CollisionWorld.CastRay 10x slower than Physics.Raycast.

I have two examples scenes, which both load the same Mesh Collider, once as a GameObject, once as a Unity.Physics.Collider. Then I perform a simple raycast on click. It looks like CollisionWorld.CastRay is 10x as slow as Physics.CastRay.

I have enabled burst, disabled Safety Checks and Native Debug Mode Compilation.


Tried it with and without burst.

CollisionWorld.CastRay is using the example code from Unity.Physics https://docs.unity3d.com/Packages/com.unity.physics@0.6/manual/collision_queries.html

Code (CSharp):

1.  
2. // Physics.CastRay
3. // bool result = Physics.Raycast(ray, out var hit);
4. Hit (-15,3, -1,0, -93,2) after 0 ms or 63 ticks
5. Hit (-15,3, -1,0, -93,2) after 0 ms or 54 ticks
6.  

Code (CSharp):

1.  
2. // CollisionWorld.CastRay
3. Hit float3(-15,33019f, 0,474617f, -73,96638f) after 0 ms or 755 ticks
4. Hit float3(-15,33019f, 0,474617f, -73,96638f) after 0 ms or 747 ticks
5.  

And while it becomes faster in an IL2CPP Release build compared to the editor, it is still much slower. I am wondering why it is so much slower and what differences could influence this. I would have expected it to be pretty much the same.

Code (CSharp):

1. // IL2CPP Release Physics.Raycast
2. Hit (3,4, -0,6, -4,6) after 0 ms or 36 ticks

Code (CSharp):

1. // IL2CPP Release CollisionWorld.CastRay
2. Hit float3(0,8347985f, 2,100365f, -58,24535f) after 0 ms or 254 ticks

I am wondering if this is because of the general degradation noticed here https://forum.unity.com/threads/unity-physics-performance-degraded-since-2020-3-16.1207051/

I am using Unity 2020.3.27f1 with the following packages

Code (csharp):

1. "com.havok.physics": "0.6.0-preview.3",
2. "com.unity.jobs": "0.8.0-preview.23",
3. "com.unity.mathematics": "1.2.5",
4. "com.unity.entities": "0.17.0-preview.42",
5. "com.unity.rendering.hybrid": "0.11.0-preview.43",
6. "com.unity.collections": "0.15.0-preview.21",
7. "com.unity.platforms": "0.5.0-preview.6",
8. "com.unity.platforms.windows": "0.10.0-preview.10",
9.  

 

Are you doing this on main thread? Because yes it's going to be reasonably slow then. Try it with burst.

 

@tertle as written I have tried it with burst.

Yes, both cases are called from the main thread from a MonoBehaviour.Update, as Physics.CastRay is a sync call as well and completes within the same frame. Otherwise it would not be that easy to compare both, wouldn't it be ?

Code (csharp):

1.  
2. private void Update(){
3.   if (Input.GetKeyDown(KeyCode.Mouse0) && Input.GetKey(KeyCode.LeftAlt))
4.     ...
5.     bool result = Physics.Raycast(ray, out var hit);
6.     ...
7.  

Code (csharp):

1.  
2. private void Update(){
3.   if (Input.GetKeyDown(KeyCode.Mouse0) && Input.GetKey(KeyCode.LeftAlt))
4.     ...
5.     var physicsWorldSystem =
6. World.DefaultGameObjectInjectionWorld.GetOrCreateSystem<Unity.Physics.Systems.BuildPhysicsWorld>();
7.     CollisionWorld collisionWorld = physicsWorldSystem.PhysicsWorld.CollisionWorld;
8.     var input = new RaycastInput
9.     {
10.       Start = RayFrom,
11.       End   = RayTo,
12.       Filter = new CollisionFilter
13.       {
14.         BelongsTo    = ~0u,
15.         CollidesWith = ~0u, // all 1s, so all layers, collide with everything
16.         GroupIndex   = 0
17.       }
18.     };
19.     bool haveHit = collisionWorld.CastRay(input, out hit);
20.     ...
21.  

 

Hmmm that is odd. Do you have your benchmark in a repo that we could run?