Thank you for filing this! Doing an actual investigation into the backend support would be the next step.

Multi-View Rendering is supposedly supported by DX12, DX11, Vulkan and OpenGL, as written on the related NVidia page. They also have a nice article about the feature.

Resources on the feature:

• DX11: I could not find any.
• DX12: View Instancing
• Vulkan: VK_KHR_multiview + SPIR-V's SPV_KHR_multiview
• OpenGL: GL_OVR_multiview, GL_OVR_multiview2, GL_EXT_multiview_timer_query, GL_EXT_multiview_texture_multisample, GL_EXT_multiview_tessellation_geometry_shader
• OpenGL ES: Above OpenGL extensions + GL_EXT_multiview_draw_buffers, GL_OVR_multiview_multisampled_render_to_texture
• WebGL: OVR_multiview2

Thank you! The next would be adding a feature to gfx-hal Features, something like:

const MULTIVIEW = 0x2000_0000_0000_0000;

If you are interested in the full thing, may also add these:

const MULTIVIEW_GEOMETRY_SHADER = 0x4000_0000_0000_0000;
const MULTIVIEW_TESSELLATION_SHADER = 0x8000_0000_0000_0000;

This is sorta optional as WebGPU isn't going to be using those anyway. It's up to you to limit the scope here, we can always follow-up and add those too.

Then basically extend the Rust API in the way it's done by VK_KHR_multiview extension:

• adding the limits
• add view_mask: u32 field to SubpassDesc
• add view_offset: u32 field to SubpassDependency
• find some way to provide the correlation masks, maybe just as an argument to create_render_pass

Finally, implement the logic on all backends, or at least update them to compile (not exposing the feature yet).