How Far The Radeon Gallium3D Driver Has Come In Five Years

[jaxxed]

And a litte bit "fail": radeonsi reached OpenGL 4.1 before 4.0. It's a little bit ironic to see that happen --> http://mesamatrix.net/#Version_OpenGL4.1-GLSL4.10 But it's absolutely great to see that driver grow, because my HD 7970 wants to play games under linux. So, I think that my f***ing windows will be killed in the next month and linux will get the big SSD. Yeah, weekend is in a few hours and I will party hard this new freedom! *MESA!!!*

I can't see that as a fail. Forward progress is never a fail, and some of the 4.0 targets were quite difficult.

[TheRealBecks]

That was ironic Truly it's no fail, but when climbing a hill you don't get first to summit without going the whole path. So it's step 2 before 1 - I can't, but AMD did it

[Serafean]

I honestly don't see the irony...

I remember when initial UVD support came through vdpau. That was for me the sign that, wanting OSS & performance, I chose well.
DPM,VCE and others followed suit, and I do not want to diminish their impact, but the initial UVD support was for me the first and strongest moment of "hey, AMD is really going through with this..."

What I do find ironic is that the reverse engineered nvc0 driver supports everything it can (speaking about Ogl), but intel and radeonSi not yet...

Anyway, congrats and thank you to everyone working on this...

[geearf]

That was ironic Truly it's no fail, but when climbing a hill you don't get first to summit without going the whole path. So it's step 2 before 1 - I can't, but AMD did it

To reach 4.1 you still need 4.0 to be done as 4.1 is just an increment of 4.0.
So we still have not reached the top of the hill, we're still climbing it, but it's a bigger hill.

What I do find ironic is that the reverse engineered nvc0 driver supports everything it can (speaking about Ogl), but intel and radeonSi not yet...

Yeah I truly don't get this, how can nvc0 be first?
Is the hardware somewhat easier to implement for or...?

[Kano]

Mesa 11 should seems to be a very interesting upgrade. Now i just want to see Intel's OpenGL 4 support...

[89c51]

Other parts of it -Video acceleration, open cl- still sucks and needs love.

[89c51]

Other parts of it like HW video acceleration OpenCL still need love. And a lot of it.

[BrollyLSSJ]

For me, VDPAU for the old HD3850 AGP was the best feature as I am still using an old P4 at my main site (do not play much or games like Jedi Knight Jedi Academy are enough). I am using Fedora 22 now (began with Fedora 18 and continually updated it with yum to 19 and 20 and with fedup from 20 to 21 and to 22. These upgrade never worked for me for debian and Kubuntu (around year 2004 to 2006). I still have Windows 7 installed on the P4, but mainly use Fedora and hope to see OpenGL 4.2 support for the r600 drive soon

I still have a Q9300 Windows 7 PC with 8GB of RAM and a GeForce 660 though. But I do not use it with Linux (I use it only when I visit my parents).

[eydee]

For me, VDPAU for the old HD3850 AGP was the best feature as I am still using an old P4 at my main site (do not play much or games like Jedi Knight Jedi Academy are enough). I am using Fedora 22 now (began with Fedora 18 and continually updated it with yum to 19 and 20 and with fedup from 20 to 21 and to 22. These upgrade never worked for me for debian and Kubuntu (around year 2004 to 2006). I still have Windows 7 installed on the P4, but mainly use Fedora and hope to see OpenGL 4.2 support for the r600 drive soon

I still have a Q9300 Windows 7 PC with 8GB of RAM and a GeForce 660 though. But I do not use it with Linux (I use it only when I visit my parents).

The hardware itself doesn't support GL4.x, so even if the driver and mesa have it, it won't work on that card. It's a DX10/GL3 level card.

[SXX⁣]

Yeah I truly don't get this, how can nvc0 be first?
Is the hardware somewhat easier to implement for or...?

Best answer for that is imirkin post there:

Ssssort of. In the sense that the functionality is exposed, yes. In the sense that it works... maybe. I don't even have a maxwell, I did do basic testing on one remotely at some point for that patchset, maybe 6-8 months ago and it didn't fail any more than fermi. But there are probably subtle differences that will have to be analyzed and fixed as they are reported.

Same goes for Kepler (GK10x) and Kepler2 (GK110/GK20x). I've already discovered one weird thing about Kepler, I'm sure there are a few more lurking. On the bright side, I do have access to a GK208 now so hopefully I'll notice issues sooner.

Unlike the professional dev teams, I don't have arbitrary access to hardware, dev boxes, not to mention documentation or time to look into things. I don't push things that I know to be totally broken, but it does often happen that one or another thing doesn't work on some revision of the hardware.

So basically patches for tesselation on RadeonSI were around for many months, but there is multiple full-time AMD developers and they won't push them into mainline until they know something may not work on specific hardware for whatever reason. Nouveau have less full-time developers with limited access to GPUs so more bugs going to be fixed after code going live and somebody who own hardware report problems.