653206 - Using "layers.acceleration.force-enabled=true" is SLOWER than with it set to false

Status: UNCONFIRMED

(Core :: Graphics, defect)

Description

[donrhummy]

User-Agent:       Mozilla/5.0 (X11; Linux i686; rv:2.0) Gecko/20100101 Firefox/4.0
Build Identifier: Mozilla/5.0 (X11; Linux i686; rv:2.0) Gecko/20100101 Firefox/4.0

Using this page:
http://ie.microsoft.com/testdrive/Performance/FishBowl/Default.html

It's FASTER without the "layers.acceleration.force-enabled=true":

WITH: 28-32fps (10 fish)
WITHOUT: 41-43fps (10 fish)

USING:
DISPLAY
----------
  Vendor:  nVidia Corporation
  Model:  GeForce 8400 GS
  2D driver:  nvidia
  3D driver:  NVIDIA 256.53

Processor (CPU)
-----------------
   AMD Sempron(tm) Processor 3100+
  Speed:  1,808.55 MHz
  Temperature:  40 °C

OS/Desktop
--------------
openSUSE 11.3
KDE 4.4.4

Reproducible: Always

Steps to Reproduce:
1. Visit the url provided
2. Change "layers.acceleration.force-enabled=true"
3. Visit the site again

Actual Results:  
Fewer fps with 10 fish (and also more fish as well)

Expected Results:  
Faster or the same speed.

Comment 1

[Benoit Jacob [:bjacob] (mostly away)]

I can't reproduce this problem here with my Quadro FX 880 M, NVIDIA 195.36.31, Core i7, today's Nightly, linux x86-64, debian stable.

Here, enabling GL layers does give better results:

         | With layers accel | Without layers accel
---------+-------------------+----------------------
10 Fish  | 60 FPS            | 60 FPS
100 Fish | 60 FPS            | 40 FPS
250 Fish | 13 FPS            | 10 FPS

One thing that might explain why it works better for me is that I have a much faster CPU. If we had a bug causing some operation to run on the CPU, that could make things slow on your machine and not on mine.

Also, there is something weird in this benchmark: why does the frame rate fall so low at 250 fish compared to 100 fish?

Comment 2

[Bas Schouten (:bas.schouten)]

(In reply to comment #1)
> I can't reproduce this problem here with my Quadro FX 880 M, NVIDIA 195.36.31,
> Core i7, today's Nightly, linux x86-64, debian stable.
> 
> Here, enabling GL layers does give better results:
> 
>          | With layers accel | Without layers accel
> ---------+-------------------+----------------------
> 10 Fish  | 60 FPS            | 60 FPS
> 100 Fish | 60 FPS            | 40 FPS
> 250 Fish | 13 FPS            | 10 FPS
> 
> One thing that might explain why it works better for me is that I have a much
> faster CPU. If we had a bug causing some operation to run on the CPU, that
> could make things slow on your machine and not on mine.
> 
> Also, there is something weird in this benchmark: why does the frame rate fall
> so low at 250 fish compared to 100 fish?

Could this be hitting readback somewhere? In that case a fast GPU would cause less blockage on the x-render work.

Comment 3

[Benoit Jacob [:bjacob] (mostly away)]

> > If we had a bug causing some operation to run on the CPU, that
> > could make things slow on your machine and not on mine.
> 
> Could this be hitting readback somewhere? In that case a fast GPU would cause
> less blockage on the x-render work.

That's what I had in mind, yes.

Comment 4

[:aceman]

Do you also see too high memory allocation as in bug 678940?

Comment 5

[Matt Woodrow (:mattwoodrow)]

Can this still be reproduced?

In firefox 4, we didn't have texture_from_pixmap, and this was causing significant performance regressions with accelerated layers.

This should be fixed now in the latest versions.

Comment 6

[:aceman]

I can't really say if it is slower or faster graphics-wise than no accel. Because it flickers a lot everywhere (elements disapearing just on mouseover, scrolling flicker, etc.). And it also consumes 1GB of RAM just with first page loaded so it is slow just from swapping and memory allocating.

Comment 7

[Stefan Fleiter (:sfleiter)]

With
NVIDIA Corporation -- GeForce 8400M GS/PCI/SSE2 -- 3.3.0 NVIDIA 280.13
and layers.acceleration.force-enabled=true
the demo hangs during "Initializing FishBowl...".

With layers.acceleration.force-enabled=false it works.