Unknown attribute error

[Beyley]

I am attempting to port the https://github.com/gfx-rs/wgpu-native/tree/master/examples/compute project to C# as an example for the WIP Silk.NET bindings, but upon attempting to compile the shader i get an error claiming '@compute' is an invalid attribute, even though every example of WGSL compute ive seen shows that attribute being used, is this possibly some problem with the bindings or some breaking change in wgpu-native and the example hasnt been updated to address it?

Code:
https://github.com/dotnet/Silk.NET/blob/7984f8aabf08c765c56f4f7f0e5eaf35e5ce2c49/src/Lab/Experiments/WebGPUTest/Program.cs

Error: Error: Internal Message: Parsing(ShaderError { source: "@group(0)\n@binding(0)\nvar<storage, read_write> v_indices: array<u32>; // this is used as both input and output for convenience\n\n// The Collatz Conjecture states that for any integer n:\n// If n is even, n = n/2\n// If n is odd, n = 3n+1\n// And repeat this process for each new n, you will always eventually reach 1.\n// Though the conjecture has not been proven, no counterexample has ever been found.\n// This function returns how many times this recurrence needs to be applied to reach 1.\nfn collatz_iterations(n_base: u32) -> u32{\n var n: u32 = n_base;\n var i: u32 = 0u;\n loop {\n if (n <= 1u) {\n break;\n }\n if (n % 2u == 0u) {\n n = n / 2u;\n }\n else {\n // Overflow? (i.e. 3*n + 1 > 0xffffffffu?)\n if (n >= 1431655765u) { // 0x55555555u\n return 4294967295u; // 0xffffffffu\n }\n\n n = 3u * n + 1u;\n }\n i = i + 1u;\n }\n return i;\n}\n\n@compute\n@workgroup_size(1)\nfn main(@builtin(global_invocation_id) global_id: vec3<u32>) {\n v_indices[global_id.x] = collatz_iterations(v_indices[global_id.x]);\n}", label: Some("Test Compute Shader"), inner: ParseError { message: "unknown attribute: 'compute'", labels: [(975..982, "unknown attribute")], notes: [] } })

[ErichDonGubler]

TL;DR: I would bet that if you use a master build of wgpu-native, you'll be able to compile this shader as you expect.

---

Questions

I gave up trying to repro this issue, because of these obstacles:

1. It's not clear to me what instructions should be followed for reproducing this locally with the given Silk.NET example. The translation looks more or less correct to me, but I wasn't able to determine how to run it.
2. It's not clear how you're consuming wgpu-native. It looks like bindings are being generated via NUKE (which is a build tool I'm not familiar with), but it's unclear where the headers used to generate *.gen.cs files and corresponding libs are coming from.

Without these, there's several leads to a fix that I see that can't be tested:

1. What version of wgpu-native are you consuming? Is there maybe a version mismatch? This seems likely, since naga (which handles WGSL input) 0.10 is the first version of W
2. Is there maybe an issue with bindings being generated?
3. Is this an environmental issue?

Bonus question: Why not file this discussion in wgpu-native, which is the layer you're actually depending on?

---

Thoughts

• The @compute attribute is specified by the WGSL spec:

  Declares the function to be an entry point for the compute shader stage of a compute pipeline.

  So, it's definitely supposed to be supported at time of writing. I even tried copying over the version in your linked source to see if that caused any problems. I saw no difference in output (looks like the only changes are a few newlines removed, compared to the original).

• The compute example works locally on my Macbook Pro with M1 silicon:

  # Check out `wgpu-native`.
  # I tested on the current `master` branch: f3228dc9ad19dc4cbe43eff65ec532e6f94f7e9f.
  $ cd wgpu-native 
  
  $ cargo build
  <snip>
      Finished dev [unoptimized + debuginfo] target(s) in 0.07s
  
  $ cd examples/compute
  $ cmake . && make
  Times: [0, 1, 7, 2]

  N.B. that the latest commit in wgpu-native's master branch is consuming wgpu 0.14 and naga 0.10, which are very recent (ATOW). AFAICT, the consumption of these versions hasn't yet been released in wgpu-native yet.

[Beyley]

TL;DR: I would bet that if you use a master build of wgpu-native, you'll be able to compile this shader as you expect.

Questions

I gave up trying to repro this issue, because of these obstacles:

1. It's not clear to me what instructions should be followed for reproducing this locally with the given Silk.NET example. The translation looks more or less correct to me, but I wasn't able to determine how to run it.

2. It's not clear how you're consuming `wgpu-native`. It looks like bindings are being generated via NUKE (which is a build tool I'm not familiar with), but it's unclear where the headers used to generate `*.gen.cs` files and corresponding libs are coming from.

Without these, there's several leads to a fix that I see that can't be tested:

1. What version of `wgpu-native` are you consuming? Is there maybe a version mismatch? This seems likely, since `naga` (which handles WGSL input) 0.10 is the first version of W

2. Is there maybe an issue with bindings being generated?

3. Is this an environmental issue?

Bonus question: Why not file this discussion in wgpu-native, which is the layer you're actually depending on?

Thoughts

* The `@compute` attribute is specified by the [WGSL spec](https://gpuweb.github.io/gpuweb/wgsl/#attribute-compute):
  > Declares the function to be an [entry point](https://gpuweb.github.io/gpuweb/wgsl/#entry-point) for the [compute shader stage](https://gpuweb.github.io/gpuweb/wgsl/#compute-shader-stage) of a [compute pipeline](https://gpuweb.github.io/gpuweb/wgsl/#gpucomputepipeline).
  
  
  So, it's definitely _supposed_ to be supported at time of writing. I even tried copying over the version in your linked source to see if that caused any problems. I saw no difference in output (looks like the only changes are a few newlines removed, compared to the original).

* The `compute` example works locally on my Macbook Pro with M1 silicon:
  ```shell
  # Check out `wgpu-native`.
  # I tested on the current `master` branch: f3228dc9ad19dc4cbe43eff65ec532e6f94f7e9f.
  $ cd wgpu-native 
  
  $ cargo build
  <snip>
      Finished dev [unoptimized + debuginfo] target(s) in 0.07s
  
  $ cd examples/compute
  $ cmake . && make
  Times: [0, 1, 7, 2]
  ```
  
  
      
        
      
  
        
      
  
      
    
  N.B. that the latest commit in `wgpu-native`'s `master` branch is consuming [`wgpu` 0.14](https://github.com/gfx-rs/wgpu/releases/tag/v0.14.0) and [`naga` 0.10](https://github.com/gfx-rs/naga/releases/tag/v0.10.0), which are very recent (ATOW). AFAICT, the consumption of these versions hasn't yet been released in `wgpu-native` yet.

My bad for not being clear about running the app locally, my WebGPU stuff is very WIP currently, and yes since you mentioned it, its very possible its a mismatch of library versions, i downloaded the latest version from github releases, which may be out of date compared to the master branch of webgpu-headers (which are the headers im parsing to generate the bindings)
And to answer the bonus question: i hadn't filed this under wgpu-native or webgpu-headers as after reading the code, it seemed more possible to be a problem with the underlying wgpu code, but i wasn't totally sure, so i just filed it here, i shall try building wgpu-native from source and come back when i have results

[Beyley]

Doh yeah you were right! i built wgpu-native from source and it seems to work now! thank you for helping me with my stupidity

[ErichDonGubler]

thank you for helping me with my stupidity

Trying the first thing that seems like it should work isn't stupid! WGSL is still technically a moving target right now, so while it's a bummer that you had an issue, it isn't necessarily surprising. I'm happy that you were able to get a fast resolution. :)