Nouveau NVC0 Gallium3D Driver Now Exposes GLSL 4.10

[phoronix]

Phoronix: Nouveau NVC0 Gallium3D Driver Now Exposes GLSL 4.10

While there's still more work to be done before advertising OpenGL 4.0~4.1 compliance, the Nouveau NVC0 Gallium3D driver is now advertising support for GLSL 410 (4.10), the GL Shading Language version to match OpenGL 4.1...

http://www.phoronix.com/scan.php?page=news_item&px=Nouveau-Gallium3D-GLSL-410

[imirkin]

Phoronix: Nouveau NVC0 Gallium3D Driver Now Exposes GLSL 4.10

http://www.phoronix.com/scan.php?pag...ium3D-GLSL-410

Clarification, lest someone misunderstand what's going on -- this is the gallium driver replying with 410 to the PIPE_CAP_GLSL query. I did that out of convenience, not because nouveau secretly supports GLSL 4.10. To counteract this, I also pushed a patch that will still limit the version actually reported to 3.30 when GL 3.3 is exposed.

That said, any interested parties can test out https://github.com/imirkin/mesa/commits/gl4-integration which will expose GL 4.1 on nvc0. It's pulling in tess, subroutine, and attrib_64bit branches to achieve this. I'll be rebasing/updating it. While Heaven 4.0 kinda works (with tess) on this branch, there are a bunch of funky triangles popping up all over the place on Fermi, and additional issues on Kepler+. [And I have an unpushed patch to fix the crazy squares that pop up in Heaven 4.0 as well. That should be making it in soon-ish.]

[kmare]

Thank you a lot for your hard work imirkin, it's very appreciated by many. Keep it up!

[GreatEmerald]

Hmm, the new gl3.txt layout doesn't include GLSL as separate items any more. Strange...

[imirkin]

Hmm, the new gl3.txt layout doesn't include GLSL as separate items any more. Strange...

That's because GLSL isn't a separate item from GL. It matches up 1:1. You can't have "done" GLSL 4.10 without also doing GL 4.1, for example.

And as far as work items, almost all changes between recent (3.x+) GL versions are fully encompassed by the relevant extensions. The big one where that's not the case is GL 3.2 which added geometry shaders which were (slightly) different from ARB_geometry_shader4, and one of the later GL 4.x's adds a GL_MAX_VERTEX_ATTRIB_STRIDE or something which doesn't come from an ext.

[GreatEmerald]

That's because GLSL isn't a separate item from GL. It matches up 1:1. You can't have "done" GLSL 4.10 without also doing GL 4.1, for example.

I thought that might be it. But couldn't the reverse happen, as in having GL 4.1 but not the bits for GLSL 4.10?

[imirkin]

I thought that might be it. But couldn't the reverse happen, as in having GL 4.1 but not the bits for GLSL 4.10?

Nope. GL 4.1 is just GL 4.0 + a bunch of extensions. Just as GL 4.0 is GL 3.3 + a bunch of extensions.

There are *very* minor pieces, like enabling #version 410 in the glsl compiler, and making sure that the right things are enabled, but that's already all done.

[OneTimeShot]

While Heaven 4.0 kinda works (with tess) on this branch, there are a bunch of funky triangles popping up all over the place on Fermi, and additional issues on Kepler+. [And I have an unpushed patch to fix the crazy squares that pop up in Heaven 4.0 as well. That should be making it in soon-ish.]

Nice - can you post a YouTube video showing your progress? I'd like to see what (if any) difference tessellation makes...

[chris200x9]

Does this help mesa as a whole or is it very nouveau specific?

[imirkin]

Nice - can you post a YouTube video showing your progress? I'd like to see what (if any) difference tessellation makes...

Way too lazy to make a video, but here is a pair of screenshots of the same scene with and without tess. Looks like they use it mostly to make the stones look more cobble-stoney.

http://imgur.com/Ffyf3xa,qaeXYmY#1

And like I said, there's still a pile of fail in tess...