pygfx
missing rgba16unorm
Describe the bug
I'm attempting to use rgba16unorm as a storage texture type per the WGSL spec: https://www.w3.org/TR/WGSL/#texel-formats but it is missing from https://wgpu-py.readthedocs.io/en/stable/wgpu_enums.html#wgpu.enums.TextureFormat and I get: KeyError: 'TextureFormat.rgba16unorm'
To Reproduce
Attempt to create an output texture via:
>>> import wgpu
>>> gpu = wgpu.utils.get_default_device()
>>> tex_out = gpu.create_texture(size=(16,16,1),
... format=wgpu.TextureFormat.rgba16unorm,
... usage=wgpu.TextureUsage.COPY_SRC | wgpu.TextureUsage.STORAGE_BINDING,
... sample_count=1)
Traceback (most recent call last):
File "<stdin>", line 2, in <module>
AttributeError: type object 'TextureFormat' has no attribute 'rgba16unorm'. Did you mean: 'rgba8unorm'?
>>>
Observed behavior
See above.
Screenshots
n/a
Your environment
Details about the environment this issue was observed on, like OS, Python version, wgpu-py version.
You can also include (the relevant sections of) the result of wgpu.diagnostics.print_report().
>>> wgpu.diagnostics.print_report()
██ system:
platform: macOS-14.7-arm64-arm-64bit
python_implementation: CPython
python: 3.10.14
██ versions:
wgpu: 0.22.2
cffi: 1.17.1
██ wgpu_native_info:
expected_version: 24.0.3.1
lib_version: 24.0.3.1
lib_path: ./resources/libwgpu_native-release.dylib
██ object_counts:
count resource_mem
Adapter: 1
BindGroup: 0
BindGroupLayout: 0
Buffer: 0
CanvasContext: 0
CommandBuffer: 0
CommandEncoder: 0
ComputePassEncoder: 0
ComputePipeline: 0
Device: 1
PipelineLayout: 0
QuerySet: 0
QuerySet: 0
Queue: 1
RenderBundle: 0
RenderBundleEncoder: 0
RenderPassEncoder: 0
RenderPipeline: 0
Sampler: 0
ShaderModule: 0
Texture: 0
TextureView: 0
total: 3 0
██ wgpu_native_counts:
count mem hub a k r el_size
Adapter: 1 8 hub: 1 1 0 8
BindGroup: 0 0 hub: 0 0 0 16
BindGroupLayout: 0 0 hub: 0 0 0 16
Buffer: 0 0 hub: 0 0 0 16
CanvasContext: 0 0 0 0 0 8
CommandBuffer: 0 0 hub: 0 0 0 8
ComputePipeline: 0 0 hub: 0 0 0 16
Device: 1 8 hub: 1 1 0 8
PipelineCache: 0 0 hub: 0 0 0 16
PipelineLayout: 0 0 hub: 0 0 0 16
QuerySet: 0 0 hub: 0 0 0 16
Queue: 1 8 hub: 1 1 0 8
RenderBundle: 0 0 hub: 0 0 0 16
RenderPipeline: 0 0 hub: 0 0 0 16
Sampler: 0 0 hub: 0 0 0 16
ShaderModule: 0 0 hub: 0 0 0 16
Texture: 0 0 hub: 0 0 0 16
TextureView: 0 0 hub: 0 0 0 16
total: 3 24
* The a, k, r are allocated, kept, and released, respectively.
* Reported memory does not include buffer/texture data.
looks like the version of webgpu.h we have in wgpu-native doesn't have it. But it's available as a native only texture format... so it's not part of the enum or mapping. You can try to just set the magic value from here: https://github.com/pygfx/wgpu-py/blob/d25edb102644f74c0d12e24996f757eac77b57fa/wgpu/resources/wgpu.h#L247 or format=lib.WGPUNativeTextureFormat_Rgba16Unorm(but I don't think lib is exposed) it might work.
Interestingly, the format is present in the WebGPU spec, so I image it will land in webgpu.h at some point.
Perhaps for things like this, we can add it to our API, even when its not in webgpu.h, when:
- WebGPU has it.
- We can support it, e.g. because its in
wgpu.h.
The enums fields are now available (added in #743), but they are not supported by the current wgpu-native yet.
They are present in the current webgpu.h (https://github.com/webgpu-native/webgpu-headers/blob/main/webgpu.h#L1012). But wgpu-native is pinned to an older version that does not have it yet.
So to get proper support for the 16unorm texture formats and friends:
- wgpu-native should update to the latest
webgpu.h. - wgpu-native should do a release.
- we should update to that version of wgpu-native.