Mesa 21.2 Lands NVIDIA's Code For Handling Alternate GBM Backends

[phoronix]

Phoronix: Mesa 21.2 Lands NVIDIA's Code For Handling Alternate GBM Backends

Earlier this year was the proposed NVIDIA code from NVIDIA for allowing Mesa's GBM to support alternative back-ends. This support is notable given that most Wayland compositors are catering to using Mesa's Generic Buffer Manager (GBM) rather than EGLStreams or other options for buffer management. That support code has now been merged into Mesa 21.2...

https://www.phoronix.com/scan.php?page=news_item&px=Mesa-21.2-Alternate-GBM

[smartalgorithm]

I always knew it

[V1tol]

Now unlucky notebook users need Wayland+PRIME offloading working.

[Vorpal]

This seems really nice. Sure it would be better if nvidia went open source / nouveau was better etc. But that doesn't seem likely in the short term at least. And given that some of us are stuck with nvidia with no viable choice (corporate laptops, machine learning research etc) I welcome anything that will make it possible to move forward from X11.

[ix900]

Mesa (open source) is being used here to do this and the proprietary haters will still complain lol.

[uid313]

This seems really nice. Sure it would be better if nvidia went open source / nouveau was better etc. But that doesn't seem likely in the short term at least. And given that some of us are stuck with nvidia with no viable choice (corporate laptops, machine learning research etc) I welcome anything that will make it possible to move forward from X11.

nouveau seems very nice, it does have OpenGL 4.5 support (and only two extensions away from achieving OpenGL 4.6 support). It does have OpenGL ES 3.1 support (and only two extensions away from achieving OpenGL ES 3.2 support). The really major problem is its lack of reclocking support which without it, it is very slow.

Is there any progress on this?
Nvidia so far seems to have no interest in helping out.
How feasible is it for for independent developers without access to any source code or documentation to implement reclocking support?

[gufide]

Seems like they abandonned the idea of a new allocator covering their needs in addition of wayland needs.

[Myownfriend]

I'm really hoping most of the driver-side work is done for this so it can be released soon. Since Nvidia's drivers are just blobs, we don't know if they already setup the driver to work without the EGLStreams code. It may just be a matter of them waiting to see what the final backend swapping code looks like so they can make last minute tweaks.

Maybe I'm just impatient but it would be cool if this worked by the time the 470 drivers are finalized.

[BwackNinja]

I'm really hoping most of the driver-side work is done for this so it can be released soon. Since Nvidia's drivers are just blobs, we don't know if they already setup the driver to work without the EGLStreams code. It may just be a matter of them waiting to see what the final backend swapping code looks like so they can make last minute tweaks.

Maybe I'm just impatient but it would be cool if this worked by the time the 470 drivers are finalized.

This was done by an Nvidia developer, and the same one who was working on liballocator previously. From the merge request[1], it was clear that he was already testing with an Nvidia backend, so this should be released soon.

[1] https://gitlab.freedesktop.org/mesa/..._requests/9902

[ssokolow]

Is there any progress on this?
Nvidia so far seems to have no interest in helping out.
How feasible is it for for independent developers without access to any source code or documentation to implement reclocking support?

For reclocking to work, the nVidia-signed firmware for the relevant component must be uploaded into the card as part of the initialization.

To get the firmware, either nVidia must provide a blob that is licensed for redistribution or the Nouveau developers must commit to writing something that can extract it from each new release of the nVidia binary driver on the end-user's machine, as if they're modding a game with no modding API, similar to how nVidia massages around the GPL taking effect on distribution using an open-source glue stub that gets compiled on the end-user's machine.

Naturally, the Nouveau developers don't want to commit to the latter.