Optimizing Performance For Intel OpenGL On Linux

[mattst88]

I find it a little weird that AMD is accomplishing tasks at a much faster rate.

I wonder what you mean by this. Intel drove development of the new GLSL compiler, and then development of GL 3.0, and then 3.1, and now Paul is completing the geometry shaders work to get us to 3.3.

If it looks like it takes less time for the one driver to gain support for a feature, it's often because someone at Intel implemented it first and did a lot of work in core Mesa to make it possible.

[duby229]

I figured as much, and I'm not surprised either. I wouldn't be surprised if they're paid MUCH better too, and were employed for longer. I would also believe it if they had more qualifications. What I find hard to believe is how their progression seems to be slowing down considerably. They seem to get hardware supported much quicker than AMD, but they seem to take much longer to improve or expand upon things. On the other hand, as far as I'm aware, Intel devs are also focusing on non-driver related code, including mesa, wayland, Qt, and X, which I assume AMD does not. Like I said though, considering intel has the upper-hand in basically every perspective when it comes to open source driver development, I find it a little weird that AMD is accomplishing tasks at a much faster rate. Even when you consider the other projects intel is involved in, such as wayland, the progression is still pretty slow. Wayland seems like a great project and I'm glad a company like intel is backing it, but they really got to stop messing with gimmicky things and focus on making it a viable replacement to X. It took way too long to get minimize working (I'm not sure if it even is working).

Don't underestimate AMD. They have an awesome team. Most of them are well known around here on Phoronix... They deserve every bit of praise that they get. Both Intel and AMD do a lot of work on Mesa. AMD did a lot to make gallium3d what it is today, and it was Intel who really made gl3 possible.

As far as I'm concerned it's a damn good thing that both Intel and AMD are employing OSS development teams. I applaud both.

[duby229]

For what it's worth, the Mesa development team at Intel has grown by nearly 10x over the last 3 years. So there is significant investment going on. It's always hard to find qualified people, and even then, at some point you hit the classic Fred Brooks problem where adding more people to a late project only makes it later.

The pace has picked up considerably, but so has the attention and importance. We're still on track to hit 3.3 by the end of the year, and working on 4.x features as well. If there are particular features people need sooner rather than later, we're open to that feedback.

OK, I'm not a programmer. Please forgive me for my inadequate understanding.

I don't understand why Intel hasn't ported it's driver to gallium3d yet. You are already compiling into mesair, why not just implement a tgsi compiler?

[mattst88]

You are already compiling into mesair

No, we're not. The i965 driver takes GLSL IR directly.

[jrch2k8]

No, we're not. The i965 driver takes GLSL IR directly.

just out of curiosity with something like iLO is possible to replace your GLSL IR with LLVM IR and reuse tom stellard/Vadim work for intel drivers? i mean im not complaining or anything like it, is just the thought of maybe if intel/amd/nouveau/freedreno/etc all focus on LLVM and gallium you could remove a lot of code from Mesa and maybe make everyone jobs easy?

i understand you had your reservations with gallium/llvm[i think i read some time ago] but as an owner of an RadeonSI i can tell you it flies, sure is behind other drivers in some areas but what works, works really fast.

dunno is just a though for the future i know there are more important thing pending to worry too much about

thanks to all FOSS graphic team