Parallel Transform Propagation (#17840)

# Objective

- Make transform propagation faster.

## Solution

- Work sharing worker threads
- Parallel tree traversal excluding leaves
- Second cache friendly wide pass over all leaves
- 3-10x faster than main

## Testing

- Tracy
- Caldera hotel is showing 3-7x faster on my M4 Max. Timing for bevy's
existing transform system shifts wildly run to run, so I don't know that
I would advertise a particular number. But this implementation is faster
in a... statistically significant way.

![image](https://github.com/user-attachments/assets/b4a48fc6-86b8-4b9c-8c5e-5b746c1d163b)

---------

Co-authored-by: Alice Cecile <alice.i.cecile@gmail.com>
Co-authored-by: François Mockers <mockersf@gmail.com>

crates/bevy_transform/Cargo.toml

@@ -12,8 +12,11 @@ keywords = ["bevy"]
 # bevy
 bevy_app = { path = "../bevy_app", version = "0.16.0-dev", default-features = false, optional = true }
 bevy_ecs = { path = "../bevy_ecs", version = "0.16.0-dev", default-features = false, optional = true }
+bevy_log = { path = "../bevy_log", version = "0.16.0-dev", default-features = false, optional = true }
 bevy_math = { path = "../bevy_math", version = "0.16.0-dev", default-features = false }
 bevy_reflect = { path = "../bevy_reflect", version = "0.16.0-dev", default-features = false, optional = true }
+bevy_tasks = { path = "../bevy_tasks", version = "0.16.0-dev", default-features = false, optional = true }
+bevy_utils = { path = "../bevy_utils", version = "0.16.0-dev", default-features = false, optional = true }
 serde = { version = "1", default-features = false, features = [
   "derive",
 ], optional = true }
@@ -30,7 +33,7 @@ approx = "0.5.1"
 [features]
 # Turning off default features leaves you with a barebones
 # definition of transform.
-default = ["std", "bevy-support", "bevy_reflect"]
+default = ["std", "bevy-support", "bevy_reflect", "async_executor"]
 
 # Functionality
 
@@ -52,6 +55,21 @@ bevy_reflect = [
   "bevy_app/bevy_reflect",
 ]
 
+# Executor Backend
+
+## Uses `async-executor` as a task execution backend.
+## This backend is incompatible with `no_std` targets.
+async_executor = [
+  "std",
+  "dep:bevy_tasks",
+  "dep:bevy_utils",
+  "bevy_tasks/async_executor",
+]
+
+## Uses `edge-executor` as a task execution backend.
+## Use this instead of `async-executor` if working on a `no_std` target.
+edge_executor = ["dep:bevy_tasks", "bevy_tasks/edge_executor"]
+
 # Platform Compatibility
 
 ## Allows access to the `std` crate. Enabling this feature will prevent compilation
@@ -60,12 +78,24 @@ bevy_reflect = [
 std = [
   "alloc",
   "bevy_app?/std",
+  "bevy_log",
   "bevy_ecs?/std",
   "bevy_math/std",
   "bevy_reflect?/std",
+  "bevy_tasks?/std",
+  "bevy_utils?/std",
   "serde?/std",
 ]
 
+## `critical-section` provides the building blocks for synchronization primitives
+## on all platforms, including `no_std`.
+critical-section = [
+  "bevy_app?/critical-section",
+  "bevy_ecs?/critical-section",
+  "bevy_tasks?/critical-section",
+  "bevy_reflect?/critical-section",
+]
+
 ## Allows access to the `alloc` crate.
 alloc = ["serde?/alloc"]
 

crates/bevy_transform/src/plugins.rs

@@ -1,4 +1,6 @@
-use crate::systems::{propagate_transforms, sync_simple_transforms};
+use crate::systems::{
+    compute_transform_leaves, propagate_parent_transforms, sync_simple_transforms,
+};
 use bevy_app::{App, Plugin, PostStartup, PostUpdate};
 use bevy_ecs::schedule::{IntoSystemConfigs, IntoSystemSetConfigs, SystemSet};
 
@@ -32,14 +34,12 @@ impl Plugin for TransformPlugin {
         .add_systems(
             PostStartup,
             (
-                sync_simple_transforms
-                    .in_set(TransformSystem::TransformPropagate)
-                    // FIXME: https://github.com/bevyengine/bevy/issues/4381
-                    // These systems cannot access the same entities,
-                    // due to subtle query filtering that is not yet correctly computed in the ambiguity detector
-                    .ambiguous_with(PropagateTransformsSet),
-                propagate_transforms.in_set(PropagateTransformsSet),
-            ),
+                propagate_parent_transforms,
+                (compute_transform_leaves, sync_simple_transforms)
+                    .ambiguous_with(TransformSystem::TransformPropagate),
+            )
+                .chain()
+                .in_set(PropagateTransformsSet),
         )
         .configure_sets(
             PostUpdate,
@@ -48,11 +48,12 @@ impl Plugin for TransformPlugin {
         .add_systems(
             PostUpdate,
             (
-                sync_simple_transforms
-                    .in_set(TransformSystem::TransformPropagate)
-                    .ambiguous_with(PropagateTransformsSet),
-                propagate_transforms.in_set(PropagateTransformsSet),
-            ),
+                propagate_parent_transforms,
+                (compute_transform_leaves, sync_simple_transforms) // TODO: Adjust the internal parallel queries to make these parallel systems more efficiently share and fill CPU time.
+                    .ambiguous_with(TransformSystem::TransformPropagate),
+            )
+                .chain()
+                .in_set(PropagateTransformsSet),
         );
     }
 }