An Initial Version Of LunarGLASS-LLVM For Mesa

[phoronix]

Phoronix: An Initial Version Of LunarGLASS-LLVM For Mesa

In October of last year there was a proposal by LunarG, a small consulting company focusing upon Gallium3D and Mesa that was formed by some of the original Tungsten Graphics crew, to create a common kernel and shader compiler stack. This stack would utilize LLVM (the Low-Level Virtual Machine) for optimizations This work was published as LunarGLASS and there is now a specification and initial implementation of it for Mesa...

http://www.phoronix.com/vr.php?view=OTA3Mw

[FireBurn]

I think that conversion path has far too many stages

Hopefully they can leverage LLVM to do more of the work in the middle

[TechMage89]

Well, the ultimate goal of this, if I understand correctly, is to deprecate the tgsi ir so it can be replaced by llvm ir, but they need to prove the llvm ir will work well for this application first.

[cb88]

Yeah this isn't really news... I mean obviously it can be done... but that doesn't mean it will be fast or efficient in any way.

I mean the current improvements in mesa as of 7.10 have my R350 running openarena @ 10-15fps and sauerbraten @5-10fps or a little more depending on the map with most effects off ... on a 2xPII@300Mhz of all things if the CPU bottlenecks can be removed I can see this getting alot faster. On the other hand since we have faster processors these days they may ignore that part and proformance on my box will regress un future mesa :C I mean I'll be happy with the drive once its running 30fps if that is even possible on this HW not that it isn't a good 2D drive already.

I should put my r350 in my sempron 3100+ box just to see what difference the system ram/CPU/AGP speed limitations is causing.

[allquixotic]

I have yet to see a user of LLVM demonstrate measurable and consistent performance gains. It may be a convenient programming paradigm, but the performance of it is way over-hyped I think.

[xeros]

I should put my r350 in my sempron 3100+ box just to see what difference the system ram/CPU/AGP speed limitations is causing.

The difference will be huge.
I get now on Xonotic with some nice effects enabled not less than 60FPS on most maps on my RV350 @ P4 3GHz @ Mesa 7.11-git.

[dnebdal]

I have yet to see a user of LLVM demonstrate measurable and consistent performance gains. It may be a convenient programming paradigm, but the performance of it is way over-hyped I think.

Which LLVM uses have had performance as the main reason for their move? The only one I can think of is that software GL renderer thing, and that one really was faster than what it's replacing.

[V!NCENT]

Sexeh... Let's wait and see. Even if it doesn't spike in performance it still reduces development burden, resulting in faster development of the driver and getting optmization work in faster. Win-win