AMD RadeonSI & R600g Gallium3D Tests On Mesa 11.0 + DRM-Next

[phoronix]

Phoronix: AMD RadeonSI & R600g Gallium3D Tests On Mesa 11.0 + DRM-Next

As some new Mesa 11.0 benchmarks to publish is looking at the performance of several Radeon GPUs on the R600g and RadeonSI Gallium3D drivers as tested out-of-the-box on Ubuntu 15.04 with the Linux 3.19 kernel and Mesa 10.5.2, then compared to the DRM-Next code for Linux 4.3 plus Mesa 11.0-rc1.

http://www.phoronix.com/vr.php?view=22078

[chuckula]

So one clear win and a bunch of clear losses. So much for the expected miracle, but at least they are getting OGL 4 features working.

[Veerappan]

Wow... yeah, that's not great. It's possible that adding new features for radeonsi has caused more faithful rendering at the cost of performance, but it's also possible that we have a regression given that almost all tests for radeonsi show a decrease.

Michael, I looked on openbenchmarking.org, but didn't find anything... do you happen to have the configure settings that you used to build LLVM with to look at (debug/release build)? Otherwise, maybe comparing just a kernel upgrade or just the mesa or llvm upgrade might help to isolate this regression.

If I had a bit more time available, I'd probably look into this one myself.

[oleid]

Hm, on the bright side: the regressions didn't make the other games unplayable, while Unigine Valley just got playable on R290^^

[boffo]

There are clearly regressions with R600

[GreatEmerald]

Well, the one test that got a performance improvement is the only one that uses GL4 features as far as I can tell.

[schmidtbag]

Normally results like this wouldn't be a problem to me. The additional GL 4.x features being available means there is more to render, and therefore the GPU has to work harder. However, I'm not sure these results are actually telling that story.

[spstarr]

That's without the experimental SI scheduler not part of LLVM 3.8.x yet, I am seeing the opposite of your tests, on Bonaire for radeonsi I get better FPS across the board.

[dungeon]

Apply this patch to rc1 and retest again It is known one and will be in rc2:

http://cgit.freedesktop.org/mesa/mes...bab900006648aa

BTW generally that mesa 11 + llvm 3.8-git does not make sense to me Really, you want RC of all (mesa 11 and llvm 3.7) or git heads of all

[Michael]

Wow... yeah, that's not great. It's possible that adding new features for radeonsi has caused more faithful rendering at the cost of performance, but it's also possible that we have a regression given that almost all tests for radeonsi show a decrease.

Michael, I looked on openbenchmarking.org, but didn't find anything... do you happen to have the configure settings that you used to build LLVM with to look at (debug/release build)? Otherwise, maybe comparing just a kernel upgrade or just the mesa or llvm upgrade might help to isolate this regression.

If I had a bit more time available, I'd probably look into this one myself.

It was a release build; optimizations enabled, assertions disabled.