Initial commit:

PhysX 3.4.0 Update @ 21294896
APEX 1.4.0 Update @ 21275617

[CL 21300167]

38 files changed, 7150 insertions, 0 deletions

diff --git a/PhysX_3.4/Source/Common/src/CmBitMap.h b/PhysX_3.4/Source/Common/src/CmBitMap.h
new file mode 100644
index 00000000..fbf6f781
--- /dev/null
+++ b/PhysX_3.4/Source/Common/src/CmBitMap.h
@@ -0,0 +1,504 @@
+// This code contains NVIDIA Confidential Information and is disclosed to you
+// under a form of NVIDIA software license agreement provided separately to you.
+//
+// Notice
+// NVIDIA Corporation and its licensors retain all intellectual property and
+// proprietary rights in and to this software and related documentation and
+// any modifications thereto. Any use, reproduction, disclosure, or
+// distribution of this software and related documentation without an express
+// license agreement from NVIDIA Corporation is strictly prohibited.
+//
+// ALL NVIDIA DESIGN SPECIFICATIONS, CODE ARE PROVIDED "AS IS.". NVIDIA MAKES
+// NO WARRANTIES, EXPRESSED, IMPLIED, STATUTORY, OR OTHERWISE WITH RESPECT TO
+// THE MATERIALS, AND EXPRESSLY DISCLAIMS ALL IMPLIED WARRANTIES OF NONINFRINGEMENT,
+// MERCHANTABILITY, AND FITNESS FOR A PARTICULAR PURPOSE.
+//
+// Information and code furnished is believed to be accurate and reliable.
+// However, NVIDIA Corporation assumes no responsibility for the consequences of use of such
+// information or for any infringement of patents or other rights of third parties that may
+// result from its use. No license is granted by implication or otherwise under any patent
+// or patent rights of NVIDIA Corporation. Details are subject to change without notice.
+// This code supersedes and replaces all information previously supplied.
+// NVIDIA Corporation products are not authorized for use as critical
+// components in life support devices or systems without express written approval of
+// NVIDIA Corporation.
+//
+// Copyright (c) 2008-2016 NVIDIA Corporation. All rights reserved.
+// Copyright (c) 2004-2008 AGEIA Technologies, Inc. All rights reserved.
+// Copyright (c) 2001-2004 NovodeX AG. All rights reserved.  
+
+
+#ifndef PX_PHYSICS_COMMON_BITMAP
+#define PX_PHYSICS_COMMON_BITMAP
+
+#include "foundation/PxAssert.h"
+#include "foundation/PxMath.h"
+#include "foundation/PxMemory.h"
+#include "PsAllocator.h"
+#include "PsUserAllocated.h"
+#include "PsIntrinsics.h"
+#include "PsMathUtils.h"
+#include "CmPhysXCommon.h"
+#include "PsBitUtils.h"
+// PX_SERIALIZATION
+#include "PxSerialFramework.h"
+//~PX_SERIALIZATION
+
+namespace physx
+{
+namespace Cm
+{
+
+	/*!
+	Hold a bitmap with operations to set,reset or test given bit.
+
+	We inhibit copy to prevent unintentional copies. If a copy is desired copy() should be used or
+	alternatively a copy constructor implemented.
+	*/
+	template<class Allocator>
+	class BitMapBase : public Ps::UserAllocated
+	{
+	//= ATTENTION! =====================================================================================
+	// Changing the data layout of this class breaks the binary serialization format.  See comments for 
+	// PX_BINARY_SERIAL_VERSION.  If a modification is required, please adjust the getBinaryMetaData 
+	// function.  If the modification is made on a custom branch, please change PX_BINARY_SERIAL_VERSION
+	// accordingly.
+	//==================================================================================================
+
+		PX_NOCOPY(BitMapBase)
+
+	public:
+
+		// PX_SERIALIZATION
+		/* todo: explicit */ BitMapBase(const PxEMPTY) 
+		{
+			if(mMap)
+				mWordCount |= PX_SIGN_BITMASK;
+		}
+
+		void	exportExtraData(PxSerializationContext& stream, void*)	
+		{
+			if(mMap && getWordCount())
+			{
+				stream.alignData(PX_SERIAL_ALIGN);
+				stream.writeData(mMap, getWordCount()*sizeof(PxU32));
+			}
+		}
+		void	importExtraData(PxDeserializationContext& context)
+		{
+			if(mMap && getWordCount())
+				mMap = context.readExtraData<PxU32, PX_SERIAL_ALIGN>(getWordCount());
+		}
+		//~PX_SERIALIZATION
+
+		//sschirm: function for placement new. Almost the same as importExtraData above, but lets you set word count and map after default construction
+		void importData(PxU32 worldCount, PxU32* words)
+		{
+			PX_ASSERT(mWordCount == 0 && !mMap);
+			mMap = words;
+			mWordCount = worldCount | PX_SIGN_BITMASK;
+		}
+		
+		PX_INLINE BitMapBase(Allocator& allocator) : mMap(0), mWordCount(0), mAllocator(allocator) {}
+
+		PX_INLINE BitMapBase() : mMap(0), mWordCount(0) {}
+
+		PX_INLINE ~BitMapBase()
+		{
+			if(mMap && !isInUserMemory())
+				mAllocator.deallocate(mMap);
+			mMap = NULL;
+		}
+
+		PX_INLINE Allocator&	getAllocator() { return mAllocator; }
+
+		PX_INLINE void growAndSet(PxU32 index)
+		{
+			extend(index+1);
+			mMap[index>>5] |= 1<<(index&31);
+		}
+
+		PX_INLINE void growAndReset(PxU32 index)
+		{
+			extend(index+1);
+			mMap[index>>5] &= ~(1<<(index&31));
+		}
+
+		PX_INLINE Ps::IntBool boundedTest(PxU32 index) const
+		{
+			return Ps::IntBool(index>>5 >= getWordCount() ? Ps::IntFalse : (mMap[index>>5]&(1<<(index&31))));
+		}
+
+		// Special optimized versions, when you _know_ your index is in range
+		PX_INLINE void set(PxU32 index)
+		{
+			PX_ASSERT(index<getWordCount()*32);
+			mMap[index>>5] |= 1<<(index&31);
+		}
+
+		PX_INLINE void reset(PxU32 index)
+		{
+			PX_ASSERT(index<getWordCount()*32);
+			mMap[index>>5] &= ~(1<<(index&31));
+		}
+
+		PX_INLINE Ps::IntBool test(PxU32 index) const
+		{
+			PX_ASSERT(index<getWordCount()*32);
+			return Ps::IntBool(mMap[index>>5]&(1<<(index&31)));
+		}
+
+		// nibble == 4 bits
+		PX_INLINE PxU32 getNibbleFast(PxU32 nibIndex) const
+		{
+			PxU32 bitIndex = nibIndex << 2;
+			PX_ASSERT(bitIndex < getWordCount()*32);
+			return (mMap[bitIndex >> 5] >> (bitIndex & 31)) & 0xf;
+		}
+
+		PX_INLINE void andNibbleFast(PxU32 nibIndex, PxU32 mask)
+		{
+			//TODO: there has to be a faster way...
+			PxU32 bitIndex = nibIndex << 2;
+			PxU32 shift = (bitIndex & 31);
+			PxU32 nibMask = 0xf << shift;
+
+			PX_ASSERT(bitIndex < getWordCount()*32);
+
+			mMap[bitIndex >> 5] &= ((mask << shift) | ~nibMask);
+		}
+
+		PX_INLINE void orNibbleFast(PxU32 nibIndex, PxU32 mask)
+		{
+			PX_ASSERT(!(mask & ~0xf)); //check extra bits are not set
+
+			PxU32 bitIndex = nibIndex << 2;
+			PxU32 shift = bitIndex & 31;
+
+			PX_ASSERT(bitIndex < getWordCount()*32);
+
+			mMap[bitIndex >> 5] |= (mask << shift);
+		}
+
+		void clear()
+		{
+			PxMemSet(mMap, 0, getWordCount()*sizeof(PxU32));
+		}
+
+		void resizeAndClear(PxU32 newBitCount)
+		{
+			extendUninitialized(newBitCount);
+			PxMemSet(mMap, 0, getWordCount()*sizeof(PxU32));
+		}
+
+		void setEmpty()
+		{
+			mMap=NULL;
+			mWordCount=0;
+		}
+
+		void setWords(PxU32* map, PxU32 wordCount)
+		{
+			mMap=map;
+			mWordCount=wordCount;
+			mWordCount |= PX_SIGN_BITMASK;
+		}
+	
+		// !!! only sets /last/ bit to value
+		void resize(PxU32 newBitCount, bool value = false)
+		{
+			PX_ASSERT(!value); // only new class supports this
+			PX_UNUSED(value);
+			extend(newBitCount);
+		}
+		PxU32 size() const { return getWordCount()*32; }
+
+		void copy(const BitMapBase& a)
+		{
+			extendUninitialized(a.getWordCount()<<5);
+			PxMemCopy(mMap, a.mMap, a.getWordCount() * sizeof(PxU32));
+			if(getWordCount() > a.getWordCount())
+				PxMemSet(mMap + a.getWordCount(), 0, (getWordCount() - a.getWordCount()) * sizeof(PxU32));
+		}
+
+		PX_INLINE PxU32 count()		const
+		{
+			// NOTE: we can probably do this faster, since the last steps in PxcBitCount32 can be defered to
+			// the end of the seq. + 64/128bits at a time + native bit counting instructions(360 is fast non micro code).
+			PxU32 count = 0;
+			PxU32 wordCount = getWordCount();
+			for(PxU32 i=0; i<wordCount; i++)
+				count += Ps::bitCount(mMap[i]);
+
+			return count;
+		}
+
+		PX_INLINE PxU32 count(PxU32 start, PxU32 length) const
+		{
+			PxU32 end = PxMin(getWordCount()<<5,start+length);
+			PxU32 count = 0;
+			for(PxU32 i=start; i<end; i++)
+				count+= (test(i)!=0);
+			return count;
+		}
+
+		//! returns 0 if no bits set (!!!)
+		PxU32 findLast() const
+		{
+			for(PxU32 i = getWordCount(); i-- > 0;)
+			{
+				if(mMap[i])
+					return (i<<5)+Ps::highestSetBit(mMap[i]);
+			}
+			return PxU32(0); 
+		}
+
+
+
+		// the obvious combiners and some used in the SDK
+
+		struct OR		{ PX_INLINE PxU32 operator()(PxU32 a, PxU32 b) {	return a|b;		}	};
+		struct AND		{ PX_INLINE PxU32 operator()(PxU32 a, PxU32 b) {	return a&b;		}	};
+		struct XOR		{ PX_INLINE PxU32 operator()(PxU32 a, PxU32 b) {	return a^b;		}	};
+
+		// we use auxiliary functions here so as not to generate combiners for every combination
+		// of allocators
+
+		template<class Combiner, class _> 
+		PX_INLINE void combineInPlace(const BitMapBase<_>& b)
+		{
+			combine1<Combiner>(b.mMap,b.getWordCount());
+		}
+
+		template<class Combiner, class _1, class _2>
+		PX_INLINE void combine(const BitMapBase<_1>& a, const BitMapBase<_2>& b)
+		{
+			combine2<Combiner>(a.mMap,a.getWordCount(),b.mMap,b.getWordCount());
+		}
+
+		PX_FORCE_INLINE const PxU32*	getWords()			const	{ return mMap;							}
+		PX_FORCE_INLINE PxU32*			getWords()					{ return mMap;							}
+		
+		// PX_SERIALIZATION
+		PX_FORCE_INLINE PxU32			getWordCount()		const	{ return mWordCount & ~PX_SIGN_BITMASK;	}
+		
+		// We need one bit to mark arrays that have been deserialized from a user-provided memory block.
+		PX_FORCE_INLINE	PxU32			isInUserMemory()	const	{ return mWordCount & PX_SIGN_BITMASK;	}
+		//~PX_SERIALIZATION
+
+		/*!
+		Iterate over indices in a bitmap
+
+		This iterator is good because it finds the set bit without looping over the cached bits upto 31 times.
+		However it does require a variable shift.
+		*/
+
+		class Iterator
+		{
+		public:
+			static const PxU32 DONE = 0xffffffff;
+
+			PX_INLINE Iterator(const BitMapBase &map) :	mBitMap(map)
+			{
+				reset();
+			}
+
+			PX_INLINE Iterator& operator=(const Iterator& other)
+			{
+				PX_ASSERT(&mBitMap == &other.mBitMap);
+				mBlock = other.mBlock;
+				mIndex = other.mIndex;
+				return *this;
+			}
+
+			PX_INLINE PxU32	getNext()
+			{
+				if(mBlock)
+				{
+					PxU32 bitIndex = mIndex<<5 | Ps::lowestSetBit(mBlock);
+					mBlock &= mBlock-1;
+					PxU32 wordCount = mBitMap.getWordCount();
+					while(!mBlock && ++mIndex < wordCount)
+						mBlock = mBitMap.mMap[mIndex];
+					return bitIndex;
+				}
+				return DONE;
+			}
+
+			PX_INLINE void reset()
+			{
+				mIndex = mBlock = 0;
+				PxU32 wordCount = mBitMap.getWordCount();
+				while(mIndex < wordCount && ((mBlock = mBitMap.mMap[mIndex]) == 0))
+					++mIndex;
+			}
+		private:
+			PxU32 mBlock, mIndex;
+			const BitMapBase& mBitMap;
+		};
+
+	// DS: faster but less general: hasBits() must be true or getNext() is illegal so it is the calling code's responsibility to ensure that getNext() is not called illegally.
+	class LoopIterator
+	{
+		PX_NOCOPY(LoopIterator)
+
+	public:
+		PX_FORCE_INLINE LoopIterator(const BitMapBase &map) : mMap(map.getWords()), mBlock(0), mIndex(-1), mWordCount(PxI32(map.getWordCount()))	{}
+
+		PX_FORCE_INLINE bool hasBits()
+		{
+			PX_ASSERT(mIndex<mWordCount);
+			while (mBlock == 0)
+			{
+				if (++mIndex == mWordCount)
+					return false;
+				mBlock = mMap[mIndex];
+			}
+			return true;
+		}
+
+		PX_FORCE_INLINE PxU32 getNext()
+		{
+			PX_ASSERT(mIndex<mWordCount && mBlock != 0);
+			PxU32 result = PxU32(mIndex) << 5 | Ps::lowestSetBit(mBlock);	// will assert if mask is zero
+			mBlock &= (mBlock - 1);
+			return result;
+		}
+
+	private:
+		const PxU32*const mMap;
+		PxU32 mBlock;		// the word we're currently scanning
+		PxI32 mIndex;		// the index of the word we're currently looking at
+		PxI32 mWordCount;
+	};
+
+	//Class to iterate over the bitmap from a particular start location rather than the beginning of the list
+	class CircularIterator
+	{
+	public:
+		static const PxU32 DONE = 0xffffffff;
+
+		PX_INLINE CircularIterator(const BitMapBase &map, PxU32 index) : mBitMap(map)
+		{
+			mIndex = mBlock = mStartIndex = 0;
+			const PxU32 wordCount = mBitMap.getWordCount();
+			if ((index << 5) < wordCount)
+			{
+				mIndex = index << 5;
+				mStartIndex = mIndex;
+			}
+
+			if (mIndex < wordCount)
+			{
+				mBlock = mBitMap.mMap[mIndex];
+				if (mBlock == 0)
+				{
+					mIndex = (mIndex + 1) % wordCount;
+					while (mIndex != mStartIndex && (mBlock = mBitMap.mMap[mIndex]) == 0)
+						mIndex = (mIndex + 1) % wordCount;
+				}
+			}
+		}
+
+		PX_INLINE PxU32	getNext()
+		{
+			if (mBlock)
+			{
+				PxU32 bitIndex = mIndex << 5 | Ps::lowestSetBit(mBlock);
+				mBlock &= mBlock - 1;
+				PxU32 wordCount = mBitMap.getWordCount();
+				while (!mBlock && (mIndex = ((mIndex+1)%wordCount)) != mStartIndex)
+					mBlock = mBitMap.mMap[mIndex];
+				return bitIndex;
+			}
+			return DONE;
+		}
+
+	private:
+		PxU32 mBlock, mIndex;
+		PxU32 mStartIndex;
+		const BitMapBase& mBitMap;
+
+		PX_NOCOPY(CircularIterator)
+	};
+
+
+
+
+	protected:
+		PxU32*			mMap;	//one bit per index
+		PxU32			mWordCount;
+		Allocator		mAllocator;
+		PxU8			mPadding[3];	// PT: "mAllocator" is empty but consumes 1 byte
+
+		void extend(PxU32 size)
+		{
+			PxU32 newWordCount = (size+31)>>5;
+			if(newWordCount > getWordCount())
+			{
+				PxU32* newMap = reinterpret_cast<PxU32*>(mAllocator.allocate(newWordCount*sizeof(PxU32), __FILE__, __LINE__));
+				if(mMap)
+				{
+					PxMemCopy(newMap, mMap, getWordCount()*sizeof(PxU32));
+					if (!isInUserMemory())
+						mAllocator.deallocate(mMap);
+				}
+				PxMemSet(newMap+getWordCount(), 0, (newWordCount-getWordCount())*sizeof(PxU32));
+				mMap = newMap;
+				// also resets the isInUserMemory bit
+				mWordCount = newWordCount;
+			}
+		}
+
+		void extendUninitialized(PxU32 size)
+		{
+			PxU32 newWordCount = (size+31)>>5;
+			if(newWordCount > getWordCount())
+			{
+				if(mMap && !isInUserMemory())
+					mAllocator.deallocate(mMap);
+				// also resets the isInUserMemory bit
+				mWordCount = newWordCount;
+				mMap = reinterpret_cast<PxU32*>(mAllocator.allocate(mWordCount*sizeof(PxU32), __FILE__, __LINE__));
+			}
+		}
+
+		template<class Combiner>
+		void combine1(const PxU32* words, PxU32 length)
+		{
+			extend(length<<5);
+			PxU32 combineLength = PxMin(getWordCount(), length);
+			for(PxU32 i=0;i<combineLength;i++)
+				mMap[i] = Combiner()(mMap[i], words[i]);
+		}
+
+		template<class Combiner>
+		void combine2(const PxU32* words1, PxU32 length1,
+			const PxU32* words2, PxU32 length2)
+		{
+			extendUninitialized(PxMax(length1,length2)<<5);
+
+			PxU32 commonSize = PxMin(length1,length2);
+
+			for(PxU32 i=0;i<commonSize;i++)
+				mMap[i] = Combiner()(words1[i],words2[i]);
+
+			for(PxU32 i=commonSize;i<length1;i++)
+				mMap[i] = Combiner()(words1[i],0);
+
+			for(PxU32 i=commonSize;i<length2;i++)
+				mMap[i] = Combiner()(0,words2[i]);
+		}
+
+		friend class Iterator;
+	};
+
+	typedef BitMapBase<Ps::NonTrackingAllocator> BitMap;
+	typedef BitMapBase<Ps::VirtualAllocator> BitMapPinned;
+
+
+} // namespace Cm
+
+}
+
+#endif
diff --git a/PhysX_3.4/Source/Common/src/CmBoxPruning.cpp b/PhysX_3.4/Source/Common/src/CmBoxPruning.cpp
new file mode 100644
index 00000000..a59ecd8d
--- /dev/null
+++ b/PhysX_3.4/Source/Common/src/CmBoxPruning.cpp
@@ -0,0 +1,197 @@
+// This code contains NVIDIA Confidential Information and is disclosed to you
+// under a form of NVIDIA software license agreement provided separately to you.
+//
+// Notice
+// NVIDIA Corporation and its licensors retain all intellectual property and
+// proprietary rights in and to this software and related documentation and
+// any modifications thereto. Any use, reproduction, disclosure, or
+// distribution of this software and related documentation without an express
+// license agreement from NVIDIA Corporation is strictly prohibited.
+//
+// ALL NVIDIA DESIGN SPECIFICATIONS, CODE ARE PROVIDED "AS IS.". NVIDIA MAKES
+// NO WARRANTIES, EXPRESSED, IMPLIED, STATUTORY, OR OTHERWISE WITH RESPECT TO
+// THE MATERIALS, AND EXPRESSLY DISCLAIMS ALL IMPLIED WARRANTIES OF NONINFRINGEMENT,
+// MERCHANTABILITY, AND FITNESS FOR A PARTICULAR PURPOSE.
+//
+// Information and code furnished is believed to be accurate and reliable.
+// However, NVIDIA Corporation assumes no responsibility for the consequences of use of such
+// information or for any infringement of patents or other rights of third parties that may
+// result from its use. No license is granted by implication or otherwise under any patent
+// or patent rights of NVIDIA Corporation. Details are subject to change without notice.
+// This code supersedes and replaces all information previously supplied.
+// NVIDIA Corporation products are not authorized for use as critical
+// components in life support devices or systems without express written approval of
+// NVIDIA Corporation.
+//
+// Copyright (c) 2008-2016 NVIDIA Corporation. All rights reserved.
+// Copyright (c) 2004-2008 AGEIA Technologies, Inc. All rights reserved.
+// Copyright (c) 2001-2004 NovodeX AG. All rights reserved.  
+
+#include "CmBoxPruning.h"
+#include "CmRadixSortBuffered.h"
+#include "PsAllocator.h"
+
+using namespace physx;
+using namespace Gu;
+using namespace Cm;
+
+///////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+/**
+ *	Bipartite box pruning. Returns a list of overlapping pairs of boxes, each box of the pair belongs to a different set.
+ *	\param		nb0		[in] number of boxes in the first set
+ *	\param		bounds0	[in] list of boxes for the first set
+ *	\param		nb1		[in] number of boxes in the second set
+ *	\param		bounds1	[in] list of boxes for the second set
+ *	\param		pairs	[out] list of overlapping pairs
+ *	\param		axes	[in] projection order (0,2,1 is often best)
+ *	\return		true if success.
+ */
+///////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+bool Cm::BipartiteBoxPruning(const PxBounds3* bounds0, PxU32 nb0, const PxBounds3* bounds1, PxU32 nb1, Ps::Array<PxU32>& pairs, const Axes& axes)
+{
+	pairs.clear();
+	// Checkings
+	if(nb0 == 0 || nb1 == 0)
+		return false;
+
+	// Catch axes
+	PxU32 Axis0 = axes.mAxis0;
+	PxU32 Axis1 = axes.mAxis1;
+	PxU32 Axis2 = axes.mAxis2;
+
+	PX_UNUSED(Axis1);
+	PX_UNUSED(Axis2);
+
+	// Allocate some temporary data
+	float* MinPosBounds0 = reinterpret_cast<float*>(PX_ALLOC_TEMP(sizeof(float)*nb0, "Gu::BipartiteBoxPruning"));
+	float* MinPosBounds1 = reinterpret_cast<float*>(PX_ALLOC_TEMP(sizeof(float)*nb1, "Gu::BipartiteBoxPruning"));
+
+	// 1) Build main lists using the primary axis
+	for(PxU32 i=0;i<nb0;i++)	MinPosBounds0[i] = bounds0[i].minimum[Axis0];
+	for(PxU32 i=0;i<nb1;i++)	MinPosBounds1[i] = bounds1[i].minimum[Axis0];
+
+	// 2) Sort the lists
+	//static RadixSort RS0, RS1;	// Static for coherence. Crashes on exit
+	RadixSortBuffered RS0, RS1;	// Static for coherence.
+
+	const PxU32* Sorted0 = RS0.Sort(MinPosBounds0, nb0).GetRanks();
+	const PxU32* Sorted1 = RS1.Sort(MinPosBounds1, nb1).GetRanks();
+
+	// 3) Prune the lists
+	PxU32 Index0, Index1;
+
+	const PxU32* const LastSorted0 = &Sorted0[nb0];
+	const PxU32* const LastSorted1 = &Sorted1[nb1];
+	const PxU32* RunningAddress0 = Sorted0;
+	const PxU32* RunningAddress1 = Sorted1;
+
+	while(RunningAddress1<LastSorted1 && Sorted0<LastSorted0)
+	{
+		Index0 = *Sorted0++;
+
+		while(RunningAddress1<LastSorted1 && MinPosBounds1[*RunningAddress1]<MinPosBounds0[Index0])	RunningAddress1++;
+
+		const PxU32* RunningAddress2_1 = RunningAddress1;
+
+		while(RunningAddress2_1<LastSorted1 && MinPosBounds1[Index1 = *RunningAddress2_1++]<=bounds0[Index0].maximum[Axis0])
+		{
+			if(bounds0[Index0].intersects(bounds1[Index1]))
+			{
+				pairs.pushBack(Index0);
+				pairs.pushBack(Index1);
+			}
+		}
+	}
+
+	////
+
+	while(RunningAddress0<LastSorted0 && Sorted1<LastSorted1)
+	{
+		Index0 = *Sorted1++;
+
+		while(RunningAddress0<LastSorted0 && MinPosBounds0[*RunningAddress0]<=MinPosBounds1[Index0])	RunningAddress0++;
+
+		const PxU32* RunningAddress2_0 = RunningAddress0;
+
+		while(RunningAddress2_0<LastSorted0 && MinPosBounds0[Index1 = *RunningAddress2_0++]<=bounds1[Index0].maximum[Axis0])
+		{
+			if(bounds0[Index1].intersects(bounds1[Index0]))
+			{
+				pairs.pushBack(Index1);
+				pairs.pushBack(Index0);
+			}
+		}
+	}
+
+	PX_FREE(MinPosBounds1);
+	PX_FREE(MinPosBounds0);
+
+	return true;
+}
+
+///////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+/**
+ *	Complete box pruning. Returns a list of overlapping pairs of boxes, each box of the pair belongs to the same set.
+ *	\param		bounds	[in] list of boxes
+ *	\param		nb		[in] number of boxes
+ *	\param		pairs	[out] list of overlapping pairs
+ *	\param		axes	[in] projection order (0,2,1 is often best)
+ *	\return		true if success.
+ */
+///////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+bool Cm::CompleteBoxPruning(const PxBounds3* bounds, PxU32 nb, Ps::Array<PxU32>& pairs, const Axes& axes)
+{
+	pairs.clear();
+
+	// Checkings
+	if(!nb)
+		return false;
+
+	// Catch axes
+	const PxU32 Axis0 = axes.mAxis0;
+	const PxU32 Axis1 = axes.mAxis1;
+	const PxU32 Axis2 = axes.mAxis2;
+
+	PX_UNUSED(Axis1);
+	PX_UNUSED(Axis2);
+
+	// Allocate some temporary data
+	float* PosList = reinterpret_cast<float*>(PX_ALLOC_TEMP(sizeof(float)*nb, "Cm::CompleteBoxPruning"));
+
+	// 1) Build main list using the primary axis
+	for(PxU32 i=0;i<nb;i++)	PosList[i] = bounds[i].minimum[Axis0];
+
+	// 2) Sort the list
+	/*static*/ RadixSortBuffered RS;	// Static for coherence
+	const PxU32* Sorted = RS.Sort(PosList, nb).GetRanks();
+
+	// 3) Prune the list
+	const PxU32* const LastSorted = &Sorted[nb];
+	const PxU32* RunningAddress = Sorted;
+	PxU32 Index0, Index1;
+	while(RunningAddress<LastSorted && Sorted<LastSorted)
+	{
+		Index0 = *Sorted++;
+
+		while(RunningAddress<LastSorted && PosList[*RunningAddress++]<PosList[Index0]);
+
+		const PxU32* RunningAddress2 = RunningAddress;
+
+		while(RunningAddress2<LastSorted && PosList[Index1 = *RunningAddress2++]<=bounds[Index0].maximum[Axis0])
+		{
+			if(Index0!=Index1)
+			{
+				if(bounds[Index0].intersects(bounds[Index1]))
+				{
+					pairs.pushBack(Index0);
+					pairs.pushBack(Index1);
+				}
+			}
+		}
+	}
+
+	PX_FREE(PosList);
+
+	return true;
+}
+
diff --git a/PhysX_3.4/Source/Common/src/CmBoxPruning.h b/PhysX_3.4/Source/Common/src/CmBoxPruning.h
new file mode 100644
index 00000000..993b92a5
--- /dev/null
+++ b/PhysX_3.4/Source/Common/src/CmBoxPruning.h
@@ -0,0 +1,49 @@
+// This code contains NVIDIA Confidential Information and is disclosed to you
+// under a form of NVIDIA software license agreement provided separately to you.
+//
+// Notice
+// NVIDIA Corporation and its licensors retain all intellectual property and
+// proprietary rights in and to this software and related documentation and
+// any modifications thereto. Any use, reproduction, disclosure, or
+// distribution of this software and related documentation without an express
+// license agreement from NVIDIA Corporation is strictly prohibited.
+//
+// ALL NVIDIA DESIGN SPECIFICATIONS, CODE ARE PROVIDED "AS IS.". NVIDIA MAKES
+// NO WARRANTIES, EXPRESSED, IMPLIED, STATUTORY, OR OTHERWISE WITH RESPECT TO
+// THE MATERIALS, AND EXPRESSLY DISCLAIMS ALL IMPLIED WARRANTIES OF NONINFRINGEMENT,
+// MERCHANTABILITY, AND FITNESS FOR A PARTICULAR PURPOSE.
+//
+// Information and code furnished is believed to be accurate and reliable.
+// However, NVIDIA Corporation assumes no responsibility for the consequences of use of such
+// information or for any infringement of patents or other rights of third parties that may
+// result from its use. No license is granted by implication or otherwise under any patent
+// or patent rights of NVIDIA Corporation. Details are subject to change without notice.
+// This code supersedes and replaces all information previously supplied.
+// NVIDIA Corporation products are not authorized for use as critical
+// components in life support devices or systems without express written approval of
+// NVIDIA Corporation.
+//
+// Copyright (c) 2008-2016 NVIDIA Corporation. All rights reserved.
+// Copyright (c) 2004-2008 AGEIA Technologies, Inc. All rights reserved.
+// Copyright (c) 2001-2004 NovodeX AG. All rights reserved.  
+
+#ifndef CM_BOXPRUNING_H
+#define CM_BOXPRUNING_H
+
+#include "foundation/PxBounds3.h"
+#include "PsArray.h"
+#include "CmPhysXCommon.h"
+#include "PxPhysXCommonConfig.h"
+#include "GuAxes.h"
+
+namespace physx
+{
+
+namespace Cm
+{
+	PX_PHYSX_COMMON_API bool CompleteBoxPruning(const PxBounds3* bounds, PxU32 nb, Ps::Array<PxU32>& pairs, const Gu::Axes& axes);
+	PX_PHYSX_COMMON_API bool BipartiteBoxPruning(const PxBounds3* bounds0, PxU32 nb0, const PxBounds3* bounds1, PxU32 nb1, Ps::Array<PxU32>& pairs, const Gu::Axes& axes);
+}
+}
+	
+#endif // CM_BOXPRUNING_H
diff --git a/PhysX_3.4/Source/Common/src/CmCollection.cpp b/PhysX_3.4/Source/Common/src/CmCollection.cpp
new file mode 100644
index 00000000..dc3634d7
--- /dev/null
+++ b/PhysX_3.4/Source/Common/src/CmCollection.cpp
@@ -0,0 +1,217 @@
+// This code contains NVIDIA Confidential Information and is disclosed to you
+// under a form of NVIDIA software license agreement provided separately to you.
+//
+// Notice
+// NVIDIA Corporation and its licensors retain all intellectual property and
+// proprietary rights in and to this software and related documentation and
+// any modifications thereto. Any use, reproduction, disclosure, or
+// distribution of this software and related documentation without an express
+// license agreement from NVIDIA Corporation is strictly prohibited.
+//
+// ALL NVIDIA DESIGN SPECIFICATIONS, CODE ARE PROVIDED "AS IS.". NVIDIA MAKES
+// NO WARRANTIES, EXPRESSED, IMPLIED, STATUTORY, OR OTHERWISE WITH RESPECT TO
+// THE MATERIALS, AND EXPRESSLY DISCLAIMS ALL IMPLIED WARRANTIES OF NONINFRINGEMENT,
+// MERCHANTABILITY, AND FITNESS FOR A PARTICULAR PURPOSE.
+//
+// Information and code furnished is believed to be accurate and reliable.
+// However, NVIDIA Corporation assumes no responsibility for the consequences of use of such
+// information or for any infringement of patents or other rights of third parties that may
+// result from its use. No license is granted by implication or otherwise under any patent
+// or patent rights of NVIDIA Corporation. Details are subject to change without notice.
+// This code supersedes and replaces all information previously supplied.
+// NVIDIA Corporation products are not authorized for use as critical
+// components in life support devices or systems without express written approval of
+// NVIDIA Corporation.
+//
+// Copyright (c) 2008-2016 NVIDIA Corporation. All rights reserved.
+// Copyright (c) 2004-2008 AGEIA Technologies, Inc. All rights reserved.
+// Copyright (c) 2001-2004 NovodeX AG. All rights reserved.  
+
+#include "CmCollection.h"
+#include "PsFoundation.h"
+
+using namespace physx;
+using namespace Cm;
+
+void Collection::add(PxBase& object, PxSerialObjectId id)
+{
+	PxSerialObjectId originId = getId(object);
+	if( originId != PX_SERIAL_OBJECT_ID_INVALID)
+	{
+		if( originId != id)
+		{
+			 physx::shdfnd::getFoundation().error(physx::PxErrorCode::eINVALID_PARAMETER, __FILE__, __LINE__,
+		        "PxCollection::add called for an object that has an associated id already present in the collection!");
+		}
+		return;		   
+	}
+	
+	if(id != PX_SERIAL_OBJECT_ID_INVALID)
+	{		
+		if(!mIds.insert(id, &object))
+		{
+		   physx::shdfnd::getFoundation().error(physx::PxErrorCode::eINVALID_PARAMETER, __FILE__, __LINE__,
+		        "PxCollection::add called with an id which is already used in the collection");
+		   return;	
+		}
+	}
+    
+	mObjects[&object] = id;	
+}
+
+void Collection::remove(PxBase& object)
+{
+	PX_CHECK_AND_RETURN(contains(object), "PxCollection::remove called for an object not contained in the collection!");
+	
+	const ObjectToIdMap::Entry* e = mObjects.find(&object);
+	if(e)
+	{
+		mIds.erase(e->second);
+	    mObjects.erase(&object);
+	}
+}
+
+bool Collection::contains(PxBase& object) const
+{	
+	return  mObjects.find(&object) != NULL;
+}
+
+void Collection::addId(PxBase& object, PxSerialObjectId id)
+{
+	PX_CHECK_AND_RETURN(contains(object), "PxCollection::addId called for object that is not contained in the collection!");
+	PX_CHECK_AND_RETURN(id != PX_SERIAL_OBJECT_ID_INVALID, "PxCollection::addId called with PxSerialObjectId being set to PX_SERIAL_OBJECT_ID_INVALID!");
+	PX_CHECK_AND_RETURN(mIds.find(id) == NULL, "PxCollection::addId called with an id which is already used in the collection!");
+	
+	const ObjectToIdMap::Entry* e = mObjects.find(&object);
+	if(e  && e->second != PX_SERIAL_OBJECT_ID_INVALID)
+		mIds.erase(e->second);
+
+	mIds.insert(id, &object);	
+	mObjects[&object] = id;
+}
+
+void Collection::removeId(PxSerialObjectId id)
+{
+	PX_CHECK_AND_RETURN(id != PX_SERIAL_OBJECT_ID_INVALID, "PxCollection::removeId called with PxSerialObjectId being set to PX_SERIAL_OBJECT_ID_INVALID!");
+	PX_CHECK_AND_RETURN(mIds.find(id), "PxCollection::removeId called with PxSerialObjectId not contained in the collection!");
+	const IdToObjectMap::Entry* e = mIds.find(id);
+	if(e)
+	{	
+		mObjects[e->second] = PX_SERIAL_OBJECT_ID_INVALID;
+	    mIds.erase(id);	   
+	}
+}
+
+PxBase* Collection::find(PxSerialObjectId id) const
+{
+	PX_CHECK_AND_RETURN_NULL(id != PX_SERIAL_OBJECT_ID_INVALID, "PxCollection::find called with PxSerialObjectId being set to PX_SERIAL_OBJECT_ID_INVALID!");
+	const IdToObjectMap::Entry* e = mIds.find(id);
+	return e ? static_cast<PxBase*>(e->second) : NULL;
+}
+
+void Collection::add(PxCollection& _collection)
+{
+	Collection& collection = static_cast<Collection&>(_collection);
+	PX_CHECK_AND_RETURN(this != &collection, "PxCollection::add(PxCollection&) called with itself!");
+
+	mObjects.reserve(mObjects.capacity() + collection.mObjects.size());
+	const ObjectToIdMap::Entry* e = collection.mObjects.getEntries();
+	for (PxU32 i = 0; i < collection.mObjects.size(); ++i)
+	{
+        PxSerialObjectId id = e[i].second;
+		if( id != PX_SERIAL_OBJECT_ID_INVALID)
+		{
+			if(!mIds.insert(id, e[i].first))
+		    {
+			    if(mIds[id] != e[i].first)
+				{
+				    PX_CHECK_MSG( false, "PxCollection::add(PxCollection&) called with conflicting id!");
+				    mObjects.insert(e[i].first, PX_SERIAL_OBJECT_ID_INVALID);
+				}
+		    }
+			else
+			   mObjects[ e[i].first ] = id;
+		}
+		else
+		    mObjects.insert(e[i].first, PX_SERIAL_OBJECT_ID_INVALID);				
+	}
+}
+
+void Collection::remove(PxCollection& _collection)
+{
+	Collection& collection = static_cast<Collection&>(_collection);
+	PX_CHECK_AND_RETURN(this != &collection, "PxCollection::remove(PxCollection&) called with itself!");
+
+	const ObjectToIdMap::Entry* e = collection.mObjects.getEntries();
+	for (PxU32 i = 0; i < collection.mObjects.size(); ++i)
+	{
+		const ObjectToIdMap::Entry* e1 = mObjects.find(e[i].first);
+		if(e1)
+		{
+		    mIds.erase(e1->second);
+	        mObjects.erase(e1->first);
+		}
+	}
+}
+
+PxU32 Collection::getNbObjects() const
+{
+	return mObjects.size();
+}
+
+PxBase& Collection::getObject(PxU32 index) const
+{
+	PX_ASSERT(index < mObjects.size());
+	return *mObjects.getEntries()[index].first;
+}
+
+PxU32 Collection::getObjects(PxBase** userBuffer, PxU32 bufferSize, PxU32 startIndex) const
+{
+	PX_CHECK_AND_RETURN_NULL(userBuffer != NULL, "PxCollection::getObjects called with userBuffer NULL!");
+	PX_CHECK_AND_RETURN_NULL(bufferSize != 0, "PxCollection::getObjects called with bufferSize 0!");
+	PxU32 dstIndex = 0;
+	const ObjectToIdMap::Entry* e  = mObjects.getEntries();
+	for (PxU32 srcIndex = startIndex; srcIndex < mObjects.size() && dstIndex < bufferSize; ++srcIndex)
+		userBuffer[dstIndex++] = e[srcIndex].first;
+
+	return dstIndex;
+}
+
+PxU32 Collection::getNbIds() const
+{
+	return mIds.size();
+}
+
+PxSerialObjectId Collection::getId(const PxBase& object) const
+{	
+	const ObjectToIdMap::Entry* e =  mObjects.find(const_cast<PxBase*>(&object));
+	return e ? e->second : PX_SERIAL_OBJECT_ID_INVALID;
+}
+
+PxU32 Collection::getIds(PxSerialObjectId* userBuffer, PxU32 bufferSize, PxU32 startIndex) const
+{
+	PX_CHECK_AND_RETURN_NULL(userBuffer != NULL, "PxCollection::getIds called with userBuffer NULL!");
+	PX_CHECK_AND_RETURN_NULL(bufferSize != 0, "PxCollection::getIds called with bufferSize 0!");
+	PxU32 dstIndex = 0;
+
+	IdToObjectMap::Iterator srcIt = (const_cast<IdToObjectMap&>(mIds)).getIterator();
+
+	while (!srcIt.done() &&  dstIndex < bufferSize)
+	{
+		if(srcIt->first != PX_SERIAL_OBJECT_ID_INVALID)
+		{
+			if(startIndex > 0)
+				startIndex--;
+			else
+				userBuffer[dstIndex++] = srcIt->first;
+		}
+		srcIt++;
+	}
+
+	return dstIndex;
+}
+
+PxCollection*	PxCreateCollection()
+{
+	return PX_NEW(Collection);
+}
diff --git a/PhysX_3.4/Source/Common/src/CmCollection.h b/PhysX_3.4/Source/Common/src/CmCollection.h
new file mode 100644
index 00000000..76f9738e
--- /dev/null
+++ b/PhysX_3.4/Source/Common/src/CmCollection.h
@@ -0,0 +1,103 @@
+// This code contains NVIDIA Confidential Information and is disclosed to you
+// under a form of NVIDIA software license agreement provided separately to you.
+//
+// Notice
+// NVIDIA Corporation and its licensors retain all intellectual property and
+// proprietary rights in and to this software and related documentation and
+// any modifications thereto. Any use, reproduction, disclosure, or
+// distribution of this software and related documentation without an express
+// license agreement from NVIDIA Corporation is strictly prohibited.
+//
+// ALL NVIDIA DESIGN SPECIFICATIONS, CODE ARE PROVIDED "AS IS.". NVIDIA MAKES
+// NO WARRANTIES, EXPRESSED, IMPLIED, STATUTORY, OR OTHERWISE WITH RESPECT TO
+// THE MATERIALS, AND EXPRESSLY DISCLAIMS ALL IMPLIED WARRANTIES OF NONINFRINGEMENT,
+// MERCHANTABILITY, AND FITNESS FOR A PARTICULAR PURPOSE.
+//
+// Information and code furnished is believed to be accurate and reliable.
+// However, NVIDIA Corporation assumes no responsibility for the consequences of use of such
+// information or for any infringement of patents or other rights of third parties that may
+// result from its use. No license is granted by implication or otherwise under any patent
+// or patent rights of NVIDIA Corporation. Details are subject to change without notice.
+// This code supersedes and replaces all information previously supplied.
+// NVIDIA Corporation products are not authorized for use as critical
+// components in life support devices or systems without express written approval of
+// NVIDIA Corporation.
+//
+// Copyright (c) 2008-2016 NVIDIA Corporation. All rights reserved.
+// Copyright (c) 2004-2008 AGEIA Technologies, Inc. All rights reserved.
+// Copyright (c) 2001-2004 NovodeX AG. All rights reserved.  
+
+
+#ifndef PX_PHYSICS_CM_COLLECTION
+#define PX_PHYSICS_CM_COLLECTION
+
+#include "CmPhysXCommon.h"
+#include "PxCollection.h"
+#include "PsHashMap.h"
+#include "PsUserAllocated.h"
+#include "PsAllocator.h"
+
+namespace physx
+{
+namespace Cm
+{	
+	template <class Key, 
+			  class Value,
+			  class HashFn = Ps::Hash<Key>, 
+			  class Allocator = Ps::NonTrackingAllocator >
+	class CollectionHashMap : public Ps::CoalescedHashMap< Key, Value, HashFn, Allocator>
+	{
+		typedef physx::shdfnd::internal::HashMapBase< Key, Value, HashFn, Allocator> MapBase;	
+		typedef Ps::Pair<const Key,Value> EntryData;
+
+		public:
+			CollectionHashMap(PxU32 initialTableSize = 64, float loadFactor = 0.75f):
+			    Ps::CoalescedHashMap< Key, Value, HashFn, Allocator>(initialTableSize,loadFactor) {}
+
+			void insertUnique(const Key& k, const Value& v)
+			{
+				PX_PLACEMENT_NEW(MapBase::mBase.insertUnique(k), EntryData)(k,v);
+			}
+	};
+
+	
+	
+	class Collection : public PxCollection, public Ps::UserAllocated
+	{
+	public:
+		typedef CollectionHashMap<PxBase*, PxSerialObjectId> ObjectToIdMap;
+		typedef CollectionHashMap<PxSerialObjectId, PxBase*> IdToObjectMap;
+					
+		virtual void						add(PxBase& object, PxSerialObjectId ref);
+		virtual	void						remove(PxBase& object);	
+		virtual bool						contains(PxBase& object) const;
+		virtual void						addId(PxBase& object, PxSerialObjectId id);
+		virtual void						removeId(PxSerialObjectId id);
+		virtual PxBase*						find(PxSerialObjectId ref) const;
+		virtual void						add(PxCollection& collection);
+		virtual void						remove(PxCollection& collection);		
+		virtual	PxU32						getNbObjects() const;
+		virtual PxBase&						getObject(PxU32 index) const;
+		virtual	PxU32						getObjects(PxBase** userBuffer, PxU32 bufferSize, PxU32 startIndex=0) const;
+
+		virtual PxU32						getNbIds() const;		
+		virtual PxSerialObjectId			getId(const PxBase& object) const;
+		virtual	PxU32						getIds(PxSerialObjectId* userBuffer, PxU32 bufferSize, PxU32 startIndex=0) const;
+
+		void								release() { PX_DELETE(this); }
+
+
+		// Only for internal use. Bypasses virtual calls, specialized behaviour.
+		PX_INLINE	void		            internalAdd(PxBase* s, PxSerialObjectId id = PX_SERIAL_OBJECT_ID_INVALID)				{ mObjects.insertUnique(s, id);	                   }
+		PX_INLINE	PxU32		            internalGetNbObjects()		 const	{ return mObjects.size();								               }
+		PX_INLINE	PxBase*		            internalGetObject(PxU32 i)	 const	{ PX_ASSERT(i<mObjects.size());	return mObjects.getEntries()[i].first; }
+		PX_INLINE	const ObjectToIdMap::Entry*	internalGetObjects() const  { return mObjects.getEntries(); 			                           }
+			
+		IdToObjectMap					    mIds;
+		ObjectToIdMap                       mObjects;
+		
+	};
+}
+}
+
+#endif
diff --git a/PhysX_3.4/Source/Common/src/CmConeLimitHelper.h b/PhysX_3.4/Source/Common/src/CmConeLimitHelper.h
new file mode 100644
index 00000000..fa57574a
--- /dev/null
+++ b/PhysX_3.4/Source/Common/src/CmConeLimitHelper.h
@@ -0,0 +1,144 @@
+// This code contains NVIDIA Confidential Information and is disclosed to you
+// under a form of NVIDIA software license agreement provided separately to you.
+//
+// Notice
+// NVIDIA Corporation and its licensors retain all intellectual property and
+// proprietary rights in and to this software and related documentation and
+// any modifications thereto. Any use, reproduction, disclosure, or
+// distribution of this software and related documentation without an express
+// license agreement from NVIDIA Corporation is strictly prohibited.
+//
+// ALL NVIDIA DESIGN SPECIFICATIONS, CODE ARE PROVIDED "AS IS.". NVIDIA MAKES
+// NO WARRANTIES, EXPRESSED, IMPLIED, STATUTORY, OR OTHERWISE WITH RESPECT TO
+// THE MATERIALS, AND EXPRESSLY DISCLAIMS ALL IMPLIED WARRANTIES OF NONINFRINGEMENT,
+// MERCHANTABILITY, AND FITNESS FOR A PARTICULAR PURPOSE.
+//
+// Information and code furnished is believed to be accurate and reliable.
+// However, NVIDIA Corporation assumes no responsibility for the consequences of use of such
+// information or for any infringement of patents or other rights of third parties that may
+// result from its use. No license is granted by implication or otherwise under any patent
+// or patent rights of NVIDIA Corporation. Details are subject to change without notice.
+// This code supersedes and replaces all information previously supplied.
+// NVIDIA Corporation products are not authorized for use as critical
+// components in life support devices or systems without express written approval of
+// NVIDIA Corporation.
+//
+// Copyright (c) 2008-2016 NVIDIA Corporation. All rights reserved.
+// Copyright (c) 2004-2008 AGEIA Technologies, Inc. All rights reserved.
+// Copyright (c) 2001-2004 NovodeX AG. All rights reserved.  
+
+
+#ifndef PX_PHYSICS_COMMON_CONELIMITHELPER
+#define PX_PHYSICS_COMMON_CONELIMITHELPER
+
+// This class contains methods for supporting the tan-quarter swing limit - that
+// is the, ellipse defined by tanQ(theta)^2/tanQ(thetaMax)^2 + tanQ(phi)^2/tanQ(phiMax)^2 = 1
+// 
+// Angles are passed as an PxVec3 swing vector with x = 0 and y and z the swing angles
+// around the y and z axes
+
+#include "CmPhysXCommon.h"
+#include "PsMathUtils.h"
+
+namespace physx
+{
+namespace Cm
+{
+	PX_FORCE_INLINE PxReal tanAdd(PxReal tan1, PxReal tan2)
+	{
+		PX_ASSERT(PxAbs(1-tan1*tan2)>1e-6f);
+		return (tan1+tan2)/(1-tan1*tan2);
+	}
+
+	// this is here because it's used in both LL and Extensions. However, it
+	// should STAY IN THE SDK CODE BASE because it's SDK-specific
+
+	class ConeLimitHelper
+	{
+	public:
+		ConeLimitHelper(PxReal tanQSwingY, PxReal tanQSwingZ, PxReal tanQPadding)
+			: mTanQYMax(tanQSwingY), mTanQZMax(tanQSwingZ), mTanQPadding(tanQPadding) {}
+
+		// whether the point is inside the (inwardly) padded cone - if it is, there's no limit
+		// constraint
+
+		PX_FORCE_INLINE bool contains(const PxVec3& tanQSwing)
+		{
+			PxReal tanQSwingYPadded = tanAdd(PxAbs(tanQSwing.y),mTanQPadding);
+			PxReal tanQSwingZPadded = tanAdd(PxAbs(tanQSwing.z),mTanQPadding);
+			return Ps::sqr(tanQSwingYPadded/mTanQYMax)+Ps::sqr(tanQSwingZPadded/mTanQZMax) <= 1;
+		}
+
+		PX_FORCE_INLINE PxVec3 clamp(const PxVec3& tanQSwing, 
+									 PxVec3& normal)
+		{
+			PxVec3 p = Ps::ellipseClamp(tanQSwing, PxVec3(0,mTanQYMax,mTanQZMax));
+			normal = PxVec3(0, p.y/Ps::sqr(mTanQYMax), p.z/Ps::sqr(mTanQZMax));
+#ifdef PX_PARANOIA_ELLIPSE_CHECK
+			PxReal err = PxAbs(Ps::sqr(p.y/mTanQYMax) + Ps::sqr(p.z/mTanQZMax) - 1);
+			PX_ASSERT(err<1e-3);
+#endif
+
+			return p;
+		}
+
+
+		// input is a swing quat, such that swing.x = twist.y = twist.z = 0, q = swing * twist
+		// The routine is agnostic to the sign of q.w (i.e. we don't need the minimal-rotation swing)
+
+		// output is an axis such that positive rotation increases the angle outward from the
+		// limit (i.e. the image of the x axis), the error is the sine of the angular difference,
+		// positive if the twist axis is inside the cone 
+
+		bool getLimit(const PxQuat& swing, PxVec3& axis, PxReal& error)
+		{
+			PX_ASSERT(swing.w>0);
+			PxVec3 twistAxis = swing.getBasisVector0();
+			PxVec3 tanQSwing = PxVec3(0, Ps::tanHalf(swing.z,swing.w), -Ps::tanHalf(swing.y,swing.w));
+			if(contains(tanQSwing))
+				return false;
+
+			PxVec3 normal, clamped = clamp(tanQSwing, normal);
+
+			// rotation vector and ellipse normal
+			PxVec3 r(0,-clamped.z,clamped.y), d(0, -normal.z, normal.y);
+
+			// the point on the cone defined by the tanQ swing vector r
+			PxVec3 p(1.f,0,0);
+			PxReal r2 = r.dot(r), a = 1-r2, b = 1/(1+r2), b2 = b*b;
+			PxReal v1 = 2*a*b2;
+			PxVec3 v2(a, 2*r.z, -2*r.y);		// a*p + 2*r.cross(p);
+			PxVec3 coneLine = v1 * v2 - p;		// already normalized
+
+			// the derivative of coneLine in the direction d	
+			PxReal rd = r.dot(d);
+			PxReal dv1 = -4*rd*(3-r2)*b2*b;
+			PxVec3 dv2(-2*rd, 2*d.z, -2*d.y);
+	
+			PxVec3 coneNormal = v1 * dv2 + dv1 * v2;
+
+			axis = coneLine.cross(coneNormal)/coneNormal.magnitude();
+			error = coneLine.cross(axis).dot(twistAxis);
+
+			PX_ASSERT(PxAbs(axis.magnitude()-1)<1e-5f);
+
+#ifdef PX_PARANOIA_ELLIPSE_CHECK
+			bool inside = Ps::sqr(tanQSwing.y/mTanQYMax) + Ps::sqr(tanQSwing.z/mTanQZMax) <= 1;
+			PX_ASSERT(inside && error>-1e-4f || !inside && error<1e-4f);
+#endif
+
+			return true;
+		}
+
+	private:
+
+
+		PxReal mTanQYMax, mTanQZMax, mTanQPadding;
+	};
+
+
+} // namespace Cm
+
+}
+
+#endif
diff --git a/PhysX_3.4/Source/Common/src/CmFlushPool.h b/PhysX_3.4/Source/Common/src/CmFlushPool.h
new file mode 100644
index 00000000..875d09f4
--- /dev/null
+++ b/PhysX_3.4/Source/Common/src/CmFlushPool.h
@@ -0,0 +1,157 @@
+// This code contains NVIDIA Confidential Information and is disclosed to you
+// under a form of NVIDIA software license agreement provided separately to you.
+//
+// Notice
+// NVIDIA Corporation and its licensors retain all intellectual property and
+// proprietary rights in and to this software and related documentation and
+// any modifications thereto. Any use, reproduction, disclosure, or
+// distribution of this software and related documentation without an express
+// license agreement from NVIDIA Corporation is strictly prohibited.
+//
+// ALL NVIDIA DESIGN SPECIFICATIONS, CODE ARE PROVIDED "AS IS.". NVIDIA MAKES
+// NO WARRANTIES, EXPRESSED, IMPLIED, STATUTORY, OR OTHERWISE WITH RESPECT TO
+// THE MATERIALS, AND EXPRESSLY DISCLAIMS ALL IMPLIED WARRANTIES OF NONINFRINGEMENT,
+// MERCHANTABILITY, AND FITNESS FOR A PARTICULAR PURPOSE.
+//
+// Information and code furnished is believed to be accurate and reliable.
+// However, NVIDIA Corporation assumes no responsibility for the consequences of use of such
+// information or for any infringement of patents or other rights of third parties that may
+// result from its use. No license is granted by implication or otherwise under any patent
+// or patent rights of NVIDIA Corporation. Details are subject to change without notice.
+// This code supersedes and replaces all information previously supplied.
+// NVIDIA Corporation products are not authorized for use as critical
+// components in life support devices or systems without express written approval of
+// NVIDIA Corporation.
+//
+// Copyright (c) 2008-2016 NVIDIA Corporation. All rights reserved.
+// Copyright (c) 2004-2008 AGEIA Technologies, Inc. All rights reserved.
+// Copyright (c) 2001-2004 NovodeX AG. All rights reserved.  
+
+
+#ifndef PX_PHYSICS_COMMON_FLUSHPOOL
+#define PX_PHYSICS_COMMON_FLUSHPOOL
+
+#include "foundation/Px.h"
+#include "PsUserAllocated.h"
+#include "CmPhysXCommon.h"
+#include "PsMutex.h"
+#include "PsArray.h"
+#include "PsBitUtils.h"
+
+/*
+Pool used to allocate variable sized tasks. It's intended to be cleared after a short period (time step).
+*/
+
+namespace physx
+{
+namespace Cm
+{
+	static const PxU32 sSpareChunkCount = 2;
+
+	class FlushPool
+	{
+		PX_NOCOPY(FlushPool)
+	public:
+		FlushPool(PxU32 chunkSize) : mChunks(PX_DEBUG_EXP("FlushPoolChunk")), mChunkIndex(0), mOffset(0), mChunkSize(chunkSize)
+		{
+			mChunks.pushBack(static_cast<PxU8*>(PX_ALLOC(mChunkSize, "PxU8")));
+		}
+
+		~FlushPool()
+		{
+			for (PxU32 i = 0; i < mChunks.size(); ++i)
+				PX_FREE(mChunks[i]);
+		}
+
+		// alignment must be a power of two
+		void* allocate(PxU32 size, PxU32 alignment=16)
+		{
+			Ps::Mutex::ScopedLock lock(mMutex);
+			return allocateNotThreadSafe(size, alignment);
+		}
+
+		// alignment must be a power of two
+		void* allocateNotThreadSafe(PxU32 size, PxU32 alignment=16)
+		{
+			PX_ASSERT(shdfnd::isPowerOfTwo(alignment));
+			PX_ASSERT(size <= mChunkSize && !mChunks.empty());
+			
+			// padding for alignment
+			size_t unalignedStart = reinterpret_cast<size_t>(mChunks[mChunkIndex]+mOffset);
+			PxU32 pad = PxU32(((unalignedStart+alignment-1)&~(size_t(alignment)-1)) - unalignedStart);
+
+			if (mOffset + size + pad > mChunkSize)
+			{
+				mChunkIndex++;
+				mOffset = 0;
+				if (mChunkIndex >= mChunks.size())
+					mChunks.pushBack(static_cast<PxU8*>(PX_ALLOC(mChunkSize, "PxU8")));
+
+				// update padding to ensure new alloc is aligned
+				unalignedStart = reinterpret_cast<size_t>(mChunks[mChunkIndex]);
+				pad = PxU32(((unalignedStart+alignment-1)&~(size_t(alignment)-1)) - unalignedStart);
+			}
+
+			void* ptr = mChunks[mChunkIndex] + mOffset + pad;
+			PX_ASSERT((reinterpret_cast<size_t>(ptr)&(size_t(alignment)-1)) == 0);
+			mOffset += size + pad;
+			return ptr;
+		}
+
+		void clear(PxU32 spareChunkCount = sSpareChunkCount)
+		{
+			Ps::Mutex::ScopedLock lock(mMutex);
+			
+			clearNotThreadSafe(spareChunkCount);
+		}
+
+		void clearNotThreadSafe(PxU32 spareChunkCount = sSpareChunkCount)
+		{
+			PX_UNUSED(spareChunkCount);
+
+			//release memory not used previously
+			PxU32 targetSize = mChunkIndex+sSpareChunkCount;
+			while (mChunks.size() > targetSize)
+				PX_FREE(mChunks.popBack());
+
+			mChunkIndex = 0;
+			mOffset = 0;
+		}
+
+		void resetNotThreadSafe()
+		{
+			PxU8* firstChunk = mChunks[0];
+
+			for (PxU32 i = 1; i < mChunks.size(); ++i)
+				PX_FREE(mChunks[i]);
+
+			mChunks.clear();
+			mChunks.pushBack(firstChunk);
+			mChunkIndex = 0;
+			mOffset = 0;
+		}
+
+		void lock()
+		{
+			mMutex.lock();
+		}
+
+		void unlock()
+		{
+			mMutex.unlock();	
+		}
+
+	private:
+		Ps::Mutex mMutex;
+		Ps::Array<PxU8*> mChunks;
+		PxU32 mChunkIndex;
+		PxU32 mOffset;
+		PxU32 mChunkSize;
+	};
+
+	
+} // namespace Cm
+
+}
+
+#endif
diff --git a/PhysX_3.4/Source/Common/src/CmIDPool.h b/PhysX_3.4/Source/Common/src/CmIDPool.h
new file mode 100644
index 00000000..cce76ab3
--- /dev/null
+++ b/PhysX_3.4/Source/Common/src/CmIDPool.h
@@ -0,0 +1,202 @@
+// This code contains NVIDIA Confidential Information and is disclosed to you
+// under a form of NVIDIA software license agreement provided separately to you.
+//
+// Notice
+// NVIDIA Corporation and its licensors retain all intellectual property and
+// proprietary rights in and to this software and related documentation and
+// any modifications thereto. Any use, reproduction, disclosure, or
+// distribution of this software and related documentation without an express
+// license agreement from NVIDIA Corporation is strictly prohibited.
+//
+// ALL NVIDIA DESIGN SPECIFICATIONS, CODE ARE PROVIDED "AS IS.". NVIDIA MAKES
+// NO WARRANTIES, EXPRESSED, IMPLIED, STATUTORY, OR OTHERWISE WITH RESPECT TO
+// THE MATERIALS, AND EXPRESSLY DISCLAIMS ALL IMPLIED WARRANTIES OF NONINFRINGEMENT,
+// MERCHANTABILITY, AND FITNESS FOR A PARTICULAR PURPOSE.
+//
+// Information and code furnished is believed to be accurate and reliable.
+// However, NVIDIA Corporation assumes no responsibility for the consequences of use of such
+// information or for any infringement of patents or other rights of third parties that may
+// result from its use. No license is granted by implication or otherwise under any patent
+// or patent rights of NVIDIA Corporation. Details are subject to change without notice.
+// This code supersedes and replaces all information previously supplied.
+// NVIDIA Corporation products are not authorized for use as critical
+// components in life support devices or systems without express written approval of
+// NVIDIA Corporation.
+//
+// Copyright (c) 2008-2016 NVIDIA Corporation. All rights reserved.
+// Copyright (c) 2004-2008 AGEIA Technologies, Inc. All rights reserved.
+// Copyright (c) 2001-2004 NovodeX AG. All rights reserved.  
+
+
+#ifndef PX_PHYSICS_COMMON_ID_POOL
+#define PX_PHYSICS_COMMON_ID_POOL
+
+#include "foundation/Px.h"
+#include "CmPhysXCommon.h"
+#include "PsArray.h"
+#include "PsUserAllocated.h"
+
+namespace physx
+{
+namespace Cm
+{
+	template<class FreeBuffer>
+	class IDPoolBase : public Ps::UserAllocated
+	{
+	protected:
+		PxU32			mCurrentID;
+		FreeBuffer		mFreeIDs;
+	public:
+        IDPoolBase() : mCurrentID(0)	{}
+
+		void	freeID(PxU32 id)
+		{
+			// Allocate on first call
+			// Add released ID to the array of free IDs
+			if(id == (mCurrentID - 1))
+				--mCurrentID;
+			else
+				mFreeIDs.pushBack(id);
+		}
+
+		void	freeAll()
+		{
+			mCurrentID = 0;
+			mFreeIDs.clear();
+		}
+
+		PxU32	getNewID()
+		{
+			// If recycled IDs are available, use them
+			const PxU32 size = mFreeIDs.size();
+			if(size)
+			{
+				// Recycle last ID
+				return mFreeIDs.popBack();
+			}
+			// Else create a new ID
+			return mCurrentID++;
+		}
+
+		PxU32 getNumUsedID()	const
+		{
+			return mCurrentID - mFreeIDs.size();
+		}
+
+		PxU32 getMaxID() const
+		{
+			return mCurrentID;
+		}
+
+	};
+
+	//This class extends IDPoolBase. This is mainly used for when it is unsafe for the application to free the id immediately so that it can
+	//defer the free process until it is safe to do so
+	template<class FreeBuffer>
+	class DeferredIDPoolBase : public IDPoolBase<FreeBuffer>
+	{
+		FreeBuffer mDeferredFreeIDs;
+	public:
+		//release an index into the deferred list
+		void	deferredFreeID(PxU32 id)
+		{
+			mDeferredFreeIDs.pushBack(id);
+		}
+
+		//release the deferred indices into the free list
+		void processDeferredIds()
+		{
+			const PxU32 deferredFreeIDCount = mDeferredFreeIDs.size();
+			for(PxU32 a = 0; a < deferredFreeIDCount;++a)
+			{
+				this->mFreeIDs.pushBack(mDeferredFreeIDs[a]);
+			}
+			mDeferredFreeIDs.clear();
+		}
+
+		//release all indices
+		void	freeAll()
+		{
+			mDeferredFreeIDs.clear();
+			IDPoolBase<FreeBuffer>::freeAll();
+		}
+	};
+
+	//This is spu friendly fixed size array
+	template <typename T, uint32_t N>
+	class InlineFixedArray
+	{
+		T mArr[N];
+		PxU32 mSize;
+	public:
+
+		InlineFixedArray() : mSize(0)
+		{
+		}
+
+		~InlineFixedArray(){}
+
+		void pushBack(const T& t)
+		{
+			PX_ASSERT(mSize < N);
+			mArr[mSize++] = t;
+		}
+
+		T popBack()
+		{
+			PX_ASSERT(mSize > 0);
+			return mArr[--mSize];
+		}
+
+		void clear() { mSize = 0; }
+
+		T& operator [] (PxU32 index) { PX_ASSERT(index < N); return mArr[index]; }
+
+		const T& operator [] (PxU32 index) const { PX_ASSERT(index < N); return mArr[index]; }
+
+		PxU32 size() const { return mSize; }
+	};
+
+
+
+	//Fix size IDPool
+	template<PxU32 Capacity>
+	class InlineIDPool : public IDPoolBase<InlineFixedArray<PxU32, Capacity> >
+	{
+	public:
+		PxU32 getNumRemainingIDs()
+		{
+			return Capacity - this->getNumUsedID();
+		}
+	};
+
+	//Dynamic resize IDPool
+	class IDPool : public IDPoolBase<Ps::Array<PxU32> >
+	{
+	};
+
+
+	//This class is used to recycle indices. It supports deferred release, so that until processDeferredIds is called,
+	//released indices will not be reallocated. This class will fail if the calling code request more id than the InlineDeferredIDPoll
+	//has. It is the calling code's responsibility to ensure that this does not happen.
+	template<PxU32 Capacity>
+	class InlineDeferredIDPool : public DeferredIDPoolBase<InlineFixedArray<PxU32, Capacity> >
+	{
+	public:
+		PxU32 getNumRemainingIDs()
+		{
+			return Capacity - this->getNumUsedID();
+		}
+	};
+
+	//Dynamic resize DeferredIDPool
+	class DeferredIDPool : public DeferredIDPoolBase<Ps::Array<PxU32> >
+	{
+
+	};
+
+} // namespace Cm
+
+}
+
+#endif
diff --git a/PhysX_3.4/Source/Common/src/CmIO.h b/PhysX_3.4/Source/Common/src/CmIO.h
new file mode 100644
index 00000000..0913a54d
--- /dev/null
+++ b/PhysX_3.4/Source/Common/src/CmIO.h
@@ -0,0 +1,136 @@
+// This code contains NVIDIA Confidential Information and is disclosed to you
+// under a form of NVIDIA software license agreement provided separately to you.
+//
+// Notice
+// NVIDIA Corporation and its licensors retain all intellectual property and
+// proprietary rights in and to this software and related documentation and
+// any modifications thereto. Any use, reproduction, disclosure, or
+// distribution of this software and related documentation without an express
+// license agreement from NVIDIA Corporation is strictly prohibited.
+//
+// ALL NVIDIA DESIGN SPECIFICATIONS, CODE ARE PROVIDED "AS IS.". NVIDIA MAKES
+// NO WARRANTIES, EXPRESSED, IMPLIED, STATUTORY, OR OTHERWISE WITH RESPECT TO
+// THE MATERIALS, AND EXPRESSLY DISCLAIMS ALL IMPLIED WARRANTIES OF NONINFRINGEMENT,
+// MERCHANTABILITY, AND FITNESS FOR A PARTICULAR PURPOSE.
+//
+// Information and code furnished is believed to be accurate and reliable.
+// However, NVIDIA Corporation assumes no responsibility for the consequences of use of such
+// information or for any infringement of patents or other rights of third parties that may
+// result from its use. No license is granted by implication or otherwise under any patent
+// or patent rights of NVIDIA Corporation. Details are subject to change without notice.
+// This code supersedes and replaces all information previously supplied.
+// NVIDIA Corporation products are not authorized for use as critical
+// components in life support devices or systems without express written approval of
+// NVIDIA Corporation.
+//
+// Copyright (c) 2008-2016 NVIDIA Corporation. All rights reserved.
+// Copyright (c) 2004-2008 AGEIA Technologies, Inc. All rights reserved.
+// Copyright (c) 2001-2004 NovodeX AG. All rights reserved.  
+
+
+#ifndef PX_PHYSICS_COMMON_IO
+#define PX_PHYSICS_COMMON_IO
+
+#include "foundation/PxIO.h"
+#include "foundation/PxAssert.h"
+#include "foundation/PxMemory.h"
+#include "CmPhysXCommon.h"
+
+namespace physx
+{
+
+	// wrappers for IO classes so that we can add extra functionality (byte counting, buffering etc)
+
+namespace Cm
+{
+
+class InputStreamReader
+{
+public:
+
+	InputStreamReader(PxInputStream& stream) : mStream(stream) {	}
+	PxU32	read(void* dest, PxU32 count)	
+	{	
+		PxU32 readLength = mStream.read(dest, count);
+
+		// zero the buffer if we didn't get all the data
+		if(readLength<count)
+			PxMemZero(reinterpret_cast<PxU8*>(dest)+readLength, count-readLength);
+	
+		return readLength;
+	}
+
+	template <typename T> T get()			
+	{		
+		T val;	
+		PxU32 length = mStream.read(&val, sizeof(T));
+		PX_ASSERT(length == sizeof(T));
+		PX_UNUSED(length);
+		return val; 
+	}
+
+
+protected:
+	PxInputStream &mStream;
+private:
+	InputStreamReader& operator=(const InputStreamReader&);
+};
+
+
+class InputDataReader : public InputStreamReader
+{
+public:
+	InputDataReader(PxInputData& data) : InputStreamReader(data) {}
+	InputDataReader &operator=(const InputDataReader &);
+
+	PxU32	length() const					{		return getData().getLength();		}
+	void	seek(PxU32 offset)				{		getData().seek(offset);				}
+	PxU32	tell()							{		return getData().tell();			}
+
+private:
+	PxInputData& getData()					{		return static_cast<PxInputData&>(mStream); }
+	const PxInputData& getData() const		{		return static_cast<const PxInputData&>(mStream); }
+};
+
+
+class OutputStreamWriter
+{
+public:
+
+	PX_INLINE OutputStreamWriter(PxOutputStream& stream) 
+	:	mStream(stream)
+	,	mCount(0)
+	{}
+
+	PX_INLINE	PxU32	write(const void* src, PxU32 offset)		
+	{		
+		PxU32 count = mStream.write(src, offset);
+		mCount += count;
+		return count;
+	}
+
+	PX_INLINE	PxU32	getStoredSize()
+	{
+		return mCount;
+	}
+
+	template<typename T> void put(const T& val)	
+	{		
+		PxU32 length = write(&val, sizeof(T));		
+		PX_ASSERT(length == sizeof(T));
+		PX_UNUSED(length);
+	}
+
+private:
+
+	OutputStreamWriter& operator=(const OutputStreamWriter&);
+	PxOutputStream& mStream;
+	PxU32 mCount;
+};
+
+
+
+}
+}
+
+#endif
diff --git a/PhysX_3.4/Source/Common/src/CmMathUtils.cpp b/PhysX_3.4/Source/Common/src/CmMathUtils.cpp
new file mode 100644
index 00000000..72d7586f
--- /dev/null
+++ b/PhysX_3.4/Source/Common/src/CmMathUtils.cpp
@@ -0,0 +1,69 @@
+// This code contains NVIDIA Confidential Information and is disclosed to you
+// under a form of NVIDIA software license agreement provided separately to you.
+//
+// Notice
+// NVIDIA Corporation and its licensors retain all intellectual property and
+// proprietary rights in and to this software and related documentation and
+// any modifications thereto. Any use, reproduction, disclosure, or
+// distribution of this software and related documentation without an express
+// license agreement from NVIDIA Corporation is strictly prohibited.
+//
+// ALL NVIDIA DESIGN SPECIFICATIONS, CODE ARE PROVIDED "AS IS.". NVIDIA MAKES
+// NO WARRANTIES, EXPRESSED, IMPLIED, STATUTORY, OR OTHERWISE WITH RESPECT TO
+// THE MATERIALS, AND EXPRESSLY DISCLAIMS ALL IMPLIED WARRANTIES OF NONINFRINGEMENT,
+// MERCHANTABILITY, AND FITNESS FOR A PARTICULAR PURPOSE.
+//
+// Information and code furnished is believed to be accurate and reliable.
+// However, NVIDIA Corporation assumes no responsibility for the consequences of use of such
+// information or for any infringement of patents or other rights of third parties that may
+// result from its use. No license is granted by implication or otherwise under any patent
+// or patent rights of NVIDIA Corporation. Details are subject to change without notice.
+// This code supersedes and replaces all information previously supplied.
+// NVIDIA Corporation products are not authorized for use as critical
+// components in life support devices or systems without express written approval of
+// NVIDIA Corporation.
+//
+// Copyright (c) 2008-2016 NVIDIA Corporation. All rights reserved.
+// Copyright (c) 2004-2008 AGEIA Technologies, Inc. All rights reserved.
+// Copyright (c) 2001-2004 NovodeX AG. All rights reserved.  
+
+#include "foundation/PxPreprocessor.h"
+#include "foundation/PxTransform.h"
+#include "foundation/PxMathUtils.h"
+
+namespace physx
+{
+
+
+PX_FOUNDATION_API PxTransform PxTransformFromPlaneEquation(const PxPlane& plane)
+{
+	PxPlane p = plane; 
+	p.normalize();
+
+	// special case handling for axis aligned planes
+	const PxReal halfsqrt2 = 0.707106781;
+	PxQuat q;
+	if(2 == (p.n.x == 0.0f) + (p.n.y == 0.0f) + (p.n.z == 0.0f)) // special handling for axis aligned planes
+	{
+		if(p.n.x > 0)		q = PxQuat(PxIdentity);
+		else if(p.n.x < 0)	q = PxQuat(0, 0, 1.0f, 0);
+		else				q = PxQuat(0.0f, -p.n.z, p.n.y, 1.0f) * halfsqrt2;
+	}
+	else q = PxShortestRotation(PxVec3(1.f,0,0), p.n);
+
+	return PxTransform(-p.n * p.d, q);
+
+}
+
+PX_FOUNDATION_API PxTransform PxTransformFromSegment(const PxVec3& p0, const PxVec3& p1, PxReal* halfHeight)
+{
+	const PxVec3 axis = p1-p0;
+	const PxReal height = axis.magnitude();
+	if(halfHeight)
+		*halfHeight = height/2;
+
+	return PxTransform((p1+p0) * 0.5f, 
+						height<1e-6f ? PxQuat(PxIdentity) : PxShortestRotation(PxVec3(1.f,0,0), axis/height));		
+}
+
+}
diff --git a/PhysX_3.4/Source/Common/src/CmMatrix34.h b/PhysX_3.4/Source/Common/src/CmMatrix34.h
new file mode 100644
index 00000000..48f4e1f0
--- /dev/null
+++ b/PhysX_3.4/Source/Common/src/CmMatrix34.h
@@ -0,0 +1,286 @@
+// This code contains NVIDIA Confidential Information and is disclosed to you
+// under a form of NVIDIA software license agreement provided separately to you.
+//
+// Notice
+// NVIDIA Corporation and its licensors retain all intellectual property and
+// proprietary rights in and to this software and related documentation and
+// any modifications thereto. Any use, reproduction, disclosure, or
+// distribution of this software and related documentation without an express
+// license agreement from NVIDIA Corporation is strictly prohibited.
+//
+// ALL NVIDIA DESIGN SPECIFICATIONS, CODE ARE PROVIDED "AS IS.". NVIDIA MAKES
+// NO WARRANTIES, EXPRESSED, IMPLIED, STATUTORY, OR OTHERWISE WITH RESPECT TO
+// THE MATERIALS, AND EXPRESSLY DISCLAIMS ALL IMPLIED WARRANTIES OF NONINFRINGEMENT,
+// MERCHANTABILITY, AND FITNESS FOR A PARTICULAR PURPOSE.
+//
+// Information and code furnished is believed to be accurate and reliable.
+// However, NVIDIA Corporation assumes no responsibility for the consequences of use of such
+// information or for any infringement of patents or other rights of third parties that may
+// result from its use. No license is granted by implication or otherwise under any patent
+// or patent rights of NVIDIA Corporation. Details are subject to change without notice.
+// This code supersedes and replaces all information previously supplied.
+// NVIDIA Corporation products are not authorized for use as critical
+// components in life support devices or systems without express written approval of
+// NVIDIA Corporation.
+//
+// Copyright (c) 2008-2016 NVIDIA Corporation. All rights reserved.
+// Copyright (c) 2004-2008 AGEIA Technologies, Inc. All rights reserved.
+// Copyright (c) 2001-2004 NovodeX AG. All rights reserved.  
+
+
+#ifndef PX_PHYSICS_COMMON_MATRIX34
+#define PX_PHYSICS_COMMON_MATRIX34
+
+#include "foundation/PxVec3.h"
+#include "foundation/PxTransform.h"
+#include "foundation/PxMat33.h"
+
+#include "CmPhysXCommon.h"
+
+namespace physx
+{
+namespace Cm
+{
+
+/*!
+Basic mathematical 3x4 matrix, implemented as a 3x3 rotation matrix and a translation
+
+See PxMat33 for the format of the rotation matrix.
+
+*/
+
+class Matrix34
+{
+public:
+	//! Default constructor
+	PX_CUDA_CALLABLE PX_FORCE_INLINE Matrix34()
+	{}
+
+	//! Construct from four base vectors
+	PX_CUDA_CALLABLE PX_FORCE_INLINE Matrix34(const PxVec3& b0, const PxVec3& b1, const PxVec3& b2, const PxVec3& b3)
+		: m(b0, b1, b2), p(b3)
+	{}
+
+	//! Construct from float[12]
+	explicit PX_CUDA_CALLABLE PX_FORCE_INLINE Matrix34(PxReal values[]):
+		m(values), p(values[9], values[10], values[11])
+	{		
+	}
+
+	//! Construct from a 3x3 matrix
+	explicit PX_CUDA_CALLABLE PX_FORCE_INLINE Matrix34(const PxMat33& other)
+		: m(other), p(PxZero)
+	{
+	}
+
+	//! Construct from a 3x3 matrix and a translation vector
+	PX_CUDA_CALLABLE PX_FORCE_INLINE Matrix34(const PxMat33& other, const PxVec3& t)
+		: m(other), p(t)
+	{}
+
+	//! Construct from a PxTransform
+	explicit PX_CUDA_CALLABLE PX_FORCE_INLINE Matrix34(const PxTransform& other):
+		m(other.q), p(other.p)
+	{
+	}
+
+	//! Construct from a quaternion
+	explicit PX_CUDA_CALLABLE PX_FORCE_INLINE Matrix34(const PxQuat& q):
+		m(q), p(PxZero)
+	{
+	}
+
+	//! Copy constructor
+	PX_CUDA_CALLABLE PX_FORCE_INLINE Matrix34(const Matrix34& other):
+		m(other.m), p(other.p)
+	{
+	}
+
+	//! Assignment operator
+	PX_CUDA_CALLABLE PX_FORCE_INLINE const Matrix34& operator=(const Matrix34& other)
+	{
+		m = other.m;
+		p = other.p;
+		return *this;
+	}
+
+	//! Set to identity matrix
+	PX_CUDA_CALLABLE PX_FORCE_INLINE void setIdentity()
+	{
+		m = PxMat33(PxIdentity);
+		p = PxVec3(0);
+	}
+	
+	// Simpler operators
+	//! Equality operator
+	PX_CUDA_CALLABLE PX_FORCE_INLINE bool operator==(const Matrix34& other) const
+	{
+		return m == other.m && p == other.p;
+	}
+
+	//! Inequality operator
+	PX_CUDA_CALLABLE PX_FORCE_INLINE bool operator!=(const Matrix34& other) const
+	{
+		return !operator==(other);
+	}
+
+	//! Unary minus
+	PX_CUDA_CALLABLE PX_FORCE_INLINE Matrix34 operator-() const
+	{
+		return Matrix34(-m, -p);
+	}
+
+	//! Add
+	PX_CUDA_CALLABLE PX_FORCE_INLINE Matrix34 operator+(const Matrix34& other) const
+	{
+		return Matrix34(m + other.m, p + other.p);
+	}
+
+	//! Subtract
+	PX_CUDA_CALLABLE PX_FORCE_INLINE Matrix34 operator-(const Matrix34& other) const
+	{
+		return Matrix34(m - other.m, p - other.p);
+	}
+
+	//! Scalar multiplication
+	PX_CUDA_CALLABLE PX_FORCE_INLINE Matrix34 operator*(PxReal scalar) const
+	{
+		return Matrix34(m*scalar, p*scalar);
+	}
+
+	friend Matrix34 operator*(PxReal, const Matrix34&);
+
+	//! Matrix multiplication
+	PX_CUDA_CALLABLE PX_FORCE_INLINE Matrix34 operator*(const Matrix34& other) const
+	{
+		//Rows from this <dot> columns from other
+		//base0 = rotate(other.m.column0) etc
+		return Matrix34(m*other.m, m*other.p + p);
+	}
+
+	//! Matrix multiplication, extend the second matrix
+	PX_CUDA_CALLABLE PX_FORCE_INLINE Matrix34 operator*(const PxMat33& other) const
+	{
+		//Rows from this <dot> columns from other
+		//base0 = transform(other.m.column0) etc
+		return Matrix34(m*other, p);
+	}
+
+	friend Matrix34 operator*(const PxMat33& a, const Matrix34& b);
+	
+	// a <op>= b operators
+
+	//! Equals-add
+	PX_CUDA_CALLABLE PX_FORCE_INLINE Matrix34& operator+=(const Matrix34& other)
+	{
+		m += other.m;
+		p += other.p;
+		return *this;
+	}
+
+	//! Equals-sub
+	PX_CUDA_CALLABLE PX_FORCE_INLINE Matrix34& operator-=(const Matrix34& other)
+	{
+		m -= other.m;
+		p -= other.p;
+		return *this;
+	}
+
+	//! Equals scalar multiplication
+	PX_CUDA_CALLABLE PX_FORCE_INLINE Matrix34& operator*=(PxReal scalar)
+	{
+		m *= scalar;
+		p *= scalar;
+
+		return *this;
+	}
+
+	//! Element access, mathematical way!
+	PX_CUDA_CALLABLE PX_FORCE_INLINE PxReal operator()(PxU32 row, PxU32 col) const
+	{
+		return (*this)[col][row];
+	}
+
+	//! Element access, mathematical way!
+	PX_CUDA_CALLABLE PX_FORCE_INLINE PxReal& operator()(PxU32 row, PxU32 col)
+	{
+		return (*this)[col][row];
+	}
+
+	// Transform etc
+	
+	//! Transform vector by matrix, equal to v' = M*v
+	PX_CUDA_CALLABLE PX_FORCE_INLINE PxVec3 rotate(const PxVec3& other) const
+	{
+		return m*other;
+	}
+
+	//! Transform vector by transpose of matrix, equal to v' = M^t*v
+	PX_CUDA_CALLABLE PX_FORCE_INLINE PxVec3 rotateTranspose(const PxVec3& other) const
+	{
+		return m.transformTranspose(other);
+	}
+
+	//! Transform point by matrix
+	PX_CUDA_CALLABLE PX_FORCE_INLINE PxVec3 transform(const PxVec3& other) const
+	{
+		return m*other + p;
+	}
+
+	//! Transform point by transposed matrix
+	PX_CUDA_CALLABLE PX_FORCE_INLINE PxVec3 transformTranspose(const PxVec3& other) const
+	{
+		return m.transformTranspose(other - p);
+	}
+
+	//! Transform point by transposed matrix
+	PX_CUDA_CALLABLE PX_FORCE_INLINE Cm::Matrix34 transformTranspose(const Cm::Matrix34& other) const
+	{
+		return Cm::Matrix34(m.transformTranspose(other.m.column0), 
+							m.transformTranspose(other.m.column1), 
+							m.transformTranspose(other.m.column2), 
+							m.transformTranspose(other.p - p));
+	}
+
+
+	//! Invert matrix treating it as a rotation+translation matrix only
+	PX_CUDA_CALLABLE PX_FORCE_INLINE Matrix34 getInverseRT() const
+	{
+		return Matrix34(m.getTranspose(), m.transformTranspose(-p));
+	}
+
+
+	// Conversion
+	//! Set matrix from quaternion
+	PX_CUDA_CALLABLE PX_FORCE_INLINE void set(const PxQuat& q) 
+	{
+		m = PxMat33(q);
+		p = PxVec3(PxZero);
+	}
+
+
+	PX_CUDA_CALLABLE PX_FORCE_INLINE		PxVec3& operator[](unsigned int num){return (&m.column0)[num];}
+	PX_CUDA_CALLABLE PX_FORCE_INLINE		PxVec3& operator[](int num)			{ return (&m.column0)[num]; }
+	PX_CUDA_CALLABLE PX_FORCE_INLINE const	PxVec3& operator[](unsigned int num) const	{ return (&m.column0)[num]; }
+	PX_CUDA_CALLABLE PX_FORCE_INLINE const	PxVec3& operator[](int num) const	{ return (&m.column0)[num]; }
+
+	//Data, see above for format!
+
+	PxMat33 m;
+	PxVec3 p;
+
+};
+
+
+//! Multiply a*b, a is extended
+PX_INLINE Matrix34 operator*(const PxMat33& a, const Matrix34& b)
+{
+	return Matrix34(a * b.m, a * b.p);
+}
+
+
+} // namespace Cm
+
+}
+
+#endif
diff --git a/PhysX_3.4/Source/Common/src/CmPhysXCommon.h b/PhysX_3.4/Source/Common/src/CmPhysXCommon.h
new file mode 100644
index 00000000..a6e911bb
--- /dev/null
+++ b/PhysX_3.4/Source/Common/src/CmPhysXCommon.h
@@ -0,0 +1,89 @@
+// This code contains NVIDIA Confidential Information and is disclosed to you
+// under a form of NVIDIA software license agreement provided separately to you.
+//
+// Notice
+// NVIDIA Corporation and its licensors retain all intellectual property and
+// proprietary rights in and to this software and related documentation and
+// any modifications thereto. Any use, reproduction, disclosure, or
+// distribution of this software and related documentation without an express
+// license agreement from NVIDIA Corporation is strictly prohibited.
+//
+// ALL NVIDIA DESIGN SPECIFICATIONS, CODE ARE PROVIDED "AS IS.". NVIDIA MAKES
+// NO WARRANTIES, EXPRESSED, IMPLIED, STATUTORY, OR OTHERWISE WITH RESPECT TO
+// THE MATERIALS, AND EXPRESSLY DISCLAIMS ALL IMPLIED WARRANTIES OF NONINFRINGEMENT,
+// MERCHANTABILITY, AND FITNESS FOR A PARTICULAR PURPOSE.
+//
+// Information and code furnished is believed to be accurate and reliable.
+// However, NVIDIA Corporation assumes no responsibility for the consequences of use of such
+// information or for any infringement of patents or other rights of third parties that may
+// result from its use. No license is granted by implication or otherwise under any patent
+// or patent rights of NVIDIA Corporation. Details are subject to change without notice.
+// This code supersedes and replaces all information previously supplied.
+// NVIDIA Corporation products are not authorized for use as critical
+// components in life support devices or systems without express written approval of
+// NVIDIA Corporation.
+//
+// Copyright (c) 2008-2016 NVIDIA Corporation. All rights reserved.
+// Copyright (c) 2004-2008 AGEIA Technologies, Inc. All rights reserved.
+// Copyright (c) 2001-2004 NovodeX AG. All rights reserved.  
+
+
+#ifndef PX_PHYSICS_COMMON
+#define PX_PHYSICS_COMMON
+
+//! \file Top level internal include file for PhysX SDK
+
+#include "Ps.h"
+
+// Enable debug visualization
+#define PX_ENABLE_DEBUG_VISUALIZATION	1
+
+// Enable simulation statistics generation
+#define PX_ENABLE_SIM_STATS 1
+
+// PT: typical "invalid" value in various CD algorithms
+#define	PX_INVALID_U32		0xffffffff
+#define PX_INVALID_U16		0xffff
+
+// PT: this used to be replicated everywhere in the code, causing bugs to sometimes reappear (e.g. TTP 3587).
+// It is better to define it in a header and use the same constant everywhere. The original value (1e-05f)
+// caused troubles (e.g. TTP 1705, TTP 306).
+#define PX_PARALLEL_TOLERANCE	1e-02f
+
+#define PX_USE_16_BIT_HANDLES 0
+
+namespace physx
+{
+	// alias shared foundation to something usable
+	namespace Ps = shdfnd;
+}
+
+#if PX_CHECKED
+	#define PX_CHECK_MSG(exp, msg)				(!!(exp) || (physx::shdfnd::getFoundation().error(physx::PxErrorCode::eINVALID_PARAMETER, __FILE__, __LINE__, msg), 0) )
+	#define PX_CHECK(exp)						PX_CHECK_MSG(exp, #exp)
+	#define PX_CHECK_AND_RETURN(exp,msg)		{ if(!(exp)) { PX_CHECK_MSG(exp, msg); return; } }
+	#define PX_CHECK_AND_RETURN_NULL(exp,msg)	{ if(!(exp)) { PX_CHECK_MSG(exp, msg); return 0; } }
+	#define PX_CHECK_AND_RETURN_VAL(exp,msg,r)	{ if(!(exp)) { PX_CHECK_MSG(exp, msg); return r; } }
+#else
+	#define PX_CHECK_MSG(exp, msg)
+	#define PX_CHECK(exp)
+	#define PX_CHECK_AND_RETURN(exp,msg)
+	#define PX_CHECK_AND_RETURN_NULL(exp,msg)
+	#define PX_CHECK_AND_RETURN_VAL(exp,msg,r)
+#endif
+
+#if PX_VC
+	// VC compiler defines __FUNCTION__ as a string literal so it is possible to concatenate it with another string
+	// Example: #define PX_CHECK_VALID(x)	PX_CHECK_MSG(physx::shdfnd::checkValid(x), __FUNCTION__ ": parameter invalid!")
+	#define PX_CHECK_VALID(x)				PX_CHECK_MSG(physx::shdfnd::checkValid(x), __FUNCTION__)
+#elif PX_GCC_FAMILY
+	// GCC compiler defines __FUNCTION__ as a variable, hence, it is NOT possible concatenate an additional string to it
+	// In GCC, __FUNCTION__ only returns the function name, using __PRETTY_FUNCTION__ will return the full function definition
+	#define PX_CHECK_VALID(x)				PX_CHECK_MSG(physx::shdfnd::checkValid(x), __PRETTY_FUNCTION__)
+#else
+	// Generic macro for other compilers
+	#define PX_CHECK_VALID(x)				PX_CHECK_MSG(physx::shdfnd::checkValid(x), __FUNCTION__)
+#endif
+
+
+#endif
diff --git a/PhysX_3.4/Source/Common/src/CmPool.h b/PhysX_3.4/Source/Common/src/CmPool.h
new file mode 100644
index 00000000..a83d33ca
--- /dev/null
+++ b/PhysX_3.4/Source/Common/src/CmPool.h
@@ -0,0 +1,292 @@
+// This code contains NVIDIA Confidential Information and is disclosed to you
+// under a form of NVIDIA software license agreement provided separately to you.
+//
+// Notice
+// NVIDIA Corporation and its licensors retain all intellectual property and
+// proprietary rights in and to this software and related documentation and
+// any modifications thereto. Any use, reproduction, disclosure, or
+// distribution of this software and related documentation without an express
+// license agreement from NVIDIA Corporation is strictly prohibited.
+//
+// ALL NVIDIA DESIGN SPECIFICATIONS, CODE ARE PROVIDED "AS IS.". NVIDIA MAKES
+// NO WARRANTIES, EXPRESSED, IMPLIED, STATUTORY, OR OTHERWISE WITH RESPECT TO
+// THE MATERIALS, AND EXPRESSLY DISCLAIMS ALL IMPLIED WARRANTIES OF NONINFRINGEMENT,
+// MERCHANTABILITY, AND FITNESS FOR A PARTICULAR PURPOSE.
+//
+// Information and code furnished is believed to be accurate and reliable.
+// However, NVIDIA Corporation assumes no responsibility for the consequences of use of such
+// information or for any infringement of patents or other rights of third parties that may
+// result from its use. No license is granted by implication or otherwise under any patent
+// or patent rights of NVIDIA Corporation. Details are subject to change without notice.
+// This code supersedes and replaces all information previously supplied.
+// NVIDIA Corporation products are not authorized for use as critical
+// components in life support devices or systems without express written approval of
+// NVIDIA Corporation.
+//
+// Copyright (c) 2008-2016 NVIDIA Corporation. All rights reserved.
+// Copyright (c) 2004-2008 AGEIA Technologies, Inc. All rights reserved.
+// Copyright (c) 2001-2004 NovodeX AG. All rights reserved.  
+
+
+#ifndef CM_POOL_H
+#define CM_POOL_H
+
+#include "PsSort.h"
+#include "PsMutex.h"
+#include "PsBasicTemplates.h"
+
+#include "CmBitMap.h"
+#include "CmPhysXCommon.h"
+
+namespace physx
+{
+namespace Cm
+{
+
+/*!
+Allocator for pools of data structures
+Also decodes indices (which can be computed from handles) into objects. To make this
+faster, the EltsPerSlab must be a power of two
+*/
+template <class T, class ArgumentType> 
+class PoolList : public Ps::AllocatorTraits<T>::Type
+{
+	typedef typename Ps::AllocatorTraits<T>::Type Alloc;
+	PX_NOCOPY(PoolList)
+public:
+	PX_INLINE PoolList(const Alloc& alloc, ArgumentType* argument, PxU32 eltsPerSlab, PxU32 maxSlabs)
+		: Alloc(alloc),
+		mEltsPerSlab(eltsPerSlab), 
+		mMaxSlabs(maxSlabs), 
+		mSlabCount(0),
+		mFreeList(0), 
+		mFreeCount(0), 
+		mSlabs(reinterpret_cast<T**>(Alloc::allocate(maxSlabs * sizeof(T*), __FILE__, __LINE__))),
+		mArgument(argument)
+	{
+		PX_ASSERT(mEltsPerSlab>0);
+		// either maxSlabs = 1 (non-resizable pool), or elts per slab must be a power of two
+		PX_ASSERT((maxSlabs==1) || ((maxSlabs < 8192) && (mEltsPerSlab & (mEltsPerSlab-1))) == 0);		
+		mLog2EltsPerSlab = 0;
+
+		if(mMaxSlabs>1)
+		{
+			for(mLog2EltsPerSlab=0; mEltsPerSlab!=PxU32(1<<mLog2EltsPerSlab); mLog2EltsPerSlab++)
+				;
+		}
+	}
+
+	PX_INLINE ~PoolList()
+	{
+		destroy();
+	}
+
+	PX_INLINE void destroy()
+	{
+		// Run all destructors
+		for(PxU32 i=0;i<mSlabCount;i++)
+		{
+			PX_ASSERT(mSlabs);
+			T* slab = mSlabs[i];
+			for(PxU32 j=0;j<mEltsPerSlab;j++)
+			{
+				slab[j].~T();
+			}
+		}
+
+		//Deallocate
+		for(PxU32 i=0;i<mSlabCount;i++)
+		{
+			PX_FREE(mSlabs[i]);
+			mSlabs[i] = NULL;
+		}
+		mSlabCount = 0;
+
+		if(mFreeList)
+			PX_FREE(mFreeList);
+		mFreeList = NULL;
+		if(mSlabs)
+		{
+			PX_FREE(mSlabs);
+			mSlabs = NULL;
+		}
+	}
+
+	PxU32 preallocate(const PxU32 nbRequired, T** elements)
+	{
+		//(1) Allocate and pull out an array of X elements
+
+		PxU32 nbToAllocate = nbRequired > mFreeCount ? nbRequired - mFreeCount : 0;
+
+		PxU32 nbElements = nbRequired - nbToAllocate;
+
+		PxMemCopy(elements, mFreeList + (mFreeCount - nbElements), sizeof(T*) * nbElements);
+		//PxU32 originalFreeCount = mFreeCount;
+		mFreeCount -= nbElements;
+
+		if (nbToAllocate)
+		{
+			PX_ASSERT(mFreeCount == 0);
+
+			PxU32 nbSlabs = (nbToAllocate + mEltsPerSlab - 1) / mEltsPerSlab; //The number of slabs we need to allocate...
+
+			if (mSlabCount + nbSlabs >= mMaxSlabs)
+				return nbElements; //Return only nbFromFree because we're not going to allocate any slabs. Seriously, we need to nuke this "maxSlabs" stuff ASAP!
+
+			//allocate our slabs...
+
+			PxU32 freeCount = mFreeCount;
+
+			for (PxU32 i = 0; i < nbSlabs; ++i)
+			{
+
+				//KS - would be great to allocate this using a single allocation but it will make releasing slabs fail later :(
+				T * mAddr = reinterpret_cast<T*>(Alloc::allocate(mEltsPerSlab * sizeof(T), __FILE__, __LINE__));
+				if (!mAddr)
+					return nbElements; //Allocation failed so only return the set of elements we could allocate from the free list
+
+				mSlabs[mSlabCount++] = mAddr;
+
+				// Make sure the usage bitmap is up-to-size
+				if (mUseBitmap.size() < mSlabCount*mEltsPerSlab)
+				{
+					mUseBitmap.resize(2 * mSlabCount*mEltsPerSlab); //set last element as not used
+					if (mFreeList)
+						PX_FREE(mFreeList);
+					mFreeList = reinterpret_cast<T**>(Alloc::allocate(2 * mSlabCount * mEltsPerSlab * sizeof(T*), __FILE__, __LINE__));
+				}
+
+				PxU32 baseIndex = (mSlabCount-1) * mEltsPerSlab;
+
+				//Now add all these to the mFreeList and elements...
+				PxI32 idx = PxI32(mEltsPerSlab - 1);
+
+				for (; idx >= PxI32(nbToAllocate); --idx)
+				{
+					mFreeList[freeCount++] = new(mAddr + idx) T(mArgument, baseIndex + idx);
+				}
+
+				PxU32 origElements = nbElements;
+				T** writeIdx = elements + nbElements;
+				for (; idx >= 0; --idx)
+				{
+					writeIdx[idx] = new(mAddr + idx) T(mArgument, baseIndex + idx);
+					nbElements++;
+				}
+
+				nbToAllocate -= (nbElements - origElements);
+			}
+
+			mFreeCount = freeCount;
+		}
+		
+		PX_ASSERT(nbElements == nbRequired);
+
+		for (PxU32 a = 0; a < nbElements; ++a)
+		{
+			mUseBitmap.set(elements[a]->getIndex());
+		}
+
+		return nbRequired;
+	}
+
+	// TODO: would be nice to add templated construct/destroy methods like ObjectPool
+
+	PX_INLINE T* get()
+	{
+		if(mFreeCount == 0 && !extend())
+			return 0;
+		T* element = mFreeList[--mFreeCount];
+		mUseBitmap.set(element->getIndex());
+		return element;
+	}
+
+	PX_INLINE void put(T* element)
+	{
+		PxU32 i = element->getIndex();
+		mUseBitmap.reset(i);
+		mFreeList[mFreeCount++] = element;
+	}
+
+	/*
+		WARNING: Unlike findByIndexFast below, this method is NOT safe to use if another thread 
+		is concurrently updating the pool (e.g. through put/get/extend/getIterator), since the
+		safety boundedTest uses mSlabCount and mUseBitmap.
+	*/
+	PX_FORCE_INLINE T* findByIndex(PxU32 index) const
+	{
+		if(index>=mSlabCount*mEltsPerSlab || !(mUseBitmap.boundedTest(index)))
+			return 0;
+		return mMaxSlabs==1 ? mSlabs[0]+index : mSlabs[index>>mLog2EltsPerSlab] + (index&(mEltsPerSlab-1));
+	}
+
+	/*
+		This call is safe to do while other threads update the pool.
+	*/
+	PX_FORCE_INLINE T* findByIndexFast(PxU32 index) const
+	{
+		PX_ASSERT(mMaxSlabs != 1);
+		return mSlabs[index>>mLog2EltsPerSlab] + (index&(mEltsPerSlab-1));
+	}
+
+	bool extend()
+	{
+		if(mSlabCount == mMaxSlabs)
+			return false;
+		T * mAddr = reinterpret_cast<T*>(Alloc::allocate(mEltsPerSlab * sizeof(T), __FILE__, __LINE__));
+		if(!mAddr)
+			return false;
+		mSlabs[mSlabCount++] = mAddr;
+
+
+
+		// Make sure the usage bitmap is up-to-size
+		if(mUseBitmap.size() < mSlabCount*mEltsPerSlab)
+		{
+			mUseBitmap.resize(2*mSlabCount*mEltsPerSlab); //set last element as not used
+			if(mFreeList)
+				PX_FREE(mFreeList);
+			mFreeList = reinterpret_cast<T**>(Alloc::allocate(2*mSlabCount * mEltsPerSlab * sizeof(T*), __FILE__, __LINE__));
+		}
+	
+		// Add to free list in descending order so that lowest indices get allocated first - 
+		// the FW context code currently *relies* on this behavior to grab the zero-index volume
+		// which can't be allocated to the user. TODO: fix this
+
+		PxU32 baseIndex = (mSlabCount-1) * mEltsPerSlab;
+		PxU32 freeCount = mFreeCount;
+		for(PxI32 i=PxI32(mEltsPerSlab-1);i>=0;i--)
+			mFreeList[freeCount++] = new(mAddr+i) T(mArgument, baseIndex+ i);
+
+		mFreeCount = freeCount;
+
+		return true;
+	}
+
+	PX_INLINE PxU32 getMaxUsedIndex()	const
+	{
+		return mUseBitmap.findLast();
+	}
+
+	PX_INLINE BitMap::Iterator getIterator() const
+	{
+		return BitMap::Iterator(mUseBitmap);
+	}
+
+private:
+	const PxU32				mEltsPerSlab;
+	const PxU32				mMaxSlabs;
+	PxU32					mSlabCount;
+	PxU32					mLog2EltsPerSlab;
+	T**						mFreeList;
+	PxU32					mFreeCount;
+	T**						mSlabs;
+	ArgumentType*			mArgument;
+	BitMap					mUseBitmap;
+};
+
+
+}
+}
+
+#endif
diff --git a/PhysX_3.4/Source/Common/src/CmPreallocatingPool.h b/PhysX_3.4/Source/Common/src/CmPreallocatingPool.h
new file mode 100644
index 00000000..da8bba92
--- /dev/null
+++ b/PhysX_3.4/Source/Common/src/CmPreallocatingPool.h
@@ -0,0 +1,435 @@
+// This code contains NVIDIA Confidential Information and is disclosed to you
+// under a form of NVIDIA software license agreement provided separately to you.
+//
+// Notice
+// NVIDIA Corporation and its licensors retain all intellectual property and
+// proprietary rights in and to this software and related documentation and
+// any modifications thereto. Any use, reproduction, disclosure, or
+// distribution of this software and related documentation without an express
+// license agreement from NVIDIA Corporation is strictly prohibited.
+//
+// ALL NVIDIA DESIGN SPECIFICATIONS, CODE ARE PROVIDED "AS IS.". NVIDIA MAKES
+// NO WARRANTIES, EXPRESSED, IMPLIED, STATUTORY, OR OTHERWISE WITH RESPECT TO
+// THE MATERIALS, AND EXPRESSLY DISCLAIMS ALL IMPLIED WARRANTIES OF NONINFRINGEMENT,
+// MERCHANTABILITY, AND FITNESS FOR A PARTICULAR PURPOSE.
+//
+// Information and code furnished is believed to be accurate and reliable.
+// However, NVIDIA Corporation assumes no responsibility for the consequences of use of such
+// information or for any infringement of patents or other rights of third parties that may
+// result from its use. No license is granted by implication or otherwise under any patent
+// or patent rights of NVIDIA Corporation. Details are subject to change without notice.
+// This code supersedes and replaces all information previously supplied.
+// NVIDIA Corporation products are not authorized for use as critical
+// components in life support devices or systems without express written approval of
+// NVIDIA Corporation.
+//
+// Copyright (c) 2008-2016 NVIDIA Corporation. All rights reserved.
+// Copyright (c) 2004-2008 AGEIA Technologies, Inc. All rights reserved.
+// Copyright (c) 2001-2004 NovodeX AG. All rights reserved.  
+
+
+#ifndef PX_PHYSICS_COMMON_PREALLOCATINGPOOL
+#define PX_PHYSICS_COMMON_PREALLOCATINGPOOL
+
+#include "foundation/Px.h"
+#include "PsUserAllocated.h"
+#include "CmPhysXCommon.h"
+#include "PsSort.h"
+#include "PsArray.h"
+
+/*
+Pool used to allocate variable sized tasks. It's intended to be cleared after a short period (time step).
+*/
+
+namespace physx
+{
+namespace Cm
+{
+
+class PreallocatingRegion
+{
+public:
+	PX_FORCE_INLINE		PreallocatingRegion() : mMemory(NULL), mFirstFree(NULL), mNbElements(0)	{}
+
+	void		init(PxU32 maxElements, PxU32 elementSize, const char* typeName)
+	{	
+		mFirstFree	= NULL;
+		mNbElements	= 0;
+		PX_ASSERT(typeName);
+		PX_UNUSED(typeName);
+		mMemory = reinterpret_cast<PxU8*>(PX_ALLOC(sizeof(PxU8)*elementSize*maxElements, typeName?typeName:"SceneSim Pool"));	// ### addActor alloc
+		PX_ASSERT(elementSize*maxElements>=sizeof(void*));
+	}
+
+	void		reset()
+	{
+		PX_FREE_AND_RESET(mMemory);
+	}
+
+	PX_FORCE_INLINE PxU8* allocateMemory(PxU32 maxElements, PxU32 elementSize)
+	{
+		if(mFirstFree)
+		{
+			PxU8* recycled = reinterpret_cast<PxU8*>(mFirstFree);
+
+			void** recycled32 = reinterpret_cast<void**>(recycled);
+			mFirstFree = *recycled32;
+
+			return recycled;
+		}
+		else
+		{
+			if(mNbElements==maxElements)
+				return NULL;	// Out of memory
+
+			const PxU32 freeIndex = mNbElements++;
+			return mMemory + freeIndex * elementSize;
+		}
+	}
+
+	void		deallocateMemory(PxU32 maxElements, PxU32 elementSize, PxU8* element)
+	{
+		PX_ASSERT(element);
+		PX_ASSERT(element>=mMemory && element<mMemory + maxElements * elementSize);
+		PX_UNUSED(elementSize);
+		PX_UNUSED(maxElements);
+
+		void** recycled32 = reinterpret_cast<void**>(element);
+		*recycled32 = mFirstFree;
+
+		mFirstFree = element;
+	}
+
+	PX_FORCE_INLINE bool operator < (const PreallocatingRegion& p) const
+	{
+		return mMemory < p.mMemory;
+	}
+
+	PX_FORCE_INLINE bool operator > (const PreallocatingRegion& p) const
+	{
+		return mMemory > p.mMemory;
+	}
+
+	PxU8*		mMemory;
+	void*		mFirstFree;
+	PxU32		mNbElements;
+};
+
+
+
+class PreallocatingRegionManager
+{
+	public:
+						PreallocatingRegionManager(PxU32 maxElements, PxU32 elementSize, const char* typeName)
+						: mMaxElements		(maxElements)
+						, mElementSize		(elementSize)
+						, mActivePoolIndex	(0)
+						, mPools(PX_DEBUG_EXP("MyPoolManagerPools"))
+						, mNeedsSorting		(true)
+						, mTypeName			(typeName)
+						{
+							PreallocatingRegion tmp;
+							tmp.init(maxElements, elementSize, mTypeName);
+							mPools.pushBack(tmp);
+						}
+
+						~PreallocatingRegionManager()
+						{
+							const PxU32 nbPools = mPools.size();
+							for(PxU32 i=0;i<nbPools;i++)
+								mPools[i].reset();
+						}
+
+	void				preAllocate(PxU32 n)
+	{
+		if(!n)
+			return;
+
+		const PxU32 nbPools = mPools.size();
+		const PxU32 maxElements = mMaxElements;
+		const PxU32 elementSize = mElementSize;
+		PxU32 availableSpace = nbPools * maxElements;
+
+		while(n>availableSpace)
+		{
+			PreallocatingRegion tmp;
+			tmp.init(maxElements, elementSize, mTypeName);
+			mPools.pushBack(tmp);
+
+			availableSpace += maxElements;
+		}
+	}
+
+	PX_FORCE_INLINE PxU8* allocateMemory()
+	{
+		PX_ASSERT(mActivePoolIndex<mPools.size());
+		PxU8* memory = mPools[mActivePoolIndex].allocateMemory(mMaxElements, mElementSize);
+		return memory ? memory : searchForMemory();
+	}
+
+	void				deallocateMemory(PxU8* element)
+	{
+		if(!element)
+			return;
+
+		if(mNeedsSorting)
+			Ps::sort(mPools.begin(), mPools.size());
+
+		const PxU32 maxElements = mMaxElements;
+		const PxU32 elementSize = mElementSize;
+		const PxU32 slabSize = maxElements * elementSize;
+		const PxU32 nbPools = mPools.size();
+
+		// O(log n) search
+		int first = 0;
+		int last = int(nbPools-1);
+
+		while(first<=last)
+		{
+			const int mid = (first+last)>>1;
+
+			PreallocatingRegion& candidate = mPools[PxU32(mid)];
+			if(contains(candidate.mMemory, slabSize, element))
+			{
+				candidate.deallocateMemory(maxElements, elementSize, element);
+				
+				// when we sorted earlier we trashed the active index, but at least this region has a free element
+				if(mNeedsSorting)	
+					mActivePoolIndex = PxU32(mid);
+
+				mNeedsSorting = false;
+				return;
+			}
+
+			if(candidate.mMemory<element)
+				first = mid+1;
+			else
+				last = mid-1;
+		}
+
+		PX_ASSERT(0);
+	}
+
+
+private:
+
+	PreallocatingRegionManager& operator=(const PreallocatingRegionManager&);
+	PxU8*				searchForMemory()
+	{
+		const PxU32 nbPools = mPools.size();
+		const PxU32 activePoolIndex = mActivePoolIndex;
+		const PxU32 maxElements = mMaxElements;
+		const PxU32 elementSize = mElementSize;
+
+
+		for(PxU32 i=0;i<nbPools;i++)
+		{
+			if(i==activePoolIndex)
+				continue;
+
+			PxU8* memory = mPools[i].allocateMemory(maxElements, elementSize);
+			if(memory)
+			{
+				mActivePoolIndex = i;
+				return memory;
+			}
+		}
+
+		mActivePoolIndex = nbPools;
+		mNeedsSorting = true;
+
+		PreallocatingRegion tmp;
+		tmp.init(maxElements, elementSize, mTypeName);
+
+		PreallocatingRegion& newPool = mPools.pushBack(tmp);		// ### addActor alloc (StaticSim, ShapeSim, SceneQueryShapeData)
+		return newPool.allocateMemory(maxElements, elementSize);
+	}
+
+
+
+	PX_FORCE_INLINE	bool contains(PxU8* memory, const PxU32 slabSize, PxU8* element)
+	{
+		return element>=memory && element<memory+slabSize;
+	}
+
+
+
+	const PxU32			mMaxElements;
+	const PxU32			mElementSize;
+	PxU32				mActivePoolIndex;
+
+	Ps::Array<PreallocatingRegion>	mPools;
+	bool				mNeedsSorting;
+	const char*			mTypeName;
+};
+
+template<class T>
+class PreallocatingPool : public Ps::UserAllocated
+{
+	PreallocatingPool<T>& operator=(const PreallocatingPool<T>&);
+
+public:
+	PreallocatingPool(PxU32 maxElements, const char* typeName) : mPool(maxElements, sizeof(T), typeName)
+	{
+	}
+
+	~PreallocatingPool()
+	{
+	}
+
+	PX_FORCE_INLINE	void	preAllocate(PxU32 n)
+	{
+		mPool.preAllocate(n);
+	}
+
+	PX_INLINE T* allocate()
+	{
+		return reinterpret_cast<T*>(mPool.allocateMemory());
+	}
+
+	PX_FORCE_INLINE T* allocateAndPrefetch()
+	{
+		T* t = reinterpret_cast<T*>(mPool.allocateMemory());
+		Ps::prefetch(t, sizeof(T));
+		return t;
+	}
+
+	PX_INLINE T* construct()
+	{
+		T* t = reinterpret_cast<T*>(mPool.allocateMemory());
+		return t ? new (t) T() : 0;
+	}
+
+	template<class A1>
+	PX_INLINE T* construct(A1& a)
+	{
+		T* t = reinterpret_cast<T*>(mPool.allocateMemory());
+		return t ? new (t) T(a) : 0;
+	}
+
+	template<class A1, class A2>
+	PX_INLINE T* construct(A1& a, A2& b)
+	{
+		T* t = reinterpret_cast<T*>(mPool.allocateMemory());
+		return t ? new (t) T(a,b) : 0;
+	}
+
+	template<class A1, class A2, class A3>
+	PX_INLINE T* construct(A1& a, A2& b, A3& c)
+	{
+		T* t = reinterpret_cast<T*>(mPool.allocateMemory());
+		return t ? new (t) T(a,b,c) : 0;
+	}
+
+	template<class A1, class A2, class A3, class A4>
+	PX_INLINE T* construct(A1& a, A2& b, A3& c, A4& d)
+	{
+		T* t = reinterpret_cast<T*>(mPool.allocateMemory());
+		return t ? new (t) T(a,b,c,d) : 0;
+	}
+
+	template<class A1, class A2, class A3, class A4, class A5>
+	PX_INLINE T* construct(A1& a, A2& b, A3& c, A4& d, A5& e)
+	{
+		T* t = reinterpret_cast<T*>(mPool.allocateMemory());
+		return t ? new (t) T(a,b,c,d,e) : 0;
+	}
+
+	////
+
+	PX_INLINE T* construct(T* t)
+	{
+		PX_ASSERT(t);
+		return new (t) T();
+	}
+
+	template<class A1>
+	PX_INLINE T* construct(T* t, A1& a)
+	{
+		PX_ASSERT(t);
+		return new (t) T(a);
+	}
+
+	template<class A1, class A2>
+	PX_INLINE T* construct(T* t, A1& a, A2& b)
+	{
+		PX_ASSERT(t);
+		return new (t) T(a,b);
+	}
+
+	template<class A1, class A2, class A3>
+	PX_INLINE T* construct(T* t, A1& a, A2& b, A3& c)
+	{
+		PX_ASSERT(t);
+		return new (t) T(a,b,c);
+	}
+
+	template<class A1, class A2, class A3, class A4>
+	PX_INLINE T* construct(T* t, A1& a, A2& b, A3& c, A4& d)
+	{
+		PX_ASSERT(t);
+		return new (t) T(a,b,c,d);
+	}
+
+	template<class A1, class A2, class A3, class A4, class A5>
+	PX_INLINE T* construct(T* t, A1& a, A2& b, A3& c, A4& d, A5& e)
+	{
+		PX_ASSERT(t);
+		return new (t) T(a,b,c,d,e);
+	}
+
+	PX_INLINE void destroy(T* const p)
+	{
+		if(p)
+		{
+			p->~T();
+			mPool.deallocateMemory(reinterpret_cast<PxU8*>(p));
+		}
+	}
+
+	PX_INLINE void releasePreallocated(T* const p)
+	{
+		if(p)
+			mPool.deallocateMemory(reinterpret_cast<PxU8*>(p));
+	}
+protected:
+	PreallocatingRegionManager	mPool;
+};
+
+template<class T>
+class BufferedPreallocatingPool : public PreallocatingPool<T>
+{
+	Ps::Array<T*> mDeletedElems;
+	PX_NOCOPY(BufferedPreallocatingPool<T>)
+public:
+	BufferedPreallocatingPool(PxU32 maxElements, const char* typeName) : PreallocatingPool<T>(maxElements, typeName)
+	{
+	}
+
+	PX_INLINE void destroy(T* const p)
+	{
+		if (p)
+		{
+			p->~T();
+			mDeletedElems.pushBack(p);
+		}
+	}
+
+	void processPendingDeletedElems()
+	{
+		for (PxU32 i = 0; i < mDeletedElems.size(); ++i)
+			this->mPool.deallocateMemory(reinterpret_cast<PxU8*>(mDeletedElems[i]));
+		mDeletedElems.clear();
+	}
+
+
+};
+
+
+
+	
+} // namespace Cm
+
+}
+
+#endif
diff --git a/PhysX_3.4/Source/Common/src/CmPriorityQueue.h b/PhysX_3.4/Source/Common/src/CmPriorityQueue.h
new file mode 100644
index 00000000..062d5958
--- /dev/null
+++ b/PhysX_3.4/Source/Common/src/CmPriorityQueue.h
@@ -0,0 +1,237 @@
+// This code contains NVIDIA Confidential Information and is disclosed to you
+// under a form of NVIDIA software license agreement provided separately to you.
+//
+// Notice
+// NVIDIA Corporation and its licensors retain all intellectual property and
+// proprietary rights in and to this software and related documentation and
+// any modifications thereto. Any use, reproduction, disclosure, or
+// distribution of this software and related documentation without an express
+// license agreement from NVIDIA Corporation is strictly prohibited.
+//
+// ALL NVIDIA DESIGN SPECIFICATIONS, CODE ARE PROVIDED "AS IS.". NVIDIA MAKES
+// NO WARRANTIES, EXPRESSED, IMPLIED, STATUTORY, OR OTHERWISE WITH RESPECT TO
+// THE MATERIALS, AND EXPRESSLY DISCLAIMS ALL IMPLIED WARRANTIES OF NONINFRINGEMENT,
+// MERCHANTABILITY, AND FITNESS FOR A PARTICULAR PURPOSE.
+//
+// Information and code furnished is believed to be accurate and reliable.
+// However, NVIDIA Corporation assumes no responsibility for the consequences of use of such
+// information or for any infringement of patents or other rights of third parties that may
+// result from its use. No license is granted by implication or otherwise under any patent
+// or patent rights of NVIDIA Corporation. Details are subject to change without notice.
+// This code supersedes and replaces all information previously supplied.
+// NVIDIA Corporation products are not authorized for use as critical
+// components in life support devices or systems without express written approval of
+// NVIDIA Corporation.
+//
+// Copyright (c) 2008-2016 NVIDIA Corporation. All rights reserved.
+// Copyright (c) 2004-2008 AGEIA Technologies, Inc. All rights reserved.
+// Copyright (c) 2001-2004 NovodeX AG. All rights reserved.  
+
+
+#ifndef PX_PHYSICS_COMMON_PRIORITYQUEUE
+#define PX_PHYSICS_COMMON_PRIORITYQUEUE
+
+#include "PsBasicTemplates.h"
+#include "CmPhysXCommon.h"
+#include "PsAllocator.h"
+#include "foundation/PxMemory.h"
+
+namespace physx
+{
+namespace Cm
+{
+	template<class Element, class Comparator = Ps::Less<Element> >
+	class PriorityQueueBase : protected Comparator // inherit so that stateless comparators take no space
+	{
+	public:
+		PriorityQueueBase(const Comparator& less, Element* elements) : Comparator(less), mHeapSize(0), mDataPtr(elements)
+		{
+		}
+		
+		~PriorityQueueBase() 
+		{
+		}
+		
+		//! Get the element with the highest priority
+		PX_FORCE_INLINE const Element top() const
+		{
+			return mDataPtr[0];
+		}
+
+		//! Get the element with the highest priority
+		PX_FORCE_INLINE Element top()
+		{
+			return mDataPtr[0];
+		}
+		
+		//! Check to whether the priority queue is empty
+		PX_FORCE_INLINE bool empty() const
+		{
+			return (mHeapSize == 0);
+		}
+		
+		//! Empty the priority queue
+		PX_FORCE_INLINE void clear()
+		{
+			mHeapSize = 0;
+		}  
+
+		//! Insert a new element into the priority queue. Only valid when size() is less than Capacity
+		PX_FORCE_INLINE void push(const Element& value)
+		{
+			PxU32 newIndex;
+			PxU32 parentIndex = parent(mHeapSize);
+
+			for (newIndex = mHeapSize; newIndex > 0 && compare(value, mDataPtr[parentIndex]); newIndex = parentIndex, parentIndex= parent(newIndex)) 
+			{
+				mDataPtr[ newIndex ] = mDataPtr[parentIndex];
+			}
+			mDataPtr[newIndex] = value; 
+			mHeapSize++;
+			PX_ASSERT(valid());
+		}
+
+		//! Delete the highest priority element. Only valid when non-empty.
+		PX_FORCE_INLINE Element pop()
+		{
+			PX_ASSERT(mHeapSize > 0);
+			PxU32 i, child;
+			//try to avoid LHS
+			PxU32 tempHs = mHeapSize-1;
+			mHeapSize = tempHs;
+			Element min = mDataPtr[0];
+			Element last = mDataPtr[tempHs];
+			
+			for (i = 0; (child = left(i)) < tempHs; i = child) 
+			{
+				/* Find highest priority child */
+				const PxU32 rightChild = child + 1;
+			
+				child += ((rightChild < tempHs) & compare((mDataPtr[rightChild]), (mDataPtr[child]))) ? 1 : 0;
+
+				if(compare(last, mDataPtr[child]))
+					break;
+
+				mDataPtr[i] = mDataPtr[child];
+			}
+			mDataPtr[ i ] = last;
+			
+			PX_ASSERT(valid());
+			return min;
+		} 
+
+		//! Make sure the priority queue sort all elements correctly
+		bool valid() const
+		{
+			const Element& min = mDataPtr[0];
+			for(PxU32 i=1; i<mHeapSize; ++i)
+			{
+				if(compare(mDataPtr[i], min))
+					return false;
+			}
+
+			return true;
+		}
+
+		//! Return number of elements in the priority queue
+		PxU32 size() const
+		{
+			return mHeapSize;
+		}
+
+	protected:
+
+		PxU32 mHeapSize;
+		Element* mDataPtr;
+		
+		PX_FORCE_INLINE bool compare(const Element& a, const Element& b) const
+		{
+			return Comparator::operator()(a,b);
+		}
+
+		static PX_FORCE_INLINE PxU32 left(PxU32 nodeIndex) 
+		{
+			return (nodeIndex << 1) + 1;
+		}
+		
+		static PX_FORCE_INLINE PxU32 parent(PxU32 nodeIndex) 
+		{
+			return (nodeIndex - 1) >> 1;
+		}
+	private:
+		PriorityQueueBase<Element, Comparator>& operator = (const PriorityQueueBase<Element, Comparator>);
+	};
+
+	template <typename Element, PxU32 Capacity, typename Comparator>
+	class InlinePriorityQueue : public PriorityQueueBase<Element, Comparator>
+	{
+		Element mData[Capacity];
+	public:
+		InlinePriorityQueue(const Comparator& less = Comparator()) : PriorityQueueBase<Element, Comparator>(less, mData)
+		{
+		}
+
+		PX_FORCE_INLINE void push(Element& elem)
+		{
+			PX_ASSERT(this->mHeapSize < Capacity);
+			PriorityQueueBase<Element, Comparator>::push(elem);
+		}
+	private:
+		InlinePriorityQueue<Element, Capacity, Comparator>& operator = (const InlinePriorityQueue<Element, Capacity, Comparator>);
+	};
+
+	template <typename Element, typename Comparator, typename Alloc = typename physx::shdfnd::AllocatorTraits<Element>::Type>
+	class PriorityQueue : public PriorityQueueBase<Element, Comparator>, protected Alloc
+	{
+		PxU32 mCapacity;
+	public:
+		PriorityQueue(const Comparator& less = Comparator(), PxU32 initialCapacity = 0, Alloc alloc = Alloc()) 
+			: PriorityQueueBase<Element, Comparator>(less, NULL), Alloc(alloc), mCapacity(initialCapacity)
+		{
+			if(initialCapacity > 0)
+				this->mDataPtr = reinterpret_cast<Element*>(Alloc::allocate(sizeof(Element)*initialCapacity, __FILE__, __LINE__));
+		}
+
+		~PriorityQueue()
+		{
+			if(this->mDataPtr)
+				this->deallocate(this->mDataPtr);
+		}
+
+		PX_FORCE_INLINE void push(Element& elem)
+		{
+			if(this->mHeapSize == mCapacity)
+			{
+				reserve((this->mHeapSize+1)*2);
+			}
+			PriorityQueueBase<Element, Comparator>::push(elem);
+		}
+
+		PX_FORCE_INLINE PxU32 capacity()
+		{
+			return mCapacity;
+		}
+
+		PX_FORCE_INLINE void reserve(const PxU32 newCapacity)
+		{
+			if(newCapacity > mCapacity)
+			{
+				Element* newElems = reinterpret_cast<Element*>(Alloc::allocate(sizeof(Element)*newCapacity, __FILE__, __LINE__));
+				if(this->mDataPtr)
+				{
+					physx::PxMemCopy(newElems, this->mDataPtr, sizeof(Element) * this->mHeapSize);
+					Alloc::deallocate(this->mDataPtr);
+				}
+				this->mDataPtr = newElems;
+				mCapacity = newCapacity;
+			}
+		}
+
+	private:
+		PriorityQueue<Element, Comparator, Alloc>& operator = (const PriorityQueue<Element, Comparator, Alloc>);
+	};
+	
+}
+}
+
+#endif
diff --git a/PhysX_3.4/Source/Common/src/CmPtrTable.cpp b/PhysX_3.4/Source/Common/src/CmPtrTable.cpp
new file mode 100644
index 00000000..9ca54e3a
--- /dev/null
+++ b/PhysX_3.4/Source/Common/src/CmPtrTable.cpp
@@ -0,0 +1,209 @@
+// This code contains NVIDIA Confidential Information and is disclosed to you
+// under a form of NVIDIA software license agreement provided separately to you.
+//
+// Notice
+// NVIDIA Corporation and its licensors retain all intellectual property and
+// proprietary rights in and to this software and related documentation and
+// any modifications thereto. Any use, reproduction, disclosure, or
+// distribution of this software and related documentation without an express
+// license agreement from NVIDIA Corporation is strictly prohibited.
+//
+// ALL NVIDIA DESIGN SPECIFICATIONS, CODE ARE PROVIDED "AS IS.". NVIDIA MAKES
+// NO WARRANTIES, EXPRESSED, IMPLIED, STATUTORY, OR OTHERWISE WITH RESPECT TO
+// THE MATERIALS, AND EXPRESSLY DISCLAIMS ALL IMPLIED WARRANTIES OF NONINFRINGEMENT,
+// MERCHANTABILITY, AND FITNESS FOR A PARTICULAR PURPOSE.
+//
+// Information and code furnished is believed to be accurate and reliable.
+// However, NVIDIA Corporation assumes no responsibility for the consequences of use of such
+// information or for any infringement of patents or other rights of third parties that may
+// result from its use. No license is granted by implication or otherwise under any patent
+// or patent rights of NVIDIA Corporation. Details are subject to change without notice.
+// This code supersedes and replaces all information previously supplied.
+// NVIDIA Corporation products are not authorized for use as critical
+// components in life support devices or systems without express written approval of
+// NVIDIA Corporation.
+//
+// Copyright (c) 2008-2016 NVIDIA Corporation. All rights reserved.
+// Copyright (c) 2004-2008 AGEIA Technologies, Inc. All rights reserved.
+// Copyright (c) 2001-2004 NovodeX AG. All rights reserved.  
+
+#include "foundation/PxAssert.h"
+#include "foundation/PxMemory.h"
+#include "CmPtrTable.h"
+#include "CmUtils.h"
+#include "PxMetaData.h"
+#include "PsBitUtils.h"
+
+using namespace physx;
+using namespace Cm;
+
+PtrTable::PtrTable()
+: mList(NULL)
+, mCount(0)
+, mOwnsMemory(true)
+, mBufferUsed(false)
+{
+}
+
+PtrTable::~PtrTable()
+{
+	PX_ASSERT(mOwnsMemory);
+	PX_ASSERT(mCount == 0);
+	PX_ASSERT(mList == NULL);
+}
+
+void PtrTable::clear(PtrTableStorageManager& sm)
+{
+	if(mOwnsMemory && mCount>1)
+	{
+		PxU32 implicitCapacity = Ps::nextPowerOfTwo(PxU32(mCount)-1);
+		sm.deallocate(mList, sizeof(void*)*implicitCapacity);
+	}
+
+	mList = NULL;
+	mOwnsMemory = true;
+	mCount = 0;
+}
+
+PxU32 PtrTable::find(const void* ptr) const
+{
+	const PxU32 nbPtrs = mCount;
+	void*const * PX_RESTRICT ptrs = getPtrs();
+
+	for(PxU32 i=0; i<nbPtrs; i++)
+	{
+		if(ptrs[i] == ptr)
+			return i;
+	}
+		
+	PX_ASSERT(0);
+	return 0xffffffff;
+}
+
+void PtrTable::exportExtraData(PxSerializationContext& stream)
+{
+	if(mCount>1)
+	{
+		stream.alignData(PX_SERIAL_ALIGN);
+		stream.writeData(mList, sizeof(void*)*mCount);
+	}
+}
+
+void PtrTable::importExtraData(PxDeserializationContext& context)
+{
+	if(mCount>1)
+		mList = context.readExtraData<void*, PX_SERIAL_ALIGN>(mCount);
+}
+
+void PtrTable::realloc(PxU32 oldCapacity, PxU32 newCapacity, PtrTableStorageManager& sm)
+{
+	PX_ASSERT((mOwnsMemory && oldCapacity) || (!mOwnsMemory && oldCapacity == 0));
+	PX_ASSERT(newCapacity);
+
+	if(mOwnsMemory && sm.canReuse(oldCapacity, newCapacity))
+		return;
+
+	void** newMem = sm.allocate(newCapacity * sizeof(void*));
+	PxMemCopy(newMem, mList, mCount * sizeof(void*));
+
+	if(mOwnsMemory)
+		sm.deallocate(mList, oldCapacity*sizeof(void*));
+
+	mList = newMem;
+	mOwnsMemory = true;
+}
+
+void PtrTable::add(void* ptr, PtrTableStorageManager& sm)
+{
+	if(mCount == 0)												// 0 -> 1, easy case
+	{
+		PX_ASSERT(mOwnsMemory);
+		PX_ASSERT(mList == NULL);
+		PX_ASSERT(!mBufferUsed);
+		mSingle = ptr;
+		mCount = 1;
+		mBufferUsed = true;
+		return;
+	}
+	
+	if(mCount == 1)												// 1 -> 2, easy case
+	{
+		PX_ASSERT(mOwnsMemory);
+		PX_ASSERT(mBufferUsed);
+
+		void* single = mSingle;
+		mList = sm.allocate(2*sizeof(void*));
+		mList[0] = single;
+		mBufferUsed = false;
+		mOwnsMemory = true;
+	} 
+	else 
+	{
+		PX_ASSERT(!mBufferUsed);
+
+		if(!mOwnsMemory)										// don't own the memory, must always alloc
+			realloc(0, Ps::nextPowerOfTwo(mCount), sm);			// we're guaranteed nextPowerOfTwo(x) > x
+
+		else if(Ps::isPowerOfTwo(mCount))						// count is at implicit capacity, so realloc
+			realloc(mCount, PxU32(mCount)*2, sm);				// ... to next higher power of 2
+
+		PX_ASSERT(mOwnsMemory);
+	}
+
+	mList[mCount++] = ptr;
+}
+
+void PtrTable::replaceWithLast(PxU32 index, PtrTableStorageManager& sm)
+{
+	PX_ASSERT(mCount!=0);
+
+	if(mCount == 1)												// 1 -> 0 easy case
+	{
+		PX_ASSERT(mOwnsMemory);
+		PX_ASSERT(mBufferUsed);
+
+		mList = NULL;
+		mCount = 0;
+		mBufferUsed = false;
+	}
+	else if(mCount == 2)										// 2 -> 1 easy case
+	{
+		PX_ASSERT(!mBufferUsed);
+		void* ptr = mList[1-index];
+		if(mOwnsMemory)
+			sm.deallocate(mList, 2*sizeof(void*));
+		mSingle = ptr;
+		mCount = 1;
+		mBufferUsed = true;
+		mOwnsMemory = true;
+	} 
+	else
+	{
+		PX_ASSERT(!mBufferUsed);
+
+		mList[index] = mList[--mCount];								// remove before adjusting memory
+
+		if(!mOwnsMemory)											// don't own the memory, must alloc
+			realloc(0, Ps::nextPowerOfTwo(PxU32(mCount)-1), sm);	// if currently a power of 2, don't jump to the next one
+
+		else if(Ps::isPowerOfTwo(mCount))							// own the memory, and implicit capacity requires that we downsize
+			realloc(PxU32(mCount)*2, PxU32(mCount), sm);			// ... from the next power of 2, which was the old implicit capacity
+
+		PX_ASSERT(mOwnsMemory);
+	}
+}
+
+void Cm::PtrTable::getBinaryMetaData(PxOutputStream& stream)
+{
+	PX_DEF_BIN_METADATA_CLASS(stream,	PtrTable)
+
+	PX_DEF_BIN_METADATA_ITEM(stream,	PtrTable, void,		mSingle,		PxMetaDataFlag::ePTR)		// PT: this is actually a union, beware
+	PX_DEF_BIN_METADATA_ITEM(stream,	PtrTable, PxU16,	mCount,			0)
+	PX_DEF_BIN_METADATA_ITEM(stream,	PtrTable, bool,		mOwnsMemory,	0)
+	PX_DEF_BIN_METADATA_ITEM(stream,	PtrTable, bool,		mBufferUsed,	0)
+
+	//------ Extra-data ------
+
+	// mList
+	PX_DEF_BIN_METADATA_EXTRA_ITEMS(stream, PtrTable, void, mBufferUsed, mCount, PxMetaDataFlag::eCONTROL_FLIP|PxMetaDataFlag::ePTR, PX_SERIAL_ALIGN)
+}
diff --git a/PhysX_3.4/Source/Common/src/CmPtrTable.h b/PhysX_3.4/Source/Common/src/CmPtrTable.h
new file mode 100644
index 00000000..c0eb3b61
--- /dev/null
+++ b/PhysX_3.4/Source/Common/src/CmPtrTable.h
@@ -0,0 +1,142 @@
+// This code contains NVIDIA Confidential Information and is disclosed to you
+// under a form of NVIDIA software license agreement provided separately to you.
+//
+// Notice
+// NVIDIA Corporation and its licensors retain all intellectual property and
+// proprietary rights in and to this software and related documentation and
+// any modifications thereto. Any use, reproduction, disclosure, or
+// distribution of this software and related documentation without an express
+// license agreement from NVIDIA Corporation is strictly prohibited.
+//
+// ALL NVIDIA DESIGN SPECIFICATIONS, CODE ARE PROVIDED "AS IS.". NVIDIA MAKES
+// NO WARRANTIES, EXPRESSED, IMPLIED, STATUTORY, OR OTHERWISE WITH RESPECT TO
+// THE MATERIALS, AND EXPRESSLY DISCLAIMS ALL IMPLIED WARRANTIES OF NONINFRINGEMENT,
+// MERCHANTABILITY, AND FITNESS FOR A PARTICULAR PURPOSE.
+//
+// Information and code furnished is believed to be accurate and reliable.
+// However, NVIDIA Corporation assumes no responsibility for the consequences of use of such
+// information or for any infringement of patents or other rights of third parties that may
+// result from its use. No license is granted by implication or otherwise under any patent
+// or patent rights of NVIDIA Corporation. Details are subject to change without notice.
+// This code supersedes and replaces all information previously supplied.
+// NVIDIA Corporation products are not authorized for use as critical
+// components in life support devices or systems without express written approval of
+// NVIDIA Corporation.
+//
+// Copyright (c) 2008-2016 NVIDIA Corporation. All rights reserved.
+// Copyright (c) 2004-2008 AGEIA Technologies, Inc. All rights reserved.
+// Copyright (c) 2001-2004 NovodeX AG. All rights reserved.  
+
+
+#ifndef PX_PHYSICS_COMMON_PTR_TABLE
+#define PX_PHYSICS_COMMON_PTR_TABLE
+
+#include "PxPhysXCommonConfig.h"
+#include "CmPhysXCommon.h"
+
+namespace physx
+{
+
+class PxSerializationContext;
+class PxDeserializationContext;
+
+
+namespace Cm
+{
+
+class PtrTableStorageManager
+{
+	// This will typically be backed by a MultiPool implementation with fallback to the user
+	// allocator. For MultiPool, when deallocating we want to know what the previously requested size was
+	// so we can release into the right pool
+
+public:
+
+	// capacity is in bytes
+
+	virtual void**	allocate(PxU32 capacity)												= 0;
+	virtual void	deallocate(void** addr, PxU32 originalCapacity)							= 0;
+
+	// whether memory allocated at one capacity can (and should) be safely reused at a different capacity
+	// allows realloc-style reuse by clients.
+
+	virtual bool	canReuse(PxU32 originalCapacity, PxU32 newCapacity)		= 0;
+protected:
+	virtual ~PtrTableStorageManager() {}
+};
+
+
+
+// specialized class to hold an array of pointers with extrinsic storage management, 
+// serialization-compatible with 3.3.1 PtrTable
+//
+// note that extrinsic storage implies you *must* clear the table before the destructor runs
+//
+// capacity is implicit: 
+// if the memory is not owned (i.e. came from deserialization) then the capacity is exactly mCount
+// else if mCount==0,  capacity is 0
+// else the capacity is the power of 2 >= mCount
+// 
+// one implication of this is that if we want to add or remove a pointer from unowned memory, we always realloc
+
+struct PX_PHYSX_COMMON_API PtrTable
+{
+//= ATTENTION! =====================================================================================
+// Changing the data layout of this class breaks the binary serialization format.  See comments for 
+// PX_BINARY_SERIAL_VERSION.  If a modification is required, please adjust the getBinaryMetaData 
+// function.  If the modification is made on a custom branch, please change PX_BINARY_SERIAL_VERSION
+// accordingly.
+//==================================================================================================
+
+	PtrTable();
+	~PtrTable();
+
+	void	add(void* ptr, PtrTableStorageManager& sm);
+	void	replaceWithLast(PxU32 index, PtrTableStorageManager& sm);
+	void	clear(PtrTableStorageManager& sm);
+
+	PxU32	find(const void* ptr) const;
+
+	PX_FORCE_INLINE PxU32		getCount()	const	{ return mCount; }
+	PX_FORCE_INLINE	void*const*	getPtrs()	const	{ return mCount == 1 ? &mSingle : mList;	}
+	PX_FORCE_INLINE	void**		getPtrs()			{ return mCount == 1 ? &mSingle : mList;	}
+
+
+	// SERIALIZATION
+
+	// 3.3.1 compatibility fixup: this implementation ALWAYS sets 'ownsMemory' if the size is 0 or 1
+	PtrTable(const PxEMPTY)
+	{
+		mOwnsMemory = mCount<2;
+		if(mCount == 0)
+			mList = NULL;
+	}
+
+	void	exportExtraData(PxSerializationContext& stream);
+	void	importExtraData(PxDeserializationContext& context);
+
+	static void getBinaryMetaData(physx::PxOutputStream& stream);
+
+private:
+	void realloc(PxU32 oldCapacity, PxU32 newCapacity, PtrTableStorageManager& sm);
+
+	union
+	{
+		void*	mSingle;
+		void**	mList;
+	};
+
+	PxU16	mCount;
+	bool	mOwnsMemory;
+	bool	mBufferUsed;		// dark magic in serialization requires this, otherwise redundant because it's logically equivalent to mCount == 1.
+
+};
+
+} // namespace Cm
+#if !PX_P64_FAMILY
+PX_COMPILE_TIME_ASSERT(sizeof(Cm::PtrTable)==8);
+#endif
+
+}
+
+#endif
diff --git a/PhysX_3.4/Source/Common/src/CmQueue.h b/PhysX_3.4/Source/Common/src/CmQueue.h
new file mode 100644
index 00000000..20f3191d
--- /dev/null
+++ b/PhysX_3.4/Source/Common/src/CmQueue.h
@@ -0,0 +1,152 @@
+// This code contains NVIDIA Confidential Information and is disclosed to you
+// under a form of NVIDIA software license agreement provided separately to you.
+//
+// Notice
+// NVIDIA Corporation and its licensors retain all intellectual property and
+// proprietary rights in and to this software and related documentation and
+// any modifications thereto. Any use, reproduction, disclosure, or
+// distribution of this software and related documentation without an express
+// license agreement from NVIDIA Corporation is strictly prohibited.
+//
+// ALL NVIDIA DESIGN SPECIFICATIONS, CODE ARE PROVIDED "AS IS.". NVIDIA MAKES
+// NO WARRANTIES, EXPRESSED, IMPLIED, STATUTORY, OR OTHERWISE WITH RESPECT TO
+// THE MATERIALS, AND EXPRESSLY DISCLAIMS ALL IMPLIED WARRANTIES OF NONINFRINGEMENT,
+// MERCHANTABILITY, AND FITNESS FOR A PARTICULAR PURPOSE.
+//
+// Information and code furnished is believed to be accurate and reliable.
+// However, NVIDIA Corporation assumes no responsibility for the consequences of use of such
+// information or for any infringement of patents or other rights of third parties that may
+// result from its use. No license is granted by implication or otherwise under any patent
+// or patent rights of NVIDIA Corporation. Details are subject to change without notice.
+// This code supersedes and replaces all information previously supplied.
+// NVIDIA Corporation products are not authorized for use as critical
+// components in life support devices or systems without express written approval of
+// NVIDIA Corporation.
+//
+// Copyright (c) 2008-2016 NVIDIA Corporation. All rights reserved.
+// Copyright (c) 2004-2008 AGEIA Technologies, Inc. All rights reserved.
+// Copyright (c) 2001-2004 NovodeX AG. All rights reserved.  
+
+
+#ifndef PX_PHYSICS_COMMON_QUEUE
+#define PX_PHYSICS_COMMON_QUEUE
+
+#include "foundation/PxAssert.h"
+#include "PsAllocator.h"
+#include "PsUserAllocated.h"
+#include "CmPhysXCommon.h"
+
+namespace physx
+{
+namespace Cm
+{
+
+	template<class T, class AllocType = Ps::NonTrackingAllocator >
+	class Queue: public Ps::UserAllocated
+	{
+	public:
+		Queue(PxU32 maxEntries);
+		~Queue();
+
+		T 		popFront();
+		T 		front();
+		T 		popBack();
+		T 		back();
+		bool	pushBack(const T& element);
+		bool	empty() const;
+		PxU32	size() const;
+
+	private:
+		T*			mJobQueue;
+		PxU32		mNum;
+		PxU32		mHead;
+		PxU32		mTail;
+		PxU32		mMaxEntries;
+		AllocType	mAllocator;
+	};
+
+	template<class T, class AllocType>
+	Queue<T, AllocType>::Queue(PxU32 maxEntries):
+		mNum(0),
+		mHead(0),
+		mTail(0),
+		mMaxEntries(maxEntries)
+	{
+		mJobQueue = reinterpret_cast<T*>(mAllocator.allocate(sizeof(T)*mMaxEntries, __FILE__, __LINE__));
+	}
+
+	template<class T, class AllocType>
+	Queue<T, AllocType>::~Queue()
+	{
+		if(mJobQueue)
+			mAllocator.deallocate(mJobQueue);
+	}
+
+	template<class T, class AllocType>
+	T Queue<T, AllocType>::popFront()
+	{
+		PX_ASSERT(mNum>0);
+
+		mNum--;
+		T& element = mJobQueue[mTail];
+		mTail = (mTail+1) % (mMaxEntries);
+		return element;
+	}
+
+	template<class T, class AllocType>
+	T Queue<T, AllocType>::front()
+	{
+		PX_ASSERT(mNum>0);
+
+		return mJobQueue[mTail];
+	}
+
+	template<class T, class AllocType>
+	T Queue<T, AllocType>::popBack()
+	{
+		PX_ASSERT(mNum>0);
+
+		mNum--;
+		mHead = (mHead-1) % (mMaxEntries);
+		return mJobQueue[mHead];
+	}
+
+	template<class T, class AllocType>
+	T Queue<T, AllocType>::back()
+	{
+		PX_ASSERT(mNum>0);
+
+		PxU32 headAccess = (mHead-1) % (mMaxEntries);
+		return mJobQueue[headAccess];
+	}
+
+	template<class T, class AllocType>
+	bool Queue<T, AllocType>::pushBack(const T& element)
+	{
+		if (mNum == mMaxEntries) return false;
+		mJobQueue[mHead] = element;
+
+		mNum++;
+		mHead = (mHead+1) % (mMaxEntries);
+
+		return true;
+	}
+
+	template<class T, class AllocType>
+	bool Queue<T, AllocType>::empty() const
+	{
+		return mNum == 0;
+	}
+
+	template<class T, class AllocType>
+	PxU32 Queue<T, AllocType>::size() const
+	{
+		return mNum;
+	}
+
+
+} // namespace Cm
+
+}
+
+#endif
diff --git a/PhysX_3.4/Source/Common/src/CmRadixSort.cpp b/PhysX_3.4/Source/Common/src/CmRadixSort.cpp
new file mode 100644
index 00000000..e5417849
--- /dev/null
+++ b/PhysX_3.4/Source/Common/src/CmRadixSort.cpp
@@ -0,0 +1,460 @@
+// This code contains NVIDIA Confidential Information and is disclosed to you
+// under a form of NVIDIA software license agreement provided separately to you.
+//
+// Notice
+// NVIDIA Corporation and its licensors retain all intellectual property and
+// proprietary rights in and to this software and related documentation and
+// any modifications thereto. Any use, reproduction, disclosure, or
+// distribution of this software and related documentation without an express
+// license agreement from NVIDIA Corporation is strictly prohibited.
+//
+// ALL NVIDIA DESIGN SPECIFICATIONS, CODE ARE PROVIDED "AS IS.". NVIDIA MAKES
+// NO WARRANTIES, EXPRESSED, IMPLIED, STATUTORY, OR OTHERWISE WITH RESPECT TO
+// THE MATERIALS, AND EXPRESSLY DISCLAIMS ALL IMPLIED WARRANTIES OF NONINFRINGEMENT,
+// MERCHANTABILITY, AND FITNESS FOR A PARTICULAR PURPOSE.
+//
+// Information and code furnished is believed to be accurate and reliable.
+// However, NVIDIA Corporation assumes no responsibility for the consequences of use of such
+// information or for any infringement of patents or other rights of third parties that may
+// result from its use. No license is granted by implication or otherwise under any patent
+// or patent rights of NVIDIA Corporation. Details are subject to change without notice.
+// This code supersedes and replaces all information previously supplied.
+// NVIDIA Corporation products are not authorized for use as critical
+// components in life support devices or systems without express written approval of
+// NVIDIA Corporation.
+//
+// Copyright (c) 2008-2016 NVIDIA Corporation. All rights reserved.
+// Copyright (c) 2004-2008 AGEIA Technologies, Inc. All rights reserved.
+// Copyright (c) 2001-2004 NovodeX AG. All rights reserved.  
+
+///////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+#include "foundation/PxMemory.h"
+#include "foundation/PxAssert.h"
+#include "CmRadixSort.h"
+
+using namespace physx;
+using namespace Cm;
+
+#if defined(__BIG_ENDIAN__) || defined(_XBOX)
+	#define H0_OFFSET	768
+	#define H1_OFFSET	512
+	#define H2_OFFSET	256
+	#define H3_OFFSET	0
+	#define BYTES_INC	(3-j)
+#else 
+	#define H0_OFFSET	0
+	#define H1_OFFSET	256
+	#define H2_OFFSET	512
+	#define H3_OFFSET	768
+	#define BYTES_INC	j
+#endif
+
+#define CREATE_HISTOGRAMS(type, buffer)														\
+	/* Clear counters/histograms */															\
+	PxMemZero(mHistogram1024, 256*4*sizeof(PxU32));											\
+																							\
+	/* Prepare to count */																	\
+	const PxU8* PX_RESTRICT p = reinterpret_cast<const PxU8*>(input);						\
+	const PxU8* PX_RESTRICT pe = &p[nb*4];													\
+	PxU32* PX_RESTRICT h0= &mHistogram1024[H0_OFFSET];	/* Histogram for first pass (LSB)*/	\
+	PxU32* PX_RESTRICT h1= &mHistogram1024[H1_OFFSET];	/* Histogram for second pass	*/	\
+	PxU32* PX_RESTRICT h2= &mHistogram1024[H2_OFFSET];	/* Histogram for third pass		*/	\
+	PxU32* PX_RESTRICT h3= &mHistogram1024[H3_OFFSET];	/* Histogram for last pass (MSB)*/	\
+																							\
+	bool AlreadySorted = true;	/* Optimism... */											\
+																							\
+	if(INVALID_RANKS)																		\
+	{																						\
+		/* Prepare for temporal coherence */												\
+		const type* PX_RESTRICT Running = reinterpret_cast<const type*>(buffer);			\
+		type PrevVal = *Running;															\
+																							\
+		while(p!=pe)																		\
+		{																					\
+			/* Read input buffer in previous sorted order */								\
+			const type Val = *Running++;													\
+			/* Check whether already sorted or not */										\
+			if(Val<PrevVal)	{ AlreadySorted = false; break; } /* Early out */				\
+			/* Update for next iteration */													\
+			PrevVal = Val;																	\
+																							\
+			/* Create histograms */															\
+			h0[*p++]++;	h1[*p++]++;	h2[*p++]++;	h3[*p++]++;									\
+		}																					\
+																							\
+		/* If all input values are already sorted, we just have to return and leave the */	\
+		/* previous list unchanged. That way the routine may take advantage of temporal */	\
+		/* coherence, for example when used to sort transparent faces.					*/	\
+		if(AlreadySorted)																	\
+		{																					\
+			mNbHits++;																		\
+			for(PxU32 i=0;i<nb;i++)	mRanks[i] = i;											\
+			return *this;																	\
+		}																					\
+	}																						\
+	else																					\
+	{																						\
+		/* Prepare for temporal coherence */												\
+		const PxU32* PX_RESTRICT Indices = mRanks;											\
+		type PrevVal = type(buffer[*Indices]);												\
+																							\
+		while(p!=pe)																		\
+		{																					\
+			/* Read input buffer in previous sorted order */								\
+			const type Val = type(buffer[*Indices++]);										\
+			/* Check whether already sorted or not */										\
+			if(Val<PrevVal)	{ AlreadySorted = false; break; } /* Early out */				\
+			/* Update for next iteration */													\
+			PrevVal = Val;																	\
+																							\
+			/* Create histograms */															\
+			h0[*p++]++;	h1[*p++]++;	h2[*p++]++;	h3[*p++]++;									\
+		}																					\
+																							\
+		/* If all input values are already sorted, we just have to return and leave the */	\
+		/* previous list unchanged. That way the routine may take advantage of temporal */	\
+		/* coherence, for example when used to sort transparent faces.					*/	\
+		if(AlreadySorted)	{ mNbHits++; return *this;	}									\
+	}																						\
+																							\
+	/* Else there has been an early out and we must finish computing the histograms */		\
+	while(p!=pe)																			\
+	{																						\
+		/* Create histograms without the previous overhead */								\
+		h0[*p++]++;	h1[*p++]++;	h2[*p++]++;	h3[*p++]++;										\
+	}
+
+PX_INLINE const PxU32* CheckPassValidity(PxU32 pass, const PxU32* mHistogram1024, PxU32 nb, const void* input, PxU8& UniqueVal)
+{
+	// Shortcut to current counters
+	const PxU32* CurCount = &mHistogram1024[pass<<8];
+
+	// Check pass validity
+
+	// If all values have the same byte, sorting is useless.
+	// It may happen when sorting bytes or words instead of dwords.
+	// This routine actually sorts words faster than dwords, and bytes
+	// faster than words. Standard running time (O(4*n))is reduced to O(2*n)
+	// for words and O(n) for bytes. Running time for floats depends on actual values...
+
+	// Get first byte
+	UniqueVal = *((reinterpret_cast<const PxU8*>(input))+pass);
+
+	// Check that byte's counter
+	if(CurCount[UniqueVal]==nb)
+		return NULL;
+
+	return CurCount;
+}
+
+///////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+/**
+ *	Constructor.
+ */
+///////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+
+RadixSort::RadixSort() : mCurrentSize(0), mRanks(NULL), mRanks2(NULL), mHistogram1024(0), mLinks256(0), mTotalCalls(0), mNbHits(0), mDeleteRanks(true)
+{
+	// Initialize indices
+	INVALIDATE_RANKS;
+}
+
+///////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+/**
+ *	Destructor.
+ */
+///////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+
+RadixSort::~RadixSort()
+{
+}
+
+
+///////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+/**
+ *	Main sort routine.
+ *	This one is for integer values. After the call, mRanks contains a list of indices in sorted order, i.e. in the order you may process your data.
+ *	\param		input	[in] a list of integer values to sort
+ *	\param		nb		[in] number of values to sort, must be < 2^31
+ *	\param		hint	[in] RADIX_SIGNED to handle negative values, RADIX_UNSIGNED if you know your input buffer only contains positive values
+ *	\return		Self-Reference
+ */
+///////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+
+RadixSort& RadixSort::Sort(const PxU32* input, PxU32 nb, RadixHint hint)
+{
+	PX_ASSERT(mHistogram1024);
+	PX_ASSERT(mLinks256);
+	PX_ASSERT(mRanks);
+	PX_ASSERT(mRanks2);
+
+	// Checkings
+	if(!input || !nb || nb&0x80000000)	return *this;
+
+	// Stats
+	mTotalCalls++;
+
+	// Create histograms (counters). Counters for all passes are created in one run.
+	// Pros:	read input buffer once instead of four times
+	// Cons:	mHistogram1024 is 4Kb instead of 1Kb
+	// We must take care of signed/unsigned values for temporal coherence.... I just
+	// have 2 code paths even if just a single opcode changes. Self-modifying code, someone?
+	if(hint==RADIX_UNSIGNED)	{ CREATE_HISTOGRAMS(PxU32, input);	}
+	else						{ CREATE_HISTOGRAMS(PxI32, input);	}
+
+	// Compute #negative values involved if needed
+	PxU32 NbNegativeValues = 0;
+	if(hint==RADIX_SIGNED)
+	{
+		// An efficient way to compute the number of negatives values we'll have to deal with is simply to sum the 128
+		// last values of the last histogram. Last histogram because that's the one for the Most Significant Byte,
+		// responsible for the sign. 128 last values because the 128 first ones are related to positive numbers.
+		PxU32* PX_RESTRICT h3= &mHistogram1024[768];
+		for(PxU32 i=128;i<256;i++)	NbNegativeValues += h3[i];	// 768 for last histogram, 128 for negative part
+	}
+
+	// Radix sort, j is the pass number (0=LSB, 3=MSB)
+	for(PxU32 j=0;j<4;j++)
+	{
+//		CHECK_PASS_VALIDITY(j);
+		PxU8 UniqueVal;
+		const PxU32* PX_RESTRICT CurCount = CheckPassValidity(j, mHistogram1024, nb, input, UniqueVal);
+
+		// Sometimes the fourth (negative) pass is skipped because all numbers are negative and the MSB is 0xFF (for example). This is
+		// not a problem, numbers are correctly sorted anyway.
+		if(CurCount)
+		{
+			PxU32** PX_RESTRICT Links256 = mLinks256;
+
+			// Should we care about negative values?
+			if(j!=3 || hint==RADIX_UNSIGNED)
+			{
+				// Here we deal with positive values only
+
+				// Create offsets
+				Links256[0] = mRanks2;
+				for(PxU32 i=1;i<256;i++)
+					Links256[i] = Links256[i-1] + CurCount[i-1];
+			}
+			else
+			{
+				// This is a special case to correctly handle negative integers. They're sorted in the right order but at the wrong place.
+
+				// Create biased offsets, in order for negative numbers to be sorted as well
+				Links256[0] = &mRanks2[NbNegativeValues];										// First positive number takes place after the negative ones
+				for(PxU32 i=1;i<128;i++)
+					Links256[i] = Links256[i-1] + CurCount[i-1];		// 1 to 128 for positive numbers
+
+				// Fixing the wrong place for negative values
+				Links256[128] = mRanks2;
+				for(PxU32 i=129;i<256;i++)
+					Links256[i] = Links256[i-1] + CurCount[i-1];
+			}
+
+			// Perform Radix Sort
+			const PxU8* PX_RESTRICT InputBytes = reinterpret_cast<const PxU8*>(input);
+            InputBytes += BYTES_INC;
+			if(INVALID_RANKS)
+			{
+				for(PxU32 i=0;i<nb;i++)
+					*Links256[InputBytes[i<<2]]++ = i;
+				VALIDATE_RANKS;
+			}
+			else
+			{
+				PxU32* PX_RESTRICT Indices		= mRanks;
+				PxU32* PX_RESTRICT IndicesEnd	= &mRanks[nb];
+				while(Indices!=IndicesEnd)
+				{
+					const PxU32 id = *Indices++;
+					*Links256[InputBytes[id<<2]]++ = id;
+				}
+			}
+
+			// Swap pointers for next pass. Valid indices - the most recent ones - are in mRanks after the swap.
+			PxU32* Tmp	= mRanks;	mRanks = mRanks2; mRanks2 = Tmp;
+		}
+	}
+	return *this;
+}
+
+///////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+/**
+ *	Main sort routine.
+ *	This one is for floating-point values. After the call, mRanks contains a list of indices in sorted order, i.e. in the order you may process your data.
+ *	\param		input2			[in] a list of floating-point values to sort
+ *	\param		nb				[in] number of values to sort, must be < 2^31
+ *	\return		Self-Reference
+ *	\warning	only sorts IEEE floating-point values
+ */
+///////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+
+RadixSort& RadixSort::Sort(const float* input2, PxU32 nb)
+{
+	PX_ASSERT(mHistogram1024);
+	PX_ASSERT(mLinks256);
+	PX_ASSERT(mRanks);
+	PX_ASSERT(mRanks2);
+
+	// Checkings
+	if(!input2 || !nb || nb&0x80000000)	return *this;
+
+	// Stats
+	mTotalCalls++;
+
+	const PxU32* PX_RESTRICT input = reinterpret_cast<const PxU32*>(input2);
+
+	// Allocate histograms & offsets on the stack
+	//PxU32 mHistogram1024[256*4];
+	//PxU32* mLinks256[256];
+
+	// Create histograms (counters). Counters for all passes are created in one run.
+	// Pros:	read input buffer once instead of four times
+	// Cons:	mHistogram1024 is 4Kb instead of 1Kb
+	// Floating-point values are always supposed to be signed values, so there's only one code path there.
+	// Please note the floating point comparison needed for temporal coherence! Although the resulting asm code
+	// is dreadful, this is surprisingly not such a performance hit - well, I suppose that's a big one on first
+	// generation Pentiums....We can't make comparison on integer representations because, as Chris said, it just
+	// wouldn't work with mixed positive/negative values....
+	{ CREATE_HISTOGRAMS(float, input2); }
+
+	// Compute #negative values involved if needed
+	PxU32 NbNegativeValues = 0;
+	// An efficient way to compute the number of negatives values we'll have to deal with is simply to sum the 128
+	// last values of the last histogram. Last histogram because that's the one for the Most Significant Byte,
+	// responsible for the sign. 128 last values because the 128 first ones are related to positive numbers.
+	// ### is that ok on Apple ?!
+	PxU32* PX_RESTRICT h3= &mHistogram1024[768];
+	for(PxU32 i=128;i<256;i++)	NbNegativeValues += h3[i];	// 768 for last histogram, 128 for negative part
+
+	// Radix sort, j is the pass number (0=LSB, 3=MSB)
+	for(PxU32 j=0;j<4;j++)
+	{
+		PxU8 UniqueVal;
+		const PxU32* PX_RESTRICT CurCount = CheckPassValidity(j, mHistogram1024, nb, input, UniqueVal);
+		// Should we care about negative values?
+		if(j!=3)
+		{
+			// Here we deal with positive values only
+//			CHECK_PASS_VALIDITY(j);
+//			const bool PerformPass = CheckPassValidity(j, mHistogram1024, nb, input);
+
+			if(CurCount)
+			{
+				PxU32** PX_RESTRICT Links256 = mLinks256;
+
+				// Create offsets
+				Links256[0] = mRanks2;
+				for(PxU32 i=1;i<256;i++)
+					Links256[i] = Links256[i-1] + CurCount[i-1];
+
+				// Perform Radix Sort
+				const PxU8* PX_RESTRICT InputBytes = reinterpret_cast<const PxU8*>(input);
+                InputBytes += BYTES_INC;
+				if(INVALID_RANKS)
+				{
+					for(PxU32 i=0;i<nb;i++)
+						*Links256[InputBytes[i<<2]]++ = i;
+					VALIDATE_RANKS;
+				}
+				else
+				{
+					PxU32* PX_RESTRICT Indices		= mRanks;
+					PxU32* PX_RESTRICT IndicesEnd	= &mRanks[nb];
+					while(Indices!=IndicesEnd)
+					{
+						const PxU32 id = *Indices++;
+						*Links256[InputBytes[id<<2]]++ = id;
+					}
+				}
+
+				// Swap pointers for next pass. Valid indices - the most recent ones - are in mRanks after the swap.
+				PxU32* Tmp	= mRanks;	mRanks = mRanks2; mRanks2 = Tmp;
+			}
+		}
+		else
+		{
+			// This is a special case to correctly handle negative values
+//			CHECK_PASS_VALIDITY(j);
+//			const bool PerformPass = CheckPassValidity(j, mHistogram1024, nb, input);
+
+			if(CurCount)
+			{
+				PxU32** PX_RESTRICT Links256 = mLinks256;
+
+				// Create biased offsets, in order for negative numbers to be sorted as well
+				Links256[0] = &mRanks2[NbNegativeValues];										// First positive number takes place after the negative ones
+				for(PxU32 i=1;i<128;i++)
+					Links256[i] = Links256[i-1] + CurCount[i-1];		// 1 to 128 for positive numbers
+
+				// We must reverse the sorting order for negative numbers!
+				Links256[255] = mRanks2;
+				for(PxU32 i=0;i<127;i++)
+					Links256[254-i] = Links256[255-i] + CurCount[255-i];		// Fixing the wrong order for negative values
+				for(PxU32 i=128;i<256;i++)
+					Links256[i] += CurCount[i];							// Fixing the wrong place for negative values
+
+				// Perform Radix Sort
+				if(INVALID_RANKS)
+				{
+					for(PxU32 i=0;i<nb;i++)
+					{
+						const PxU32 Radix = input[i]>>24;				// Radix byte, same as above. AND is useless here (PxU32).
+						// ### cmp to be killed. Not good. Later.
+						if(Radix<128)	*Links256[Radix]++ = i;			// Number is positive, same as above
+						else			*(--Links256[Radix]) = i;		// Number is negative, flip the sorting order
+					}
+					VALIDATE_RANKS;
+				}
+				else
+				{
+					const PxU32* PX_RESTRICT Ranks = mRanks;
+					for(PxU32 i=0;i<nb;i++)
+					{
+						const PxU32 Radix = input[Ranks[i]]>>24;				// Radix byte, same as above. AND is useless here (PxU32).
+                        // ### cmp to be killed. Not good. Later.
+						if(Radix<128)	*Links256[Radix]++ = Ranks[i];			// Number is positive, same as above
+						else			*(--Links256[Radix]) = Ranks[i];		// Number is negative, flip the sorting order
+					}
+				}
+				// Swap pointers for next pass. Valid indices - the most recent ones - are in mRanks after the swap.
+				PxU32* Tmp	= mRanks;	mRanks = mRanks2; mRanks2 = Tmp;
+			}
+			else
+			{
+				// The pass is useless, yet we still have to reverse the order of current list if all values are negative.
+				if(UniqueVal>=128)
+				{
+					if(INVALID_RANKS)
+					{
+						// ###Possible?
+						for(PxU32 i=0;i<nb;i++)	mRanks2[i] = nb-i-1;
+						VALIDATE_RANKS;
+					}
+					else
+					{
+						for(PxU32 i=0;i<nb;i++)	mRanks2[i] = mRanks[nb-i-1];
+					}
+
+					// Swap pointers for next pass. Valid indices - the most recent ones - are in mRanks after the swap.
+					PxU32* Tmp	= mRanks;	mRanks = mRanks2; mRanks2 = Tmp;
+				}
+			}
+		}
+	}
+	return *this;
+}
+
+bool RadixSort::SetBuffers(PxU32* ranks0, PxU32* ranks1, PxU32* histogram1024, PxU32** links256)
+{
+	if(!ranks0 || !ranks1 || !histogram1024 || !links256)	return false;
+
+	mRanks			= ranks0;
+	mRanks2			= ranks1;
+	mHistogram1024	= histogram1024;
+	mLinks256		= links256;
+	mDeleteRanks	= false;
+	INVALIDATE_RANKS;
+	return true;
+}
+
diff --git a/PhysX_3.4/Source/Common/src/CmRadixSort.h b/PhysX_3.4/Source/Common/src/CmRadixSort.h
new file mode 100644
index 00000000..4bde7f7a
--- /dev/null
+++ b/PhysX_3.4/Source/Common/src/CmRadixSort.h
@@ -0,0 +1,100 @@
+// This code contains NVIDIA Confidential Information and is disclosed to you
+// under a form of NVIDIA software license agreement provided separately to you.
+//
+// Notice
+// NVIDIA Corporation and its licensors retain all intellectual property and
+// proprietary rights in and to this software and related documentation and
+// any modifications thereto. Any use, reproduction, disclosure, or
+// distribution of this software and related documentation without an express
+// license agreement from NVIDIA Corporation is strictly prohibited.
+//
+// ALL NVIDIA DESIGN SPECIFICATIONS, CODE ARE PROVIDED "AS IS.". NVIDIA MAKES
+// NO WARRANTIES, EXPRESSED, IMPLIED, STATUTORY, OR OTHERWISE WITH RESPECT TO
+// THE MATERIALS, AND EXPRESSLY DISCLAIMS ALL IMPLIED WARRANTIES OF NONINFRINGEMENT,
+// MERCHANTABILITY, AND FITNESS FOR A PARTICULAR PURPOSE.
+//
+// Information and code furnished is believed to be accurate and reliable.
+// However, NVIDIA Corporation assumes no responsibility for the consequences of use of such
+// information or for any infringement of patents or other rights of third parties that may
+// result from its use. No license is granted by implication or otherwise under any patent
+// or patent rights of NVIDIA Corporation. Details are subject to change without notice.
+// This code supersedes and replaces all information previously supplied.
+// NVIDIA Corporation products are not authorized for use as critical
+// components in life support devices or systems without express written approval of
+// NVIDIA Corporation.
+//
+// Copyright (c) 2008-2016 NVIDIA Corporation. All rights reserved.
+// Copyright (c) 2004-2008 AGEIA Technologies, Inc. All rights reserved.
+// Copyright (c) 2001-2004 NovodeX AG. All rights reserved.  
+
+#ifndef CM_RADIX_SORT_H
+#define CM_RADIX_SORT_H
+
+#include "PxPhysXCommonConfig.h"
+
+namespace physx
+{
+namespace Cm
+{
+
+	enum RadixHint
+	{
+		RADIX_SIGNED,		//!< Input values are signed
+		RADIX_UNSIGNED,		//!< Input values are unsigned
+
+		RADIX_FORCE_DWORD = 0x7fffffff
+	};
+
+#define INVALIDATE_RANKS	mCurrentSize|=0x80000000
+#define VALIDATE_RANKS		mCurrentSize&=0x7fffffff
+#define CURRENT_SIZE		(mCurrentSize&0x7fffffff)
+#define INVALID_RANKS		(mCurrentSize&0x80000000)
+
+	class PX_PHYSX_COMMON_API RadixSort
+	{
+		public:
+										RadixSort();
+		virtual							~RadixSort();
+		// Sorting methods
+						RadixSort&		Sort(const PxU32* input, PxU32 nb, RadixHint hint=RADIX_SIGNED);
+						RadixSort&		Sort(const float* input, PxU32 nb);
+
+		//! Access to results. mRanks is a list of indices in sorted order, i.e. in the order you may further process your data
+		PX_FORCE_INLINE	const PxU32*	GetRanks()			const	{ return mRanks;		}
+
+		//! mIndices2 gets trashed on calling the sort routine, but otherwise you can recycle it the way you want.
+		PX_FORCE_INLINE	PxU32*			GetRecyclable()		const	{ return mRanks2;		}
+
+		//! Returns the total number of calls to the radix sorter.
+		PX_FORCE_INLINE	PxU32			GetNbTotalCalls()	const	{ return mTotalCalls;	}
+		//! Returns the number of eraly exits due to temporal coherence.
+		PX_FORCE_INLINE	PxU32			GetNbHits()			const	{ return mNbHits;		}
+
+						bool			SetBuffers(PxU32* ranks0, PxU32* ranks1, PxU32* histogram1024, PxU32** links256);
+		private:
+										RadixSort(const RadixSort& object);
+										RadixSort& operator=(const RadixSort& object);
+		protected:
+						PxU32			mCurrentSize;		//!< Current size of the indices list
+						PxU32*			mRanks;				//!< Two lists, swapped each pass
+						PxU32*			mRanks2;
+						PxU32*			mHistogram1024;
+						PxU32**			mLinks256;
+		// Stats
+						PxU32			mTotalCalls;		//!< Total number of calls to the sort routine
+						PxU32			mNbHits;			//!< Number of early exits due to coherence
+
+						// Stack-radix
+						bool			mDeleteRanks;		//!<
+	};
+
+	#define StackRadixSort(name, ranks0, ranks1)	\
+		RadixSort name;								\
+		PxU32 histogramBuffer[1024];				\
+		PxU32* linksBuffer[256];					\
+		name.SetBuffers(ranks0, ranks1, histogramBuffer, linksBuffer);
+}
+
+}
+
+#endif // CM_RADIX_SORT_H
diff --git a/PhysX_3.4/Source/Common/src/CmRadixSortBuffered.cpp b/PhysX_3.4/Source/Common/src/CmRadixSortBuffered.cpp
new file mode 100644
index 00000000..9efd9fb7
--- /dev/null
+++ b/PhysX_3.4/Source/Common/src/CmRadixSortBuffered.cpp
@@ -0,0 +1,142 @@
+// This code contains NVIDIA Confidential Information and is disclosed to you
+// under a form of NVIDIA software license agreement provided separately to you.
+//
+// Notice
+// NVIDIA Corporation and its licensors retain all intellectual property and
+// proprietary rights in and to this software and related documentation and
+// any modifications thereto. Any use, reproduction, disclosure, or
+// distribution of this software and related documentation without an express
+// license agreement from NVIDIA Corporation is strictly prohibited.
+//
+// ALL NVIDIA DESIGN SPECIFICATIONS, CODE ARE PROVIDED "AS IS.". NVIDIA MAKES
+// NO WARRANTIES, EXPRESSED, IMPLIED, STATUTORY, OR OTHERWISE WITH RESPECT TO
+// THE MATERIALS, AND EXPRESSLY DISCLAIMS ALL IMPLIED WARRANTIES OF NONINFRINGEMENT,
+// MERCHANTABILITY, AND FITNESS FOR A PARTICULAR PURPOSE.
+//
+// Information and code furnished is believed to be accurate and reliable.
+// However, NVIDIA Corporation assumes no responsibility for the consequences of use of such
+// information or for any infringement of patents or other rights of third parties that may
+// result from its use. No license is granted by implication or otherwise under any patent
+// or patent rights of NVIDIA Corporation. Details are subject to change without notice.
+// This code supersedes and replaces all information previously supplied.
+// NVIDIA Corporation products are not authorized for use as critical
+// components in life support devices or systems without express written approval of
+// NVIDIA Corporation.
+//
+// Copyright (c) 2008-2016 NVIDIA Corporation. All rights reserved.
+// Copyright (c) 2004-2008 AGEIA Technologies, Inc. All rights reserved.
+// Copyright (c) 2001-2004 NovodeX AG. All rights reserved.  
+
+
+///////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+#include "CmRadixSortBuffered.h"
+#include "PsAllocator.h"
+
+using namespace physx;
+using namespace Cm;
+
+RadixSortBuffered::RadixSortBuffered()
+: RadixSort()
+{
+}
+
+RadixSortBuffered::~RadixSortBuffered()
+{
+	// Release everything
+	if(mDeleteRanks)
+	{
+		PX_FREE_AND_RESET(mRanks2);
+		PX_FREE_AND_RESET(mRanks);
+	}
+}
+
+///////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+/**
+ *	Resizes the inner lists.
+ *	\param		nb	[in] new size (number of dwords)
+ *	\return		true if success
+ */
+///////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+bool RadixSortBuffered::Resize(PxU32 nb)
+{
+	if(mDeleteRanks)
+	{
+		// Free previously used ram
+		PX_FREE_AND_RESET(mRanks2);
+		PX_FREE_AND_RESET(mRanks);
+
+		// Get some fresh one
+		mRanks	= reinterpret_cast<PxU32*>(PX_ALLOC(sizeof(PxU32)*nb, "RadixSortBuffered:mRanks"));
+		mRanks2	= reinterpret_cast<PxU32*>(PX_ALLOC(sizeof(PxU32)*nb, "RadixSortBuffered:mRanks2"));
+	}
+
+	return true;
+}
+
+PX_INLINE void RadixSortBuffered::CheckResize(PxU32 nb)
+{
+	PxU32 CurSize = CURRENT_SIZE;
+	if(nb!=CurSize)
+	{
+		if(nb>CurSize)	Resize(nb);
+		mCurrentSize = nb;
+		INVALIDATE_RANKS;
+	}
+}
+
+///////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+/**
+ *	Main sort routine.
+ *	This one is for integer values. After the call, mRanks contains a list of indices in sorted order, i.e. in the order you may process your data.
+ *	\param		input	[in] a list of integer values to sort
+ *	\param		nb		[in] number of values to sort, must be < 2^31
+ *	\param		hint	[in] RADIX_SIGNED to handle negative values, RADIX_UNSIGNED if you know your input buffer only contains positive values
+ *	\return		Self-Reference
+ */
+///////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+RadixSortBuffered& RadixSortBuffered::Sort(const PxU32* input, PxU32 nb, RadixHint hint)
+{
+	// Checkings
+	if(!input || !nb || nb&0x80000000)	return *this;
+
+	// Resize lists if needed
+	CheckResize(nb);
+
+	//Set histogram buffers.
+	PxU32 histogram[1024];
+	PxU32* links[256];
+	mHistogram1024=histogram;
+	mLinks256=links;
+
+	RadixSort::Sort(input,nb,hint);
+	return *this;
+}
+
+///////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+/**
+ *	Main sort routine.
+ *	This one is for floating-point values. After the call, mRanks contains a list of indices in sorted order, i.e. in the order you may process your data.
+ *	\param		input2			[in] a list of floating-point values to sort
+ *	\param		nb				[in] number of values to sort, must be < 2^31
+ *	\return		Self-Reference
+ *	\warning	only sorts IEEE floating-point values
+ */
+///////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+RadixSortBuffered& RadixSortBuffered::Sort(const float* input2, PxU32 nb)
+{
+	// Checkings
+	if(!input2 || !nb || nb&0x80000000)	return *this;
+
+	// Resize lists if needed
+	CheckResize(nb);
+
+	//Set histogram buffers.
+	PxU32 histogram[1024];
+	PxU32* links[256];
+	mHistogram1024=histogram;
+	mLinks256=links;
+
+	RadixSort::Sort(input2,nb);
+	return *this;
+}
+
diff --git a/PhysX_3.4/Source/Common/src/CmRadixSortBuffered.h b/PhysX_3.4/Source/Common/src/CmRadixSortBuffered.h
new file mode 100644
index 00000000..aa76bfbb
--- /dev/null
+++ b/PhysX_3.4/Source/Common/src/CmRadixSortBuffered.h
@@ -0,0 +1,61 @@
+// This code contains NVIDIA Confidential Information and is disclosed to you
+// under a form of NVIDIA software license agreement provided separately to you.
+//
+// Notice
+// NVIDIA Corporation and its licensors retain all intellectual property and
+// proprietary rights in and to this software and related documentation and
+// any modifications thereto. Any use, reproduction, disclosure, or
+// distribution of this software and related documentation without an express
+// license agreement from NVIDIA Corporation is strictly prohibited.
+//
+// ALL NVIDIA DESIGN SPECIFICATIONS, CODE ARE PROVIDED "AS IS.". NVIDIA MAKES
+// NO WARRANTIES, EXPRESSED, IMPLIED, STATUTORY, OR OTHERWISE WITH RESPECT TO
+// THE MATERIALS, AND EXPRESSLY DISCLAIMS ALL IMPLIED WARRANTIES OF NONINFRINGEMENT,
+// MERCHANTABILITY, AND FITNESS FOR A PARTICULAR PURPOSE.
+//
+// Information and code furnished is believed to be accurate and reliable.
+// However, NVIDIA Corporation assumes no responsibility for the consequences of use of such
+// information or for any infringement of patents or other rights of third parties that may
+// result from its use. No license is granted by implication or otherwise under any patent
+// or patent rights of NVIDIA Corporation. Details are subject to change without notice.
+// This code supersedes and replaces all information previously supplied.
+// NVIDIA Corporation products are not authorized for use as critical
+// components in life support devices or systems without express written approval of
+// NVIDIA Corporation.
+//
+// Copyright (c) 2008-2016 NVIDIA Corporation. All rights reserved.
+// Copyright (c) 2004-2008 AGEIA Technologies, Inc. All rights reserved.
+// Copyright (c) 2001-2004 NovodeX AG. All rights reserved.  
+
+#ifndef CM_RADIX_SORT_BUFFERED_H
+#define CM_RADIX_SORT_BUFFERED_H
+
+#include "CmPhysXCommon.h"
+#include "CmRadixSort.h"
+
+namespace physx
+{
+namespace Cm
+{
+	class PX_PHYSX_COMMON_API RadixSortBuffered : public RadixSort
+	{
+	public:
+							RadixSortBuffered();
+							~RadixSortBuffered();
+
+		RadixSortBuffered&	Sort(const PxU32* input, PxU32 nb, RadixHint hint=RADIX_SIGNED);
+		RadixSortBuffered&	Sort(const float* input, PxU32 nb);
+
+	private:
+							RadixSortBuffered(const RadixSortBuffered& object);
+							RadixSortBuffered& operator=(const RadixSortBuffered& object);
+
+		// Internal methods
+		void				CheckResize(PxU32 nb);
+		bool				Resize(PxU32 nb);
+	};
+}
+
+}
+
+#endif // CM_RADIX_SORT_BUFFERED_H
diff --git a/PhysX_3.4/Source/Common/src/CmReaderWriterLock.h b/PhysX_3.4/Source/Common/src/CmReaderWriterLock.h
new file mode 100644
index 00000000..e1b2a3c5
--- /dev/null
+++ b/PhysX_3.4/Source/Common/src/CmReaderWriterLock.h
@@ -0,0 +1,202 @@
+// This code contains NVIDIA Confidential Information and is disclosed to you
+// under a form of NVIDIA software license agreement provided separately to you.
+//
+// Notice
+// NVIDIA Corporation and its licensors retain all intellectual property and
+// proprietary rights in and to this software and related documentation and
+// any modifications thereto. Any use, reproduction, disclosure, or
+// distribution of this software and related documentation without an express
+// license agreement from NVIDIA Corporation is strictly prohibited.
+//
+// ALL NVIDIA DESIGN SPECIFICATIONS, CODE ARE PROVIDED "AS IS.". NVIDIA MAKES
+// NO WARRANTIES, EXPRESSED, IMPLIED, STATUTORY, OR OTHERWISE WITH RESPECT TO
+// THE MATERIALS, AND EXPRESSLY DISCLAIMS ALL IMPLIED WARRANTIES OF NONINFRINGEMENT,
+// MERCHANTABILITY, AND FITNESS FOR A PARTICULAR PURPOSE.
+//
+// Information and code furnished is believed to be accurate and reliable.
+// However, NVIDIA Corporation assumes no responsibility for the consequences of use of such
+// information or for any infringement of patents or other rights of third parties that may
+// result from its use. No license is granted by implication or otherwise under any patent
+// or patent rights of NVIDIA Corporation. Details are subject to change without notice.
+// This code supersedes and replaces all information previously supplied.
+// NVIDIA Corporation products are not authorized for use as critical
+// components in life support devices or systems without express written approval of
+// NVIDIA Corporation.
+//
+// Copyright (c) 2008-2016 NVIDIA Corporation. All rights reserved.
+// Copyright (c) 2004-2008 AGEIA Technologies, Inc. All rights reserved.
+// Copyright (c) 2001-2004 NovodeX AG. All rights reserved.  
+
+
+#ifndef PX_PHYSICS_COMMON_READERWRITERLOCK
+#define PX_PHYSICS_COMMON_READERWRITERLOCK
+
+//#define PX_RAYCAST_READWRITE_LOCK
+
+/*
+Implements a full reader writer lock, i.e. multiple readers or a single writer.
+
+Below is an implementation using just a mutex for non win32 platforms.
+
+Disable the full reader/writer lock by default. Using the full reader writer lock
+results in better performance when a lot of parallel raycasts are performed but is
+more costly when only a few are performed and slows down serial raycasts.
+*/
+
+namespace physx
+{
+namespace Cm
+{
+
+#ifdef PX_RAYCAST_READWRITE_LOCK
+	class ReaderWriterLock
+	{
+	public:
+
+		PX_INLINE ReaderWriterLock()
+		{
+			hReaderEvent=CreateEvent(NULL,TRUE,FALSE,NULL);
+			PX_ASSERT(hReaderEvent!=NULL);
+
+			hMutex = CreateEvent(NULL,FALSE,TRUE,NULL);
+			PX_ASSERT(hMutex!=NULL);
+
+			InitializeCriticalSection(&writerMutex);
+			counter = -1;
+			recursionCounter=0;
+		}
+		PX_INLINE ~ReaderWriterLock()
+		{
+			if(hReaderEvent!=NULL)
+				CloseHandle(hReaderEvent);
+			if(hMutex!=NULL)
+				CloseHandle(hMutex);
+			DeleteCriticalSection(&writerMutex);
+		}
+
+		PX_INLINE void lockReader()
+		{
+			if(InterlockedIncrement(&counter) == 0)
+			{
+				WaitForSingleObject(hMutex, INFINITE);
+				SetEvent(hReaderEvent);
+			}
+
+			WaitForSingleObject(hReaderEvent,INFINITE);
+		}
+
+		PX_INLINE void lockWriter()
+		{
+			EnterCriticalSection(&writerMutex);
+
+			//we may already have the global mutex(really an event so we have to handle recursion ourselves)
+			recursionCounter++;
+			if(recursionCounter==1)
+				WaitForSingleObject(hMutex, INFINITE);
+		}
+		PX_INLINE void unlockReader()
+		{
+			if(InterlockedDecrement(&counter) < 0)
+			{
+				ResetEvent(hReaderEvent);
+				SetEvent(hMutex);
+			}
+		}
+		PX_INLINE void unlockWriter()
+		{
+			recursionCounter--;
+			if(recursionCounter==0)
+				SetEvent(hMutex);
+
+			LeaveCriticalSection(&writerMutex);
+		}
+
+	private:
+
+		HANDLE hReaderEvent;
+		HANDLE hMutex;
+		CRITICAL_SECTION writerMutex;
+		LONG counter;//count the number of readers in the lock.
+		LONG recursionCounter;//handle recursive writer locking
+	};
+#else
+	/*
+	TODO: implement proper reader writer lock for other platforms, eg cell os has a native reader writer lock.
+	*/
+
+	class ReaderWriterLock
+	{
+		PX_NOCOPY(ReaderWriterLock)
+	public:
+
+		PX_INLINE void lockReader()
+		{
+			//lock.lock();
+			lock.lockReader();
+		}
+		PX_INLINE void lockWriter()
+		{
+			//lock.lock();
+			lock.lockWriter();
+		}
+		PX_INLINE void unlockReader()
+		{
+			//lock.unlock();
+			lock.unlockReader();
+		}
+		PX_INLINE void unlockWriter()
+		{
+			//lock.unlock();
+			lock.unlockWriter();
+		}
+
+	private:
+
+		//Ps::Mutex lock;
+
+		Ps::AtomicRwLock lock;
+	};
+#endif
+
+	class TakeReaderLock
+	{
+	public:
+
+		TakeReaderLock(ReaderWriterLock &l) : lock(l)
+		{
+			lock.lockReader();
+		}
+		~TakeReaderLock()
+		{
+			lock.unlockReader();
+		}
+
+	private:
+		TakeReaderLock& operator=(const TakeReaderLock&);
+		ReaderWriterLock &lock;
+	};
+
+	class TakeWriterLock
+	{
+	public:
+
+		TakeWriterLock(ReaderWriterLock &l) : lock(l)
+		{
+			lock.lockWriter();
+		}
+		~TakeWriterLock()
+		{
+			lock.unlockWriter();
+		}
+
+	private:
+		TakeWriterLock& operator=(const TakeWriterLock&);
+		ReaderWriterLock &lock;
+	};
+
+
+} // namespace Cm
+
+}
+
+#endif
diff --git a/PhysX_3.4/Source/Common/src/CmRefCountable.h b/PhysX_3.4/Source/Common/src/CmRefCountable.h
new file mode 100644
index 00000000..44087247
--- /dev/null
+++ b/PhysX_3.4/Source/Common/src/CmRefCountable.h
@@ -0,0 +1,102 @@
+// This code contains NVIDIA Confidential Information and is disclosed to you
+// under a form of NVIDIA software license agreement provided separately to you.
+//
+// Notice
+// NVIDIA Corporation and its licensors retain all intellectual property and
+// proprietary rights in and to this software and related documentation and
+// any modifications thereto. Any use, reproduction, disclosure, or
+// distribution of this software and related documentation without an express
+// license agreement from NVIDIA Corporation is strictly prohibited.
+//
+// ALL NVIDIA DESIGN SPECIFICATIONS, CODE ARE PROVIDED "AS IS.". NVIDIA MAKES
+// NO WARRANTIES, EXPRESSED, IMPLIED, STATUTORY, OR OTHERWISE WITH RESPECT TO
+// THE MATERIALS, AND EXPRESSLY DISCLAIMS ALL IMPLIED WARRANTIES OF NONINFRINGEMENT,
+// MERCHANTABILITY, AND FITNESS FOR A PARTICULAR PURPOSE.
+//
+// Information and code furnished is believed to be accurate and reliable.
+// However, NVIDIA Corporation assumes no responsibility for the consequences of use of such
+// information or for any infringement of patents or other rights of third parties that may
+// result from its use. No license is granted by implication or otherwise under any patent
+// or patent rights of NVIDIA Corporation. Details are subject to change without notice.
+// This code supersedes and replaces all information previously supplied.
+// NVIDIA Corporation products are not authorized for use as critical
+// components in life support devices or systems without express written approval of
+// NVIDIA Corporation.
+//
+// Copyright (c) 2008-2016 NVIDIA Corporation. All rights reserved.
+// Copyright (c) 2004-2008 AGEIA Technologies, Inc. All rights reserved.
+// Copyright (c) 2001-2004 NovodeX AG. All rights reserved.  
+
+
+#ifndef PX_PHYSICS_COMMON_REFCOUNTABLE
+#define PX_PHYSICS_COMMON_REFCOUNTABLE
+
+#include "foundation/PxAssert.h"
+#include "PsAtomic.h"
+
+namespace physx
+{
+namespace Cm
+{
+
+	// simple thread-safe reference count
+	// when the ref count is zero, the object is in an undefined state (pending delete)
+
+	class RefCountable
+	{
+	//= ATTENTION! =====================================================================================
+	// Changing the data layout of this class breaks the binary serialization format.  See comments for 
+	// PX_BINARY_SERIAL_VERSION.  If a modification is required, please adjust the getBinaryMetaData 
+	// function.  If the modification is made on a custom branch, please change PX_BINARY_SERIAL_VERSION
+	// accordingly.
+	//==================================================================================================
+	public:
+// PX_SERIALIZATION
+		RefCountable(const PxEMPTY) : mRefCount(1)	{}
+		static	void	getBinaryMetaData(PxOutputStream& stream);
+//~PX_SERIALIZATION
+		explicit RefCountable(PxU32 initialCount = 1)
+			: mRefCount(PxI32(initialCount))
+		{
+			PX_ASSERT(mRefCount!=0);
+		}
+
+		virtual ~RefCountable() {}
+
+		/**
+		Calls 'delete this;'. It needs to be overloaded for classes also deriving from 
+		PxBase and call 'Cm::deletePxBase(this);' instead.
+		*/
+		virtual	void onRefCountZero()
+		{
+			delete this;
+		}
+
+		void incRefCount()
+		{
+			physx::shdfnd::atomicIncrement(&mRefCount);
+			// value better be greater than 1, or we've created a ref to an undefined object
+			PX_ASSERT(mRefCount>1);
+		}
+
+		void decRefCount()
+		{
+			PX_ASSERT(mRefCount>0);
+			if(physx::shdfnd::atomicDecrement(&mRefCount) == 0)
+				onRefCountZero();
+		}
+
+		PX_FORCE_INLINE PxU32 getRefCount() const
+		{
+			return PxU32(mRefCount);
+		}
+	private:
+		volatile PxI32 mRefCount;
+	};
+
+
+} // namespace Cm
+
+}
+
+#endif
diff --git a/PhysX_3.4/Source/Common/src/CmRenderBuffer.h b/PhysX_3.4/Source/Common/src/CmRenderBuffer.h
new file mode 100644
index 00000000..b07fc1f8
--- /dev/null
+++ b/PhysX_3.4/Source/Common/src/CmRenderBuffer.h
@@ -0,0 +1,132 @@
+// This code contains NVIDIA Confidential Information and is disclosed to you
+// under a form of NVIDIA software license agreement provided separately to you.
+//
+// Notice
+// NVIDIA Corporation and its licensors retain all intellectual property and
+// proprietary rights in and to this software and related documentation and
+// any modifications thereto. Any use, reproduction, disclosure, or
+// distribution of this software and related documentation without an express
+// license agreement from NVIDIA Corporation is strictly prohibited.
+//
+// ALL NVIDIA DESIGN SPECIFICATIONS, CODE ARE PROVIDED "AS IS.". NVIDIA MAKES
+// NO WARRANTIES, EXPRESSED, IMPLIED, STATUTORY, OR OTHERWISE WITH RESPECT TO
+// THE MATERIALS, AND EXPRESSLY DISCLAIMS ALL IMPLIED WARRANTIES OF NONINFRINGEMENT,
+// MERCHANTABILITY, AND FITNESS FOR A PARTICULAR PURPOSE.
+//
+// Information and code furnished is believed to be accurate and reliable.
+// However, NVIDIA Corporation assumes no responsibility for the consequences of use of such
+// information or for any infringement of patents or other rights of third parties that may
+// result from its use. No license is granted by implication or otherwise under any patent
+// or patent rights of NVIDIA Corporation. Details are subject to change without notice.
+// This code supersedes and replaces all information previously supplied.
+// NVIDIA Corporation products are not authorized for use as critical
+// components in life support devices or systems without express written approval of
+// NVIDIA Corporation.
+//
+// Copyright (c) 2008-2016 NVIDIA Corporation. All rights reserved.
+// Copyright (c) 2004-2008 AGEIA Technologies, Inc. All rights reserved.
+// Copyright (c) 2001-2004 NovodeX AG. All rights reserved.  
+
+
+#ifndef PX_FOUNDATION_PSRENDERBUFFER_H
+#define PX_FOUNDATION_PSRENDERBUFFER_H
+
+#include "common/PxRenderBuffer.h"
+#include "CmPhysXCommon.h"
+#include "PsArray.h"
+#include "PsUserAllocated.h"
+
+namespace physx
+{
+namespace Cm
+{
+	/**
+	Implementation of PxRenderBuffer.
+	*/
+	class RenderBuffer : public PxRenderBuffer, public Ps::UserAllocated
+	{
+
+		template <typename T>
+		void append(Ps::Array<T>& dst, const T* src, PxU32 count)
+		{
+			dst.reserve(dst.size() + count);
+			for(const T* end=src+count; src<end; ++src)
+				dst.pushBack(*src);
+		}
+
+	public:
+
+		RenderBuffer() :
+			mPoints(PX_DEBUG_EXP("renderBufferPoints")),
+			mLines(PX_DEBUG_EXP("renderBufferLines")),
+			mTriangles(PX_DEBUG_EXP("renderBufferTriangles")),
+			mTexts(PX_DEBUG_EXP("renderBufferTexts")),
+			mCharBuf(PX_DEBUG_EXP("renderBufferCharBuf"))
+		{}
+		
+
+		virtual PxU32 getNbPoints() const { return mPoints.size(); }
+		virtual const PxDebugPoint* getPoints() const { return mPoints.begin(); }
+		virtual PxU32 getNbLines() const { return mLines.size(); }
+		virtual const PxDebugLine* getLines() const { return mLines.begin(); }
+		virtual PxU32 getNbTriangles() const { return mTriangles.size(); }
+		virtual const PxDebugTriangle* getTriangles() const { return mTriangles.begin(); }
+		virtual PxU32 getNbTexts() const { return mTexts.size(); }
+		virtual const PxDebugText* getTexts() const { return mTexts.begin(); }
+
+		virtual void append(const PxRenderBuffer& other)
+		{
+			append(mPoints, other.getPoints(), other.getNbPoints());
+			append(mLines, other.getLines(), other.getNbLines());
+			append(mTriangles, other.getTriangles(), other.getNbTriangles());
+			append(mTexts, other.getTexts(), other.getNbTexts());
+		}
+
+		virtual void clear()
+		{
+			mPoints.clear(); 
+			mLines.clear();
+			mTriangles.clear();
+			mTexts.clear();
+			mCharBuf.clear();
+		}
+
+		bool empty() const 
+		{
+			return mPoints.empty() && mLines.empty() && mTriangles.empty() && mTexts.empty()&& mCharBuf.empty();
+		}
+
+		void shift(const PxVec3& delta)
+		{
+			for(PxU32 i=0; i < mPoints.size(); i++)
+				mPoints[i].pos += delta;
+
+			for(PxU32 i=0; i < mLines.size(); i++)
+			{
+				mLines[i].pos0 += delta;
+				mLines[i].pos1 += delta;
+			}
+
+			for(PxU32 i=0; i < mTriangles.size(); i++)
+			{
+				mTriangles[i].pos0 += delta;
+				mTriangles[i].pos1 += delta;
+				mTriangles[i].pos2 += delta;
+			}
+
+			for(PxU32 i=0; i < mTexts.size(); i++)
+				mTexts[i].position += delta;
+		}
+
+		Ps::Array<PxDebugPoint>		mPoints;
+		Ps::Array<PxDebugLine>		mLines;
+		Ps::Array<PxDebugTriangle>	mTriangles;
+		Ps::Array<PxDebugText>		mTexts;
+		Ps::Array<char>				mCharBuf;
+	};
+
+} // Cm
+
+}
+
+#endif
diff --git a/PhysX_3.4/Source/Common/src/CmRenderOutput.cpp b/PhysX_3.4/Source/Common/src/CmRenderOutput.cpp
new file mode 100644
index 00000000..2226d137
--- /dev/null
+++ b/PhysX_3.4/Source/Common/src/CmRenderOutput.cpp
@@ -0,0 +1,301 @@
+// This code contains NVIDIA Confidential Information and is disclosed to you
+// under a form of NVIDIA software license agreement provided separately to you.
+//
+// Notice
+// NVIDIA Corporation and its licensors retain all intellectual property and
+// proprietary rights in and to this software and related documentation and
+// any modifications thereto. Any use, reproduction, disclosure, or
+// distribution of this software and related documentation without an express
+// license agreement from NVIDIA Corporation is strictly prohibited.
+//
+// ALL NVIDIA DESIGN SPECIFICATIONS, CODE ARE PROVIDED "AS IS.". NVIDIA MAKES
+// NO WARRANTIES, EXPRESSED, IMPLIED, STATUTORY, OR OTHERWISE WITH RESPECT TO
+// THE MATERIALS, AND EXPRESSLY DISCLAIMS ALL IMPLIED WARRANTIES OF NONINFRINGEMENT,
+// MERCHANTABILITY, AND FITNESS FOR A PARTICULAR PURPOSE.
+//
+// Information and code furnished is believed to be accurate and reliable.
+// However, NVIDIA Corporation assumes no responsibility for the consequences of use of such
+// information or for any infringement of patents or other rights of third parties that may
+// result from its use. No license is granted by implication or otherwise under any patent
+// or patent rights of NVIDIA Corporation. Details are subject to change without notice.
+// This code supersedes and replaces all information previously supplied.
+// NVIDIA Corporation products are not authorized for use as critical
+// components in life support devices or systems without express written approval of
+// NVIDIA Corporation.
+//
+// Copyright (c) 2008-2016 NVIDIA Corporation. All rights reserved.
+// Copyright (c) 2004-2008 AGEIA Technologies, Inc. All rights reserved.
+// Copyright (c) 2001-2004 NovodeX AG. All rights reserved.  
+
+
+#include "foundation/PxMat44.h"
+#include "CmRenderOutput.h"
+#include "PsMathUtils.h"
+#include "PsString.h"
+#include <stdarg.h>
+
+#if PX_VC
+#pragma warning(disable: 4342 ) // behavior change: 'function' called, but a member operator was called in previous versions
+#pragma warning(disable: 4996 ) // intentionally suppressing this warning message
+#endif
+
+using namespace physx;
+using namespace Cm;
+
+
+namespace physx
+{
+namespace Cm
+{
+	RenderOutput& RenderOutput::operator<<(Primitive prim)
+	{
+		mPrim = prim; mVertexCount = 0; return *this;
+	}
+
+	RenderOutput& RenderOutput::operator<<(PxU32 color)
+	{
+		mColor = color; return *this;
+	}
+
+	RenderOutput& RenderOutput::operator<<(const PxMat44& transform)
+	{
+		mTransform = transform; return *this;
+	}
+
+	RenderOutput& RenderOutput::operator<<(const PxTransform&t)
+	{
+		mTransform = PxMat44(t);
+		return *this;
+	}
+
+	RenderOutput& RenderOutput::operator<<(PxVec3 vertex)
+	{
+		// apply transformation
+		vertex = mTransform.transform(vertex);
+		++mVertexCount;
+
+		// add primitive to render buffer
+		switch(mPrim)
+		{
+		case POINTS:
+			mBuffer.mPoints.pushBack(PxDebugPoint(vertex, mColor)); break;
+		case LINES:
+			if(mVertexCount == 2)
+			{
+				mBuffer.mLines.pushBack(PxDebugLine(mVertex0, vertex, mColor)); 
+				mVertexCount = 0;
+			}
+			break;
+		case LINESTRIP:
+			if(mVertexCount >= 2)
+				mBuffer.mLines.pushBack(PxDebugLine(mVertex0, vertex, mColor)); 
+			break;
+		case TRIANGLES:
+			if(mVertexCount == 3)
+			{
+				mBuffer.mTriangles.pushBack(PxDebugTriangle(mVertex1, mVertex0, vertex, mColor)); 
+				mVertexCount = 0;
+			}
+			break;
+		case TRIANGLESTRIP:
+			if(mVertexCount >= 3)
+				mBuffer.mTriangles.pushBack(PxDebugTriangle(
+					(mVertexCount & 0x1) ? mVertex0 : mVertex1, 
+					(mVertexCount & 0x1) ? mVertex1 : mVertex0, vertex, mColor)); 
+			break;
+		case TEXT:	
+			break;
+		}
+
+		// cache the last 2 vertices (for strips)
+		if(1 < mVertexCount)
+		{
+			mVertex1 = mVertex0;
+			mVertex0 = vertex;
+		} else {
+			mVertex0 = vertex;
+		}
+		return *this;
+	}
+
+	DebugText::DebugText(const PxVec3& position_, PxReal size_, const char* string, ...) 
+	: position(position_), size(size_) 
+	{
+		va_list argList;
+		va_start(argList, string);
+		if(0 >= Ps::vsnprintf(buffer, sBufferSize-1, string, argList))
+			buffer[sBufferSize-1] = 0; // terminate string
+		va_end(argList);
+	}
+
+	RenderOutput& RenderOutput::operator<<(const DebugText& text)
+	{
+		const PxU32 n = PxU32(strlen(text.buffer));
+		const PxU32 newCharBufSize = mBuffer.mCharBuf.size()+n+1;
+		if(mBuffer.mCharBuf.capacity() < newCharBufSize)
+		{
+			char* oldBuf = mBuffer.mCharBuf.begin(); 
+			mBuffer.mCharBuf.reserve(newCharBufSize);
+			intptr_t diff =  mBuffer.mCharBuf.begin() - oldBuf;
+			for (PxU32 i = 0; i < mBuffer.mTexts.size(); ++i)
+				mBuffer.mTexts[i].string += diff;
+		}
+		mBuffer.mTexts.pushBack(PxDebugText(mTransform.transform(text.position), text.size, mColor, mBuffer.mCharBuf.end()));
+		for(size_t i=0; i<=n; ++i)
+			mBuffer.mCharBuf.pushBack(text.buffer[i]);
+		return *this;
+	}
+
+	RenderOutput& operator<<(RenderOutput& out, const DebugBox& box)
+	{
+		if(box.wireframe)
+		{
+			out << RenderOutput::LINESTRIP;
+			out << PxVec3(box.minimum.x, box.minimum.y, box.minimum.z);
+			out << PxVec3(box.maximum.x, box.minimum.y, box.minimum.z);
+			out << PxVec3(box.maximum.x, box.maximum.y, box.minimum.z);
+			out << PxVec3(box.minimum.x, box.maximum.y, box.minimum.z);
+			out << PxVec3(box.minimum.x, box.minimum.y, box.minimum.z);
+			out << PxVec3(box.minimum.x, box.minimum.y, box.maximum.z);
+			out << PxVec3(box.maximum.x, box.minimum.y, box.maximum.z);
+			out << PxVec3(box.maximum.x, box.maximum.y, box.maximum.z);
+			out << PxVec3(box.minimum.x, box.maximum.y, box.maximum.z);
+			out << PxVec3(box.minimum.x, box.minimum.y, box.maximum.z);
+			out << RenderOutput::LINES;
+			out << PxVec3(box.maximum.x, box.minimum.y, box.minimum.z); 
+			out << PxVec3(box.maximum.x, box.minimum.y, box.maximum.z);
+			out << PxVec3(box.maximum.x, box.maximum.y, box.minimum.z); 
+			out << PxVec3(box.maximum.x, box.maximum.y, box.maximum.z);
+			out << PxVec3(box.minimum.x, box.maximum.y, box.minimum.z); 
+			out << PxVec3(box.minimum.x, box.maximum.y, box.maximum.z);
+		}
+		else
+		{
+			out << RenderOutput::TRIANGLESTRIP;
+			out << PxVec3(box.minimum.x, box.minimum.y, box.minimum.z); // 0
+			out << PxVec3(box.minimum.x, box.maximum.y, box.minimum.z); // 2
+			out << PxVec3(box.maximum.x, box.minimum.y, box.minimum.z); // 1
+			out << PxVec3(box.maximum.x, box.maximum.y, box.minimum.z); // 3
+			out << PxVec3(box.maximum.x, box.maximum.y, box.maximum.z); // 7
+			out << PxVec3(box.minimum.x, box.maximum.y, box.minimum.z); // 2
+			out << PxVec3(box.minimum.x, box.maximum.y, box.maximum.z); // 6
+			out << PxVec3(box.minimum.x, box.minimum.y, box.minimum.z); // 0
+			out << PxVec3(box.minimum.x, box.minimum.y, box.maximum.z); // 4
+			out << PxVec3(box.maximum.x, box.minimum.y, box.minimum.z); // 1
+			out << PxVec3(box.maximum.x, box.minimum.y, box.maximum.z); // 5
+			out << PxVec3(box.maximum.x, box.maximum.y, box.maximum.z); // 7
+			out << PxVec3(box.minimum.x, box.minimum.y, box.maximum.z); // 4
+			out << PxVec3(box.minimum.x, box.maximum.y, box.maximum.z); // 6
+		}
+		return out;
+	}
+
+	RenderOutput& operator<<(RenderOutput& out, const DebugArrow& arrow)
+	{
+		PxVec3 t0 = arrow.tip - arrow.base, t1, t2; 
+
+		t0.normalize();
+		Ps::normalToTangents(t0, t1, t2);
+
+		const PxReal tipAngle = 0.25f;
+		t1 *= arrow.headLength * tipAngle; 
+		t2 *= arrow.headLength * tipAngle * PxSqrt(3.0f); 
+		PxVec3 headBase = arrow.tip - t0 * arrow.headLength;
+
+		out << RenderOutput::LINES;
+		out << arrow.base << arrow.tip;
+
+		out << RenderOutput::TRIANGLESTRIP;
+		out << arrow.tip;
+		out << headBase + t1 + t1;
+		out << headBase - t1 - t2;
+		out << headBase - t1 + t2;
+		out << arrow.tip;
+		out << headBase + t1 + t1;
+		return out;
+	}
+
+	RenderOutput& operator<<(RenderOutput& out, const DebugBasis& basis)
+	{
+		const PxReal headLength = basis.extends.magnitude() * 0.15f;
+		out << basis.colorX << DebugArrow(PxVec3(0.0f), PxVec3(basis.extends.x, 0, 0), headLength);
+		out << basis.colorY << DebugArrow(PxVec3(0.0f), PxVec3(0, basis.extends.y, 0), headLength);
+		out << basis.colorZ << DebugArrow(PxVec3(0.0f), PxVec3(0, 0, basis.extends.z), headLength);
+		return out;
+	}
+
+	RenderOutput& operator<<(RenderOutput& out, const DebugCircle& circle)
+	{
+		const PxF32 step = PxTwoPi/circle.nSegments;
+		PxF32 angle = 0;
+		out << RenderOutput::LINESTRIP;
+		for(PxU32 i=0; i<circle.nSegments; i++, angle += step)
+			out << PxVec3(circle.radius * PxSin(angle), circle.radius * PxCos(angle), 0);
+		out << PxVec3(0, circle.radius, 0);
+		return out;
+	}
+
+	RenderOutput& operator<<(RenderOutput& out, const DebugArc& arc)
+	{
+		const PxF32 step = (arc.maxAngle - arc.minAngle) / arc.nSegments;
+		PxF32 angle = arc.minAngle;
+		out << RenderOutput::LINESTRIP;
+		for(PxU32 i=0; i<arc.nSegments; i++, angle += step)
+			out << PxVec3(arc.radius * PxSin(angle), arc.radius * PxCos(angle), 0);
+		out << PxVec3(arc.radius * PxSin(arc.maxAngle), arc.radius * PxCos(arc.maxAngle), 0);
+		return out;
+	}
+
+	// PT: I need those functions available here so that I don't have to duplicate all the code in other modules like the CCT.
+	// PT: TODO: move other functions here as well
+
+	RenderOutput& RenderOutput::outputCapsule(PxF32 radius, PxF32 halfHeight, const PxMat44& absPose)
+	{
+		RenderOutput& out = *this;
+
+		const PxVec3 vleft2(-halfHeight, 0.0f, 0.0f);
+		PxMat44 left2 = absPose;
+		left2.column3 += PxVec4(left2.rotate(vleft2), 0.0f);
+		out << left2 << Cm::DebugArc(100, radius, PxPi, PxTwoPi);
+
+		PxMat44 rotPose = left2;
+		Ps::swap(rotPose.column1, rotPose.column2);
+		rotPose.column1 = -rotPose.column1;
+		out << rotPose << Cm::DebugArc(100, radius, PxPi, PxTwoPi);
+
+		Ps::swap(rotPose.column0, rotPose.column2);
+		rotPose.column0 = -rotPose.column0;
+		out << rotPose << Cm::DebugCircle(100, radius);
+
+		const PxVec3 vright2(halfHeight, 0.0f, 0.0f);
+		PxMat44 right2 = absPose;
+		right2.column3 += PxVec4(right2.rotate(vright2), 0.0f);
+		out << right2 << Cm::DebugArc(100, radius, 0.0f, PxPi);
+
+		rotPose = right2;
+		Ps::swap(rotPose.column1, rotPose.column2);
+		rotPose.column1 = -rotPose.column1;
+		out << rotPose << Cm::DebugArc(100, radius, 0.0f, PxPi);
+
+		Ps::swap(rotPose.column0, rotPose.column2);
+		rotPose.column0 = -rotPose.column0;
+		out << rotPose << Cm::DebugCircle(100, radius);
+
+		out << absPose;
+		out.outputSegment(	absPose.transform(PxVec3(-halfHeight,  radius, 0)),
+							absPose.transform(PxVec3( halfHeight,  radius, 0)));
+		out.outputSegment(	absPose.transform(PxVec3(-halfHeight, -radius, 0)),
+							absPose.transform(PxVec3( halfHeight, -radius, 0)));
+		out.outputSegment(	absPose.transform(PxVec3(-halfHeight,  0, radius)),
+							absPose.transform(PxVec3( halfHeight,  0, radius)));
+		out.outputSegment(	absPose.transform(PxVec3(-halfHeight, 0, -radius)),
+							absPose.transform(PxVec3( halfHeight, 0, -radius)));
+
+		return *this;
+	}
+
+
+
+} // Cm
+
+}
diff --git a/PhysX_3.4/Source/Common/src/CmRenderOutput.h b/PhysX_3.4/Source/Common/src/CmRenderOutput.h
new file mode 100644
index 00000000..a121cb27
--- /dev/null
+++ b/PhysX_3.4/Source/Common/src/CmRenderOutput.h
@@ -0,0 +1,191 @@
+// This code contains NVIDIA Confidential Information and is disclosed to you
+// under a form of NVIDIA software license agreement provided separately to you.
+//
+// Notice
+// NVIDIA Corporation and its licensors retain all intellectual property and
+// proprietary rights in and to this software and related documentation and
+// any modifications thereto. Any use, reproduction, disclosure, or
+// distribution of this software and related documentation without an express
+// license agreement from NVIDIA Corporation is strictly prohibited.
+//
+// ALL NVIDIA DESIGN SPECIFICATIONS, CODE ARE PROVIDED "AS IS.". NVIDIA MAKES
+// NO WARRANTIES, EXPRESSED, IMPLIED, STATUTORY, OR OTHERWISE WITH RESPECT TO
+// THE MATERIALS, AND EXPRESSLY DISCLAIMS ALL IMPLIED WARRANTIES OF NONINFRINGEMENT,
+// MERCHANTABILITY, AND FITNESS FOR A PARTICULAR PURPOSE.
+//
+// Information and code furnished is believed to be accurate and reliable.
+// However, NVIDIA Corporation assumes no responsibility for the consequences of use of such
+// information or for any infringement of patents or other rights of third parties that may
+// result from its use. No license is granted by implication or otherwise under any patent
+// or patent rights of NVIDIA Corporation. Details are subject to change without notice.
+// This code supersedes and replaces all information previously supplied.
+// NVIDIA Corporation products are not authorized for use as critical
+// components in life support devices or systems without express written approval of
+// NVIDIA Corporation.
+//
+// Copyright (c) 2008-2016 NVIDIA Corporation. All rights reserved.
+// Copyright (c) 2004-2008 AGEIA Technologies, Inc. All rights reserved.
+// Copyright (c) 2001-2004 NovodeX AG. All rights reserved.  
+
+
+#ifndef PX_FOUNDATION_PSRENDEROUTPUT_H
+#define PX_FOUNDATION_PSRENDEROUTPUT_H
+
+#include "foundation/PxMat44.h"
+#include "CmRenderBuffer.h"
+
+namespace physx
+{
+namespace Cm
+{
+	struct DebugText;
+
+#if PX_VC 
+    #pragma warning(push)
+	#pragma warning( disable : 4251 ) // class needs to have dll-interface to be used by clients of class
+#endif
+	/**
+	Output stream to fill RenderBuffer
+	*/
+	class PX_PHYSX_COMMON_API RenderOutput
+	{
+	public:
+
+		enum Primitive {
+			POINTS,
+			LINES,
+			LINESTRIP,
+			TRIANGLES,
+			TRIANGLESTRIP,
+			TEXT
+		};
+
+		RenderOutput(RenderBuffer& buffer) 
+			: mPrim(POINTS), mColor(0), mVertex0(0.0f), mVertex1(0.0f)
+			, mVertexCount(0), mTransform(PxMat44(PxIdentity)), mBuffer(buffer) 
+		{}
+
+		RenderOutput& operator<<(Primitive prim);
+		RenderOutput& operator<<(PxU32 color); // 0xbbggrr
+		RenderOutput& operator<<(const PxMat44& transform);
+		RenderOutput& operator<<(const PxTransform&);
+
+		RenderOutput& operator<<(PxVec3 vertex);
+		RenderOutput& operator<<(const DebugText& text);
+
+		PX_FORCE_INLINE PxDebugLine* reserveSegments(PxU32 nbSegments)
+		{
+			const PxU32 currentSize = mBuffer.mLines.size();
+			mBuffer.mLines.resizeUninitialized(currentSize + nbSegments);
+			return mBuffer.mLines.begin() + currentSize;
+		}
+
+		// PT: using the operators is just too slow.
+		PX_FORCE_INLINE	void outputSegment(const PxVec3& v0, const PxVec3& v1)
+		{
+			// PT: TODO: replace pushBack with something faster like the versions below
+			mBuffer.mLines.pushBack(PxDebugLine(v0, v1, mColor));
+
+			// PT: TODO: use the "pushBackUnsafe" version or replace with ICE containers
+/*			PxDebugLine& l = mBuffer.mLines.insert();
+			l.pos0	= v0;
+			l.pos1	= v1;
+			l.color0 = l.color1 = mColor;
+
+			PxDebugLine& l = mBuffer.mLines.pushBackUnsafe();
+			l.pos0	= v0;
+			l.pos1	= v1;
+			l.color0 = l.color1 = mColor;*/
+		}
+
+		RenderOutput&	outputCapsule(PxF32 radius, PxF32 halfHeight, const PxMat44& absPose);
+
+	private:
+
+		RenderOutput& operator=(const RenderOutput&);
+
+		Primitive		mPrim;
+		PxU32			mColor;
+		PxVec3			mVertex0, mVertex1;
+		PxU32			mVertexCount;
+		PxMat44			mTransform;
+		RenderBuffer&	mBuffer;
+	};
+
+	/** debug render helper types */
+	struct PX_PHYSX_COMMON_API DebugText 
+	{
+		DebugText(const PxVec3& position, PxReal size, const char* string, ...); 
+		static const int sBufferSize = 1008; // sizeof(DebugText)==1kB
+		char buffer[sBufferSize]; 
+		PxVec3 position;
+		PxReal size;
+	};
+
+	struct DebugBox 
+	{
+		explicit DebugBox(const PxVec3& extents, bool wireframe_ = true) 
+		: minimum(-extents), maximum(extents), wireframe(wireframe_) {}
+
+		explicit DebugBox(const PxVec3& pos, const PxVec3& extents, bool wireframe_ = true) 
+		: minimum(pos-extents), maximum(pos+extents), wireframe(wireframe_) {}
+
+		explicit DebugBox(const PxBounds3& bounds, bool wireframe_ = true)
+		: minimum(bounds.minimum), maximum(bounds.maximum), wireframe(wireframe_) {}
+
+		PxVec3 minimum, maximum;
+		bool wireframe;
+	};
+	PX_PHYSX_COMMON_API RenderOutput& operator<<(RenderOutput& out, const DebugBox& box);
+
+	struct DebugArrow 
+	{
+		DebugArrow(const PxVec3& pos, const PxVec3& vec) 
+		: base(pos), tip(pos+vec), headLength(vec.magnitude()*0.15f) {}
+
+		DebugArrow(const PxVec3& pos, const PxVec3& vec, PxReal headLength_) 
+		: base(pos), tip(pos+vec), headLength(headLength_) {}
+
+		PxVec3 base, tip;
+		PxReal headLength;
+	};
+	PX_PHYSX_COMMON_API RenderOutput& operator<<(RenderOutput& out, const DebugArrow& arrow);
+
+	struct DebugBasis
+	{
+		DebugBasis(const PxVec3& ext, PxU32 cX = PxDebugColor::eARGB_RED, 
+			PxU32 cY = PxDebugColor::eARGB_GREEN, PxU32 cZ = PxDebugColor::eARGB_BLUE) 
+		: extends(ext), colorX(cX), colorY(cY), colorZ(cZ) {}
+		PxVec3 extends;
+		PxU32 colorX, colorY, colorZ;
+	};
+	PX_PHYSX_COMMON_API RenderOutput& operator<<(RenderOutput& out, const DebugBasis& basis);
+
+#if PX_VC 
+     #pragma warning(pop) 
+#endif
+
+	struct DebugCircle
+	{
+		DebugCircle(PxU32 s, PxReal r) 
+		: nSegments(s), radius(r) {}
+		PxU32 nSegments;
+		PxReal radius;
+	};
+	PX_PHYSX_COMMON_API RenderOutput& operator<<(RenderOutput& out, const DebugCircle& circle);
+
+	struct DebugArc
+	{
+		DebugArc(PxU32 s, PxReal r, PxReal minAng, PxReal maxAng) 
+		: nSegments(s), radius(r), minAngle(minAng), maxAngle(maxAng) {}
+		PxU32 nSegments;
+		PxReal radius;
+		PxReal minAngle, maxAngle;
+	};
+	PX_PHYSX_COMMON_API RenderOutput& operator<<(RenderOutput& out, const DebugArc& arc);
+
+} // namespace Cm
+
+}
+
+#endif
diff --git a/PhysX_3.4/Source/Common/src/CmScaling.h b/PhysX_3.4/Source/Common/src/CmScaling.h
new file mode 100644
index 00000000..3753b678
--- /dev/null
+++ b/PhysX_3.4/Source/Common/src/CmScaling.h
@@ -0,0 +1,227 @@
+// This code contains NVIDIA Confidential Information and is disclosed to you
+// under a form of NVIDIA software license agreement provided separately to you.
+//
+// Notice
+// NVIDIA Corporation and its licensors retain all intellectual property and
+// proprietary rights in and to this software and related documentation and
+// any modifications thereto. Any use, reproduction, disclosure, or
+// distribution of this software and related documentation without an express
+// license agreement from NVIDIA Corporation is strictly prohibited.
+//
+// ALL NVIDIA DESIGN SPECIFICATIONS, CODE ARE PROVIDED "AS IS.". NVIDIA MAKES
+// NO WARRANTIES, EXPRESSED, IMPLIED, STATUTORY, OR OTHERWISE WITH RESPECT TO
+// THE MATERIALS, AND EXPRESSLY DISCLAIMS ALL IMPLIED WARRANTIES OF NONINFRINGEMENT,
+// MERCHANTABILITY, AND FITNESS FOR A PARTICULAR PURPOSE.
+//
+// Information and code furnished is believed to be accurate and reliable.
+// However, NVIDIA Corporation assumes no responsibility for the consequences of use of such
+// information or for any infringement of patents or other rights of third parties that may
+// result from its use. No license is granted by implication or otherwise under any patent
+// or patent rights of NVIDIA Corporation. Details are subject to change without notice.
+// This code supersedes and replaces all information previously supplied.
+// NVIDIA Corporation products are not authorized for use as critical
+// components in life support devices or systems without express written approval of
+// NVIDIA Corporation.
+//
+// Copyright (c) 2008-2016 NVIDIA Corporation. All rights reserved.
+// Copyright (c) 2004-2008 AGEIA Technologies, Inc. All rights reserved.
+// Copyright (c) 2001-2004 NovodeX AG. All rights reserved.  
+
+
+#ifndef PX_PHYSICS_COMMON_SCALING
+#define PX_PHYSICS_COMMON_SCALING
+
+#include "foundation/PxBounds3.h"
+#include "foundation/PxMat33.h"
+#include "geometry/PxMeshScale.h"
+#include "CmMatrix34.h"
+#include "CmUtils.h"
+#include "PsMathUtils.h"
+
+namespace physx
+{
+namespace Cm
+{
+	// class that can perform scaling fast.  Relatively large size, generated from PxMeshScale on demand.
+	// CS: I've removed most usages of this class, because most of the time only one-way transform is needed.
+	// If you only need a temporary FastVertex2ShapeScaling, setup your transform as PxMat34Legacy and use
+	// normal matrix multiplication or a transform() overload to convert points and bounds between spaces.
+	class FastVertex2ShapeScaling
+	{
+	public:
+		PX_INLINE FastVertex2ShapeScaling()
+		{
+			//no scaling by default:
+			vertex2ShapeSkew = PxMat33(PxIdentity);
+			shape2VertexSkew = PxMat33(PxIdentity);
+			mFlipNormal = false;
+		}
+
+		PX_INLINE explicit FastVertex2ShapeScaling(const PxMeshScale& scale)
+		{
+			init(scale);
+		}
+
+		PX_INLINE FastVertex2ShapeScaling(const PxVec3& scale, const PxQuat& rotation)
+		{
+			init(scale, rotation);
+		}
+
+		PX_INLINE void init(const PxMeshScale& scale)
+		{
+			init(scale.scale, scale.rotation);
+		}
+
+		PX_INLINE void setIdentity()
+		{
+			vertex2ShapeSkew = PxMat33(PxIdentity);
+			shape2VertexSkew = PxMat33(PxIdentity);
+			mFlipNormal = false;
+		}
+
+		PX_INLINE void init(const PxVec3& scale, const PxQuat& rotation)
+		{
+			// TODO: may want to optimize this for cases where we have uniform or axis aligned scaling!
+			// That would introduce branches and it's unclear to me whether that's faster than just doing the math.
+			// Lazy computation would be another option, at the cost of introducing even more branches.
+
+			const PxMat33 R(rotation);
+			vertex2ShapeSkew = R.getTranspose();
+			const PxMat33 diagonal = PxMat33::createDiagonal(scale);
+			vertex2ShapeSkew = vertex2ShapeSkew * diagonal;
+			vertex2ShapeSkew = vertex2ShapeSkew * R;
+
+			/*
+			The inverse, is, explicitly:
+			shape2VertexSkew.setTransposed(R);
+			shape2VertexSkew.multiplyDiagonal(PxVec3(1.0f/scale.x, 1.0f/scale.y, 1.0f/scale.z));
+			shape2VertexSkew *= R;
+
+			It may be competitive to compute the inverse -- though this has a branch in it:
+			*/
+			
+			shape2VertexSkew = vertex2ShapeSkew.getInverse();			
+
+			mFlipNormal = ((scale.x * scale.y * scale.z) < 0.0f);
+		}
+
+		PX_FORCE_INLINE void flipNormal(PxVec3& v1, PxVec3& v2) const
+		{
+			if (mFlipNormal)
+			{
+				PxVec3 tmp = v1; v1 = v2; v2 = tmp;
+			}
+		}
+
+		PX_FORCE_INLINE PxVec3 operator*  (const PxVec3& src) const
+		{
+			return vertex2ShapeSkew * src;
+		}
+		PX_FORCE_INLINE PxVec3 operator%  (const PxVec3& src) const
+		{
+			return shape2VertexSkew * src;
+		}
+
+		PX_FORCE_INLINE const PxMat33& getVertex2ShapeSkew() const
+		{
+			return vertex2ShapeSkew;
+		}
+
+		PX_FORCE_INLINE const PxMat33& getShape2VertexSkew() const
+		{
+			return shape2VertexSkew;
+		}
+
+		PX_INLINE Cm::Matrix34 getVertex2WorldSkew(const Cm::Matrix34& shape2world) const
+		{
+			const Cm::Matrix34 vertex2worldSkew = shape2world * getVertex2ShapeSkew();
+			//vertex2worldSkew = shape2world * [vertex2shapeSkew, 0]
+			//[aR at] * [bR bt] = [aR * bR		aR * bt + at]  NOTE: order of operations important so it works when this ?= left ?= right.
+			return vertex2worldSkew;
+		}
+
+		PX_INLINE Cm::Matrix34 getWorld2VertexSkew(const Cm::Matrix34& shape2world) const
+		{
+			//world2vertexSkew = shape2vertex * invPQ(shape2world)
+			//[aR 0] * [bR' -bR'*bt] = [aR * bR'		-aR * bR' * bt + 0]
+
+			const PxMat33 rotate( shape2world[0], shape2world[1], shape2world[2] );
+			const PxMat33 M = getShape2VertexSkew() * rotate.getTranspose();
+			return Cm::Matrix34(M[0], M[1], M[2], -M * shape2world[3]);			
+		}
+
+		//! Transforms a shape space OBB to a vertex space OBB.  All 3 params are in and out.
+		void transformQueryBounds(PxVec3& center, PxVec3& extents, PxMat33& basis)	const
+		{
+			basis.column0 = shape2VertexSkew * (basis.column0 * extents.x);
+			basis.column1 = shape2VertexSkew * (basis.column1 * extents.y);
+			basis.column2 = shape2VertexSkew * (basis.column2 * extents.z);
+
+			center = shape2VertexSkew * center;
+			extents = Ps::optimizeBoundingBox(basis);
+		}
+
+		void transformPlaneToShapeSpace(const PxVec3& nIn, const PxReal dIn, PxVec3& nOut, PxReal& dOut)	const
+		{
+			const PxVec3 tmp = shape2VertexSkew.transformTranspose(nIn);
+			const PxReal denom = 1.0f / tmp.magnitude();
+			nOut = tmp * denom;
+			dOut = dIn * denom;
+		}
+
+		PX_FORCE_INLINE bool flipsNormal() const { return mFlipNormal; }
+
+	private:
+		PxMat33 vertex2ShapeSkew;
+		PxMat33 shape2VertexSkew;
+		bool mFlipNormal;
+	};
+
+	PX_FORCE_INLINE void getScaledVertices(PxVec3* v, const PxVec3& v0, const PxVec3& v1, const PxVec3& v2, bool idtMeshScale, const Cm::FastVertex2ShapeScaling& scaling)
+	{
+		if(idtMeshScale)
+		{
+			v[0] = v0;
+			v[1] = v1;
+			v[2] = v2;
+		}
+		else
+		{
+			const PxI32 winding = scaling.flipsNormal() ? 1 : 0;
+			v[0] = scaling * v0;
+			v[1+winding] = scaling * v1;
+			v[2-winding] = scaling * v2;
+		}
+	}
+
+} // namespace Cm
+
+
+PX_INLINE Cm::Matrix34 operator*(const PxTransform& transform, const PxMeshScale& scale) 
+{
+	return Cm::Matrix34(PxMat33(transform.q) * scale.toMat33(), transform.p);
+}
+
+PX_INLINE Cm::Matrix34 operator*(const PxMeshScale& scale, const PxTransform& transform) 
+{
+	const PxMat33 scaleMat = scale.toMat33();
+	const PxMat33 t = PxMat33(transform.q);
+	const PxMat33 r = scaleMat * t;
+	const PxVec3 p = scaleMat * transform.p;
+	return Cm::Matrix34(r, p);
+}
+
+PX_INLINE Cm::Matrix34 operator*(const Cm::Matrix34& transform, const PxMeshScale& scale) 
+{
+	return Cm::Matrix34(transform.m * scale.toMat33(), transform.p);
+}
+
+PX_INLINE Cm::Matrix34 operator*(const PxMeshScale& scale, const Cm::Matrix34& transform) 
+{
+	const PxMat33 scaleMat = scale.toMat33();
+	return Cm::Matrix34(scaleMat * transform.m, scaleMat * transform.p);
+}
+
+}
+
+#endif
diff --git a/PhysX_3.4/Source/Common/src/CmSpatialVector.h b/PhysX_3.4/Source/Common/src/CmSpatialVector.h
new file mode 100644
index 00000000..4f140d4e
--- /dev/null
+++ b/PhysX_3.4/Source/Common/src/CmSpatialVector.h
@@ -0,0 +1,184 @@
+// This code contains NVIDIA Confidential Information and is disclosed to you
+// under a form of NVIDIA software license agreement provided separately to you.
+//
+// Notice
+// NVIDIA Corporation and its licensors retain all intellectual property and
+// proprietary rights in and to this software and related documentation and
+// any modifications thereto. Any use, reproduction, disclosure, or
+// distribution of this software and related documentation without an express
+// license agreement from NVIDIA Corporation is strictly prohibited.
+//
+// ALL NVIDIA DESIGN SPECIFICATIONS, CODE ARE PROVIDED "AS IS.". NVIDIA MAKES
+// NO WARRANTIES, EXPRESSED, IMPLIED, STATUTORY, OR OTHERWISE WITH RESPECT TO
+// THE MATERIALS, AND EXPRESSLY DISCLAIMS ALL IMPLIED WARRANTIES OF NONINFRINGEMENT,
+// MERCHANTABILITY, AND FITNESS FOR A PARTICULAR PURPOSE.
+//
+// Information and code furnished is believed to be accurate and reliable.
+// However, NVIDIA Corporation assumes no responsibility for the consequences of use of such
+// information or for any infringement of patents or other rights of third parties that may
+// result from its use. No license is granted by implication or otherwise under any patent
+// or patent rights of NVIDIA Corporation. Details are subject to change without notice.
+// This code supersedes and replaces all information previously supplied.
+// NVIDIA Corporation products are not authorized for use as critical
+// components in life support devices or systems without express written approval of
+// NVIDIA Corporation.
+//
+// Copyright (c) 2008-2016 NVIDIA Corporation. All rights reserved.
+// Copyright (c) 2004-2008 AGEIA Technologies, Inc. All rights reserved.
+// Copyright (c) 2001-2004 NovodeX AG. All rights reserved.  
+
+
+#ifndef PX_PHYSICS_COMMON_VECTOR
+#define PX_PHYSICS_COMMON_VECTOR
+
+#include "foundation/PxVec3.h"
+#include "CmPhysXCommon.h"
+#include "PsVecMath.h"
+
+/*!
+Combination of two R3 vectors.
+*/
+
+namespace physx
+{
+namespace Cm
+{
+PX_ALIGN_PREFIX(16)
+class SpatialVector
+{
+public:
+	//! Default constructor
+	PX_FORCE_INLINE SpatialVector()
+	{}
+
+	//! Construct from two PxcVectors
+	PX_FORCE_INLINE SpatialVector(const PxVec3& lin, const PxVec3& ang)
+		: linear(lin), pad0(0.0f), angular(ang), pad1(0.0f)
+	{
+	}
+
+	PX_FORCE_INLINE ~SpatialVector()
+	{}
+
+
+
+	// PT: this one is very important. Without it, the Xbox compiler generates weird "float-to-int" and "int-to-float" LHS
+	// each time we copy a SpatialVector (see for example PIX on "solveSimpleGroupA" without this operator).
+	PX_FORCE_INLINE	void	operator = (const SpatialVector& v)
+	{
+		linear = v.linear;
+		pad0 = 0.0f;
+		angular = v.angular;
+		pad1 = 0.0f;
+	}
+
+
+	static PX_FORCE_INLINE SpatialVector zero() {	return SpatialVector(PxVec3(0),PxVec3(0)); }
+
+	PX_FORCE_INLINE SpatialVector operator+(const SpatialVector& v) const
+	{
+		return SpatialVector(linear+v.linear,angular+v.angular);
+	}
+
+	PX_FORCE_INLINE SpatialVector operator-(const SpatialVector& v) const
+	{
+		return SpatialVector(linear-v.linear,angular-v.angular);
+	}
+
+	PX_FORCE_INLINE SpatialVector operator-() const
+	{
+		return SpatialVector(-linear,-angular);
+	}
+
+
+	PX_FORCE_INLINE SpatialVector operator *(PxReal s) const
+	{	
+		return SpatialVector(linear*s,angular*s);	
+	}
+		
+	PX_FORCE_INLINE void operator+=(const SpatialVector& v)
+	{
+		linear+=v.linear;
+		angular+=v.angular;
+	}
+
+	PX_FORCE_INLINE void operator-=(const SpatialVector& v)
+	{
+		linear-=v.linear;
+		angular-=v.angular;
+	}
+
+	PX_FORCE_INLINE PxReal magnitude()	const
+	{
+		return angular.magnitude() + linear.magnitude();
+	}
+
+	PX_FORCE_INLINE PxReal dot(const SpatialVector& v) const
+	{
+		return linear.dot(v.linear) + angular.dot(v.angular);
+	}
+		
+	PX_FORCE_INLINE bool isFinite() const
+	{
+		return linear.isFinite() && angular.isFinite();
+	}
+
+	PX_FORCE_INLINE Cm::SpatialVector scale(PxReal l, PxReal a) const
+	{
+		return Cm::SpatialVector(linear*l, angular*a);
+	}
+
+	PxVec3 linear;
+	PxReal pad0;
+	PxVec3 angular;
+	PxReal pad1;
+}
+PX_ALIGN_SUFFIX(16);
+
+
+PX_ALIGN_PREFIX(16)
+struct SpatialVectorV
+{
+	Ps::aos::Vec3V linear;
+	Ps::aos::Vec3V angular;
+
+	PX_FORCE_INLINE SpatialVectorV() {}
+	PX_FORCE_INLINE SpatialVectorV(PxZERO): linear(Ps::aos::V3Zero()), angular(Ps::aos::V3Zero()) {}
+	PX_FORCE_INLINE SpatialVectorV(const Cm::SpatialVector& v): linear(Ps::aos::V3LoadU(v.linear)), angular(Ps::aos::V3LoadU(v.angular)) {}
+	PX_FORCE_INLINE SpatialVectorV(const Ps::aos::Vec3VArg l, const Ps::aos::Vec3VArg a): linear(l), angular(a) {}
+	PX_FORCE_INLINE SpatialVectorV(const SpatialVectorV& other): linear(other.linear), angular(other.angular) {}
+	PX_FORCE_INLINE SpatialVectorV& operator=(const SpatialVectorV& other) { linear = other.linear; angular = other.angular; return *this; }
+
+	PX_FORCE_INLINE SpatialVectorV operator+(const SpatialVectorV& other) const { return SpatialVectorV(Ps::aos::V3Add(linear,other.linear),
+																								  Ps::aos::V3Add(angular, other.angular)); }
+	
+	PX_FORCE_INLINE SpatialVectorV& operator+=(const SpatialVectorV& other) { linear = Ps::aos::V3Add(linear,other.linear); 
+																			  angular = Ps::aos::V3Add(angular, other.angular);
+																			  return *this;
+																			}
+																								    
+	PX_FORCE_INLINE SpatialVectorV operator-(const SpatialVectorV& other) const { return SpatialVectorV(Ps::aos::V3Sub(linear,other.linear),
+																								  Ps::aos::V3Sub(angular, other.angular)); }
+	
+	PX_FORCE_INLINE SpatialVectorV operator-() const { return SpatialVectorV(Ps::aos::V3Neg(linear), Ps::aos::V3Neg(angular)); }
+
+	PX_FORCE_INLINE SpatialVectorV operator*(Ps::aos::FloatVArg r) const { return SpatialVectorV(Ps::aos::V3Scale(linear,r), Ps::aos::V3Scale(angular,r)); }
+
+	PX_FORCE_INLINE SpatialVectorV& operator-=(const SpatialVectorV& other) { linear = Ps::aos::V3Sub(linear,other.linear); 
+																			  angular = Ps::aos::V3Sub(angular, other.angular);
+																			  return *this;
+																			}
+
+	PX_FORCE_INLINE Ps::aos::FloatV dot(const SpatialVectorV& other) const { return Ps::aos::FAdd(Ps::aos::V3Dot(linear, other.linear), Ps::aos::V3Dot(angular, other.angular)); }
+
+	
+}PX_ALIGN_SUFFIX(16);
+
+} // namespace Cm
+
+PX_COMPILE_TIME_ASSERT(sizeof(Cm::SpatialVector) == 32);
+PX_COMPILE_TIME_ASSERT(sizeof(Cm::SpatialVectorV) == 32);
+
+}
+
+#endif
diff --git a/PhysX_3.4/Source/Common/src/CmTask.h b/PhysX_3.4/Source/Common/src/CmTask.h
new file mode 100644
index 00000000..84186b59
--- /dev/null
+++ b/PhysX_3.4/Source/Common/src/CmTask.h
@@ -0,0 +1,257 @@
+// This code contains NVIDIA Confidential Information and is disclosed to you
+// under a form of NVIDIA software license agreement provided separately to you.
+//
+// Notice
+// NVIDIA Corporation and its licensors retain all intellectual property and
+// proprietary rights in and to this software and related documentation and
+// any modifications thereto. Any use, reproduction, disclosure, or
+// distribution of this software and related documentation without an express
+// license agreement from NVIDIA Corporation is strictly prohibited.
+//
+// ALL NVIDIA DESIGN SPECIFICATIONS, CODE ARE PROVIDED "AS IS.". NVIDIA MAKES
+// NO WARRANTIES, EXPRESSED, IMPLIED, STATUTORY, OR OTHERWISE WITH RESPECT TO
+// THE MATERIALS, AND EXPRESSLY DISCLAIMS ALL IMPLIED WARRANTIES OF NONINFRINGEMENT,
+// MERCHANTABILITY, AND FITNESS FOR A PARTICULAR PURPOSE.
+//
+// Information and code furnished is believed to be accurate and reliable.
+// However, NVIDIA Corporation assumes no responsibility for the consequences of use of such
+// information or for any infringement of patents or other rights of third parties that may
+// result from its use. No license is granted by implication or otherwise under any patent
+// or patent rights of NVIDIA Corporation. Details are subject to change without notice.
+// This code supersedes and replaces all information previously supplied.
+// NVIDIA Corporation products are not authorized for use as critical
+// components in life support devices or systems without express written approval of
+// NVIDIA Corporation.
+//
+// Copyright (c) 2008-2016 NVIDIA Corporation. All rights reserved.
+// Copyright (c) 2004-2008 AGEIA Technologies, Inc. All rights reserved.
+// Copyright (c) 2001-2004 NovodeX AG. All rights reserved.  
+
+
+#ifndef PX_PHYSICS_COMMON_TASK
+#define PX_PHYSICS_COMMON_TASK
+
+#include "task/PxTask.h"
+#include "CmPhysXCommon.h"
+#include "PsUserAllocated.h"
+#include "PsAtomic.h"
+#include "PsMutex.h"
+#include "PsInlineArray.h"
+#include "PsFPU.h"
+
+namespace physx
+{
+namespace Cm
+{
+	// wrapper around the public PxLightCpuTask
+	// internal SDK tasks should be inherited from
+	// this and override the runInternal() method
+	// to ensure that the correct floating point 
+	// state is set / reset during execution
+	class Task : public physx::PxLightCpuTask
+	{
+	public:
+
+		virtual void run()
+		{
+			PX_SIMD_GUARD;
+			runInternal();
+		}
+
+		virtual void runInternal()=0;
+	};
+
+	// same as Cm::Task but inheriting from physx::PxBaseTask
+	// instead of PxLightCpuTask
+	class BaseTask : public physx::PxBaseTask
+	{
+	public:
+
+		virtual void run()
+		{
+			PX_SIMD_GUARD;
+			runInternal();
+		}
+
+		virtual void runInternal()=0;
+	};
+
+	template <class T, void (T::*Fn)(physx::PxBaseTask*) >
+	class DelegateTask : public Cm::Task, public shdfnd::UserAllocated
+	{
+	public:
+
+		DelegateTask(T* obj, const char* name) : 
+		  mObj(obj), mName(name) { }
+
+		virtual void runInternal()
+		{
+			(mObj->*Fn)(mCont);
+		}
+
+		virtual const char* getName() const
+		{
+			return mName;
+		}
+
+		void setObject(T* obj) { mObj = obj; }
+
+	private:
+		T* mObj;
+		const char* mName;
+	};
+
+
+	/**
+	\brief A task that maintains a list of dependent tasks.
+	
+	This task maintains a list of dependent tasks that have their reference counts 
+	reduced on completion of the task.
+
+	The refcount is incremented every time a dependent task is added.
+	*/
+	class FanoutTask : public Cm::BaseTask
+	{
+		PX_NOCOPY(FanoutTask)
+	public:
+		FanoutTask(const char* name) : Cm::BaseTask(), mRefCount(0), mName(name), mNotifySubmission(false) {}
+
+		virtual void runInternal() {}
+
+		virtual const char* getName() const { return mName; }
+
+		/**
+		Swap mDependents with mReferencesToRemove when refcount goes to 0.
+		*/
+		virtual void removeReference()
+		{
+			shdfnd::Mutex::ScopedLock lock(mMutex);
+			if (!physx::shdfnd::atomicDecrement(&mRefCount))
+			{
+				// prevents access to mReferencesToRemove until release
+				physx::shdfnd::atomicIncrement(&mRefCount);
+				mNotifySubmission = false;
+				PX_ASSERT(mReferencesToRemove.empty());
+				for (PxU32 i = 0; i < mDependents.size(); i++)
+					mReferencesToRemove.pushBack(mDependents[i]);
+				mDependents.clear();
+				mTm->getCpuDispatcher()->submitTask(*this);
+			}
+		}
+
+		/** 
+		\brief Increases reference count
+		*/
+		virtual void addReference()
+		{
+			shdfnd::Mutex::ScopedLock lock(mMutex);
+			physx::shdfnd::atomicIncrement(&mRefCount);
+			mNotifySubmission = true;
+		}
+
+		/** 
+		\brief Return the ref-count for this task 
+		*/
+		PX_INLINE PxI32 getReference() const
+		{
+			return mRefCount;
+		}
+
+		/**
+		Sets the task manager. Doesn't increase the reference count.
+		*/
+		PX_INLINE void setTaskManager(physx::PxTaskManager& tm)
+		{
+			mTm = &tm;
+		}
+
+		/**
+		Adds a dependent task. It also sets the task manager querying it from the dependent task.  
+		The refcount is incremented every time a dependent task is added.
+		*/
+		PX_INLINE void addDependent(physx::PxBaseTask& dependent)
+		{
+			shdfnd::Mutex::ScopedLock lock(mMutex);
+			physx::shdfnd::atomicIncrement(&mRefCount);
+			mTm = dependent.getTaskManager();
+			mDependents.pushBack(&dependent);
+			dependent.addReference();
+			mNotifySubmission = true;
+		}
+
+		/**
+		Reduces reference counts of the continuation task and the dependent tasks, also 
+		clearing the copy of continuation and dependents task list.
+		*/
+		virtual void release()
+		{
+			Ps::InlineArray<physx::PxBaseTask*, 10> referencesToRemove;
+
+			{
+				shdfnd::Mutex::ScopedLock lock(mMutex);
+
+				const PxU32 contCount = mReferencesToRemove.size(); 
+				referencesToRemove.reserve(contCount);
+				for (PxU32 i=0; i < contCount; ++i)
+					referencesToRemove.pushBack(mReferencesToRemove[i]);
+				
+				mReferencesToRemove.clear();
+				// allow access to mReferencesToRemove again
+				if (mNotifySubmission)
+				{
+					removeReference();
+				}
+				else
+				{
+					physx::shdfnd::atomicDecrement(&mRefCount);
+				}
+
+				// the scoped lock needs to get freed before the continuation tasks get (potentially) submitted because
+				// those continuation tasks might trigger events that delete this task and corrupt the memory of the
+				// mutex (for example, assume this task is a member of the scene then the submitted tasks cause the simulation 
+				// to finish and then the scene gets released which in turn will delete this task. When this task then finally
+				// continues the heap memory will be corrupted.
+			}
+
+			for (PxU32 i=0; i < referencesToRemove.size(); ++i)
+				referencesToRemove[i]->removeReference();
+		}
+
+	protected:
+		volatile PxI32 mRefCount;
+		const char* mName;
+		Ps::InlineArray<physx::PxBaseTask*, 4> mDependents;
+		Ps::InlineArray<physx::PxBaseTask*, 4> mReferencesToRemove;
+		bool mNotifySubmission;
+		Ps::Mutex mMutex; // guarding mDependents and mNotifySubmission
+	};
+
+
+	/**
+	\brief Specialization of FanoutTask class in order to provide the delegation mechanism.
+	*/
+	template <class T, void (T::*Fn)(physx::PxBaseTask*) >
+	class DelegateFanoutTask : public FanoutTask, public shdfnd::UserAllocated
+	{
+	public:
+
+		DelegateFanoutTask(T* obj, const char* name) : 
+		  FanoutTask(name), mObj(obj) { }
+
+		  virtual void runInternal()
+		  {
+			  physx::PxBaseTask* continuation = mReferencesToRemove.empty() ? NULL : mReferencesToRemove[0];
+			  (mObj->*Fn)(continuation);
+		  }
+
+		  void setObject(T* obj) { mObj = obj; }
+
+	private:
+		T* mObj;
+	};
+
+} // namespace Cm
+
+}
+
+#endif
diff --git a/PhysX_3.4/Source/Common/src/CmTaskPool.h b/PhysX_3.4/Source/Common/src/CmTaskPool.h
new file mode 100644
index 00000000..34e6f086
--- /dev/null
+++ b/PhysX_3.4/Source/Common/src/CmTaskPool.h
@@ -0,0 +1,143 @@
+// This code contains NVIDIA Confidential Information and is disclosed to you
+// under a form of NVIDIA software license agreement provided separately to you.
+//
+// Notice
+// NVIDIA Corporation and its licensors retain all intellectual property and
+// proprietary rights in and to this software and related documentation and
+// any modifications thereto. Any use, reproduction, disclosure, or
+// distribution of this software and related documentation without an express
+// license agreement from NVIDIA Corporation is strictly prohibited.
+//
+// ALL NVIDIA DESIGN SPECIFICATIONS, CODE ARE PROVIDED "AS IS.". NVIDIA MAKES
+// NO WARRANTIES, EXPRESSED, IMPLIED, STATUTORY, OR OTHERWISE WITH RESPECT TO
+// THE MATERIALS, AND EXPRESSLY DISCLAIMS ALL IMPLIED WARRANTIES OF NONINFRINGEMENT,
+// MERCHANTABILITY, AND FITNESS FOR A PARTICULAR PURPOSE.
+//
+// Information and code furnished is believed to be accurate and reliable.
+// However, NVIDIA Corporation assumes no responsibility for the consequences of use of such
+// information or for any infringement of patents or other rights of third parties that may
+// result from its use. No license is granted by implication or otherwise under any patent
+// or patent rights of NVIDIA Corporation. Details are subject to change without notice.
+// This code supersedes and replaces all information previously supplied.
+// NVIDIA Corporation products are not authorized for use as critical
+// components in life support devices or systems without express written approval of
+// NVIDIA Corporation.
+//
+// Copyright (c) 2008-2016 NVIDIA Corporation. All rights reserved.
+// Copyright (c) 2004-2008 AGEIA Technologies, Inc. All rights reserved.
+// Copyright (c) 2001-2004 NovodeX AG. All rights reserved.  
+
+
+#ifndef PX_PHYSICS_COMMON_TASKPOOL
+#define PX_PHYSICS_COMMON_TASKPOOL
+
+#include "foundation/Px.h"
+#include "PsMutex.h"
+#include "PsSList.h"
+#include "PsAllocator.h"
+#include "PsArray.h"
+
+class PxTask;
+
+/*
+Implimentation of a thread safe task pool. (PxTask derived classes).
+
+T is the actual type of the task(currently NphaseTask or GroupSolveTask).
+*/
+
+namespace Cm
+{
+	template<class T> class TaskPool : public Ps::AlignedAllocator<16>
+	{
+		const static PxU32 TaskPoolSlabSize=64;
+
+	public:
+
+		typedef Ps::SListEntry TaskPoolItem;
+
+		PX_INLINE TaskPool() : slabArray(PX_DEBUG_EXP("taskPoolSlabArray"))
+		{
+			//we have to ensure that the list header is 16byte aligned for win64.
+			freeTasks = (Ps::SList*)allocate(sizeof(Ps::SList), __FILE__, __LINE__);
+			PX_PLACEMENT_NEW(freeTasks, Ps::SList)();
+
+			slabArray.reserve(16);
+		}
+
+		~TaskPool()
+		{
+			Ps::Mutex::ScopedLock lock(slabAllocMutex);
+
+			freeTasks->flush();
+
+			for(PxU32 i=0;i<slabArray.size();i++)
+			{
+				// call destructors
+				for(PxU32 j=0; j<TaskPoolSlabSize; j++)
+					slabArray[i][j].~T();
+
+				deallocate(slabArray[i]);
+			}
+
+			slabArray.clear();
+
+			if(freeTasks!=NULL)
+			{
+				freeTasks->~SList();
+				deallocate(freeTasks);
+				freeTasks = NULL;
+			}
+		}
+
+		T *allocTask()
+		{
+			T *rv = static_cast<T *>(freeTasks->pop());
+			if(rv == NULL)
+				return static_cast<T *>(allocateSlab());
+			else
+				return rv;
+		}
+		void freeTask(T *task)
+		{
+			freeTasks->push(*task);
+		}
+
+	private:
+
+		T *allocateSlab()
+		{
+			//ack, convoluted memory macros.
+
+			//T *newSlab=new T[TaskPoolSlabSize];
+
+			// we must align this memory.
+			T *newSlab=(T *)allocate(sizeof(T)*TaskPoolSlabSize, __FILE__, __LINE__);
+
+			new (newSlab) T();
+
+			//we keep one for the caller
+			// and build a list of tasks and insert in the free list
+			for(PxU32 i=1;i<TaskPoolSlabSize;i++)
+			{
+				new (&(newSlab[i])) T();
+				freeTasks->push(newSlab[i]);
+			}
+
+			Ps::Mutex::ScopedLock lock(slabAllocMutex);
+			slabArray.pushBack(newSlab);
+
+			return newSlab;
+		}
+
+		Ps::Mutex slabAllocMutex;
+		Ps::Array<T *> slabArray;
+
+		Ps::SList *freeTasks;
+
+	};
+
+
+} // namespace Cm
+
+
+#endif
diff --git a/PhysX_3.4/Source/Common/src/CmTmpMem.h b/PhysX_3.4/Source/Common/src/CmTmpMem.h
new file mode 100644
index 00000000..998590b9
--- /dev/null
+++ b/PhysX_3.4/Source/Common/src/CmTmpMem.h
@@ -0,0 +1,94 @@
+// This code contains NVIDIA Confidential Information and is disclosed to you
+// under a form of NVIDIA software license agreement provided separately to you.
+//
+// Notice
+// NVIDIA Corporation and its licensors retain all intellectual property and
+// proprietary rights in and to this software and related documentation and
+// any modifications thereto. Any use, reproduction, disclosure, or
+// distribution of this software and related documentation without an express
+// license agreement from NVIDIA Corporation is strictly prohibited.
+//
+// ALL NVIDIA DESIGN SPECIFICATIONS, CODE ARE PROVIDED "AS IS.". NVIDIA MAKES
+// NO WARRANTIES, EXPRESSED, IMPLIED, STATUTORY, OR OTHERWISE WITH RESPECT TO
+// THE MATERIALS, AND EXPRESSLY DISCLAIMS ALL IMPLIED WARRANTIES OF NONINFRINGEMENT,
+// MERCHANTABILITY, AND FITNESS FOR A PARTICULAR PURPOSE.
+//
+// Information and code furnished is believed to be accurate and reliable.
+// However, NVIDIA Corporation assumes no responsibility for the consequences of use of such
+// information or for any infringement of patents or other rights of third parties that may
+// result from its use. No license is granted by implication or otherwise under any patent
+// or patent rights of NVIDIA Corporation. Details are subject to change without notice.
+// This code supersedes and replaces all information previously supplied.
+// NVIDIA Corporation products are not authorized for use as critical
+// components in life support devices or systems without express written approval of
+// NVIDIA Corporation.
+//
+// Copyright (c) 2008-2016 NVIDIA Corporation. All rights reserved.
+// Copyright (c) 2004-2008 AGEIA Technologies, Inc. All rights reserved.
+// Copyright (c) 2001-2004 NovodeX AG. All rights reserved.  
+
+
+#ifndef PX_PHYSICS_COMMON_TMPMEM
+#define PX_PHYSICS_COMMON_TMPMEM
+
+#include "CmPhysXCommon.h"
+#include "PsAllocator.h"
+
+namespace physx
+{
+namespace Cm
+{
+	// dsequeira: we should be able to use PX_ALLOCA or Ps::InlineArray for this, but both seem slightly flawed:
+	//
+	//	PX_ALLOCA has non-configurable fallback threshold and uses _alloca, which means the allocation is necessarily
+	//	function-scope rather than block-scope (sometimes useful, mostly not.)
+	//
+	//	Ps::InlineArray touches all memory on resize (a general flaw in the array class which badly needs fixing)
+	//
+	//	Todo: fix both the above issues.
+
+	template<typename T, PxU32 stackLimit>
+	class TmpMem
+	{
+
+	public:
+		PX_FORCE_INLINE TmpMem(PxU32 size):
+		  mPtr(size<=stackLimit?mStackBuf : reinterpret_cast<T*>(PX_ALLOC(size*sizeof(T), "char")))
+		{
+		}
+
+		PX_FORCE_INLINE ~TmpMem()
+		{
+			if(mPtr!=mStackBuf)
+				PX_FREE(mPtr);
+		}
+
+		PX_FORCE_INLINE T& operator*() const
+	    {
+			return *mPtr;
+		}
+
+		PX_FORCE_INLINE T* operator->() const
+		{
+			return mPtr;
+		}
+	
+		PX_FORCE_INLINE T& operator[](PxU32 index)
+		{
+			return mPtr[index];
+		}
+
+		T* getBase()
+		{
+			return mPtr;
+		}
+
+	private:
+		T mStackBuf[stackLimit];
+		T* mPtr;
+	};
+}
+
+}
+
+#endif
diff --git a/PhysX_3.4/Source/Common/src/CmTransformUtils.h b/PhysX_3.4/Source/Common/src/CmTransformUtils.h
new file mode 100644
index 00000000..ddb84b1a
--- /dev/null
+++ b/PhysX_3.4/Source/Common/src/CmTransformUtils.h
@@ -0,0 +1,145 @@
+// This code contains NVIDIA Confidential Information and is disclosed to you
+// under a form of NVIDIA software license agreement provided separately to you.
+//
+// Notice
+// NVIDIA Corporation and its licensors retain all intellectual property and
+// proprietary rights in and to this software and related documentation and
+// any modifications thereto. Any use, reproduction, disclosure, or
+// distribution of this software and related documentation without an express
+// license agreement from NVIDIA Corporation is strictly prohibited.
+//
+// ALL NVIDIA DESIGN SPECIFICATIONS, CODE ARE PROVIDED "AS IS.". NVIDIA MAKES
+// NO WARRANTIES, EXPRESSED, IMPLIED, STATUTORY, OR OTHERWISE WITH RESPECT TO
+// THE MATERIALS, AND EXPRESSLY DISCLAIMS ALL IMPLIED WARRANTIES OF NONINFRINGEMENT,
+// MERCHANTABILITY, AND FITNESS FOR A PARTICULAR PURPOSE.
+//
+// Information and code furnished is believed to be accurate and reliable.
+// However, NVIDIA Corporation assumes no responsibility for the consequences of use of such
+// information or for any infringement of patents or other rights of third parties that may
+// result from its use. No license is granted by implication or otherwise under any patent
+// or patent rights of NVIDIA Corporation. Details are subject to change without notice.
+// This code supersedes and replaces all information previously supplied.
+// NVIDIA Corporation products are not authorized for use as critical
+// components in life support devices or systems without express written approval of
+// NVIDIA Corporation.
+//
+// Copyright (c) 2008-2016 NVIDIA Corporation. All rights reserved.
+// Copyright (c) 2004-2008 AGEIA Technologies, Inc. All rights reserved.
+// Copyright (c) 2001-2004 NovodeX AG. All rights reserved.  
+
+
+#ifndef PX_PHYSICS_COMMON_TRANSFORMUTILS
+#define PX_PHYSICS_COMMON_TRANSFORMUTILS
+
+#include "PsVecMath.h"
+
+namespace
+{
+
+using namespace physx::shdfnd::aos;
+
+// V3PrepareCross would help here, but it's not on all platforms yet...
+
+PX_FORCE_INLINE void transformFast(const FloatVArg wa, const Vec3VArg va, const Vec3VArg pa, 
+							   const FloatVArg wb, const Vec3VArg vb, const Vec3VArg pb, 
+							   FloatV& wo, Vec3V& vo, Vec3V& po)
+{
+	wo = FSub(FMul(wa, wb), V3Dot(va, vb));
+	vo = V3ScaleAdd(va, wb, V3ScaleAdd(vb, wa, V3Cross(va, vb)));
+
+	const Vec3V t1 = V3Scale(pb, FScaleAdd(wa, wa, FLoad(-0.5f)));
+	const Vec3V t2 = V3ScaleAdd(V3Cross(va, pb), wa, t1);
+	const Vec3V t3 = V3ScaleAdd(va, V3Dot(va, pb), t2);
+
+	po = V3ScaleAdd(t3, FLoad(2.f), pa);
+}
+
+PX_FORCE_INLINE void transformInvFast(const FloatVArg wa, const Vec3VArg va, const Vec3VArg pa, 
+								      const FloatVArg wb, const Vec3VArg vb, const Vec3VArg pb, 
+								      FloatV& wo, Vec3V& vo, Vec3V& po)
+{
+	wo = FScaleAdd(wa, wb, V3Dot(va, vb));
+	vo = V3NegScaleSub(va, wb, V3ScaleAdd(vb, wa, V3Cross(vb, va)));
+
+	const Vec3V pt = V3Sub(pb, pa);
+	const Vec3V t1 = V3Scale(pt, FScaleAdd(wa, wa, FLoad(-0.5f)));
+	const Vec3V t2 = V3ScaleAdd(V3Cross(pt, va), wa, t1);
+	const Vec3V t3 = V3ScaleAdd(va, V3Dot(va, pt), t2);
+	po = V3Add(t3,t3);
+}
+
+}
+
+
+
+
+
+namespace physx
+{
+namespace Cm
+{
+
+PX_FORCE_INLINE void getStaticGlobalPoseAligned(const PxTransform& actor2World, const PxTransform& shape2Actor, PxTransform& outTransform)
+{
+	using namespace shdfnd::aos;
+
+	PX_ASSERT((size_t(&actor2World)&15) == 0);
+	PX_ASSERT((size_t(&shape2Actor)&15) == 0);
+	PX_ASSERT((size_t(&outTransform)&15) == 0);
+
+	const Vec3V actor2WorldPos = V3LoadA(actor2World.p);
+	const QuatV actor2WorldRot = QuatVLoadA(&actor2World.q.x);
+
+	const Vec3V shape2ActorPos = V3LoadA(shape2Actor.p);
+	const QuatV shape2ActorRot = QuatVLoadA(&shape2Actor.q.x);
+	
+	Vec3V v,p;
+	FloatV w; 
+
+	transformFast(V4GetW(actor2WorldRot), Vec3V_From_Vec4V(actor2WorldRot), actor2WorldPos,
+				  V4GetW(shape2ActorRot), Vec3V_From_Vec4V(shape2ActorRot), shape2ActorPos,
+				  w, v, p);
+
+	V3StoreA(p, outTransform.p);
+	V4StoreA(V4SetW(v,w), &outTransform.q.x);
+}
+
+
+PX_FORCE_INLINE void getDynamicGlobalPoseAligned(const PxTransform& body2World, const PxTransform& shape2Actor, const PxTransform& body2Actor, PxTransform& outTransform)
+{
+	PX_ASSERT((size_t(&body2World)&15) == 0);
+	PX_ASSERT((size_t(&shape2Actor)&15) == 0);
+	PX_ASSERT((size_t(&body2Actor)&15) == 0);
+	PX_ASSERT((size_t(&outTransform)&15) == 0);
+
+	using namespace shdfnd::aos;
+
+	const Vec3V shape2ActorPos = V3LoadA(shape2Actor.p);
+	const QuatV shape2ActorRot = QuatVLoadA(&shape2Actor.q.x);
+
+	const Vec3V body2ActorPos = V3LoadA(body2Actor.p);
+	const QuatV body2ActorRot = QuatVLoadA(&body2Actor.q.x);
+
+	const Vec3V body2WorldPos = V3LoadA(body2World.p);
+	const QuatV body2WorldRot = QuatVLoadA(&body2World.q.x);
+
+	Vec3V v1, p1, v2, p2;
+	FloatV w1, w2; 
+
+	transformInvFast(V4GetW(body2ActorRot), Vec3V_From_Vec4V(body2ActorRot), body2ActorPos,
+					 V4GetW(shape2ActorRot), Vec3V_From_Vec4V(shape2ActorRot), shape2ActorPos,
+					 w1, v1, p1);
+
+	transformFast(V4GetW(body2WorldRot), Vec3V_From_Vec4V(body2WorldRot), body2WorldPos,
+			      w1, v1, p1,
+			      w2, v2, p2);
+
+	V3StoreA(p2, outTransform.p);
+	V4StoreA(V4SetW(v2, w2), &outTransform.q.x);
+}
+
+
+}
+}
+
+#endif
diff --git a/PhysX_3.4/Source/Common/src/CmUtils.h b/PhysX_3.4/Source/Common/src/CmUtils.h
new file mode 100644
index 00000000..ad911fd7
--- /dev/null
+++ b/PhysX_3.4/Source/Common/src/CmUtils.h
@@ -0,0 +1,293 @@
+// This code contains NVIDIA Confidential Information and is disclosed to you
+// under a form of NVIDIA software license agreement provided separately to you.
+//
+// Notice
+// NVIDIA Corporation and its licensors retain all intellectual property and
+// proprietary rights in and to this software and related documentation and
+// any modifications thereto. Any use, reproduction, disclosure, or
+// distribution of this software and related documentation without an express
+// license agreement from NVIDIA Corporation is strictly prohibited.
+//
+// ALL NVIDIA DESIGN SPECIFICATIONS, CODE ARE PROVIDED "AS IS.". NVIDIA MAKES
+// NO WARRANTIES, EXPRESSED, IMPLIED, STATUTORY, OR OTHERWISE WITH RESPECT TO
+// THE MATERIALS, AND EXPRESSLY DISCLAIMS ALL IMPLIED WARRANTIES OF NONINFRINGEMENT,
+// MERCHANTABILITY, AND FITNESS FOR A PARTICULAR PURPOSE.
+//
+// Information and code furnished is believed to be accurate and reliable.
+// However, NVIDIA Corporation assumes no responsibility for the consequences of use of such
+// information or for any infringement of patents or other rights of third parties that may
+// result from its use. No license is granted by implication or otherwise under any patent
+// or patent rights of NVIDIA Corporation. Details are subject to change without notice.
+// This code supersedes and replaces all information previously supplied.
+// NVIDIA Corporation products are not authorized for use as critical
+// components in life support devices or systems without express written approval of
+// NVIDIA Corporation.
+//
+// Copyright (c) 2008-2016 NVIDIA Corporation. All rights reserved.
+// Copyright (c) 2004-2008 AGEIA Technologies, Inc. All rights reserved.
+// Copyright (c) 2001-2004 NovodeX AG. All rights reserved.  
+
+
+#ifndef PX_PHYSICS_COMMON_UTILS
+#define PX_PHYSICS_COMMON_UTILS
+
+
+#include "foundation/PxVec3.h"
+#include "foundation/PxMat33.h"
+#include "foundation/PxBounds3.h"
+#include "CmPhysXCommon.h"
+#include "PxBase.h"
+#include "PsInlineArray.h"
+#include "PsArray.h"
+#include "PsAllocator.h"
+
+namespace physx
+{
+namespace Cm
+{
+
+template<class DstType, class SrcType>
+PX_FORCE_INLINE PxU32 getArrayOfPointers(DstType** PX_RESTRICT userBuffer, PxU32 bufferSize, PxU32 startIndex, SrcType*const* PX_RESTRICT src, PxU32 size)
+{
+	const PxU32 remainder = PxU32(PxMax<PxI32>(PxI32(size - startIndex), 0));
+	const PxU32 writeCount = PxMin(remainder, bufferSize);
+	src += startIndex;
+	for(PxU32 i=0;i<writeCount;i++)
+		userBuffer[i] = static_cast<DstType*>(src[i]);
+	return writeCount;
+}
+
+PX_INLINE void transformInertiaTensor(const PxVec3& invD, const PxMat33& M, PxMat33& mIInv)
+{
+	const float	axx = invD.x*M(0,0), axy = invD.x*M(1,0), axz = invD.x*M(2,0);
+	const float	byx = invD.y*M(0,1), byy = invD.y*M(1,1), byz = invD.y*M(2,1);
+	const float	czx = invD.z*M(0,2), czy = invD.z*M(1,2), czz = invD.z*M(2,2);
+
+	mIInv(0,0) = axx*M(0,0) + byx*M(0,1) + czx*M(0,2);
+	mIInv(1,1) = axy*M(1,0) + byy*M(1,1) + czy*M(1,2);
+	mIInv(2,2) = axz*M(2,0) + byz*M(2,1) + czz*M(2,2);
+
+	mIInv(0,1) = mIInv(1,0)	= axx*M(1,0) + byx*M(1,1) + czx*M(1,2);
+	mIInv(0,2) = mIInv(2,0)	= axx*M(2,0) + byx*M(2,1) + czx*M(2,2);
+	mIInv(1,2) = mIInv(2,1)	= axy*M(2,0) + byy*M(2,1) + czy*M(2,2);
+}
+
+// PT: TODO: refactor this with PxBounds3 header
+PX_FORCE_INLINE PxVec3 basisExtent(const PxVec3& basis0, const PxVec3& basis1, const PxVec3& basis2, const PxVec3& extent)
+{
+	// extended basis vectors
+	const PxVec3 c0 = basis0 * extent.x;
+	const PxVec3 c1 = basis1 * extent.y;
+	const PxVec3 c2 = basis2 * extent.z;
+
+	// find combination of base vectors that produces max. distance for each component = sum of abs()
+	return PxVec3 (	PxAbs(c0.x) + PxAbs(c1.x) + PxAbs(c2.x),
+					PxAbs(c0.y) + PxAbs(c1.y) + PxAbs(c2.y),
+					PxAbs(c0.z) + PxAbs(c1.z) + PxAbs(c2.z));
+}
+
+PX_FORCE_INLINE PxBounds3 basisExtent(const PxVec3& center, const PxVec3& basis0, const PxVec3& basis1, const PxVec3& basis2, const PxVec3& extent)
+{
+	const PxVec3 w = basisExtent(basis0, basis1, basis2, extent);
+	return PxBounds3(center - w, center + w);
+}
+
+PX_FORCE_INLINE	bool isValid(const PxVec3& c, const PxVec3& e)
+{
+	return (c.isFinite() && e.isFinite() && (((e.x >= 0.0f) && (e.y >= 0.0f) && (e.z >= 0.0f)) ||
+											((e.x == -PX_MAX_BOUNDS_EXTENTS) &&
+											(e.y == -PX_MAX_BOUNDS_EXTENTS) &&
+											(e.z == -PX_MAX_BOUNDS_EXTENTS))));
+}
+
+PX_FORCE_INLINE	bool isEmpty(const PxVec3& c, const PxVec3& e)
+{
+	PX_UNUSED(c);
+	PX_ASSERT(isValid(c, e));
+	return e.x<0.0f;
+}
+
+// Array with externally managed storage.
+// Allocation and resize policy are managed by the owner, 
+// Very minimal functionality right now, just POD types
+
+template <typename T, 
+		  typename Owner, 
+		  typename IndexType,
+		  void (Owner::*realloc)(T*& currentMem, IndexType& currentCapacity, IndexType size, IndexType requiredMinCapacity)>
+class OwnedArray
+{
+public:
+	OwnedArray() 
+	: mData(0)
+	, mCapacity(0) 
+	, mSize(0)
+	{}
+
+	~OwnedArray()		// owner must call releaseMem before destruction
+	{
+		PX_ASSERT(mCapacity==0);
+	}
+
+	void pushBack(T& element, Owner& owner)
+	{
+		// there's a failure case if here if we push an existing element which causes a resize -
+		// a rare case not worth coding around; if you need it, copy the element then push it.
+
+		PX_ASSERT(&element<mData || &element>=mData+mSize);
+		if(mSize==mCapacity)
+			(owner.*realloc)(mData, mCapacity, mSize, PxU16(mSize+1));
+
+		PX_ASSERT(mData && mSize<mCapacity);
+		mData[mSize++] = element;
+	}
+
+	IndexType size() const
+	{
+		return mSize;
+	}
+
+	void replaceWithLast(IndexType index)
+	{
+		PX_ASSERT(index<mSize);
+		mData[index] = mData[--mSize];
+	}
+
+	T* begin() const
+	{
+		return mData;
+	}
+
+	T* end() const
+	{
+		return mData+mSize;
+	}
+
+	T& operator [](IndexType index)
+	{
+		PX_ASSERT(index<mSize);
+		return mData[index];
+	}
+
+	const T& operator [](IndexType index) const
+	{
+		PX_ASSERT(index<mSize);
+		return mData[index];
+	}
+
+	void reserve(IndexType capacity, Owner &owner)
+	{
+		if(capacity>=mCapacity)
+			(owner.*realloc)(mData, mCapacity, mSize, capacity);
+	}
+
+	void releaseMem(Owner &owner)
+	{
+		mSize = 0;
+		(owner.*realloc)(mData, mCapacity, 0, 0);
+	}
+
+private:
+	T*					mData;
+	IndexType			mCapacity;
+	IndexType			mSize;
+
+	// just in case someone tries to use a non-POD in here
+	union FailIfNonPod
+	{
+		T t;
+		int x;
+	};
+};
+
+/**
+Any object deriving from PxBase needs to call this function instead of 'delete object;'. 
+
+We don't want implement 'operator delete' in PxBase because that would impose how
+memory of derived classes is allocated. Even though most or all of the time derived classes will 
+be user allocated, we don't want to put UserAllocatable into the API and derive from that.
+*/
+template<typename T>
+PX_INLINE void deletePxBase(T* object)
+{
+	if(object->getBaseFlags() & PxBaseFlag::eOWNS_MEMORY)
+		PX_DELETE(object);
+	else
+		object->~T();
+}
+
+#if PX_CHECKED
+/**
+Mark a specified amount of memory with 0xcd pattern. This is used to check that the meta data 
+definition for serialized classes is complete in checked builds.
+*/
+PX_INLINE void markSerializedMem(void* ptr, PxU32 byteSize)
+{
+	for (PxU32 i = 0; i < byteSize; ++i)
+      reinterpret_cast<PxU8*>(ptr)[i] = 0xcd;
+}
+
+/**
+Macro to instantiate a type for serialization testing. 
+Note: Only use PX_NEW_SERIALIZED once in a scope.
+*/
+#define PX_NEW_SERIALIZED(v,T) 															        \
+    void* _buf = physx::shdfnd::ReflectionAllocator<T>().allocate(sizeof(T),__FILE__,__LINE__);  \
+	Cm::markSerializedMem(_buf, sizeof(T));                                                      \
+    v = PX_PLACEMENT_NEW(_buf, T)
+
+#else
+PX_INLINE void markSerializedMem(void*, PxU32){}
+
+#define PX_NEW_SERIALIZED(v,T)  v = PX_NEW(T)
+#endif
+
+template<typename T, class Alloc>
+struct ArrayAccess: public Ps::Array<T, Alloc> 
+{
+	void store(PxSerializationContext& context) const
+	{
+		if(this->mData && (this->mSize || this->capacity()))
+			context.writeData(this->mData, this->capacity()*sizeof(T));
+	}
+
+	void load(PxDeserializationContext& context)
+	{
+		if(this->mData && (this->mSize || this->capacity()))
+			this->mData = context.readExtraData<T>(this->capacity());
+	}
+};
+
+template<typename T, typename Alloc>
+void exportArray(const Ps::Array<T, Alloc>& a, PxSerializationContext& context)
+{
+	static_cast<const ArrayAccess<T, Alloc>&>(a).store(context);
+}
+
+template<typename T, typename Alloc>
+void importArray(Ps::Array<T, Alloc>& a, PxDeserializationContext& context)
+{
+	static_cast<ArrayAccess<T, Alloc>&>(a).load(context);
+}
+
+template<typename T, PxU32 N, typename Alloc>
+void exportInlineArray(const Ps::InlineArray<T, N, Alloc>& a, PxSerializationContext& context)
+{
+	if(!a.isInlined())
+		Cm::exportArray(a, context);
+}
+
+template<typename T, PxU32 N, typename Alloc>
+void importInlineArray(Ps::InlineArray<T, N, Alloc>& a, PxDeserializationContext& context)
+{
+	if(!a.isInlined())
+		Cm::importArray(a, context);
+}
+
+} // namespace Cm
+
+
+
+}
+
+#endif
diff --git a/PhysX_3.4/Source/Common/src/CmVisualization.cpp b/PhysX_3.4/Source/Common/src/CmVisualization.cpp
new file mode 100644
index 00000000..3fdfec85
--- /dev/null
+++ b/PhysX_3.4/Source/Common/src/CmVisualization.cpp
@@ -0,0 +1,159 @@
+// This code contains NVIDIA Confidential Information and is disclosed to you
+// under a form of NVIDIA software license agreement provided separately to you.
+//
+// Notice
+// NVIDIA Corporation and its licensors retain all intellectual property and
+// proprietary rights in and to this software and related documentation and
+// any modifications thereto. Any use, reproduction, disclosure, or
+// distribution of this software and related documentation without an express
+// license agreement from NVIDIA Corporation is strictly prohibited.
+//
+// ALL NVIDIA DESIGN SPECIFICATIONS, CODE ARE PROVIDED "AS IS.". NVIDIA MAKES
+// NO WARRANTIES, EXPRESSED, IMPLIED, STATUTORY, OR OTHERWISE WITH RESPECT TO
+// THE MATERIALS, AND EXPRESSLY DISCLAIMS ALL IMPLIED WARRANTIES OF NONINFRINGEMENT,
+// MERCHANTABILITY, AND FITNESS FOR A PARTICULAR PURPOSE.
+//
+// Information and code furnished is believed to be accurate and reliable.
+// However, NVIDIA Corporation assumes no responsibility for the consequences of use of such
+// information or for any infringement of patents or other rights of third parties that may
+// result from its use. No license is granted by implication or otherwise under any patent
+// or patent rights of NVIDIA Corporation. Details are subject to change without notice.
+// This code supersedes and replaces all information previously supplied.
+// NVIDIA Corporation products are not authorized for use as critical
+// components in life support devices or systems without express written approval of
+// NVIDIA Corporation.
+//
+// Copyright (c) 2008-2016 NVIDIA Corporation. All rights reserved.
+// Copyright (c) 2004-2008 AGEIA Technologies, Inc. All rights reserved.
+// Copyright (c) 2001-2004 NovodeX AG. All rights reserved.  
+
+#include "foundation/PxTransform.h"
+#include "CmPhysXCommon.h"
+#include "CmRenderOutput.h"
+#include "CmVisualization.h"
+
+using namespace physx;
+using namespace Cm;
+
+void Cm::visualizeJointFrames(RenderOutput& out,
+							  PxReal scale,
+							  const PxTransform& parent,
+							  const PxTransform& child)
+{
+	if(scale==0.0f)
+		return;
+
+	out << parent << Cm::DebugBasis(PxVec3(scale, scale, scale) * 1.5f,
+		PxU32(PxDebugColor::eARGB_DARKRED), PxU32(PxDebugColor::eARGB_DARKGREEN), PxU32(PxDebugColor::eARGB_DARKBLUE));
+	out << child << Cm::DebugBasis(PxVec3(scale, scale, scale));	
+}
+
+void Cm::visualizeLinearLimit(RenderOutput& out,
+							  PxReal scale,
+							  const PxTransform& t0,
+							  const PxTransform& /*t1*/,
+							  PxReal value,
+							  bool active)
+{
+	if(scale==0.0f)
+		return;
+
+	// debug circle is around z-axis, and we want it around x-axis
+	PxTransform r(t0.p+value*t0.q.getBasisVector0(), t0.q*PxQuat(PxPi/2,PxVec3(0,1.f,0)));
+	out << (active ? PxDebugColor::eARGB_RED : PxDebugColor::eARGB_GREY);
+	out << PxTransform(PxIdentity);
+	out << Cm::DebugArrow(t0.p,r.p);
+
+	out << r << Cm::DebugCircle(20, scale*0.3f);
+}
+
+void Cm::visualizeAngularLimit(RenderOutput& out,
+							   PxReal scale,
+							   const PxTransform& t,
+							   PxReal lower,
+							   PxReal upper,
+							   bool active)
+{
+	if(scale==0.0f)
+		return;
+
+	out << t << (active ? PxDebugColor::eARGB_RED : PxDebugColor::eARGB_GREY);
+	
+	out << Cm::RenderOutput::LINES 
+		<< PxVec3(0) << PxVec3(0, PxCos(lower), PxSin(lower)) * scale
+		<< PxVec3(0) << PxVec3(0, PxCos(upper), PxSin(upper)) * scale;
+
+	out << Cm::RenderOutput::LINESTRIP;
+	PxReal angle = lower, step = (upper-lower)/20;
+
+	for(PxU32 i=0; i<=20; i++, angle += step)
+		out << PxVec3(0, PxCos(angle), PxSin(angle)) * scale;
+}
+
+void Cm::visualizeLimitCone(RenderOutput& out,
+							PxReal scale,
+							const PxTransform& t,
+							PxReal tanQSwingY,
+							PxReal tanQSwingZ,
+							bool active)
+{
+	if(scale==0.0f)
+		return;
+
+	out << t << (active ? PxDebugColor::eARGB_RED : PxDebugColor::eARGB_GREY);	
+	out << Cm::RenderOutput::LINES;
+
+	PxVec3 prev(0,0,0);
+	
+	const int LINES = 32;
+
+	for(PxU32 i=0;i<=32;i++)
+	{
+		PxReal angle = 2*PxPi/LINES*i;
+		PxReal c = PxCos(angle), s = PxSin(angle);
+		PxVec3 rv(0,-tanQSwingZ*s, tanQSwingY*c);
+		PxReal rv2 = rv.magnitudeSquared();
+		PxQuat q = PxQuat(0,2*rv.y,2*rv.z,1-rv2) * (1/(1+rv2));
+		PxVec3 a = q.rotate(PxVec3(1.0f,0,0)) * scale;
+
+		out << prev << a << PxVec3(0) << a;
+		prev = a;
+	}
+}
+
+void Cm::visualizeDoubleCone(Cm::RenderOutput& out,
+							 PxReal scale,
+							 const PxTransform& t,
+							 PxReal angle,
+							 bool active)
+{
+	if(scale==0.0f)
+		return;
+
+	out << t << (active ? PxDebugColor::eARGB_RED : PxDebugColor::eARGB_GREY);	
+
+	const PxReal height = PxSin(angle);//, radius = cos(angle);
+
+	const PxU32 LINES = 32;
+
+	out << Cm::RenderOutput::LINESTRIP;
+
+	const PxReal step = PxPi*2/LINES;
+
+	for(PxU32 i=0; i<=LINES; i++)
+		out << PxVec3(height, PxCos(step * i), PxSin(step * i)) * scale;
+
+	angle = 0;
+	out << Cm::RenderOutput::LINESTRIP;
+	for(PxU32 i=0; i<=LINES; i++, angle += PxPi*2/LINES)
+		out << PxVec3(-height, PxCos(step * i), PxSin(step * i)) * scale;
+
+	angle = 0;
+	out << Cm::RenderOutput::LINES;
+	for(PxU32 i=0;i<LINES;i++, angle += PxPi*2/LINES)
+	{
+		out << PxVec3(0) << PxVec3(-height, PxCos(step * i), PxSin(step * i)) * scale;
+		out << PxVec3(0) << PxVec3(height, PxCos(step * i), PxSin(step * i)) * scale;
+	}
+}
+
diff --git a/PhysX_3.4/Source/Common/src/CmVisualization.h b/PhysX_3.4/Source/Common/src/CmVisualization.h
new file mode 100644
index 00000000..ecedfb17
--- /dev/null
+++ b/PhysX_3.4/Source/Common/src/CmVisualization.h
@@ -0,0 +1,125 @@
+// This code contains NVIDIA Confidential Information and is disclosed to you
+// under a form of NVIDIA software license agreement provided separately to you.
+//
+// Notice
+// NVIDIA Corporation and its licensors retain all intellectual property and
+// proprietary rights in and to this software and related documentation and
+// any modifications thereto. Any use, reproduction, disclosure, or
+// distribution of this software and related documentation without an express
+// license agreement from NVIDIA Corporation is strictly prohibited.
+//
+// ALL NVIDIA DESIGN SPECIFICATIONS, CODE ARE PROVIDED "AS IS.". NVIDIA MAKES
+// NO WARRANTIES, EXPRESSED, IMPLIED, STATUTORY, OR OTHERWISE WITH RESPECT TO
+// THE MATERIALS, AND EXPRESSLY DISCLAIMS ALL IMPLIED WARRANTIES OF NONINFRINGEMENT,
+// MERCHANTABILITY, AND FITNESS FOR A PARTICULAR PURPOSE.
+//
+// Information and code furnished is believed to be accurate and reliable.
+// However, NVIDIA Corporation assumes no responsibility for the consequences of use of such
+// information or for any infringement of patents or other rights of third parties that may
+// result from its use. No license is granted by implication or otherwise under any patent
+// or patent rights of NVIDIA Corporation. Details are subject to change without notice.
+// This code supersedes and replaces all information previously supplied.
+// NVIDIA Corporation products are not authorized for use as critical
+// components in life support devices or systems without express written approval of
+// NVIDIA Corporation.
+//
+// Copyright (c) 2008-2016 NVIDIA Corporation. All rights reserved.
+// Copyright (c) 2004-2008 AGEIA Technologies, Inc. All rights reserved.
+// Copyright (c) 2001-2004 NovodeX AG. All rights reserved.  
+
+
+#ifndef PX_PHYSICS_COMMON_VISUALIZATION
+#define PX_PHYSICS_COMMON_VISUALIZATION
+
+#include "foundation/PxTransform.h"
+#include "CmPhysXCommon.h"
+#include "PxConstraintDesc.h"
+
+namespace physx
+{
+
+class RenderOutput;
+
+namespace Cm
+{
+	PX_PHYSX_COMMON_API void visualizeJointFrames(RenderOutput& out,
+							  PxReal scale,
+							  const PxTransform& parent,
+							  const PxTransform& child);
+
+	PX_PHYSX_COMMON_API void visualizeLinearLimit(RenderOutput& out,
+							  PxReal scale,
+							  const PxTransform& t0,
+							  const PxTransform& t1,
+							  PxReal value,
+							  bool active);
+
+	PX_PHYSX_COMMON_API void visualizeAngularLimit(RenderOutput& out,
+							   PxReal scale,
+							   const PxTransform& t0,
+							   PxReal lower,
+							   PxReal upper,
+							   bool active);
+
+
+	PX_PHYSX_COMMON_API void visualizeLimitCone(RenderOutput& out,
+							PxReal scale,
+							const PxTransform& t,
+							PxReal ySwing,
+							PxReal zSwing,
+							bool active);
+
+	PX_PHYSX_COMMON_API void visualizeDoubleCone(RenderOutput& out,
+							 PxReal scale,
+							 const PxTransform& t,
+							 PxReal angle,
+							 bool active);
+	
+
+	struct ConstraintImmediateVisualizer : public PxConstraintVisualizer
+	{
+		PxF32			mFrameScale;
+		PxF32			mLimitScale;
+		RenderOutput&	mCmOutput;
+
+		//Not possible to implement
+		ConstraintImmediateVisualizer& operator=( const ConstraintImmediateVisualizer& );
+
+		ConstraintImmediateVisualizer( PxF32 _frameScale, PxF32 _limitScale, RenderOutput& _output )
+			: mFrameScale( _frameScale )
+			, mLimitScale( _limitScale )
+			//, mCmOutput(static_cast<RenderBuffer &>(_output))
+			, mCmOutput( _output )
+		{
+		}
+
+		virtual void visualizeJointFrames( const PxTransform& parent, const PxTransform& child )
+		{
+			Cm::visualizeJointFrames(mCmOutput, mFrameScale, parent, child );
+		}
+
+		virtual void visualizeLinearLimit( const PxTransform& t0, const PxTransform& t1, PxReal value, bool active )
+		{
+			Cm::visualizeLinearLimit( mCmOutput, mLimitScale, t0, t1, value, active );
+		}
+
+		virtual void visualizeAngularLimit( const PxTransform& t0, PxReal lower, PxReal upper, bool active)
+		{
+			Cm::visualizeAngularLimit( mCmOutput, mLimitScale, t0, lower, upper, active );
+		}
+
+		virtual void visualizeLimitCone( const PxTransform& t, PxReal ySwing, PxReal zSwing, bool active)
+		{
+			Cm::visualizeLimitCone( mCmOutput, mLimitScale, t, ySwing, zSwing, active );
+		}
+
+		virtual void visualizeDoubleCone( const PxTransform& t, PxReal angle, bool active)
+		{
+			Cm::visualizeDoubleCone( mCmOutput, mLimitScale, t, angle, active );
+		}
+	};
+}
+
+}
+
+#endif
diff --git a/PhysX_3.4/Source/Common/src/windows/CmWindowsDelayLoadHook.cpp b/PhysX_3.4/Source/Common/src/windows/CmWindowsDelayLoadHook.cpp
new file mode 100644
index 00000000..42a50c0e
--- /dev/null
+++ b/PhysX_3.4/Source/Common/src/windows/CmWindowsDelayLoadHook.cpp
@@ -0,0 +1,82 @@
+// This code contains NVIDIA Confidential Information and is disclosed to you
+// under a form of NVIDIA software license agreement provided separately to you.
+//
+// Notice
+// NVIDIA Corporation and its licensors retain all intellectual property and
+// proprietary rights in and to this software and related documentation and
+// any modifications thereto. Any use, reproduction, disclosure, or
+// distribution of this software and related documentation without an express
+// license agreement from NVIDIA Corporation is strictly prohibited.
+//
+// ALL NVIDIA DESIGN SPECIFICATIONS, CODE ARE PROVIDED "AS IS.". NVIDIA MAKES
+// NO WARRANTIES, EXPRESSED, IMPLIED, STATUTORY, OR OTHERWISE WITH RESPECT TO
+// THE MATERIALS, AND EXPRESSLY DISCLAIMS ALL IMPLIED WARRANTIES OF NONINFRINGEMENT,
+// MERCHANTABILITY, AND FITNESS FOR A PARTICULAR PURPOSE.
+//
+// Information and code furnished is believed to be accurate and reliable.
+// However, NVIDIA Corporation assumes no responsibility for the consequences of use of such
+// information or for any infringement of patents or other rights of third parties that may
+// result from its use. No license is granted by implication or otherwise under any patent
+// or patent rights of NVIDIA Corporation. Details are subject to change without notice.
+// This code supersedes and replaces all information previously supplied.
+// NVIDIA Corporation products are not authorized for use as critical
+// components in life support devices or systems without express written approval of
+// NVIDIA Corporation.
+//
+// Copyright (c) 2008-2016 NVIDIA Corporation. All rights reserved.
+// Copyright (c) 2004-2008 AGEIA Technologies, Inc. All rights reserved.
+// Copyright (c) 2001-2004 NovodeX AG. All rights reserved.  
+
+
+#include "windows/PxWindowsDelayLoadHook.h"
+#include "windows/PsWindowsInclude.h"
+#include "windows/CmWindowsLoadLibrary.h"
+
+// Prior to Visual Studio 2015 Update 3, these hooks were non-const.
+#define DELAYIMP_INSECURE_WRITABLE_HOOKS
+#include <delayimp.h>
+
+static const physx::PxDelayLoadHook* gDelayLoadHook = NULL;
+
+void physx::PxSetPhysXCommonDelayLoadHook(const physx::PxDelayLoadHook* hook)
+{
+	gDelayLoadHook = hook;
+}
+
+using namespace physx;
+
+#pragma comment(lib, "delayimp")
+
+FARPROC WINAPI delayHook(unsigned dliNotify, PDelayLoadInfo pdli)
+{
+	switch (dliNotify) {
+	case dliStartProcessing :
+		break;
+
+	case dliNotePreLoadLibrary :
+		{
+			return Cm::physXCommonDliNotePreLoadLibrary(pdli->szDll,gDelayLoadHook);
+		}
+		break;
+
+	case dliNotePreGetProcAddress :
+		break;
+
+	case dliFailLoadLib :
+		break;
+
+	case dliFailGetProc :
+		break;
+
+	case dliNoteEndProcessing :
+		break;
+
+	default :
+
+		return NULL;
+	}
+
+	return NULL;
+}
+
+PfnDliHook __pfnDliNotifyHook2 = delayHook;
diff --git a/PhysX_3.4/Source/Common/src/windows/CmWindowsModuleUpdateLoader.cpp b/PhysX_3.4/Source/Common/src/windows/CmWindowsModuleUpdateLoader.cpp
new file mode 100644
index 00000000..6ff93bc4
--- /dev/null
+++ b/PhysX_3.4/Source/Common/src/windows/CmWindowsModuleUpdateLoader.cpp
@@ -0,0 +1,130 @@
+// This code contains NVIDIA Confidential Information and is disclosed to you
+// under a form of NVIDIA software license agreement provided separately to you.
+//
+// Notice
+// NVIDIA Corporation and its licensors retain all intellectual property and
+// proprietary rights in and to this software and related documentation and
+// any modifications thereto. Any use, reproduction, disclosure, or
+// distribution of this software and related documentation without an express
+// license agreement from NVIDIA Corporation is strictly prohibited.
+//
+// ALL NVIDIA DESIGN SPECIFICATIONS, CODE ARE PROVIDED "AS IS.". NVIDIA MAKES
+// NO WARRANTIES, EXPRESSED, IMPLIED, STATUTORY, OR OTHERWISE WITH RESPECT TO
+// THE MATERIALS, AND EXPRESSLY DISCLAIMS ALL IMPLIED WARRANTIES OF NONINFRINGEMENT,
+// MERCHANTABILITY, AND FITNESS FOR A PARTICULAR PURPOSE.
+//
+// Information and code furnished is believed to be accurate and reliable.
+// However, NVIDIA Corporation assumes no responsibility for the consequences of use of such
+// information or for any infringement of patents or other rights of third parties that may
+// result from its use. No license is granted by implication or otherwise under any patent
+// or patent rights of NVIDIA Corporation. Details are subject to change without notice.
+// This code supersedes and replaces all information previously supplied.
+// NVIDIA Corporation products are not authorized for use as critical
+// components in life support devices or systems without express written approval of
+// NVIDIA Corporation.
+//
+// Copyright (c) 2008-2016 NVIDIA Corporation. All rights reserved.
+// Copyright (c) 2004-2008 AGEIA Technologies, Inc. All rights reserved.
+// Copyright (c) 2001-2004 NovodeX AG. All rights reserved.  
+
+#include "PsFoundation.h"
+
+#ifdef SUPPORT_UPDATE_LOADER_LOGGING
+#if PX_X86
+#define NX_USE_SDK_DLLS
+#include "PhysXUpdateLoader.h"
+#endif
+#endif /* SUPPORT_UPDATE_LOADER_LOGGING */
+
+#include "windows/CmWindowsModuleUpdateLoader.h"
+#include "windows/CmWindowsLoadLibrary.h"
+
+
+namespace physx { namespace Cm {
+
+#if PX_VC
+#pragma warning(disable: 4191)	//'operator/operation' : unsafe conversion from 'type of expression' to 'type required'
+#endif
+
+
+typedef HMODULE (*GetUpdatedModule_FUNC)(const char*, const char*);
+
+#ifdef SUPPORT_UPDATE_LOADER_LOGGING
+#if PX_X86
+typedef void (*setLogging_FUNC)(PXUL_ErrorCode, pt2LogFunc);
+
+static void LogMessage(PXUL_ErrorCode messageType, char* message)
+{
+	switch(messageType)
+	{
+	case PXUL_ERROR_MESSAGES:
+		getFoundation().error(PxErrorCode::eINTERNAL_ERROR, __FILE__, __LINE__, 
+			"PhysX Update Loader Error: %s.", message);
+		break;
+	case PXUL_WARNING_MESSAGES:
+		getFoundation().error(PX_WARN, "PhysX Update Loader Warning: %s.", message);
+		break;
+	case PXUL_INFO_MESSAGES:
+		getFoundation().error(PX_INFO, "PhysX Update Loader Information: %s.", message);
+		break;
+	default:
+		getFoundation().error(PxErrorCode::eINTERNAL_ERROR, __FILE__, __LINE__,
+			"Unknown message type from update loader.");
+		break;
+	}
+}
+#endif
+#endif /* SUPPORT_UPDATE_LOADER_LOGGING */
+
+CmModuleUpdateLoader::CmModuleUpdateLoader(const char* updateLoaderDllName)
+	: mGetUpdatedModuleFunc(NULL)
+{
+	mUpdateLoaderDllHandle = loadLibrary(updateLoaderDllName);
+
+	if (mUpdateLoaderDllHandle != NULL)
+	{
+		mGetUpdatedModuleFunc = GetProcAddress(mUpdateLoaderDllHandle, "GetUpdatedModule");
+
+#ifdef SUPPORT_UPDATE_LOADER_LOGGING
+#if PX_X86
+		setLogging_FUNC setLoggingFunc;
+		setLoggingFunc = (setLogging_FUNC)GetProcAddress(mUpdateLoaderDllHandle, "setLoggingFunction");
+		if(setLoggingFunc != NULL)		
+		{
+           setLoggingFunc(PXUL_ERROR_MESSAGES, LogMessage);
+        }
+#endif
+#endif /* SUPPORT_UPDATE_LOADER_LOGGING */
+	}
+}
+
+CmModuleUpdateLoader::~CmModuleUpdateLoader()
+{
+	if (mUpdateLoaderDllHandle != NULL)
+	{
+		FreeLibrary(mUpdateLoaderDllHandle);
+		mUpdateLoaderDllHandle = NULL;
+	}
+}
+
+HMODULE CmModuleUpdateLoader::LoadModule(const char* moduleName, const char* appGUID)
+{
+	HMODULE result = NULL;
+
+	if (mGetUpdatedModuleFunc != NULL)
+	{
+		// Try to get the module through PhysXUpdateLoader
+		GetUpdatedModule_FUNC getUpdatedModuleFunc = (GetUpdatedModule_FUNC)mGetUpdatedModuleFunc;
+		result = getUpdatedModuleFunc(moduleName, appGUID);
+	}
+	else
+	{
+		// If no PhysXUpdateLoader, just load the DLL directly
+		result = loadLibrary(moduleName);
+	}
+
+	return result;
+}
+
+}; // end of namespace
+}; // end of namespace