Mesa 7.12 Is Now Where The Fun Is At

[phoronix]

Phoronix: Mesa 7.12 Is Now Where The Fun Is At

Mesa 7.11 has been branched in preparation for its release next month. What this means is that now Mesa Git master is for Mesa 7.12, which will be the graphics code in-development until next January when it's either released as 7.12, or Mesa 8.0 should the OpenGL 3.0 support land by year's end...

http://www.phoronix.com/vr.php?view=OTYxMg

[pejakm]

What else would you like to see in Mesa 7.12/8.0?

Radeon drivers beats the hell out of Catalyst!

[DanL]

What else would you like to see in Mesa 7.12/8.0?

pipe-video improvements, pipe-video improvements, pipe-video improvements, pipe-video improvements, pipe-video improvements... etc.

[lolren]

What else would you like to see in Mesa 7.12/8.0?
kde working on samsung nc10 (intel 945 ) with effects

[allquixotic]

Michael,

Any word out of LunarGLASS? I know we have GLSL to IR in-progress as a competing implementation, but using LLVM is also an interesting alternative, and there's ample time now to get this into Mesa 8.0 / 7.12.

Using a minimum number of bytecode translation layers (enough to provide a general abstraction across hardware, but lean enough for real-world performance) would be a significant improvement over what we have now. Bonus points if you manage not to break half the GLSL programs that ran on the old stack, too.

One of the last holdouts I'm hoping to see support for on Evergreen is Hyper-Z. If that needs mesa bits, I hope they make it into mesa 7.12 and Linux 3.1.

Also, in light of the recent focus on Phoronix about power consumption, I would love if additional power reductions could be made with desktop radeons. I don't care if the card goes full blast while actively playing an intensive 3d game, but I'd like to see the power consumption on a composited desktop go down. If your browser is just sitting there making no / almost no screen updates, the card shouldn't have to go full blast just to keep it that way. And sometimes I lock my screen to go AFK for 15 - 30 minutes and would like the card to go down to idle power (25W, iirc?) during that time.

It's not that I am really concerned about my electricity bill, but more about the life of the card. The VRMs are rated for 5000 hours at maximum operating temperature, and the cooler they stay, the longer they last. So running the card hotter than necessary for the workload is inching the VRMs closer to the MTBF point, at a faster rate than would be if using proprietary drivers that manage power better.

[Drago]

Michael,

Any word out of LunarGLASS? I know we have GLSL to IR in-progress as a competing implementation, but using LLVM is also an interesting alternative, and there's ample time now to get this into Mesa 8.0 / 7.12.

I don't know for 7xxx series, but for sure 8xxx will depart from VLIW design, and will be SIMD. So guess there LLVM will kick ass!

[smitty3268]

but i really think GL 3 support can happen soon now. The main blocker is GLSL 1.3, and there's been some talk on the mailing list the last few weeks that make me think this is moving forward. Then even the rest of GL 3.1 - 3.4 shouldn't be that much work (12 months?) - geometry shaders are the big thing left to do, as far as i know.

As far as what i'd like to see beyond GL3 support. Mostly improvements to the radeon drivers, such as texture tiling and power management, which are likely kernel issues, and performance improvements. Like an optimizing compiler in r600g that would actually reorder instructions to take advantage of the VLIW hardware architecture. And catching up with getting current hardware fully working - NI/Cayman seems to be the last stumbling block right now.

[Azpegath]

but i really think GL 3 support can happen soon now. The main blocker is GLSL 1.3, and there's been some talk on the mailing list the last few weeks that make me think this is moving forward. Then even the rest of GL 3.1 - 3.4 shouldn't be that much work (12 months?) - geometry shaders are the big thing left to do, as far as i know.

As far as what i'd like to see beyond GL3 support. Mostly improvements to the radeon drivers, such as texture tiling and power management, which are likely kernel issues, and performance improvements. Like an optimizing compiler in r600g that would actually reorder instructions to take advantage of the VLIW hardware architecture. And catching up with getting current hardware fully working - NI/Cayman seems to be the last stumbling block right now.

I agree, OpenGL 3 goes before performance improvements. I want to throw the fixed pipeline out the window! I fully agree on the radeon stuff as well, but I'm very biased

[tom.higgy]

I'd quite like to see improvements to pipe-video, power management and performance using r300g, although I'm guessing the focus is now on newer hardware.

It doesn't make sense for me to upgrade my hardware at present as I don't do anything particularly intensive besides playing 0ad and watching the occasional HD video, which my machine can handle fairly well.

[Qaridarium]

I don't know for 7xxx series, but for sure 8xxx will depart from VLIW design, and will be SIMD. So guess there LLVM will kick ass!

in the past they use VLIW+SIMD and i think they use RISC+SIMD in the hd8000 line.