I just tried playing FoFiX with KMS and r600 dri driver. It's quite stable and good working (50-60 fps), but when I tried to go through song list it hanged. Following message repeated itself many times in dmesg:
Nothing could I do other than switching to VT console and issuing reboot (I like KMS ).Code:Dec 9 21:19:21 wielkieg kernel: [ 7142.780009] [drm:radeon_ib_get] *ERROR* radeon: IB(7:0x0000000010171000:14937) Dec 9 21:19:21 wielkieg kernel: [ 7142.780013] [drm:radeon_ib_get] *ERROR* radeon: GPU lockup detected, fail to get a IB Dec 9 21:19:21 wielkieg kernel: [ 7142.780015] [drm:radeon_cs_ioctl] *ERROR* Failed to get ib !
System (kubuntu lucid):
X from xorgedgers, including (from git's, most from 2009-12-08):
X, mesa, libdrm, xf86-video-ati, ...
I have the same errors, but with openarena and tremulous
I'm on a HD3650 (rv635) AGP, mobo has VIA K8M800
Using KMS on Karmic + 2.6.32 final + xorg-edgers ppa updated as of a couple of days ago.
Default AGP mode and speed, no change in UFS/DFS, AGP aperture set to 256M (although this was more required for UMS)
HD4850 using xorg-edgers ppa:
Playable at "normal" settings; some effects cause severe framerate drops and game can sometimes CTD (try running demo1 for instance); effects like water reflection don't render at all
Does not even get to the menu
Alien Arena 7.32
Runs, but gameplay is too slow even at the lowest settings
Urban Terror 4.1
Runs perfectly at max settings
Runs perfectly at max settings
Does not run
Crashes while loading the menu
Eschalon Book 1
Runs, but really slow
Runs with no visible errors, but too slow to be playable
Secret Maryo Chronicles
Initial launch screen has rendering errors, but game runs
Disable shaders and set texture quality to medium; game runs fine but I did encounter a few random crashes
This is all with radeon and no KMS, Kwin effects in the background and all games at 1680x1050. Radeonhd for some reason is way too slow and has more errors (can't even run the Kwin cube without chocking).
What is the status of GLSL in r600_dri? I know that it can be experimentally enabled by uncommenting something in one source file, but it's incomplete/unstable. Will it be enabled in legacy dri driver ever?
For example FoFiX works quite good, but without GLSL it looks weird in some cases (like fretboard being bright yellow in which case I can't distingish notes from fretboard). Additionally AFAIK wine in many places makes heavy use of GLSL.
Status of GLSL is "initial code has been pushed, testing and fixing are underway".
Right now I believe uncommenting the flag also indicates that GL 2.0 support is available - which implies some other extensions are also present which may not be the case. I don't know if the majority of games :
1. check for GLSL and use it, even if GL level is 1.5
2. look for GL 2.0 then only use a subset of the GL 2.0 functionality (ie GLSL), implying that announcing GL 2.0 is probably a Good Thing even if some core extensions are yet to be implemented
3. Look for GL 2.0 and actually use all of the GL 2.0 core extensions without further checking, implying that exposing GL 2.0 at this point would be a Bad Thing even if GLSL was ready to go
So, yes it will be enabled, but only when it works well enough to turn it on by default. We're still learning about which functions and extensions are actually used -- for example the Noise functions don't seem to get used by applications, presumably because most drivers do not seem to implement them.
I grepped through the sources of the wined3d driver, and it appears that wine isn't using the noise functions. Still no luck getting Half-Life 2 running with -dxlevel 80 though.
btw. Wine does appear to use the dFdx/dFdy functions, which if I'm not mistaken haven't been implemented yet.
Last edited by monraaf; 12-11-2009 at 06:26 PM.
Richard was working on implementation of the noise functions and was making good progress, but we all started to have nagging doubts about whether they would ever get used by applications. As far as we can see the "industry standard implementation" is to return 0.
Richard has parked the Noise work and is working on other functions until we figure out how to proceed. I believe the dFdx/dFdy (gradient ?) functions were next on the list after noise, although Richard may work on fixing some other bugs first.
Last edited by bridgman; 12-11-2009 at 06:51 PM.