Bug 27954

Summary: presentations in Impress really slow, when images are included
Product: xorg Reporter: Martin Stolpe <martinstolpe>
Component: Lib/otherAssignee: Xorg Project Team <xorg-team>
Status: RESOLVED FIXED QA Contact: Xorg Project Team <xorg-team>
Severity: normal    
Priority: medium CC: bugs.freedesktop
Version: 7.5 (2009.10)   
Hardware: Other   
OS: All   
Whiteboard:
i915 platform: i915 features:
Attachments:
Description Flags
oprofile output
none
xorg log none

Description Martin Stolpe 2010-05-04 00:12:40 UTC
Created attachment 35403 [details]
oprofile output

When I start an Impress presentation it takes a lot of time to load slides which include images. I've attached an oprofile output which shows the problem.

kernel: 2.6.33.3
xorg-server: 1.7.6
xf86-video-ati: git b2ecd1bb507ed4e0d9ba6ebd498c51da2e8136fa
mesa: git a12614362a79d676eeb1c3030ed52057bcb8f4f8
pixman: 0.18
Comment 1 Martin Stolpe 2010-05-04 00:15:17 UTC
Created attachment 35404 [details]
xorg log
Comment 2 Michel Dänzer 2010-05-04 00:26:37 UTC
It's a RENDER Composite operation software fallback. Rebuilding the radeon driver with the

#define RADEON_TRACE_FALL 0

line in radeon_exa.c changed to 1 might give some information in the log file about why the fallback is occurring, but there will probably be a lot of debugging output so please try to isolate the parts corresponding to the problem.
Comment 3 Martin Stolpe 2010-05-04 16:06:58 UTC
Those three lines are repeated over and over again in the log file:
R300CheckComposite: Component alpha not supported with source alpha and source value blending.
R300CheckCompositeTexture: REPEAT_NONE unsupported for transformed xRGB source
R300CheckCompositeTexture: Unsupported picture format 0x1011000

The first line appears also when I don't use Impress. But the other two seem to appear only when I use Impress.
Comment 4 Michel Dänzer 2010-05-05 02:54:13 UTC
(In reply to comment #3)
> R300CheckCompositeTexture: REPEAT_NONE unsupported for transformed xRGB source

This is most likely the relevant one for images, assuming there's scaling or other transformations involved. Unfortunately, the RENDER extension specifies relatively unusual semantics for RepeatNone pictures with no alpha channel, which can make it tricky to accelerate transformed Composite operations using them. If Impress just needs scaling, it would probably be better served with RepeatPad instead of RepeatNone.
Comment 5 Martin Stolpe 2010-05-05 06:31:08 UTC
I have filed a bug at openoffice.org: http://www.openoffice.org/issues/show_bug.cgi?id=111399
Comment 6 Martin Stolpe 2010-09-27 13:11:42 UTC
There's no speed problem when using gallium + xorg state tracker.
Comment 7 Martin Stolpe 2010-10-12 14:23:35 UTC
Seems to have been fixed in the meantime.

Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct. How we collect and use information is described in our Privacy Policy.