Open-Source 2D, 3D For ATI Radeon HD 5000 Series GPUs

[whizse]

FYI Andre has committed a number of fixes to the 600c Evergreen support - agd5f's initial impression of the updated code is that the functionality on Evergreen is now pretty close to what you get on 6xx/7xx (although he mentioned some problems with mipmaps).

Might be a good point to give the 600c Evergreen code another try.

Interesting! The r300g side of things is almost boringly fast and complete now, I'm tempted to order a new card...

[rohcQaH]

Let's see what has improved in mesa classic since my last post.

• neverball looks good. Only minor issues.
• geartrain has the correct colours again, but wrong geometry (like the camera was at the wrong position). It was this way before, then it exchanged correct geometry for broken colors, now it's back again. Weird. Would you like a git bisect?
• piglit: still didn't finish a full run

[agd5f]

geartrain looks fine here (other than the problems with the gear teeth on the back gear, but that's a bug in geartrain itself). Make sure you 'make clean' before rebuilding.

[pingufunkybeat]

Now, that's really cool. My next card will be an evergreen one

[whizse]

Might be time to add an "Evergreen" column to http://www.x.org/wiki/RadeonProgram and start filling in the blanks?

[darkbasic]

but next time the amd-OS driver team will be much faster because the difference on R900 are not so high because its just a refresh on the biggest parts.

They told the same thing with "evergreen" -.-

[rohcQaH]

geartrain looks fine here

Just booted into the oss drivers again and re-checked, now it looks fine. Weird.

(other than the problems with the gear teeth on the back gear, but that's a bug in geartrain itself).

I know, I fixed those Which is the only reason why I cared about geartrain anyway.

Make sure you 'make clean' before rebuilding.

gentoo's git ebuilds are a bit more thorough than make clean.


ioquake3: background-gfx in the menus are broken, but start working after playing a game. The game itself works. Mipmaps seem broken: screenshot.
Still looks a lot better than these from 5 days ago.
Strafing doesn't work as well as it used to when I last played it. Surely a driver bug.

doom3: Went to the menu screen, <1 fps. Didn't enter a game because it's asking about my CD key which I don't have here (where'd my ~/.doom3 go?)

ut: ~0.5fps, but the intro sequence (ingame-gfx) does render correctly.

ut2k4: menu works as fast as it should. Starting a game results in

ut2004-bin: radeon_texture.c:94: radeonFreeTexImageData: Assertion `!image->base.Data' failed.

I re-tested dynpm. It obediently upped the frequencies to maximum when starting the first game, but didn't lower them again, not even on my desktop with virtually nothing running. Back to profile / low for now.

I'm pretty sure there's still a nasty bug hidden somewhere, writing to parts of the video memory it shouldn't write to. I experienced it less frequently than before, but there still were small changes (red lines) in some pixmaps after running 3d apps. I wish I knew how to reproduce it or how to boil it down to a testcase :/


I'm amazed by the amount of complex games that run. Just a few days ago glxgears could force my computer into a reboot, now I can frag stuff. Nice!

[agd5f]

I think the last two major issues for getting on par with r6xx/r7xx are fixing mipmaps and stencil buffers.

[bridgman]

They told the same thing with "evergreen" -.-

Yep, and that's how it happened. Roughly ~31 months from launch to GL2 for 6xx, ~19 months for 7xx, <12 months for Evergreen... although it was arguably more like 8 months since work didn't really start until the end of 2009.

[bridgman]

... and no, we're not going to make any schedule estimates for chips we haven't even launched yet