Cast benchmark (#817)

- optimized ray and shape cast : 2x faster
- update mass options instead of automatic mass
- fixes for 32-bit Windows build
- b2TreeStats for measuring query performance
- reduced tree node size from 48-bytes to 40-bytes
- fixes for 32-bit Neon
- MSVC warning level 4 and fixes
#814, #815, #813, #809

Author: erincatto

.github/workflows/build.yml

@@ -75,7 +75,6 @@ jobs:
         arch: x64
 
     - name: Configure CMake
-      # enkiTS is failing ASAN on windows
       run: cmake -B ${{github.workspace}}/build -DCMAKE_BUILD_TYPE=${{env.BUILD_TYPE}} -DBOX2D_SAMPLES=OFF -DBOX2D_SANITIZE=ON -DBUILD_SHARED_LIBS=OFF
       # run: cmake -B ${{github.workspace}}/build -DCMAKE_BUILD_TYPE=${{env.BUILD_TYPE}} -DBOX2D_SAMPLES=OFF -DBUILD_SHARED_LIBS=OFF
 
@@ -85,4 +84,22 @@ jobs:
     - name: Test
       working-directory: ${{github.workspace}}/build
       run: ./bin/${{env.BUILD_TYPE}}/test
+      
+  samples-windows:
+    name: windows
+    runs-on: windows-latest
+    steps:
+
+    - uses: actions/checkout@v4
+
+    - name: Setup MSVC dev command prompt
+      uses: TheMrMilchmann/setup-msvc-dev@v3
+      with:
+        arch: x64
+
+    - name: Configure CMake
+      run: cmake -B ${{github.workspace}}/build -DCMAKE_BUILD_TYPE=Release -DBOX2D_SAMPLES=ON -DBUILD_SHARED_LIBS=OFF -DBOX2D_UNIT_TESTS=OFF
+      
+    - name: Build
+      run: cmake --build ${{github.workspace}}/build --config Release
 

include/box2d/box2d.h

@@ -51,48 +51,53 @@ B2_API b2SensorEvents b2World_GetSensorEvents( b2WorldId worldId );
 B2_API b2ContactEvents b2World_GetContactEvents( b2WorldId worldId );
 
 /// Overlap test for all shapes that *potentially* overlap the provided AABB
-B2_API void b2World_OverlapAABB( b2WorldId worldId, b2AABB aabb, b2QueryFilter filter, b2OverlapResultFcn* fcn, void* context );
+B2_API b2TreeStats b2World_OverlapAABB( b2WorldId worldId, b2AABB aabb, b2QueryFilter filter, b2OverlapResultFcn* fcn,
+											  void* context );
 
 /// Overlap test for for all shapes that overlap the provided circle
-B2_API void b2World_OverlapCircle( b2WorldId worldId, const b2Circle* circle, b2Transform transform, b2QueryFilter filter,
-								   b2OverlapResultFcn* fcn, void* context );
+B2_API b2TreeStats b2World_OverlapCircle( b2WorldId worldId, const b2Circle* circle, b2Transform transform,
+												b2QueryFilter filter, b2OverlapResultFcn* fcn, void* context );
 
 /// Overlap test for all shapes that overlap the provided capsule
-B2_API void b2World_OverlapCapsule( b2WorldId worldId, const b2Capsule* capsule, b2Transform transform, b2QueryFilter filter,
-									b2OverlapResultFcn* fcn, void* context );
+B2_API b2TreeStats b2World_OverlapCapsule( b2WorldId worldId, const b2Capsule* capsule, b2Transform transform,
+												 b2QueryFilter filter, b2OverlapResultFcn* fcn, void* context );
 
 /// Overlap test for all shapes that overlap the provided polygon
-B2_API void b2World_OverlapPolygon( b2WorldId worldId, const b2Polygon* polygon, b2Transform transform, b2QueryFilter filter,
-									b2OverlapResultFcn* fcn, void* context );
+B2_API b2TreeStats b2World_OverlapPolygon( b2WorldId worldId, const b2Polygon* polygon, b2Transform transform,
+												 b2QueryFilter filter, b2OverlapResultFcn* fcn, void* context );
 
 /// Cast a ray into the world to collect shapes in the path of the ray.
 /// Your callback function controls whether you get the closest point, any point, or n-points.
 /// The ray-cast ignores shapes that contain the starting point.
+/// @note The callback function may receive shapes in any order
 /// @param worldId The world to cast the ray against
 /// @param origin The start point of the ray
 /// @param translation The translation of the ray from the start point to the end point
 /// @param filter Contains bit flags to filter unwanted shapes from the results
 /// @param fcn A user implemented callback function
 /// @param context A user context that is passed along to the callback function
-/// @note The callback function may receive shapes in any order
-B2_API void b2World_CastRay( b2WorldId worldId, b2Vec2 origin, b2Vec2 translation, b2QueryFilter filter, b2CastResultFcn* fcn,
-							 void* context );
+///	@return traversal performance counters
+B2_API b2TreeStats b2World_CastRay( b2WorldId worldId, b2Vec2 origin, b2Vec2 translation, b2QueryFilter filter,
+										  b2CastResultFcn* fcn, void* context );
 
 /// Cast a ray into the world to collect the closest hit. This is a convenience function.
 /// This is less general than b2World_CastRay() and does not allow for custom filtering.
 B2_API b2RayResult b2World_CastRayClosest( b2WorldId worldId, b2Vec2 origin, b2Vec2 translation, b2QueryFilter filter );
 
 /// Cast a circle through the world. Similar to a cast ray except that a circle is cast instead of a point.
-B2_API void b2World_CastCircle( b2WorldId worldId, const b2Circle* circle, b2Transform originTransform, b2Vec2 translation,
-								b2QueryFilter filter, b2CastResultFcn* fcn, void* context );
+///	@see b2World_CastRay
+B2_API b2TreeStats b2World_CastCircle( b2WorldId worldId, const b2Circle* circle, b2Transform originTransform,
+											 b2Vec2 translation, b2QueryFilter filter, b2CastResultFcn* fcn, void* context );
 
 /// Cast a capsule through the world. Similar to a cast ray except that a capsule is cast instead of a point.
-B2_API void b2World_CastCapsule( b2WorldId worldId, const b2Capsule* capsule, b2Transform originTransform, b2Vec2 translation,
-								 b2QueryFilter filter, b2CastResultFcn* fcn, void* context );
+///	@see b2World_CastRay
+B2_API b2TreeStats b2World_CastCapsule( b2WorldId worldId, const b2Capsule* capsule, b2Transform originTransform,
+											  b2Vec2 translation, b2QueryFilter filter, b2CastResultFcn* fcn, void* context );
 
 /// Cast a polygon through the world. Similar to a cast ray except that a polygon is cast instead of a point.
-B2_API void b2World_CastPolygon( b2WorldId worldId, const b2Polygon* polygon, b2Transform originTransform, b2Vec2 translation,
-								 b2QueryFilter filter, b2CastResultFcn* fcn, void* context );
+///	@see b2World_CastRay
+B2_API b2TreeStats b2World_CastPolygon( b2WorldId worldId, const b2Polygon* polygon, b2Transform originTransform,
+											  b2Vec2 translation, b2QueryFilter filter, b2CastResultFcn* fcn, void* context );
 
 /// Enable/disable sleep. If your application does not need sleeping, you can gain some performance
 /// by disabling sleep completely at the world level.
@@ -183,6 +188,9 @@ B2_API b2Counters b2World_GetCounters( b2WorldId worldId );
 /// Dump memory stats to box2d_memory.txt
 B2_API void b2World_DumpMemoryStats( b2WorldId worldId );
 
+/// todo testing
+B2_API void b2World_RebuildStaticTree( b2WorldId worldId );
+
 /** @} */
 
 /**
@@ -466,8 +474,10 @@ B2_API b2ShapeId b2CreateCapsuleShape( b2BodyId bodyId, const b2ShapeDef* def, c
 /// @return the shape id for accessing the shape
 B2_API b2ShapeId b2CreatePolygonShape( b2BodyId bodyId, const b2ShapeDef* def, const b2Polygon* polygon );
 
-/// Destroy a shape
-B2_API void b2DestroyShape( b2ShapeId shapeId );
+/// Destroy a shape. You may defer the body mass update which can improve performance if several shapes on a
+///	body are destroyed at once.
+///	@see b2Body_ApplyMassFromShapes
+B2_API void b2DestroyShape( b2ShapeId shapeId, bool updateBodyMass );
 
 /// Shape identifier validation. Provides validation for up to 64K allocations.
 B2_API bool b2Shape_IsValid( b2ShapeId id );
@@ -492,9 +502,9 @@ B2_API void b2Shape_SetUserData( b2ShapeId shapeId, void* userData );
 B2_API void* b2Shape_GetUserData( b2ShapeId shapeId );
 
 /// Set the mass density of a shape, typically in kg/m^2.
-/// This will not update the mass properties on the parent body.
+/// This will optionally update the mass properties on the parent body.
 /// @see b2ShapeDef::density, b2Body_ApplyMassFromShapes
-B2_API void b2Shape_SetDensity( b2ShapeId shapeId, float density );
+B2_API void b2Shape_SetDensity( b2ShapeId shapeId, float density, bool updateBodyMass );
 
 /// Get the density of a shape, typically in kg/m^2
 B2_API float b2Shape_GetDensity( b2ShapeId shapeId );

include/box2d/collision.h

@@ -29,7 +29,7 @@ typedef struct b2Hull b2Hull;
 /// don't use more vertices.
 #define b2_maxPolygonVertices 8
 
-/// Low level ray-cast input data
+/// Low level ray cast input data
 typedef struct b2RayCastInput
 {
 	/// Start point of the ray cast
@@ -63,7 +63,7 @@ typedef struct b2ShapeCastInput
 	float maxFraction;
 } b2ShapeCastInput;
 
-/// Low level ray-cast or shape-cast output data
+/// Low level ray cast or shape-cast output data
 typedef struct b2CastOutput
 {
 	/// The surface normal at the hit point
@@ -566,16 +566,16 @@ B2_API b2Manifold b2CollideSegmentAndPolygon( const b2Segment* segmentA, b2Trans
 											  b2Transform xfB );
 
 /// Compute the contact manifold between a chain segment and a circle
-B2_API b2Manifold b2CollideChainSegmentAndCircle( const b2ChainSegment* segmentA, b2Transform xfA,
-												   const b2Circle* circleB, b2Transform xfB );
+B2_API b2Manifold b2CollideChainSegmentAndCircle( const b2ChainSegment* segmentA, b2Transform xfA, const b2Circle* circleB,
+												  b2Transform xfB );
 
 /// Compute the contact manifold between a chain segment and a capsule
-B2_API b2Manifold b2CollideChainSegmentAndCapsule( const b2ChainSegment* segmentA, b2Transform xfA,
-													const b2Capsule* capsuleB, b2Transform xfB, b2DistanceCache* cache );
+B2_API b2Manifold b2CollideChainSegmentAndCapsule( const b2ChainSegment* segmentA, b2Transform xfA, const b2Capsule* capsuleB,
+												   b2Transform xfB, b2DistanceCache* cache );
 
 /// Compute the contact manifold between a chain segment and a rounded polygon
-B2_API b2Manifold b2CollideChainSegmentAndPolygon( const b2ChainSegment* segmentA, b2Transform xfA,
-													const b2Polygon* polygonB, b2Transform xfB, b2DistanceCache* cache );
+B2_API b2Manifold b2CollideChainSegmentAndPolygon( const b2ChainSegment* segmentA, b2Transform xfA, const b2Polygon* polygonB,
+												   b2Transform xfB, b2DistanceCache* cache );
 
 /**@}*/
 
@@ -602,8 +602,7 @@ B2_API b2Manifold b2CollideChainSegmentAndPolygon( const b2ChainSegment* segment
 /// The default category bit for a tree proxy. Used for collision filtering.
 #define b2_defaultCategoryBits ( 1 )
 
-/// Convenience mask bits to use when you don't need collision filtering and just want
-/// all results.
+/// Convenience mask bits to use when you don't need collision filtering and just want all results.
 #define b2_defaultMaskBits ( UINT64_MAX )
 
 /// A node in the dynamic tree. This is private data placed here for performance reasons.
@@ -617,31 +616,27 @@ typedef struct b2TreeNode
 
 	union
 	{
-		/// The node parent index
+		/// The node parent index (allocated node)
 		int32_t parent;
 
-		/// The node freelist next index
+		/// The node freelist next index (free node)
 		int32_t next;
 	}; // 4
 
-	/// Child 1 index
+	/// Child 1 index (internal node)
 	int32_t child1; // 4
 
-	/// Child 2 index
-	int32_t child2; // 4
-
-	/// User data
-	// todo could be union with child index
-	int32_t userData; // 4
-
-	/// Leaf = 0, free node = -1
-	int16_t height; // 2
+	union
+	{
+		/// Child 2 index (internal node)
+		int32_t child2;
 
-	/// Has the AABB been enlarged?
-	bool enlarged; // 1
+		/// User data (leaf node)
+		int32_t userData;
+	}; // 4
 
-	/// Padding for clarity
-	char pad[5];
+	uint16_t height; // 2
+	uint16_t flags;	 // 2
 } b2TreeNode;
 
 /// The dynamic tree structure. This should be considered private data.
@@ -682,6 +677,16 @@ typedef struct b2DynamicTree
 	int32_t rebuildCapacity;
 } b2DynamicTree;
 
+/// These are performance results returned by dynamic tree queries.
+typedef struct b2TreeStats
+{
+	/// Number of internal nodes visited during the query
+	int32_t nodeVisits;
+
+	/// Number of leaf nodes visited during the query
+	int32_t leafVisits;
+} b2TreeStats;
+
 /// Constructing the tree initializes the node pool.
 B2_API b2DynamicTree b2DynamicTree_Create( void );
 
@@ -705,49 +710,53 @@ B2_API void b2DynamicTree_EnlargeProxy( b2DynamicTree* tree, int32_t proxyId, b2
 typedef bool b2TreeQueryCallbackFcn( int32_t proxyId, int32_t userData, void* context );
 
 /// Query an AABB for overlapping proxies. The callback class is called for each proxy that overlaps the supplied AABB.
-B2_API void b2DynamicTree_Query( const b2DynamicTree* tree, b2AABB aabb, uint64_t maskBits, b2TreeQueryCallbackFcn* callback,
-								 void* context );
+///	@return performance data
+B2_API b2TreeStats b2DynamicTree_Query( const b2DynamicTree* tree, b2AABB aabb, uint64_t maskBits,
+											  b2TreeQueryCallbackFcn* callback, void* context );
 
-/// This function receives clipped raycast input for a proxy. The function
+/// This function receives clipped ray cast input for a proxy. The function
 /// returns the new ray fraction.
 /// - return a value of 0 to terminate the ray cast
 /// - return a value less than input->maxFraction to clip the ray
 /// - return a value of input->maxFraction to continue the ray cast without clipping
 typedef float b2TreeRayCastCallbackFcn( const b2RayCastInput* input, int32_t proxyId, int32_t userData, void* context );
 
-/// Ray-cast against the proxies in the tree. This relies on the callback
-/// to perform a exact ray-cast in the case were the proxy contains a shape.
+/// Ray cast against the proxies in the tree. This relies on the callback
+/// to perform a exact ray cast in the case were the proxy contains a shape.
 /// The callback also performs the any collision filtering. This has performance
 /// roughly equal to k * log(n), where k is the number of collisions and n is the
 /// number of proxies in the tree.
 /// Bit-wise filtering using mask bits can greatly improve performance in some scenarios.
+///	However, this filtering may be approximate, so the user should still apply filtering to results.
 /// @param tree the dynamic tree to ray cast
-/// @param input the ray-cast input data. The ray extends from p1 to p1 + maxFraction * (p2 - p1)
-/// @param maskBits filter bits: `bool accept = (maskBits & node->categoryBits) != 0;`
+/// @param input the ray cast input data. The ray extends from p1 to p1 + maxFraction * (p2 - p1)
+/// @param maskBits mask bit hint: `bool accept = (maskBits & node->categoryBits) != 0;`
 /// @param callback a callback class that is called for each proxy that is hit by the ray
 /// @param context user context that is passed to the callback
-B2_API void b2DynamicTree_RayCast( const b2DynamicTree* tree, const b2RayCastInput* input, uint64_t maskBits,
-								   b2TreeRayCastCallbackFcn* callback, void* context );
+///	@return performance data
+B2_API b2TreeStats b2DynamicTree_RayCast( const b2DynamicTree* tree, const b2RayCastInput* input, uint64_t maskBits,
+												b2TreeRayCastCallbackFcn* callback, void* context );
 
-/// This function receives clipped ray-cast input for a proxy. The function
+/// This function receives clipped ray cast input for a proxy. The function
 /// returns the new ray fraction.
-/// - return a value of 0 to terminate the ray-cast
+/// - return a value of 0 to terminate the ray cast
 /// - return a value less than input->maxFraction to clip the ray
 /// - return a value of input->maxFraction to continue the ray cast without clipping
 typedef float b2TreeShapeCastCallbackFcn( const b2ShapeCastInput* input, int32_t proxyId, int32_t userData, void* context );
 
-/// Ray-cast against the proxies in the tree. This relies on the callback
-/// to perform a exact ray-cast in the case were the proxy contains a shape.
+/// Ray cast against the proxies in the tree. This relies on the callback
+/// to perform a exact ray cast in the case were the proxy contains a shape.
 /// The callback also performs the any collision filtering. This has performance
 /// roughly equal to k * log(n), where k is the number of collisions and n is the
 /// number of proxies in the tree.
 /// @param tree the dynamic tree to ray cast
-/// @param input the ray-cast input data. The ray extends from p1 to p1 + maxFraction * (p2 - p1).
+/// @param input the ray cast input data. The ray extends from p1 to p1 + maxFraction * (p2 - p1).
 /// @param maskBits filter bits: `bool accept = (maskBits & node->categoryBits) != 0;`
 /// @param callback a callback class that is called for each proxy that is hit by the shape
 /// @param context user context that is passed to the callback
-B2_API void b2DynamicTree_ShapeCast( const b2DynamicTree* tree, const b2ShapeCastInput* input, uint64_t maskBits,
-									 b2TreeShapeCastCallbackFcn* callback, void* context );
+///	@return performance data
+B2_API b2TreeStats b2DynamicTree_ShapeCast( const b2DynamicTree* tree, const b2ShapeCastInput* input, uint64_t maskBits,
+												  b2TreeShapeCastCallbackFcn* callback, void* context );
 
 /// Validate this tree. For testing.
 B2_API void b2DynamicTree_Validate( const b2DynamicTree* tree );
@@ -781,7 +790,6 @@ B2_API void b2DynamicTree_ShiftOrigin( b2DynamicTree* tree, b2Vec2 newOrigin );
 B2_API int b2DynamicTree_GetByteCount( const b2DynamicTree* tree );
 
 /// Get proxy user data
-/// @return the proxy user data or 0 if the id is invalid
 B2_INLINE int32_t b2DynamicTree_GetUserData( const b2DynamicTree* tree, int32_t proxyId )
 {
 	return tree->nodes[proxyId].userData;

include/box2d/types.h

@@ -58,6 +58,8 @@ typedef struct b2RayResult
 	b2Vec2 point;
 	b2Vec2 normal;
 	float fraction;
+	int nodeVisits;
+	int leafVisits;
 	bool hit;
 } b2RayResult;
 
@@ -220,10 +222,6 @@ typedef struct b2BodyDef
 	/// Used to disable a body. A disabled body does not move or collide.
 	bool isEnabled;
 
-	/// Automatically compute mass and related properties on this body from shapes.
-	/// Triggers whenever a shape is add/removed/changed. Default is true.
-	bool automaticMass;
-
 	/// This allows this body to bypass rotational speed limits. Should only be used
 	/// for circular objects, like wheels.
 	bool allowFastRotation;
@@ -367,6 +365,9 @@ typedef struct b2ShapeDef
 	/// This is implicitly always true for sensors.
 	bool forceContactCreation;
 
+	/// Should the body update the mass properties when this shape is created. Default is true.
+	bool updateBodyMass;
+
 	/// Used internally to detect a valid definition. DO NOT SET.
 	int32_t internalValue;
 } b2ShapeDef;

samples/main.cpp

@@ -68,7 +68,7 @@ void* AllocFcn( uint32_t size, int32_t alignment )
 	size_t sizeAligned = ( ( size - 1 ) | ( alignment - 1 ) ) + 1;
 	assert( ( sizeAligned & ( alignment - 1 ) ) == 0 );
 
-#if defined( _WIN64 )
+#if defined( _WIN64 ) || defined( _WIN32 )
 	void* ptr = _aligned_malloc( sizeAligned, alignment );
 #else
 	void* ptr = aligned_alloc( alignment, sizeAligned );
@@ -79,7 +79,7 @@ void* AllocFcn( uint32_t size, int32_t alignment )
 
 void FreeFcn( void* mem )
 {
-#if defined( _WIN64 )
+#if defined( _WIN64 ) || defined( _WIN32 )
 	_aligned_free( mem );
 #else
 	free( mem );