Nouveau Adds ARB_copy_image, Intel Adds Another OpenGL 4.3 Extension Too

[phoronix]

Phoronix: Nouveau Adds ARB_copy_image, Intel Adds Another OpenGL 4.3 Extension Too

Following AMD adding ARB_copy_image support to the RadeonSI driver, Nouveau's NV50 and NVC0 Gallium3D drivers have now been wired up for this OpenGL extension...

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

[justmy2cents]

this race is getting interesting. radeonsi that overtake nouveau got overtaken back in one day.

[cbxbiker61]

this race is getting interesting. radeonsi that overtake nouveau got overtaken back in one day.

I'm not sure if it's a race, it seems more like synergy between the different teams. In my view this is open source at it's best.

[Namenlos]

(slightly offtopic)Spoiler: From the mailinglist it seems like the last few patches of the GL_ARB_arrays_of_arrays extension will be completely reviewed mid to end November. So if everything goes well we will have that extension before December and i965 will be only one extension away from OpenGL ES 3.1, which seems like the biggest priority for the Intel folks, after that tessellation is probably very high on the list, for OpenGL 4.0, OpenGL ES 3.1 and the AEP (Android).

[jf33]

this race is getting interesting. radeonsi that overtake nouveau got overtaken back in one day.

It's not a race at all. It's more like AMD implemented the ARB_copy_image extension in hard work and after that, Ilia Mirkin only had to enable it also for nouveau (since both drivers use gallium). I don't want to disesteem Ilia Mirkin, who did a lot of good work for Mesa in the past, but this achievement was easy to get.

[Gusar]

I find it a bit amusing that two people apparently took justmy2cents' comment about a "race" seriously. I'm quite sure justmy2cents meant it in a playful tone. Because of course there isn't a race, but the mesamatrix page contains a table at the top listing how many extensions each driver supports, so it's easy to get into the spirit of a race and root for a specific driver

[Ancurio]

I find it a bit amusing that two people apparently took justmy2cents' comment about a "race" seriously. I'm quite sure justmy2cents meant it in a playful tone. Because of course there isn't a race, but the mesamatrix page contains a table at the top listing how many extensions each driver supports, so it's easy to get into the spirit of a race and root for a specific driver

I still think painting this as a competition is a bit disrespectful towards the developers. They're doing the exact opposite, working together to get to a common goal, and some of them even do so in their free time.

[Kayden]

i965 will be only one extension away from OpenGL ES 3.1, which seems like the biggest priority for the Intel folks, after that tessellation is probably very high on the list, for OpenGL 4.0, OpenGL ES 3.1 and the AEP (Android).

Yup. I've been working on tessellation for a month or two now. Ilia's been helping me on IRC.

At this point, I have 99% of the Piglit tests passing on Broadwell, and understand why the remaining 2 tests fail. Most of dEQP seems pretty happy as well. I've got two demo applications working. There are apparently a couple of bugs left that break application rendering pretty horribly (triangles everywhere!), so I'm not able to run Tessmark, Heaven, or Shadow of Mordor quite yet. Hopefully soon!

Things left to do (in some order):
- Find and squash that one awful bug
- Fix isoline problems
- Deal with SSO programs that want to mix-and-match TCS and TES independently
- Implement a dummy TCS for when the application doesn't provide one (really weird case...ES doesn't even allow this)
- Backport to Haswell / Ivybridge (add more workarounds), forward port to Skylake (hopefully just testing)
- Enable SIMD8 evaluation shader mode (optional - hopefully better performance)
- Make TCS output shadowing more competent and re-enable it (optional, better performance)
- Fix and re-enable varying packing for TCS inputs (optional)
- Experiment with SIMD8 control shaders (optional, for performance)
- Lots of performance tuning

It's really exciting to think where we'll be in a year. Most of the free software drivers should have a ton of GL 4.x done.

[Gusar]

I still think painting this as a competition is a bit disrespectful towards the developers. They're doing the exact opposite, working together to get to a common goal, and some of them even do so in their free time.

Err, the whole point of my post was to highlight that it's *not* being painted as a competition by anyone. Jeez, people, lighten up. I respect all mesa devs, those paid to work on it and especially the volunteers like Ilia. I'm sure justmy2cents does too. We're happy to see the progress being made and are having some fun.