Open-Source Radeon HD 6000 Series Still Borked

[darkbasic]

Perhaps even me

That would be really awesome, I reported multilib related bugs lots of times but nobody ever cared. First thing to do should be porting mesa-9999 ebuild in sync with the x11 overlay one, otherwise it's completely useless (it lacks tons of use flags and much more). Also libdrm-9999 didn't work last time I tried.

[curaga]

Hey Q, clean your PM box

[crazycheese]

I'm still searching a way to keep 32bit mesa and libdrm in sync with the main 64bit driver. On debian I do use multiarch xserver/mesa from xorg-edgers ppa and dpkg from 'pu/multiarch/full' branch (but natty already ships it), fedora already has multiarch since 6 years, but what about gentoo? multilib overlay sucks, nobody keeps the mesa/libdrm/xf86-video-ati 9999 ebuilds in sync and they simply don't work. Also, if you use external overlays you can forgot about ia32 because multilib is not in portage and so nobody has to take care of it.

Heh, sounds like more like lack of manpower and interest to me. Sure someone would jump-in, matter of time. Perhaps even me

I have a professional ultra wide gamut (107% of AdobeRGB) 30" 2560x1600 monitor, it should be enough

Also,
9) How does performance compare when using an old dual core (Athlon64 X2 3800+) vs the most powerful cpu available (Sandy Bridge 2500K)? It will be one of my first tests, I want to know how much cpu limited is R600g.

Nice!
I guess the main problem is currently very unoptimized, very cpu bound opengl extensions.. At least when I had 4770, supertuxkart had used cpu massively. Not to the state of complete cpu fallback, but still.

[darkbasic]

6) wine (sometimes, 2d and 3d - say fallout2 and painkiller)

I'm still searching a way to keep 32bit mesa and libdrm in sync with the main 64bit driver. On debian I do use multiarch xserver/mesa from xorg-edgers ppa and dpkg from 'pu/multiarch/full' branch (but natty already ships it), fedora already has multiarch since 6 years, but what about gentoo? multilib overlay sucks, nobody keeps the mesa/libdrm/xf86-video-ati 9999 ebuilds in sync and they simply don't work. Also, if you use external overlays you can forgot about ia32 because multilib is not in portage and so nobody has to take care of it.

P.S.
oibaf, do you plan to support multiarch in your ppa?

7) monitor support

I have a professional ultra wide gamut (107% of AdobeRGB) 30" 2560x1600 monitor, it should be enough

Also,
9) How does performance compare when using an old dual core (Athlon64 X2 3800+) vs the most powerful cpu available (Sandy Bridge 2500K)? It will be one of my first tests, I want to know how much cpu limited is R600g.

[crazycheese]

In the first post of my PPA I added some instructions on how to debug problems. But what do you mean exactly with I assume you work via freedesktop and distributed debug pool? I only provide these packages and feedback through the forum, there is nothing more than that.

Yes, Im aware. I use gentoo though, but I will be able to apply thanks to you keeping the revision logs. With "debug pool" I mean a pool, similar to current bug pool on freedesktop, that would list the tasks to perform(and instruction on how) some test on current drivers, directed to people that use open drivers but have no possiblilty to program the changes. Ie, bringing hackers and users together.

[oibaf]

I assume you work via freedesktop.. Do you have some sort of debugging pool - a list with task and detailed instructions, so users with cards can perform the tests and report. Some sort of distributed debug pool. Many people do not program, but many have the hardware to test.
I know about fedora on this subject btw.

In the first post of my PPA I added some instructions on how to debug problems. But what do you mean exactly with I assume you work via freedesktop and distributed debug pool? I only provide these packages and feedback through the forum, there is nothing more than that.

[crazycheese]

Yes, I will. I'm really, really curious to know if I can get more than 10fps in unigine heaven with R600g

Ty, very much!

i can check your personal checklist with my hd5670...

just ask your needs.

I know. But purchasing 5870 monster is something not made overnight... given the electricity draw amount over time alone. Its vliw5 card, driver performance is very heavy affected by optimization as far as I understand. Michael did splendid review on nvidia vs amd powerdraw and catalyst vs radeon performance. But still
for both drivers on 5870 or very similar architecture
1) powerdraw in idle and in performance modes, speed and intelligence of switching (you need external 10? measurement device for this)
2) 2D, compiz and 3D bugs
3) -"- features
4) -"- performance(in native games and compiz)
5) opencl... please?
6) wine (sometimes, 2d and 3d - say fallout2 and painkiller)
7) monitor support
8) detailed setup, its ok to have 160 page, but have it covered. Preferably in one place.

[crazycheese]

If you are using my PPA these warnings are showed because I build mesa with --enable-debug. These are probably errors in the application, and should be reported to Uningine.


This looks indeed a problem with the glsl-to-tgsi branch which I am currently merging in the PPA.

EDIT: I reported this to Bryan Cain, hopefully he will fix it.

I assume you work via freedesktop.. Do you have some sort of debugging pool - a list with task and detailed instructions, so users with cards can perform the tests and report. Some sort of distributed debug pool. Many people do not program, but many have the hardware to test.
I know about fedora on this subject btw.

[oibaf]

i try it for you: 0FPS
...
Mesa warning: glDraw[Range]Elements(start 297, end 1072, count 2253, type 0x1403, indices=0x9c0)
end is out of bounds (max=1071) Element Buffer 59 (size 7002)
This should probably be fixed in the application.
Mesa warning: glDraw[Range]Elements(start 297, end 1072, count 2253, type 0x1403, indices=0x9c0)
end is out of bounds (max=1071) Element Buffer 59 (size 7002)
This should probably be fixed in the application.

If you are using my PPA these warnings are showed because I build mesa with --enable-debug. These are probably errors in the application, and should be reported to Uningine.

state_tracker/st_glsl_to_tgsi.cpp:2886:simplify_cmp: Assertion `inst->dst.index < 256' failed.
Trace/breakpoint trap

This looks indeed a problem with the glsl-to-tgsi branch which I am currently merging in the PPA.

EDIT: I reported this to Bryan Cain, hopefully he will fix it.

[smitty3268]

state_tracker/st_glsl_to_tgsi.cpp:2886:simplify_cmp: Assertion `inst->dst.index < 256' failed.
Trace/breakpoint trap

Wasn't the st_glsl_to_tgsi.cpp file added by Plombo in his branch getting rid of the Mesa IR? I don't think it's present in git master or 7.11.

Looks like that optimization pass is the first one getting called in his code, before a lot of the code removal passes. Possibly Unigine's shaders just use more than 256 temps before being simplified. Whatever the cause, this looks like a problem in his branch only. You could try bumping that constant up to 512 to see if it fixes things.