[RV635] RadeonHD 3670, unable to show entire color scheme

[Mario Limonciello]

See attached screenshot for an example.

[Mario Limonciello (superm1)]

This is on Intrepid, fully updated to June 26 2008, with the open source -ati driver.

[Mario Limonciello (superm1)]

Well actually that screenshot doesn't do justice. It shows up with several "contours" on the display.

[Mario Limonciello (superm1)]

This image really does show it:

<bryce> banding
<mario_limonciell> because that only happens with the 'ati' driver
<mario_limonciell> not with the fglrx
<bryce> yeah we used to have that in inkscape with gradient patterns
<mario_limonciell> what causes it?
<bryce> color interpolation bugs,
<bryce> numerical issues in gradient printing code

[Mario Limonciello (superm1)]

In order to rule it out being the background, this was verified in inkscape by making a large circle with a gradient from one color to another. It shows very similar banding results.

[Tormod Volden (tormodvolden)]

Thanks for your report. Can you please try the newest driver in intrepid and if it's still the same file a bug upstream on bugs.freedesktop.org (Product=xorg, Component=Driver/Radeon), attaching your Xorg.0.log?

[Mario Limonciello (superm1)]

Created an attachment (id=18239)
Xorg log

This is tested on Ubuntu Intrepid, using the default xorg.conf which defaults to the radeon driver.

This results in banding across the display.

[Mario Limonciello (superm1)]

It happens using the latest xserver-xorg-video-ati (1.6.9+git20080802.1f3eee36-1ubuntu1) too.

[Bryce Harrington (bryce)]

Created an attachment (id=18698)
Photo showing banding

Mario sent me the hardware in question; I've taken a photo of the screen to illustrate the banding.

This is with -ati version 1:6.9.0+git20080802.1f3eee36-1ubuntu1. I'll test with the a3cc1d7a once it's built.

[Bryce Harrington (bryce)]

Some additional details...

This seems to only occur with gradients. I can reproduce the issue both in Inkscape and Gimp when using their gradients.

[agd5f (agd5f)]

Dithering is probably enabled for that output. You'd need to fix up dfp_disable_dither() in atombios_output.c for DCE 3.0 cards.

[Bryce Harrington (bryce)]

Noticed that the log didn't specify the pci info:

01:00.0 VGA compatible controller [0300]: ATI Technologies Inc Radeon Mobility HD 3670 [1002:9593]
 Subsystem: Dell Device [1028:0272]

[Bryce Harrington (bryce)]

Created an attachment (id=18990)
rv635_dithering.patch

I'm having some trouble determining what register needs to be poked. I've tried all the ones in the current code, and tried finding some docs to see if there is a new BIT_DEPTH_CONTROL register, but no luck so far. Maybe something has to be done in addition to setting the bit depth?

[Bryce Harrington (bryce)]

I've also tried each of the following without luck:

            OUTREG(0x6710, 0);

            OUTREG(0x6f10, 0);

[Bryce Harrington (bryce)]

Created an attachment (id=19105)
Enable dithering on rv635 hardware

After a whole bunch of experimentation and advice from Alex, this patch enables dithering and greatly diminishes the banding. (The banding's still there, but you have to look close to see it.)

[Bryce Harrington (bryce)]

Alex, please let me know if there are changes you'd like to see to this patch before it can be committed to upstream git. I've incorporated it into Ubuntu's -ati package.

[Bryce Harrington (bryce)]

I've put in a patch to enable dithering, which makes it look a lot better, although it'd be preferable not to need dithering at all. The patch is upstream, so maybe a better fix will come in time, but at least the issue isn't as noticeable anymore.

[Oibaf (oibaf)]

Is this bug supposed to be fixed with current git?

[agd5f (agd5f)]

(In reply to comment #9)
> Is this bug supposed to be fixed with current git?
>

Ajax' commit (d744dc013cf5ca3141c0fdf55e36d713fe9f53ba) should fix it I think.

[Bryce Harrington (bryce)]

I finally got a chance to re-test this. Unfortunately, it did not eliminate the banding. At least it seems to be no worse than the patch I'd posted earlier, but the banding is still present if you look closely.

[agd5f (agd5f)]

Is this still an issue with ati git master?

[Timothy R. Chavez (timrchavez)]

The bug task for the somerville project has been removed by an automated script. This bug has been cloned on that project and is available here: https://bugs.launchpad.net/bugs/1305630