Mesa 18.3.3 Released With Fixes For RadeonSI/RADV, OpenGL Driver Bugs

[phoronix]

Phoronix: Mesa 18.3.3 Released With Fixes For RadeonSI/RADV, OpenGL Driver Bugs

Mesa 18.3.3 was released today as the newest stable release for the current Mesa 18.3 series from Q4...

http://www.phoronix.com/scan.php?page=news_item&px=Mesa-18.3.3-Released

[dungeon]

If only llvm's L350532 would be backported to llvm 7:

https://reviews.llvm.org/rL350532

otherwise this 18.3 is unusable for me, full of corruptions using AMDGPU... since september and now good morning february

Now because seems mostly RadV and Nine are affected, AMD seems is not interested into this... who will backport this i have no idea, at least we have Phoronix to post our PITA with mainly borked AMD drivers

Anything that involves AMD developers helping to troubleshoot code that emulates our customers platforms gets into murky territory, and so we generally stay away from it. Same for anything emulating Windows or DX.

[Masush5]

Yeah, the current llvm release situation is rather unfortunate. On the stable series you wait either months or forever for critical bugfixes (and lets be real, most of the time it's forever, i.e. the next major release), and on master you get bugfixes and sometimes perf improvements, but also frequent regressions. It would be nice if they would do more frequent point releases and backport some of the fixes.

[dungeon]

Well, LLVM have 2 stable releases annually, Mesa have 4... of which maybe only one is really stable, maybe two. This one 18.3 does not compile against unreleased llvm8, llvm7 have a bug (in combination with 18.3, not with 18.2), so to me it is still paperware (un)release.

Pretty much same like kernel, maybe one is good annually for an end average Joe user, maybe two... but nothing much more than this

It is fine only for development and if you have several patches ready to apply anytime anywhere

Maybe i will end up recommending 18.3.x on its last point, but we will see about that

For now i could only piss off on these "stables", move it into shit basket and for AMD developers i have more music since they don't read bugs

[clapbr]

Well, LLVM have 2 stable releases annually, Mesa have 4... of which maybe only one is really stable, maybe two. This one 18.3 does not compile against unreleased llvm8, llvm7 have a bug (in combination with 18.3, not with 18.2), so to me it is still paperware (un)release.

Pretty much same like kernel, maybe one is good annually for an end average Joe user, maybe two... but nothing much more than this

It is fine only for development and if you have several patches ready to apply anytime anywhere

Maybe i will end up recommending 18.3.x on its last point, but we will see about that

For now i could only piss off on these "stables", move it into shit basket and for AMD developers i have more music since they don't read bugs

I still wait on this and related issues - https://bugs.freedesktop.org/show_bug.cgi?id=102646
I have two different setups/displays affected, a 75hz and a 144hz. Have to run both at 60hz or force clocks to high and deal with constant high temps plus having to reapply it when sleeping or switching resolutions.

[moilami]

If only llvm's L350532 would be backported to llvm 7:

https://reviews.llvm.org/rL350532

otherwise this 18.3 is unusable for me, full of corruptions using AMDGPU... since september and now good morning february

Now because seems mostly RadV and Nine are affected, AMD seems is not interested into this... who will backport this i have no idea, at least we have Phoronix to post our PITA with mainly borked AMD drivers

What? Since when people who use Linux, Wine, Dolphin. etc. with AMD graphics cards are not customers? If they are not customers, then, what are they? It is not like they would get their AMD graphics cards free.

[moilami]

I don't remember seeing ever such arrogant comment by Bridgman.


*

Originally posted by bridgman View Post


"Anything that involves AMD developers helping to troubleshoot code that emulates our customers platforms gets into murky territory, and so we generally stay away from it. Same for anything emulating Windows or DX."

[bridgman]

If only llvm's L350532 would be backported to llvm 7:

Now because seems mostly RadV and Nine are affected...

Do you know if there is a bug ticket filed against Nine or radeonsi/Nine ?

https://bugs.freedesktop.org/show_bug.cgi?id=108594

RH and Valve developers are working on this already so I don't think it's fair to say "only AMD can make this happen".

Video unavailable The uploader has not made this video available in your country.

The second video was nice though.

[dungeon]

The second video was nice though.

Of course, point was that so you get it, but can't see it - just like this stable mesa release compiled against llvm 7 from Twilight Zone

Otherwise it is only usable to sit in a shit basket

[brent]

Yeah, the LLVM release situation completely sucks. I've been saying that on these forums for quite some time. There are various options, and they're all quite simple. Ideally, LLVM should make more patch releases and make it easier for developers to get fixes merged into stables branches. This would not only help AMDGPU. This is unlikely to happen, though. I am not the first person to suggest this, leading LLVM developers don't seem to care.

Realistically, either a patchset against stable LLVM might work (published by AMD or Mesa for packaging with distributions) might work or maintaining a separate fork of LLVM. It might also be a good idea to make the LLVM version used by AMDGPU independent from system-wide LLVM. Distribution updates of LLVM tend to lag because so much stuff is using it and every major update can break a lot of stuff.

Overall, LLVM release management is quite bad if you consider how important the project is.