MozillaZine

Direct2D/DirectWrite Accelerated Rendering For Firefox

Discussion about official Mozilla Firefox builds
_Alexander

User avatar
 
Posts: 1197
Joined: April 1st, 2010, 2:24 pm
Location: Your augmented reality

Post Posted October 30th, 2011, 10:02 am

If IE9 works fine then it is not a D2D problem, but a D2D implementation problem.

EDIT: It seems that the developers with the graphic know-how are currently working on OpenGL/WebGL - there seems to be relatively little work done (that gets into Nightlies) for D2D/D3D10 layers and the Direct3D9. Hell the root cause of backing out the GIF performance fixes might have been machines running old drivers that offer bad Direct3D9 performance (something that I recall them noticing a long time ago).
http://magneticpudding.com/ <- My Blog
i5 3570k @ 4.5 Ghz / NV 660 / 32GB DDR3 / 1080p LCD / SSD (120 + 180) / W8 ||| Atom N270 / NV ION / 3GB DDR3 / SSD / 1366x768 / W8

spirits247
 
Posts: 59
Joined: October 6th, 2009, 6:30 am

Post Posted November 2nd, 2011, 9:27 am

Here is the reported issue with Bugzilla - no response out of ATI, unlike nVidia who at least have addressed the issue.

https://bugzilla.mozilla.org/show_bug.cgi?id=698301

Hope this gets confirmed.

Zlip792

User avatar
 
Posts: 1338
Joined: May 7th, 2011, 1:47 pm
Location: Pakistan

Post Posted November 2nd, 2011, 9:59 am

Mozilla folks who are facing TDR issues with 280 and 285 series should try this Registry trick under Windows 7:
Code: Select all
Go here in your registry:

[HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\GraphicsDrivers]

Then create a a REG_DWORD file called TdrDelay and set the value to 60

ex. "TdrDelay"=dword:00000060 (60 seconds)

Then navigate to
[HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\GraphicsDrivers\DCI]

and Create a Value called

"Timeout"=dword:00000060

Cru_N_cher
 
Posts: 377
Joined: January 31st, 2010, 11:15 am

Post Posted November 2nd, 2011, 11:44 am

@Hera
Sure it's a D2D implementation issue problem seems that they care more about the techdemo performance including most ie speed reading ect then actual current real performance on different existing pages rendering.
Also the GIF patch was taken out again because of a 3.3% regression in tscroll :( (back on the drawing board) though this seems only to be XP specific so Vista/7 shouldn't suffer from this (in theory) so without taking responsibility this should include the current fix http://ftp.mozilla.org/pub/mozilla.org/ ... f492ead84/
if its working is easy to see it should reduce core utilization on testcases by half compared to 1 whole core still not optimal result for Windows Vista/7 but working on this is going so deep and can brake all kinds of stuff that this is already a very good result under those circumstances :)

Omega X

User avatar
 
Posts: 7470
Joined: October 18th, 2007, 2:38 pm
Location: A Parallel Dimension...

Post Posted November 2nd, 2011, 11:51 am

That bug was not backed out. Its being supplemented by a separate fix.
https://bugzilla.mozilla.org/show_bug.cgi?id=691775#c4
Latest: Firefox/40.0.2 *ESR/38.2.0 - Mobile/40.0 - Thunderbird/38.2.0 - SeaMonkey/2.33.1
Nightly: Nightly/43.0 - Mobile/43.0 - Daily/43.0 - SeaMonkey/2.35a1

Cru_N_cher
 
Posts: 377
Joined: January 31st, 2010, 11:15 am

Post Posted November 2nd, 2011, 11:59 am

ahh yes sorry hmm shouldn't it be in the current nightlies then by now ?

Omega X

User avatar
 
Posts: 7470
Joined: October 18th, 2007, 2:38 pm
Location: A Parallel Dimension...

Post Posted November 2nd, 2011, 12:01 pm

No, inbound pushed after the cutoff. It'll be in tomorrow's nightly.
Latest: Firefox/40.0.2 *ESR/38.2.0 - Mobile/40.0 - Thunderbird/38.2.0 - SeaMonkey/2.33.1
Nightly: Nightly/43.0 - Mobile/43.0 - Daily/43.0 - SeaMonkey/2.35a1

Cru_N_cher
 
Posts: 377
Joined: January 31st, 2010, 11:15 am

Post Posted November 2nd, 2011, 12:05 pm

Omega X do you know if they test the D2D part separately in talos, and what is used to test performance (especially GPU cycles and rendering bottlenecks to slow or inefficient GPU/CPU copy) is it a generic test or automated test of different GPUs even ?

I would really love to see some chart showing the CPU vs GPU Performance difference on non techdemos, my perception at least on Sandy Bridge is the CPU beats the GPU by a lot in Generic Browsing Percepted Performance currently (completely different scenarios) and that's not the case with IE9 (it can hold @ least Performance stable on the GPU, and it yet has to show to overload it in any of those tests Firefox D2D does it very often). Though Performance of Chrome/Chromium doesn't look much better in this case when ANGLE gets used.

Omega X

User avatar
 
Posts: 7470
Joined: October 18th, 2007, 2:38 pm
Location: A Parallel Dimension...

Post Posted November 2nd, 2011, 1:01 pm

I know that there are automated tests for HWA. And I know that they do tests the various backends regularly when there's work being done on them. But the public graph server doesn't show specific D2D test results AFAIK. Either that or I ended up at the wrong resource.
Latest: Firefox/40.0.2 *ESR/38.2.0 - Mobile/40.0 - Thunderbird/38.2.0 - SeaMonkey/2.33.1
Nightly: Nightly/43.0 - Mobile/43.0 - Daily/43.0 - SeaMonkey/2.35a1

Cru_N_cher
 
Posts: 377
Joined: January 31st, 2010, 11:15 am

Post Posted November 3rd, 2011, 8:03 am

Omega X did i understood correctly and they backed it out completely now because of that regression so it wont be neither in Aurora nor Nigthly, im just testing it myself and the improvement is even better then the first fix about 3x lower pressure on the CPU with the GIF testcase :)

Though with Direct2D it looks even more disastrous compared to the CPU load (which is with the patch and Software rendering on Aero @ 8%) we have around 40% GPU load (almost half the GPU) (Direct2d,DWM) in those GIF testcase (not dependent on the CPU patch) :(

_Alexander

User avatar
 
Posts: 1197
Joined: April 1st, 2010, 2:24 pm
Location: Your augmented reality

Post Posted November 3rd, 2011, 9:29 am

Cru_N_cher wrote:Omega X did i understood correctly and they backed it out completely now because of that regression so it wont be neither in Aurora nor Nigthly, im just testing it myself and the improvement is even better then the first fix about 3x lower pressure on the CPU with the GIF testcase :)

Though with Direct2D it looks even more disastrous compared to the CPU load (which is with the patch and Software rendering on Aero @ 8%) we have around 40% GPU load (almost half the GPU) (Direct2d,DWM) in those GIF testcase (not dependent on the CPU patch) :(

I note that there has essentially been no progress on GIF performance for ~half a year since I filed the original bug report, then eventually they started working on it and still do not see it as a priority.
There is a grave lack of care on Mozilla's part for bugs which talk about common web elements causing Firefox to saturate the CPU.
:evil:
http://magneticpudding.com/ <- My Blog
i5 3570k @ 4.5 Ghz / NV 660 / 32GB DDR3 / 1080p LCD / SSD (120 + 180) / W8 ||| Atom N270 / NV ION / 3GB DDR3 / SSD / 1366x768 / W8

Omega X

User avatar
 
Posts: 7470
Joined: October 18th, 2007, 2:38 pm
Location: A Parallel Dimension...

Post Posted November 3rd, 2011, 11:31 am

Cru_N_cher wrote:Omega X did i understood correctly and they backed it out completely now because of that regression so it wont be neither in Aurora nor Nigthly, im just testing it myself and the improvement is even better then the first fix about 3x lower pressure on the CPU with the GIF testcase :)

Though with Direct2D it looks even more disastrous compared to the CPU load (which is with the patch and Software rendering on Aero @ 8%) we have around 40% GPU load (almost half the GPU) (Direct2d,DWM) in those GIF testcase (not dependent on the CPU patch)


The backout came much later even though there's a working patch for the XP issue.

With the patch GPU load is 15% here and it spikes intermittently, not constant. Firefox's CPU usage was 1%.
Latest: Firefox/40.0.2 *ESR/38.2.0 - Mobile/40.0 - Thunderbird/38.2.0 - SeaMonkey/2.33.1
Nightly: Nightly/43.0 - Mobile/43.0 - Daily/43.0 - SeaMonkey/2.35a1

_Alexander

User avatar
 
Posts: 1197
Joined: April 1st, 2010, 2:24 pm
Location: Your augmented reality

Post Posted November 3rd, 2011, 1:03 pm

Omega X wrote:
Cru_N_cher wrote:Omega X did i understood correctly and they backed it out completely now because of that regression so it wont be neither in Aurora nor Nigthly, im just testing it myself and the improvement is even better then the first fix about 3x lower pressure on the CPU with the GIF testcase :)

Though with Direct2D it looks even more disastrous compared to the CPU load (which is with the patch and Software rendering on Aero @ 8%) we have around 40% GPU load (almost half the GPU) (Direct2d,DWM) in those GIF testcase (not dependent on the CPU patch)


The backout came much later even though there's a working patch for the XP issue.

With the patch GPU load is 15% here and it spikes intermittently, not constant. Firefox's CPU usage was 1%.

This fares how to IE9?
http://magneticpudding.com/ <- My Blog
i5 3570k @ 4.5 Ghz / NV 660 / 32GB DDR3 / 1080p LCD / SSD (120 + 180) / W8 ||| Atom N270 / NV ION / 3GB DDR3 / SSD / 1366x768 / W8

RNiK

User avatar
 
Posts: 561
Joined: August 9th, 2006, 6:47 am
Location: Forette City, Italy

Post Posted November 3rd, 2011, 1:57 pm

Hera wrote:
Cru_N_cher wrote:Omega X did i understood correctly and they backed it out completely now because of that regression so it wont be neither in Aurora nor Nigthly, im just testing it myself and the improvement is even better then the first fix about 3x lower pressure on the CPU with the GIF testcase :)

Though with Direct2D it looks even more disastrous compared to the CPU load (which is with the patch and Software rendering on Aero @ 8%) we have around 40% GPU load (almost half the GPU) (Direct2d,DWM) in those GIF testcase (not dependent on the CPU patch) :(

I note that there has essentially been no progress on GIF performance for ~half a year since I filed the original bug report, then eventually they started working on it and still do not see it as a priority.
There is a grave lack of care on Mozilla's part for bugs which talk about common web elements causing Firefox to saturate the CPU.
:evil:

You should look better. :P

There's been lot of activity and a HUGE amount of work around Bug 666446. However, fixing the bug revealed lots of weak aspects about Firefox image handling, so the developers are considering a more complex approach to the problem. In the long-term view development this is good 8-), unfortunately we will have to wait some more weeks to see it land. :|
MondoWin ==> Italian site for information about MS Windows tweaking

spirits247
 
Posts: 59
Joined: October 6th, 2009, 6:30 am

Post Posted November 4th, 2011, 7:12 am

Since this thread is discussing many diffferent aspects of Hardware Acceleration, I may start a new thread just covering Firefox and GPU temps/fan speeds whilst scrolling to stick to the same topic.

There is a temporary solution:

http://forums.mozillazine.org/viewtopic.php?f=38&t=2340977

Creating a profile and switching between to force the card to stay at low clocks whilst surfing with Firefox.

I find this so annoying - fan and temps going up and down whilst surfing/scrolling even with hardware acceleration disabled, how does stuff like this get past testing? It was immediately obvious to me after two minutes of using Firefox.

Return to Firefox Builds


Who is online

Users browsing this forum: Bing [Bot] and 4 guests