Imagination Tech Publishes Open-Source PowerVR Vulkan Driver For Mesa

[ayumu]

Obviously it won't be as good as a native driver.

There's no such thing as a native opengl driver. Neither hardware is designed around opengl, nor opengl "drivers" hit the hardware directly. In mesa, you get to pick which abstraction of the hardware to hit.

Vulkan is a very good choice for any hardware designed today. I expect future hardware will settle on Zink for their opengl support.

[Quackdoc]

There's no such thing as a native opengl driver. Neither hardware is designed around opengl, nor opengl "drivers" hit the hardware directly. In mesa, you get to pick which abstraction of the hardware to hit.

Vulkan is a very good choice for any hardware designed today. I expect future hardware will settle on Zink for their opengl support.

This isn't true. gallium drivers would be considered native gallium drivers.

Now with the change to a abstracting the gallium, for zink it would be GL->Gallium->zink/VK->vulkan driver. vs GL->Gallium->driver.

zink will add overhead vs using whatever commands the gpu understands directly from gallium. it is overhead that is absolutely noticeable for now. will it always be? I have not a clue

[microcode]

When I was commenting this about their recent hardware releases, didn't think it was already in the pipeline lol. Cool, good to see that some things sometimes change. Hopefully this doesn't stop with chips for Chromebooks, where Google has done a lot of work making upstream Mesa attractive for SoC vendors.

[ms178]

Zero chance of that.

Right in the sense that at least Nvidia is not willing to. But as the hacker group might release the source code of their proprietary driver any second now, that might make it a source of inspiration for open source developers to look at.

[Tomin]

Right in the sense that at least Nvidia is not willing to. But as the hacker group might release the source code of their proprietary driver any second now, that might make it a source of inspiration for open source developers to look at.

Open source developers should never look at Nvidia's closed source driver. It contains proprietary code and thus it's better to base the driver on reverse engineered information or on information that Nvidia has released to avoid any copyright issues. Generally this kind of leaks just hurt Open Source projects. Also it's likely that Nvidia can't just release their driver Open Source. It's possible that Nvidia contracts which prevent them from open sourcing parts of the driver. Remember that AMD didn't just go and open source their driver but rewrote the whole thing Open Source from the beginning.

Anyway most of the issues with nouveau nowadays are due to Nvidia refusing to provide all the needed binary blobs in a way that nouveau could use. Figuring out ISA etc. is not too difficult and after that implementing compilers etc. is just work. Thus I'd question how much it would help even if nouveau developers were allowed to read Nvidia's driver code.

[Quackdoc]

Open source developers should never look at Nvidia's closed source driver. It contains proprietary code and thus it's better to base the driver on reverse engineered information or on information that Nvidia has released to avoid any copyright issues. Generally this kind of leaks just hurt Open Source projects. Also it's likely that Nvidia can't just release their driver Open Source. It's possible that Nvidia contracts which prevent them from open sourcing parts of the driver. Remember that AMD didn't just go and open source their driver but rewrote the whole thing Open Source from the beginning.

Anyway most of the issues with nouveau nowadays are due to Nvidia refusing to provide all the needed binary blobs in a way that nouveau could use. Figuring out ISA etc. is not too difficult and after that implementing compilers etc. is just work. Thus I'd question how much it would help even if nouveau developers were allowed to read Nvidia's driver code.

IMO licensing issues are one reason, and the fact that nvidia uses software locks for enterprise features is another potential issue with open sourcing their driver, I mean vgpu was already cracked, who knows what other goodies might become available if they do

[coder]

Open source developers should never look at Nvidia's closed source driver. It contains proprietary code and thus it's better to base the driver on reverse engineered information or on information that Nvidia has released to avoid any copyright issues.

This. 100%. An IP lawyer once told me about the dangers of making a derivative-work. Just because some code is dumped on the public internet doesn't invalidate its copyright. If you download it or share it, you're committing a copyright violation. Use that code in some other driver, and that driver is now potentially a derivative work. It's best for open source drivers not to be tainted by such things, or else we could end up with something like that SCO lawsuit situation (although I think that was mostly about patents, not copyrights - also a risk increased by not using a cleanroom implementation).

Anyway, it's not only proprietary GPU makers who have smart people working on their drivers.

The best case scenario is probably where someone reads the proprietary driver to help them document the hardware. Then, another person might use information from that document to enhance the open source driver. As long as the information isn't about the Nvidia driver, but just the hardware, you might avoid the derivative-work issue. So, even that wouldn't be as good as Nvidia just releasing full hardware details.

Figuring out ISA etc. is not too difficult and after that implementing compilers etc. is just work. Thus I'd question how much it would help even if nouveau developers were allowed to read Nvidia's driver code.

I'm sure there's a lot on information in the proprietary driver about how to get better performance from the hardware. Also, these GPUs have lots of fixed-function units that the open source devs might not be able to make complete sense of how to use, fully.

[coder]

IMO licensing issues are one reason, and the fact that nvidia uses software locks for enterprise features is another potential issue with open sourcing their driver, I mean vgpu was already cracked, who knows what other goodies might become available if they do

Well, here's a case where a code dump might not help at all. Depends on whether they simply commit their keys to the source tree, which I rather doubt. If I were them, I'd manage the keys in a separate system.

Also, it makes me wonder whether Nvidia could or did copyright their keys. If so, then even having them wouldn't let you use them in the open source driver.

[brad0]

Right in the sense that at least Nvidia is not willing to. But as the hacker group might release the source code of their proprietary driver any second now, that might make it a source of inspiration for open source developers to look at.

Open source projects almost always try their best to stay away from code like this. There is too great of a chance of copying proprietary information that makes it very clear that they did look at their closed source code and tainting the open source code.

[brad0]

But not those of third parties, and not for code that doesn't directly use the patent. Note how the M1 doesn't support OpenGL directly either.

It's not a matter of supporting OpenGL directly. It just doesn't have hardware features that OpenGL requires, and of course it doesn't when Apple uses Metal not OpenGL or Vulkan.