This commit implements opt-in GPU frustum culling, built on top of the
infrastructure in https://github.com/bevyengine/bevy/pull/12773. To
enable it on a camera, add the `GpuCulling` component to it. To
additionally disable CPU frustum culling, add the `NoCpuCulling`
component. Note that adding `GpuCulling` without `NoCpuCulling`
*currently* does nothing useful. The reason why `GpuCulling` doesn't
automatically imply `NoCpuCulling` is that I intend to follow this patch
up with GPU two-phase occlusion culling, and CPU frustum culling plus
GPU occlusion culling seems like a very commonly-desired mode.
Adding the `GpuCulling` component to a view puts that view into
*indirect mode*. This mode makes all drawcalls indirect, relying on the
mesh preprocessing shader to allocate instances dynamically. In indirect
mode, the `PreprocessWorkItem` `output_index` points not to a
`MeshUniform` instance slot but instead to a set of `wgpu`
`IndirectParameters`, from which it allocates an instance slot
dynamically if frustum culling succeeds. Batch building has been updated
to allocate and track indirect parameter slots, and the AABBs are now
supplied to the GPU as `MeshCullingData`.
A small amount of code relating to the frustum culling has been borrowed
from meshlets and moved into `maths.wgsl`. Note that standard Bevy
frustum culling uses AABBs, while meshlets use bounding spheres; this
means that not as much code can be shared as one might think.
This patch doesn't provide any way to perform GPU culling on shadow
maps, to avoid making this patch bigger than it already is. That can be
a followup.
## Changelog
### Added
* Frustum culling can now optionally be done on the GPU. To enable it,
add the `GpuCulling` component to a camera.
* To disable CPU frustum culling, add `NoCpuCulling` to a camera. Note
that `GpuCulling` doesn't automatically imply `NoCpuCulling`.
# Objective
- There is an unfortunate lack of dragons in the meshlet docs.
- Dragons are symbolic of majesty, power, storms, and meshlets.
- A dragon habitat such as our docs requires cultivation to ensure each
winged lizard reaches their fullest, fiery selves.
## Solution
- Fix the link to the dragon image.
- The link originally targeted the `meshlet` branch, but that was later
deleted after it was merged into `main`.
---
## Changelog
- Added a dragon back into the `MeshletPlugin` documentation.
Keeping track of explicit visibility per cluster between frames does not
work with LODs, and leads to worse culling (using the final depth buffer
from the previous frame is more accurate).
Instead, we need to generate a second depth pyramid after the second
raster pass, and then use that in the first culling pass in the next
frame to test if a cluster would have been visible last frame or not.
As part of these changes, the write_index_buffer pass has been folded
into the culling pass for a large performance gain, and to avoid
tracking a lot of extra state that would be needed between passes.
Prepass previous model/view stuff was adapted to work with meshlets as
well.
Also fixed a bug with materials, and other misc improvements.
---------
Co-authored-by: François <mockersf@gmail.com>
Co-authored-by: atlas dostal <rodol@rivalrebels.com>
Co-authored-by: vero <email@atlasdostal.com>
Co-authored-by: Patrick Walton <pcwalton@mimiga.net>
Co-authored-by: Robert Swain <robert.swain@gmail.com>
# Objective
Fix https://github.com/bevyengine/bevy/issues/11799 and improve
`CameraProjectionPlugin`
## Solution
`CameraProjectionPlugin` is now an all-in-one plugin for adding a custom
`CameraProjection`. I also added `PbrProjectionPlugin` which is like
`CameraProjectionPlugin` but for PBR.
P.S. I'd like to get this merged after
https://github.com/bevyengine/bevy/pull/11766.
---
## Changelog
- Changed `CameraProjectionPlugin` to be an all-in-one plugin for adding
a `CameraProjection`
- Removed `VisibilitySystems::{UpdateOrthographicFrusta,
UpdatePerspectiveFrusta, UpdateProjectionFrusta}`, now replaced with
`VisibilitySystems::UpdateFrusta`
- Added `PbrProjectionPlugin` for projection-specific PBR functionality.
## Migration Guide
`VisibilitySystems`'s `UpdateOrthographicFrusta`,
`UpdatePerspectiveFrusta`, and `UpdateProjectionFrusta` variants were
removed, they were replaced with `VisibilitySystems::UpdateFrusta`
# Objective
- clean up extract_mesh_(gpu/cpu)_building
## Solution
- gpu_building no need to hold `prev_render_mesh_instances`
- using `insert_unique_unchecked` instead of simple insert as we know
all entities are unique
- direcly get `previous_input_index ` in par_loop
## Performance
this should also bring a slight performance win.
cargo run --release --example many_cubes --features bevy/trace_tracy --
--no-frustum-culling
`extract_meshes_for_gpu_building`
---------
Co-authored-by: Patrick Walton <pcwalton@mimiga.net>
# Objective
- bevy usually use `Parallel::scope` to collect items from `par_iter`,
but `scope` will be called with every satifified items. it will cause a
lot of unnecessary lookup.
## Solution
- similar to Rayon ,we introduce `for_each_init` for `par_iter` which
only be invoked when spawn a task for a group of items.
---
## Changelog
- added `for_each_init`
## Performance
`check_visibility ` in `many_foxes `
~40% performance gain in `check_visibility`.
---------
Co-authored-by: James Liu <contact@jamessliu.com>
# Objective
- `MeshPipelineKey` use some bits for two things
- First commit in this PR adds an assertion that doesn't work currently
on main
- This leads to some mesh topology not working anymore, for example
`LineStrip`
- With examples `lines`, there should be two groups of lines, the blue
one doesn't display currently
## Solution
- Change the `MeshPipelineKey` to be backed by a `u64` instead, to have
enough bits
# Objective
- The docs says the WireframeColor is supposed to override the default
global color but it doesn't.
## Solution
- Use WireframeColor to override global color like docs said it was
supposed to do.
- Updated the example to document this feature
- I also took the opportunity to clean up the code a bit
Fixes#13032
# Objective
Make visibility system ordering explicit. Fixes#12953.
## Solution
Specify `CheckVisibility` happens after all other `VisibilitySystems`
sets have happened.
---------
Co-authored-by: Elabajaba <Elabajaba@users.noreply.github.com>
# Objective
When trying to be generic over `Material + Default`, the lack of a
`Default` impl for `ExtendedMaterial`, even when both of its components
implement `Default`, is problematic. I think was probably just
overlooked.
## Solution
Impl `Default` if the material and extension both impl `Default`.
---
## Changelog
## Migration Guide
glTF files that contain lights currently panic when loaded into Bevy,
because Bevy tries to reflect on `Cascades`, which accidentally wasn't
registered.
# Objective
- Fixes#12976
## Solution
This one is a doozy.
- Run `cargo +beta clippy --workspace --all-targets --all-features` and
fix all issues
- This includes:
- Moving inner attributes to be outer attributes, when the item in
question has both inner and outer attributes
- Use `ptr::from_ref` in more scenarios
- Extend the valid idents list used by `clippy:doc_markdown` with more
names
- Use `Clone::clone_from` when possible
- Remove redundant `ron` import
- Add backticks to **so many** identifiers and items
- I'm sorry whoever has to review this
---
## Changelog
- Added links to more identifiers in documentation.
[Alpha to coverage] (A2C) replaces alpha blending with a
hardware-specific multisample coverage mask when multisample
antialiasing is in use. It's a simple form of [order-independent
transparency] that relies on MSAA. ["Anti-aliased Alpha Test: The
Esoteric Alpha To Coverage"] is a good summary of the motivation for and
best practices relating to A2C.
This commit implements alpha to coverage support as a new variant for
`AlphaMode`. You can supply `AlphaMode::AlphaToCoverage` as the
`alpha_mode` field in `StandardMaterial` to use it. When in use, the
standard material shader automatically applies the texture filtering
method from ["Anti-aliased Alpha Test: The Esoteric Alpha To Coverage"].
Objects with alpha-to-coverage materials are binned in the opaque pass,
as they're fully order-independent.
The `transparency_3d` example has been updated to feature an object with
alpha to coverage. Happily, the example was already using MSAA.
This is part of #2223, as far as I can tell.
[Alpha to coverage]: https://en.wikipedia.org/wiki/Alpha_to_coverage
[order-independent transparency]:
https://en.wikipedia.org/wiki/Order-independent_transparency
["Anti-aliased Alpha Test: The Esoteric Alpha To Coverage"]:
https://bgolus.medium.com/anti-aliased-alpha-test-the-esoteric-alpha-to-coverage-8b177335ae4f
---
## Changelog
### Added
* The `AlphaMode` enum now supports `AlphaToCoverage`, to provide
limited order-independent transparency when multisample antialiasing is
in use.
# Objective
- `cargo run --release --example bevymark -- --benchmark --waves 160
--per-wave 1000 --mode mesh2d` runs slower and slower over time due to
`no_gpu_preprocessing::write_batched_instance_buffer<bevy_sprite::mesh2d::mesh::Mesh2dPipeline>`
taking longer and longer because the `BatchedInstanceBuffer` is not
cleared
## Solution
- Split the `clear_batched_instance_buffers` system into CPU and GPU
versions
- Use the CPU version for 2D meshes
`Sprite`, `Text`, and `Handle<MeshletMesh>` were types of renderable
entities that the new segregated visible entity system didn't handle, so
they didn't appear.
Because `bevy_text` depends on `bevy_sprite`, and the visibility
computation of text happens in the latter crate, I had to introduce a
new marker component, `SpriteSource`. `SpriteSource` marks entities that
aren't themselves sprites but become sprites during rendering. I added
this component to `Text2dBundle`. Unfortunately, this is technically a
breaking change, although I suspect it won't break anybody in practice
except perhaps editors.
Fixes#12935.
## Changelog
### Changed
* `Text2dBundle` now includes a new marker component, `SpriteSource`.
Bevy uses this internally to optimize visibility calculation.
## Migration Guide
* `Text` now requires a `SpriteSource` marker component in order to
appear. This component has been added to `Text2dBundle`.
This commit splits `VisibleEntities::entities` into four separate lists:
one for lights, one for 2D meshes, one for 3D meshes, and one for UI
elements. This allows `queue_material_meshes` and similar methods to
avoid examining entities that are obviously irrelevant. In particular,
this separation helps scenes with many skinned meshes, as the individual
bones are considered visible entities but have no rendered appearance.
Internally, `VisibleEntities::entities` is a `HashMap` from the `TypeId`
representing a `QueryFilter` to the appropriate `Entity` list. I had to
do this because `VisibleEntities` is located within an upstream crate
from the crates that provide lights (`bevy_pbr`) and 2D meshes
(`bevy_sprite`). As an added benefit, this setup allows apps to provide
their own types of renderable components, by simply adding a specialized
`check_visibility` to the schedule.
This provides a 16.23% end-to-end speedup on `many_foxes` with 10,000
foxes (24.06 ms/frame to 20.70 ms/frame).
## Migration guide
* `check_visibility` and `VisibleEntities` now store the four types of
renderable entities--2D meshes, 3D meshes, lights, and UI
elements--separately. If your custom rendering code examines
`VisibleEntities`, it will now need to specify which type of entity it's
interested in using the `WithMesh2d`, `WithMesh`, `WithLight`, and
`WithNode` types respectively. If your app introduces a new type of
renderable entity, you'll need to add an explicit call to
`check_visibility` to the schedule to accommodate your new component or
components.
## Analysis
`many_foxes`, 10,000 foxes: `main`:
`many_foxes`, 10,000 foxes, this branch:
`queue_material_meshes` (yellow = this branch, red = `main`):
`queue_shadows` (yellow = this branch, red = `main`):
I ported the two existing PCF techniques to the cubemap domain as best I
could. Generally, the technique is to create a 2D orthonormal basis
using Gram-Schmidt normalization, then apply the technique over that
basis. The results look fine, though the shadow bias often needs
adjusting.
For comparison, Unity uses a 4-tap pattern for PCF on point lights of
(1, 1, 1), (-1, -1, 1), (-1, 1, -1), (1, -1, -1). I tried this but
didn't like the look, so I went with the design above, which ports the
2D techniques to the 3D domain. There's surprisingly little material on
point light PCF.
I've gone through every example using point lights and verified that the
shadow maps look fine, adjusting biases as necessary.
Fixes#3628.
---
## Changelog
### Added
* Shadows from point lights now support percentage-closer filtering
(PCF), and as a result look less aliased.
### Changed
* `ShadowFilteringMethod::Castano13` and
`ShadowFilteringMethod::Jimenez14` have been renamed to
`ShadowFilteringMethod::Gaussian` and `ShadowFilteringMethod::Temporal`
respectively.
## Migration Guide
* `ShadowFilteringMethod::Castano13` and
`ShadowFilteringMethod::Jimenez14` have been renamed to
`ShadowFilteringMethod::Gaussian` and `ShadowFilteringMethod::Temporal`
respectively.
# Objective
Fixes#11996
The deprecated shape Quad's flip field role migrated to
StandardMaterial's flip/flipped methods
## Solution
flip/flipping methods of StandardMaterial is applicable to any mesh
---
## Changelog
- Added flip and flipped methods to the StandardMaterial implementation
- Added FLIP_HORIZONTAL, FLIP_VERTICAL, FLIP_X, FLIP_Y, FLIP_Z constants
## Migration Guide
Instead of using `Quad::flip` field, call `flipped(true, false)` method
on the StandardMaterial instance when adding the mesh.
---------
Co-authored-by: BD103 <59022059+BD103@users.noreply.github.com>
Currently, `MeshUniform`s are rather large: 160 bytes. They're also
somewhat expensive to compute, because they involve taking the inverse
of a 3x4 matrix. Finally, if a mesh is present in multiple views, that
mesh will have a separate `MeshUniform` for each and every view, which
is wasteful.
This commit fixes these issues by introducing the concept of a *mesh
input uniform* and adding a *mesh uniform building* compute shader pass.
The `MeshInputUniform` is simply the minimum amount of data needed for
the GPU to compute the full `MeshUniform`. Most of this data is just the
transform and is therefore only 64 bytes. `MeshInputUniform`s are
computed during the *extraction* phase, much like skins are today, in
order to avoid needlessly copying transforms around on CPU. (In fact,
the render app has been changed to only store the translation of each
mesh; it no longer cares about any other part of the transform, which is
stored only on the GPU and the main world.) Before rendering, the
`build_mesh_uniforms` pass runs to expand the `MeshInputUniform`s to the
full `MeshUniform`.
The mesh uniform building pass does the following, all on GPU:
1. Copy the appropriate fields of the `MeshInputUniform` to the
`MeshUniform` slot. If a single mesh is present in multiple views, this
effectively duplicates it into each view.
2. Compute the inverse transpose of the model transform, used for
transforming normals.
3. If applicable, copy the mesh's transform from the previous frame for
TAA. To support this, we double-buffer the `MeshInputUniform`s over two
frames and swap the buffers each frame. The `MeshInputUniform`s for the
current frame contain the index of that mesh's `MeshInputUniform` for
the previous frame.
This commit produces wins in virtually every CPU part of the pipeline:
`extract_meshes`, `queue_material_meshes`,
`batch_and_prepare_render_phase`, and especially
`write_batched_instance_buffer` are all faster. Shrinking the amount of
CPU data that has to be shuffled around speeds up the entire rendering
process.
| Benchmark | This branch | `main` | Speedup |
|------------------------|-------------|---------|---------|
| `many_cubes -nfc` | 17.259 | 24.529 | 42.12% |
| `many_cubes -nfc -vpi` | 302.116 | 312.123 | 3.31% |
| `many_foxes` | 3.227 | 3.515 | 8.92% |
Because mesh uniform building requires compute shader, and WebGL 2 has
no compute shader, the existing CPU mesh uniform building code has been
left as-is. Many types now have both CPU mesh uniform building and GPU
mesh uniform building modes. Developers can opt into the old CPU mesh
uniform building by setting the `use_gpu_uniform_builder` option on
`PbrPlugin` to `false`.
Below are graphs of the CPU portions of `many-cubes
--no-frustum-culling`. Yellow is this branch, red is `main`.
`extract_meshes`:
It's notable that we get a small win even though we're now writing to a
GPU buffer.
`queue_material_meshes`:
There's a bit of a regression here; not sure what's causing it. In any
case it's very outweighed by the other gains.
`batch_and_prepare_render_phase`:
There's a huge win here, enough to make batching basically drop off the
profile.
`write_batched_instance_buffer`:
There's a massive improvement here, as expected. Note that a lot of it
simply comes from the fact that `MeshInputUniform` is `Pod`. (This isn't
a maintainability problem in my view because `MeshInputUniform` is so
simple: just 16 tightly-packed words.)
## Changelog
### Added
* Per-mesh instance data is now generated on GPU with a compute shader
instead of CPU, resulting in rendering performance improvements on
platforms where compute shaders are supported.
## Migration guide
* Custom render phases now need multiple systems beyond just
`batch_and_prepare_render_phase`. Code that was previously creating
custom render phases should now add a `BinnedRenderPhasePlugin` or
`SortedRenderPhasePlugin` as appropriate instead of directly adding
`batch_and_prepare_render_phase`.
# Objective
- Replace `RenderMaterials` / `RenderMaterials2d` / `RenderUiMaterials`
with `RenderAssets` to enable implementing changes to one thing,
`RenderAssets`, that applies to all use cases rather than duplicating
changes everywhere for multiple things that should be one thing.
- Adopts #8149
## Solution
- Make RenderAsset generic over the destination type rather than the
source type as in #8149
- Use `RenderAssets<PreparedMaterial<M>>` etc for render materials
---
## Changelog
- Changed:
- The `RenderAsset` trait is now implemented on the destination type.
Its `SourceAsset` associated type refers to the type of the source
asset.
- `RenderMaterials`, `RenderMaterials2d`, and `RenderUiMaterials` have
been replaced by `RenderAssets<PreparedMaterial<M>>` and similar.
## Migration Guide
- `RenderAsset` is now implemented for the destination type rather that
the source asset type. The source asset type is now the `RenderAsset`
trait's `SourceAsset` associated type.
# Objective
Minimize the number of dependencies low in the tree.
## Solution
* Remove the dependency on rustc-hash in bevy_ecs (not used) and
bevy_macro_utils (only used in one spot).
* Deduplicate the dependency on `sha1_smol` with the existing blake3
dependency already being used for bevy_asset.
* Remove the unused `ron` dependency on `bevy_app`
* Make the `serde` dependency for `bevy_ecs` optional. It's only used
for serializing Entity.
* Change the `wgpu` dependency to `wgpu-types`, and make it optional for
`bevy_color`.
* Remove the unused `thread-local` dependency on `bevy_render`.
* Make multiple dependencies for `bevy_tasks` optional and enabled only
when running with the `multi-threaded` feature. Preferably they'd be
disabled all the time on wasm, but I couldn't find a clean way to do
this.
---
## Changelog
TODO
## Migration Guide
TODO
# Objective
- Fix a potential out-of-bounds access in the `pbr_functions.wgsl`
shader.
## Solution
- Correctly compute the `GpuLights::directional_lights` array length.
## Comments
I think this solves this comment in the code, but need someone to test
it:
```rust
//NOTE: When running bevy on Adreno GPU chipsets in WebGL, any value above 1 will result in a crash
// when loading the wgsl "pbr_functions.wgsl" in the function apply_fog.
```
# Objective
- Upload previous frame's inverse_view matrix to the GPU for use with
https://github.com/bevyengine/bevy/pull/12898.
---
## Changelog
- Added `prepass_bindings::previous_view_uniforms.inverse_view`.
- Renamed `prepass_bindings::previous_view_proj` to
`prepass_bindings::previous_view_uniforms.view_proj`.
- Renamed `PreviousViewProjectionUniformOffset` to
`PreviousViewUniformOffset`.
- Renamed `PreviousViewProjection` to `PreviousViewData`.
## Migration Guide
- Renamed `prepass_bindings::previous_view_proj` to
`prepass_bindings::previous_view_uniforms.view_proj`.
- Renamed `PreviousViewProjectionUniformOffset` to
`PreviousViewUniformOffset`.
- Renamed `PreviousViewProjection` to `PreviousViewData`.
# Objective
make morph targets and tonemapping more tolerant of delayed image
loading.
neither of these actually fail currently unless using a bespoke loader
(and even then it would be rare), but i am working on adding throttling
for asset gpu uploads (as a stopgap until we can do proper asset
streaming) and they break with that.
## Solution
when a mesh with morph targets is uploaded to the gpu, the prepare
function uploads the morph target texture if it's available, otherwise
it uploads without morph targets. this is generally fine as long as
morph targets are typically loaded from bytes (in gltf loader), but may
fail for a custom loader if the asset server async-loads the target
texture and the texture is not available yet. the mesh fails to render
and doesn't update when the image is loaded
-> if morph targets are specified but not ready yet, retry mesh upload
next frame
tonemapping `unwrap`s on the lookup table image. this is never a problem
since the image is added via `include_bytes!`, but could be a problem in
future with asset gpu throttling/streaming.
-> if the lookup texture is not yet available, use a fallback
-> in the node, check if the fallback was used before caching the bind
group
# Objective
- #12791 broke example `irradiance_volumes`
- Fixes#12876
```
wgpu error: Validation Error
Caused by:
In Device::create_render_pipeline
note: label = `pbr_opaque_mesh_pipeline`
Color state [0] is invalid
Sample count 8 is not supported by format Rgba8UnormSrgb on this device. The WebGPU spec guarentees [1, 4] samples are supported by this format. With the TEXTURE_ADAPTER_SPECIFIC_FORMAT_FEATURES feature your device supports [1, 2, 4].
```
## Solution
- Shift bits a bit more
# Objective
Since BufferVec was first introduced, `bytemuck` has added additional
traits with fewer restrictions than `Pod`. Within BufferVec, we only
rely on the constraints of `bytemuck::cast_slice` to a `u8` slice, which
now only requires `T: NoUninit` which is a strict superset of `Pod`
types.
## Solution
Change out the `Pod` generic type constraint with `NoUninit`. Also
taking the opportunity to substitute `cast_slice` with
`must_cast_slice`, which avoids a runtime panic in place of a compile
time failure if `T` cannot be used.
---
## Changelog
Changed: `BufferVec` now supports working with types containing
`NoUninit` but not `Pod` members.
Changed: `BufferVec` will now fail to compile if used with a type that
cannot be safely read from. Most notably, this includes ZSTs, which
would previously always panic at runtime.
This commit makes the following optimizations:
## `MeshPipelineKey`/`BaseMeshPipelineKey` split
`MeshPipelineKey` has been split into `BaseMeshPipelineKey`, which lives
in `bevy_render` and `MeshPipelineKey`, which lives in `bevy_pbr`.
Conceptually, `BaseMeshPipelineKey` is a superclass of
`MeshPipelineKey`. For `BaseMeshPipelineKey`, the bits start at the
highest (most significant) bit and grow downward toward the lowest bit;
for `MeshPipelineKey`, the bits start at the lowest bit and grow upward
toward the highest bit. This prevents them from colliding.
The goal of this is to avoid having to reassemble bits of the pipeline
key for every mesh every frame. Instead, we can just use a bitwise or
operation to combine the pieces that make up a `MeshPipelineKey`.
## `specialize_slow`
Previously, all of `specialize()` was marked as `#[inline]`. This
bloated `queue_material_meshes` unnecessarily, as a large chunk of it
ended up being a slow path that was rarely hit. This commit refactors
the function to move the slow path to `specialize_slow()`.
Together, these two changes shave about 5% off `queue_material_meshes`:
## Migration Guide
- The `primitive_topology` field on `GpuMesh` is now an accessor method:
`GpuMesh::primitive_topology()`.
- For performance reasons, `MeshPipelineKey` has been split into
`BaseMeshPipelineKey`, which lives in `bevy_render`, and
`MeshPipelineKey`, which lives in `bevy_pbr`. These two should be
combined with bitwise-or to produce the final `MeshPipelineKey`.
# Objective
This is a necessary precursor to #9122 (this was split from that PR to
reduce the amount of code to review all at once).
Moving `!Send` resource ownership to `App` will make it unambiguously
`!Send`. `SubApp` must be `Send`, so it can't wrap `App`.
## Solution
Refactor `App` and `SubApp` to not have a recursive relationship. Since
`SubApp` no longer wraps `App`, once `!Send` resources are moved out of
`World` and into `App`, `SubApp` will become unambiguously `Send`.
There could be less code duplication between `App` and `SubApp`, but
that would break `App` method chaining.
## Changelog
- `SubApp` no longer wraps `App`.
- `App` fields are no longer publicly accessible.
- `App` can no longer be converted into a `SubApp`.
- Various methods now return references to a `SubApp` instead of an
`App`.
## Migration Guide
- To construct a sub-app, use `SubApp::new()`. `App` can no longer
convert into `SubApp`.
- If you implemented a trait for `App`, you may want to implement it for
`SubApp` as well.
- If you're accessing `app.world` directly, you now have to use
`app.world()` and `app.world_mut()`.
- `App::sub_app` now returns `&SubApp`.
- `App::sub_app_mut` now returns `&mut SubApp`.
- `App::get_sub_app` now returns `Option<&SubApp>.`
- `App::get_sub_app_mut` now returns `Option<&mut SubApp>.`
Today, we sort all entities added to all phases, even the phases that
don't strictly need sorting, such as the opaque and shadow phases. This
results in a performance loss because our `PhaseItem`s are rather large
in memory, so sorting is slow. Additionally, determining the boundaries
of batches is an O(n) process.
This commit makes Bevy instead applicable place phase items into *bins*
keyed by *bin keys*, which have the invariant that everything in the
same bin is potentially batchable. This makes determining batch
boundaries O(1), because everything in the same bin can be batched.
Instead of sorting each entity, we now sort only the bin keys. This
drops the sorting time to near-zero on workloads with few bins like
`many_cubes --no-frustum-culling`. Memory usage is improved too, with
batch boundaries and dynamic indices now implicit instead of explicit.
The improved memory usage results in a significant win even on
unbatchable workloads like `many_cubes --no-frustum-culling
--vary-material-data-per-instance`, presumably due to cache effects.
Not all phases can be binned; some, such as transparent and transmissive
phases, must still be sorted. To handle this, this commit splits
`PhaseItem` into `BinnedPhaseItem` and `SortedPhaseItem`. Most of the
logic that today deals with `PhaseItem`s has been moved to
`SortedPhaseItem`. `BinnedPhaseItem` has the new logic.
Frame time results (in ms/frame) are as follows:
| Benchmark | `binning` | `main` | Speedup |
| ------------------------ | --------- | ------- | ------- |
| `many_cubes -nfc -vpi` | 232.179 | 312.123 | 34.43% |
| `many_cubes -nfc` | 25.874 | 30.117 | 16.40% |
| `many_foxes` | 3.276 | 3.515 | 7.30% |
(`-nfc` is short for `--no-frustum-culling`; `-vpi` is short for
`--vary-per-instance`.)
---
## Changelog
### Changed
* Render phases have been split into binned and sorted phases. Binned
phases, such as the common opaque phase, achieve improved CPU
performance by avoiding the sorting step.
## Migration Guide
- `PhaseItem` has been split into `BinnedPhaseItem` and
`SortedPhaseItem`. If your code has custom `PhaseItem`s, you will need
to migrate them to one of these two types. `SortedPhaseItem` requires
the fewest code changes, but you may want to pick `BinnedPhaseItem` if
your phase doesn't require sorting, as that enables higher performance.
## Tracy graphs
`many-cubes --no-frustum-culling`, `main` branch:
<img width="1064" alt="Screenshot 2024-03-12 180037"
src="https://github.com/bevyengine/bevy/assets/157897/e1180ce8-8e89-46d2-85e3-f59f72109a55">
`many-cubes --no-frustum-culling`, this branch:
<img width="1064" alt="Screenshot 2024-03-12 180011"
src="https://github.com/bevyengine/bevy/assets/157897/0899f036-6075-44c5-a972-44d95895f46c">
You can see that `batch_and_prepare_binned_render_phase` is a much
smaller fraction of the time. Zooming in on that function, with yellow
being this branch and red being `main`, we see:
<img width="1064" alt="Screenshot 2024-03-12 175832"
src="https://github.com/bevyengine/bevy/assets/157897/0dfc8d3f-49f4-496e-8825-a66e64d356d0">
The binning happens in `queue_material_meshes`. Again with yellow being
this branch and red being `main`:
<img width="1064" alt="Screenshot 2024-03-12 175755"
src="https://github.com/bevyengine/bevy/assets/157897/b9b20dc1-11c8-400c-a6cc-1c2e09c1bb96">
We can see that there is a small regression in `queue_material_meshes`
performance, but it's not nearly enough to outweigh the large gains in
`batch_and_prepare_binned_render_phase`.
---------
Co-authored-by: James Liu <contact@jamessliu.com>
This commit changes the `StandardMaterialKey` to be based on a set of
bitflags instead of a structure. We hash it every frame for every mesh,
and `#[derive(Hash)]` doesn't generate particularly efficient code for
large structures full of small types. Packing it into a single `u64`
therefore results in a roughly 10% speedup in `queue_material_meshes` on
`many_cubes --no-frustum-culling`.
# Objective
Fixes#12727. All parts that `PersistentGpuBuffer` interact with should
be 100% safe both on the CPU and the GPU: `Queue::write_buffer_with`
zeroes out the slice being written to and when uploading to the GPU, and
all slice writes are bounds checked on the CPU side.
## Solution
Make `PersistentGpuBufferable` a safe trait. Enforce it's correct
implementation via assertions. Re-enable `forbid(unsafe_code)` on
`bevy_pbr`.
# Objective
- Fixes#12712
## Solution
- Move the `float_ord.rs` file to `bevy_math`
- Change any `bevy_utils::FloatOrd` statements to `bevy_math::FloatOrd`
---
## Changelog
- Moved `FloatOrd` from `bevy_utils` to `bevy_math`
## Migration Guide
- References to `bevy_utils::FloatOrd` should be changed to
`bevy_math::FloatOrd`
# Objective
Resolves#3824. `unsafe` code should be the exception, not the norm in
Rust. It's obviously needed for various use cases as it's interfacing
with platforms and essentially running the borrow checker at runtime in
the ECS, but the touted benefits of Bevy is that we are able to heavily
leverage Rust's safety, and we should be holding ourselves accountable
to that by minimizing our unsafe footprint.
## Solution
Deny `unsafe_code` workspace wide. Add explicit exceptions for the
following crates, and forbid it in almost all of the others.
* bevy_ecs - Obvious given how much unsafe is needed to achieve
performant results
* bevy_ptr - Works with raw pointers, even more low level than bevy_ecs.
* bevy_render - due to needing to integrate with wgpu
* bevy_window - due to needing to integrate with raw_window_handle
* bevy_utils - Several unsafe utilities used by bevy_ecs. Ideally moved
into bevy_ecs instead of made publicly usable.
* bevy_reflect - Required for the unsafe type casting it's doing.
* bevy_transform - for the parallel transform propagation
* bevy_gizmos - For the SystemParam impls it has.
* bevy_assets - To support reflection. Might not be required, not 100%
sure yet.
* bevy_mikktspace - due to being a conversion from a C library. Pending
safe rewrite.
* bevy_dynamic_plugin - Inherently unsafe due to the dynamic loading
nature.
Several uses of unsafe were rewritten, as they did not need to be using
them:
* bevy_text - a case of `Option::unchecked` could be rewritten as a
normal for loop and match instead of an iterator.
* bevy_color - the Pod/Zeroable implementations were replaceable with
bytemuck's derive macros.
# Objective
Currently the built docs only shows the logo and favicon for the top
level `bevy` crate. This makes views like
https://docs.rs/bevy_ecs/latest/bevy_ecs/ look potentially unrelated to
the project at first glance.
## Solution
Reproduce the docs attributes for every crate that Bevy publishes.
Ideally this would be done with some workspace level Cargo.toml control,
but AFAICT, such support does not exist.
# Objective
Follow up from PR #12369 to extract lighting structs from light/mod.rs
into their own file.
Part of the Purdue Refactoring Team's goals issue #12349
## Solution
- Moved PointLight from light/mod.rs to light/point_light.rs
- Moved SpotLight from light/mod.rs to light/spot_light.rs
- Moved DirectionalLight from light/mod.rs to light/directional_light.rs
Fixes#12600
## Solution
Removed Into<AssetId<T>> for Handle<T> as proposed in Issue
conversation, fixed dependent code
## Migration guide
If you use passing Handle by value as AssetId, you should pass reference
or call .id() method on it
Before (0.13):
`assets.insert(handle, value);`
After (0.14):
`assets.insert(&handle, value);`
or
`assets.insert(handle.id(), value);`
# Objective
- Not all materials need shadow, but a queue_shadows system is always
added to the `Render` schedule and executed
## Solution
- Make a setting for shadows, it defaults to true
## Changelog
- Added `shadows_enabled` setting to `MaterialPlugin`
## Migration Guide
- `MaterialPlugin` now has a `shadows_enabled` setting, if you didn't
spawn the plugin using `::default()` or `..default()`, you'll need to
set it. `shadows_enabled: true` is the same behavior as the previous
version, and also the default value.
# Objective
It's useful to have access to render pipeline statistics, since they
provide more information than FPS alone. For example, the number of
drawn triangles can be used to debug culling and LODs. The number of
fragment shader invocations can provide a more stable alternative metric
than GPU elapsed time.
See also: Render node GPU timing overlay #8067, which doesn't provide
pipeline statistics, but adds a nice overlay.
## Solution
Add `RenderDiagnosticsPlugin`, which enables collecting pipeline
statistics and CPU & GPU timings.
---
## Changelog
- Add `RenderDiagnosticsPlugin`
- Add `RenderContext::diagnostic_recorder` method
---------
Co-authored-by: Alice Cecile <alice.i.cecile@gmail.com>
# Objective
Improve code quality involving fixedbitset.
## Solution
Update to fixedbitset 0.5. Use the new `grow_and_insert` function
instead of `grow` and `insert` functions separately.
This should also speed up most of the set operations involving
fixedbitset. They should be ~2x faster, but testing this against the
stress tests seems to show little to no difference. The multithreaded
executor doesn't seem to be all that much faster in many_cubes and
many_foxes. These use cases are likely dominated by other operations or
the bitsets aren't big enough to make them the bottleneck.
This introduces a duplicate dependency due to petgraph and wgpu, but the
former may take some time to update.
## Changelog
Removed: `Access::grow`
## Migration Guide
`Access::grow` has been removed. It's no longer needed. Remove all
references to it.
# Objective
assets that don't load before they get removed are retried forever,
causing buffer churn and slowdown.
## Solution
stop trying to prepare dead assets.
# Objective
Beginning of refactoring of light.rs in bevy_pbr, as per issue #12349
Create and move light.rs to its own directory, and extract AmbientLight
struct.
## Solution
- moved light.rs to light/mod.rs
- extracted AmbientLight struct to light/ambient_light.rs
