That Radeon Performance Regression For R9 290 Might Be Nailed

[dungeon]

I guess MrCooper is on vacation, it will be fixed once he return back

[dungeon]

But in whole, it is very weird to see everywhere releases that does not work

llvm 3.9, mesa 12, linux 4.7 - all borked

[MrCooper]

My guess is that people don't really look at the kernel bugzilla for radeon bug as most of the stuff append on mesa. My bug just rotted there as a Wine bug (it was kind of) but with the fps loss it was matching the phoronix bench, just that people just did not try to reverse that patch by looking only in mesa bug, my guess.

Your bug report is about a single game using the nine state tracker. If you knew that the same kernel change affected other cases as well, please make that clearer in the future.

It was Axel Davy (main developer of the nine state tracker) who suggested that the problem is between nine and Wine. His explanation made sense to me (and still does at least for some D3D applications, though I don't know if EVE Online is one of them).

BTW, the same kernel change resulted in significant performance boosts in many benchmarks with the majority of GPUs, so I'm not sure simply reverting it is a good solution.

Maybe the solution should be a unified location for all the graphic stack (like mesa and drm in the same bug tracker).

We certainly prefer kernel driver issues to be reported in the freedesktop.org bugzilla as well. In addition to the reasons mentioned by Alex, our accounts in the kernel.org bugzilla are too restricted to properly manage the reports.

[MrCooper]

But in whole, it is very weird to see everywhere releases that does not work

llvm 3.9, mesa 12, linux 4.7 - all borked

That's an overstatement, to say the least. All non-trivial software has bugs. It just depends whether any of them happen to affect something you care about.

[dungeon]

That's an overstatement, to say the least.

It is not overstatement, as i also spotted unfixed regressions in all mentioned components... and all that took part of being in those shiny new releases.

[dungeon]

In the end it is what it is, but i don't like to pretend this is the best stack for most people to use while it is proven worse

[theghost]

But in whole, it is very weird to see everywhere releases that does not work

llvm 3.9, mesa 12, linux 4.7 - all borked

I think that's a general problem in the Linux world. Most of these things could be easily spotted by running a ci-system (where Phoronix Test Suite delivers some nice tests), e.g. against Mesa-git etc. But as the development contributions seems to happen mostly independently or decentralized, no one sees his-/herself in charge to maintain and run such systems. If there such systems, e.g. in companies like Red Hat, Canonical or SUSE, they probably just run against their own stable OS versions.

Even if there is a poor soul who discovers such regressions (like Micheal) it remains unfixed for months. That's what's worse...

[duby229]

In the end it is what it is, but i don't like to pretend this is the best stack for most people to use while it is proven worse

Wrong. All you are doing is stating your own bias. The facts are very different from what you say.

[dungeon]

Wrong. All you are doing is stating your own bias. The facts are very different from what you say.

And facts are what? Here was one day check of it, and found regressions one in llvm another in mesa:

https://bugs.freedesktop.org/show_bug.cgi?id=97038

It seems it won't be fixed for llvm 3.9 and here mesa regression from april:

https://bugs.freedesktop.org/show_bug.cgi?id=97039

Tested on Kabini, Kaveri and Bonaire... do ypu want more?

[dungeon]

And if you ask me all that started because of regression with vdpau in 11.2

https://bugs.freedesktop.org/show_bug.cgi?id=96860

Which is actually fixed in 12, but can't be backported to eariler too... so if you want fixed vdpau you should use 12, but else things with GL will be slow or borked there, etc... So i can't use any release, without couple patches workarounds out of my pocket