diff --git a/Extras/AllBulletDemosOSX/AllBulletDemos.xcodeproj/project.pbxproj b/Extras/AllBulletDemosOSX/AllBulletDemos.xcodeproj/project.pbxproj
index 6915e5075..448c5c699 100644
--- a/Extras/AllBulletDemosOSX/AllBulletDemos.xcodeproj/project.pbxproj
+++ b/Extras/AllBulletDemosOSX/AllBulletDemos.xcodeproj/project.pbxproj
@@ -181,6 +181,7 @@
 		8D11072B0486CEB800E47090 /* InfoPlist.strings in Resources */ = {isa = PBXBuildFile; fileRef = 089C165CFE840E0CC02AAC07 /* InfoPlist.strings */; };
 		8D11072D0486CEB800E47090 /* main.m in Sources */ = {isa = PBXBuildFile; fileRef = 29B97316FDCFA39411CA2CEA /* main.m */; settings = {ATTRIBUTES = (); }; };
 		8D11072F0486CEB800E47090 /* Cocoa.framework in Frameworks */ = {isa = PBXBuildFile; fileRef = 1058C7A1FEA54F0111CA2CBB /* Cocoa.framework */; };
+		BCEDD8EC112AF72F00F49466 /* btSerializer.cpp in Sources */ = {isa = PBXBuildFile; fileRef = BCEDD8EB112AF72F00F49466 /* btSerializer.cpp */; };
 /* End PBXBuildFile section */
 
 /* Begin PBXFileReference section */
@@ -569,6 +570,8 @@
 		7AE465E40F4CA2DC007715D6 /* ForkLiftDemo.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; name = ForkLiftDemo.h; path = ../../Demos/ForkLiftDemo/ForkLiftDemo.h; sourceTree = SOURCE_ROOT; };
 		8D1107310486CEB800E47090 /* Info.plist */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = text.plist.xml; path = Info.plist; sourceTree = "<group>"; };
 		8D1107320486CEB800E47090 /* AllBulletDemos.app */ = {isa = PBXFileReference; explicitFileType = wrapper.application; includeInIndex = 0; path = AllBulletDemos.app; sourceTree = BUILT_PRODUCTS_DIR; };
+		BCEDD8EA112AF72F00F49466 /* btSerializer.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; path = btSerializer.h; sourceTree = "<group>"; };
+		BCEDD8EB112AF72F00F49466 /* btSerializer.cpp */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.cpp.cpp; path = btSerializer.cpp; sourceTree = "<group>"; };
 /* End PBXFileReference section */
 
 /* Begin PBXFrameworksBuildPhase section */
@@ -1075,6 +1078,8 @@
 		7ACE567D0DDFA1E600CA06EA /* LinearMath */ = {
 			isa = PBXGroup;
 			children = (
+				BCEDD8EA112AF72F00F49466 /* btSerializer.h */,
+				BCEDD8EB112AF72F00F49466 /* btSerializer.cpp */,
 				7ACE567E0DDFA1E600CA06EA /* btAabbUtil2.h */,
 				7ACE567F0DDFA1E600CA06EA /* btAlignedAllocator.cpp */,
 				7ACE56800DDFA1E600CA06EA /* btAlignedAllocator.h */,
@@ -1398,6 +1403,7 @@
 				7A2B5DB5105483A90069469C /* btGeneric6DofSpringConstraint.cpp in Sources */,
 				7A2B5DB6105483A90069469C /* btHinge2Constraint.cpp in Sources */,
 				7A2B5DB910548D9F0069469C /* btUniversalConstraint.cpp in Sources */,
+				BCEDD8EC112AF72F00F49466 /* btSerializer.cpp in Sources */,
 			);
 			runOnlyForDeploymentPostprocessing = 0;
 		};
diff --git a/src/BulletCollision/CollisionDispatch/btCollisionObject.h b/src/BulletCollision/CollisionDispatch/btCollisionObject.h
index aa618b535..be3d5a710 100644
--- a/src/BulletCollision/CollisionDispatch/btCollisionObject.h
+++ b/src/BulletCollision/CollisionDispatch/btCollisionObject.h
@@ -119,7 +119,8 @@ public:
 		CF_NO_CONTACT_RESPONSE = 4,
 		CF_CUSTOM_MATERIAL_CALLBACK = 8,//this allows per-triangle material (friction/restitution)
 		CF_CHARACTER_OBJECT = 16,
-		CF_DISABLE_VISUALIZE_OBJECT = 32 //disable debug drawing
+		CF_DISABLE_VISUALIZE_OBJECT = 32, //disable debug drawing
+		CF_DISABLE_SPU_COLLISION_PROCESSING = 64//disable parallel/SPU processing
 	};
 
 	enum	CollisionObjectTypes
diff --git a/src/BulletDynamics/Dynamics/btDiscreteDynamicsWorld.cpp b/src/BulletDynamics/Dynamics/btDiscreteDynamicsWorld.cpp
index 2beb3f50f..1c9b74b7e 100644
--- a/src/BulletDynamics/Dynamics/btDiscreteDynamicsWorld.cpp
+++ b/src/BulletDynamics/Dynamics/btDiscreteDynamicsWorld.cpp
@@ -19,6 +19,7 @@ subject to the following restrictions:
 //collision detection
 #include "BulletCollision/CollisionDispatch/btCollisionDispatcher.h"
 #include "BulletCollision/BroadphaseCollision/btSimpleBroadphase.h"
+#include "BulletCollision/BroadphaseCollision/btCollisionAlgorithm.h"
 #include "BulletCollision/CollisionShapes/btCollisionShape.h"
 #include "BulletCollision/CollisionDispatch/btSimulationIslandManager.h"
 #include "LinearMath/btTransformUtil.h"
@@ -738,7 +739,7 @@ void	btDiscreteDynamicsWorld::calculateSimulationIslands()
 }
 
 
-#include "BulletCollision/BroadphaseCollision/btCollisionAlgorithm.h"
+
 
 class btClosestNotMeConvexResultCallback : public btCollisionWorld::ClosestConvexResultCallback
 {
diff --git a/src/BulletMultiThreaded/SpuGatheringCollisionDispatcher.cpp b/src/BulletMultiThreaded/SpuGatheringCollisionDispatcher.cpp
index d9c90c1ca..ee0832f12 100644
--- a/src/BulletMultiThreaded/SpuGatheringCollisionDispatcher.cpp
+++ b/src/BulletMultiThreaded/SpuGatheringCollisionDispatcher.cpp
@@ -127,7 +127,10 @@ public:
 				{
 					int	proxyType0 = colObj0->getCollisionShape()->getShapeType();
 					int	proxyType1 = colObj1->getCollisionShape()->getShapeType();
-					if (m_dispatcher->supportsDispatchPairOnSpu(proxyType0,proxyType1))
+					if (m_dispatcher->supportsDispatchPairOnSpu(proxyType0,proxyType1) 
+						&& (colObj0->getCollisionFlags() != btCollisionObject::CF_DISABLE_SPU_COLLISION_PROCESSING) 
+						&& (colObj1->getCollisionFlags() != btCollisionObject::CF_DISABLE_SPU_COLLISION_PROCESSING) 
+						)
 					{
 						int so = sizeof(SpuContactManifoldCollisionAlgorithm);
 #ifdef ALLOCATE_SEPARATELY
diff --git a/src/BulletMultiThreaded/SpuNarrowPhaseCollisionTask/SpuGatheringCollisionTask.cpp b/src/BulletMultiThreaded/SpuNarrowPhaseCollisionTask/SpuGatheringCollisionTask.cpp
index 4ca2e8757..40feb132b 100644
--- a/src/BulletMultiThreaded/SpuNarrowPhaseCollisionTask/SpuGatheringCollisionTask.cpp
+++ b/src/BulletMultiThreaded/SpuNarrowPhaseCollisionTask/SpuGatheringCollisionTask.cpp
@@ -870,10 +870,13 @@ void	handleCollisionPair(SpuCollisionPairInput& collisionPairInput, CollisionTas
 		for (int i = 0; i < childShapeCount0; ++i)
 		{
 			btCompoundShapeChild& childShape0 = lsMem.compoundShapeData[0].gSubshapes[i];
+			btAssert(!btBroadphaseProxy::isCompound(childShape0.m_childShapeType));
 
 			for (int j = 0; j < childShapeCount1; ++j)
 			{
 				btCompoundShapeChild& childShape1 = lsMem.compoundShapeData[1].gSubshapes[j];
+				btAssert(!btBroadphaseProxy::isCompound(childShape1.m_childShapeType));
+
 
 				/* Create a new collision pair input struct using the two child shapes */
 				SpuCollisionPairInput cinput (collisionPairInput);
@@ -886,9 +889,10 @@ void	handleCollisionPair(SpuCollisionPairInput& collisionPairInput, CollisionTas
 				cinput.m_shapeType1 = childShape1.m_childShapeType;
 				cinput.m_collisionMargin1 = childShape1.m_childMargin;
 				/* Recursively call handleCollisionPair () with new collision pair input */
+				
 				handleCollisionPair(cinput, lsMem, spuContacts,			
 					(ppu_address_t)childShape0.m_childShape, lsMem.compoundShapeData[0].gSubshapeShape[i], 
-					(ppu_address_t)childShape1.m_childShape, lsMem.compoundShapeData[1].gSubshapeShape[j], false); // bug fix: changed index to j.
+					(ppu_address_t)childShape1.m_childShape, lsMem.compoundShapeData[1].gSubshapeShape[j], false);
 			}
 		}
 	}
@@ -910,7 +914,7 @@ void	handleCollisionPair(SpuCollisionPairInput& collisionPairInput, CollisionTas
 		for (int i = 0; i < childShapeCount; ++i)
 		{
 			btCompoundShapeChild& childShape = lsMem.compoundShapeData[0].gSubshapes[i];
-
+			btAssert(!btBroadphaseProxy::isCompound(childShape.m_childShapeType));
 			// Dma the child shape
 			dmaCollisionShape (&lsMem.compoundShapeData[0].gSubshapeShape[i], (ppu_address_t)childShape.m_childShape, 1, childShape.m_childShapeType);
 			cellDmaWaitTagStatusAll(DMA_MASK(1));
@@ -942,6 +946,7 @@ void	handleCollisionPair(SpuCollisionPairInput& collisionPairInput, CollisionTas
 		for (int i = 0; i < childShapeCount; ++i)
 		{
 			btCompoundShapeChild& childShape = lsMem.compoundShapeData[0].gSubshapes[i];
+			btAssert(!btBroadphaseProxy::isCompound(childShape.m_childShapeType));
 			// Dma the child shape
 			dmaCollisionShape (&lsMem.compoundShapeData[0].gSubshapeShape[i], (ppu_address_t)childShape.m_childShape, 1, childShape.m_childShapeType);
 			cellDmaWaitTagStatusAll(DMA_MASK(1));