Merge branch 'master' into texture-preview

14 files changed, 395 insertions, 243 deletions

diff --git a/indra/llmath/CMakeLists.txt b/indra/llmath/CMakeLists.txt
index 552e820127..4617309606 100644
--- a/indra/llmath/CMakeLists.txt
+++ b/indra/llmath/CMakeLists.txt
@@ -4,12 +4,14 @@ project(llmath)
 
 include(00-Common)
 include(LLCommon)
+include(LLMeshOptimizer)
 include(bugsplat)
 include(Boost)
 
 include_directories(
     ${LLCOMMON_INCLUDE_DIRS}
     ${LLCOMMON_SYSTEM_INCLUDE_DIRS}
+    ${LLMESHOPTIMIZER_INCLUDE_DIRS}
     )
 
 set(llmath_SOURCE_FILES
@@ -109,6 +111,7 @@ add_library (llmath ${llmath_SOURCE_FILES})
 
 target_link_libraries(llmath
     ${LLCOMMON_LIBRARIES}
+    ${LLMESHOPTIMIZER_LIBRARIES}
     )
 
 # Add tests
diff --git a/indra/llmath/llmatrix4a.h b/indra/llmath/llmatrix4a.h
index 7ba347062f..2cf50e9cd2 100644
--- a/indra/llmath/llmatrix4a.h
+++ b/indra/llmath/llmatrix4a.h
@@ -36,6 +36,26 @@ class LLMatrix4a
 public:
 	LL_ALIGN_16(LLVector4a mMatrix[4]);
 
+    LLMatrix4a()
+    {
+
+    }
+
+    explicit LLMatrix4a(const LLMatrix4& val)
+    {
+        loadu(val);
+    }
+
+    inline F32* getF32ptr()
+    {
+        return (F32*) &mMatrix;
+    }
+
+    inline const F32* getF32ptr() const
+    {
+        return (F32*)&mMatrix;
+    }
+
 	inline void clear()
 	{
 		mMatrix[0].clear();
@@ -44,14 +64,29 @@ public:
 		mMatrix[3].clear();
 	}
 
+    inline void setIdentity()
+    {
+        mMatrix[0].set(1.f, 0.f, 0.f, 0.f);
+        mMatrix[1].set(0.f, 1.f, 0.f, 0.f);
+        mMatrix[2].set(0.f, 0.f, 1.f, 0.f);
+        mMatrix[3].set(0.f, 0.f, 0.f, 1.f);
+    }
+
 	inline void loadu(const LLMatrix4& src)
 	{
 		mMatrix[0] = _mm_loadu_ps(src.mMatrix[0]);
 		mMatrix[1] = _mm_loadu_ps(src.mMatrix[1]);
 		mMatrix[2] = _mm_loadu_ps(src.mMatrix[2]);
 		mMatrix[3] = _mm_loadu_ps(src.mMatrix[3]);
-		
 	}
+    
+    inline void loadu(const F32* src)
+    {
+        mMatrix[0] = _mm_loadu_ps(src);
+        mMatrix[1] = _mm_loadu_ps(src+4);
+        mMatrix[2] = _mm_loadu_ps(src+8);
+        mMatrix[3] = _mm_loadu_ps(src+12);
+    }
 
 	inline void loadu(const LLMatrix3& src)
 	{
@@ -105,7 +140,7 @@ public:
 		mMatrix[3].setAdd(a.mMatrix[3],d3);
 	}
 
-	inline void rotate(const LLVector4a& v, LLVector4a& res)
+	inline void rotate(const LLVector4a& v, LLVector4a& res) const
 	{
 		LLVector4a y,z;
 
@@ -151,6 +186,8 @@ public:
     {
         affineTransformSSE(v,res);
     }
+
+    const LLVector4a& getTranslation() const { return mMatrix[3]; }
 };
 
 inline LLVector4a rowMul(const LLVector4a &row, const LLMatrix4a &mat)
@@ -176,6 +213,15 @@ inline void matMul(const LLMatrix4a &a, const LLMatrix4a &b, LLMatrix4a &res)
     res.mMatrix[3] = row3;
 }
 
+//Faster version of matMul wehere res must not be a or b
+inline void matMulUnsafe(const LLMatrix4a &a, const LLMatrix4a &b, LLMatrix4a &res)
+{
+    res.mMatrix[0] = rowMul(a.mMatrix[0], b);
+    res.mMatrix[1] = rowMul(a.mMatrix[1], b);
+    res.mMatrix[2] = rowMul(a.mMatrix[2], b);
+    res.mMatrix[3] = rowMul(a.mMatrix[3], b);
+}
+
 inline std::ostream& operator<<(std::ostream& s, const LLMatrix4a& m)
 {
     s << "[" << m.mMatrix[0] << ", " << m.mMatrix[1] << ", " << m.mMatrix[2] << ", " << m.mMatrix[3] << "]";
diff --git a/indra/llmath/lloctree.h b/indra/llmath/lloctree.h
index 0e2f62f9db..318ee65cc0 100644
--- a/indra/llmath/lloctree.h
+++ b/indra/llmath/lloctree.h
@@ -34,6 +34,9 @@
 
 #define OCT_ERRS LL_WARNS("OctreeErrors")
 
+#define OCTREE_DEBUG_COLOR_REMOVE   0x0000FF // r
+#define OCTREE_DEBUG_COLOR_INSERT   0x00FF00 // g
+#define OCTREE_DEBUG_COLOR_BALANCE  0xFF0000 // b
 
 extern U32 gOctreeMaxCapacity;
 extern float gOctreeMinSize;
@@ -45,101 +48,98 @@ extern float gOctreeMinSize;
 #define LL_OCTREE_MAX_CAPACITY 128
 #endif*/
 
-template <class T> class LLOctreeNode;
+// T is the type of the element referenced by the octree node.
+// T_PTR determines how pointers to elements are stored internally.
+// LLOctreeNode<T, LLPointer<T>> assumes ownership of inserted elements and
+// deletes elements removed from the tree.
+// LLOctreeNode<T, T*> doesn't take ownership of inserted elements, so the API
+// user is responsible for managing the storage lifecycle of elements added to
+// the tree.
+template <class T, typename T_PTR> class LLOctreeNode;
 
-template <class T>
+template <class T, typename T_PTR>
 class LLOctreeListener: public LLTreeListener<T>
 {
 public:
 	typedef LLTreeListener<T> BaseType;
-	typedef LLOctreeNode<T> oct_node;
+    typedef LLOctreeNode<T, T_PTR> oct_node;
 
 	virtual void handleChildAddition(const oct_node* parent, oct_node* child) = 0;
 	virtual void handleChildRemoval(const oct_node* parent, const oct_node* child) = 0;
 };
 
-template <class T>
+template <class T, typename T_PTR>
 class LLOctreeTraveler
 {
 public:
-	virtual void traverse(const LLOctreeNode<T>* node);
-	virtual void visit(const LLOctreeNode<T>* branch) = 0;
+    virtual void traverse(const LLOctreeNode<T, T_PTR>* node);
+    virtual void visit(const LLOctreeNode<T, T_PTR>* branch) = 0;
 };
 
-template <class T>
-class LLOctreeTravelerDepthFirst : public LLOctreeTraveler<T>
+template <class T, typename T_PTR>
+class LLOctreeTravelerDepthFirst : public LLOctreeTraveler<T, T_PTR>
 {
 public:
-	virtual void traverse(const LLOctreeNode<T>* node);
+	virtual void traverse(const LLOctreeNode<T, T_PTR>* node) override;
 };
 
-template <class T>
-class LLOctreeNode : public LLTreeNode<T>
+template <class T, typename T_PTR>
+class alignas(16) LLOctreeNode : public LLTreeNode<T>
 {
+    LL_ALIGN_NEW
 public:
 
-	typedef LLOctreeTraveler<T>									oct_traveler;
-	typedef LLTreeTraveler<T>									tree_traveler;
-	typedef std::vector< LLPointer<T> >							element_list;		// note:  don't remove the whitespace between "> >"
-	typedef LLPointer<T>*										element_iter;
-	typedef const LLPointer<T>*									const_element_iter;
+    typedef LLOctreeTraveler<T, T_PTR>                          oct_traveler;
+    typedef LLTreeTraveler<T>                                   tree_traveler;
+    typedef std::vector<T_PTR>                                  element_list;
+    typedef typename element_list::iterator                     element_iter;
+    typedef typename element_list::const_iterator               const_element_iter;
 	typedef typename std::vector<LLTreeListener<T>*>::iterator	tree_listener_iter;
-	typedef LLOctreeNode<T>**									child_list;
-	typedef LLOctreeNode<T>**									child_iter;
+    typedef LLOctreeNode<T, T_PTR>**                            child_list;
+    typedef LLOctreeNode<T, T_PTR>**                            child_iter;
 
-	typedef LLTreeNode<T>		BaseType;
-	typedef LLOctreeNode<T>		oct_node;
-	typedef LLOctreeListener<T>	oct_listener;
+    typedef LLTreeNode<T>               BaseType;
+    typedef LLOctreeNode<T, T_PTR>      oct_node;
+    typedef LLOctreeListener<T, T_PTR>  oct_listener;
 
-	void* operator new(size_t size)
-	{
-		return ll_aligned_malloc_16(size);
-	}
-
-	void operator delete(void* ptr)
-	{
-		ll_aligned_free_16(ptr);
-	}
+    enum
+    {
+        NO_CHILD_NODES = 255 // Note: This is an U8 to match the max value in mChildMap[]
+    };
 
 	LLOctreeNode(	const LLVector4a& center, 
 					const LLVector4a& size, 
 					BaseType* parent, 
-					U8 octant = 255)
+					U8 octant = NO_CHILD_NODES)
 	:	mParent((oct_node*)parent), 
 		mOctant(octant) 
 	{ 
 		llassert(size[0] >= gOctreeMinSize*0.5f);
-		//always keep a NULL terminated list to avoid out of bounds exceptions in debug builds
-		mData.push_back(NULL);
-		mDataEnd = &mData[0];
 
 		mCenter = center;
 		mSize = size;
 
 		updateMinMax();
-		if ((mOctant == 255) && mParent)
+		if ((mOctant == NO_CHILD_NODES) && mParent)
 		{
 			mOctant = ((oct_node*) mParent)->getOctant(mCenter);
 		}
 
-		mElementCount = 0;
-
 		clearChildren();
 	}
 
-	virtual ~LLOctreeNode()								
+	virtual ~LLOctreeNode()
 	{ 
-		BaseType::destroyListeners(); 
+		BaseType::destroyListeners();
 		
-		for (U32 i = 0; i < mElementCount; ++i)
+        const U32 element_count = getElementCount();
+        for (U32 i = 0; i < element_count; ++i)
 		{
 			mData[i]->setBinIndex(-1);
 			mData[i] = NULL;
 		}
 
 		mData.clear();
-		mData.push_back(NULL);
-		mDataEnd = &mData[0];
 
 		for (U32 i = 0; i < getChildCount(); i++)
 		{
@@ -168,7 +168,7 @@ public:
 		return rad <= mSize[0]*2.f && isInside(pos); 
 	}
 
-	inline bool isInside(T* data) const			
+	inline bool isInside(T* data) const
 	{ 
 		return isInside(data->getPositionGroup(), data->getBinRadius());
 	}
@@ -239,14 +239,12 @@ public:
 	void accept(oct_traveler* visitor)				{ visitor->visit(this); }
 	virtual bool isLeaf() const						{ return mChildCount == 0; }
 	
-	U32 getElementCount() const						{ return mElementCount; }
-	bool isEmpty() const							{ return mElementCount == 0; }
-	element_list& getData()							{ return mData; }
-	const element_list& getData() const				{ return mData; }
-	element_iter getDataBegin()						{ return &mData[0]; }
-	element_iter getDataEnd()						{ return mDataEnd; }
-	const_element_iter getDataBegin() const			{ return &mData[0]; }
-	const_element_iter getDataEnd() const			{ return mDataEnd; }
+    U32 getElementCount() const                     { return (U32)mData.size(); }
+    bool isEmpty() const                            { return mData.empty(); }
+    element_iter getDataBegin()                     { return mData.begin(); }
+    element_iter getDataEnd()                       { return mData.end(); }
+    const_element_iter getDataBegin() const         { return mData.cbegin(); }
+    const_element_iter getDataEnd() const           { return mData.cend(); }
 		
 	U32 getChildCount()	const						{ return mChildCount; }
 	oct_node* getChild(U32 index)					{ return mChild[index]; }
@@ -262,9 +260,9 @@ public:
 		for (U32 i = 0; i < 8; i++)
 		{
 			U8 idx = mChildMap[i];
-			if (idx != 255)
+			if (idx != NO_CHILD_NODES)
 			{
-				LLOctreeNode<T>* child = mChild[idx];
+                oct_node* child = mChild[idx];
 
 				if (child->getOctant() != i)
 				{
@@ -282,10 +280,10 @@ public:
 
 	oct_node* getNodeAt(const LLVector4a& pos, const F32& rad)
 	{ 
-		LLOctreeNode<T>* node = this;
+        oct_node* node = this;
 
 		if (node->isInside(pos, rad))
-		{		
+		{
 			//do a quick search by octant
 			U8 octant = node->getOctant(pos);
 			
@@ -295,7 +293,7 @@ public:
 			// the data
 			U8 next_node = node->mChildMap[octant];
 			
-			while (next_node != 255 && node->getSize()[0] >= rad)
+			while (next_node != NO_CHILD_NODES && node->getSize()[0] >= rad)
 			{	
 				node = node->getChild(next_node);
 				octant = node->getOctant(pos);
@@ -304,7 +302,7 @@ public:
 		}
 		else if (!node->contains(rad) && node->getParent())
 		{ //if we got here, data does not exist in this node
-			return ((LLOctreeNode<T>*) node->getParent())->getNodeAt(pos, rad);
+            return ((oct_node*) node->getParent())->getNodeAt(pos, rad);
 		}
 
 		return node;
@@ -312,12 +310,14 @@ public:
 	
 	virtual bool insert(T* data)
 	{
+        //LL_PROFILE_ZONE_NAMED_COLOR("Octree::insert()",OCTREE_DEBUG_COLOR_INSERT);
+
 		if (data == NULL || data->getBinIndex() != -1)
 		{
 			OCT_ERRS << "!!! INVALID ELEMENT ADDED TO OCTREE BRANCH !!!" << LL_ENDL;
 			return false;
 		}
-		LLOctreeNode<T>* parent = getOctParent();
+        oct_node* parent = getOctParent();
 
 		//is it here?
 		if (isInside(data->getPositionGroup()))
@@ -325,11 +325,8 @@ public:
 			if ((((getElementCount() < gOctreeMaxCapacity || getSize()[0] <= gOctreeMinSize) && contains(data->getBinRadius())) ||
 				(data->getBinRadius() > getSize()[0] &&	parent && parent->getElementCount() >= gOctreeMaxCapacity))) 
 			{ //it belongs here
-				mData.push_back(NULL);
-				mData[mElementCount] = data;
-				mElementCount++;
-				mDataEnd = &mData[mElementCount];
-				data->setBinIndex(mElementCount-1);
+                mData.push_back(data);
+                data->setBinIndex(getElementCount() - 1);
 				BaseType::insert(data);
 				return true;
 			}
@@ -353,7 +350,7 @@ public:
 				size.mul(0.5f);
 		        		
 				//push center in direction of data
-				LLOctreeNode<T>::pushCenter(center, size, data);
+                oct_node::pushCenter(center, size, data);
 
 				// handle case where floating point number gets too small
 				LLVector4a val;
@@ -365,11 +362,8 @@ public:
 
 				if( lt == 0x7 )
 				{
-					mData.push_back(NULL);
-					mData[mElementCount] = data;
-					mElementCount++;
-					mDataEnd = &mData[mElementCount];
-					data->setBinIndex(mElementCount-1);
+                    mData.push_back(data);
+                    data->setBinIndex(getElementCount() - 1);
 					BaseType::insert(data);
 					return true;
 				}
@@ -395,7 +389,7 @@ public:
 
 				llassert(size[0] >= gOctreeMinSize*0.5f);
 				//make the new kid
-				child = new LLOctreeNode<T>(center, size, this);
+                child = new oct_node(center, size, this);
 				addChild(child);
 								
 				child->insert(data);
@@ -428,28 +422,25 @@ public:
 	}
 
 	void _remove(T* data, S32 i)
-	{ //precondition -- mElementCount > 0, idx is in range [0, mElementCount)
+    { //precondition -- getElementCount() > 0, idx is in range [0, getElementCount())
 
-		mElementCount--;
 		data->setBinIndex(-1); 
 		
-		if (mElementCount > 0)
+        const U32 new_element_count = getElementCount() - 1;
+		if (new_element_count > 0)
 		{
-			if (mElementCount != i)
+			if (new_element_count != i)
 			{
-				mData[i] = mData[mElementCount]; //might unref data, do not access data after this point
+				mData[i] = mData[new_element_count]; //might unref data, do not access data after this point
 				mData[i]->setBinIndex(i);
 			}
 
-			mData[mElementCount] = NULL;
+			mData[new_element_count] = NULL;
 			mData.pop_back();
-			mDataEnd = &mData[mElementCount];
 		}
 		else
 		{
 			mData.clear();
-			mData.push_back(NULL);
-			mDataEnd = &mData[0];
 		}
 
 		this->notifyRemoval(data);
@@ -458,9 +449,11 @@ public:
 
 	bool remove(T* data)
 	{
+        //LL_PROFILE_ZONE_NAMED_COLOR("Octree::remove()", OCTREE_DEBUG_COLOR_REMOVE);
+
 		S32 i = data->getBinIndex();
 
-		if (i >= 0 && i < mElementCount)
+        if (i >= 0 && i < getElementCount())
 		{
 			if (mData[i] == data)
 			{ //found it
@@ -503,7 +496,8 @@ public:
 
 	void removeByAddress(T* data)
 	{
-        for (U32 i = 0; i < mElementCount; ++i)
+        const U32 element_count = getElementCount();
+        for (U32 i = 0; i < element_count; ++i)
 		{
 			if (mData[i] == data)
 			{ //we have data
@@ -515,7 +509,7 @@ public:
 		
 		for (U32 i = 0; i < getChildCount(); i++)
 		{	//we don't contain data, so pass this guy down
-			LLOctreeNode<T>* child = (LLOctreeNode<T>*) getChild(i);
+            oct_node* child = (oct_node*) getChild(i);
 			child->removeByAddress(data);
 		}
 	}
@@ -523,9 +517,7 @@ public:
 	void clearChildren()
 	{
 		mChildCount = 0;
-
-		U32* foo = (U32*) mChildMap;
-		foo[0] = foo[1] = 0xFFFFFFFF;
+		memset(mChildMap, NO_CHILD_NODES, sizeof(mChildMap));
 	}
 
 	void validate()
@@ -616,11 +608,9 @@ public:
 		--mChildCount;
 
 		mChild[index] = mChild[mChildCount];
-		
 
 		//rebuild child map
-		U32* foo = (U32*) mChildMap;
-		foo[0] = foo[1] = 0xFFFFFFFF;
+		memset(mChildMap, NO_CHILD_NODES, sizeof(mChildMap));
 
 		for (U32 i = 0; i < mChildCount; ++i)
 		{
@@ -656,7 +646,7 @@ public:
 		OCT_ERRS << "Octree failed to delete requested child." << LL_ENDL;
 	}
 
-protected:	
+protected:
 	typedef enum
 	{
 		CENTER = 0,
@@ -673,23 +663,20 @@ protected:
 	oct_node* mParent;
 	U8 mOctant;
 
-	LLOctreeNode<T>* mChild[8];
+    oct_node* mChild[8];
 	U8 mChildMap[8];
 	U32 mChildCount;
 
 	element_list mData;
-	element_iter mDataEnd;
-	U32 mElementCount;
-		
 }; 
 
 //just like a regular node, except it might expand on insert and compress on balance
-template <class T>
-class LLOctreeRoot : public LLOctreeNode<T>
+template <class T, typename T_PTR>
+class LLOctreeRoot : public LLOctreeNode<T, T_PTR>
 {
 public:
-	typedef LLOctreeNode<T>	BaseType;
-	typedef LLOctreeNode<T>		oct_node;
+    typedef LLOctreeNode<T, T_PTR> BaseType;
+    typedef LLOctreeNode<T, T_PTR> oct_node;
 
 	LLOctreeRoot(const LLVector4a& center, 
 				 const LLVector4a& size, 
@@ -698,11 +685,13 @@ public:
 	{
 	}
 	
-	bool balance()
+	bool balance() override
 	{	
+        //LL_PROFILE_ZONE_NAMED_COLOR("Octree::balance()",OCTREE_DEBUG_COLOR_BALANCE);
+
 		if (this->getChildCount() == 1 && 
 			!(this->mChild[0]->isLeaf()) &&
-			this->mChild[0]->getElementCount() == 0) 
+			this->mChild[0]->getElementCount() == 0)
 		{ //if we have only one child and that child is an empty branch, make that child the root
 			oct_node* child = this->mChild[0];
 					
@@ -732,7 +721,7 @@ public:
 	}
 
 	// LLOctreeRoot::insert
-	bool insert(T* data)
+	bool insert(T* data) override
 	{
 		if (data == NULL) 
 		{
@@ -768,7 +757,7 @@ public:
 			oct_node* node = this->getNodeAt(data);
 			if (node == this)
 			{
-				LLOctreeNode<T>::insert(data);
+                oct_node::insert(data);
 			}
 			else if (node->isInside(data->getPositionGroup()))
 			{
@@ -788,13 +777,13 @@ public:
 				LLVector4a center, size;
 				center = this->getCenter();
 				size = this->getSize();
-				LLOctreeNode<T>::pushCenter(center, size, data);
+                oct_node::pushCenter(center, size, data);
 				this->setCenter(center);
 				size.mul(2.f);
 				this->setSize(size);
 				this->updateMinMax();
 			}
-			LLOctreeNode<T>::insert(data);
+            oct_node::insert(data);
 		}
 		else
 		{
@@ -806,7 +795,7 @@ public:
 
 				//expand this node
 				LLVector4a newcenter(center);
-				LLOctreeNode<T>::pushCenter(newcenter, size, data);
+                oct_node::pushCenter(newcenter, size, data);
 				this->setCenter(newcenter);
 				LLVector4a size2 = size;
 				size2.mul(2.f);
@@ -816,11 +805,11 @@ public:
 				llassert(size[0] >= gOctreeMinSize);
 
 				//copy our children to a new branch
-				LLOctreeNode<T>* newnode = new LLOctreeNode<T>(center, size, this);
+                oct_node* newnode = new oct_node(center, size, this);
 				
 				for (U32 i = 0; i < this->getChildCount(); i++)
 				{
-					LLOctreeNode<T>* child = this->getChild(i);
+                    oct_node* child = this->getChild(i);
 					newnode->addChild(child);
 				}
 
@@ -835,13 +824,19 @@ public:
 
 		return false;
 	}
+
+    bool isLeaf() const override
+    {
+        // root can't be a leaf
+        return false;
+    }
 };
 
 //========================
 //		LLOctreeTraveler
 //========================
-template <class T>
-void LLOctreeTraveler<T>::traverse(const LLOctreeNode<T>* node)
+template <class T, typename T_PTR>
+void LLOctreeTraveler<T, T_PTR>::traverse(const LLOctreeNode<T, T_PTR>* node)
 {
 	node->accept(this);
 	for (U32 i = 0; i < node->getChildCount(); i++)
@@ -850,8 +845,8 @@ void LLOctreeTraveler<T>::traverse(const LLOctreeNode<T>* node)
 	}
 }
 
-template <class T>
-void LLOctreeTravelerDepthFirst<T>::traverse(const LLOctreeNode<T>* node)
+template <class T, typename T_PTR>
+void LLOctreeTravelerDepthFirst<T, T_PTR>::traverse(const LLOctreeNode<T, T_PTR>* node)
 {
 	for (U32 i = 0; i < node->getChildCount(); i++)
 	{
diff --git a/indra/llmath/llrigginginfo.h b/indra/llmath/llrigginginfo.h
index b3d6bc2d19..059c6ae082 100644
--- a/indra/llmath/llrigginginfo.h
+++ b/indra/llmath/llrigginginfo.h
@@ -34,9 +34,9 @@
 
 // Extents are in joint space
 // isRiggedTo is based on the state of all currently associated rigged meshes
-LL_ALIGN_PREFIX(16)
-class LLJointRiggingInfo
+class alignas(16) LLJointRiggingInfo
 {
+    LL_ALIGN_NEW
 public:
     LLJointRiggingInfo();
     bool isRiggedTo() const;
@@ -45,31 +45,10 @@ public:
     const LLVector4a *getRiggedExtents() const;
     void merge(const LLJointRiggingInfo& other);
 
-	void* operator new(size_t size)
-	{
-		return ll_aligned_malloc_16(size);
-	}
-
-	void operator delete(void* ptr)
-	{
-		ll_aligned_free_16(ptr);
-	}
-
-	void* operator new[](size_t size)
-	{
-		return ll_aligned_malloc_16(size);
-	}
-
-	void operator delete[](void* ptr)
-	{
-		ll_aligned_free_16(ptr);
-	}
-
-
 private:
-	LL_ALIGN_16(LLVector4a mRiggedExtents[2]);
+	LLVector4a mRiggedExtents[2];
     bool mIsRiggedTo;
-} LL_ALIGN_POSTFIX(16);
+};
 
 // For storing all the rigging info associated with a given avatar or
 // object, keyed by joint_num.
diff --git a/indra/llmath/llvector4a.h b/indra/llmath/llvector4a.h
index 27abf39537..53c8f604f6 100644
--- a/indra/llmath/llvector4a.h
+++ b/indra/llmath/llvector4a.h
@@ -46,11 +46,10 @@ class LLRotation;
 // of this writing, July 08, 2010) about getting it implemented before you resort to
 // LLVector3/LLVector4. 
 /////////////////////////////////
-struct LLVector4a;
 
-LL_ALIGN_PREFIX(16)
-struct LLVector4a
+class alignas(16) LLVector4a
 {
+    LL_ALIGN_NEW
 public:
 
 	///////////////////////////////////
@@ -138,10 +137,10 @@ public:
 	// BASIC GET/SET 
 	////////////////////////////////////
 	
-	// Return a "this" as an F32 pointer. Do not use unless you have a very good reason.  (Not sure? Ask Falcon)
+	// Return a "this" as an F32 pointer.
 	inline F32* getF32ptr();
 	
-	// Return a "this" as a const F32 pointer. Do not use unless you have a very good reason.  (Not sure? Ask Falcon)
+	// Return a "this" as a const F32 pointer.
 	inline const F32* const getF32ptr() const;
 	
 	// Read-only access a single float in this vector. Do not use in proximity to any function call that manipulates
@@ -324,7 +323,7 @@ public:
     
 private:
 	LLQuad mQ;
-} LL_ALIGN_POSTFIX(16);
+};
 
 inline void update_min_max(LLVector4a& min, LLVector4a& max, const LLVector4a& p)
 {
diff --git a/indra/llmath/llvector4a.inl b/indra/llmath/llvector4a.inl
index 69d3d01efe..8be1c1b114 100644
--- a/indra/llmath/llvector4a.inl
+++ b/indra/llmath/llvector4a.inl
@@ -58,13 +58,13 @@ inline void LLVector4a::store4a(F32* dst) const
 // BASIC GET/SET 
 ////////////////////////////////////
 
-// Return a "this" as an F32 pointer. Do not use unless you have a very good reason.  (Not sure? Ask Falcon)
+// Return a "this" as an F32 pointer.
 F32* LLVector4a::getF32ptr()
 {
 	return (F32*) &mQ;
 }
 
-// Return a "this" as a const F32 pointer. Do not use unless you have a very good reason.  (Not sure? Ask Falcon)
+// Return a "this" as a const F32 pointer.
 const F32* const LLVector4a::getF32ptr() const
 {
 	return (const F32* const) &mQ;
diff --git a/indra/llmath/llvolume.cpp b/indra/llmath/llvolume.cpp
index 13b65dfaa0..f43d07ce5e 100644
--- a/indra/llmath/llvolume.cpp
+++ b/indra/llmath/llvolume.cpp
@@ -49,6 +49,7 @@
 #include "llsdserialize.h"
 #include "llvector4a.h"
 #include "llmatrix4a.h"
+#include "llmeshoptimizer.h"
 #include "lltimer.h"
 
 #define DEBUG_SILHOUETTE_BINORMALS 0
@@ -88,7 +89,7 @@ const F32 SKEW_MAX	=  0.95f;
 const F32 SCULPT_MIN_AREA = 0.002f;
 const S32 SCULPT_MIN_AREA_DETAIL = 1;
 
-BOOL gDebugGL = FALSE;
+BOOL gDebugGL = FALSE; // See settings.xml "RenderDebugGL"
 
 BOOL check_same_clock_dir( const LLVector3& pt1, const LLVector3& pt2, const LLVector3& pt3, const LLVector3& norm)
 {    
@@ -370,7 +371,7 @@ BOOL LLTriangleRayIntersect(const LLVector3& vert0, const LLVector3& vert1, cons
 	}
 }
 
-class LLVolumeOctreeRebound : public LLOctreeTravelerDepthFirst<LLVolumeTriangle>
+class LLVolumeOctreeRebound : public LLOctreeTravelerDepthFirst<LLVolumeTriangle, LLVolumeTriangle*>
 {
 public:
 	const LLVolumeFace* mFace;
@@ -380,9 +381,10 @@ public:
 		mFace = face;
 	}
 
-	virtual void visit(const LLOctreeNode<LLVolumeTriangle>* branch)
+    virtual void visit(const LLOctreeNode<LLVolumeTriangle, LLVolumeTriangle*>* branch)
 	{ //this is a depth first traversal, so it's safe to assum all children have complete
 		//bounding data
+	LL_PROFILE_ZONE_SCOPED_CATEGORY_VOLUME
 
 		LLVolumeOctreeListener* node = (LLVolumeOctreeListener*) branch->getListener(0);
 
@@ -397,8 +399,7 @@ public:
 			min = *(tri->mV[0]);
 			max = *(tri->mV[0]);
 			
-			for (LLOctreeNode<LLVolumeTriangle>::const_element_iter iter = 
-				branch->getDataBegin(); iter != branch->getDataEnd(); ++iter)
+            for (LLOctreeNode<LLVolumeTriangle, LLVolumeTriangle*>::const_element_iter iter = branch->getDataBegin(); iter != branch->getDataEnd(); ++iter)
 			{ //for each triangle in node
 
 				//stretch by triangles in node
@@ -413,7 +414,7 @@ public:
 				max.setMax(max, *tri->mV[2]);
 			}
 		}
-		else if (!branch->isLeaf())
+		else if (branch->getChildCount() > 0)
 		{ //no data, but child nodes exist
 			LLVolumeOctreeListener* child = (LLVolumeOctreeListener*) branch->getChild(0)->getListener(0);
 
@@ -423,7 +424,7 @@ public:
 		}
 		else
 		{
-			LL_ERRS() << "Empty leaf" << LL_ENDL;
+            llassert(!branch->isLeaf()); // Empty leaf
 		}
 
 		for (S32 i = 0; i < branch->getChildCount(); ++i)
@@ -682,7 +683,7 @@ LLProfile::Face* LLProfile::addHole(const LLProfileParams& params, BOOL flat, F3
 
 	Face *face = addFace(mTotalOut, mTotal-mTotalOut,0,LL_FACE_INNER_SIDE, flat);
 
-	static LLAlignedArray<LLVector4a,64> pt;
+	static thread_local LLAlignedArray<LLVector4a,64> pt;
 	pt.resize(mTotal) ;
 
 	for (S32 i=mTotalOut;i<mTotal;i++)
@@ -822,6 +823,8 @@ S32 LLProfile::getNumPoints(const LLProfileParams& params, BOOL path_open,F32 de
 BOOL LLProfile::generate(const LLProfileParams& params, BOOL path_open,F32 detail, S32 split,
 						 BOOL is_sculpted, S32 sculpt_size)
 {
+	LL_PROFILE_ZONE_SCOPED_CATEGORY_VOLUME
+
 	if ((!mDirty) && (!is_sculpted))
 	{
 		return FALSE;
@@ -1302,6 +1305,8 @@ S32 LLPath::getNumNGonPoints(const LLPathParams& params, S32 sides, F32 startOff
 
 void LLPath::genNGon(const LLPathParams& params, S32 sides, F32 startOff, F32 end_scale, F32 twist_scale)
 {
+	LL_PROFILE_ZONE_SCOPED_CATEGORY_VOLUME
+
 	// Generates a circular path, starting at (1, 0, 0), counterclockwise along the xz plane.
 	static const F32 tableScale[] = { 1, 1, 1, 0.5f, 0.707107f, 0.53f, 0.525f, 0.5f };
 
@@ -1536,6 +1541,8 @@ S32 LLPath::getNumPoints(const LLPathParams& params, F32 detail)
 BOOL LLPath::generate(const LLPathParams& params, F32 detail, S32 split,
 					  BOOL is_sculpted, S32 sculpt_size)
 {
+	LL_PROFILE_ZONE_SCOPED_CATEGORY_VOLUME
+
 	if ((!mDirty) && (!is_sculpted))
 	{
 		return FALSE;
@@ -1617,9 +1624,6 @@ BOOL LLPath::generate(const LLPathParams& params, F32 detail, S32 split,
 			//genNGon(params, llfloor(MIN_DETAIL_FACES * detail), 4.f, 0.f);
 			genNGon(params, llfloor(MIN_DETAIL_FACES * detail));
 
-			F32 t     = 0.f;
-			F32 tStep = 1.0f / mPath.size();
-
 			F32 toggle = 0.5f;
 			for (S32 i=0;i<(S32)mPath.size();i++)
 			{
@@ -1628,7 +1632,6 @@ BOOL LLPath::generate(const LLPathParams& params, F32 detail, S32 split,
 					toggle = -0.5f;
 				else
 					toggle = 0.5f;
-				t += tStep;
 			}
 		}
 
@@ -2112,6 +2115,8 @@ LLVolume::~LLVolume()
 
 BOOL LLVolume::generate()
 {
+	LL_PROFILE_ZONE_SCOPED_CATEGORY_VOLUME
+
 	LL_CHECK_MEMORY
 	llassert_always(mProfilep);
 	
@@ -2370,6 +2375,8 @@ bool LLVolumeFace::VertexData::compareNormal(const LLVolumeFace::VertexData& rhs
 
 bool LLVolume::unpackVolumeFaces(std::istream& is, S32 size)
 {
+	LL_PROFILE_ZONE_SCOPED_CATEGORY_VOLUME
+
 	//input stream is now pointing at a zlib compressed block of LLSD
 	//decompress block
 	LLSD mdl;
@@ -2415,6 +2422,13 @@ bool LLVolume::unpackVolumeFaces(std::istream& is, S32 size)
 
 			//copy out indices
             S32 num_indices = idx.size() / 2;
+            const S32 indices_to_discard = num_indices % 3;
+            if (indices_to_discard > 0)
+            {
+                // Invalid number of triangle indices
+                LL_WARNS() << "Incomplete triangle discarded from face! Indices count " << num_indices << " was not divisible by 3. face index: " << i << " Total: " << face_count << LL_ENDL;
+                num_indices -= indices_to_discard;
+            }
             face.resizeIndices(num_indices);
 
             if (num_indices > 2 && !face.mIndices)
@@ -2430,8 +2444,7 @@ bool LLVolume::unpackVolumeFaces(std::istream& is, S32 size)
 			}
 
 			U16* indices = (U16*) &(idx[0]);
-			U32 count = idx.size()/2;
-			for (U32 j = 0; j < count; ++j)
+            for (U32 j = 0; j < num_indices; ++j)
 			{
 				face.mIndices[j] = indices[j];
 			}
@@ -2776,6 +2789,8 @@ S32	LLVolume::getNumFaces() const
 
 void LLVolume::createVolumeFaces()
 {
+	LL_PROFILE_ZONE_SCOPED_CATEGORY_VOLUME
+
 	if (mGenerateSingleFace)
 	{
 		// do nothing
@@ -3741,6 +3756,8 @@ void LLVolume::generateSilhouetteVertices(std::vector<LLVector3> &vertices,
 										  const LLMatrix3& norm_mat_in,
 										  S32 face_mask)
 {
+	LL_PROFILE_ZONE_SCOPED_CATEGORY_VOLUME
+
 	LLMatrix4a mat;
 	mat.loadu(mat_in);
 
@@ -3819,8 +3836,8 @@ void LLVolume::generateSilhouetteVertices(std::vector<LLVector3> &vertices,
 #if DEBUG_SILHOUETTE_EDGE_MAP
 
 			//for each triangle
-			U32 count = face.mNumIndices;
-			for (U32 j = 0; j < count/3; j++) {
+            U32 tri_count = face.mNumIndices / 3;
+            for (U32 j = 0; j < tri_count; j++) {
 				//get vertices
 				S32 v1 = face.mIndices[j*3+0];
 				S32 v2 = face.mIndices[j*3+1];
@@ -3838,7 +3855,7 @@ void LLVolume::generateSilhouetteVertices(std::vector<LLVector3> &vertices,
 						continue;
 					}
 
-					if (nIndex >= (S32) count/3) {
+                    if (nIndex >= (S32)tri_count) {
 						continue;
 					}
 					//get neighbor vertices
@@ -4130,13 +4147,13 @@ S32 LLVolume::lineSegmentIntersect(const LLVector4a& start, const LLVector4a& en
 			}
 			else
 			{
-				if (!face.mOctree)
+                if (!face.getOctree())
 				{
 					face.createOctree();
 				}
 			
 				LLOctreeTriangleRayIntersect intersect(start, dir, &face, &closest_t, intersection, tex_coord, normal, tangent_out);
-				intersect.traverse(face.mOctree);
+                intersect.traverse(face.getOctree());
 				if (intersect.mHitFace)
 				{
 					hit_face = i;
@@ -4691,6 +4708,7 @@ LLVolumeFace::LLVolumeFace() :
 #endif
     mWeightsScrubbed(FALSE),
 	mOctree(NULL),
+    mOctreeTriangles(NULL),
 	mOptimized(FALSE)
 {
 	mExtents = (LLVector4a*) ll_aligned_malloc_16(sizeof(LLVector4a)*3);
@@ -4720,8 +4738,9 @@ LLVolumeFace::LLVolumeFace(const LLVolumeFace& src)
     mJointIndices(NULL),
 #endif
     mWeightsScrubbed(FALSE),
-	mOctree(NULL)
-{ 
+    mOctree(NULL),
+    mOctreeTriangles(NULL)
+{
 	mExtents = (LLVector4a*) ll_aligned_malloc_16(sizeof(LLVector4a)*3);
 	mCenter = mExtents+2;
 	*this = src;
@@ -4861,15 +4880,15 @@ void LLVolumeFace::freeData()
 	mJustWeights = NULL;
 #endif
 
-	delete mOctree;
-	mOctree = NULL;
+    destroyOctree();
 }
 
 BOOL LLVolumeFace::create(LLVolume* volume, BOOL partial_build)
 {
+	LL_PROFILE_ZONE_SCOPED_CATEGORY_VOLUME
+
 	//tree for this face is no longer valid
-	delete mOctree;
-	mOctree = NULL;
+    destroyOctree();
 
 	LL_CHECK_MEMORY
 	BOOL ret = FALSE ;
@@ -4935,6 +4954,50 @@ bool LLVolumeFace::VertexMapData::ComparePosition::operator()(const LLVector3& a
 	return a.mV[2] < b.mV[2];
 }
 
+void LLVolumeFace::remap()
+{
+    // Generate a remap buffer
+    std::vector<unsigned int> remap(mNumVertices);
+    S32 remap_vertices_count = LLMeshOptimizer::generateRemapMultiU16(&remap[0],
+        mIndices,
+        mNumIndices,
+        mPositions,
+        mNormals,
+        mTexCoords,
+        mNumVertices);
+
+    // Allocate new buffers
+    S32 size = ((mNumIndices * sizeof(U16)) + 0xF) & ~0xF;
+    U16* remap_indices = (U16*)ll_aligned_malloc_16(size);
+
+    S32 tc_bytes_size = ((remap_vertices_count * sizeof(LLVector2)) + 0xF) & ~0xF;
+    LLVector4a* remap_positions = (LLVector4a*)ll_aligned_malloc<64>(sizeof(LLVector4a) * 2 * remap_vertices_count + tc_bytes_size);
+    LLVector4a* remap_normals = remap_positions + remap_vertices_count;
+    LLVector2* remap_tex_coords = (LLVector2*)(remap_normals + remap_vertices_count);
+
+    // Fill the buffers
+    LLMeshOptimizer::remapIndexBufferU16(remap_indices, mIndices, mNumIndices, &remap[0]);
+    LLMeshOptimizer::remapPositionsBuffer(remap_positions, mPositions, mNumVertices, &remap[0]);
+    LLMeshOptimizer::remapNormalsBuffer(remap_normals, mNormals, mNumVertices, &remap[0]);
+    LLMeshOptimizer::remapUVBuffer(remap_tex_coords, mTexCoords, mNumVertices, &remap[0]);
+
+    // Free unused buffers
+    ll_aligned_free_16(mIndices);
+    ll_aligned_free<64>(mPositions);
+
+    // Tangets are now invalid
+    ll_aligned_free_16(mTangents);
+    mTangents = NULL;
+
+    // Assign new values
+    mIndices = remap_indices;
+    mPositions = remap_positions;
+    mNormals = remap_normals;
+    mTexCoords = remap_tex_coords;
+    mNumVertices = remap_vertices_count;
+    mNumAllocatedVertices = remap_vertices_count;
+}
+
 void LLVolumeFace::optimize(F32 angle_cutoff)
 {
 	LLVolumeFace new_face;
@@ -5536,21 +5599,29 @@ bool LLVolumeFace::cacheOptimize()
 
 void LLVolumeFace::createOctree(F32 scaler, const LLVector4a& center, const LLVector4a& size)
 {
-	if (mOctree)
+	LL_PROFILE_ZONE_SCOPED_CATEGORY_VOLUME
+
+    if (getOctree())
 	{
 		return;
 	}
 
-	mOctree = new LLOctreeRoot<LLVolumeTriangle>(center, size, NULL);
+    llassert(mNumIndices % 3 == 0);
+
+    mOctree = new LLOctreeRoot<LLVolumeTriangle, LLVolumeTriangle*>(center, size, NULL);
 	new LLVolumeOctreeListener(mOctree);
+    const U32 num_triangles = mNumIndices / 3;
+    // Initialize all the triangles we need
+    mOctreeTriangles = new LLVolumeTriangle[num_triangles];
 
-	for (U32 i = 0; i < mNumIndices; i+= 3)
+    for (U32 triangle_index = 0; triangle_index < num_triangles; ++triangle_index)
 	{ //for each triangle
-		LLPointer<LLVolumeTriangle> tri = new LLVolumeTriangle();
+        const U32 index = triangle_index * 3;
+        LLVolumeTriangle* tri = &mOctreeTriangles[triangle_index];
 				
-		const LLVector4a& v0 = mPositions[mIndices[i]];
-		const LLVector4a& v1 = mPositions[mIndices[i+1]];
-		const LLVector4a& v2 = mPositions[mIndices[i+2]];
+		const LLVector4a& v0 = mPositions[mIndices[index]];
+		const LLVector4a& v1 = mPositions[mIndices[index + 1]];
+		const LLVector4a& v2 = mPositions[mIndices[index + 2]];
 
 		//store pointers to vertex data
 		tri->mV[0] = &v0;
@@ -5558,9 +5629,9 @@ void LLVolumeFace::createOctree(F32 scaler, const LLVector4a& center, const LLVe
 		tri->mV[2] = &v2;
 
 		//store indices
-		tri->mIndex[0] = mIndices[i];
-		tri->mIndex[1] = mIndices[i+1];
-		tri->mIndex[2] = mIndices[i+2];
+		tri->mIndex[0] = mIndices[index];
+		tri->mIndex[1] = mIndices[index + 1];
+		tri->mIndex[2] = mIndices[index + 2];
 
 		//get minimum point
 		LLVector4a min = v0;
@@ -5603,6 +5674,19 @@ void LLVolumeFace::createOctree(F32 scaler, const LLVector4a& center, const LLVe
 	}
 }
 
+void LLVolumeFace::destroyOctree()
+{
+    delete mOctree;
+    mOctree = NULL;
+    delete[] mOctreeTriangles;
+    mOctreeTriangles = NULL;
+}
+
+const LLOctreeNode<LLVolumeTriangle, LLVolumeTriangle*>* LLVolumeFace::getOctree() const
+{
+    return mOctree;
+}
+
 
 void LLVolumeFace::swapData(LLVolumeFace& rhs)
 {
@@ -5745,7 +5829,16 @@ BOOL LLVolumeFace::createUnCutCubeCap(LLVolume* volume, BOOL partial_build)
 		resizeIndices(grid_size*grid_size*6);
 		if (!volume->isMeshAssetLoaded())
 		{
-			mEdge.resize(grid_size*grid_size * 6);
+            S32 size = grid_size * grid_size * 6;
+            try
+            {
+                mEdge.resize(size);
+            }
+            catch (std::bad_alloc&)
+            {
+                LL_WARNS("LLVOLUME") << "Resize of mEdge to " << size << " failed" << LL_ENDL;
+                return false;
+            }
 		}
 
 		U16* out = mIndices;
@@ -6309,6 +6402,8 @@ void CalculateTangentArray(U32 vertexCount, const LLVector4a *vertex, const LLVe
 
 void LLVolumeFace::createTangents()
 {
+	LL_PROFILE_ZONE_SCOPED_CATEGORY_VOLUME
+
 	if (!mTangents)
 	{
 		allocateTangents(mNumVertices);
@@ -6349,9 +6444,9 @@ void LLVolumeFace::resizeVertices(S32 num_verts)
 	if (num_verts)
 	{
 		//pad texture coordinate block end to allow for QWORD reads
-		S32 size = ((num_verts*sizeof(LLVector2)) + 0xF) & ~0xF;
+		S32 tc_size = ((num_verts*sizeof(LLVector2)) + 0xF) & ~0xF;
 
-		mPositions = (LLVector4a*) ll_aligned_malloc<64>(sizeof(LLVector4a)*2*num_verts+size);
+		mPositions = (LLVector4a*) ll_aligned_malloc<64>(sizeof(LLVector4a)*2*num_verts+tc_size);
 		mNormals = mPositions+num_verts;
 		mTexCoords = (LLVector2*) (mNormals+num_verts);
 
@@ -6463,6 +6558,7 @@ void LLVolumeFace::allocateJointIndices(S32 num_verts)
 void LLVolumeFace::resizeIndices(S32 num_indices)
 {
 	ll_aligned_free_16(mIndices);
+    llassert(num_indices % 3 == 0);
 	
 	if (num_indices)
 	{
@@ -6522,6 +6618,8 @@ void LLVolumeFace::fillFromLegacyData(std::vector<LLVolumeFace::VertexData>& v,
 
 BOOL LLVolumeFace::createSide(LLVolume* volume, BOOL partial_build)
 {
+	LL_PROFILE_ZONE_SCOPED_CATEGORY_VOLUME
+
 	LL_CHECK_MEMORY
 	BOOL flat = mTypeMask & FLAT_MASK;
 
@@ -6554,7 +6652,15 @@ BOOL LLVolumeFace::createSide(LLVolume* volume, BOOL partial_build)
 
 		if (!volume->isMeshAssetLoaded())
 		{
-			mEdge.resize(num_indices);
+            try
+            {
+                mEdge.resize(num_indices);
+            }
+            catch (std::bad_alloc&)
+            {
+                LL_WARNS("LLVOLUME") << "Resize of mEdge to " << num_indices << " failed" << LL_ENDL;
+                return false;
+            }
 		}
 	}
 
@@ -6589,13 +6695,19 @@ BOOL LLVolumeFace::createSide(LLVolume* volume, BOOL partial_build)
 			else
 			{
 				// Get s value for tex-coord.
-				if (!flat)
+                S32 index = mBeginS + s;
+                if (index >= profile.size())
+                {
+                    // edge?
+                    ss = flat ? 1.f - begin_stex : 1.f;
+                }
+				else if (!flat)
 				{
-					ss = profile[mBeginS + s][2];
+					ss = profile[index][2];
 				}
 				else
 				{
-					ss = profile[mBeginS + s][2] - begin_stex;
+					ss = profile[index][2] - begin_stex;
 				}
 			}
 
@@ -6781,8 +6893,16 @@ BOOL LLVolumeFace::createSide(LLVolume* volume, BOOL partial_build)
 
 	LLVector4a* norm = mNormals;
 
-	static LLAlignedArray<LLVector4a, 64> triangle_normals;
-	triangle_normals.resize(count);
+    static thread_local LLAlignedArray<LLVector4a, 64> triangle_normals;
+    try
+    {
+        triangle_normals.resize(count);
+    }
+    catch (std::bad_alloc&)
+    {
+        LL_WARNS("LLVOLUME") << "Resize of triangle_normals to " << count << " failed" << LL_ENDL;
+        return false;
+    }
 	LLVector4a* output = triangle_normals.mArray;
 	LLVector4a* end_output = output+count;
 
@@ -7014,6 +7134,8 @@ BOOL LLVolumeFace::createSide(LLVolume* volume, BOOL partial_build)
 void CalculateTangentArray(U32 vertexCount, const LLVector4a *vertex, const LLVector4a *normal,
         const LLVector2 *texcoord, U32 triangleCount, const U16* index_array, LLVector4a *tangent)
 {
+	LL_PROFILE_ZONE_SCOPED_CATEGORY_VOLUME
+
     //LLVector4a *tan1 = new LLVector4a[vertexCount * 2];
 	LLVector4a* tan1 = (LLVector4a*) ll_aligned_malloc_16(vertexCount*2*sizeof(LLVector4a));
 	// new(tan1) LLVector4a;
diff --git a/indra/llmath/llvolume.h b/indra/llmath/llvolume.h
index a77e8c08c6..3ccaed47f1 100644
--- a/indra/llmath/llvolume.h
+++ b/indra/llmath/llvolume.h
@@ -35,7 +35,8 @@ class LLVolumeParams;
 class LLProfile;
 class LLPath;
 
-template <class T> class LLOctreeNode;
+template<class T> class LLPointer;
+template <class T, typename T_PTR> class LLOctreeNode;
 
 class LLVolumeFace;
 class LLVolume;
@@ -902,10 +903,17 @@ public:
 		typedef std::map<LLVector3, std::vector<VertexMapData>, VertexMapData::ComparePosition > PointMap;
 	};
 
+    // Eliminates non unique triangles, takes positions,
+    // normals and texture coordinates into account.
+    void remap();
+
 	void optimize(F32 angle_cutoff = 2.f);
 	bool cacheOptimize();
 
 	void createOctree(F32 scaler = 0.25f, const LLVector4a& center = LLVector4a(0,0,0), const LLVector4a& size = LLVector4a(0.5f,0.5f,0.5f));
+    void destroyOctree();
+    // Get a reference to the octree, which may be null
+    const LLOctreeNode<LLVolumeTriangle, LLVolumeTriangle*>* getOctree() const;
 
 	enum
 	{
@@ -936,17 +944,23 @@ public:
 	LLVector4a* mCenter;
 	LLVector2   mTexCoordExtents[2]; //minimum and maximum of texture coordinates of the face.
 
-	S32 mNumVertices;
+	S32 mNumVertices; // num vertices == num normals == num texcoords
 	S32 mNumAllocatedVertices;
 	S32 mNumIndices;
 
-	LLVector4a* mPositions;
-	LLVector4a* mNormals;
+	LLVector4a* mPositions; // Contains vertices, nortmals and texcoords
+	LLVector4a* mNormals; // pointer into mPositions
 	LLVector4a* mTangents;
-	LLVector2*  mTexCoords;
+	LLVector2*  mTexCoords; // pointer into mPositions
+
+	// mIndices contains mNumIndices amount of elements.
+	// It contains triangles, each 3 indices describe one triangle.
+    // If mIndices contains {0, 2, 3, 1, 2, 4}, it means there
+    // are two triangles {0, 2, 3} and {1, 2, 4} with values being
+    // indexes for mPositions/mNormals/mTexCoords
 	U16* mIndices;
 
-	//vertex buffer filled in by LLFace to cache this volume face geometry in vram 
+	// vertex buffer filled in by LLFace to cache this volume face geometry in vram 
 	// (declared as a LLPointer to LLRefCount to avoid dependency on LLVertexBuffer)
 	mutable LLPointer<LLRefCount> mVertexBuffer; 
 
@@ -967,13 +981,14 @@ public:
     // Which joints are rigged to, and the bounding box of any rigged
     // vertices per joint.
     LLJointRiggingInfoTab mJointRiggingInfoTab;
-    
-	LLOctreeNode<LLVolumeTriangle>* mOctree;
 
 	//whether or not face has been cache optimized
 	BOOL mOptimized;
 
 private:
+    LLOctreeNode<LLVolumeTriangle, LLVolumeTriangle*>* mOctree;
+    LLVolumeTriangle* mOctreeTriangles;
+
 	BOOL createUnCutCubeCap(LLVolume* volume, BOOL partial_build = FALSE);
 	BOOL createCap(LLVolume* volume, BOOL partial_build = FALSE);
 	BOOL createSide(LLVolume* volume, BOOL partial_build = FALSE);
diff --git a/indra/llmath/llvolumeoctree.cpp b/indra/llmath/llvolumeoctree.cpp
index fb232d5f6c..6894d04d3c 100644
--- a/indra/llmath/llvolumeoctree.cpp
+++ b/indra/llmath/llvolumeoctree.cpp
@@ -75,7 +75,7 @@ BOOL LLLineSegmentBoxIntersect(const LLVector4a& start, const LLVector4a& end, c
 }
 
 
-LLVolumeOctreeListener::LLVolumeOctreeListener(LLOctreeNode<LLVolumeTriangle>* node)
+LLVolumeOctreeListener::LLVolumeOctreeListener(LLOctreeNode<LLVolumeTriangle, LLVolumeTriangle*>* node)
 {
 	node->addListener(this);
 }
@@ -85,13 +85,12 @@ LLVolumeOctreeListener::~LLVolumeOctreeListener()
 
 }
 	
-void LLVolumeOctreeListener::handleChildAddition(const LLOctreeNode<LLVolumeTriangle>* parent, 
-	LLOctreeNode<LLVolumeTriangle>* child)
+void LLVolumeOctreeListener::handleChildAddition(const LLOctreeNode<LLVolumeTriangle, LLVolumeTriangle*>* parent, 
+    LLOctreeNode<LLVolumeTriangle, LLVolumeTriangle*>* child)
 {
 	new LLVolumeOctreeListener(child);
 }
 
-
 LLOctreeTriangleRayIntersect::LLOctreeTriangleRayIntersect(const LLVector4a& start, const LLVector4a& dir, 
 							   const LLVolumeFace* face, F32* closest_t,
 							   LLVector4a* intersection,LLVector2* tex_coord, LLVector4a* normal, LLVector4a* tangent)
@@ -108,7 +107,7 @@ LLOctreeTriangleRayIntersect::LLOctreeTriangleRayIntersect(const LLVector4a& sta
 	mEnd.setAdd(mStart, mDir);
 }
 
-void LLOctreeTriangleRayIntersect::traverse(const LLOctreeNode<LLVolumeTriangle>* node)
+void LLOctreeTriangleRayIntersect::traverse(const LLOctreeNode<LLVolumeTriangle, LLVolumeTriangle*>* node)
 {
 	LLVolumeOctreeListener* vl = (LLVolumeOctreeListener*) node->getListener(0);
 
@@ -122,9 +121,9 @@ void LLOctreeTriangleRayIntersect::traverse(const LLOctreeNode<LLVolumeTriangle>
 	}
 }
 
-void LLOctreeTriangleRayIntersect::visit(const LLOctreeNode<LLVolumeTriangle>* node)
+void LLOctreeTriangleRayIntersect::visit(const LLOctreeNode<LLVolumeTriangle, LLVolumeTriangle*>* node)
 {
-	for (LLOctreeNode<LLVolumeTriangle>::const_element_iter iter = 
+    for (typename LLOctreeNode<LLVolumeTriangle, LLVolumeTriangle*>::const_element_iter iter =
 			node->getDataBegin(); iter != node->getDataEnd(); ++iter)
 	{
 		const LLVolumeTriangle* tri = *iter;
@@ -219,7 +218,7 @@ const F32& LLVolumeTriangle::getBinRadius() const
 
 //TEST CODE
 
-void LLVolumeOctreeValidate::visit(const LLOctreeNode<LLVolumeTriangle>* branch)
+void LLVolumeOctreeValidate::visit(const LLOctreeNode<LLVolumeTriangle, LLVolumeTriangle*>* branch)
 {
 	LLVolumeOctreeListener* node = (LLVolumeOctreeListener*) branch->getListener(0);
 
@@ -256,7 +255,7 @@ void LLVolumeOctreeValidate::visit(const LLOctreeNode<LLVolumeTriangle>* branch)
 	}
 
 	//children fit, check data
-	for (LLOctreeNode<LLVolumeTriangle>::const_element_iter iter = branch->getDataBegin(); 
+    for (typename LLOctreeNode<LLVolumeTriangle, LLVolumeTriangle*>::const_element_iter iter = branch->getDataBegin();
 			iter != branch->getDataEnd(); ++iter)
 	{
 		const LLVolumeTriangle* tri = *iter;
@@ -273,4 +272,3 @@ void LLVolumeOctreeValidate::visit(const LLOctreeNode<LLVolumeTriangle>* branch)
 	}
 }
 
-
diff --git a/indra/llmath/llvolumeoctree.h b/indra/llmath/llvolumeoctree.h
index 13150028d8..d65bca5e52 100644
--- a/indra/llmath/llvolumeoctree.h
+++ b/indra/llmath/llvolumeoctree.h
@@ -34,19 +34,10 @@
 #include "llvolume.h"
 #include "llvector4a.h"
 
-class LLVolumeTriangle : public LLRefCount
+class alignas(16) LLVolumeTriangle : public LLRefCount
 {
+    LL_ALIGN_NEW
 public:
-	void* operator new(size_t size)
-	{
-		return ll_aligned_malloc_16(size);
-	}
-
-	void operator delete(void* ptr)
-	{
-		ll_aligned_free_16(ptr);
-	}
-
 	LLVolumeTriangle()
 	{
 		mBinIndex = -1;	
@@ -86,21 +77,11 @@ public:
 
 };
 
-class LLVolumeOctreeListener : public LLOctreeListener<LLVolumeTriangle>
+class alignas(16) LLVolumeOctreeListener : public LLOctreeListener<LLVolumeTriangle, LLVolumeTriangle*>
 {
+    LL_ALIGN_NEW
 public:
-	
-	void* operator new(size_t size)
-	{
-		return ll_aligned_malloc_16(size);
-	}
-
-	void operator delete(void* ptr)
-	{
-		ll_aligned_free_16(ptr);
-	}
-
-	LLVolumeOctreeListener(LLOctreeNode<LLVolumeTriangle>* node);
+    LLVolumeOctreeListener(LLOctreeNode<LLVolumeTriangle, LLVolumeTriangle*>* node);
 	~LLVolumeOctreeListener();
 	
 	LLVolumeOctreeListener(const LLVolumeOctreeListener& rhs)
@@ -115,11 +96,9 @@ public:
 	}
 
 	 //LISTENER FUNCTIONS
-	virtual void handleChildAddition(const LLOctreeNode<LLVolumeTriangle>* parent, 
-		LLOctreeNode<LLVolumeTriangle>* child);
+    virtual void handleChildAddition(const LLOctreeNode<LLVolumeTriangle, LLVolumeTriangle*>* parent, LLOctreeNode<LLVolumeTriangle, LLVolumeTriangle*>* child);
 	virtual void handleStateChange(const LLTreeNode<LLVolumeTriangle>* node) { }
-	virtual void handleChildRemoval(const LLOctreeNode<LLVolumeTriangle>* parent, 
-			const LLOctreeNode<LLVolumeTriangle>* child) {	}
+    virtual void handleChildRemoval(const LLOctreeNode<LLVolumeTriangle, LLVolumeTriangle*>* parent, const LLOctreeNode<LLVolumeTriangle, LLVolumeTriangle*>* child) { }
 	virtual void handleInsertion(const LLTreeNode<LLVolumeTriangle>* node, LLVolumeTriangle* tri) { }
 	virtual void handleRemoval(const LLTreeNode<LLVolumeTriangle>* node, LLVolumeTriangle* tri) { }
 	virtual void handleDestruction(const LLTreeNode<LLVolumeTriangle>* node) { }
@@ -130,7 +109,7 @@ public:
 	LL_ALIGN_16(LLVector4a mExtents[2]); // extents (min, max) of this node and all its children
 };
 
-class LLOctreeTriangleRayIntersect : public LLOctreeTraveler<LLVolumeTriangle>
+class LLOctreeTriangleRayIntersect : public LLOctreeTraveler<LLVolumeTriangle, LLVolumeTriangle*>
 {
 public:
 	const LLVolumeFace* mFace;
@@ -148,14 +127,14 @@ public:
 								   const LLVolumeFace* face, F32* closest_t,
 								   LLVector4a* intersection,LLVector2* tex_coord, LLVector4a* normal, LLVector4a* tangent);
 
-	void traverse(const LLOctreeNode<LLVolumeTriangle>* node);
+    void traverse(const LLOctreeNode<LLVolumeTriangle, LLVolumeTriangle*>* node);
 
-	virtual void visit(const LLOctreeNode<LLVolumeTriangle>* node);
+    virtual void visit(const LLOctreeNode<LLVolumeTriangle, LLVolumeTriangle*>* node);
 };
 
-class LLVolumeOctreeValidate : public LLOctreeTraveler<LLVolumeTriangle>
+class LLVolumeOctreeValidate : public LLOctreeTraveler<LLVolumeTriangle, LLVolumeTriangle*>
 {
-	virtual void visit(const LLOctreeNode<LLVolumeTriangle>* branch);
+    virtual void visit(const LLOctreeNode<LLVolumeTriangle, LLVolumeTriangle*>* branch);
 };
 
 #endif
diff --git a/indra/llmath/m4math.cpp b/indra/llmath/m4math.cpp
index 3baf1bad18..6e40dae30b 100644
--- a/indra/llmath/m4math.cpp
+++ b/indra/llmath/m4math.cpp
@@ -32,8 +32,7 @@
 #include "m4math.h"
 #include "m3math.h"
 #include "llquaternion.h"
-
-
+#include "llmatrix4a.h"
 
 
 // LLMatrix4
@@ -115,6 +114,12 @@ LLMatrix4::LLMatrix4(const LLQuaternion &q)
 	*this = initRotation(q);
 }
 
+LLMatrix4::LLMatrix4(const LLMatrix4a& mat)
+    : LLMatrix4(mat.getF32ptr())
+{
+    
+}
+
 LLMatrix4::LLMatrix4(const LLQuaternion &q, const LLVector4 &pos)
 {
 	*this = initRotTrans(q, pos);
diff --git a/indra/llmath/m4math.h b/indra/llmath/m4math.h
index bf60adb9b6..b9da970cde 100644
--- a/indra/llmath/m4math.h
+++ b/indra/llmath/m4math.h
@@ -32,6 +32,7 @@
 class LLVector4;
 class LLMatrix3;
 class LLQuaternion;
+class LLMatrix4a;
 
 // NOTA BENE: Currently assuming a right-handed, x-forward, y-left, z-up universe
 
@@ -104,6 +105,7 @@ public:
 	explicit LLMatrix4(const F32 *mat);								// Initializes Matrix to values in mat
 	explicit LLMatrix4(const LLMatrix3 &mat);						// Initializes Matrix to values in mat and sets position to (0,0,0)
 	explicit LLMatrix4(const LLQuaternion &q);						// Initializes Matrix with rotation q and sets position to (0,0,0)
+    explicit LLMatrix4(const LLMatrix4a& mat);
 
 	LLMatrix4(const LLMatrix3 &mat, const LLVector4 &pos);	// Initializes Matrix to values in mat and pos
 
diff --git a/indra/llmath/v3math.cpp b/indra/llmath/v3math.cpp
index b04c67d926..93010d2250 100644
--- a/indra/llmath/v3math.cpp
+++ b/indra/llmath/v3math.cpp
@@ -316,6 +316,12 @@ LLVector3::LLVector3(const LLVector4 &vec)
 	mV[VZ] = (F32)vec.mV[VZ];
 }
 
+LLVector3::LLVector3(const LLVector4a& vec)
+    : LLVector3(vec.getF32ptr())
+{
+
+}
+
 LLVector3::LLVector3(const LLSD& sd)
 {
 	setValue(sd);
diff --git a/indra/llmath/v3math.h b/indra/llmath/v3math.h
index 6f857d7061..068f489020 100644
--- a/indra/llmath/v3math.h
+++ b/indra/llmath/v3math.h
@@ -33,6 +33,7 @@
 #include "llsd.h"
 class LLVector2;
 class LLVector4;
+class LLVector4a;
 class LLMatrix3;
 class LLMatrix4;
 class LLVector3d;
@@ -62,7 +63,9 @@ class LLVector3
 		explicit LLVector3(const LLVector2 &vec);				// Initializes LLVector3 to (vec[0]. vec[1], 0)
 		explicit LLVector3(const LLVector3d &vec);				// Initializes LLVector3 to (vec[0]. vec[1], vec[2])
 		explicit LLVector3(const LLVector4 &vec);				// Initializes LLVector4 to (vec[0]. vec[1], vec[2])
-		explicit LLVector3(const LLSD& sd);
+        explicit LLVector3(const LLVector4a& vec);              // Initializes LLVector4 to (vec[0]. vec[1], vec[2])
+        explicit LLVector3(const LLSD& sd);
+        
 
 		LLSD getValue() const;