diff --git a/Demos/Gpu2dDemo/BasicDemo.cpp b/Demos/Gpu2dDemo/BasicDemo.cpp
index 5c5dca2f8..8a16cdc67 100644
--- a/Demos/Gpu2dDemo/BasicDemo.cpp
+++ b/Demos/Gpu2dDemo/BasicDemo.cpp
@@ -19,6 +19,7 @@ subject to the following restrictions:
 #include "BulletMultiThreaded/btGpuUtilsSharedCode.h"
 
 
+
 #include "BulletCollision/CollisionDispatch/btEmptyCollisionAlgorithm.h"
 
 #include "btGpuDemoPairCache.h"
diff --git a/Demos/Gpu3dDemo/BasicDemo3d.cpp b/Demos/Gpu3dDemo/BasicDemo3d.cpp
index 7ad7d2150..13952c43c 100644
--- a/Demos/Gpu3dDemo/BasicDemo3d.cpp
+++ b/Demos/Gpu3dDemo/BasicDemo3d.cpp
@@ -13,9 +13,6 @@ subject to the following restrictions:
 3. This notice may not be removed or altered from any source distribution.
 */
 
-#include "BulletMultiThreaded/btGpuDefines.h"
-#include "BulletMultiThreaded/btGpuUtilsSharedDefs.h"
-#include "BulletMultiThreaded/btGpuUtilsSharedCode.h"
 
 #include "BulletMultiThreaded/SpuNarrowPhaseCollisionTask/SpuGatheringCollisionTask.h"
 #include "BulletMultiThreaded/SpuContactManifoldCollisionAlgorithm.h"
diff --git a/Demos/Gpu3dDemo/CMakeLists.txt b/Demos/Gpu3dDemo/CMakeLists.txt
index 2b3093d43..50656031c 100644
--- a/Demos/Gpu3dDemo/CMakeLists.txt
+++ b/Demos/Gpu3dDemo/CMakeLists.txt
@@ -16,7 +16,7 @@ ${BULLET_PHYSICS_SOURCE_DIR}/src ${BULLET_PHYSICS_SOURCE_DIR}/Demos/OpenGL }
 )
 
 LINK_LIBRARIES(
-OpenGLSupport  BulletDynamics  BulletCollision LinearMath    ${GLUT_glut_LIBRARY} ${OPENGL_gl_LIBRARY} ${OPENGL_glu_LIBRARY}
+OpenGLSupport BulletMultiThreaded  BulletDynamics  BulletCollision LinearMath    ${GLUT_glut_LIBRARY} ${OPENGL_gl_LIBRARY} ${OPENGL_glu_LIBRARY}
 )
 
 ADD_EXECUTABLE(AppGpu3dDemo
diff --git a/config.h.in b/config.h.in
index 6c30c429a..2f10be9b4 100644
--- a/config.h.in
+++ b/config.h.in
@@ -60,6 +60,10 @@
 /* Define to 1 if you have the <windows.h> header file. */
 #undef HAVE_WINDOWS_H
 
+/* Define to the sub-directory in which libtool stores uninstalled libraries.
+   */
+#undef LT_OBJDIR
+
 /* Name of package */
 #undef PACKAGE
 
diff --git a/src/BulletCollision/BroadphaseCollision/btDbvt.h b/src/BulletCollision/BroadphaseCollision/btDbvt.h
index 412241a46..d733fedf2 100644
--- a/src/BulletCollision/BroadphaseCollision/btDbvt.h
+++ b/src/BulletCollision/BroadphaseCollision/btDbvt.h
@@ -57,7 +57,7 @@ subject to the following restrictions:
 // Specific methods implementation
 
 //SSE gives errors on a MSVC 7.1
-#ifdef BT_USE_SSE
+#if defined (BT_USE_SSE) && defined (WIN32)
 #define DBVT_SELECT_IMPL		DBVT_IMPL_SSE
 #define DBVT_MERGE_IMPL			DBVT_IMPL_SSE
 #define DBVT_INT0_IMPL			DBVT_IMPL_SSE
diff --git a/src/BulletMultiThreaded/btGpu3DGridBroadphase.cpp b/src/BulletMultiThreaded/btGpu3DGridBroadphase.cpp
index 69bf94cc0..ea0f7e980 100644
--- a/src/BulletMultiThreaded/btGpu3DGridBroadphase.cpp
+++ b/src/BulletMultiThreaded/btGpu3DGridBroadphase.cpp
@@ -13,6 +13,11 @@ subject to the following restrictions:
 3. This notice may not be removed or altered from any source distribution.
 */
 
+///The 3 following lines include the CPU implementation of the kernels, keep them in this order.
+#include "BulletMultiThreaded/btGpuDefines.h"
+#include "BulletMultiThreaded/btGpuUtilsSharedDefs.h"
+#include "BulletMultiThreaded/btGpuUtilsSharedCode.h"
+
 
 
 #include "LinearMath/btAlignedAllocator.h"
diff --git a/src/BulletSoftBody/btSoftRigidDynamicsWorld.cpp b/src/BulletSoftBody/btSoftRigidDynamicsWorld.cpp
index 47f072e6d..8f3656829 100644
--- a/src/BulletSoftBody/btSoftRigidDynamicsWorld.cpp
+++ b/src/BulletSoftBody/btSoftRigidDynamicsWorld.cpp
@@ -102,13 +102,13 @@ void	btSoftRigidDynamicsWorld::solveSoftBodiesConstraints()
 	}	
 }
 
-void	btSoftRigidDynamicsWorld::addSoftBody(btSoftBody* body)
+void	btSoftRigidDynamicsWorld::addSoftBody(btSoftBody* body,short int collisionFilterGroup,short int collisionFilterMask)
 {
 	m_softBodies.push_back(body);
 
 	btCollisionWorld::addCollisionObject(body,
-		btBroadphaseProxy::DefaultFilter,
-		btBroadphaseProxy::AllFilter);
+		collisionFilterGroup,
+		collisionFilterMask);
 
 }
 
diff --git a/src/BulletSoftBody/btSoftRigidDynamicsWorld.h b/src/BulletSoftBody/btSoftRigidDynamicsWorld.h
index 2cb9d712e..6b7ed5dea 100644
--- a/src/BulletSoftBody/btSoftRigidDynamicsWorld.h
+++ b/src/BulletSoftBody/btSoftRigidDynamicsWorld.h
@@ -50,7 +50,7 @@ public:
 
 	virtual void	debugDrawWorld();
 
-	void	addSoftBody(btSoftBody* body);
+	void	addSoftBody(btSoftBody* body,short int collisionFilterGroup=btBroadphaseProxy::DefaultFilter,short int collisionFilterMask=btBroadphaseProxy::AllFilter);
 
 	void	removeSoftBody(btSoftBody* body);
 
diff --git a/src/LinearMath/btScalar.h b/src/LinearMath/btScalar.h
index a6d66d61e..36f54b369 100644
--- a/src/LinearMath/btScalar.h
+++ b/src/LinearMath/btScalar.h
@@ -128,6 +128,31 @@ inline int	btGetVersion()
 #else
 	//non-windows systems
 
+#if (defined (__APPLE__) && defined (__i386__))
+	#define BT_USE_SSE
+	#include <emmintrin.h>
+
+	#define SIMD_FORCE_INLINE inline
+///@todo: check out alignment methods for other platforms/compilers
+	#define ATTRIBUTE_ALIGNED16(a) a __attribute__ ((aligned (16)))
+	#define ATTRIBUTE_ALIGNED128(a) a __attribute__ ((aligned (128)))
+	#ifndef assert
+	#include <assert.h>
+	#endif
+
+	#if defined(DEBUG) || defined (_DEBUG)
+		#define btAssert assert
+	#else
+		#define btAssert(x)
+	#endif
+
+	//btFullAssert is optional, slows down a lot
+	#define btFullAssert(x)
+	#define btLikely(_c)  _c
+	#define btUnlikely(_c) _c
+
+#else
+
 		#define SIMD_FORCE_INLINE inline
 		///@todo: check out alignment methods for other platforms/compilers
 		///#define ATTRIBUTE_ALIGNED16(a) a __attribute__ ((aligned (16)))
@@ -148,7 +173,7 @@ inline int	btGetVersion()
 		#define btFullAssert(x)
 		#define btLikely(_c)  _c
 		#define btUnlikely(_c) _c
-
+#endif //__APPLE__ 
 
 #endif // LIBSPE2