Native unclipped depth on supported platforms (#16095)

# Objective
- Fixes #16078

## Solution

- Rename things to clarify that we _want_ unclipped depth for
directional light shadow views, and need some way of disabling the GPU's
builtin depth clipping
- Use DEPTH_CLIP_CONTROL instead of the fragment shader emulation on
supported platforms
- Pass only the clip position depth instead of the whole clip position
between vertex->fragment shader (no idea if this helps performance or
not, compiler might optimize it anyways)
- Meshlets
- HW raster always uses DEPTH_CLIP_CONTROL since it targets a more
limited set of platforms
- SW raster was not handling DEPTH_CLAMP_ORTHO correctly, it ended up
pretty much doing nothing.
- This PR made me realize that SW raster technically should have depth
clipping for all views that are not directional light shadows, but I
decided not to bother writing it. I'm not sure that it ever matters in
practice. If proven otherwise, I can add it.

## Testing

- Did you test these changes? If so, how?
- Lighting example. Both opaque (no fragment shader) and alpha masked
geometry (fragment shader emulation) are working with
depth_clip_control, and both work when it's turned off. Also tested
meshlet example.
- Are there any parts that need more testing?
  - Performance. I can't figure out a good test scene.
- How can other people (reviewers) test your changes? Is there anything
specific they need to know?
- Toggle depth_clip_control_supported in prepass/mod.rs line 323 to turn
this PR on or off.
- If relevant, what platforms did you test these changes on, and are
there any important ones you can't test?
  - Native

---

## Migration Guide
- `MeshPipelineKey::DEPTH_CLAMP_ORTHO` is now
`MeshPipelineKey::UNCLIPPED_DEPTH_ORTHO`
- The `DEPTH_CLAMP_ORTHO` shaderdef has been renamed to
`UNCLIPPED_DEPTH_ORTHO_EMULATION`
- `clip_position_unclamped: vec4<f32>` is now `unclipped_depth: f32`

crates/bevy_pbr/src/meshlet/mod.rs

@@ -142,6 +142,7 @@ impl MeshletPlugin {
         WgpuFeatures::SHADER_INT64_ATOMIC_MIN_MAX
             | WgpuFeatures::SHADER_INT64
             | WgpuFeatures::SUBGROUP
+            | WgpuFeatures::DEPTH_CLIP_CONTROL
             | WgpuFeatures::PUSH_CONSTANTS
     }
 }

crates/bevy_pbr/src/meshlet/pipelines.rs

@@ -34,10 +34,9 @@ pub struct MeshletPipelines {
     downsample_depth_second_shadow_view: CachedComputePipelineId,
     visibility_buffer_software_raster: CachedComputePipelineId,
     visibility_buffer_software_raster_depth_only: CachedComputePipelineId,
-    visibility_buffer_software_raster_depth_only_clamp_ortho: CachedComputePipelineId,
     visibility_buffer_hardware_raster: CachedRenderPipelineId,
     visibility_buffer_hardware_raster_depth_only: CachedRenderPipelineId,
-    visibility_buffer_hardware_raster_depth_only_clamp_ortho: CachedRenderPipelineId,
+    visibility_buffer_hardware_raster_depth_only_unclipped: CachedRenderPipelineId,
     resolve_depth: CachedRenderPipelineId,
     resolve_depth_shadow_view: CachedRenderPipelineId,
     resolve_material_depth: CachedRenderPipelineId,
@@ -215,29 +214,6 @@ impl FromWorld for MeshletPipelines {
                 },
             ),
 
-            visibility_buffer_software_raster_depth_only_clamp_ortho: pipeline_cache
-                .queue_compute_pipeline(ComputePipelineDescriptor {
-                    label: Some(
-                        "meshlet_visibility_buffer_software_raster_depth_only_clamp_ortho_pipeline"
-                            .into(),
-                    ),
-                    layout: vec![visibility_buffer_raster_layout.clone()],
-                    push_constant_ranges: vec![],
-                    shader: MESHLET_VISIBILITY_BUFFER_SOFTWARE_RASTER_SHADER_HANDLE,
-                    shader_defs: vec![
-                        "MESHLET_VISIBILITY_BUFFER_RASTER_PASS".into(),
-                        "DEPTH_CLAMP_ORTHO".into(),
-                        if remap_1d_to_2d_dispatch_layout.is_some() {
-                            "MESHLET_2D_DISPATCH"
-                        } else {
-                            ""
-                        }
-                        .into(),
-                    ],
-                    entry_point: "rasterize_cluster".into(),
-                    zero_initialize_workgroup_memory: false,
-                }),
-
             visibility_buffer_hardware_raster: pipeline_cache.queue_render_pipeline(
                 RenderPipelineDescriptor {
                     label: Some("meshlet_visibility_buffer_hardware_raster_pipeline".into()),
@@ -324,10 +300,10 @@ impl FromWorld for MeshletPipelines {
                 },
             ),
 
-            visibility_buffer_hardware_raster_depth_only_clamp_ortho: pipeline_cache
+            visibility_buffer_hardware_raster_depth_only_unclipped: pipeline_cache
                 .queue_render_pipeline(RenderPipelineDescriptor {
                     label: Some(
-                        "meshlet_visibility_buffer_hardware_raster_depth_only_clamp_ortho_pipeline"
+                        "meshlet_visibility_buffer_hardware_raster_depth_only_unclipped_pipeline"
                             .into(),
                     ),
                     layout: vec![visibility_buffer_raster_layout],
@@ -337,10 +313,7 @@ impl FromWorld for MeshletPipelines {
                     }],
                     vertex: VertexState {
                         shader: MESHLET_VISIBILITY_BUFFER_HARDWARE_RASTER_SHADER_HANDLE,
-                        shader_defs: vec![
+                        shader_defs: vec!["MESHLET_VISIBILITY_BUFFER_RASTER_PASS".into()],
-                            "MESHLET_VISIBILITY_BUFFER_RASTER_PASS".into(),
-                            "DEPTH_CLAMP_ORTHO".into(),
-                        ],
                         entry_point: "vertex".into(),
                         buffers: vec![],
                     },
@@ -349,7 +322,7 @@ impl FromWorld for MeshletPipelines {
                         strip_index_format: None,
                         front_face: FrontFace::Ccw,
                         cull_mode: Some(Face::Back),
-                        unclipped_depth: false,
+                        unclipped_depth: true,
                         polygon_mode: PolygonMode::Fill,
                         conservative: false,
                     },
@@ -357,10 +330,7 @@ impl FromWorld for MeshletPipelines {
                     multisample: MultisampleState::default(),
                     fragment: Some(FragmentState {
                         shader: MESHLET_VISIBILITY_BUFFER_HARDWARE_RASTER_SHADER_HANDLE,
-                        shader_defs: vec![
+                        shader_defs: vec!["MESHLET_VISIBILITY_BUFFER_RASTER_PASS".into()],
-                            "MESHLET_VISIBILITY_BUFFER_RASTER_PASS".into(),
-                            "DEPTH_CLAMP_ORTHO".into(),
-                        ],
                         entry_point: "fragment".into(),
                         targets: vec![Some(ColorTargetState {
                             format: TextureFormat::R8Uint,
@@ -484,7 +454,6 @@ impl MeshletPipelines {
         &ComputePipeline,
         &ComputePipeline,
         &ComputePipeline,
-        &ComputePipeline,
         &RenderPipeline,
         &RenderPipeline,
         &RenderPipeline,
@@ -506,14 +475,11 @@ impl MeshletPipelines {
             pipeline_cache.get_compute_pipeline(pipeline.visibility_buffer_software_raster)?,
             pipeline_cache
                 .get_compute_pipeline(pipeline.visibility_buffer_software_raster_depth_only)?,
-            pipeline_cache.get_compute_pipeline(
-                pipeline.visibility_buffer_software_raster_depth_only_clamp_ortho,
-            )?,
             pipeline_cache.get_render_pipeline(pipeline.visibility_buffer_hardware_raster)?,
             pipeline_cache
                 .get_render_pipeline(pipeline.visibility_buffer_hardware_raster_depth_only)?,
             pipeline_cache.get_render_pipeline(
-                pipeline.visibility_buffer_hardware_raster_depth_only_clamp_ortho,
+                pipeline.visibility_buffer_hardware_raster_depth_only_unclipped,
             )?,
             pipeline_cache.get_render_pipeline(pipeline.resolve_depth)?,
             pipeline_cache.get_render_pipeline(pipeline.resolve_depth_shadow_view)?,

crates/bevy_pbr/src/meshlet/visibility_buffer_hardware_raster.wgsl

@@ -23,9 +23,6 @@ struct VertexOutput {
 #ifdef MESHLET_VISIBILITY_BUFFER_RASTER_PASS_OUTPUT
     @location(0) @interpolate(flat) packed_ids: u32,
 #endif
-#ifdef DEPTH_CLAMP_ORTHO
-    @location(0) unclamped_clip_depth: f32,
-#endif
 }
 
 @vertex
@@ -45,19 +42,12 @@ fn vertex(@builtin(instance_index) instance_index: u32, @builtin(vertex_index) v
     let vertex_position = get_meshlet_vertex_position(&meshlet, vertex_id);
     let world_from_local = affine3_to_square(instance_uniform.world_from_local);
     let world_position = mesh_position_local_to_world(world_from_local, vec4(vertex_position, 1.0));
-    var clip_position = view.clip_from_world * vec4(world_position.xyz, 1.0);
+    let clip_position = view.clip_from_world * vec4(world_position.xyz, 1.0);
-#ifdef DEPTH_CLAMP_ORTHO
-    let unclamped_clip_depth = clip_position.z;
-    clip_position.z = min(clip_position.z, 1.0);
-#endif
 
     return VertexOutput(
         clip_position,
 #ifdef MESHLET_VISIBILITY_BUFFER_RASTER_PASS_OUTPUT
         (cluster_id << 7u) | triangle_id,
-#endif
-#ifdef DEPTH_CLAMP_ORTHO
-        unclamped_clip_depth,
 #endif
     );
 }
@@ -70,9 +60,6 @@ fn fragment(vertex_output: VertexOutput) {
     let depth = bitcast<u32>(vertex_output.position.z);
     let visibility = (u64(depth) << 32u) | u64(vertex_output.packed_ids);
     atomicMax(&meshlet_visibility_buffer[frag_coord_1d], visibility);
-#else ifdef DEPTH_CLAMP_ORTHO
-    let depth = bitcast<u32>(vertex_output.unclamped_clip_depth);
-    atomicMax(&meshlet_visibility_buffer[frag_coord_1d], depth);
 #else
     let depth = bitcast<u32>(vertex_output.position.z);
     atomicMax(&meshlet_visibility_buffer[frag_coord_1d], depth);
@@ -84,9 +71,6 @@ fn dummy_vertex() -> VertexOutput {
         vec4(divide(0.0, 0.0)), // NaN vertex position
 #ifdef MESHLET_VISIBILITY_BUFFER_RASTER_PASS_OUTPUT
         0u,
-#endif
-#ifdef DEPTH_CLAMP_ORTHO
-        0.0,
 #endif
     );
 }

crates/bevy_pbr/src/meshlet/visibility_buffer_raster_node.rs

@@ -85,10 +85,9 @@ impl Node for MeshletVisibilityBufferRasterPassNode {
             downsample_depth_second_shadow_view_pipeline,
             visibility_buffer_software_raster_pipeline,
             visibility_buffer_software_raster_depth_only_pipeline,
-            visibility_buffer_software_raster_depth_only_clamp_ortho,
             visibility_buffer_hardware_raster_pipeline,
             visibility_buffer_hardware_raster_depth_only_pipeline,
-            visibility_buffer_hardware_raster_depth_only_clamp_ortho,
+            visibility_buffer_hardware_raster_depth_only_unclipped_pipeline,
             resolve_depth_pipeline,
             resolve_depth_shadow_view_pipeline,
             resolve_material_depth_pipeline,
@@ -223,18 +222,11 @@ impl Node for MeshletVisibilityBufferRasterPassNode {
                 continue;
             };
 
-            let (
+            let shadow_visibility_buffer_hardware_raster_pipeline =
-                shadow_visibility_buffer_software_raster_pipeline,
+                if let LightEntity::Directional { .. } = light_type {
-                shadow_visibility_buffer_hardware_raster_pipeline,
+                    visibility_buffer_hardware_raster_depth_only_unclipped_pipeline
-            ) = match light_type {
+                } else {
-                LightEntity::Directional { .. } => (
+                    visibility_buffer_hardware_raster_depth_only_pipeline
-                    visibility_buffer_software_raster_depth_only_clamp_ortho,
-                    visibility_buffer_hardware_raster_depth_only_clamp_ortho,
-                ),
-                _ => (
-                    visibility_buffer_software_raster_depth_only_pipeline,
-                    visibility_buffer_hardware_raster_depth_only_pipeline,
-                ),
                 };
 
             render_context.command_encoder().push_debug_group(&format!(
@@ -270,7 +262,7 @@ impl Node for MeshletVisibilityBufferRasterPassNode {
                 &meshlet_view_resources.dummy_render_target.default_view,
                 meshlet_view_bind_groups,
                 view_offset,
-                shadow_visibility_buffer_software_raster_pipeline,
+                visibility_buffer_software_raster_depth_only_pipeline,
                 shadow_visibility_buffer_hardware_raster_pipeline,
                 None,
                 meshlet_view_resources.raster_cluster_rightmost_slot,
@@ -306,7 +298,7 @@ impl Node for MeshletVisibilityBufferRasterPassNode {
                 &meshlet_view_resources.dummy_render_target.default_view,
                 meshlet_view_bind_groups,
                 view_offset,
-                shadow_visibility_buffer_software_raster_pipeline,
+                visibility_buffer_software_raster_depth_only_pipeline,
                 shadow_visibility_buffer_hardware_raster_pipeline,
                 None,
                 meshlet_view_resources.raster_cluster_rightmost_slot,

crates/bevy_pbr/src/meshlet/visibility_buffer_software_raster.wgsl

@@ -57,10 +57,7 @@ fn rasterize_cluster(
             // Project vertex to viewport space
             let world_position = mesh_position_local_to_world(world_from_local, vec4(vertex_position, 1.0));
             let clip_position = view.clip_from_world * vec4(world_position.xyz, 1.0);
-            var ndc_position = clip_position.xyz / clip_position.w;
+            let ndc_position = clip_position.xyz / clip_position.w;
-#ifdef DEPTH_CLAMP_ORTHO
-            ndc_position.z = 1.0 / clip_position.z;
-#endif
             let viewport_position_xy = ndc_to_uv(ndc_position.xy) * view.viewport.zw;
 
             // Write vertex to workgroup shared memory
@@ -176,9 +173,6 @@ fn write_visibility_buffer_pixel(x: f32, y: f32, z: f32, packed_ids: u32) {
     let depth = bitcast<u32>(z);
     let visibility = (u64(depth) << 32u) | u64(packed_ids);
     atomicMax(&meshlet_visibility_buffer[frag_coord_1d], visibility);
-#else ifdef DEPTH_CLAMP_ORTHO
-    let depth = bitcast<u32>(1.0 / z);
-    atomicMax(&meshlet_visibility_buffer[frag_coord_1d], depth);
 #else
     let depth = bitcast<u32>(z);
     atomicMax(&meshlet_visibility_buffer[frag_coord_1d], depth);

crates/bevy_pbr/src/prepass/mod.rs

@@ -251,6 +251,7 @@ pub struct PrepassPipeline<M: Material> {
     pub deferred_material_vertex_shader: Option<Handle<Shader>>,
     pub deferred_material_fragment_shader: Option<Handle<Shader>>,
     pub material_pipeline: MaterialPipeline<M>,
+    pub depth_clip_control_supported: bool,
     _marker: PhantomData<M>,
 }
 
@@ -289,6 +290,10 @@ impl<M: Material> FromWorld for PrepassPipeline<M> {
 
         let mesh_pipeline = world.resource::<MeshPipeline>();
 
+        let depth_clip_control_supported = render_device
+            .features()
+            .contains(WgpuFeatures::DEPTH_CLIP_CONTROL);
+
         PrepassPipeline {
             view_layout_motion_vectors,
             view_layout_no_motion_vectors,
@@ -315,6 +320,7 @@ impl<M: Material> FromWorld for PrepassPipeline<M> {
             },
             material_layout: M::bind_group_layout(render_device),
             material_pipeline: world.resource::<MaterialPipeline<M>>().clone(),
+            depth_clip_control_supported,
             _marker: PhantomData,
         }
     }
@@ -379,8 +385,14 @@ where
             vertex_attributes.push(Mesh::ATTRIBUTE_POSITION.at_shader_location(0));
         }
 
-        if key.mesh_key.contains(MeshPipelineKey::DEPTH_CLAMP_ORTHO) {
+        // For directional light shadow map views, use unclipped depth via either the native GPU feature,
-            shader_defs.push("DEPTH_CLAMP_ORTHO".into());
+        // or emulated by setting depth in the fragment shader for GPUs that don't support it natively.
+        let emulate_unclipped_depth = key
+            .mesh_key
+            .contains(MeshPipelineKey::UNCLIPPED_DEPTH_ORTHO)
+            && !self.depth_clip_control_supported;
+        if emulate_unclipped_depth {
+            shader_defs.push("UNCLIPPED_DEPTH_ORTHO_EMULATION".into());
             // PERF: This line forces the "prepass fragment shader" to always run in
             // common scenarios like "directional light calculation". Doing so resolves
             // a pretty nasty depth clamping bug, but it also feels a bit excessive.
@@ -389,6 +401,10 @@ where
             // https://github.com/bevyengine/bevy/pull/8877
             shader_defs.push("PREPASS_FRAGMENT".into());
         }
+        let unclipped_depth = key
+            .mesh_key
+            .contains(MeshPipelineKey::UNCLIPPED_DEPTH_ORTHO)
+            && self.depth_clip_control_supported;
 
         if layout.0.contains(Mesh::ATTRIBUTE_UV_0) {
             shader_defs.push("VERTEX_UVS".into());
@@ -488,10 +504,10 @@ where
         }
 
         // The fragment shader is only used when the normal prepass or motion vectors prepass
-        // is enabled or the material uses alpha cutoff values and doesn't rely on the standard
+        // is enabled, the material uses alpha cutoff values and doesn't rely on the standard
-        // prepass shader or we are clamping the orthographic depth.
+        // prepass shader, or we are emulating unclipped depth in the fragment shader.
         let fragment_required = !targets.is_empty()
-            || key.mesh_key.contains(MeshPipelineKey::DEPTH_CLAMP_ORTHO)
+            || emulate_unclipped_depth
             || (key.mesh_key.contains(MeshPipelineKey::MAY_DISCARD)
                 && self.prepass_material_fragment_shader.is_some());
 
@@ -544,7 +560,7 @@ where
                 strip_index_format: None,
                 front_face: FrontFace::Ccw,
                 cull_mode: None,
-                unclipped_depth: false,
+                unclipped_depth,
                 polygon_mode: PolygonMode::Fill,
                 conservative: false,
             },

crates/bevy_pbr/src/prepass/prepass.wgsl

@@ -76,10 +76,10 @@ fn vertex(vertex_no_morph: Vertex) -> VertexOutput {
 
     out.world_position = mesh_functions::mesh_position_local_to_world(world_from_local, vec4<f32>(vertex.position, 1.0));
     out.position = position_world_to_clip(out.world_position.xyz);
-#ifdef DEPTH_CLAMP_ORTHO
+#ifdef UNCLIPPED_DEPTH_ORTHO_EMULATION
-    out.clip_position_unclamped = out.position;
+    out.unclipped_depth = out.position.z;
-    out.position.z = min(out.position.z, 1.0);
+    out.position.z = min(out.position.z, 1.0); // Clamp depth to avoid clipping
-#endif // DEPTH_CLAMP_ORTHO
+#endif // UNCLIPPED_DEPTH_ORTHO_EMULATION
 
 #ifdef VERTEX_UVS_A
     out.uv = vertex.uv;
@@ -173,9 +173,9 @@ fn fragment(in: VertexOutput) -> FragmentOutput {
     out.normal = vec4(in.world_normal * 0.5 + vec3(0.5), 1.0);
 #endif
 
-#ifdef DEPTH_CLAMP_ORTHO
+#ifdef UNCLIPPED_DEPTH_ORTHO_EMULATION
-    out.frag_depth = in.clip_position_unclamped.z;
+    out.frag_depth = in.unclipped_depth;
-#endif // DEPTH_CLAMP_ORTHO
+#endif // UNCLIPPED_DEPTH_ORTHO_EMULATION
 
 #ifdef MOTION_VECTOR_PREPASS
     let clip_position_t = view.unjittered_clip_from_world * in.world_position;

crates/bevy_pbr/src/prepass/prepass_io.wgsl

@@ -60,9 +60,9 @@ struct VertexOutput {
     @location(5) previous_world_position: vec4<f32>,
 #endif
 
-#ifdef DEPTH_CLAMP_ORTHO
+#ifdef UNCLIPPED_DEPTH_ORTHO_EMULATION
-    @location(6) clip_position_unclamped: vec4<f32>,
+    @location(6) unclipped_depth: f32,
-#endif // DEPTH_CLAMP_ORTHO
+#endif // UNCLIPPED_DEPTH_ORTHO_EMULATION
 #ifdef VERTEX_OUTPUT_INSTANCE_INDEX
     @location(7) instance_index: u32,
 #endif
@@ -87,8 +87,8 @@ struct FragmentOutput {
     @location(3) deferred_lighting_pass_id: u32,
 #endif
 
-#ifdef DEPTH_CLAMP_ORTHO
+#ifdef UNCLIPPED_DEPTH_ORTHO_EMULATION
     @builtin(frag_depth) frag_depth: f32,
-#endif // DEPTH_CLAMP_ORTHO
+#endif // UNCLIPPED_DEPTH_ORTHO_EMULATION
 }
 #endif //PREPASS_FRAGMENT

crates/bevy_pbr/src/render/light.rs

@@ -1554,7 +1554,7 @@ pub fn queue_shadows<M: Material>(
                     .expect("Failed to get spot light visible entities"),
             };
             let mut light_key = MeshPipelineKey::DEPTH_PREPASS;
-            light_key.set(MeshPipelineKey::DEPTH_CLAMP_ORTHO, is_directional_light);
+            light_key.set(MeshPipelineKey::UNCLIPPED_DEPTH_ORTHO, is_directional_light);
 
             // NOTE: Lights with shadow mapping disabled will have no visible entities
             // so no meshes will be queued

crates/bevy_pbr/src/render/mesh.rs

@@ -1489,7 +1489,9 @@ bitflags::bitflags! {
                                                             // See: https://www.khronos.org/opengl/wiki/Early_Fragment_Test
         const ENVIRONMENT_MAP                   = 1 << 8;
         const SCREEN_SPACE_AMBIENT_OCCLUSION    = 1 << 9;
-        const DEPTH_CLAMP_ORTHO                 = 1 << 10;
+        const UNCLIPPED_DEPTH_ORTHO             = 1 << 10; // Disables depth clipping for use with directional light shadow views
+                                                            // Emulated via fragment shader depth on hardware that doesn't support it natively
+                                                            // See: https://www.w3.org/TR/webgpu/#depth-clipping and https://therealmjp.github.io/posts/shadow-maps/#disabling-z-clipping
         const TEMPORAL_JITTER                   = 1 << 11;
         const READS_VIEW_TRANSMISSION_TEXTURE   = 1 << 12;
         const LIGHTMAPPED                       = 1 << 13;

crates/bevy_pbr/src/render/pbr_prepass.wgsl

@@ -32,9 +32,9 @@ fn fragment(
 
     var out: prepass_io::FragmentOutput;
 
-#ifdef DEPTH_CLAMP_ORTHO
+#ifdef UNCLIPPED_DEPTH_ORTHO_EMULATION
-    out.frag_depth = in.clip_position_unclamped.z;
+    out.frag_depth = in.unclipped_depth;
-#endif // DEPTH_CLAMP_ORTHO
+#endif // UNCLIPPED_DEPTH_ORTHO_EMULATION
 
 #ifdef NORMAL_PREPASS
     // NOTE: Unlit bit not set means == 0 is true, so the true case is if lit