Parallel Mesh Collection (#22297)

# Objective

- Speed up `collect_meshes_for_gpu_building`, a bottleneck for scenes
with many moving meshes.

## Solution

- Parallelize the gather step for mesh collection.
- Immediately start up a task for serial collection of meshes, which
cannot be parallelized.
- Spawn many tasks for gathering meshes, and send batches of these to
the collection task
- This allows the serial collection step to start immediately, instead
of being delayed until after all collection is finished.

## Testing

- Built a new `bevymark_3d` stress test for benchmarking dynamic 3d mesh
scenes. This is not currently covered by our stress tests.
#22298
- With 200k meshes, this drops total frame times from 16.4ms to 12.3ms
(-4.1ms)

<img width="631" height="377" alt="image"
src="https://github.com/user-attachments/assets/ff9de860-e1f2-4f98-8075-0b3720a82913"
/>

- Mesh collection itself drops from 7.9ms to 3.6ms (-4.3ms)

<img width="623" height="372" alt="image"
src="https://github.com/user-attachments/assets/cf52687a-503b-4f70-8e99-0ad2e63f4da0"
/>

Author: aevyrie

crates/bevy_pbr/Cargo.toml

@@ -22,7 +22,7 @@ bluenoise_texture = ["bevy_image/ktx2", "bevy_image/zstd"]
 shader_format_glsl = ["bevy_shader/shader_format_glsl"]
 trace = ["bevy_render/trace"]
 # Enables the meshlet renderer for dense high-poly scenes (experimental)
-meshlet = ["dep:lz4_flex", "dep:range-alloc", "dep:bevy_tasks"]
+meshlet = ["dep:lz4_flex", "dep:range-alloc"]
 # Enables processing meshes into meshlet meshes
 meshlet_processor = [
   "meshlet",
@@ -56,7 +56,7 @@ bevy_render = { path = "../bevy_render", version = "0.19.0-dev", features = [
   "morph",
 ] }
 bevy_camera = { path = "../bevy_camera", version = "0.19.0-dev" }
-bevy_tasks = { path = "../bevy_tasks", version = "0.19.0-dev", optional = true }
+bevy_tasks = { path = "../bevy_tasks", version = "0.19.0-dev" }
 bevy_transform = { path = "../bevy_transform", version = "0.19.0-dev" }
 bevy_utils = { path = "../bevy_utils", version = "0.19.0-dev" }
 bevy_platform = { path = "../bevy_platform", version = "0.19.0-dev", default-features = false, features = [

crates/bevy_pbr/src/render/mesh.rs

@@ -21,6 +21,12 @@ rayon = ["dep:rayon", "hashbrown/rayon"]
 
 # Platform Compatibility
 
+## Provides an implementation of `block_on` from `futures-lite`.
+futures-lite = ["std", "dep:futures-lite", "futures-lite?/std"]
+
+## Provides an implementation of `block_on` from `async-io`.
+async-io = ["std", "dep:async-io"]
+
 ## Allows access to the `std` crate. Enabling this feature will prevent compilation
 ## on `no_std` targets, but provides access to certain additional features on
 ## supported platforms.
@@ -72,6 +78,8 @@ hashbrown = { version = "0.16.1", features = [
 ], optional = true, default-features = false }
 serde = { version = "1", default-features = false, optional = true }
 rayon = { version = "1", default-features = false, optional = true }
+futures-lite = { version = "2.0.1", default-features = false, optional = true }
+async-io = { version = "2.0.0", optional = true }
 
 [target.'cfg(target_arch = "wasm32")'.dependencies]
 web-time = { version = "1.1", default-features = false, optional = true }

crates/bevy_platform/Cargo.toml

@@ -0,0 +1,32 @@
+//! Platform-aware future utilities.
+
+crate::cfg::switch! {
+    #[cfg(feature = "async-io")] => {
+        pub use async_io::block_on;
+    }
+    #[cfg(feature = "futures-lite")] => {
+        pub use futures_lite::future::block_on;
+    }
+    _ => {
+        /// Blocks on the supplied `future`.
+        /// This implementation will busy-wait until it is completed.
+        /// Consider enabling the `async-io` or `futures-lite` features.
+        pub fn block_on<T>(future: impl Future<Output = T>) -> T {
+            use core::task::{Poll, Context};
+
+            // Pin the future on the stack.
+            let mut future = core::pin::pin!(future);
+
+            // We don't care about the waker as we're just going to poll as fast as possible.
+            let cx = &mut Context::from_waker(core::task::Waker::noop());
+
+            // Keep polling until the future is ready.
+            loop {
+                match future.as_mut().poll(cx) {
+                    Poll::Ready(output) => return output,
+                    Poll::Pending => core::hint::spin_loop(),
+                }
+            }
+        }
+    }
+}

crates/bevy_platform/src/future.rs

@@ -21,6 +21,7 @@ cfg::alloc! {
 
 pub mod cell;
 pub mod cfg;
+pub mod future;
 pub mod hash;
 pub mod sync;
 pub mod thread;

crates/bevy_platform/src/lib.rs

@@ -25,11 +25,11 @@ multi_threaded = [
 async_executor = ["bevy_platform/std", "dep:async-executor", "futures-lite"]
 
 # Provide an implementation of `block_on` from  `futures-lite`.
-futures-lite = ["bevy_platform/std", "futures-lite/std"]
+futures-lite = ["bevy_platform/futures-lite"]
 
 # Use async-io's implementation of block_on instead of futures-lite's implementation.
 # This is preferred if your application uses async-io.
-async-io = ["bevy_platform/std", "dep:async-io"]
+async-io = ["bevy_platform/async-io"]
 
 [dependencies]
 bevy_platform = { path = "../bevy_platform", version = "0.19.0-dev", default-features = false, features = [
@@ -46,7 +46,6 @@ derive_more = { version = "2", default-features = false, features = [
 ] }
 async-executor = { version = "1.11", optional = true }
 async-channel = { version = "2.3.0", optional = true }
-async-io = { version = "2.0.0", optional = true }
 concurrent-queue = { version = "2.0.0", optional = true }
 atomic-waker = { version = "1", default-features = false }
 crossbeam-queue = { version = "0.3", default-features = false, features = [

crates/bevy_tasks/Cargo.toml

@@ -28,15 +28,6 @@ pub mod cfg {
             conditional_send
         }
 
-        #[cfg(feature = "async-io")] => {
-            /// Indicates `async-io` will be used for the implementation of `block_on`.
-            async_io
-        }
-
-        #[cfg(feature = "futures-lite")] => {
-            /// Indicates `futures-lite` will be used for the implementation of `block_on`.
-            futures_lite
-        }
     }
 }
 
@@ -114,36 +105,7 @@ cfg::multi_threaded! {
     }
 }
 
-cfg::switch! {
-    cfg::async_io => {
-        pub use async_io::block_on;
-    }
-    cfg::futures_lite => {
-        pub use futures_lite::future::block_on;
-    }
-    _ => {
-        /// Blocks on the supplied `future`.
-        /// This implementation will busy-wait until it is completed.
-        /// Consider enabling the `async-io` or `futures-lite` features.
-        pub fn block_on<T>(future: impl Future<Output = T>) -> T {
-            use core::task::{Poll, Context};
-
-            // Pin the future on the stack.
-            let mut future = core::pin::pin!(future);
-
-            // We don't care about the waker as we're just going to poll as fast as possible.
-            let cx = &mut Context::from_waker(core::task::Waker::noop());
-
-            // Keep polling until the future is ready.
-            loop {
-                match future.as_mut().poll(cx) {
-                    Poll::Ready(output) => return output,
-                    Poll::Pending => core::hint::spin_loop(),
-                }
-            }
-        }
-    }
-}
+pub use bevy_platform::future::block_on;
 
 /// The tasks prelude.
 ///

crates/bevy_tasks/src/lib.rs

@@ -9,11 +9,13 @@ license = "MIT OR Apache-2.0"
 keywords = ["bevy"]
 
 [features]
-default = ["parallel"]
+default = ["parallel", "buffered_channel"]
 
 # Provides access to the `Parallel` type.
 parallel = ["bevy_platform/std", "dep:thread_local"]
 
+buffered_channel = ["bevy_platform/std", "dep:async-channel"]
+
 std = ["disqualified/alloc"]
 
 debug = ["bevy_platform/alloc"]
@@ -23,9 +25,13 @@ bevy_platform = { path = "../bevy_platform", version = "0.19.0-dev", default-fea
 
 disqualified = { version = "1.0", default-features = false }
 thread_local = { version = "1.0", optional = true }
+async-channel = { version = "2.3.0", optional = true }
 
 [dev-dependencies]
 static_assertions = "1.1.0"
+bevy_ecs = { path = "../bevy_ecs", version = "0.19.0-dev", default-features = false }
+bevy_app = { path = "../bevy_app", version = "0.19.0-dev", default-features = false }
+bevy_tasks = { path = "../bevy_tasks", version = "0.19.0-dev", default-features = false }
 
 [lints]
 workspace = true