Composite Quesiton

Huang, FrankR FrankR.Huang at amd.com
Tue May 18 18:53:40 PDT 2010


This bug is definitely a serious bug. Right now Use "EXANoComposite" can avoid this that will lead to SW rendering. "NoAccel" option will crash the X with our driver.

I am debugging on this bug now. Right now I already have some progress on this bug with my simple application. From the SRC result, we know that geode does more SW rendering on the region outside the 20x20 mask. I am trying to see why it does that. Then I'll see the wrong result on OVER result.

The rendercheck program is good. But it is so huge to use, I'll debug it using my program, then turn to it.


Thanks,
Frank 

-----Original Message-----
From: Jonathan Morton [mailto:jonathan.morton at movial.com] 
Sent: 2010年5月18日 21:12
To: Huang, FrankR
Cc: xorg-devel at lists.x.org
Subject: RE: Composite Quesiton

On Tue, 2010-05-18 at 11:10 +0800, Huang, FrankR wrote:
> I have used OVER operation to produce what I want on RS690 platform.
> But in the geode driver, I found that it can not do the SRC and OVER
> rendering correctly.

That would be a fairly serious bug.  Does it work with Option "NoAccel"
set?

> I am now firstly focusing the SRC operation. From your explanation, a
> (1,0,1,0) to (0,0,0,0) transfer is done outside the 20x20 mask region.
> Is it done by Xserver or driver? I think it is done by driver, is it
> correct?

It's an inherent part of the XRender spec.  Whether the driver, hardware
or xserver does any particular operation depends on whether the driver
claims to be able to accelerate it.

> Any reference code to see how the driver to handle it? I want to fix
> the bug in the geode driver.

My advice is to obtain the Rendercheck program.  It runs an insane
number of very small tests to (almost) exhaustively exercise XRender.
When a test fails, it displays the inputs, expected output and actual
output.  Run it on an unaccelerated driver first, to prove to yourself
that it works.

-- 
------
From: Jonathan Morton
      jonathan.morton at movial.com






More information about the xorg-devel mailing list