • Posts: 6946
    • Show only replies by Piru
    • http://www.iki.fi/sintonen/
Re: @Bernd_afa: OWB (68k) optimizations
« Reply #57 on: January 12, 2010, 08:01:52 PM »
Quote from: bernd_afa;537896
And i have hope in 1-2 years when the MOS/OS4 developers maybe see there cant make enough money with the OS they stop the lots work spend on browser.

As "making money" isn't the goal I'm afraid we can't grant that wish.
 

Offline samo79

  • Full Member
  • ***
  • Join Date: Jul 2002
  • Posts: 124
    • Show only replies by samo79
    • http://www.betatesting.it/backforthefuture/
Re: @Bernd_afa: OWB (68k) optimizations
« Reply #58 on: January 12, 2010, 08:10:18 PM »
Pardon but how you can't make a speed comparision in a YouTube video ??

I tried also NetSurf on 68k but I can't say if speed are really better than the OS4 release as I play with it on an old x86 PC in emulation via WinUAE, but aniway in one thing i'm sure 100%, OWB under OS4 is perfectly for navigation and there aren't any issue on this way

Maybe we can speak about UI problems or other issue but not about time rendering

Latest OWB's rendering on my Flex 800 is faster than Firefox 3.0 on an high-end PCs, but you are free to believe anythings ;-)
 

Offline kolla

Re: @Bernd_afa: OWB (68k) optimizations
« Reply #59 on: January 12, 2010, 10:37:38 PM »
Quote from: Piru;537908
As "making money" isn't the goal I'm afraid we can't grant that wish.


So the price tag of MorphOS is there to keep the number of whining users down - that's what I've been suspecting :lol:
B5D6A1D019D5D45BCC56F4782AC220D8B3E2A6CC
---
A3000/060CSPPC+CVPPC/128MB + 256MB BigRAM/Deneb USB
A4000/CS060/Mediator4000Di/Voodoo5/128MB
A1200/Blz1260/IndyAGA/192MB
A1200/Blz1260/64MB
A1200/Blz1230III/32MB
A1200/ACA1221
A600/V600v2/Subway USB
A600/Apollo630/32MB
A600/A6095
CD32/SX32/32MB/Plipbox
CD32/TF328
A500/V500v2
A500/MTec520
CDTV
MiSTer, MiST, FleaFPGAs and original Minimig
Peg1, SAM440 and Mac minis with MorphOS