Viewer should warn & not start when Rgba32Float is not renderable

[DobromirM]

Describe the bug
Hovering over a 3D view causes it to freeze until the cursor is removed from it, returning the following error:

Demo:
https://github.com/user-attachments/assets/e05afc1c-1024-4cac-adc7-44b5e6df8bba

Backtrace

[2024-12-15T16:26:42Z ERROR wgpu_core::device::global] Device::create_texture error: Texture usages TextureUsages(RENDER_ATTACHMENT) are not allowed on a texture of type Rgba32Float due to downlevel restrictions
[2024-12-15T16:26:42Z ERROR wgpu_core::device::global] Texture::create_view(Id(28,3)) error: Texture with 'DepthCopyWorkaround::readable_texture' label is invalid
[2024-12-15T16:26:42Z ERROR wgpu_core::device::global] Device::create_render_pipeline error: Color state [0] is invalid
[2024-12-15T16:26:42Z WARN  re_renderer::error_handling::wgpu_core_error] unknown error cause cause=InvalidFormatUsages(TextureUsages(RENDER_ATTACHMENT), Rgba32Float, true)
[2024-12-15T16:26:42Z ERROR re_renderer::error_handling::error_tracker] Wgpu validation error 426: Validation Error
    
    Caused by:
      In Device::create_texture, label = 'DepthCopyWorkaround::readable_texture'
        Texture usages TextureUsages(RENDER_ATTACHMENT) are not allowed on a texture of type Rgba32Float due to downlevel restrictions

Desktop:

 OS: Arch Linux 
 Kernel: x86_64 Linux 6.12.3-arch1-1
 DE: KDE
 WM: KWin
 Icon Theme: breeze
 CPU: 12th Gen Intel Core i7-12700H @ 20x 4.6GHz [50.0°C]
 GPU: Mesa Intel(R) Iris(R) Xe Graphics (ADL GT2)

Rerun version

rerun-cli 0.20.3 (default map_view native_viewer web_viewer) [rustc 1.83.0 (90b35a623 2024-11-26), LLVM 19.1.1] x86_64-unknown-linux-gnu
Video features: av1 default ffmpeg serde

[DobromirM]

Building from source and reading through the logs shows that wgpu is setting the adapter backend to Gl.

[2024-12-15T16:26:39Z WARN  wgpu_hal::vulkan::instance] Unable to find extension: VK_KHR_surface
[2024-12-15T16:26:39Z WARN  wgpu_hal::vulkan::instance] Unable to find extension: VK_KHR_xlib_surface
[2024-12-15T16:26:39Z WARN  wgpu_hal::vulkan::instance] Unable to find extension: VK_KHR_xcb_surface
[2024-12-15T16:26:39Z WARN  wgpu_hal::vulkan::instance] Unable to find extension: VK_KHR_wayland_surface
[2024-12-15T16:26:39Z WARN  wgpu_hal::vulkan::instance] Unable to find extension: VK_EXT_swapchain_colorspace
[2024-12-15T16:26:39Z WARN  wgpu_hal::vulkan::instance] Unable to find extension: VK_KHR_get_physical_device_properties2
[2024-12-15T16:26:39Z WARN  wgpu_hal::vulkan::instance] InstanceFlags::VALIDATION requested, but unable to find layer: VK_LAYER_KHRONOS_validation
[2024-12-15T16:26:39Z ERROR wgpu_hal::vulkan::instance] GENERAL [Loader Message (0x0)]
        vkCreateInstance: Found no drivers!
[2024-12-15T16:26:39Z ERROR wgpu_hal::vulkan::instance]         objects: (type: INSTANCE, hndl: 0x56ed7ba3cc60, name: ?)
[2024-12-15T16:26:39Z WARN  wgpu_hal::gles::egl] Re-initializing Gles context due to Wayland window
[2024-12-15T16:26:39Z WARN  wgpu_hal::gles::adapter] Detected skylake derivative running on mesa i915. Clears to srgb textures will use manual shader clears.
[2024-12-15T16:26:39Z WARN  wgpu_hal::gles::adapter] Detected skylake derivative running on mesa i915. Clears to srgb textures will use manual shader clears.
[2024-12-15T16:26:39Z DEBUG egui_wgpu] Picked the only available wgpu adapter: backend: Gl, device_type: IntegratedGpu, name: "Mesa Intel(R) Iris(R) Xe Graphics (ADL GT2)", driver_info: "4.6 (Core Profile) Mesa 24.3.1-arch1.3", vendor: 0x8086
[2024-12-15T16:26:39Z DEBUG re_renderer::context] wgpu adapter backend: Gl, device_type: IntegratedGpu, name: "Mesa Intel(R) Iris(R) Xe Graphics (ADL GT2)", driver_info: "4.6 (Core Profile) Mesa 24.3.1-arch1.3"

I found that disabling the DepthReadbackWorkaround in picking_layer.rs, by setting it to None, fixes the issue for that particular case.

[DobromirM]

Alternatively, after updating both vulkan-intel and vulkan-icd-loader on my system, wgpu sets the backend to Vulkan and the problem is not encountered.

[Wumpf]

after updating both vulkan-intel and vulkan-icd-loader on my system, wgpu sets the backend to Vulkan and the problem is not encountered

The error we're getting is literally means that the driver/gpu was not capable enough to do a very basic rendering operation which we need to do for picking detection. Very happy to hear that updating the drivers resolved this and it picks now Vulkan - OpenGL is really just a fallback for us that we'd like to phase out eventually.

I figure we should not start the application to begin with - keeping the issue open and changing the title accordingly. I don't think there's more we can do here.