AMD R600g Tessellation Support Lands In Mesa Git

[bug77]

Someone please tell me, cause I always tend to forget: what is the purpose of all this work? And by all this, I mean all of the open-source graphics stacks. If it is not maintained by the vendors themselves, what good is it t become feature complete (or not even) by the time even the vendors themselves deem their hardware deprecated ("legacy" is the nice term).

It is just a waist of time. If I were an ATI 5xxxx card owner (and I still am, but use Windows), I'd have already upgraded to an R2xx GPU and lack again a new set of features that will be complete once I sell my junk, etc.

First, this is a pet project for many developers that gain lots of knowledge/skills working on graphics drivers.
Second, an old card with even basic support is still better than an old card with no support at all.
Third, there's this dream that at some point all drivers will be open source and kernel and X/Wayland compatibility will be a no issue. Imho, this will never happen because graphics drivers are far, far more complicated beasts than other drivers and graphics cards are being released at a very quick rate. But that's just my opinion, I'd love if reality proved me wrong.

[haagch]

I have a 6870 (Barts). One that will not expose 4.x because of a lack of fp64. My question is, do cards like mine still accel tessellation even if not exposing GL4 support?

Depends.
Tessellation is in OpenGL core since 4.0, so some developers might make it easy and check for OpenGL >= 4.0 and enable tessellation if it is available.
But "the OpenGL extension way" is to test whether the tessellation extension is available regardless of the OpenGL version.

[chrisr]

[W]hat is the purpose of all this work? And by all this, I mean all of the open-source graphics stacks. If it is not maintained by the vendors themselves, what good is it t become feature complete (or not even) by the time even the vendors themselves deem their hardware deprecated ("legacy" is the nice term).

Is this a Troll post? Are you not aware that many of the developers here are employed by AMD ("the vendor"), and that much of this work is made possible by the effort that AMD makes to release their GPU specifications to the Open Source community?

I am speaking as someone who owns HD4890, HD4650, HD4670 and HD6450 cards, and am very happy that they all work beautifully, and that AMD doesn't constantly try to b*tch-slap me into buying the "latest and greatest". (Obviously I will upgrade at least some of these cards at some point, although the HD4670 is AGP and my docking station doesn't have enough room for a card with a massive heat-sink...)

I think the real problem here is that Mesa is still catching up, although the gap has narrowed considerably in recent years.

[bug77]

... AMD doesn't constantly try to b*tch-slap me into buying the "latest and greatest".

Neither do nvidia or intel.

[jf33]

Depends.
Tessellation is in OpenGL core since 4.0, so some developers might make it easy and check for OpenGL >= 4.0 and enable tessellation if it is available.

Which is not a real problem, since you can override the advertised OpenGL version with an environment variable (forgot the name of it). Since almost nobody uses fp64, you can be pretty sure that OpenGL 4.1 games run properly on all HD 5xxx/6xxx gpus.

[Mike Frett]

I have the box for my 6670 right here and it clearly states OpenGL 4.1. So what's going on?

[chrisr]

Neither do nvidia or intel.

And yet this is still my perception of drivers for Windows, which is being touted as an alternative to using an Open Source stack.

[bug77]

And yet this is still my perception of drivers for Windows, which is being touted as an alternative to using an Open Source stack.

I wasn't aware Windows drivers are touted as an alternative to Linux open source drivers. Even so, you can download Windows 10 drivers for intel 900 series which is ~10 years old and for nvidia 8000 series which is 9 years old so I still don't see how either of these "try to b*tch-slap me into buying the "latest and greatest"".

[chrisr]

I wasn't aware Windows drivers are touted as an alternative to Linux open source drivers.

Are you being deliberately obtuse? Windows itself is touted as an alternative to using an Open Source stack running on Linux, and yet the availability (or not) of working Windows drivers for the particular version of Windows that you happen to own (i.e. not necessarily Windows 10) does put you completely at the mercy of the vendors.

[Nille_kungen]

Yes. With this new driver code simpliest way to test on real app is if you try Unigine Heaven benchmark and enable tessellation. GL 4 context is not required there, but tessellation should work.

https://unigine.com/products/benchmarks/heaven/

Or anybody with EG/NI can test, just use this new driver code then run heaven with tessellation enabled... that is enough to test if it works - basically if it render OK (and does that in all of 3 tess modes) there and does not lock up GPU, etc. then it is fine...

I got an NI in my laptop APU and mesa master installed and pts so what do you want me to test.
I started the Unigine Heaven benchmark but how should tesselation be be enabled? (tried "phoronix-test-suite benchmark unigine-heaven", does F3 work when run this way since i didn't see anything different?)

glxinfo does show the card as OpenGL4.1
Tried trine3 but there's some rendering issues (same as before tess when using stock 11.0.6 with mesa overrides), with my radeonsi everything is as it should.