bevy/examples/stress_tests/many_lights.rs

189 lines
6.1 KiB
Rust
Raw Normal View History

//! Simple benchmark to test rendering many point lights.
//! Run with `WGPU_SETTINGS_PRIO=webgl2` to restrict to uniform buffers and max 256 lights.
use std::f64::consts::PI;
Use storage buffers for clustered forward point lights (#3989) # Objective - Make use of storage buffers, where they are available, for clustered forward bindings to support far more point lights in a scene - Fixes #3605 - Based on top of #4079 This branch on an M1 Max can keep 60fps with about 2150 point lights of radius 1m in the Sponza scene where I've been testing. The bottleneck is mostly assigning lights to clusters which grows faster than linearly (I think 1000 lights was about 1.5ms and 5000 was 7.5ms). I have seen papers and presentations leveraging compute shaders that can get this up to over 1 million. That said, I think any further optimisations should probably be done in a separate PR. ## Solution - Add `RenderDevice` to the `Material` and `SpecializedMaterial` trait `::key()` functions to allow setting flags on the keys depending on feature/limit availability - Make `GpuPointLights` and `ViewClusterBuffers` into enums containing `UniformVec` and `StorageBuffer` variants. Implement the necessary API on them to make usage the same for both cases, and the only difference is at initialisation time. - Appropriate shader defs in the shader code to handle the two cases ## Context on some decisions / open questions - I'm using `max_storage_buffers_per_shader_stage >= 3` as a check to see if storage buffers are supported. I was thinking about diving into 'binding resource management' but it feels like we don't have enough use cases to understand the problem yet, and it is mostly a separate concern to this PR, so I think it should be handled separately. - Should `ViewClusterBuffers` and `ViewClusterBindings` be merged, duplicating the count variables into the enum variants? Co-authored-by: Carter Anderson <mcanders1@gmail.com>
2022-04-07 16:16:35 +00:00
use bevy::{
color::palettes::css::DEEP_PINK,
Use storage buffers for clustered forward point lights (#3989) # Objective - Make use of storage buffers, where they are available, for clustered forward bindings to support far more point lights in a scene - Fixes #3605 - Based on top of #4079 This branch on an M1 Max can keep 60fps with about 2150 point lights of radius 1m in the Sponza scene where I've been testing. The bottleneck is mostly assigning lights to clusters which grows faster than linearly (I think 1000 lights was about 1.5ms and 5000 was 7.5ms). I have seen papers and presentations leveraging compute shaders that can get this up to over 1 million. That said, I think any further optimisations should probably be done in a separate PR. ## Solution - Add `RenderDevice` to the `Material` and `SpecializedMaterial` trait `::key()` functions to allow setting flags on the keys depending on feature/limit availability - Make `GpuPointLights` and `ViewClusterBuffers` into enums containing `UniformVec` and `StorageBuffer` variants. Implement the necessary API on them to make usage the same for both cases, and the only difference is at initialisation time. - Appropriate shader defs in the shader code to handle the two cases ## Context on some decisions / open questions - I'm using `max_storage_buffers_per_shader_stage >= 3` as a check to see if storage buffers are supported. I was thinking about diving into 'binding resource management' but it feels like we don't have enough use cases to understand the problem yet, and it is mostly a separate concern to this PR, so I think it should be handled separately. - Should `ViewClusterBuffers` and `ViewClusterBindings` be merged, duplicating the count variables into the enum variants? Co-authored-by: Carter Anderson <mcanders1@gmail.com>
2022-04-07 16:16:35 +00:00
diagnostic::{FrameTimeDiagnosticsPlugin, LogDiagnosticsPlugin},
math::{DVec2, DVec3},
pbr::{ExtractedPointLight, GlobalClusterableObjectMeta},
Use storage buffers for clustered forward point lights (#3989) # Objective - Make use of storage buffers, where they are available, for clustered forward bindings to support far more point lights in a scene - Fixes #3605 - Based on top of #4079 This branch on an M1 Max can keep 60fps with about 2150 point lights of radius 1m in the Sponza scene where I've been testing. The bottleneck is mostly assigning lights to clusters which grows faster than linearly (I think 1000 lights was about 1.5ms and 5000 was 7.5ms). I have seen papers and presentations leveraging compute shaders that can get this up to over 1 million. That said, I think any further optimisations should probably be done in a separate PR. ## Solution - Add `RenderDevice` to the `Material` and `SpecializedMaterial` trait `::key()` functions to allow setting flags on the keys depending on feature/limit availability - Make `GpuPointLights` and `ViewClusterBuffers` into enums containing `UniformVec` and `StorageBuffer` variants. Implement the necessary API on them to make usage the same for both cases, and the only difference is at initialisation time. - Appropriate shader defs in the shader code to handle the two cases ## Context on some decisions / open questions - I'm using `max_storage_buffers_per_shader_stage >= 3` as a check to see if storage buffers are supported. I was thinking about diving into 'binding resource management' but it feels like we don't have enough use cases to understand the problem yet, and it is mostly a separate concern to this PR, so I think it should be handled separately. - Should `ViewClusterBuffers` and `ViewClusterBindings` be merged, duplicating the count variables into the enum variants? Co-authored-by: Carter Anderson <mcanders1@gmail.com>
2022-04-07 16:16:35 +00:00
prelude::*,
render::{camera::ScalingMode, Render, RenderApp, RenderSet},
window::{PresentMode, WindowResolution},
winit::{UpdateMode, WinitSettings},
Use storage buffers for clustered forward point lights (#3989) # Objective - Make use of storage buffers, where they are available, for clustered forward bindings to support far more point lights in a scene - Fixes #3605 - Based on top of #4079 This branch on an M1 Max can keep 60fps with about 2150 point lights of radius 1m in the Sponza scene where I've been testing. The bottleneck is mostly assigning lights to clusters which grows faster than linearly (I think 1000 lights was about 1.5ms and 5000 was 7.5ms). I have seen papers and presentations leveraging compute shaders that can get this up to over 1 million. That said, I think any further optimisations should probably be done in a separate PR. ## Solution - Add `RenderDevice` to the `Material` and `SpecializedMaterial` trait `::key()` functions to allow setting flags on the keys depending on feature/limit availability - Make `GpuPointLights` and `ViewClusterBuffers` into enums containing `UniformVec` and `StorageBuffer` variants. Implement the necessary API on them to make usage the same for both cases, and the only difference is at initialisation time. - Appropriate shader defs in the shader code to handle the two cases ## Context on some decisions / open questions - I'm using `max_storage_buffers_per_shader_stage >= 3` as a check to see if storage buffers are supported. I was thinking about diving into 'binding resource management' but it feels like we don't have enough use cases to understand the problem yet, and it is mostly a separate concern to this PR, so I think it should be handled separately. - Should `ViewClusterBuffers` and `ViewClusterBindings` be merged, duplicating the count variables into the enum variants? Co-authored-by: Carter Anderson <mcanders1@gmail.com>
2022-04-07 16:16:35 +00:00
};
use rand::{thread_rng, Rng};
Use storage buffers for clustered forward point lights (#3989) # Objective - Make use of storage buffers, where they are available, for clustered forward bindings to support far more point lights in a scene - Fixes #3605 - Based on top of #4079 This branch on an M1 Max can keep 60fps with about 2150 point lights of radius 1m in the Sponza scene where I've been testing. The bottleneck is mostly assigning lights to clusters which grows faster than linearly (I think 1000 lights was about 1.5ms and 5000 was 7.5ms). I have seen papers and presentations leveraging compute shaders that can get this up to over 1 million. That said, I think any further optimisations should probably be done in a separate PR. ## Solution - Add `RenderDevice` to the `Material` and `SpecializedMaterial` trait `::key()` functions to allow setting flags on the keys depending on feature/limit availability - Make `GpuPointLights` and `ViewClusterBuffers` into enums containing `UniformVec` and `StorageBuffer` variants. Implement the necessary API on them to make usage the same for both cases, and the only difference is at initialisation time. - Appropriate shader defs in the shader code to handle the two cases ## Context on some decisions / open questions - I'm using `max_storage_buffers_per_shader_stage >= 3` as a check to see if storage buffers are supported. I was thinking about diving into 'binding resource management' but it feels like we don't have enough use cases to understand the problem yet, and it is mostly a separate concern to this PR, so I think it should be handled separately. - Should `ViewClusterBuffers` and `ViewClusterBindings` be merged, duplicating the count variables into the enum variants? Co-authored-by: Carter Anderson <mcanders1@gmail.com>
2022-04-07 16:16:35 +00:00
fn main() {
App::new()
.add_plugins((
DefaultPlugins.set(WindowPlugin {
primary_window: Some(Window {
resolution: WindowResolution::new(1920.0, 1080.0)
.with_scale_factor_override(1.0),
title: "many_lights".into(),
present_mode: PresentMode::AutoNoVsync,
..default()
}),
Plugins own their settings. Rework PluginGroup trait. (#6336) # Objective Fixes #5884 #2879 Alternative to #2988 #5885 #2886 "Immutable" Plugin settings are currently represented as normal ECS resources, which are read as part of plugin init. This presents a number of problems: 1. If a user inserts the plugin settings resource after the plugin is initialized, it will be silently ignored (and use the defaults instead) 2. Users can modify the plugin settings resource after the plugin has been initialized. This creates a false sense of control over settings that can no longer be changed. (1) and (2) are especially problematic and confusing for the `WindowDescriptor` resource, but this is a general problem. ## Solution Immutable Plugin settings now live on each Plugin struct (ex: `WindowPlugin`). PluginGroups have been reworked to support overriding plugin values. This also removes the need for the `add_plugins_with` api, as the `add_plugins` api can use the builder pattern directly. Settings that can be used at runtime continue to be represented as ECS resources. Plugins are now configured like this: ```rust app.add_plugin(AssetPlugin { watch_for_changes: true, ..default() }) ``` PluginGroups are now configured like this: ```rust app.add_plugins(DefaultPlugins .set(AssetPlugin { watch_for_changes: true, ..default() }) ) ``` This is an alternative to #2988, which is similar. But I personally prefer this solution for a couple of reasons: * ~~#2988 doesn't solve (1)~~ #2988 does solve (1) and will panic in that case. I was wrong! * This PR directly ties plugin settings to Plugin types in a 1:1 relationship, rather than a loose "setup resource" <-> plugin coupling (where the setup resource is consumed by the first plugin that uses it). * I'm not a huge fan of overloading the ECS resource concept and implementation for something that has very different use cases and constraints. ## Changelog - PluginGroups can now be configured directly using the builder pattern. Individual plugin values can be overridden by using `plugin_group.set(SomePlugin {})`, which enables overriding default plugin values. - `WindowDescriptor` plugin settings have been moved to `WindowPlugin` and `AssetServerSettings` have been moved to `AssetPlugin` - `app.add_plugins_with` has been replaced by using `add_plugins` with the builder pattern. ## Migration Guide The `WindowDescriptor` settings have been moved from a resource to `WindowPlugin::window`: ```rust // Old (Bevy 0.8) app .insert_resource(WindowDescriptor { width: 400.0, ..default() }) .add_plugins(DefaultPlugins) // New (Bevy 0.9) app.add_plugins(DefaultPlugins.set(WindowPlugin { window: WindowDescriptor { width: 400.0, ..default() }, ..default() })) ``` The `AssetServerSettings` resource has been removed in favor of direct `AssetPlugin` configuration: ```rust // Old (Bevy 0.8) app .insert_resource(AssetServerSettings { watch_for_changes: true, ..default() }) .add_plugins(DefaultPlugins) // New (Bevy 0.9) app.add_plugins(DefaultPlugins.set(AssetPlugin { watch_for_changes: true, ..default() })) ``` `add_plugins_with` has been replaced by `add_plugins` in combination with the builder pattern: ```rust // Old (Bevy 0.8) app.add_plugins_with(DefaultPlugins, |group| group.disable::<AssetPlugin>()); // New (Bevy 0.9) app.add_plugins(DefaultPlugins.build().disable::<AssetPlugin>()); ```
2022-10-24 21:20:33 +00:00
..default()
Windows as Entities (#5589) # Objective Fix https://github.com/bevyengine/bevy/issues/4530 - Make it easier to open/close/modify windows by setting them up as `Entity`s with a `Window` component. - Make multiple windows very simple to set up. (just add a `Window` component to an entity and it should open) ## Solution - Move all properties of window descriptor to ~components~ a component. - Replace `WindowId` with `Entity`. - ~Use change detection for components to update backend rather than events/commands. (The `CursorMoved`/`WindowResized`/... events are kept for user convenience.~ Check each field individually to see what we need to update, events are still kept for user convenience. --- ## Changelog - `WindowDescriptor` renamed to `Window`. - Width/height consolidated into a `WindowResolution` component. - Requesting maximization/minimization is done on the [`Window::state`] field. - `WindowId` is now `Entity`. ## Migration Guide - Replace `WindowDescriptor` with `Window`. - Change `width` and `height` fields in a `WindowResolution`, either by doing ```rust WindowResolution::new(width, height) // Explicitly // or using From<_> for tuples for convenience (1920., 1080.).into() ``` - Replace any `WindowCommand` code to just modify the `Window`'s fields directly and creating/closing windows is now by spawning/despawning an entity with a `Window` component like so: ```rust let window = commands.spawn(Window { ... }).id(); // open window commands.entity(window).despawn(); // close window ``` ## Unresolved - ~How do we tell when a window is minimized by a user?~ ~Currently using the `Resize(0, 0)` as an indicator of minimization.~ No longer attempting to tell given how finnicky this was across platforms, now the user can only request that a window be maximized/minimized. ## Future work - Move `exit_on_close` functionality out from windowing and into app(?) - https://github.com/bevyengine/bevy/issues/5621 - https://github.com/bevyengine/bevy/issues/7099 - https://github.com/bevyengine/bevy/issues/7098 Co-authored-by: Carter Anderson <mcanders1@gmail.com>
2023-01-19 00:38:28 +00:00
}),
FrameTimeDiagnosticsPlugin,
LogDiagnosticsPlugin::default(),
LogVisibleLights,
))
.insert_resource(WinitSettings {
focused_mode: UpdateMode::Continuous,
unfocused_mode: UpdateMode::Continuous,
})
.add_systems(Startup, setup)
.add_systems(Update, (move_camera, print_light_count))
Use storage buffers for clustered forward point lights (#3989) # Objective - Make use of storage buffers, where they are available, for clustered forward bindings to support far more point lights in a scene - Fixes #3605 - Based on top of #4079 This branch on an M1 Max can keep 60fps with about 2150 point lights of radius 1m in the Sponza scene where I've been testing. The bottleneck is mostly assigning lights to clusters which grows faster than linearly (I think 1000 lights was about 1.5ms and 5000 was 7.5ms). I have seen papers and presentations leveraging compute shaders that can get this up to over 1 million. That said, I think any further optimisations should probably be done in a separate PR. ## Solution - Add `RenderDevice` to the `Material` and `SpecializedMaterial` trait `::key()` functions to allow setting flags on the keys depending on feature/limit availability - Make `GpuPointLights` and `ViewClusterBuffers` into enums containing `UniformVec` and `StorageBuffer` variants. Implement the necessary API on them to make usage the same for both cases, and the only difference is at initialisation time. - Appropriate shader defs in the shader code to handle the two cases ## Context on some decisions / open questions - I'm using `max_storage_buffers_per_shader_stage >= 3` as a check to see if storage buffers are supported. I was thinking about diving into 'binding resource management' but it feels like we don't have enough use cases to understand the problem yet, and it is mostly a separate concern to this PR, so I think it should be handled separately. - Should `ViewClusterBuffers` and `ViewClusterBindings` be merged, duplicating the count variables into the enum variants? Co-authored-by: Carter Anderson <mcanders1@gmail.com>
2022-04-07 16:16:35 +00:00
.run();
}
fn setup(
mut commands: Commands,
mut meshes: ResMut<Assets<Mesh>>,
mut materials: ResMut<Assets<StandardMaterial>>,
) {
warn!(include_str!("warning_string.txt"));
Use storage buffers for clustered forward point lights (#3989) # Objective - Make use of storage buffers, where they are available, for clustered forward bindings to support far more point lights in a scene - Fixes #3605 - Based on top of #4079 This branch on an M1 Max can keep 60fps with about 2150 point lights of radius 1m in the Sponza scene where I've been testing. The bottleneck is mostly assigning lights to clusters which grows faster than linearly (I think 1000 lights was about 1.5ms and 5000 was 7.5ms). I have seen papers and presentations leveraging compute shaders that can get this up to over 1 million. That said, I think any further optimisations should probably be done in a separate PR. ## Solution - Add `RenderDevice` to the `Material` and `SpecializedMaterial` trait `::key()` functions to allow setting flags on the keys depending on feature/limit availability - Make `GpuPointLights` and `ViewClusterBuffers` into enums containing `UniformVec` and `StorageBuffer` variants. Implement the necessary API on them to make usage the same for both cases, and the only difference is at initialisation time. - Appropriate shader defs in the shader code to handle the two cases ## Context on some decisions / open questions - I'm using `max_storage_buffers_per_shader_stage >= 3` as a check to see if storage buffers are supported. I was thinking about diving into 'binding resource management' but it feels like we don't have enough use cases to understand the problem yet, and it is mostly a separate concern to this PR, so I think it should be handled separately. - Should `ViewClusterBuffers` and `ViewClusterBindings` be merged, duplicating the count variables into the enum variants? Co-authored-by: Carter Anderson <mcanders1@gmail.com>
2022-04-07 16:16:35 +00:00
const LIGHT_RADIUS: f32 = 0.3;
New Exposure and Lighting Defaults (and calibrate examples) (#11868) # Objective After adding configurable exposure, we set the default ev100 value to `7` (indoor). This brought us out of sync with Blender's configuration and defaults. This PR changes the default to `9.7` (bright indoor or very overcast outdoors), as I calibrated in #11577. This feels like a very reasonable default. The other changes generally center around tweaking Bevy's lighting defaults and examples to play nicely with this number, alongside a few other tweaks and improvements. Note that for artistic reasons I have reverted some examples, which changed to directional lights in #11581, back to point lights. Fixes #11577 --- ## Changelog - Changed `Exposure::ev100` from `7` to `9.7` to better match Blender - Renamed `ExposureSettings` to `Exposure` - `Camera3dBundle` now includes `Exposure` for discoverability - Bumped `FULL_DAYLIGHT ` and `DIRECT_SUNLIGHT` to represent the middle-to-top of those ranges instead of near the bottom - Added new `AMBIENT_DAYLIGHT` constant and set that as the new `DirectionalLight` default illuminance. - `PointLight` and `SpotLight` now have a default `intensity` of 1,000,000 lumens. This makes them actually useful in the context of the new "semi-outdoor" exposure and puts them in the "cinema lighting" category instead of the "common household light" category. They are also reasonably close to the Blender default. - `AmbientLight` default has been bumped from `20` to `80`. ## Migration Guide - The increased `Exposure::ev100` means that all existing 3D lighting will need to be adjusted to match (DirectionalLights, PointLights, SpotLights, EnvironmentMapLights, etc). Or alternatively, you can adjust the `Exposure::ev100` on your cameras to work nicely with your current lighting values. If you are currently relying on default intensity values, you might need to change the intensity to achieve the same effect. Note that in Bevy 0.12, point/spot lights had a different hard coded ev100 value than directional lights. In Bevy 0.13, they use the same ev100, so if you have both in your scene, the _scale_ between these light types has changed and you will likely need to adjust one or both of them.
2024-02-15 20:42:48 +00:00
const LIGHT_INTENSITY: f32 = 1000.0;
Use storage buffers for clustered forward point lights (#3989) # Objective - Make use of storage buffers, where they are available, for clustered forward bindings to support far more point lights in a scene - Fixes #3605 - Based on top of #4079 This branch on an M1 Max can keep 60fps with about 2150 point lights of radius 1m in the Sponza scene where I've been testing. The bottleneck is mostly assigning lights to clusters which grows faster than linearly (I think 1000 lights was about 1.5ms and 5000 was 7.5ms). I have seen papers and presentations leveraging compute shaders that can get this up to over 1 million. That said, I think any further optimisations should probably be done in a separate PR. ## Solution - Add `RenderDevice` to the `Material` and `SpecializedMaterial` trait `::key()` functions to allow setting flags on the keys depending on feature/limit availability - Make `GpuPointLights` and `ViewClusterBuffers` into enums containing `UniformVec` and `StorageBuffer` variants. Implement the necessary API on them to make usage the same for both cases, and the only difference is at initialisation time. - Appropriate shader defs in the shader code to handle the two cases ## Context on some decisions / open questions - I'm using `max_storage_buffers_per_shader_stage >= 3` as a check to see if storage buffers are supported. I was thinking about diving into 'binding resource management' but it feels like we don't have enough use cases to understand the problem yet, and it is mostly a separate concern to this PR, so I think it should be handled separately. - Should `ViewClusterBuffers` and `ViewClusterBindings` be merged, duplicating the count variables into the enum variants? Co-authored-by: Carter Anderson <mcanders1@gmail.com>
2022-04-07 16:16:35 +00:00
const RADIUS: f32 = 50.0;
const N_LIGHTS: usize = 100_000;
Migrate meshes and materials to required components (#15524) # Objective A big step in the migration to required components: meshes and materials! ## Solution As per the [selected proposal](https://hackmd.io/@bevy/required_components/%2Fj9-PnF-2QKK0on1KQ29UWQ): - Deprecate `MaterialMesh2dBundle`, `MaterialMeshBundle`, and `PbrBundle`. - Add `Mesh2d` and `Mesh3d` components, which wrap a `Handle<Mesh>`. - Add `MeshMaterial2d<M: Material2d>` and `MeshMaterial3d<M: Material>`, which wrap a `Handle<M>`. - Meshes *without* a mesh material should be rendered with a default material. The existence of a material is determined by `HasMaterial2d`/`HasMaterial3d`, which is required by `MeshMaterial2d`/`MeshMaterial3d`. This gets around problems with the generics. Previously: ```rust commands.spawn(MaterialMesh2dBundle { mesh: meshes.add(Circle::new(100.0)).into(), material: materials.add(Color::srgb(7.5, 0.0, 7.5)), transform: Transform::from_translation(Vec3::new(-200., 0., 0.)), ..default() }); ``` Now: ```rust commands.spawn(( Mesh2d(meshes.add(Circle::new(100.0))), MeshMaterial2d(materials.add(Color::srgb(7.5, 0.0, 7.5))), Transform::from_translation(Vec3::new(-200., 0., 0.)), )); ``` If the mesh material is missing, previously nothing was rendered. Now, it renders a white default `ColorMaterial` in 2D and a `StandardMaterial` in 3D (this can be overridden). Below, only every other entity has a material: ![Näyttökuva 2024-09-29 181746](https://github.com/user-attachments/assets/5c8be029-d2fe-4b8c-ae89-17a72ff82c9a) ![Näyttökuva 2024-09-29 181918](https://github.com/user-attachments/assets/58adbc55-5a1e-4c7d-a2c7-ed456227b909) Why white? This is still open for discussion, but I think white makes sense for a *default* material, while *invalid* asset handles pointing to nothing should have something like a pink material to indicate that something is broken (I don't handle that in this PR yet). This is kind of a mix of Godot and Unity: Godot just renders a white material for non-existent materials, while Unity renders nothing when no materials exist, but renders pink for invalid materials. I can also change the default material to pink if that is preferable though. ## Testing I ran some 2D and 3D examples to test if anything changed visually. I have not tested all examples or features yet however. If anyone wants to test more extensively, it would be appreciated! ## Implementation Notes - The relationship between `bevy_render` and `bevy_pbr` is weird here. `bevy_render` needs `Mesh3d` for its own systems, but `bevy_pbr` has all of the material logic, and `bevy_render` doesn't depend on it. I feel like the two crates should be refactored in some way, but I think that's out of scope for this PR. - I didn't migrate meshlets to required components yet. That can probably be done in a follow-up, as this is already a huge PR. - It is becoming increasingly clear to me that we really, *really* want to disallow raw asset handles as components. They caused me a *ton* of headache here already, and it took me a long time to find every place that queried for them or inserted them directly on entities, since there were no compiler errors for it. If we don't remove the `Component` derive, I expect raw asset handles to be a *huge* footgun for users as we transition to wrapper components, especially as handles as components have been the norm so far. I personally consider this to be a blocker for 0.15: we need to migrate to wrapper components for asset handles everywhere, and remove the `Component` derive. Also see https://github.com/bevyengine/bevy/issues/14124. --- ## Migration Guide Asset handles for meshes and mesh materials must now be wrapped in the `Mesh2d` and `MeshMaterial2d` or `Mesh3d` and `MeshMaterial3d` components for 2D and 3D respectively. Raw handles as components no longer render meshes. Additionally, `MaterialMesh2dBundle`, `MaterialMeshBundle`, and `PbrBundle` have been deprecated. Instead, use the mesh and material components directly. Previously: ```rust commands.spawn(MaterialMesh2dBundle { mesh: meshes.add(Circle::new(100.0)).into(), material: materials.add(Color::srgb(7.5, 0.0, 7.5)), transform: Transform::from_translation(Vec3::new(-200., 0., 0.)), ..default() }); ``` Now: ```rust commands.spawn(( Mesh2d(meshes.add(Circle::new(100.0))), MeshMaterial2d(materials.add(Color::srgb(7.5, 0.0, 7.5))), Transform::from_translation(Vec3::new(-200., 0., 0.)), )); ``` If the mesh material is missing, a white default material is now used. Previously, nothing was rendered if the material was missing. The `WithMesh2d` and `WithMesh3d` query filter type aliases have also been removed. Simply use `With<Mesh2d>` or `With<Mesh3d>`. --------- Co-authored-by: Tim Blackbird <justthecooldude@gmail.com> Co-authored-by: Carter Anderson <mcanders1@gmail.com>
2024-10-01 21:33:17 +00:00
commands.spawn((
Mesh3d(meshes.add(Sphere::new(RADIUS).mesh().ico(9).unwrap())),
MeshMaterial3d(materials.add(Color::WHITE)),
Transform::from_scale(Vec3::NEG_ONE),
));
Use storage buffers for clustered forward point lights (#3989) # Objective - Make use of storage buffers, where they are available, for clustered forward bindings to support far more point lights in a scene - Fixes #3605 - Based on top of #4079 This branch on an M1 Max can keep 60fps with about 2150 point lights of radius 1m in the Sponza scene where I've been testing. The bottleneck is mostly assigning lights to clusters which grows faster than linearly (I think 1000 lights was about 1.5ms and 5000 was 7.5ms). I have seen papers and presentations leveraging compute shaders that can get this up to over 1 million. That said, I think any further optimisations should probably be done in a separate PR. ## Solution - Add `RenderDevice` to the `Material` and `SpecializedMaterial` trait `::key()` functions to allow setting flags on the keys depending on feature/limit availability - Make `GpuPointLights` and `ViewClusterBuffers` into enums containing `UniformVec` and `StorageBuffer` variants. Implement the necessary API on them to make usage the same for both cases, and the only difference is at initialisation time. - Appropriate shader defs in the shader code to handle the two cases ## Context on some decisions / open questions - I'm using `max_storage_buffers_per_shader_stage >= 3` as a check to see if storage buffers are supported. I was thinking about diving into 'binding resource management' but it feels like we don't have enough use cases to understand the problem yet, and it is mostly a separate concern to this PR, so I think it should be handled separately. - Should `ViewClusterBuffers` and `ViewClusterBindings` be merged, duplicating the count variables into the enum variants? Co-authored-by: Carter Anderson <mcanders1@gmail.com>
2022-04-07 16:16:35 +00:00
Deprecate shapes in `bevy_render::mesh::shape` (#11773) # Objective #11431 and #11688 implemented meshing support for Bevy's new geometric primitives. The next step is to deprecate the shapes in `bevy_render::mesh::shape` and to later remove them completely for 0.14. ## Solution Deprecate the shapes and reduce code duplication by utilizing the primitive meshing API for the old shapes where possible. Note that some shapes have behavior that can't be exactly reproduced with the new primitives yet: - `Box` is more of an AABB with min/max extents - `Plane` supports a subdivision count - `Quad` has a `flipped` property These types have not been changed to utilize the new primitives yet. --- ## Changelog - Deprecated all shapes in `bevy_render::mesh::shape` - Changed all examples to use new primitives for meshing ## Migration Guide Bevy has previously used rendering-specific types like `UVSphere` and `Quad` for primitive mesh shapes. These have now been deprecated to use the geometric primitives newly introduced in version 0.13. Some examples: ```rust let before = meshes.add(shape::Box::new(5.0, 0.15, 5.0)); let after = meshes.add(Cuboid::new(5.0, 0.15, 5.0)); let before = meshes.add(shape::Quad::default()); let after = meshes.add(Rectangle::default()); let before = meshes.add(shape::Plane::from_size(5.0)); // The surface normal can now also be specified when using `new` let after = meshes.add(Plane3d::default().mesh().size(5.0, 5.0)); let before = meshes.add( Mesh::try_from(shape::Icosphere { radius: 0.5, subdivisions: 5, }) .unwrap(), ); let after = meshes.add(Sphere::new(0.5).mesh().ico(5).unwrap()); ```
2024-02-08 18:01:34 +00:00
let mesh = meshes.add(Cuboid::default());
Use storage buffers for clustered forward point lights (#3989) # Objective - Make use of storage buffers, where they are available, for clustered forward bindings to support far more point lights in a scene - Fixes #3605 - Based on top of #4079 This branch on an M1 Max can keep 60fps with about 2150 point lights of radius 1m in the Sponza scene where I've been testing. The bottleneck is mostly assigning lights to clusters which grows faster than linearly (I think 1000 lights was about 1.5ms and 5000 was 7.5ms). I have seen papers and presentations leveraging compute shaders that can get this up to over 1 million. That said, I think any further optimisations should probably be done in a separate PR. ## Solution - Add `RenderDevice` to the `Material` and `SpecializedMaterial` trait `::key()` functions to allow setting flags on the keys depending on feature/limit availability - Make `GpuPointLights` and `ViewClusterBuffers` into enums containing `UniformVec` and `StorageBuffer` variants. Implement the necessary API on them to make usage the same for both cases, and the only difference is at initialisation time. - Appropriate shader defs in the shader code to handle the two cases ## Context on some decisions / open questions - I'm using `max_storage_buffers_per_shader_stage >= 3` as a check to see if storage buffers are supported. I was thinking about diving into 'binding resource management' but it feels like we don't have enough use cases to understand the problem yet, and it is mostly a separate concern to this PR, so I think it should be handled separately. - Should `ViewClusterBuffers` and `ViewClusterBindings` be merged, duplicating the count variables into the enum variants? Co-authored-by: Carter Anderson <mcanders1@gmail.com>
2022-04-07 16:16:35 +00:00
let material = materials.add(StandardMaterial {
base_color: DEEP_PINK.into(),
Use storage buffers for clustered forward point lights (#3989) # Objective - Make use of storage buffers, where they are available, for clustered forward bindings to support far more point lights in a scene - Fixes #3605 - Based on top of #4079 This branch on an M1 Max can keep 60fps with about 2150 point lights of radius 1m in the Sponza scene where I've been testing. The bottleneck is mostly assigning lights to clusters which grows faster than linearly (I think 1000 lights was about 1.5ms and 5000 was 7.5ms). I have seen papers and presentations leveraging compute shaders that can get this up to over 1 million. That said, I think any further optimisations should probably be done in a separate PR. ## Solution - Add `RenderDevice` to the `Material` and `SpecializedMaterial` trait `::key()` functions to allow setting flags on the keys depending on feature/limit availability - Make `GpuPointLights` and `ViewClusterBuffers` into enums containing `UniformVec` and `StorageBuffer` variants. Implement the necessary API on them to make usage the same for both cases, and the only difference is at initialisation time. - Appropriate shader defs in the shader code to handle the two cases ## Context on some decisions / open questions - I'm using `max_storage_buffers_per_shader_stage >= 3` as a check to see if storage buffers are supported. I was thinking about diving into 'binding resource management' but it feels like we don't have enough use cases to understand the problem yet, and it is mostly a separate concern to this PR, so I think it should be handled separately. - Should `ViewClusterBuffers` and `ViewClusterBindings` be merged, duplicating the count variables into the enum variants? Co-authored-by: Carter Anderson <mcanders1@gmail.com>
2022-04-07 16:16:35 +00:00
..default()
});
// NOTE: This pattern is good for testing performance of culling as it provides roughly
// the same number of visible meshes regardless of the viewing angle.
// NOTE: f64 is used to avoid precision issues that produce visual artifacts in the distribution
let golden_ratio = 0.5f64 * (1.0f64 + 5.0f64.sqrt());
Use `spawn_batch` in `many_lights` example (#11979) # Objective - The `many_lights` example uses a for-loop around `commands.spawn`. - It is generally recommended to use `spawn_batch` instead to lazily spawn entities, because it doesn't massively grow the command queue. ## Solution - Use `spawn_batch` in `many_lights` example. --- ## Discussion - `thread_rng` is called for each light spawned. This is a simple thread-local `Rc` clone, so it should compile down to a copy and an increment + decrement instruction. - I created `golden_ration` outside of the closure and `move`d it in. This should just be a copy and hopefully will get const-evaluated away. Would it be better to just move it into the closure itself? ## Performance Using `spawn_batch` seems to decrease time-to-first-`Update` by 0.1s: 1.3s to 1.2s. <details> <summary>Raw data and how it was collected.</summary> Before: - 2024-02-19T15:18:57.650987Z to 2024-02-19T15:18:58.912244Z : 1.3 - 2024-02-19T15:19:25.277135Z to 2024-02-19T15:19:26.542092Z : 1.3 - 2024-02-19T15:19:46.841460Z to 2024-02-19T15:19:48.137560Z : 1.3 After: - 2024-02-19T15:17:05.749521Z to 2024-02-19T15:17:06.993221Z : 1.2 - 2024-02-19T15:17:38.153049Z to 2024-02-19T15:17:39.393760Z : 1.2 - 2024-02-19T15:18:10.691562Z to 2024-02-19T15:18:11.891430Z : 1.2 To time performance, I tracked the time from the first `Startup` logged message to the first `Update` logged message. ```shell $ cargo run --release --example many_lights Compiling bevy v0.13.0 (/Users/bdeep/dev/bevy/bevy) Finished release [optimized] target(s) in 1.54s Running `target/release/examples/many_lights` # THIS TIME 2024-02-19T15:30:13.429609Z INFO bevy_render::renderer: AdapterInfo { name: "Apple M1", vendor: 0, device: 0, device_type: IntegratedGpu, driver: "", driver_info: "", backend: Metal } 2024-02-19T15:30:13.566856Z INFO bevy_winit::system: Creating new window "many_lights" (0v1) 2024-02-19T15:30:13.592371Z WARN many_lights: This is a stress test used to push Bevy to its limit and debug performance issues. It is not representative of an actual game. It must be run in release mode using --release or it will be very slow. 2024-02-19T15:30:13.592572Z INFO bevy_diagnostic::system_information_diagnostics_plugin::internal: SystemInfo { os: "MacOS 14.2.1 ", kernel: "23.2.0", cpu: "Apple M1", core_count: "8", memory: "16.0 GiB" } # TO THIS TIME 2024-02-19T15:30:15.429900Z INFO many_lights: Lights: 100000 2024-02-19T15:30:15.430139Z INFO bevy diagnostic: fps : 0.982693 (avg 43.026557) 2024-02-19T15:30:15.430157Z INFO bevy diagnostic: frame_time : 1017.611750ms (avg 149.456476ms) 2024-02-19T15:30:15.430165Z INFO bevy diagnostic: frame_count: 12.000000 (avg 6.000000) ``` </details>
2024-02-26 16:02:27 +00:00
// Spawn N_LIGHTS many lights
commands.spawn_batch((0..N_LIGHTS).map(move |i| {
let mut rng = thread_rng();
Use storage buffers for clustered forward point lights (#3989) # Objective - Make use of storage buffers, where they are available, for clustered forward bindings to support far more point lights in a scene - Fixes #3605 - Based on top of #4079 This branch on an M1 Max can keep 60fps with about 2150 point lights of radius 1m in the Sponza scene where I've been testing. The bottleneck is mostly assigning lights to clusters which grows faster than linearly (I think 1000 lights was about 1.5ms and 5000 was 7.5ms). I have seen papers and presentations leveraging compute shaders that can get this up to over 1 million. That said, I think any further optimisations should probably be done in a separate PR. ## Solution - Add `RenderDevice` to the `Material` and `SpecializedMaterial` trait `::key()` functions to allow setting flags on the keys depending on feature/limit availability - Make `GpuPointLights` and `ViewClusterBuffers` into enums containing `UniformVec` and `StorageBuffer` variants. Implement the necessary API on them to make usage the same for both cases, and the only difference is at initialisation time. - Appropriate shader defs in the shader code to handle the two cases ## Context on some decisions / open questions - I'm using `max_storage_buffers_per_shader_stage >= 3` as a check to see if storage buffers are supported. I was thinking about diving into 'binding resource management' but it feels like we don't have enough use cases to understand the problem yet, and it is mostly a separate concern to this PR, so I think it should be handled separately. - Should `ViewClusterBuffers` and `ViewClusterBindings` be merged, duplicating the count variables into the enum variants? Co-authored-by: Carter Anderson <mcanders1@gmail.com>
2022-04-07 16:16:35 +00:00
let spherical_polar_theta_phi = fibonacci_spiral_on_sphere(golden_ratio, i, N_LIGHTS);
let unit_sphere_p = spherical_polar_to_cartesian(spherical_polar_theta_phi);
Use `spawn_batch` in `many_lights` example (#11979) # Objective - The `many_lights` example uses a for-loop around `commands.spawn`. - It is generally recommended to use `spawn_batch` instead to lazily spawn entities, because it doesn't massively grow the command queue. ## Solution - Use `spawn_batch` in `many_lights` example. --- ## Discussion - `thread_rng` is called for each light spawned. This is a simple thread-local `Rc` clone, so it should compile down to a copy and an increment + decrement instruction. - I created `golden_ration` outside of the closure and `move`d it in. This should just be a copy and hopefully will get const-evaluated away. Would it be better to just move it into the closure itself? ## Performance Using `spawn_batch` seems to decrease time-to-first-`Update` by 0.1s: 1.3s to 1.2s. <details> <summary>Raw data and how it was collected.</summary> Before: - 2024-02-19T15:18:57.650987Z to 2024-02-19T15:18:58.912244Z : 1.3 - 2024-02-19T15:19:25.277135Z to 2024-02-19T15:19:26.542092Z : 1.3 - 2024-02-19T15:19:46.841460Z to 2024-02-19T15:19:48.137560Z : 1.3 After: - 2024-02-19T15:17:05.749521Z to 2024-02-19T15:17:06.993221Z : 1.2 - 2024-02-19T15:17:38.153049Z to 2024-02-19T15:17:39.393760Z : 1.2 - 2024-02-19T15:18:10.691562Z to 2024-02-19T15:18:11.891430Z : 1.2 To time performance, I tracked the time from the first `Startup` logged message to the first `Update` logged message. ```shell $ cargo run --release --example many_lights Compiling bevy v0.13.0 (/Users/bdeep/dev/bevy/bevy) Finished release [optimized] target(s) in 1.54s Running `target/release/examples/many_lights` # THIS TIME 2024-02-19T15:30:13.429609Z INFO bevy_render::renderer: AdapterInfo { name: "Apple M1", vendor: 0, device: 0, device_type: IntegratedGpu, driver: "", driver_info: "", backend: Metal } 2024-02-19T15:30:13.566856Z INFO bevy_winit::system: Creating new window "many_lights" (0v1) 2024-02-19T15:30:13.592371Z WARN many_lights: This is a stress test used to push Bevy to its limit and debug performance issues. It is not representative of an actual game. It must be run in release mode using --release or it will be very slow. 2024-02-19T15:30:13.592572Z INFO bevy_diagnostic::system_information_diagnostics_plugin::internal: SystemInfo { os: "MacOS 14.2.1 ", kernel: "23.2.0", cpu: "Apple M1", core_count: "8", memory: "16.0 GiB" } # TO THIS TIME 2024-02-19T15:30:15.429900Z INFO many_lights: Lights: 100000 2024-02-19T15:30:15.430139Z INFO bevy diagnostic: fps : 0.982693 (avg 43.026557) 2024-02-19T15:30:15.430157Z INFO bevy diagnostic: frame_time : 1017.611750ms (avg 149.456476ms) 2024-02-19T15:30:15.430165Z INFO bevy diagnostic: frame_count: 12.000000 (avg 6.000000) ``` </details>
2024-02-26 16:02:27 +00:00
(
PointLight {
Use storage buffers for clustered forward point lights (#3989) # Objective - Make use of storage buffers, where they are available, for clustered forward bindings to support far more point lights in a scene - Fixes #3605 - Based on top of #4079 This branch on an M1 Max can keep 60fps with about 2150 point lights of radius 1m in the Sponza scene where I've been testing. The bottleneck is mostly assigning lights to clusters which grows faster than linearly (I think 1000 lights was about 1.5ms and 5000 was 7.5ms). I have seen papers and presentations leveraging compute shaders that can get this up to over 1 million. That said, I think any further optimisations should probably be done in a separate PR. ## Solution - Add `RenderDevice` to the `Material` and `SpecializedMaterial` trait `::key()` functions to allow setting flags on the keys depending on feature/limit availability - Make `GpuPointLights` and `ViewClusterBuffers` into enums containing `UniformVec` and `StorageBuffer` variants. Implement the necessary API on them to make usage the same for both cases, and the only difference is at initialisation time. - Appropriate shader defs in the shader code to handle the two cases ## Context on some decisions / open questions - I'm using `max_storage_buffers_per_shader_stage >= 3` as a check to see if storage buffers are supported. I was thinking about diving into 'binding resource management' but it feels like we don't have enough use cases to understand the problem yet, and it is mostly a separate concern to this PR, so I think it should be handled separately. - Should `ViewClusterBuffers` and `ViewClusterBindings` be merged, duplicating the count variables into the enum variants? Co-authored-by: Carter Anderson <mcanders1@gmail.com>
2022-04-07 16:16:35 +00:00
range: LIGHT_RADIUS,
intensity: LIGHT_INTENSITY,
Migrate from `LegacyColor` to `bevy_color::Color` (#12163) # Objective - As part of the migration process we need to a) see the end effect of the migration on user ergonomics b) check for serious perf regressions c) actually migrate the code - To accomplish this, I'm going to attempt to migrate all of the remaining user-facing usages of `LegacyColor` in one PR, being careful to keep a clean commit history. - Fixes #12056. ## Solution I've chosen to use the polymorphic `Color` type as our standard user-facing API. - [x] Migrate `bevy_gizmos`. - [x] Take `impl Into<Color>` in all `bevy_gizmos` APIs - [x] Migrate sprites - [x] Migrate UI - [x] Migrate `ColorMaterial` - [x] Migrate `MaterialMesh2D` - [x] Migrate fog - [x] Migrate lights - [x] Migrate StandardMaterial - [x] Migrate wireframes - [x] Migrate clear color - [x] Migrate text - [x] Migrate gltf loader - [x] Register color types for reflection - [x] Remove `LegacyColor` - [x] Make sure CI passes Incidental improvements to ease migration: - added `Color::srgba_u8`, `Color::srgba_from_array` and friends - added `set_alpha`, `is_fully_transparent` and `is_fully_opaque` to the `Alpha` trait - add and immediately deprecate (lol) `Color::rgb` and friends in favor of more explicit and consistent `Color::srgb` - standardized on white and black for most example text colors - added vector field traits to `LinearRgba`: ~~`Add`, `Sub`, `AddAssign`, `SubAssign`,~~ `Mul<f32>` and `Div<f32>`. Multiplications and divisions do not scale alpha. `Add` and `Sub` have been cut from this PR. - added `LinearRgba` and `Srgba` `RED/GREEN/BLUE` - added `LinearRgba_to_f32_array` and `LinearRgba::to_u32` ## Migration Guide Bevy's color types have changed! Wherever you used a `bevy::render::Color`, a `bevy::color::Color` is used instead. These are quite similar! Both are enums storing a color in a specific color space (or to be more precise, using a specific color model). However, each of the different color models now has its own type. TODO... - `Color::rgba`, `Color::rgb`, `Color::rbga_u8`, `Color::rgb_u8`, `Color::rgb_from_array` are now `Color::srgba`, `Color::srgb`, `Color::srgba_u8`, `Color::srgb_u8` and `Color::srgb_from_array`. - `Color::set_a` and `Color::a` is now `Color::set_alpha` and `Color::alpha`. These are part of the `Alpha` trait in `bevy_color`. - `Color::is_fully_transparent` is now part of the `Alpha` trait in `bevy_color` - `Color::r`, `Color::set_r`, `Color::with_r` and the equivalents for `g`, `b` `h`, `s` and `l` have been removed due to causing silent relatively expensive conversions. Convert your `Color` into the desired color space, perform your operations there, and then convert it back into a polymorphic `Color` enum. - `Color::hex` is now `Srgba::hex`. Call `.into` or construct a `Color::Srgba` variant manually to convert it. - `WireframeMaterial`, `ExtractedUiNode`, `ExtractedDirectionalLight`, `ExtractedPointLight`, `ExtractedSpotLight` and `ExtractedSprite` now store a `LinearRgba`, rather than a polymorphic `Color` - `Color::rgb_linear` and `Color::rgba_linear` are now `Color::linear_rgb` and `Color::linear_rgba` - The various CSS color constants are no longer stored directly on `Color`. Instead, they're defined in the `Srgba` color space, and accessed via `bevy::color::palettes::css`. Call `.into()` on them to convert them into a `Color` for quick debugging use, and consider using the much prettier `tailwind` palette for prototyping. - The `LIME_GREEN` color has been renamed to `LIMEGREEN` to comply with the standard naming. - Vector field arithmetic operations on `Color` (add, subtract, multiply and divide by a f32) have been removed. Instead, convert your colors into `LinearRgba` space, and perform your operations explicitly there. This is particularly relevant when working with emissive or HDR colors, whose color channel values are routinely outside of the ordinary 0 to 1 range. - `Color::as_linear_rgba_f32` has been removed. Call `LinearRgba::to_f32_array` instead, converting if needed. - `Color::as_linear_rgba_u32` has been removed. Call `LinearRgba::to_u32` instead, converting if needed. - Several other color conversion methods to transform LCH or HSL colors into float arrays or `Vec` types have been removed. Please reimplement these externally or open a PR to re-add them if you found them particularly useful. - Various methods on `Color` such as `rgb` or `hsl` to convert the color into a specific color space have been removed. Convert into `LinearRgba`, then to the color space of your choice. - Various implicitly-converting color value methods on `Color` such as `r`, `g`, `b` or `h` have been removed. Please convert it into the color space of your choice, then check these properties. - `Color` no longer implements `AsBindGroup`. Store a `LinearRgba` internally instead to avoid conversion costs. --------- Co-authored-by: Alice Cecile <alice.i.cecil@gmail.com> Co-authored-by: Afonso Lage <lage.afonso@gmail.com> Co-authored-by: Rob Parrett <robparrett@gmail.com> Co-authored-by: Zachary Harrold <zac@harrold.com.au>
2024-02-29 19:35:12 +00:00
color: Color::hsl(rng.gen_range(0.0..360.0), 1.0, 0.5),
Use storage buffers for clustered forward point lights (#3989) # Objective - Make use of storage buffers, where they are available, for clustered forward bindings to support far more point lights in a scene - Fixes #3605 - Based on top of #4079 This branch on an M1 Max can keep 60fps with about 2150 point lights of radius 1m in the Sponza scene where I've been testing. The bottleneck is mostly assigning lights to clusters which grows faster than linearly (I think 1000 lights was about 1.5ms and 5000 was 7.5ms). I have seen papers and presentations leveraging compute shaders that can get this up to over 1 million. That said, I think any further optimisations should probably be done in a separate PR. ## Solution - Add `RenderDevice` to the `Material` and `SpecializedMaterial` trait `::key()` functions to allow setting flags on the keys depending on feature/limit availability - Make `GpuPointLights` and `ViewClusterBuffers` into enums containing `UniformVec` and `StorageBuffer` variants. Implement the necessary API on them to make usage the same for both cases, and the only difference is at initialisation time. - Appropriate shader defs in the shader code to handle the two cases ## Context on some decisions / open questions - I'm using `max_storage_buffers_per_shader_stage >= 3` as a check to see if storage buffers are supported. I was thinking about diving into 'binding resource management' but it feels like we don't have enough use cases to understand the problem yet, and it is mostly a separate concern to this PR, so I think it should be handled separately. - Should `ViewClusterBuffers` and `ViewClusterBindings` be merged, duplicating the count variables into the enum variants? Co-authored-by: Carter Anderson <mcanders1@gmail.com>
2022-04-07 16:16:35 +00:00
..default()
},
Transform::from_translation((RADIUS as f64 * unit_sphere_p).as_vec3()),
)
Use `spawn_batch` in `many_lights` example (#11979) # Objective - The `many_lights` example uses a for-loop around `commands.spawn`. - It is generally recommended to use `spawn_batch` instead to lazily spawn entities, because it doesn't massively grow the command queue. ## Solution - Use `spawn_batch` in `many_lights` example. --- ## Discussion - `thread_rng` is called for each light spawned. This is a simple thread-local `Rc` clone, so it should compile down to a copy and an increment + decrement instruction. - I created `golden_ration` outside of the closure and `move`d it in. This should just be a copy and hopefully will get const-evaluated away. Would it be better to just move it into the closure itself? ## Performance Using `spawn_batch` seems to decrease time-to-first-`Update` by 0.1s: 1.3s to 1.2s. <details> <summary>Raw data and how it was collected.</summary> Before: - 2024-02-19T15:18:57.650987Z to 2024-02-19T15:18:58.912244Z : 1.3 - 2024-02-19T15:19:25.277135Z to 2024-02-19T15:19:26.542092Z : 1.3 - 2024-02-19T15:19:46.841460Z to 2024-02-19T15:19:48.137560Z : 1.3 After: - 2024-02-19T15:17:05.749521Z to 2024-02-19T15:17:06.993221Z : 1.2 - 2024-02-19T15:17:38.153049Z to 2024-02-19T15:17:39.393760Z : 1.2 - 2024-02-19T15:18:10.691562Z to 2024-02-19T15:18:11.891430Z : 1.2 To time performance, I tracked the time from the first `Startup` logged message to the first `Update` logged message. ```shell $ cargo run --release --example many_lights Compiling bevy v0.13.0 (/Users/bdeep/dev/bevy/bevy) Finished release [optimized] target(s) in 1.54s Running `target/release/examples/many_lights` # THIS TIME 2024-02-19T15:30:13.429609Z INFO bevy_render::renderer: AdapterInfo { name: "Apple M1", vendor: 0, device: 0, device_type: IntegratedGpu, driver: "", driver_info: "", backend: Metal } 2024-02-19T15:30:13.566856Z INFO bevy_winit::system: Creating new window "many_lights" (0v1) 2024-02-19T15:30:13.592371Z WARN many_lights: This is a stress test used to push Bevy to its limit and debug performance issues. It is not representative of an actual game. It must be run in release mode using --release or it will be very slow. 2024-02-19T15:30:13.592572Z INFO bevy_diagnostic::system_information_diagnostics_plugin::internal: SystemInfo { os: "MacOS 14.2.1 ", kernel: "23.2.0", cpu: "Apple M1", core_count: "8", memory: "16.0 GiB" } # TO THIS TIME 2024-02-19T15:30:15.429900Z INFO many_lights: Lights: 100000 2024-02-19T15:30:15.430139Z INFO bevy diagnostic: fps : 0.982693 (avg 43.026557) 2024-02-19T15:30:15.430157Z INFO bevy diagnostic: frame_time : 1017.611750ms (avg 149.456476ms) 2024-02-19T15:30:15.430165Z INFO bevy diagnostic: frame_count: 12.000000 (avg 6.000000) ``` </details>
2024-02-26 16:02:27 +00:00
}));
Use storage buffers for clustered forward point lights (#3989) # Objective - Make use of storage buffers, where they are available, for clustered forward bindings to support far more point lights in a scene - Fixes #3605 - Based on top of #4079 This branch on an M1 Max can keep 60fps with about 2150 point lights of radius 1m in the Sponza scene where I've been testing. The bottleneck is mostly assigning lights to clusters which grows faster than linearly (I think 1000 lights was about 1.5ms and 5000 was 7.5ms). I have seen papers and presentations leveraging compute shaders that can get this up to over 1 million. That said, I think any further optimisations should probably be done in a separate PR. ## Solution - Add `RenderDevice` to the `Material` and `SpecializedMaterial` trait `::key()` functions to allow setting flags on the keys depending on feature/limit availability - Make `GpuPointLights` and `ViewClusterBuffers` into enums containing `UniformVec` and `StorageBuffer` variants. Implement the necessary API on them to make usage the same for both cases, and the only difference is at initialisation time. - Appropriate shader defs in the shader code to handle the two cases ## Context on some decisions / open questions - I'm using `max_storage_buffers_per_shader_stage >= 3` as a check to see if storage buffers are supported. I was thinking about diving into 'binding resource management' but it feels like we don't have enough use cases to understand the problem yet, and it is mostly a separate concern to this PR, so I think it should be handled separately. - Should `ViewClusterBuffers` and `ViewClusterBindings` be merged, duplicating the count variables into the enum variants? Co-authored-by: Carter Anderson <mcanders1@gmail.com>
2022-04-07 16:16:35 +00:00
// camera
match std::env::args().nth(1).as_deref() {
Some("orthographic") => commands.spawn((
Camera3d::default(),
Projection::from(OrthographicProjection {
scaling_mode: ScalingMode::FixedHorizontal(20.0),
Split OrthographicProjection::default into 2d & 3d (Adopted) (#15073) Adopted PR from dmlary, all credit to them! https://github.com/bevyengine/bevy/pull/9915 Original description: # Objective The default value for `near` in `OrthographicProjection` should be different for 2d & 3d. For 2d using `near = -1000` allows bevy users to build up scenes using background `z = 0`, and foreground elements `z > 0` similar to css. However in 3d `near = -1000` results in objects behind the camera being rendered. Using `near = 0` works for 3d, but forces 2d users to assign `z <= 0` for rendered elements, putting the background at some arbitrary negative value. There is no common value for `near` that doesn't result in a footgun or usability issue for either 2d or 3d, so they should have separate values. There was discussion about other options in the discord [0](https://discord.com/channels/691052431525675048/1154114310042292325), but splitting `default()` into `default_2d()` and `default_3d()` seemed like the lowest cost approach. Related/past work https://github.com/bevyengine/bevy/issues/9138, https://github.com/bevyengine/bevy/pull/9214, https://github.com/bevyengine/bevy/pull/9310, https://github.com/bevyengine/bevy/pull/9537 (thanks to @Selene-Amanita for the list) ## Solution This commit splits `OrthographicProjection::default` into `default_2d` and `default_3d`. ## Migration Guide - In initialization of `OrthographicProjection`, change `..default()` to `..OrthographicProjection::default_2d()` or `..OrthographicProjection::default_3d()` Example: ```diff --- a/examples/3d/orthographic.rs +++ b/examples/3d/orthographic.rs @@ -20,7 +20,7 @@ fn setup( projection: OrthographicProjection { scale: 3.0, scaling_mode: ScalingMode::FixedVertical(2.0), - ..default() + ..OrthographicProjection::default_3d() } .into(), transform: Transform::from_xyz(5.0, 5.0, 5.0).looking_at(Vec3::ZERO, Vec3::Y), ``` --------- Co-authored-by: David M. Lary <dmlary@gmail.com> Co-authored-by: Jan Hohenheim <jan@hohenheim.ch>
2024-09-09 15:51:28 +00:00
..OrthographicProjection::default_3d()
}),
)),
_ => commands.spawn(Camera3d::default()),
};
Use storage buffers for clustered forward point lights (#3989) # Objective - Make use of storage buffers, where they are available, for clustered forward bindings to support far more point lights in a scene - Fixes #3605 - Based on top of #4079 This branch on an M1 Max can keep 60fps with about 2150 point lights of radius 1m in the Sponza scene where I've been testing. The bottleneck is mostly assigning lights to clusters which grows faster than linearly (I think 1000 lights was about 1.5ms and 5000 was 7.5ms). I have seen papers and presentations leveraging compute shaders that can get this up to over 1 million. That said, I think any further optimisations should probably be done in a separate PR. ## Solution - Add `RenderDevice` to the `Material` and `SpecializedMaterial` trait `::key()` functions to allow setting flags on the keys depending on feature/limit availability - Make `GpuPointLights` and `ViewClusterBuffers` into enums containing `UniformVec` and `StorageBuffer` variants. Implement the necessary API on them to make usage the same for both cases, and the only difference is at initialisation time. - Appropriate shader defs in the shader code to handle the two cases ## Context on some decisions / open questions - I'm using `max_storage_buffers_per_shader_stage >= 3` as a check to see if storage buffers are supported. I was thinking about diving into 'binding resource management' but it feels like we don't have enough use cases to understand the problem yet, and it is mostly a separate concern to this PR, so I think it should be handled separately. - Should `ViewClusterBuffers` and `ViewClusterBindings` be merged, duplicating the count variables into the enum variants? Co-authored-by: Carter Anderson <mcanders1@gmail.com>
2022-04-07 16:16:35 +00:00
// add one cube, the only one with strong handles
// also serves as a reference point during rotation
Migrate meshes and materials to required components (#15524) # Objective A big step in the migration to required components: meshes and materials! ## Solution As per the [selected proposal](https://hackmd.io/@bevy/required_components/%2Fj9-PnF-2QKK0on1KQ29UWQ): - Deprecate `MaterialMesh2dBundle`, `MaterialMeshBundle`, and `PbrBundle`. - Add `Mesh2d` and `Mesh3d` components, which wrap a `Handle<Mesh>`. - Add `MeshMaterial2d<M: Material2d>` and `MeshMaterial3d<M: Material>`, which wrap a `Handle<M>`. - Meshes *without* a mesh material should be rendered with a default material. The existence of a material is determined by `HasMaterial2d`/`HasMaterial3d`, which is required by `MeshMaterial2d`/`MeshMaterial3d`. This gets around problems with the generics. Previously: ```rust commands.spawn(MaterialMesh2dBundle { mesh: meshes.add(Circle::new(100.0)).into(), material: materials.add(Color::srgb(7.5, 0.0, 7.5)), transform: Transform::from_translation(Vec3::new(-200., 0., 0.)), ..default() }); ``` Now: ```rust commands.spawn(( Mesh2d(meshes.add(Circle::new(100.0))), MeshMaterial2d(materials.add(Color::srgb(7.5, 0.0, 7.5))), Transform::from_translation(Vec3::new(-200., 0., 0.)), )); ``` If the mesh material is missing, previously nothing was rendered. Now, it renders a white default `ColorMaterial` in 2D and a `StandardMaterial` in 3D (this can be overridden). Below, only every other entity has a material: ![Näyttökuva 2024-09-29 181746](https://github.com/user-attachments/assets/5c8be029-d2fe-4b8c-ae89-17a72ff82c9a) ![Näyttökuva 2024-09-29 181918](https://github.com/user-attachments/assets/58adbc55-5a1e-4c7d-a2c7-ed456227b909) Why white? This is still open for discussion, but I think white makes sense for a *default* material, while *invalid* asset handles pointing to nothing should have something like a pink material to indicate that something is broken (I don't handle that in this PR yet). This is kind of a mix of Godot and Unity: Godot just renders a white material for non-existent materials, while Unity renders nothing when no materials exist, but renders pink for invalid materials. I can also change the default material to pink if that is preferable though. ## Testing I ran some 2D and 3D examples to test if anything changed visually. I have not tested all examples or features yet however. If anyone wants to test more extensively, it would be appreciated! ## Implementation Notes - The relationship between `bevy_render` and `bevy_pbr` is weird here. `bevy_render` needs `Mesh3d` for its own systems, but `bevy_pbr` has all of the material logic, and `bevy_render` doesn't depend on it. I feel like the two crates should be refactored in some way, but I think that's out of scope for this PR. - I didn't migrate meshlets to required components yet. That can probably be done in a follow-up, as this is already a huge PR. - It is becoming increasingly clear to me that we really, *really* want to disallow raw asset handles as components. They caused me a *ton* of headache here already, and it took me a long time to find every place that queried for them or inserted them directly on entities, since there were no compiler errors for it. If we don't remove the `Component` derive, I expect raw asset handles to be a *huge* footgun for users as we transition to wrapper components, especially as handles as components have been the norm so far. I personally consider this to be a blocker for 0.15: we need to migrate to wrapper components for asset handles everywhere, and remove the `Component` derive. Also see https://github.com/bevyengine/bevy/issues/14124. --- ## Migration Guide Asset handles for meshes and mesh materials must now be wrapped in the `Mesh2d` and `MeshMaterial2d` or `Mesh3d` and `MeshMaterial3d` components for 2D and 3D respectively. Raw handles as components no longer render meshes. Additionally, `MaterialMesh2dBundle`, `MaterialMeshBundle`, and `PbrBundle` have been deprecated. Instead, use the mesh and material components directly. Previously: ```rust commands.spawn(MaterialMesh2dBundle { mesh: meshes.add(Circle::new(100.0)).into(), material: materials.add(Color::srgb(7.5, 0.0, 7.5)), transform: Transform::from_translation(Vec3::new(-200., 0., 0.)), ..default() }); ``` Now: ```rust commands.spawn(( Mesh2d(meshes.add(Circle::new(100.0))), MeshMaterial2d(materials.add(Color::srgb(7.5, 0.0, 7.5))), Transform::from_translation(Vec3::new(-200., 0., 0.)), )); ``` If the mesh material is missing, a white default material is now used. Previously, nothing was rendered if the material was missing. The `WithMesh2d` and `WithMesh3d` query filter type aliases have also been removed. Simply use `With<Mesh2d>` or `With<Mesh3d>`. --------- Co-authored-by: Tim Blackbird <justthecooldude@gmail.com> Co-authored-by: Carter Anderson <mcanders1@gmail.com>
2024-10-01 21:33:17 +00:00
commands.spawn((
Mesh3d(mesh),
MeshMaterial3d(material),
Transform {
translation: Vec3::new(0.0, RADIUS, 0.0),
Use storage buffers for clustered forward point lights (#3989) # Objective - Make use of storage buffers, where they are available, for clustered forward bindings to support far more point lights in a scene - Fixes #3605 - Based on top of #4079 This branch on an M1 Max can keep 60fps with about 2150 point lights of radius 1m in the Sponza scene where I've been testing. The bottleneck is mostly assigning lights to clusters which grows faster than linearly (I think 1000 lights was about 1.5ms and 5000 was 7.5ms). I have seen papers and presentations leveraging compute shaders that can get this up to over 1 million. That said, I think any further optimisations should probably be done in a separate PR. ## Solution - Add `RenderDevice` to the `Material` and `SpecializedMaterial` trait `::key()` functions to allow setting flags on the keys depending on feature/limit availability - Make `GpuPointLights` and `ViewClusterBuffers` into enums containing `UniformVec` and `StorageBuffer` variants. Implement the necessary API on them to make usage the same for both cases, and the only difference is at initialisation time. - Appropriate shader defs in the shader code to handle the two cases ## Context on some decisions / open questions - I'm using `max_storage_buffers_per_shader_stage >= 3` as a check to see if storage buffers are supported. I was thinking about diving into 'binding resource management' but it feels like we don't have enough use cases to understand the problem yet, and it is mostly a separate concern to this PR, so I think it should be handled separately. - Should `ViewClusterBuffers` and `ViewClusterBindings` be merged, duplicating the count variables into the enum variants? Co-authored-by: Carter Anderson <mcanders1@gmail.com>
2022-04-07 16:16:35 +00:00
scale: Vec3::splat(5.0),
..default()
},
Migrate meshes and materials to required components (#15524) # Objective A big step in the migration to required components: meshes and materials! ## Solution As per the [selected proposal](https://hackmd.io/@bevy/required_components/%2Fj9-PnF-2QKK0on1KQ29UWQ): - Deprecate `MaterialMesh2dBundle`, `MaterialMeshBundle`, and `PbrBundle`. - Add `Mesh2d` and `Mesh3d` components, which wrap a `Handle<Mesh>`. - Add `MeshMaterial2d<M: Material2d>` and `MeshMaterial3d<M: Material>`, which wrap a `Handle<M>`. - Meshes *without* a mesh material should be rendered with a default material. The existence of a material is determined by `HasMaterial2d`/`HasMaterial3d`, which is required by `MeshMaterial2d`/`MeshMaterial3d`. This gets around problems with the generics. Previously: ```rust commands.spawn(MaterialMesh2dBundle { mesh: meshes.add(Circle::new(100.0)).into(), material: materials.add(Color::srgb(7.5, 0.0, 7.5)), transform: Transform::from_translation(Vec3::new(-200., 0., 0.)), ..default() }); ``` Now: ```rust commands.spawn(( Mesh2d(meshes.add(Circle::new(100.0))), MeshMaterial2d(materials.add(Color::srgb(7.5, 0.0, 7.5))), Transform::from_translation(Vec3::new(-200., 0., 0.)), )); ``` If the mesh material is missing, previously nothing was rendered. Now, it renders a white default `ColorMaterial` in 2D and a `StandardMaterial` in 3D (this can be overridden). Below, only every other entity has a material: ![Näyttökuva 2024-09-29 181746](https://github.com/user-attachments/assets/5c8be029-d2fe-4b8c-ae89-17a72ff82c9a) ![Näyttökuva 2024-09-29 181918](https://github.com/user-attachments/assets/58adbc55-5a1e-4c7d-a2c7-ed456227b909) Why white? This is still open for discussion, but I think white makes sense for a *default* material, while *invalid* asset handles pointing to nothing should have something like a pink material to indicate that something is broken (I don't handle that in this PR yet). This is kind of a mix of Godot and Unity: Godot just renders a white material for non-existent materials, while Unity renders nothing when no materials exist, but renders pink for invalid materials. I can also change the default material to pink if that is preferable though. ## Testing I ran some 2D and 3D examples to test if anything changed visually. I have not tested all examples or features yet however. If anyone wants to test more extensively, it would be appreciated! ## Implementation Notes - The relationship between `bevy_render` and `bevy_pbr` is weird here. `bevy_render` needs `Mesh3d` for its own systems, but `bevy_pbr` has all of the material logic, and `bevy_render` doesn't depend on it. I feel like the two crates should be refactored in some way, but I think that's out of scope for this PR. - I didn't migrate meshlets to required components yet. That can probably be done in a follow-up, as this is already a huge PR. - It is becoming increasingly clear to me that we really, *really* want to disallow raw asset handles as components. They caused me a *ton* of headache here already, and it took me a long time to find every place that queried for them or inserted them directly on entities, since there were no compiler errors for it. If we don't remove the `Component` derive, I expect raw asset handles to be a *huge* footgun for users as we transition to wrapper components, especially as handles as components have been the norm so far. I personally consider this to be a blocker for 0.15: we need to migrate to wrapper components for asset handles everywhere, and remove the `Component` derive. Also see https://github.com/bevyengine/bevy/issues/14124. --- ## Migration Guide Asset handles for meshes and mesh materials must now be wrapped in the `Mesh2d` and `MeshMaterial2d` or `Mesh3d` and `MeshMaterial3d` components for 2D and 3D respectively. Raw handles as components no longer render meshes. Additionally, `MaterialMesh2dBundle`, `MaterialMeshBundle`, and `PbrBundle` have been deprecated. Instead, use the mesh and material components directly. Previously: ```rust commands.spawn(MaterialMesh2dBundle { mesh: meshes.add(Circle::new(100.0)).into(), material: materials.add(Color::srgb(7.5, 0.0, 7.5)), transform: Transform::from_translation(Vec3::new(-200., 0., 0.)), ..default() }); ``` Now: ```rust commands.spawn(( Mesh2d(meshes.add(Circle::new(100.0))), MeshMaterial2d(materials.add(Color::srgb(7.5, 0.0, 7.5))), Transform::from_translation(Vec3::new(-200., 0., 0.)), )); ``` If the mesh material is missing, a white default material is now used. Previously, nothing was rendered if the material was missing. The `WithMesh2d` and `WithMesh3d` query filter type aliases have also been removed. Simply use `With<Mesh2d>` or `With<Mesh3d>`. --------- Co-authored-by: Tim Blackbird <justthecooldude@gmail.com> Co-authored-by: Carter Anderson <mcanders1@gmail.com>
2024-10-01 21:33:17 +00:00
));
Use storage buffers for clustered forward point lights (#3989) # Objective - Make use of storage buffers, where they are available, for clustered forward bindings to support far more point lights in a scene - Fixes #3605 - Based on top of #4079 This branch on an M1 Max can keep 60fps with about 2150 point lights of radius 1m in the Sponza scene where I've been testing. The bottleneck is mostly assigning lights to clusters which grows faster than linearly (I think 1000 lights was about 1.5ms and 5000 was 7.5ms). I have seen papers and presentations leveraging compute shaders that can get this up to over 1 million. That said, I think any further optimisations should probably be done in a separate PR. ## Solution - Add `RenderDevice` to the `Material` and `SpecializedMaterial` trait `::key()` functions to allow setting flags on the keys depending on feature/limit availability - Make `GpuPointLights` and `ViewClusterBuffers` into enums containing `UniformVec` and `StorageBuffer` variants. Implement the necessary API on them to make usage the same for both cases, and the only difference is at initialisation time. - Appropriate shader defs in the shader code to handle the two cases ## Context on some decisions / open questions - I'm using `max_storage_buffers_per_shader_stage >= 3` as a check to see if storage buffers are supported. I was thinking about diving into 'binding resource management' but it feels like we don't have enough use cases to understand the problem yet, and it is mostly a separate concern to this PR, so I think it should be handled separately. - Should `ViewClusterBuffers` and `ViewClusterBindings` be merged, duplicating the count variables into the enum variants? Co-authored-by: Carter Anderson <mcanders1@gmail.com>
2022-04-07 16:16:35 +00:00
}
// NOTE: This epsilon value is apparently optimal for optimizing for the average
// nearest-neighbor distance. See:
// http://extremelearning.com.au/how-to-evenly-distribute-points-on-a-sphere-more-effectively-than-the-canonical-fibonacci-lattice/
// for details.
const EPSILON: f64 = 0.36;
fn fibonacci_spiral_on_sphere(golden_ratio: f64, i: usize, n: usize) -> DVec2 {
DVec2::new(
PI * 2. * (i as f64 / golden_ratio),
ops::acos((1.0 - 2.0 * (i as f64 + EPSILON) / (n as f64 - 1.0 + 2.0 * EPSILON)) as f32)
as f64,
Use storage buffers for clustered forward point lights (#3989) # Objective - Make use of storage buffers, where they are available, for clustered forward bindings to support far more point lights in a scene - Fixes #3605 - Based on top of #4079 This branch on an M1 Max can keep 60fps with about 2150 point lights of radius 1m in the Sponza scene where I've been testing. The bottleneck is mostly assigning lights to clusters which grows faster than linearly (I think 1000 lights was about 1.5ms and 5000 was 7.5ms). I have seen papers and presentations leveraging compute shaders that can get this up to over 1 million. That said, I think any further optimisations should probably be done in a separate PR. ## Solution - Add `RenderDevice` to the `Material` and `SpecializedMaterial` trait `::key()` functions to allow setting flags on the keys depending on feature/limit availability - Make `GpuPointLights` and `ViewClusterBuffers` into enums containing `UniformVec` and `StorageBuffer` variants. Implement the necessary API on them to make usage the same for both cases, and the only difference is at initialisation time. - Appropriate shader defs in the shader code to handle the two cases ## Context on some decisions / open questions - I'm using `max_storage_buffers_per_shader_stage >= 3` as a check to see if storage buffers are supported. I was thinking about diving into 'binding resource management' but it feels like we don't have enough use cases to understand the problem yet, and it is mostly a separate concern to this PR, so I think it should be handled separately. - Should `ViewClusterBuffers` and `ViewClusterBindings` be merged, duplicating the count variables into the enum variants? Co-authored-by: Carter Anderson <mcanders1@gmail.com>
2022-04-07 16:16:35 +00:00
)
}
fn spherical_polar_to_cartesian(p: DVec2) -> DVec3 {
let (sin_theta, cos_theta) = p.x.sin_cos();
let (sin_phi, cos_phi) = p.y.sin_cos();
DVec3::new(cos_theta * sin_phi, sin_theta * sin_phi, cos_phi)
}
// System for rotating the camera
fn move_camera(time: Res<Time>, mut camera_query: Query<&mut Transform, With<Camera>>) {
let mut camera_transform = camera_query.single_mut();
let delta = time.delta_seconds() * 0.15;
camera_transform.rotate_z(delta);
camera_transform.rotate_x(delta);
Use storage buffers for clustered forward point lights (#3989) # Objective - Make use of storage buffers, where they are available, for clustered forward bindings to support far more point lights in a scene - Fixes #3605 - Based on top of #4079 This branch on an M1 Max can keep 60fps with about 2150 point lights of radius 1m in the Sponza scene where I've been testing. The bottleneck is mostly assigning lights to clusters which grows faster than linearly (I think 1000 lights was about 1.5ms and 5000 was 7.5ms). I have seen papers and presentations leveraging compute shaders that can get this up to over 1 million. That said, I think any further optimisations should probably be done in a separate PR. ## Solution - Add `RenderDevice` to the `Material` and `SpecializedMaterial` trait `::key()` functions to allow setting flags on the keys depending on feature/limit availability - Make `GpuPointLights` and `ViewClusterBuffers` into enums containing `UniformVec` and `StorageBuffer` variants. Implement the necessary API on them to make usage the same for both cases, and the only difference is at initialisation time. - Appropriate shader defs in the shader code to handle the two cases ## Context on some decisions / open questions - I'm using `max_storage_buffers_per_shader_stage >= 3` as a check to see if storage buffers are supported. I was thinking about diving into 'binding resource management' but it feels like we don't have enough use cases to understand the problem yet, and it is mostly a separate concern to this PR, so I think it should be handled separately. - Should `ViewClusterBuffers` and `ViewClusterBindings` be merged, duplicating the count variables into the enum variants? Co-authored-by: Carter Anderson <mcanders1@gmail.com>
2022-04-07 16:16:35 +00:00
}
// System for printing the number of meshes on every tick of the timer
fn print_light_count(time: Res<Time>, mut timer: Local<PrintingTimer>, lights: Query<&PointLight>) {
timer.0.tick(time.delta());
if timer.0.just_finished() {
info!("Lights: {}", lights.iter().len());
Use storage buffers for clustered forward point lights (#3989) # Objective - Make use of storage buffers, where they are available, for clustered forward bindings to support far more point lights in a scene - Fixes #3605 - Based on top of #4079 This branch on an M1 Max can keep 60fps with about 2150 point lights of radius 1m in the Sponza scene where I've been testing. The bottleneck is mostly assigning lights to clusters which grows faster than linearly (I think 1000 lights was about 1.5ms and 5000 was 7.5ms). I have seen papers and presentations leveraging compute shaders that can get this up to over 1 million. That said, I think any further optimisations should probably be done in a separate PR. ## Solution - Add `RenderDevice` to the `Material` and `SpecializedMaterial` trait `::key()` functions to allow setting flags on the keys depending on feature/limit availability - Make `GpuPointLights` and `ViewClusterBuffers` into enums containing `UniformVec` and `StorageBuffer` variants. Implement the necessary API on them to make usage the same for both cases, and the only difference is at initialisation time. - Appropriate shader defs in the shader code to handle the two cases ## Context on some decisions / open questions - I'm using `max_storage_buffers_per_shader_stage >= 3` as a check to see if storage buffers are supported. I was thinking about diving into 'binding resource management' but it feels like we don't have enough use cases to understand the problem yet, and it is mostly a separate concern to this PR, so I think it should be handled separately. - Should `ViewClusterBuffers` and `ViewClusterBindings` be merged, duplicating the count variables into the enum variants? Co-authored-by: Carter Anderson <mcanders1@gmail.com>
2022-04-07 16:16:35 +00:00
}
}
struct LogVisibleLights;
impl Plugin for LogVisibleLights {
fn build(&self, app: &mut App) {
let Some(render_app) = app.get_sub_app_mut(RenderApp) else {
return;
Use storage buffers for clustered forward point lights (#3989) # Objective - Make use of storage buffers, where they are available, for clustered forward bindings to support far more point lights in a scene - Fixes #3605 - Based on top of #4079 This branch on an M1 Max can keep 60fps with about 2150 point lights of radius 1m in the Sponza scene where I've been testing. The bottleneck is mostly assigning lights to clusters which grows faster than linearly (I think 1000 lights was about 1.5ms and 5000 was 7.5ms). I have seen papers and presentations leveraging compute shaders that can get this up to over 1 million. That said, I think any further optimisations should probably be done in a separate PR. ## Solution - Add `RenderDevice` to the `Material` and `SpecializedMaterial` trait `::key()` functions to allow setting flags on the keys depending on feature/limit availability - Make `GpuPointLights` and `ViewClusterBuffers` into enums containing `UniformVec` and `StorageBuffer` variants. Implement the necessary API on them to make usage the same for both cases, and the only difference is at initialisation time. - Appropriate shader defs in the shader code to handle the two cases ## Context on some decisions / open questions - I'm using `max_storage_buffers_per_shader_stage >= 3` as a check to see if storage buffers are supported. I was thinking about diving into 'binding resource management' but it feels like we don't have enough use cases to understand the problem yet, and it is mostly a separate concern to this PR, so I think it should be handled separately. - Should `ViewClusterBuffers` and `ViewClusterBindings` be merged, duplicating the count variables into the enum variants? Co-authored-by: Carter Anderson <mcanders1@gmail.com>
2022-04-07 16:16:35 +00:00
};
render_app.add_systems(Render, print_visible_light_count.in_set(RenderSet::Prepare));
Use storage buffers for clustered forward point lights (#3989) # Objective - Make use of storage buffers, where they are available, for clustered forward bindings to support far more point lights in a scene - Fixes #3605 - Based on top of #4079 This branch on an M1 Max can keep 60fps with about 2150 point lights of radius 1m in the Sponza scene where I've been testing. The bottleneck is mostly assigning lights to clusters which grows faster than linearly (I think 1000 lights was about 1.5ms and 5000 was 7.5ms). I have seen papers and presentations leveraging compute shaders that can get this up to over 1 million. That said, I think any further optimisations should probably be done in a separate PR. ## Solution - Add `RenderDevice` to the `Material` and `SpecializedMaterial` trait `::key()` functions to allow setting flags on the keys depending on feature/limit availability - Make `GpuPointLights` and `ViewClusterBuffers` into enums containing `UniformVec` and `StorageBuffer` variants. Implement the necessary API on them to make usage the same for both cases, and the only difference is at initialisation time. - Appropriate shader defs in the shader code to handle the two cases ## Context on some decisions / open questions - I'm using `max_storage_buffers_per_shader_stage >= 3` as a check to see if storage buffers are supported. I was thinking about diving into 'binding resource management' but it feels like we don't have enough use cases to understand the problem yet, and it is mostly a separate concern to this PR, so I think it should be handled separately. - Should `ViewClusterBuffers` and `ViewClusterBindings` be merged, duplicating the count variables into the enum variants? Co-authored-by: Carter Anderson <mcanders1@gmail.com>
2022-04-07 16:16:35 +00:00
}
}
// System for printing the number of meshes on every tick of the timer
fn print_visible_light_count(
time: Res<Time>,
Use storage buffers for clustered forward point lights (#3989) # Objective - Make use of storage buffers, where they are available, for clustered forward bindings to support far more point lights in a scene - Fixes #3605 - Based on top of #4079 This branch on an M1 Max can keep 60fps with about 2150 point lights of radius 1m in the Sponza scene where I've been testing. The bottleneck is mostly assigning lights to clusters which grows faster than linearly (I think 1000 lights was about 1.5ms and 5000 was 7.5ms). I have seen papers and presentations leveraging compute shaders that can get this up to over 1 million. That said, I think any further optimisations should probably be done in a separate PR. ## Solution - Add `RenderDevice` to the `Material` and `SpecializedMaterial` trait `::key()` functions to allow setting flags on the keys depending on feature/limit availability - Make `GpuPointLights` and `ViewClusterBuffers` into enums containing `UniformVec` and `StorageBuffer` variants. Implement the necessary API on them to make usage the same for both cases, and the only difference is at initialisation time. - Appropriate shader defs in the shader code to handle the two cases ## Context on some decisions / open questions - I'm using `max_storage_buffers_per_shader_stage >= 3` as a check to see if storage buffers are supported. I was thinking about diving into 'binding resource management' but it feels like we don't have enough use cases to understand the problem yet, and it is mostly a separate concern to this PR, so I think it should be handled separately. - Should `ViewClusterBuffers` and `ViewClusterBindings` be merged, duplicating the count variables into the enum variants? Co-authored-by: Carter Anderson <mcanders1@gmail.com>
2022-04-07 16:16:35 +00:00
mut timer: Local<PrintingTimer>,
visible: Query<&ExtractedPointLight>,
global_light_meta: Res<GlobalClusterableObjectMeta>,
Use storage buffers for clustered forward point lights (#3989) # Objective - Make use of storage buffers, where they are available, for clustered forward bindings to support far more point lights in a scene - Fixes #3605 - Based on top of #4079 This branch on an M1 Max can keep 60fps with about 2150 point lights of radius 1m in the Sponza scene where I've been testing. The bottleneck is mostly assigning lights to clusters which grows faster than linearly (I think 1000 lights was about 1.5ms and 5000 was 7.5ms). I have seen papers and presentations leveraging compute shaders that can get this up to over 1 million. That said, I think any further optimisations should probably be done in a separate PR. ## Solution - Add `RenderDevice` to the `Material` and `SpecializedMaterial` trait `::key()` functions to allow setting flags on the keys depending on feature/limit availability - Make `GpuPointLights` and `ViewClusterBuffers` into enums containing `UniformVec` and `StorageBuffer` variants. Implement the necessary API on them to make usage the same for both cases, and the only difference is at initialisation time. - Appropriate shader defs in the shader code to handle the two cases ## Context on some decisions / open questions - I'm using `max_storage_buffers_per_shader_stage >= 3` as a check to see if storage buffers are supported. I was thinking about diving into 'binding resource management' but it feels like we don't have enough use cases to understand the problem yet, and it is mostly a separate concern to this PR, so I think it should be handled separately. - Should `ViewClusterBuffers` and `ViewClusterBindings` be merged, duplicating the count variables into the enum variants? Co-authored-by: Carter Anderson <mcanders1@gmail.com>
2022-04-07 16:16:35 +00:00
) {
timer.0.tick(time.delta());
if timer.0.just_finished() {
info!(
"Visible Lights: {}, Rendered Lights: {}",
visible.iter().len(),
global_light_meta.entity_to_index.len()
);
}
}
struct PrintingTimer(Timer);
impl Default for PrintingTimer {
fn default() -> Self {
Self(Timer::from_seconds(1.0, TimerMode::Repeating))
Use storage buffers for clustered forward point lights (#3989) # Objective - Make use of storage buffers, where they are available, for clustered forward bindings to support far more point lights in a scene - Fixes #3605 - Based on top of #4079 This branch on an M1 Max can keep 60fps with about 2150 point lights of radius 1m in the Sponza scene where I've been testing. The bottleneck is mostly assigning lights to clusters which grows faster than linearly (I think 1000 lights was about 1.5ms and 5000 was 7.5ms). I have seen papers and presentations leveraging compute shaders that can get this up to over 1 million. That said, I think any further optimisations should probably be done in a separate PR. ## Solution - Add `RenderDevice` to the `Material` and `SpecializedMaterial` trait `::key()` functions to allow setting flags on the keys depending on feature/limit availability - Make `GpuPointLights` and `ViewClusterBuffers` into enums containing `UniformVec` and `StorageBuffer` variants. Implement the necessary API on them to make usage the same for both cases, and the only difference is at initialisation time. - Appropriate shader defs in the shader code to handle the two cases ## Context on some decisions / open questions - I'm using `max_storage_buffers_per_shader_stage >= 3` as a check to see if storage buffers are supported. I was thinking about diving into 'binding resource management' but it feels like we don't have enough use cases to understand the problem yet, and it is mostly a separate concern to this PR, so I think it should be handled separately. - Should `ViewClusterBuffers` and `ViewClusterBindings` be merged, duplicating the count variables into the enum variants? Co-authored-by: Carter Anderson <mcanders1@gmail.com>
2022-04-07 16:16:35 +00:00
}
}