Intel's Gallium3D Linux Driver Now Exposes OpenGL 4.6

[phoronix]

Phoronix: Intel's Gallium3D Linux Driver Now Exposes OpenGL 4.6

Just a few weeks after Intel's i965 OpenGL driver in Mesa added GL 4.6 support, the "Iris" Gallium3D driver is now exposing OpenGL 4.6 support too...

http://www.phoronix.com/scan.php?page=news_item&px=Intel-Gallium3D-OpenGL-4.6

[nuetzel]

And after all the heavy-lifting done with the infrastructure changes for getting the SPIR-V extensions in place for Gallium3D chances are that RadeonSI could follow in time.

[FireBurn]

And after all the heavy-lifting done with the infrastructure changes for getting the SPIR-V extensions in place for Gallium3D chances are that RadeonSI could follow in time.

I'm pretty sure Marek has been working on exactly this. It would mean switching radeonsi over to using NIR rather than TGSI

[smitty3268]

I'm pretty sure Marek has been working on exactly this. It would mean switching radeonsi over to using NIR rather than TGSI

There were comments by developers a few weeks ago that there were only a couple regressions left in the NIR backend. Expect them to switch over to that by default for 19.3. I would assume GL 4.6 support should be pretty simple to enable at that point as well.

[tildearrow]

I like how the Mesa Intel driver always is ahead of the Windows driver for a period of time.
Intel took like a year to support 4.5 on Windows, while Linux already did it in 2016.
This time it took Intel a few weeks to have 4.6 on Windows.

[Kayden]

To be fair, the Mesa driver had conformant 4.6 a long time ago. We just didn't land it because we knew there were some pretty major gaps in the SPIR-V testing and wrote piles and piles of tests. I can't speak to anyone else's quality—maybe other vendors got it right the first time—but the extra testing was really helpful for us. We found a lot of bugs above and beyond the initial CTS (turns out supporting a whole new shading language front-end is hard!).

There's always a balance between "take time to get it right" and "ship it quickly, but fix it before anyone notices". Everyone in the industry plays that game to some extent. You win a lot of splash with the latter approach, but you earn a reputation for solid quality with the former. There's a difficult balance to be found there. I think we've done a decent job balancing it over the years, but I am biased. :)

Kudos to everyone that got this done. SPIR-V support in GL was a huge amount of work, even on top of the original SPIR-V support for Vulkan.

[nuetzel]

To be fair, the Mesa driver had conformant 4.6 a long time ago. We just didn't land it because we knew there were some pretty major gaps in the SPIR-V testing and wrote piles and piles of tests. I can't speak to anyone else's quality—maybe other vendors got it right the first time—but the extra testing was really helpful for us. We found a lot of bugs above and beyond the initial CTS (turns out supporting a whole new shading language front-end is hard!).

There's always a balance between "take time to get it right" and "ship it quickly, but fix it before anyone notices". Everyone in the industry plays that game to some extent. You win a lot of splash with the latter approach, but you earn a reputation for solid quality with the former. There's a difficult balance to be found there. I think we've done a decent job balancing it over the years, but I am biased. :)

Kudos to everyone that got this done. SPIR-V support in GL was a huge amount of work, even on top of the original SPIR-V support for Vulkan.

Yeah!

Very BIG thank you to Intel and igalia!

[pal666]

are there known (app)users of 4.6?

[nuetzel]

are there known (app)users of 4.6?

We have what we have. - And one point more crossed. Time for other jobs.