bevy/examples/shader/specialized_mesh_pipeline.rs
Patrick Walton 5adf831b42
Add a bindless mode to AsBindGroup. (#16368)
This patch adds the infrastructure necessary for Bevy to support
*bindless resources*, by adding a new `#[bindless]` attribute to
`AsBindGroup`.

Classically, only a single texture (or sampler, or buffer) can be
attached to each shader binding. This means that switching materials
requires breaking a batch and issuing a new drawcall, even if the mesh
is otherwise identical. This adds significant overhead not only in the
driver but also in `wgpu`, as switching bind groups increases the amount
of validation work that `wgpu` must do.

*Bindless resources* are the typical solution to this problem. Instead
of switching bindings between each texture, the renderer instead
supplies a large *array* of all textures in the scene up front, and the
material contains an index into that array. This pattern is repeated for
buffers and samplers as well. The renderer now no longer needs to switch
binding descriptor sets while drawing the scene.

Unfortunately, as things currently stand, this approach won't quite work
for Bevy. Two aspects of `wgpu` conspire to make this ideal approach
unacceptably slow:

1. In the DX12 backend, all binding arrays (bindless resources) must
have a constant size declared in the shader, and all textures in an
array must be bound to actual textures. Changing the size requires a
recompile.

2. Changing even one texture incurs revalidation of all textures, a
process that takes time that's linear in the total size of the binding
array.

This means that declaring a large array of textures big enough to
encompass the entire scene is presently unacceptably slow. For example,
if you declare 4096 textures, then `wgpu` will have to revalidate all
4096 textures if even a single one changes. This process can take
multiple frames.

To work around this problem, this PR groups bindless resources into
small *slabs* and maintains a free list for each. The size of each slab
for the bindless arrays associated with a material is specified via the
`#[bindless(N)]` attribute. For instance, consider the following
declaration:

```rust
#[derive(AsBindGroup)]
#[bindless(16)]
struct MyMaterial {
    #[buffer(0)]
    color: Vec4,
    #[texture(1)]
    #[sampler(2)]
    diffuse: Handle<Image>,
}
```

The `#[bindless(N)]` attribute specifies that, if bindless arrays are
supported on the current platform, each resource becomes a binding array
of N instances of that resource. So, for `MyMaterial` above, the `color`
attribute is exposed to the shader as `binding_array<vec4<f32>, 16>`,
the `diffuse` texture is exposed to the shader as
`binding_array<texture_2d<f32>, 16>`, and the `diffuse` sampler is
exposed to the shader as `binding_array<sampler, 16>`. Inside the
material's vertex and fragment shaders, the applicable index is
available via the `material_bind_group_slot` field of the `Mesh`
structure. So, for instance, you can access the current color like so:

```wgsl
// `uniform` binding arrays are a non-sequitur, so `uniform` is automatically promoted
// to `storage` in bindless mode.
@group(2) @binding(0) var<storage> material_color: binding_array<Color, 4>;
...
@fragment
fn fragment(in: VertexOutput) -> @location(0) vec4<f32> {
    let color = material_color[mesh[in.instance_index].material_bind_group_slot];
    ...
}
```

Note that portable shader code can't guarantee that the current platform
supports bindless textures. Indeed, bindless mode is only available in
Vulkan and DX12. The `BINDLESS` shader definition is available for your
use to determine whether you're on a bindless platform or not. Thus a
portable version of the shader above would look like:

```wgsl
#ifdef BINDLESS
@group(2) @binding(0) var<storage> material_color: binding_array<Color, 4>;
#else // BINDLESS
@group(2) @binding(0) var<uniform> material_color: Color;
#endif // BINDLESS
...
@fragment
fn fragment(in: VertexOutput) -> @location(0) vec4<f32> {
#ifdef BINDLESS
    let color = material_color[mesh[in.instance_index].material_bind_group_slot];
#else // BINDLESS
    let color = material_color;
#endif // BINDLESS
    ...
}
```

Importantly, this PR *doesn't* update `StandardMaterial` to be bindless.
So, for example, `scene_viewer` will currently not run any faster. I
intend to update `StandardMaterial` to use bindless mode in a follow-up
patch.

A new example, `shaders/shader_material_bindless`, has been added to
demonstrate how to use this new feature.

Here's a Tracy profile of `submit_graph_commands` of this patch and an
additional patch (not submitted yet) that makes `StandardMaterial` use
bindless. Red is those patches; yellow is `main`. The scene was Bistro
Exterior with a hack that forces all textures to opaque. You can see a
1.47x mean speedup.
![Screenshot 2024-11-12
161713](https://github.com/user-attachments/assets/4334b362-42c8-4d64-9cfb-6835f019b95c)

## Migration Guide

* `RenderAssets::prepare_asset` now takes an `AssetId` parameter.
* Bin keys now have Bevy-specific material bind group indices instead of
`wgpu` material bind group IDs, as part of the bindless change. Use the
new `MaterialBindGroupAllocator` to map from bind group index to bind
group ID.
2024-12-03 18:00:34 +00:00

354 lines
15 KiB
Rust

//! Demonstrates how to define and use specialized mesh pipeline
//!
//! This example shows how to use the built-in [`SpecializedMeshPipeline`]
//! functionality with a custom [`RenderCommand`] to allow custom mesh rendering with
//! more flexibility than the material api.
//!
//! [`SpecializedMeshPipeline`] let's you customize the entire pipeline used when rendering a mesh.
use bevy::{
core_pipeline::core_3d::{Opaque3d, Opaque3dBinKey, CORE_3D_DEPTH_FORMAT},
math::{vec3, vec4},
pbr::{
DrawMesh, MeshPipeline, MeshPipelineKey, MeshPipelineViewLayoutKey, RenderMeshInstances,
SetMeshBindGroup, SetMeshViewBindGroup,
},
prelude::*,
render::{
extract_component::{ExtractComponent, ExtractComponentPlugin},
mesh::{Indices, MeshVertexBufferLayoutRef, PrimitiveTopology, RenderMesh},
render_asset::{RenderAssetUsages, RenderAssets},
render_phase::{
AddRenderCommand, BinnedRenderPhaseType, DrawFunctions, SetItemPipeline,
ViewBinnedRenderPhases,
},
render_resource::{
ColorTargetState, ColorWrites, CompareFunction, DepthStencilState, Face, FragmentState,
FrontFace, MultisampleState, PipelineCache, PolygonMode, PrimitiveState,
RenderPipelineDescriptor, SpecializedMeshPipeline, SpecializedMeshPipelineError,
SpecializedMeshPipelines, TextureFormat, VertexState,
},
view::{self, ExtractedView, RenderVisibleEntities, ViewTarget, VisibilitySystems},
Render, RenderApp, RenderSet,
},
};
const SHADER_ASSET_PATH: &str = "shaders/specialized_mesh_pipeline.wgsl";
fn main() {
App::new()
.add_plugins(DefaultPlugins)
.add_plugins(CustomRenderedMeshPipelinePlugin)
.add_systems(Startup, setup)
.run();
}
/// Spawns the objects in the scene.
fn setup(mut commands: Commands, mut meshes: ResMut<Assets<Mesh>>) {
// Build a custom triangle mesh with colors
// We define a custom mesh because the examples only uses a limited
// set of vertex attributes for simplicity
let mesh = Mesh::new(
PrimitiveTopology::TriangleList,
RenderAssetUsages::default(),
)
.with_inserted_indices(Indices::U32(vec![0, 1, 2]))
.with_inserted_attribute(
Mesh::ATTRIBUTE_POSITION,
vec![
vec3(-0.5, -0.5, 0.0),
vec3(0.5, -0.5, 0.0),
vec3(0.0, 0.25, 0.0),
],
)
.with_inserted_attribute(
Mesh::ATTRIBUTE_COLOR,
vec![
vec4(1.0, 0.0, 0.0, 1.0),
vec4(0.0, 1.0, 0.0, 1.0),
vec4(0.0, 0.0, 1.0, 1.0),
],
);
// spawn 3 triangles to show that batching works
for (x, y) in [-0.5, 0.0, 0.5].into_iter().zip([-0.25, 0.5, -0.25]) {
// Spawn an entity with all the required components for it to be rendered with our custom pipeline
commands.spawn((
// We use a marker component to identify the mesh that will be rendered
// with our specialized pipeline
CustomRenderedEntity,
// We need to add the mesh handle to the entity
Mesh3d(meshes.add(mesh.clone())),
Transform::from_xyz(x, y, 0.0),
));
}
// Spawn the camera.
commands.spawn((
Camera3d::default(),
// Move the camera back a bit to see all the triangles
Transform::from_xyz(0.0, 0.0, 3.0).looking_at(Vec3::ZERO, Vec3::Y),
));
}
// When writing custom rendering code it's generally recommended to use a plugin.
// The main reason for this is that it gives you access to the finish() hook
// which is called after rendering resources are initialized.
struct CustomRenderedMeshPipelinePlugin;
impl Plugin for CustomRenderedMeshPipelinePlugin {
fn build(&self, app: &mut App) {
app.add_plugins(ExtractComponentPlugin::<CustomRenderedEntity>::default())
.add_systems(
PostUpdate,
// Make sure to tell Bevy to check our entity for visibility. Bevy won't
// do this by default, for efficiency reasons.
// This will do things like frustum culling and hierarchy visibility
view::check_visibility::<WithCustomRenderedEntity>
.in_set(VisibilitySystems::CheckVisibility),
);
// We make sure to add these to the render app, not the main app.
let Some(render_app) = app.get_sub_app_mut(RenderApp) else {
return;
};
render_app
// This is needed to tell bevy about your custom pipeline
.init_resource::<SpecializedMeshPipelines<CustomMeshPipeline>>()
// We need to use a custom draw command so we need to register it
.add_render_command::<Opaque3d, DrawSpecializedPipelineCommands>()
.add_systems(Render, queue_custom_mesh_pipeline.in_set(RenderSet::Queue));
}
fn finish(&self, app: &mut App) {
let Some(render_app) = app.get_sub_app_mut(RenderApp) else {
return;
};
// Creating this pipeline needs the RenderDevice and RenderQueue
// which are only available once rendering plugins are initialized.
render_app.init_resource::<CustomMeshPipeline>();
}
}
/// A marker component that represents an entity that is to be rendered using
/// our specialized pipeline.
///
/// Note the [`ExtractComponent`] trait implementation. This is necessary to
/// tell Bevy that this object should be pulled into the render world.
#[derive(Clone, Component, ExtractComponent)]
struct CustomRenderedEntity;
/// The custom draw commands that Bevy executes for each entity we enqueue into
/// the render phase.
type DrawSpecializedPipelineCommands = (
// Set the pipeline
SetItemPipeline,
// Set the view uniform at bind group 0
SetMeshViewBindGroup<0>,
// Set the mesh uniform at bind group 1
SetMeshBindGroup<1>,
// Draw the mesh
DrawMesh,
);
/// A query filter that tells [`view::check_visibility`] about our custom
/// rendered entity.
type WithCustomRenderedEntity = With<CustomRenderedEntity>;
// This contains the state needed to speciazlize a mesh pipeline
#[derive(Resource)]
struct CustomMeshPipeline {
/// The base mesh pipeline defined by bevy
///
/// This isn't required, but if you want to use a bevy `Mesh` it's easier when you
/// have access to the base `MeshPipeline` that bevy already defines
mesh_pipeline: MeshPipeline,
/// Stores the shader used for this pipeline directly on the pipeline.
/// This isn't required, it's only done like this for simplicity.
shader_handle: Handle<Shader>,
}
impl FromWorld for CustomMeshPipeline {
fn from_world(world: &mut World) -> Self {
// Load the shader
let shader_handle: Handle<Shader> = world.resource::<AssetServer>().load(SHADER_ASSET_PATH);
Self {
mesh_pipeline: MeshPipeline::from_world(world),
shader_handle,
}
}
}
impl SpecializedMeshPipeline for CustomMeshPipeline {
/// Pipeline use keys to determine how to specialize it.
/// The key is also used by the pipeline cache to determine if
/// it needs to create a new pipeline or not
///
/// In this example we just use the base `MeshPipelineKey` defined by bevy, but this could be anything.
/// For example, if you want to make a pipeline with a procedural shader you could add the Handle<Shader> to the key.
type Key = MeshPipelineKey;
fn specialize(
&self,
mesh_key: Self::Key,
layout: &MeshVertexBufferLayoutRef,
) -> Result<RenderPipelineDescriptor, SpecializedMeshPipelineError> {
// Define the vertex attributes based on a standard bevy [`Mesh`]
let mut vertex_attributes = Vec::new();
if layout.0.contains(Mesh::ATTRIBUTE_POSITION) {
// Make sure this matches the shader location
vertex_attributes.push(Mesh::ATTRIBUTE_POSITION.at_shader_location(0));
}
if layout.0.contains(Mesh::ATTRIBUTE_COLOR) {
// Make sure this matches the shader location
vertex_attributes.push(Mesh::ATTRIBUTE_COLOR.at_shader_location(1));
}
// This will automatically generate the correct `VertexBufferLayout` based on the vertex attributes
let vertex_buffer_layout = layout.0.get_layout(&vertex_attributes)?;
Ok(RenderPipelineDescriptor {
label: Some("Specialized Mesh Pipeline".into()),
layout: vec![
// Bind group 0 is the view uniform
self.mesh_pipeline
.get_view_layout(MeshPipelineViewLayoutKey::from(mesh_key))
.clone(),
// Bind group 1 is the mesh uniform
self.mesh_pipeline.mesh_layouts.model_only.clone(),
],
push_constant_ranges: vec![],
vertex: VertexState {
shader: self.shader_handle.clone(),
shader_defs: vec![],
entry_point: "vertex".into(),
// Customize how to store the meshes' vertex attributes in the vertex buffer
buffers: vec![vertex_buffer_layout],
},
fragment: Some(FragmentState {
shader: self.shader_handle.clone(),
shader_defs: vec![],
entry_point: "fragment".into(),
targets: vec![Some(ColorTargetState {
// This isn't required, but bevy supports HDR and non-HDR rendering
// so it's generally recommended to specialize the pipeline for that
format: if mesh_key.contains(MeshPipelineKey::HDR) {
ViewTarget::TEXTURE_FORMAT_HDR
} else {
TextureFormat::bevy_default()
},
// For this example we only use opaque meshes,
// but if you wanted to use alpha blending you would need to set it here
blend: None,
write_mask: ColorWrites::ALL,
})],
}),
primitive: PrimitiveState {
topology: mesh_key.primitive_topology(),
front_face: FrontFace::Ccw,
cull_mode: Some(Face::Back),
polygon_mode: PolygonMode::Fill,
..default()
},
// Note that if your view has no depth buffer this will need to be
// changed.
depth_stencil: Some(DepthStencilState {
format: CORE_3D_DEPTH_FORMAT,
depth_write_enabled: true,
depth_compare: CompareFunction::GreaterEqual,
stencil: default(),
bias: default(),
}),
// It's generally recommended to specialize your pipeline for MSAA,
// but it's not always possible
multisample: MultisampleState {
count: mesh_key.msaa_samples(),
..MultisampleState::default()
},
zero_initialize_workgroup_memory: false,
})
}
}
/// A render-world system that enqueues the entity with custom rendering into
/// the opaque render phases of each view.
#[allow(clippy::too_many_arguments)]
fn queue_custom_mesh_pipeline(
pipeline_cache: Res<PipelineCache>,
custom_mesh_pipeline: Res<CustomMeshPipeline>,
mut opaque_render_phases: ResMut<ViewBinnedRenderPhases<Opaque3d>>,
opaque_draw_functions: Res<DrawFunctions<Opaque3d>>,
mut specialized_mesh_pipelines: ResMut<SpecializedMeshPipelines<CustomMeshPipeline>>,
views: Query<(Entity, &RenderVisibleEntities, &ExtractedView, &Msaa), With<ExtractedView>>,
render_meshes: Res<RenderAssets<RenderMesh>>,
render_mesh_instances: Res<RenderMeshInstances>,
) {
// Get the id for our custom draw function
let draw_function_id = opaque_draw_functions
.read()
.id::<DrawSpecializedPipelineCommands>();
// Render phases are per-view, so we need to iterate over all views so that
// the entity appears in them. (In this example, we have only one view, but
// it's good practice to loop over all views anyway.)
for (view_entity, view_visible_entities, view, msaa) in views.iter() {
let Some(opaque_phase) = opaque_render_phases.get_mut(&view_entity) else {
continue;
};
// Create the key based on the view. In this case we only care about MSAA and HDR
let view_key = MeshPipelineKey::from_msaa_samples(msaa.samples())
| MeshPipelineKey::from_hdr(view.hdr);
// Find all the custom rendered entities that are visible from this
// view.
for &(render_entity, visible_entity) in view_visible_entities
.get::<WithCustomRenderedEntity>()
.iter()
{
// Get the mesh instance
let Some(mesh_instance) = render_mesh_instances.render_mesh_queue_data(visible_entity)
else {
continue;
};
// Get the mesh data
let Some(mesh) = render_meshes.get(mesh_instance.mesh_asset_id) else {
continue;
};
// Specialize the key for the current mesh entity
// For this example we only specialize based on the mesh topology
// but you could have more complex keys and that's where you'd need to create those keys
let mut mesh_key = view_key;
mesh_key |= MeshPipelineKey::from_primitive_topology(mesh.primitive_topology());
// Finally, we can specialize the pipeline based on the key
let pipeline_id = specialized_mesh_pipelines
.specialize(
&pipeline_cache,
&custom_mesh_pipeline,
mesh_key,
&mesh.layout,
)
// This should never with this example, but if your pipeline specialization
// can fail you need to handle the error here
.expect("Failed to specialize mesh pipeline");
// Add the mesh with our specialized pipeline
opaque_phase.add(
Opaque3dBinKey {
draw_function: draw_function_id,
pipeline: pipeline_id,
// The asset ID is arbitrary; we simply use [`AssetId::invalid`],
// but you can use anything you like. Note that the asset ID need
// not be the ID of a [`Mesh`].
asset_id: AssetId::<Mesh>::invalid().untyped(),
material_bind_group_index: None,
lightmap_image: None,
},
(render_entity, visible_entity),
// This example supports batching, but if your pipeline doesn't
// support it you can use `BinnedRenderPhaseType::UnbatchableMesh`
BinnedRenderPhaseType::BatchableMesh,
);
}
}
}