Add BufferVec, an higher-performance alternative to StorageBuffer, and make GpuArrayBuffer use it. (#13199)

This is an adoption of #12670 plus some documentation fixes. See that PR
for more details.

---

## Changelog

* Renamed `BufferVec` to `RawBufferVec` and added a new `BufferVec`
type.

## Migration Guide
`BufferVec` has been renamed to `RawBufferVec` and a new similar type
has taken the `BufferVec` name.

---------

Co-authored-by: Patrick Walton <pcwalton@mimiga.net>
Co-authored-by: Alice Cecile <alice.i.cecile@gmail.com>
Co-authored-by: IceSentry <IceSentry@users.noreply.github.com>

crates/bevy_pbr/src/render/mesh.rs

@@ -331,7 +331,7 @@ pub struct MeshCullingData {
 /// To avoid wasting CPU time in the CPU culling case, this buffer will be empty
 /// if GPU culling isn't in use.
 #[derive(Resource, Deref, DerefMut)]
-pub struct MeshCullingDataBuffer(BufferVec<MeshCullingData>);
+pub struct MeshCullingDataBuffer(RawBufferVec<MeshCullingData>);
 
 impl MeshUniform {
     pub fn new(mesh_transforms: &MeshTransforms, maybe_lightmap_uv_rect: Option<Rect>) -> Self {
@@ -685,7 +685,7 @@ impl RenderMeshInstanceGpuBuilder {
         self,
         entity: Entity,
         render_mesh_instances: &mut EntityHashMap<RenderMeshInstanceGpu>,
-        current_input_buffer: &mut BufferVec<MeshInputUniform>,
+        current_input_buffer: &mut RawBufferVec<MeshInputUniform>,
     ) -> usize {
         // Push the mesh input uniform.
         let current_uniform_index = current_input_buffer.push(MeshInputUniform {
@@ -742,7 +742,7 @@ impl MeshCullingData {
 impl Default for MeshCullingDataBuffer {
     #[inline]
     fn default() -> Self {
-        Self(BufferVec::new(BufferUsages::STORAGE))
+        Self(RawBufferVec::new(BufferUsages::STORAGE))
     }
 }
 

crates/bevy_pbr/src/render/morph.rs

@@ -6,7 +6,7 @@ use bevy_ecs::prelude::*;
 use bevy_render::{
     batching::NoAutomaticBatching,
     mesh::morph::{MeshMorphWeights, MAX_MORPH_WEIGHTS},
-    render_resource::{BufferUsages, BufferVec},
+    render_resource::{BufferUsages, RawBufferVec},
     renderer::{RenderDevice, RenderQueue},
     view::ViewVisibility,
     Extract,
@@ -23,13 +23,13 @@ pub struct MorphIndices(EntityHashMap<MorphIndex>);
 
 #[derive(Resource)]
 pub struct MorphUniform {
-    pub buffer: BufferVec<f32>,
+    pub buffer: RawBufferVec<f32>,
 }
 
 impl Default for MorphUniform {
     fn default() -> Self {
         Self {
-            buffer: BufferVec::new(BufferUsages::UNIFORM),
+            buffer: RawBufferVec::new(BufferUsages::UNIFORM),
         }
     }
 }
@@ -53,14 +53,14 @@ const fn can_align(step: usize, target: usize) -> bool {
 
 const WGPU_MIN_ALIGN: usize = 256;
 
-/// Align a [`BufferVec`] to `N` bytes by padding the end with `T::default()` values.
+/// Align a [`RawBufferVec`] to `N` bytes by padding the end with `T::default()` values.
-fn add_to_alignment<T: NoUninit + Default>(buffer: &mut BufferVec<T>) {
+fn add_to_alignment<T: NoUninit + Default>(buffer: &mut RawBufferVec<T>) {
     let n = WGPU_MIN_ALIGN;
     let t_size = mem::size_of::<T>();
     if !can_align(n, t_size) {
         // This panic is stripped at compile time, due to n, t_size and can_align being const
         panic!(
-            "BufferVec should contain only types with a size multiple or divisible by {n}, \
+            "RawBufferVec should contain only types with a size multiple or divisible by {n}, \
             {} has a size of {t_size}, which is neither multiple or divisible by {n}",
             std::any::type_name::<T>()
         );

crates/bevy_pbr/src/render/skin.rs

@@ -6,7 +6,7 @@ use bevy_math::Mat4;
 use bevy_render::{
     batching::NoAutomaticBatching,
     mesh::skinning::{SkinnedMesh, SkinnedMeshInverseBindposes},
-    render_resource::{BufferUsages, BufferVec},
+    render_resource::{BufferUsages, RawBufferVec},
     renderer::{RenderDevice, RenderQueue},
     view::ViewVisibility,
     Extract,
@@ -36,13 +36,13 @@ pub struct SkinIndices(EntityHashMap<SkinIndex>);
 // Notes on implementation: see comment on top of the `extract_skins` system.
 #[derive(Resource)]
 pub struct SkinUniform {
-    pub buffer: BufferVec<Mat4>,
+    pub buffer: RawBufferVec<Mat4>,
 }
 
 impl Default for SkinUniform {
     fn default() -> Self {
         Self {
-            buffer: BufferVec::new(BufferUsages::UNIFORM),
+            buffer: RawBufferVec::new(BufferUsages::UNIFORM),
         }
     }
 }

crates/bevy_render/src/batching/gpu_preprocessing.rs

@@ -21,7 +21,7 @@ use crate::{
         BinnedPhaseItem, BinnedRenderPhase, BinnedRenderPhaseBatch, CachedRenderPipelinePhaseItem,
         PhaseItemExtraIndex, SortedPhaseItem, SortedRenderPhase, UnbatchableBinnedEntityIndices,
     },
-    render_resource::{BufferVec, GpuArrayBufferable, UninitBufferVec},
+    render_resource::{BufferVec, GpuArrayBufferable, RawBufferVec, UninitBufferVec},
     renderer::{RenderAdapter, RenderDevice, RenderQueue},
     view::{GpuCulling, ViewTarget},
     Render, RenderApp, RenderSet,
@@ -101,7 +101,7 @@ where
     /// The uniform data inputs for the current frame.
     ///
     /// These are uploaded during the extraction phase.
-    pub current_input_buffer: BufferVec<BDI>,
+    pub current_input_buffer: RawBufferVec<BDI>,
 
     /// The uniform data inputs for the previous frame.
     ///
@@ -110,7 +110,7 @@ where
     /// can spawn or despawn between frames. Instead, each current buffer
     /// data input uniform is expected to contain the index of the
     /// corresponding buffer data input uniform in this list.
-    pub previous_input_buffer: BufferVec<BDI>,
+    pub previous_input_buffer: RawBufferVec<BDI>,
 }
 
 /// The buffer of GPU preprocessing work items for a single view.
@@ -247,8 +247,8 @@ where
         BatchedInstanceBuffers {
             data_buffer: UninitBufferVec::new(BufferUsages::STORAGE),
             work_item_buffers: EntityHashMap::default(),
-            current_input_buffer: BufferVec::new(BufferUsages::STORAGE),
+            current_input_buffer: RawBufferVec::new(BufferUsages::STORAGE),
-            previous_input_buffer: BufferVec::new(BufferUsages::STORAGE),
+            previous_input_buffer: RawBufferVec::new(BufferUsages::STORAGE),
         }
     }
 

crates/bevy_render/src/render_resource/buffer_vec.rs

@@ -1,11 +1,15 @@
-use std::marker::PhantomData;
+use std::{iter, marker::PhantomData};
 
 use crate::{
     render_resource::Buffer,
     renderer::{RenderDevice, RenderQueue},
 };
 use bytemuck::{must_cast_slice, NoUninit};
-use wgpu::BufferUsages;
+use encase::{
+    internal::{WriteInto, Writer},
+    ShaderType,
+};
+use wgpu::{BufferAddress, BufferUsages};
 
 use super::GpuArrayBufferable;
 
@@ -19,9 +23,9 @@ use super::GpuArrayBufferable;
 /// Index, vertex, and instance-rate vertex buffers have no alignment nor padding requirements and
 /// so this helper type is a good choice for them.
 ///
-/// The contained data is stored in system RAM. Calling [`reserve`](BufferVec::reserve)
+/// The contained data is stored in system RAM. Calling [`reserve`](RawBufferVec::reserve)
 /// allocates VRAM from the [`RenderDevice`].
-/// [`write_buffer`](BufferVec::write_buffer) queues copying of the data
+/// [`write_buffer`](RawBufferVec::write_buffer) queues copying of the data
 /// from system RAM to VRAM.
 ///
 /// Other options for storing GPU-accessible data are:
@@ -32,7 +36,7 @@ use super::GpuArrayBufferable;
 /// * [`GpuArrayBuffer`](crate::render_resource::GpuArrayBuffer)
 /// * [`BufferVec`]
 /// * [`Texture`](crate::render_resource::Texture)
-pub struct BufferVec<T: NoUninit> {
+pub struct RawBufferVec<T: NoUninit> {
     values: Vec<T>,
     buffer: Option<Buffer>,
     capacity: usize,
@@ -42,7 +46,7 @@ pub struct BufferVec<T: NoUninit> {
     label_changed: bool,
 }
 
-impl<T: NoUninit> BufferVec<T> {
+impl<T: NoUninit> RawBufferVec<T> {
     pub const fn new(buffer_usage: BufferUsages) -> Self {
         Self {
             values: Vec::new(),
@@ -81,7 +85,7 @@ impl<T: NoUninit> BufferVec<T> {
         index
     }
 
-    pub fn append(&mut self, other: &mut BufferVec<T>) {
+    pub fn append(&mut self, other: &mut RawBufferVec<T>) {
         self.values.append(&mut other.values);
     }
 
@@ -108,7 +112,7 @@ impl<T: NoUninit> BufferVec<T> {
     /// once it is done using them (typically 1-2 frames).
     ///
     /// In addition to any [`BufferUsages`] provided when
-    /// the `BufferVec` was created, the buffer on the [`RenderDevice`]
+    /// the `RawBufferVec` was created, the buffer on the [`RenderDevice`]
     /// is marked as [`BufferUsages::COPY_DST`](BufferUsages).
     pub fn reserve(&mut self, capacity: usize, device: &RenderDevice) {
         if capacity > self.capacity || self.label_changed {
@@ -116,7 +120,7 @@ impl<T: NoUninit> BufferVec<T> {
             let size = self.item_size * capacity;
             self.buffer = Some(device.create_buffer(&wgpu::BufferDescriptor {
                 label: self.label.as_deref(),
-                size: size as wgpu::BufferAddress,
+                size: size as BufferAddress,
                 usage: BufferUsages::COPY_DST | self.buffer_usage,
                 mapped_at_creation: false,
             }));
@@ -127,7 +131,7 @@ impl<T: NoUninit> BufferVec<T> {
     /// Queues writing of data from system RAM to VRAM using the [`RenderDevice`]
     /// and the provided [`RenderQueue`].
     ///
-    /// Before queuing the write, a [`reserve`](BufferVec::reserve) operation
+    /// Before queuing the write, a [`reserve`](RawBufferVec::reserve) operation
     /// is executed.
     pub fn write_buffer(&mut self, device: &RenderDevice, queue: &RenderQueue) {
         if self.values.is_empty() {
@@ -158,20 +162,177 @@ impl<T: NoUninit> BufferVec<T> {
     }
 }
 
-impl<T: NoUninit> Extend<T> for BufferVec<T> {
+impl<T: NoUninit> Extend<T> for RawBufferVec<T> {
     #[inline]
     fn extend<I: IntoIterator<Item = T>>(&mut self, iter: I) {
         self.values.extend(iter);
     }
 }
 
+/// Like [`RawBufferVec`], but doesn't require that the data type `T` be
+/// [`NoUninit`].
+///
+/// This is a high-performance data structure that you should use whenever
+/// possible if your data is more complex than is suitable for [`RawBufferVec`].
+/// The [`ShaderType`] trait from the `encase` library is used to ensure that
+/// the data is correctly aligned for use by the GPU.
+///
+/// For performance reasons, unlike [`RawBufferVec`], this type doesn't allow
+/// CPU access to the data after it's been added via [`BufferVec::push`]. If you
+/// need CPU access to the data, consider another type, such as
+/// [`StorageBuffer`].
+pub struct BufferVec<T>
+where
+    T: ShaderType + WriteInto,
+{
+    data: Vec<u8>,
+    buffer: Option<Buffer>,
+    capacity: usize,
+    buffer_usage: BufferUsages,
+    label: Option<String>,
+    label_changed: bool,
+    phantom: PhantomData<T>,
+}
+
+impl<T> BufferVec<T>
+where
+    T: ShaderType + WriteInto,
+{
+    /// Creates a new [`BufferVec`] with the given [`BufferUsages`].
+    pub const fn new(buffer_usage: BufferUsages) -> Self {
+        Self {
+            data: vec![],
+            buffer: None,
+            capacity: 0,
+            buffer_usage,
+            label: None,
+            label_changed: false,
+            phantom: PhantomData,
+        }
+    }
+
+    /// Returns a handle to the buffer, if the data has been uploaded.
+    #[inline]
+    pub fn buffer(&self) -> Option<&Buffer> {
+        self.buffer.as_ref()
+    }
+
+    /// Returns the amount of space that the GPU will use before reallocating.
+    #[inline]
+    pub fn capacity(&self) -> usize {
+        self.capacity
+    }
+
+    /// Returns the number of items that have been pushed to this buffer.
+    #[inline]
+    pub fn len(&self) -> usize {
+        self.data.len() / u64::from(T::min_size()) as usize
+    }
+
+    /// Returns true if the buffer is empty.
+    #[inline]
+    pub fn is_empty(&self) -> bool {
+        self.data.is_empty()
+    }
+
+    /// Adds a new value and returns its index.
+    pub fn push(&mut self, value: T) -> usize {
+        let element_size = u64::from(T::min_size()) as usize;
+        let offset = self.data.len();
+
+        // TODO: Consider using unsafe code to push uninitialized, to prevent
+        // the zeroing. It shows up in profiles.
+        self.data.extend(iter::repeat(0).take(element_size));
+
+        // Take a slice of the new data for `write_into` to use. This is
+        // important: it hoists the bounds check up here so that the compiler
+        // can eliminate all the bounds checks that `write_into` will emit.
+        let mut dest = &mut self.data[offset..(offset + element_size)];
+        value.write_into(&mut Writer::new(&value, &mut dest, 0).unwrap());
+
+        offset / u64::from(T::min_size()) as usize
+    }
+
+    /// Changes the debugging label of the buffer.
+    ///
+    /// The next time the buffer is updated (via [`reserve`]), Bevy will inform
+    /// the driver of the new label.
+    pub fn set_label(&mut self, label: Option<&str>) {
+        let label = label.map(str::to_string);
+
+        if label != self.label {
+            self.label_changed = true;
+        }
+
+        self.label = label;
+    }
+
+    /// Returns the label
+    pub fn get_label(&self) -> Option<&str> {
+        self.label.as_deref()
+    }
+
+    /// Creates a [`Buffer`] on the [`RenderDevice`] with size
+    /// at least `std::mem::size_of::<T>() * capacity`, unless such a buffer already exists.
+    ///
+    /// If a [`Buffer`] exists, but is too small, references to it will be discarded,
+    /// and a new [`Buffer`] will be created. Any previously created [`Buffer`]s
+    /// that are no longer referenced will be deleted by the [`RenderDevice`]
+    /// once it is done using them (typically 1-2 frames).
+    ///
+    /// In addition to any [`BufferUsages`] provided when
+    /// the `BufferVec` was created, the buffer on the [`RenderDevice`]
+    /// is marked as [`BufferUsages::COPY_DST`](BufferUsages).
+    pub fn reserve(&mut self, capacity: usize, device: &RenderDevice) {
+        if capacity <= self.capacity && !self.label_changed {
+            return;
+        }
+
+        self.capacity = capacity;
+        let size = u64::from(T::min_size()) as usize * capacity;
+        self.buffer = Some(device.create_buffer(&wgpu::BufferDescriptor {
+            label: self.label.as_deref(),
+            size: size as BufferAddress,
+            usage: BufferUsages::COPY_DST | self.buffer_usage,
+            mapped_at_creation: false,
+        }));
+        self.label_changed = false;
+    }
+
+    /// Queues writing of data from system RAM to VRAM using the [`RenderDevice`]
+    /// and the provided [`RenderQueue`].
+    ///
+    /// Before queuing the write, a [`reserve`](BufferVec::reserve) operation is
+    /// executed.
+    pub fn write_buffer(&mut self, device: &RenderDevice, queue: &RenderQueue) {
+        if self.data.is_empty() {
+            return;
+        }
+
+        self.reserve(self.data.len() / u64::from(T::min_size()) as usize, device);
+
+        let Some(buffer) = &self.buffer else { return };
+        queue.write_buffer(buffer, 0, &self.data);
+    }
+
+    /// Reduces the length of the buffer.
+    pub fn truncate(&mut self, len: usize) {
+        self.data.truncate(u64::from(T::min_size()) as usize * len);
+    }
+
+    /// Removes all elements from the buffer.
+    pub fn clear(&mut self) {
+        self.data.clear();
+    }
+}
+
 /// Like a [`BufferVec`], but only reserves space on the GPU for elements
 /// instead of initializing them CPU-side.
 ///
 /// This type is useful when you're accumulating "output slots" for a GPU
 /// compute shader to write into.
 ///
-/// The type `T` need not be [`NoUninit`], unlike [`BufferVec`]; it only has to
+/// The type `T` need not be [`NoUninit`], unlike [`RawBufferVec`]; it only has to
 /// be [`GpuArrayBufferable`].
 pub struct UninitBufferVec<T>
 where

crates/bevy_render/src/render_resource/gpu_array_buffer.rs

@@ -1,6 +1,6 @@
 use super::{
     binding_types::{storage_buffer_read_only, uniform_buffer_sized},
-    BindGroupLayoutEntryBuilder, StorageBuffer,
+    BindGroupLayoutEntryBuilder, BufferVec,
 };
 use crate::{
     render_resource::batched_uniform_buffer::BatchedUniformBuffer,
@@ -10,7 +10,7 @@ use bevy_ecs::{prelude::Component, system::Resource};
 use encase::{private::WriteInto, ShaderSize, ShaderType};
 use nonmax::NonMaxU32;
 use std::marker::PhantomData;
-use wgpu::BindingResource;
+use wgpu::{BindingResource, BufferUsages};
 
 /// Trait for types able to go in a [`GpuArrayBuffer`].
 pub trait GpuArrayBufferable: ShaderType + ShaderSize + WriteInto + Clone {}
@@ -18,21 +18,23 @@ impl<T: ShaderType + ShaderSize + WriteInto + Clone> GpuArrayBufferable for T {}
 
 /// Stores an array of elements to be transferred to the GPU and made accessible to shaders as a read-only array.
 ///
-/// On platforms that support storage buffers, this is equivalent to [`StorageBuffer<Vec<T>>`].
+/// On platforms that support storage buffers, this is equivalent to
-/// Otherwise, this falls back to a dynamic offset uniform buffer with the largest
+/// [`BufferVec<T>`]. Otherwise, this falls back to a dynamic offset
-/// array of T that fits within a uniform buffer binding (within reasonable limits).
+/// uniform buffer with the largest array of T that fits within a uniform buffer
+/// binding (within reasonable limits).
 ///
 /// Other options for storing GPU-accessible data are:
 /// * [`StorageBuffer`]
 /// * [`DynamicStorageBuffer`](crate::render_resource::DynamicStorageBuffer)
 /// * [`UniformBuffer`](crate::render_resource::UniformBuffer)
 /// * [`DynamicUniformBuffer`](crate::render_resource::DynamicUniformBuffer)
+/// * [`RawBufferVec`](crate::render_resource::RawBufferVec)
 /// * [`BufferVec`](crate::render_resource::BufferVec)
 /// * [`Texture`](crate::render_resource::Texture)
 #[derive(Resource)]
 pub enum GpuArrayBuffer<T: GpuArrayBufferable> {
     Uniform(BatchedUniformBuffer<T>),
-    Storage(StorageBuffer<Vec<T>>),
+    Storage(BufferVec<T>),
 }
 
 impl<T: GpuArrayBufferable> GpuArrayBuffer<T> {
@@ -41,14 +43,14 @@ impl<T: GpuArrayBufferable> GpuArrayBuffer<T> {
         if limits.max_storage_buffers_per_shader_stage == 0 {
             GpuArrayBuffer::Uniform(BatchedUniformBuffer::new(&limits))
         } else {
-            GpuArrayBuffer::Storage(StorageBuffer::default())
+            GpuArrayBuffer::Storage(BufferVec::new(BufferUsages::STORAGE))
         }
     }
 
     pub fn clear(&mut self) {
         match self {
             GpuArrayBuffer::Uniform(buffer) => buffer.clear(),
-            GpuArrayBuffer::Storage(buffer) => buffer.get_mut().clear(),
+            GpuArrayBuffer::Storage(buffer) => buffer.clear(),
         }
     }
 
@@ -56,9 +58,7 @@ impl<T: GpuArrayBufferable> GpuArrayBuffer<T> {
         match self {
             GpuArrayBuffer::Uniform(buffer) => buffer.push(value),
             GpuArrayBuffer::Storage(buffer) => {
-                let buffer = buffer.get_mut();
+                let index = buffer.push(value) as u32;
-                let index = buffer.len() as u32;
-                buffer.push(value);
                 GpuArrayBufferIndex {
                     index,
                     dynamic_offset: None,
@@ -91,7 +91,9 @@ impl<T: GpuArrayBufferable> GpuArrayBuffer<T> {
     pub fn binding(&self) -> Option<BindingResource> {
         match self {
             GpuArrayBuffer::Uniform(buffer) => buffer.binding(),
-            GpuArrayBuffer::Storage(buffer) => buffer.binding(),
+            GpuArrayBuffer::Storage(buffer) => {
+                buffer.buffer().map(|buffer| buffer.as_entire_binding())
+            }
         }
     }
 

crates/bevy_render/src/render_resource/storage_buffer.rs

@@ -24,6 +24,8 @@ use wgpu::{util::BufferInitDescriptor, BindingResource, BufferBinding, BufferUsa
 /// * [`UniformBuffer`](crate::render_resource::UniformBuffer)
 /// * [`DynamicUniformBuffer`](crate::render_resource::DynamicUniformBuffer)
 /// * [`GpuArrayBuffer`](crate::render_resource::GpuArrayBuffer)
+/// * [`RawBufferVec`](crate::render_resource::RawBufferVec)
+/// * [`BufferVec`](crate::render_resource::BufferVec)
 /// * [`BufferVec`](crate::render_resource::BufferVec)
 /// * [`Texture`](crate::render_resource::Texture)
 ///
@@ -154,6 +156,8 @@ impl<T: ShaderType + WriteInto> StorageBuffer<T> {
 /// * [`UniformBuffer`](crate::render_resource::UniformBuffer)
 /// * [`DynamicUniformBuffer`](crate::render_resource::DynamicUniformBuffer)
 /// * [`GpuArrayBuffer`](crate::render_resource::GpuArrayBuffer)
+/// * [`RawBufferVec`](crate::render_resource::RawBufferVec)
+/// * [`BufferVec`](crate::render_resource::BufferVec)
 /// * [`BufferVec`](crate::render_resource::BufferVec)
 /// * [`Texture`](crate::render_resource::Texture)
 ///

crates/bevy_render/src/render_resource/uniform_buffer.rs

@@ -31,6 +31,7 @@ use super::IntoBinding;
 /// * [`DynamicStorageBuffer`](crate::render_resource::DynamicStorageBuffer)
 /// * [`DynamicUniformBuffer`]
 /// * [`GpuArrayBuffer`](crate::render_resource::GpuArrayBuffer)
+/// * [`RawBufferVec`](crate::render_resource::RawBufferVec)
 /// * [`BufferVec`](crate::render_resource::BufferVec)
 /// * [`Texture`](crate::render_resource::Texture)
 ///
@@ -168,6 +169,7 @@ impl<'a, T: ShaderType + WriteInto> IntoBinding<'a> for &'a UniformBuffer<T> {
 /// * [`UniformBuffer`]
 /// * [`DynamicUniformBuffer`]
 /// * [`GpuArrayBuffer`](crate::render_resource::GpuArrayBuffer)
+/// * [`RawBufferVec`](crate::render_resource::RawBufferVec)
 /// * [`BufferVec`](crate::render_resource::BufferVec)
 /// * [`Texture`](crate::render_resource::Texture)
 ///

crates/bevy_sprite/src/render/mod.rs

@@ -413,16 +413,16 @@ impl SpriteInstance {
 #[derive(Resource)]
 pub struct SpriteMeta {
     view_bind_group: Option<BindGroup>,
-    sprite_index_buffer: BufferVec<u32>,
+    sprite_index_buffer: RawBufferVec<u32>,
-    sprite_instance_buffer: BufferVec<SpriteInstance>,
+    sprite_instance_buffer: RawBufferVec<SpriteInstance>,
 }
 
 impl Default for SpriteMeta {
     fn default() -> Self {
         Self {
             view_bind_group: None,
-            sprite_index_buffer: BufferVec::<u32>::new(BufferUsages::INDEX),
+            sprite_index_buffer: RawBufferVec::<u32>::new(BufferUsages::INDEX),
-            sprite_instance_buffer: BufferVec::<SpriteInstance>::new(BufferUsages::VERTEX),
+            sprite_instance_buffer: RawBufferVec::<SpriteInstance>::new(BufferUsages::VERTEX),
         }
     }
 }

crates/bevy_ui/src/render/mod.rs

@@ -834,16 +834,16 @@ struct UiVertex {
 
 #[derive(Resource)]
 pub struct UiMeta {
-    vertices: BufferVec<UiVertex>,
+    vertices: RawBufferVec<UiVertex>,
-    indices: BufferVec<u32>,
+    indices: RawBufferVec<u32>,
     view_bind_group: Option<BindGroup>,
 }
 
 impl Default for UiMeta {
     fn default() -> Self {
         Self {
-            vertices: BufferVec::new(BufferUsages::VERTEX),
+            vertices: RawBufferVec::new(BufferUsages::VERTEX),
-            indices: BufferVec::new(BufferUsages::INDEX),
+            indices: RawBufferVec::new(BufferUsages::INDEX),
             view_bind_group: None,
         }
     }

crates/bevy_ui/src/render/ui_material_pipeline.rs

@@ -91,7 +91,7 @@ where
 
 #[derive(Resource)]
 pub struct UiMaterialMeta<M: UiMaterial> {
-    vertices: BufferVec<UiMaterialVertex>,
+    vertices: RawBufferVec<UiMaterialVertex>,
     view_bind_group: Option<BindGroup>,
     marker: PhantomData<M>,
 }
@@ -99,7 +99,7 @@ pub struct UiMaterialMeta<M: UiMaterial> {
 impl<M: UiMaterial> Default for UiMaterialMeta<M> {
     fn default() -> Self {
         Self {
-            vertices: BufferVec::new(BufferUsages::VERTEX),
+            vertices: RawBufferVec::new(BufferUsages::VERTEX),
             view_bind_group: Default::default(),
             marker: PhantomData,
         }