NVIDIA+CMU Develop New Shading Language & Compiler Framework

[phoronix]

Phoronix: NVIDIA+CMU Develop New Shading Language & Compiler Framework

NVIDIA and Carnegie Mellon University have developed a new open-source shading language in step with a new compiler framework...

http://www.phoronix.com/scan.php?page=news_item&px=NVIDIA-CMU-Spire

[cj.wijtmans]

why ?

[starshipeleven]

why ?

Because world domination.

What's weird is that it isn't NVIDIA-specific. They must feel very confident that all high-end stuff is done on NVIDIA anyway.

[chuckula]

why ?

If you bothered to look at the link you'd see substantial performance improvements using the new open source shader language that's compatible with OGL and Vulkan.

So I guess there's no reason to do any research in the area of graphics if you don't want improved performance.

But for literally everybody else it seems like the reason of why to do the research is pretty obvious.

[fuzz]

This would be neat to see benchmarks on AMD vs NVIDIA.

[ddriver]

If you bothered to look at the link you'd see substantial performance improvements using the new open source shader language that's compatible with OGL and Vulkan.

If you bothered to understand the content of the link, you'd realize that it is not that much of a "new shader language" but more of a shader preprocessor that allows to include individual modules, which for some reason core shader languages never bothered to implement, although some 3rd party tools have made that.

It is still all GLSL or SPIR-V or whatever shaders, but rather than copy and pasting huge walls of code to reuse existing functionality and thereby introduce useless junk, you can use some sort of modularity directives, which the preprocessor uses to inline only the bare essentials for what you need.

In short, it is not a new language, it is some very helpful and long overdue syntax sugar. And it is realized in the form of a preprocessor parser and a code generator. You still use the same old shading language.

[duby229]

Because world domination.

What's weird is that it isn't NVIDIA-specific. They must feel very confident that all high-end stuff is done on NVIDIA anyway.

Remember vdpau? It turned out to be huge success. I had always hoped they would learn a huge lesson from that. If nVidia can keep releasing useful and usable code then it's gonna get used. So lets just keep hoping.

[tuke81]

Remember vdpau? It turned out to be huge success. I had always hoped they would learn a huge lesson from that. If nVidia can keep releasing useful and usable code then it's gonna get used. So lets just keep hoping.

While yes it was great success, I don't think nvidia will develop it in future anymore(no hevc 10bit nor vp9). If I remember correctly Vdpau developer left the company and they have cross platform proprietary nvdec/nvenc now. Which makes me wonder if they even add vdpau support for volta, well maybe they will just for compatibility.

[duby229]

While yes it was great success, I don't think nvidia will develop it in future anymore(no hevc 10bit nor vp9). If I remember correctly Vdpau developer left the company and they have cross platform proprietary nvdec/nvenc now. Which makes me wonder if they even add vdpau support for volta, well maybe they will just for compatibility.

Oh well, that's not good to hear. I keep hoping for better, and I thought vdpau was a rare example of that. So I guess I'm a bit disappointed to hear this.

[emblemparade]

I'm very impressed by Spire. Until now, shader languages have been basically variants of the C language, which is familiar and good, but offers poor reusability. Spire makes it much easier to make algorithms reusable. I think this will prove especially important for Vulkan, which is so bare bones.

Apparently compiling C-like code makes it hard to optimize for modern GPUs. so with Spire we can create more optimized shader output.

And open source? This is 100% win.