New batching mechnaism and sprite optimizations

Author: Davier

crates/bevy_core_pipeline/src/lib.rs

@@ -15,6 +15,8 @@ pub use main_pass_2d::*;
 pub use main_pass_3d::*;
 pub use main_pass_driver::*;
 
+use std::ops::Range;
+
 use bevy_app::{App, Plugin};
 use bevy_core::FloatOrd;
 use bevy_ecs::prelude::*;
@@ -23,8 +25,8 @@ use bevy_render::{
     color::Color,
     render_graph::{EmptyNode, RenderGraph, SlotInfo, SlotType},
     render_phase::{
-        sort_phase_system, CachedPipelinePhaseItem, DrawFunctionId, DrawFunctions, EntityPhaseItem,
-        PhaseItem, RenderPhase,
+        batch_phase_system, sort_phase_system, BatchedPhaseItem, CachedPipelinePhaseItem,
+        DrawFunctionId, DrawFunctions, EntityPhaseItem, PhaseItem, RenderPhase,
     },
     render_resource::*,
     renderer::RenderDevice,
@@ -84,6 +86,11 @@ pub mod clear_graph {
 #[derive(Default)]
 pub struct CorePipelinePlugin;
 
+#[derive(Debug, Hash, PartialEq, Eq, Clone, SystemLabel)]
+pub enum CorePipelineRenderSystems {
+    SortTransparent2d,
+}
+
 impl Plugin for CorePipelinePlugin {
     fn build(&self, app: &mut App) {
         app.init_resource::<ClearColor>();
@@ -97,7 +104,16 @@ impl Plugin for CorePipelinePlugin {
             .add_system_to_stage(RenderStage::Extract, extract_clear_color)
             .add_system_to_stage(RenderStage::Extract, extract_core_pipeline_camera_phases)
             .add_system_to_stage(RenderStage::Prepare, prepare_core_views_system)
-            .add_system_to_stage(RenderStage::PhaseSort, sort_phase_system::<Transparent2d>)
+            .add_system_to_stage(
+                RenderStage::PhaseSort,
+                sort_phase_system::<Transparent2d>
+                    .label(CorePipelineRenderSystems::SortTransparent2d),
+            )
+            .add_system_to_stage(
+                RenderStage::PhaseSort,
+                batch_phase_system::<Transparent2d>
+                    .after(CorePipelineRenderSystems::SortTransparent2d),
+            )
             .add_system_to_stage(RenderStage::PhaseSort, sort_phase_system::<Opaque3d>)
             .add_system_to_stage(RenderStage::PhaseSort, sort_phase_system::<AlphaMask3d>)
             .add_system_to_stage(RenderStage::PhaseSort, sort_phase_system::<Transparent3d>);
@@ -160,6 +176,8 @@ pub struct Transparent2d {
     pub entity: Entity,
     pub pipeline: CachedPipelineId,
     pub draw_function: DrawFunctionId,
+    /// Range in the index buffer of this item
+    pub batch_range: Option<Range<u32>>,
 }
 
 impl PhaseItem for Transparent2d {
@@ -190,6 +208,16 @@ impl CachedPipelinePhaseItem for Transparent2d {
     }
 }
 
+impl BatchedPhaseItem for Transparent2d {
+    fn batch_range(&self) -> &Option<Range<u32>> {
+        &self.batch_range
+    }
+
+    fn batch_range_mut(&mut self) -> &mut Option<Range<u32>> {
+        &mut self.batch_range
+    }
+}
+
 pub struct Opaque3d {
     pub distance: f32,
     pub pipeline: CachedPipelineId,

crates/bevy_pbr/src/render/mesh.rs

@@ -8,7 +8,7 @@ use bevy_ecs::{
     prelude::*,
     system::{lifetimeless::*, SystemParamItem},
 };
-use bevy_math::Mat4;
+use bevy_math::{Mat4, Size};
 use bevy_reflect::TypeUuid;
 use bevy_render::{
     mesh::{GpuBufferInfo, Mesh},
@@ -328,6 +328,10 @@ impl FromWorld for MeshPipeline {
                 texture,
                 texture_view,
                 sampler,
+                size: Size::new(
+                    image.texture_descriptor.size.width as f32,
+                    image.texture_descriptor.size.height as f32,
+                ),
             }
         };
         MeshPipeline {

crates/bevy_render/src/render_phase/draw.rs

@@ -13,7 +13,7 @@ use bevy_ecs::{
 };
 use bevy_utils::HashMap;
 use parking_lot::{RwLock, RwLockReadGuard, RwLockWriteGuard};
-use std::{any::TypeId, fmt::Debug, hash::Hash};
+use std::{any::TypeId, fmt::Debug, hash::Hash, ops::Range};
 
 /// A draw function which is used to draw a specific [`PhaseItem`].
 ///
@@ -166,6 +166,40 @@ pub trait CachedPipelinePhaseItem: PhaseItem {
     fn cached_pipeline(&self) -> CachedPipelineId;
 }
 
+pub trait BatchedPhaseItem: EntityPhaseItem {
+    /// Range in the index buffer of this item
+    fn batch_range(&self) -> &Option<Range<u32>>;
+
+    /// Range in the index buffer of this item
+    fn batch_range_mut(&mut self) -> &mut Option<Range<u32>>;
+
+    /// Batches another item within this item if they are compatible.
+    /// Items can be batched together if they have the same entity, and consecutive ranges.
+    /// If batching is successful, the `other` item should be discarded from the render pass.
+    #[inline]
+    fn add_to_batch(&mut self, other: &Self) -> BatchResult {
+        let self_entity = self.entity();
+        if let (Some(self_batch_range), Some(other_batch_range)) = (
+            self.batch_range_mut().as_mut(),
+            other.batch_range().as_ref(),
+        ) {
+            if self_entity == other.entity() && self_batch_range.end == other_batch_range.start {
+                // If the items are compatible, join their range into `self`
+                self_batch_range.end = other_batch_range.end;
+                return BatchResult::Success;
+            }
+        }
+        BatchResult::IncompatibleItems
+    }
+}
+
+pub enum BatchResult {
+    /// The `other` item was batched into `self`
+    Success,
+    /// `self` and `other` cannot be batched together
+    IncompatibleItems,
+}
+
 impl<P: EntityPhaseItem, E: EntityRenderCommand> RenderCommand<P> for E {
     type Param = E::Param;
 

crates/bevy_render/src/render_phase/mod.rs

@@ -31,9 +31,44 @@ impl<I: PhaseItem> RenderPhase<I> {
     }
 }
 
+impl<I: BatchedPhaseItem> RenderPhase<I> {
+    /// Batches the compatible [`BatchedPhaseItem`]s of this render phase
+    pub fn batch(&mut self) {
+        // TODO: this could be done in-place
+        let mut items = std::mem::take(&mut self.items);
+        let mut items = items.drain(..);
+
+        self.items.reserve(items.len());
+
+        // Start the first batch from the first item
+        if let Some(mut current_batch) = items.next() {
+            // Batch following items until we find an incompatible item
+            for next_item in items {
+                if matches!(
+                    current_batch.add_to_batch(&next_item),
+                    BatchResult::IncompatibleItems
+                ) {
+                    // Store the completed batch, and start a new one from the incompatible item
+                    self.items.push(current_batch);
+                    current_batch = next_item;
+                }
+            }
+            // Store the last batch
+            self.items.push(current_batch);
+        }
+    }
+}
+
 /// This system sorts all [`RenderPhases`](RenderPhase) for the [`PhaseItem`] type.
 pub fn sort_phase_system<I: PhaseItem>(mut render_phases: Query<&mut RenderPhase<I>>) {
     for mut phase in render_phases.iter_mut() {
         phase.sort();
     }
 }
+
+/// This batches the [`PhaseItem`]s of a [`RenderPhases`](RenderPhase).
+pub fn batch_phase_system<I: BatchedPhaseItem>(mut render_phases: Query<&mut RenderPhase<I>>) {
+    for mut phase in render_phases.iter_mut() {
+        phase.batch();
+    }
+}

crates/bevy_render/src/texture/image.rs

@@ -7,6 +7,7 @@ use crate::{
 };
 use bevy_asset::HandleUntyped;
 use bevy_ecs::system::{lifetimeless::SRes, SystemParamItem};
+use bevy_math::Size;
 use bevy_reflect::TypeUuid;
 use thiserror::Error;
 use wgpu::{
@@ -373,12 +374,13 @@ impl TextureFormatPixelInfo for TextureFormat {
 }
 
 /// The GPU-representation of an [`Image`].
-/// Consists of the [`Texture`], its [`TextureView`] and the corresponding [`Sampler`].
+/// Consists of the [`Texture`], its [`TextureView`] and the corresponding [`Sampler`], and the texture's [`Size`].
 #[derive(Debug, Clone)]
 pub struct GpuImage {
     pub texture: Texture,
     pub texture_view: TextureView,
     pub sampler: Sampler,
+    pub size: Size,
 }
 
 impl RenderAsset for Image {
@@ -426,10 +428,15 @@ impl RenderAsset for Image {
         );
 
         let texture_view = texture.create_view(&TextureViewDescriptor::default());
+        let size = Size::new(
+            image.texture_descriptor.size.width as f32,
+            image.texture_descriptor.size.height as f32,
+        );
         Ok(GpuImage {
             texture,
             texture_view,
             sampler,
+            size,
         })
     }
 }

crates/bevy_render/src/view/visibility/mod.rs

@@ -28,7 +28,7 @@ impl Default for Visibility {
 }
 
 /// Algorithmically-computed indication of whether an entity is visible and should be extracted for rendering
-#[derive(Component, Clone, Reflect)]
+#[derive(Component, Clone, Reflect, Debug)]
 #[reflect(Component)]
 pub struct ComputedVisibility {
     pub is_visible: bool,

crates/bevy_sprite/src/lib.rs

@@ -48,7 +48,6 @@ pub const SPRITE_SHADER_HANDLE: HandleUntyped =
 #[derive(Debug, Hash, PartialEq, Eq, Clone, SystemLabel)]
 pub enum SpriteSystem {
     ExtractSprites,
-    PrepareSprites,
 }
 
 impl Plugin for SpritePlugin {
@@ -66,18 +65,14 @@ impl Plugin for SpritePlugin {
             .init_resource::<SpritePipeline>()
             .init_resource::<SpecializedPipelines<SpritePipeline>>()
             .init_resource::<SpriteMeta>()
-            .init_resource::<Extracted2dItems>()
+            .init_resource::<ExtractedSprites>()
             .init_resource::<SpriteAssetEvents>()
             .add_render_command::<Transparent2d, DrawSprite>()
             .add_system_to_stage(
                 RenderStage::Extract,
                 render::extract_sprites.label(SpriteSystem::ExtractSprites),
             )
             .add_system_to_stage(RenderStage::Extract, render::extract_sprite_events)
-            .add_system_to_stage(
-                RenderStage::Prepare,
-                render::prepare_sprites.label(SpriteSystem::PrepareSprites),
-            )
             .add_system_to_stage(RenderStage::Queue, queue_sprites);
     }
 }

crates/bevy_sprite/src/mesh2d/material.rs

@@ -24,7 +24,7 @@ use bevy_render::{
         SpecializedPipeline, SpecializedPipelines,
     },
     renderer::RenderDevice,
-    view::{ComputedVisibility, ExtractedView, Msaa, Visibility, VisibleEntities},
+    view::{ComputedVisibility, Msaa, Visibility, VisibleEntities},
     RenderApp, RenderStage,
 };
 use bevy_transform::components::{GlobalTransform, Transform};
@@ -265,20 +265,17 @@ pub fn queue_material2d_meshes<M: SpecializedMaterial2d>(
     render_meshes: Res<RenderAssets<Mesh>>,
     render_materials: Res<RenderAssets<M>>,
     material2d_meshes: Query<(&Handle<M>, &Mesh2dHandle, &Mesh2dUniform)>,
-    mut views: Query<(
-        &ExtractedView,
-        &VisibleEntities,
-        &mut RenderPhase<Transparent2d>,
-    )>,
+    mut views: Query<(&VisibleEntities, &mut RenderPhase<Transparent2d>)>,
 ) {
-    for (view, visible_entities, mut transparent_phase) in views.iter_mut() {
+    if material2d_meshes.is_empty() {
+        return;
+    }
+    for (visible_entities, mut transparent_phase) in views.iter_mut() {
         let draw_transparent_pbr = transparent_draw_functions
             .read()
             .get_id::<DrawMaterial2d<M>>()
             .unwrap();
 
-        let inverse_view_matrix = view.transform.compute_matrix().inverse();
-        let inverse_view_row_2 = inverse_view_matrix.row(2);
         let mesh_key = Mesh2dPipelineKey::from_msaa_samples(msaa.samples);
 
         for visible_entity in &visible_entities.entities {
@@ -302,9 +299,7 @@ pub fn queue_material2d_meshes<M: SpecializedMaterial2d>(
                         (mesh2d_key, specialized_key),
                     );
 
-                    // NOTE: row 2 of the inverse view matrix dotted with column 3 of the model matrix
-                    // gives the z component of translation of the mesh in view space
-                    let mesh_z = inverse_view_row_2.dot(mesh2d_uniform.transform.col(3));
+                    let mesh_z = mesh2d_uniform.transform.w_axis.z;
                     transparent_phase.add(Transparent2d {
                         entity: *visible_entity,
                         draw_function: draw_transparent_pbr,
@@ -314,6 +309,8 @@ pub fn queue_material2d_meshes<M: SpecializedMaterial2d>(
                         // -z in front of the camera, the largest distance is -far with values increasing toward the
                         // camera. As such we can just use mesh_z as the distance
                         sort_key: FloatOrd(mesh_z),
+                        // This material is not batched
+                        batch_range: None,
                     });
                 }
             }

crates/bevy_sprite/src/mesh2d/mesh.rs

@@ -4,7 +4,7 @@ use bevy_ecs::{
     prelude::*,
     system::{lifetimeless::*, SystemParamItem},
 };
-use bevy_math::Mat4;
+use bevy_math::{Mat4, Size};
 use bevy_reflect::TypeUuid;
 use bevy_render::{
     mesh::{GpuBufferInfo, Mesh},
@@ -15,12 +15,10 @@ use bevy_render::{
     renderer::{RenderDevice, RenderQueue},
     texture::{BevyDefault, GpuImage, Image, TextureFormatPixelInfo},
     view::{ComputedVisibility, ExtractedView, ViewUniform, ViewUniformOffset, ViewUniforms},
-    RenderApp, RenderStage, RenderWorld,
+    RenderApp, RenderStage,
 };
 use bevy_transform::components::GlobalTransform;
 
-use crate::{Extracted2dItem, Extracted2dItems, SpriteSystem};
-
 #[derive(Default, Clone, Component)]
 pub struct Mesh2dHandle(pub Handle<Mesh>);
 
@@ -63,10 +61,7 @@ impl Plugin for Mesh2dRenderPlugin {
         app.sub_app_mut(RenderApp)
             .init_resource::<Mesh2dPipeline>()
             .init_resource::<SpecializedPipelines<Mesh2dPipeline>>()
-            .add_system_to_stage(
-                RenderStage::Extract,
-                extract_mesh2d.after(SpriteSystem::ExtractSprites),
-            )
+            .add_system_to_stage(RenderStage::Extract, extract_mesh2d)
             .add_system_to_stage(RenderStage::Queue, queue_mesh2d_bind_group)
             .add_system_to_stage(RenderStage::Queue, queue_mesh2d_view_bind_groups);
     }
@@ -90,19 +85,16 @@ bitflags::bitflags! {
 }
 
 pub fn extract_mesh2d(
-    mut render_world: ResMut<RenderWorld>,
     mut commands: Commands,
     mut previous_len: Local<usize>,
     query: Query<(Entity, &ComputedVisibility, &GlobalTransform, &Mesh2dHandle)>,
 ) {
-    let mut extracted_sprites = render_world.get_resource_mut::<Extracted2dItems>().unwrap();
     let mut values = Vec::with_capacity(*previous_len);
     for (entity, computed_visibility, transform, handle) in query.iter() {
         if !computed_visibility.is_visible {
             continue;
         }
         let transform = transform.compute_matrix();
-        let z = transform.col(3).z;
         values.push((
             entity,
             (
@@ -114,9 +106,6 @@ pub fn extract_mesh2d(
                 },
             ),
         ));
-        // Break sprite batches
-        // FIXME: this doesn't account for camera position
-        extracted_sprites.items.push(Extracted2dItem::Other { z });
     }
     *previous_len = values.len();
     commands.insert_or_spawn_batch(values);
@@ -202,6 +191,10 @@ impl FromWorld for Mesh2dPipeline {
                 texture,
                 texture_view,
                 sampler,
+                size: Size::new(
+                    image.texture_descriptor.size.width as f32,
+                    image.texture_descriptor.size.height as f32,
+                ),
             }
         };
         Mesh2dPipeline {