[om2]

We sincerely think a text-based language is better for the web. It’s honestly weird that anyone thinks a binary format is a good webby choice. Both our browser engine people and our GPU software people agree.

Khronos basically said in that meeting that it would be fine to fork SPIR-V, which would solve Apple’s and Microsoft’s issues with their IPR framework. We’ve also discussed using a textual form of the SPIR-V format. We’ve offered all sorts of compromises. It’s Google that isn’t willing to budge, even stating in a WebGPU meeting that they never even considered what compromises would be acceptable to them. Encourage Google to be open to meeting in the middle and maybe we will get somewhere.

[mantap]

I do agree that a text based format is better in many cases. A byte code format would suck for debugging. People want to type their shaders directly into a browser window and have compilation happen in less than a frame. Shader compilation needs to be fast enough that it takes less than a frame because people do have things like colour pickers in their in-browser shader editor where you can drag the value around and it changes the text. To emulate this without browser support would require a lot of work.

But that text based format should be GLSL because that's what everybody's shaders are already written in for WebGL and obviously there will be a transition period where both WebGL and WebGPU will have to be supported (which is easy since most people use a library such as Babylon or Three).

Having a text based language that is not GLSL is pointless IMO. You have the drawbacks of both a bytecode language (need to ship a compiler with the page to compile GLSL into WSL) and textual formats.

As an outsider the most logical option is to support both GLSL and SPIR-V.

[raphlinus]

As someone who really wants to use this stuff, I just want mommy and daddy to stop fighting. I personally lean towards a SPIR-V approach for a number of reasons, including faster shader compilation and a more unified ecosystem. I also see the advantages of a high level text format, but feel like that could get polyfilled in 100k or so of wasm in the most important cases.

But it's clear that either approach can work. From what I'm seeing so far, flipping a coin and doing "disagree and commit" would be a healthier process than what's going on right now.

[flohofwoe]

Why would a shader byte code standard prevent text based shaders?

The only difference would be that (runtime) compilation happens in an optional Javascript / WASM module instead of being baked into the standard and browsers, and Apple can ship such a WHLSL compiler with Safari, so that startup is faster there for WebGPU code which uses WHLSL shaders, but please don't force it on everybody else.

And the "web people" are used now to compile stuff for quite some time (see TypeScript, WebPack, JS Minifiers, etc etc). Just one more compile step won't make much of a difference, but offer a lot of freedom.

(I guess in the end, byte-code versus text representation isn't that much different, but it is important that it is a good compilation target, let's not repeat the long journey from transpiled JS to asm.js to WASM).

[royjacobs]

Would that be a textual format that is essentially already compiled, or do you mean a textual format that is just source (like glsl)? The latter historically has had a bunch of issues with compiler incompatibility across vendors, so I assume you mean the former. But what is then the benefit of having essentially textual bytecode compared to binary?

[blurbleblurble]

> We sincerely think a text-based language is better for the web.

I believe that you and the webkit/metal devs are sincere in that. Apple has a culture of human focused design. And I can tell that a lot of work went into making WSL a good, human friendly GPU shader language that would meet a lot of peoples' needs.

> It’s honestly weird that anyone thinks a binary format is a good webby choice.

Okay so is it weird that common image formats like PNG or JPG are binary formats? Those are the best "webby choices" for images, because they are capable of representing the necessary information to display high quality images on many platforms, and can be generated from a whole vast range of tools that export them. It's not weird at all. You wouldn't argue that the most "webby" choice of image format is ascii text images, just because someone can edit them with a text editor. PNG gives you access to a whole ecosystem of image editing tools that do stuff a text editor won't ever be able to do, producing images that ascii will just never realistically duplicate, even if theoretically it somehow could.

¯\_(ツ)_/¯

What does SPIR-V give us that makes it a good "webby choice" even though it's a binary format? It's a portable, cross-platform Rosetta Stone of GPU functionality, in ways that a high level language like WSL or GLSL by definition can never be. They are just not the same.

The tooling that is already beginning to emerge around SPIR-V will enable possibilities that WSL just won't realistically ever be able to enable, ever. It's highly unlikely that people will develop transpilers for WSL that are as flexible and seamless as the compilers people are already making for SPIR-V. WSL is too opinionated. It's an inherently opinionated high level language.

And once the GPU Web project gives the go-ahead, people are gonna be even more motivated to work on SPIR-V tooling. It will give shader and GPU compute developers freedom to choose from a whole range of high level constructs and preexisting tooling from whatever language best fits their specific application, so that they can focus energy on their actual problems instead of just wrestling with the tooling. On top of that, people will still be able to use and quickly re-purpose shaders written in HLSL, GLSL or WSL.

WSL looks like a fine language for doing linear algebra in low dimensions so that you can rotate some colorful geometry for your brand or whatever.

But it cannot and will not ever be what SPIR-V is. Telling developers that WSL is plenty fine because your specific team of devs have that opinion is like one artist telling another that they ought to be perfectly content with crayons and markers. Sure, you can do some nice stuff with crayons and markers, but there's a whole world of visual art mediums out there once you remove that constraint. What about ceramics? And I've talked to quite a few developers who are excited about SPIR-V. You all might not be that into SPIR-V but you're burying your heads in the sand if you can't see that a lot of people are really into it.

Let me share why I'm so invested personally that I'd sit down and write this unusually long letter:

WebGPU is critical to the direction of some of the work I'm most excited about doing while I'm here on this planet. My goal is to produce interactive multimedia art using Clifford algebra and other interesting mathematical structures, targeting the web browser as a distribution platform. I've been working on this stuff for 10 years now and feel like I have another 10 years to go before I'm really ready to illustrate the stories and concepts I want to express. Soon I'd like to have some real tools that can enable collaboration with other artists and story tellers. But so far I've mostly just been been prototyping.

I've written code to visualize non-euclidean transformations and geometries using conformal geometric algebra, and to do computations in all kinds of other algebras using GLSL, Python (Sage/SymPy), Julia and C. I made a Blender modifier that does conformal inversion. I've written transpilation tools to convert SymPy expressions to GLSL, and used these to create generative art. I wrote an abstract algebra code generator that spits out algebraic glslify compilable modules from SymPy expressions.

But it's so cumbersome and I'm pretty tired. I bumped into the limits of OpenGL ES a while ago. I want to use features from newer versions of OpenGL on the web but it's not possible. WebGL is stalled. WebCL is a little better but it's even more stalled.

Plus, even if I could use the latest versions of OpenGL or OpenCL, generating code in these languages to do the math I want is really cumbersome. There are much better languages for doing math, and people have been writing all kinds of great math libraries in these languages for decades. I want to use those when I'm writing shaders and GPU compute programs. Realistically, I'm going to need to use existing tools if I want to complete the work I dream of, and SPIR-V is beginning to make it possible to use them in a cross platform way.

If WSL becomes the only possible shader language for WebGPU, I'll basically pivot away from the goal of targeting the web browser and move into cross platform desktop/mobile app development, which I'm just not nearly as excited about. It wouldn't be the end of my dreams, but it'd mean giving up on the web browser.

> We’ve offered all sorts of compromises. It’s Google that isn’t willing to budge.

On a personal level, I trust your sincerity and respect your passion, but having gone through a bunch of meeting minutes and I just didn't get that impression, and am left feeling pretty disappointed all in all with how Apple as a corporation has engaged in and publicly represented the process.

All that to say, I'm really optimistic about the progress you all have made in getting to the point of meeting with Khronos and hashing things out. Serious props on that.

And I hope you don't take my attitude toward WSL as being dismissive of it entirely. It honestly looks great for what it is. It just doesn't suit my needs.

https://drive.google.com/drive/folders/1tM5KAZ7S52cv3jnpPOZs...

---------

Relevant meeting minutes

Google/Mozilla people discussing WSL:

https://docs.google.com/document/d/1BOvJKklz-4PZN2StCU56nN4m....

Apple people making their case for WSL:

https://docs.google.com/document/d/1CmKo59tjZwmePVrFpHpIG0W5....

https://docs.google.com/document/d/1opv8MIK94DNIKU5qbgeqlkkT....

The GPU Web group prepares for the Khronos meeting:

https://docs.google.com/document/d/1RI5hgOFuOH0-v9MPxuoV6w2L....

Meeting with the Khronos laison:

https://docs.google.com/document/d/1F6ns6I3zs-2JL_dT9hOkX_25....