merge

46 files changed, 2153 insertions, 1200 deletions

diff --git a/indra/llmath/CMakeLists.txt b/indra/llmath/CMakeLists.txt
index 367486eee7..dda07133d5 100644
--- a/indra/llmath/CMakeLists.txt
+++ b/indra/llmath/CMakeLists.txt
@@ -22,6 +22,7 @@ set(llmath_SOURCE_FILES
     llsphere.cpp
     llvolume.cpp
     llvolumemgr.cpp
+    llvolumeoctree.cpp
     llsdutil_math.cpp
     m3math.cpp
     m4math.cpp
@@ -66,6 +67,7 @@ set(llmath_HEADER_FILES
     llmatrix4a.h
     llvolume.h
     llvolumemgr.h
+    llvolumeoctree.h
     llsdutil_math.h
     m3math.h
     m4math.h
diff --git a/indra/llmath/llcamera.cpp b/indra/llmath/llcamera.cpp
index 487ed6451f..6b56e4870e 100644
--- a/indra/llmath/llcamera.cpp
+++ b/indra/llmath/llcamera.cpp
@@ -48,10 +48,10 @@ LLCamera::LLCamera() :
 	mPlaneCount(6),
 	mFrustumCornerDist(0.f)
 {
+	alignPlanes();
 	calculateFrustumPlanes();
 } 
 
-
 LLCamera::LLCamera(F32 vertical_fov_rads, F32 aspect_ratio, S32 view_height_in_pixels, F32 near_plane, F32 far_plane) :
 	LLCoordFrame(),
 	mViewHeightInPixels(view_height_in_pixels),
@@ -59,6 +59,7 @@ LLCamera::LLCamera(F32 vertical_fov_rads, F32 aspect_ratio, S32 view_height_in_p
 	mPlaneCount(6),
 	mFrustumCornerDist(0.f)
 {
+	alignPlanes();
 	mAspect = llclamp(aspect_ratio, MIN_ASPECT_RATIO, MAX_ASPECT_RATIO);
 	mNearPlane = llclamp(near_plane, MIN_NEAR_PLANE, MAX_NEAR_PLANE);
 	if(far_plane < 0) far_plane = DEFAULT_FAR_PLANE;
@@ -67,6 +68,23 @@ LLCamera::LLCamera(F32 vertical_fov_rads, F32 aspect_ratio, S32 view_height_in_p
 	setView(vertical_fov_rads);
 } 
 
+LLCamera::~LLCamera()
+{
+
+}
+
+const LLCamera& LLCamera::operator=(const LLCamera& rhs)
+{
+	memcpy(this, &rhs, sizeof(LLCamera));
+	alignPlanes();
+	LLVector4a::memcpyNonAliased16((F32*) mAgentPlanes, (F32*) rhs.mAgentPlanes, 4*7);
+	return *this;
+}
+
+void LLCamera::alignPlanes()
+{
+	mAgentPlanes = (LLPlane*) LL_NEXT_ALIGNED_ADDRESS<U8>(mAgentPlaneBuffer);
+}
 
 // ---------------- LLCamera::getFoo() member functions ----------------
 
@@ -91,8 +109,8 @@ F32 LLCamera::getMaxView() const
 void LLCamera::setUserClipPlane(LLPlane plane)
 {
 	mPlaneCount = 7;
-	mAgentPlanes[6].p = plane;
-	mAgentPlanes[6].mask = calcPlaneMask(plane);
+	mAgentPlanes[6] = plane;
+	mPlaneMask[6] = calcPlaneMask(plane);
 }
 
 void LLCamera::disableUserClipPlane()
@@ -164,129 +182,66 @@ size_t LLCamera::readFrustumFromBuffer(const char *buffer)
 
 // ---------------- test methods  ---------------- 
 
-S32 LLCamera::AABBInFrustum(const LLVector3 &center, const LLVector3& radius) 
-{
-	static const LLVector3 scaler[] = {
-		LLVector3(-1,-1,-1),
-		LLVector3( 1,-1,-1),
-		LLVector3(-1, 1,-1),
-		LLVector3( 1, 1,-1),
-		LLVector3(-1,-1, 1),
-		LLVector3( 1,-1, 1),
-		LLVector3(-1, 1, 1),
-		LLVector3( 1, 1, 1)
+S32 LLCamera::AABBInFrustum(const LLVector4a &center, const LLVector4a& radius) 
+{
+	static const LLVector4a scaler[] = {
+		LLVector4a(-1,-1,-1),
+		LLVector4a( 1,-1,-1),
+		LLVector4a(-1, 1,-1),
+		LLVector4a( 1, 1,-1),
+		LLVector4a(-1,-1, 1),
+		LLVector4a( 1,-1, 1),
+		LLVector4a(-1, 1, 1),
+		LLVector4a( 1, 1, 1)
 	};
 
 	U8 mask = 0;
 	S32 result = 2;
 
-	/*if (mFrustumCornerDist > 0.f && radius.magVecSquared() > mFrustumCornerDist * mFrustumCornerDist)
-	{ //box is larger than frustum, check frustum quads against box planes
-
-		static const LLVector3 dir[] = 
-		{
-			LLVector3(1, 0, 0),
-			LLVector3(-1, 0, 0),
-			LLVector3(0, 1, 0),
-			LLVector3(0, -1, 0),
-			LLVector3(0, 0, 1),
-			LLVector3(0, 0, -1)
-		};
-
-		U32 quads[] = 
+	for (U32 i = 0; i < mPlaneCount; i++)
+	{
+		mask = mPlaneMask[i];
+		if (mask == 0xff)
 		{
-			0, 1, 2, 3,
-			0, 1, 5, 4,
-			2, 3, 7, 6,
-			3, 0, 7, 4,
-			1, 2, 6, 4,
-			4, 5, 6, 7
-		};
-
-		result = 0;
-
-		BOOL total_inside = TRUE;
-		for (U32 i = 0; i < 6; i++)
-		{ 
-			LLVector3 p = center + radius.scaledVec(dir[i]);
-			F32 d = -p*dir[i];
-
-			for (U32 j = 0; j <	6; j++)
-			{ //for each quad
-				F32 dist = mAgentFrustum[quads[j*4+0]]*dir[i] + d;
-				if (dist > 0)
-				{ //at least one frustum point is outside the AABB
-					total_inside = FALSE;
-					for (U32 k = 1; k < 4; k++)
-					{ //for each other point on quad
-						if ( mAgentFrustum[quads[j*4+k]]*dir[i]+d  <= 0.f)
-						{ //quad is straddling some plane of AABB
-							return 1;
-						}
-					}
-				}
-				else
-				{
-					for (U32 k = 1; k < 4; k++)
-					{
-						if (mAgentFrustum[quads[j*4+k]]*dir[i]+d > 0.f)
-						{
-							return 1;
-						}
-					}
-				}
-			}
+			continue;
 		}
 
-		if (total_inside)
+		const LLPlane& p = mAgentPlanes[i];
+		const LLVector4a& n = reinterpret_cast<const LLVector4a&>(p);
+		float d = p.mV[3];
+		LLVector4a rscale;
+		rscale.setMul(radius, scaler[mask]);
+
+		LLVector4a minp, maxp;
+		minp.setSub(center, rscale);
+		maxp.setAdd(center, rscale);
+
+		if (n.dot3(minp) > -d) 
 		{
-			result = 1;
+			return 0;
 		}
-	}
-	else*/
-	{
-		for (U32 i = 0; i < mPlaneCount; i++)
+	
+		if (n.dot3(maxp) > -d)
 		{
-			mask = mAgentPlanes[i].mask;
-			if (mask == 0xff)
-			{
-				continue;
-			}
-			LLPlane p = mAgentPlanes[i].p;
-			LLVector3 n = LLVector3(p);
-			float d = p.mV[3];
-			LLVector3 rscale = radius.scaledVec(scaler[mask]);
-
-			LLVector3 minp = center - rscale;
-			LLVector3 maxp = center + rscale;
-
-			if (n * minp > -d) 
-			{
-				return 0;
-			}
-		
-			if (n * maxp > -d)
-			{
-				result = 1;
-			}
+			result = 1;
 		}
 	}
 
-	
 	return result;
 }
 
-S32 LLCamera::AABBInFrustumNoFarClip(const LLVector3 &center, const LLVector3& radius) 
-{
-	static const LLVector3 scaler[] = {
-		LLVector3(-1,-1,-1),
-		LLVector3( 1,-1,-1),
-		LLVector3(-1, 1,-1),
-		LLVector3( 1, 1,-1),
-		LLVector3(-1,-1, 1),
-		LLVector3( 1,-1, 1),
-		LLVector3(-1, 1, 1),
-		LLVector3( 1, 1, 1)
+
+S32 LLCamera::AABBInFrustumNoFarClip(const LLVector4a& center, const LLVector4a& radius) 
+{
+	static const LLVector4a scaler[] = {
+		LLVector4a(-1,-1,-1),
+		LLVector4a( 1,-1,-1),
+		LLVector4a(-1, 1,-1),
+		LLVector4a( 1, 1,-1),
+		LLVector4a(-1,-1, 1),
+		LLVector4a( 1,-1, 1),
+		LLVector4a(-1, 1, 1),
+		LLVector4a( 1, 1, 1)
 	};
 
 	U8 mask = 0;
@@ -299,25 +254,28 @@ S32 LLCamera::AABBInFrustumNoFarClip(const LLVector3 &center, const LLVector3& r
 			continue;
 		}
 
-		mask = mAgentPlanes[i].mask;
+		mask = mPlaneMask[i];
 		if (mask == 0xff)
 		{
 			continue;
 		}
-		LLPlane p = mAgentPlanes[i].p;
-		LLVector3 n = LLVector3(p);
+
+		const LLPlane& p = mAgentPlanes[i];
+		const LLVector4a& n = reinterpret_cast<const LLVector4a&>(p);
 		float d = p.mV[3];
-		LLVector3 rscale = radius.scaledVec(scaler[mask]);
+		LLVector4a rscale;
+		rscale.setMul(radius, scaler[mask]);
 
-		LLVector3 minp = center - rscale;
-		LLVector3 maxp = center + rscale;
+		LLVector4a minp, maxp;
+		minp.setSub(center, rscale);
+		maxp.setAdd(center, rscale);
 
-		if (n * minp > -d) 
+		if (n.dot3(minp) > -d) 
 		{
 			return 0;
 		}
 	
-		if (n * maxp > -d)
+		if (n.dot3(maxp) > -d)
 		{
 			result = 1;
 		}
@@ -447,12 +405,12 @@ int LLCamera::sphereInFrustum(const LLVector3 &sphere_center, const F32 radius)
 	int res = 2;
 	for (int i = 0; i < 6; i++)
 	{
-		if (mAgentPlanes[i].mask == 0xff)
+		if (mPlaneMask[i] == 0xff)
 		{
 			continue;
 		}
 
-		float d = mAgentPlanes[i].p.dist(sphere_center);
+		float d = mAgentPlanes[i].dist(sphere_center);
 
 		if (d > radius) 
 		{
@@ -644,12 +602,14 @@ void LLCamera::ignoreAgentFrustumPlane(S32 idx)
 		return;
 	}
 
-	mAgentPlanes[idx].mask = 0xff;
-	mAgentPlanes[idx].p.clearVec();
+	mPlaneMask[idx] = 0xff;
+	mAgentPlanes[idx].clearVec();
 }
 
 void LLCamera::calcAgentFrustumPlanes(LLVector3* frust)
 {
+	alignPlanes();
+
 	for (int i = 0; i < 8; i++)
 	{
 		mAgentFrustum[i] = frust[i];
@@ -662,27 +622,27 @@ void LLCamera::calcAgentFrustumPlanes(LLVector3* frust)
 	//order of planes is important, keep most likely to fail in the front of the list
 
 	//near - frust[0], frust[1], frust[2]
-	mAgentPlanes[2].p = planeFromPoints(frust[0], frust[1], frust[2]);
+	mAgentPlanes[2] = planeFromPoints(frust[0], frust[1], frust[2]);
 
 	//far  
-	mAgentPlanes[5].p = planeFromPoints(frust[5], frust[4], frust[6]);
+	mAgentPlanes[5] = planeFromPoints(frust[5], frust[4], frust[6]);
 
 	//left  
-	mAgentPlanes[0].p = planeFromPoints(frust[4], frust[0], frust[7]);
+	mAgentPlanes[0] = planeFromPoints(frust[4], frust[0], frust[7]);
 
 	//right  
-	mAgentPlanes[1].p = planeFromPoints(frust[1], frust[5], frust[6]);
+	mAgentPlanes[1] = planeFromPoints(frust[1], frust[5], frust[6]);
 
 	//top  
-	mAgentPlanes[4].p = planeFromPoints(frust[3], frust[2], frust[6]);
+	mAgentPlanes[4] = planeFromPoints(frust[3], frust[2], frust[6]);
 
 	//bottom  
-	mAgentPlanes[3].p = planeFromPoints(frust[1], frust[0], frust[4]);
+	mAgentPlanes[3] = planeFromPoints(frust[1], frust[0], frust[4]);
 
 	//cache plane octant facing mask for use in AABBInFrustum
 	for (U32 i = 0; i < mPlaneCount; i++)
 	{
-		mAgentPlanes[i].mask = calcPlaneMask(mAgentPlanes[i].p);
+		mPlaneMask[i] = calcPlaneMask(mAgentPlanes[i]);
 	}
 }
 
diff --git a/indra/llmath/llcamera.h b/indra/llmath/llcamera.h
index d6c5f7bbb1..c40e819dcf 100644
--- a/indra/llmath/llcamera.h
+++ b/indra/llmath/llcamera.h
@@ -37,6 +37,7 @@
 #include "llmath.h"
 #include "llcoordframe.h"
 #include "llplane.h"
+#include "llvector4a.h"
 
 const F32 DEFAULT_FIELD_OF_VIEW 	= 60.f * DEG_TO_RAD;
 const F32 DEFAULT_ASPECT_RATIO 		= 640.f / 480.f;
@@ -79,6 +80,14 @@ class LLCamera
 : 	public LLCoordFrame
 {
 public:
+	
+	LLCamera(const LLCamera& rhs)
+	{
+		*this = rhs;
+	}
+
+	const LLCamera& operator=(const LLCamera& rhs);
+	
 	enum {
 		PLANE_LEFT = 0,
 		PLANE_RIGHT = 1,
@@ -129,13 +138,9 @@ private:
 	LLPlane mWorldPlanes[PLANE_NUM];
 	LLPlane mHorizPlanes[HORIZ_PLANE_NUM];
 
-	struct frustum_plane
-	{
-		frustum_plane() : mask(0) {}
-		LLPlane p;
-		U8 mask;
-	};
-	frustum_plane mAgentPlanes[7];  //frustum planes in agent space a la gluUnproject (I'm a bastard, I know) - DaveP
+	LLPlane* mAgentPlanes;  //frustum planes in agent space a la gluUnproject (I'm a bastard, I know) - DaveP
+	U8 mAgentPlaneBuffer[sizeof(LLPlane)*8];
+	U8 mPlaneMask[7];
 
 	U32 mPlaneCount;  //defaults to 6, if setUserClipPlane is called, uses user supplied clip plane in
 
@@ -143,12 +148,14 @@ private:
 public:
 	LLVector3 mAgentFrustum[8];  //8 corners of 6-plane frustum
 	F32	mFrustumCornerDist;		//distance to corner of frustum against far clip plane
-	LLPlane& getAgentPlane(U32 idx) { return mAgentPlanes[idx].p; }
+	LLPlane& getAgentPlane(U32 idx) { return mAgentPlanes[idx]; }
 
 public:
 	LLCamera();
 	LLCamera(F32 vertical_fov_rads, F32 aspect_ratio, S32 view_height_in_pixels, F32 near_plane, F32 far_plane);
-	virtual ~LLCamera(){} // no-op virtual destructor
+	virtual ~LLCamera();
+	
+	void alignPlanes();
 
 	void setUserClipPlane(LLPlane plane);
 	void disableUserClipPlane();
@@ -199,8 +206,8 @@ public:
 	S32 sphereInFrustum(const LLVector3 &center, const F32 radius) const;
 	S32 pointInFrustum(const LLVector3 &point) const { return sphereInFrustum(point, 0.0f); }
 	S32 sphereInFrustumFull(const LLVector3 &center, const F32 radius) const { return sphereInFrustum(center, radius); }
-	S32 AABBInFrustum(const LLVector3 &center, const LLVector3& radius);
-	S32 AABBInFrustumNoFarClip(const LLVector3 &center, const LLVector3& radius);
+	S32 AABBInFrustum(const LLVector4a& center, const LLVector4a& radius);
+	S32 AABBInFrustumNoFarClip(const LLVector4a& center, const LLVector4a& radius);
 
 	//does a quick 'n dirty sphere-sphere check
 	S32 sphereInFrustumQuick(const LLVector3 &sphere_center, const F32 radius); 
diff --git a/indra/llmath/lloctree.h b/indra/llmath/lloctree.h
index 2f34fb1bb0..ae2259dba0 100644
--- a/indra/llmath/lloctree.h
+++ b/indra/llmath/lloctree.h
@@ -35,6 +35,7 @@
 
 #include "lltreenode.h"
 #include "v3math.h"
+#include "llvector4a.h"
 #include <vector>
 #include <set>
 
@@ -44,7 +45,7 @@
 #define OCT_ERRS LL_WARNS("OctreeErrors")
 #endif
 
-#define LL_OCTREE_PARANOIA_CHECK 0
+#define LL_OCTREE_PARANOIA_CHECK 1
 #if LL_DARWIN
 #define LL_OCTREE_MAX_CAPACITY 32
 #else
@@ -73,6 +74,13 @@ public:
 };
 
 template <class T>
+class LLOctreeTravelerDepthFirst : public LLOctreeTraveler<T>
+{
+public:
+	virtual void traverse(const LLOctreeNode<T>* node);
+};
+
+template <class T>
 class LLOctreeNode : public LLTreeNode<T>
 {
 public:
@@ -87,23 +95,22 @@ public:
 	typedef LLOctreeNode<T>		oct_node;
 	typedef LLOctreeListener<T>	oct_listener;
 
-	static const U8 OCTANT_POSITIVE_X = 0x01;
-	static const U8 OCTANT_POSITIVE_Y = 0x02;
-	static const U8 OCTANT_POSITIVE_Z = 0x04;
-		
-	LLOctreeNode(	LLVector3d center, 
-					LLVector3d size, 
+	LLOctreeNode(	const LLVector4a& center, 
+					const LLVector4a& size, 
 					BaseType* parent, 
-					U8 octant = 255)
+					S32 octant = -1)
 	:	mParent((oct_node*)parent), 
-		mCenter(center), 
-		mSize(size), 
 		mOctant(octant) 
 	{ 
+		mD = (LLVector4a*) _mm_malloc(sizeof(LLVector4a)*4, 16);
+
+		mD[CENTER] = center;
+		mD[SIZE] = size;
+
 		updateMinMax();
-		if ((mOctant == 255) && mParent)
+		if ((mOctant == -1) && mParent)
 		{
-			mOctant = ((oct_node*) mParent)->getOctant(mCenter.mdV);
+			mOctant = ((oct_node*) mParent)->getOctant(mD[CENTER]);
 		}
 
 		clearChildren();
@@ -117,43 +124,30 @@ public:
 		{
 			delete getChild(i);
 		} 
+
+		_mm_free(mD);
 	}
 
 	inline const BaseType* getParent()	const			{ return mParent; }
-	inline void setParent(BaseType* parent)			{ mParent = (oct_node*) parent; }
-	inline const LLVector3d& getCenter() const			{ return mCenter; }
-	inline const LLVector3d& getSize() const			{ return mSize; }
-	inline void setCenter(LLVector3d center)			{ mCenter = center; }
-	inline void setSize(LLVector3d size)				{ mSize = size; }
-    inline oct_node* getNodeAt(T* data)				{ return getNodeAt(data->getPositionGroup(), data->getBinRadius()); }
-	inline U8 getOctant() const						{ return mOctant; }
-	inline void setOctant(U8 octant)					{ mOctant = octant; }
+	inline void setParent(BaseType* parent)				{ mParent = (oct_node*) parent; }
+	inline const LLVector4a& getCenter() const			{ return mD[CENTER]; }
+	inline const LLVector4a& getSize() const			{ return mD[SIZE]; }
+	inline void setCenter(const LLVector4a& center)		{ mD[CENTER] = center; }
+	inline void setSize(const LLVector4a& size)			{ mD[SIZE] = size; }
+    inline oct_node* getNodeAt(T* data)					{ return getNodeAt(data->getPositionGroup(), data->getBinRadius()); }
+	inline S32 getOctant() const						{ return mOctant; }
+	inline void setOctant(S32 octant)					{ mOctant = octant; }
 	inline const oct_node*	getOctParent() const		{ return (const oct_node*) getParent(); }
 	inline oct_node* getOctParent() 					{ return (oct_node*) getParent(); }
 	
-	U8 getOctant(const F64 pos[]) const	//get the octant pos is in
+	S32 getOctant(const LLVector4a& pos) const			//get the octant pos is in
 	{
-		U8 ret = 0;
-
-		if (pos[0] > mCenter.mdV[0])
-		{
-			ret |= OCTANT_POSITIVE_X;
-		}
-		if (pos[1] > mCenter.mdV[1])
-		{
-			ret |= OCTANT_POSITIVE_Y;
-		}
-		if (pos[2] > mCenter.mdV[2])
-		{
-			ret |= OCTANT_POSITIVE_Z;
-		}
-
-		return ret;
+		return pos.greaterThan4(mD[CENTER]).getComparisonMask() & 0x7;
 	}
 	
-	inline bool isInside(const LLVector3d& pos, const F64& rad) const
+	inline bool isInside(const LLVector4a& pos, const F32& rad) const
 	{
-		return rad <= mSize.mdV[0]*2.0 && isInside(pos); 
+		return rad <= mD[SIZE][0]*2.f && isInside(pos); 
 	}
 
 	inline bool isInside(T* data) const			
@@ -161,29 +155,27 @@ public:
 		return isInside(data->getPositionGroup(), data->getBinRadius());
 	}
 
-	bool isInside(const LLVector3d& pos) const
+	bool isInside(const LLVector4a& pos) const
 	{
-		const F64& x = pos.mdV[0];
-		const F64& y = pos.mdV[1];
-		const F64& z = pos.mdV[2];
-			
-		if (x > mMax.mdV[0] || x <= mMin.mdV[0] ||
-			y > mMax.mdV[1] || y <= mMin.mdV[1] ||
-			z > mMax.mdV[2] || z <= mMin.mdV[2])
+		S32 gt = pos.greaterThan4(mD[MAX]).getComparisonMask() & 0x7;
+		if (gt)
 		{
 			return false;
 		}
-		
+
+		S32 lt = pos.lessEqual4(mD[MIN]).getComparisonMask() & 0x7;
+		if (lt)
+		{
+			return false;
+		}
+				
 		return true;
 	}
 	
 	void updateMinMax()
 	{
-		for (U32 i = 0; i < 3; i++)
-		{
-			mMax.mdV[i] = mCenter.mdV[i] + mSize.mdV[i];
-			mMin.mdV[i] = mCenter.mdV[i] - mSize.mdV[i];
-		}
+		mD[MAX].setAdd(mD[CENTER], mD[SIZE]);
+		mD[MIN].setSub(mD[CENTER], mD[SIZE]);
 	}
 
 	inline oct_listener* getOctListener(U32 index) 
@@ -196,34 +188,34 @@ public:
 		return contains(xform->getBinRadius());
 	}
 
-	bool contains(F64 radius)
+	bool contains(F32 radius)
 	{
 		if (mParent == NULL)
 		{	//root node contains nothing
 			return false;
 		}
 
-		F64 size = mSize.mdV[0];
-		F64 p_size = size * 2.0;
+		F32 size = mD[SIZE][0];
+		F32 p_size = size * 2.f;
 
-		return (radius <= 0.001 && size <= 0.001) ||
+		return (radius <= 0.001f && size <= 0.001f) ||
 				(radius <= p_size && radius > size);
 	}
 
-	static void pushCenter(LLVector3d &center, const LLVector3d &size, const T* data)
+	static void pushCenter(LLVector4a &center, const LLVector4a &size, const T* data)
 	{
-		const LLVector3d& pos = data->getPositionGroup();
-		for (U32 i = 0; i < 3; i++)
-		{
-			if (pos.mdV[i] > center.mdV[i])
-			{
-				center.mdV[i] += size.mdV[i];
-			}
-			else 
-			{
-				center.mdV[i] -= size.mdV[i];
-			}
-		}
+		const LLVector4a& pos = data->getPositionGroup();
+
+		LLVector4a gt = pos.greaterThan4(center);
+
+		LLVector4a up;
+		up.mQ = _mm_and_ps(size.mQ, gt.mQ);
+
+		LLVector4a down;
+		down.mQ = _mm_andnot_ps(gt.mQ, size.mQ);
+
+		center.add(up);
+		center.sub(down);
 	}
 
 	void accept(oct_traveler* visitor)				{ visitor->visit(this); }
@@ -242,21 +234,21 @@ public:
 	void accept(tree_traveler* visitor) const		{ visitor->visit(this); }
 	void accept(oct_traveler* visitor) const		{ visitor->visit(this); }
 	
-	oct_node* getNodeAt(const LLVector3d& pos, const F64& rad)
+	oct_node* getNodeAt(const LLVector4a& pos, const F32& rad)
 	{ 
 		LLOctreeNode<T>* node = this;
 
 		if (node->isInside(pos, rad))
 		{		
 			//do a quick search by octant
-			U8 octant = node->getOctant(pos.mdV);
+			S32 octant = node->getOctant(pos);
 			BOOL keep_going = TRUE;
 
 			//traverse the tree until we find a node that has no node
 			//at the appropriate octant or is smaller than the object.  
 			//by definition, that node is the smallest node that contains 
 			// the data
-			while (keep_going && node->getSize().mdV[0] >= rad)
+			while (keep_going && node->getSize()[0] >= rad)
 			{	
 				keep_going = FALSE;
 				for (U32 i = 0; i < node->getChildCount() && !keep_going; i++)
@@ -264,7 +256,7 @@ public:
 					if (node->getChild(i)->getOctant() == octant)
 					{
 						node = node->getChild(i);
-						octant = node->getOctant(pos.mdV);
+						octant = node->getOctant(pos);
 						keep_going = TRUE;
 					}
 				}
@@ -282,7 +274,7 @@ public:
 	{
 		if (data == NULL)
 		{
-			//OCT_ERRS << "!!! INVALID ELEMENT ADDED TO OCTREE BRANCH !!!" << llendl;
+			OCT_ERRS << "!!! INVALID ELEMENT ADDED TO OCTREE BRANCH !!!" << llendl;
 			return false;
 		}
 		LLOctreeNode<T>* parent = getOctParent();
@@ -292,7 +284,7 @@ public:
 		{
 			if (getElementCount() < LL_OCTREE_MAX_CAPACITY &&
 				(contains(data->getBinRadius()) ||
-				(data->getBinRadius() > getSize().mdV[0] &&
+				(data->getBinRadius() > getSize()[0] &&
 				parent && parent->getElementCount() >= LL_OCTREE_MAX_CAPACITY))) 
 			{ //it belongs here
 #if LL_OCTREE_PARANOIA_CHECK
@@ -323,16 +315,22 @@ public:
 				}
 				
 				//it's here, but no kids are in the right place, make a new kid
-				LLVector3d center(getCenter());
-				LLVector3d size(getSize()*0.5);
+				LLVector4a center = getCenter();
+				LLVector4a size = getSize();
+				size.mul(0.5f);
 		        		
 				//push center in direction of data
 				LLOctreeNode<T>::pushCenter(center, size, data);
 
 				// handle case where floating point number gets too small
-				if( llabs(center.mdV[0] - getCenter().mdV[0]) < F_APPROXIMATELY_ZERO &&
-					llabs(center.mdV[1] - getCenter().mdV[1]) < F_APPROXIMATELY_ZERO &&
-					llabs(center.mdV[2] - getCenter().mdV[2]) < F_APPROXIMATELY_ZERO)
+				LLVector4a val;
+				val.setSub(center, getCenter());
+				val.setAbs(val);
+				LLVector4a app_zero;
+				app_zero.mQ = F_APPROXIMATELY_ZERO_4A;
+				S32 lt = val.lessThan4(app_zero).getComparisonMask() & 0x7;
+
+				if( lt == 0x7 )
 				{
 					mData.insert(data);
 					BaseType::insert(data);
@@ -350,7 +348,7 @@ public:
 				//make sure no existing node matches this position
 				for (U32 i = 0; i < getChildCount(); i++)
 				{
-					if (mChild[i]->getCenter() == center)
+					if (mChild[i]->getCenter().equal3(center))
 					{
 						OCT_ERRS << "Octree detected duplicate child center and gave up." << llendl;
 						return false;
@@ -368,7 +366,7 @@ public:
 		else 
 		{
 			//it's not in here, give it to the root
-			//OCT_ERRS << "Octree insertion failed, starting over from root!" << llendl;
+			OCT_ERRS << "Octree insertion failed, starting over from root!" << llendl;
 
 			oct_node* node = this;
 
@@ -475,13 +473,19 @@ public:
 	void addChild(oct_node* child, BOOL silent = FALSE) 
 	{
 #if LL_OCTREE_PARANOIA_CHECK
+
+		if (child->getSize().equal3(getSize()))
+		{
+			OCT_ERRS << "Child size is same as parent size!" << llendl;
+		}
+
 		for (U32 i = 0; i < getChildCount(); i++)
 		{
-			if(mChild[i]->getSize() != child->getSize()) 
+			if(!mChild[i]->getSize().equal3(child->getSize())) 
 			{
 				OCT_ERRS <<"Invalid octree child size." << llendl;
 			}
-			if (mChild[i]->getCenter() == child->getCenter())
+			if (mChild[i]->getCenter().equal3(child->getCenter()))
 			{
 				OCT_ERRS <<"Duplicate octree child position." << llendl;
 			}
@@ -506,7 +510,7 @@ public:
 		}
 	}
 
-	void removeChild(U8 index, BOOL destroy = FALSE)
+	void removeChild(S32 index, BOOL destroy = FALSE)
 	{
 		for (U32 i = 0; i < this->getListenerCount(); i++)
 		{
@@ -547,18 +551,26 @@ public:
 			}
 		}
 
-		//OCT_ERRS << "Octree failed to delete requested child." << llendl;
+		OCT_ERRS << "Octree failed to delete requested child." << llendl;
 	}
 
 protected:	
+	typedef enum
+	{
+		CENTER = 0,
+		SIZE = 1,
+		MAX = 2,
+		MIN = 3
+	} eDName;
+
+	LLVector4a* mD;
+	
+	oct_node* mParent;
+	S32 mOctant;
+
 	child_list mChild;
 	element_list mData;
-	oct_node* mParent;
-	LLVector3d mCenter;
-	LLVector3d mSize;
-	LLVector3d mMax;
-	LLVector3d mMin;
-	U8 mOctant;
+		
 };
 
 //just like a regular node, except it might expand on insert and compress on balance
@@ -569,9 +581,9 @@ public:
 	typedef LLOctreeNode<T>	BaseType;
 	typedef LLOctreeNode<T>		oct_node;
 
-	LLOctreeRoot(	LLVector3d center, 
-					LLVector3d size, 
-					BaseType* parent)
+	LLOctreeRoot(const LLVector4a& center, 
+				 const LLVector4a& size, 
+				 BaseType* parent)
 	:	BaseType(center, size, parent)
 	{
 	}
@@ -612,28 +624,33 @@ public:
 	{
 		if (data == NULL) 
 		{
-			//OCT_ERRS << "!!! INVALID ELEMENT ADDED TO OCTREE ROOT !!!" << llendl;
+			OCT_ERRS << "!!! INVALID ELEMENT ADDED TO OCTREE ROOT !!!" << llendl;
 			return false;
 		}
 		
 		if (data->getBinRadius() > 4096.0)
 		{
-			//OCT_ERRS << "!!! ELEMENT EXCEEDS MAXIMUM SIZE IN OCTREE ROOT !!!" << llendl;
+			OCT_ERRS << "!!! ELEMENT EXCEEDS MAXIMUM SIZE IN OCTREE ROOT !!!" << llendl;
 			return false;
 		}
 		
-		const F64 MAX_MAG = 1024.0*1024.0;
+		LLVector4a MAX_MAG;
+		MAX_MAG.splat(1024.f*1024.f);
+
+		const LLVector4a& v = data->getPositionGroup();
+
+		LLVector4a val;
+		val.setSub(v, mD[CENTER]);
+		val.setAbs(val);
+		S32 lt = val.lessThan4(MAX_MAG).getComparisonMask() & 0x7;
 
-		const LLVector3d& v = data->getPositionGroup();
-		if (!(fabs(v.mdV[0]-this->mCenter.mdV[0]) < MAX_MAG &&
-		      fabs(v.mdV[1]-this->mCenter.mdV[1]) < MAX_MAG &&
-		      fabs(v.mdV[2]-this->mCenter.mdV[2]) < MAX_MAG))
+		if (lt != 0x7)
 		{
-			//OCT_ERRS << "!!! ELEMENT EXCEEDS RANGE OF SPATIAL PARTITION !!!" << llendl;
+			OCT_ERRS << "!!! ELEMENT EXCEEDS RANGE OF SPATIAL PARTITION !!!" << llendl;
 			return false;
 		}
 
-		if (this->getSize().mdV[0] > data->getBinRadius() && isInside(data->getPositionGroup()))
+		if (this->getSize()[0] > data->getBinRadius() && isInside(data->getPositionGroup()))
 		{
 			//we got it, just act like a branch
 			oct_node* node = getNodeAt(data);
@@ -649,31 +666,34 @@ public:
 		else if (this->getChildCount() == 0)
 		{
 			//first object being added, just wrap it up
-			while (!(this->getSize().mdV[0] > data->getBinRadius() && isInside(data->getPositionGroup())))
+			while (!(this->getSize()[0] > data->getBinRadius() && isInside(data->getPositionGroup())))
 			{
-				LLVector3d center, size;
+				LLVector4a center, size;
 				center = this->getCenter();
 				size = this->getSize();
 				LLOctreeNode<T>::pushCenter(center, size, data);
 				this->setCenter(center);
-				this->setSize(size*2);
+				size.mul(2.f);
+				this->setSize(size);
 				this->updateMinMax();
 			}
 			LLOctreeNode<T>::insert(data);
 		}
 		else
 		{
-			while (!(this->getSize().mdV[0] > data->getBinRadius() && isInside(data->getPositionGroup())))
+			while (!(this->getSize()[0] > data->getBinRadius() && isInside(data->getPositionGroup())))
 			{
 				//the data is outside the root node, we need to grow
-				LLVector3d center(this->getCenter());
-				LLVector3d size(this->getSize());
+				LLVector4a center(this->getCenter());
+				LLVector4a size(this->getSize());
 
 				//expand this node
-				LLVector3d newcenter(center);
+				LLVector4a newcenter(center);
 				LLOctreeNode<T>::pushCenter(newcenter, size, data);
 				this->setCenter(newcenter);
-				this->setSize(size*2);
+				LLVector4a size2 = size;
+				size2.mul(2.f);
+				this->setSize(size2);
 				this->updateMinMax();
 
 				//copy our children to a new branch
@@ -710,4 +730,15 @@ void LLOctreeTraveler<T>::traverse(const LLOctreeNode<T>* node)
 		traverse(node->getChild(i));
 	}
 }
+
+template <class T>
+void LLOctreeTravelerDepthFirst<T>::traverse(const LLOctreeNode<T>* node)
+{
+	for (U32 i = 0; i < node->getChildCount(); i++)
+	{
+		traverse(node->getChild(i));
+	}
+	node->accept(this);
+}
+
 #endif
diff --git a/indra/llmath/lltreenode.h b/indra/llmath/lltreenode.h
index ee9836241a..e6d2521b2a 100644
--- a/indra/llmath/lltreenode.h
+++ b/indra/llmath/lltreenode.h
@@ -34,6 +34,9 @@
 
 #include "stdtypes.h"
 #include "xform.h"
+#include "llpointer.h"
+#include "llrefcount.h"
+
 #include <vector>
 
 template <class T> class LLTreeNode;
diff --git a/indra/llmath/llvolume.cpp b/indra/llmath/llvolume.cpp
index 9b6e2488e6..72833c019f 100644
--- a/indra/llmath/llvolume.cpp
+++ b/indra/llmath/llvolume.cpp
@@ -45,8 +45,10 @@
 #include "m4math.h"
 #include "m3math.h"
 #include "llmatrix4a.h"
+#include "lloctree.h"
 #include "lldarray.h"
 #include "llvolume.h"
+#include "llvolumeoctree.h"
 #include "llstl.h"
 #include "llsdserialize.h"
 #include "llvector4a.h"
@@ -133,21 +135,20 @@ BOOL LLLineSegmentBoxIntersect(const F32* start, const F32* end, const F32* cent
 }
 
 
+
 // intersect test between triangle vert0, vert1, vert2 and a ray from orig in direction dir.
 // returns TRUE if intersecting and returns barycentric coordinates in intersection_a, intersection_b,
 // and returns the intersection point along dir in intersection_t.
 
 // Moller-Trumbore algorithm
 BOOL LLTriangleRayIntersect(const LLVector4a& vert0, const LLVector4a& vert1, const LLVector4a& vert2, const LLVector4a& orig, const LLVector4a& dir,
-							F32* intersection_a, F32* intersection_b, F32* intersection_t, BOOL two_sided)
+							F32& intersection_a, F32& intersection_b, F32& intersection_t)
 {
-	F32 u, v, t;
 	
 	/* find vectors for two edges sharing vert0 */
 	LLVector4a edge1;
 	edge1.setSub(vert1, vert0);
 	
-	
 	LLVector4a edge2;
 	edge2.setSub(vert2, vert0);
 
@@ -156,87 +157,116 @@ BOOL LLTriangleRayIntersect(const LLVector4a& vert0, const LLVector4a& vert1, co
 	pvec.setCross3(dir, edge2);
 
 	/* if determinant is near zero, ray lies in plane of triangle */
-	F32 det = edge1.dot3(pvec);
-
-	if (!two_sided)
+	LLVector4a det;
+	det.setAllDot3(edge1, pvec);
+	
+	if (det.greaterEqual4(LLVector4a::getApproximatelyZero()).getComparisonMask() & 0x7)
 	{
-		if (det < F_APPROXIMATELY_ZERO)
-		{
-			return FALSE;
-		}
-
 		/* calculate distance from vert0 to ray origin */
 		LLVector4a tvec;
 		tvec.setSub(orig, vert0);
 
 		/* calculate U parameter and test bounds */
-		u = tvec.dot3(pvec);	
+		LLVector4a u;
+		u.setAllDot3(tvec,pvec);
 
-		if (u < 0.f || u > det)
+		if ((u.greaterEqual4(LLVector4a::getZero()).getComparisonMask() & 0x7) &&
+			(u.lessEqual4(det).getComparisonMask() & 0x7))
 		{
-			return FALSE;
+			/* prepare to test V parameter */
+			LLVector4a qvec;
+			qvec.setCross3(tvec, edge1);
+			
+			/* calculate V parameter and test bounds */
+			LLVector4a v;
+			v.setAllDot3(dir, qvec);
+
+			
+			//if (!(v < 0.f || u + v > det))
+
+			LLVector4a sum_uv;
+			sum_uv.setAdd(u, v);
+
+			S32 v_gequal = v.greaterEqual4(LLVector4a::getZero()).getComparisonMask() & 0x7;
+			S32 sum_lequal = sum_uv.lessEqual4(det).getComparisonMask() & 0x7;
+
+			if (v_gequal  && sum_lequal)
+			{
+				/* calculate t, scale parameters, ray intersects triangle */
+				LLVector4a t;
+				t.setAllDot3(edge2,qvec);
+
+				t.div(det);
+				u.div(det);
+				v.div(det);
+				
+				intersection_a = u[0];
+				intersection_b = v[0];
+				intersection_t = t[0];
+				return TRUE;
+			}
 		}
-	
-		/* prepare to test V parameter */
-		LLVector4a qvec;
-		qvec.setCross3(tvec, edge1);
+	}
 		
-		/* calculate V parameter and test bounds */
-		v = dir.dot3(qvec);
-		if (v < 0.f || u + v > det)
-		{
-			return FALSE;
-		}
+	return FALSE;
+} 
 
-		/* calculate t, scale parameters, ray intersects triangle */
-		t = edge2.dot3(qvec);
-		F32 inv_det = 1.0 / det;
-		t *= inv_det;
-		u *= inv_det;
-		v *= inv_det;
-	}
+BOOL LLTriangleRayIntersectTwoSided(const LLVector4a& vert0, const LLVector4a& vert1, const LLVector4a& vert2, const LLVector4a& orig, const LLVector4a& dir,
+							F32& intersection_a, F32& intersection_b, F32& intersection_t)
+{
+	F32 u, v, t;
 	
-	else // two sided
-			{
-		if (det > -F_APPROXIMATELY_ZERO && det < F_APPROXIMATELY_ZERO)
-				{
-			return FALSE;
-				}
-		F32 inv_det = 1.0 / det;
+	/* find vectors for two edges sharing vert0 */
+	LLVector4a edge1;
+	edge1.setSub(vert1, vert0);
+	
+	
+	LLVector4a edge2;
+	edge2.setSub(vert2, vert0);
 
-		/* calculate distance from vert0 to ray origin */
-		LLVector4a tvec;
-		tvec.setSub(orig, vert0);
-		
-		/* calculate U parameter and test bounds */
-		u = (tvec.dot3(pvec)) * inv_det;
-		if (u < 0.f || u > 1.f)
-		{
-			return FALSE;
-			}
+	/* begin calculating determinant - also used to calculate U parameter */
+	LLVector4a pvec;
+	pvec.setCross3(dir, edge2);
 
-		/* prepare to test V parameter */
-		LLVector4a qvec;
-		qvec.setSub(tvec, edge1);
-		
-		/* calculate V parameter and test bounds */
-		v = (dir.dot3(qvec)) * inv_det;
-		
-		if (v < 0.f || u + v > 1.f)
-		{
-			return FALSE;
-		}
+	/* if determinant is near zero, ray lies in plane of triangle */
+	F32 det = edge1.dot3(pvec);
 
-		/* calculate t, ray intersects triangle */
-		t = (edge2.dot3(qvec)) * inv_det;
+	
+	if (det > -F_APPROXIMATELY_ZERO && det < F_APPROXIMATELY_ZERO)
+	{
+		return FALSE;
+	}
+
+	F32 inv_det = 1.f / det;
+
+	/* calculate distance from vert0 to ray origin */
+	LLVector4a tvec;
+	tvec.setSub(orig, vert0);
+	
+	/* calculate U parameter and test bounds */
+	u = (tvec.dot3(pvec)) * inv_det;
+	if (u < 0.f || u > 1.f)
+	{
+		return FALSE;
+	}
+
+	/* prepare to test V parameter */
+	tvec.sub(edge1);
+		
+	/* calculate V parameter and test bounds */
+	v = (dir.dot3(tvec)) * inv_det;
+	
+	if (v < 0.f || u + v > 1.f)
+	{
+		return FALSE;
 	}
+
+	/* calculate t, ray intersects triangle */
+	t = (edge2.dot3(tvec)) * inv_det;
 	
-	if (intersection_a != NULL)
-		*intersection_a = u;
-	if (intersection_b != NULL)
-		*intersection_b = v;
-	if (intersection_t != NULL)
-		*intersection_t = t;
+	intersection_a = u;
+	intersection_b = v;
+	intersection_t = t;
 	
 	
 	return TRUE;
@@ -244,7 +274,7 @@ BOOL LLTriangleRayIntersect(const LLVector4a& vert0, const LLVector4a& vert1, co
 
 //helper for non-aligned vectors
 BOOL LLTriangleRayIntersect(const LLVector3& vert0, const LLVector3& vert1, const LLVector3& vert2, const LLVector3& orig, const LLVector3& dir,
-							F32* intersection_a, F32* intersection_b, F32* intersection_t, BOOL two_sided)
+							F32& intersection_a, F32& intersection_b, F32& intersection_t, BOOL two_sided)
 {
 	LLVector4a vert0a, vert1a, vert2a, origa, dira;
 	vert0a.load3(vert0.mV);
@@ -253,10 +283,91 @@ BOOL LLTriangleRayIntersect(const LLVector3& vert0, const LLVector3& vert1, cons
 	origa.load3(orig.mV);
 	dira.load3(dir.mV);
 
-	return LLTriangleRayIntersect(vert0a, vert1a, vert2a, origa, dira, 
-			intersection_a, intersection_b, intersection_t, two_sided);
+	if (two_sided)
+	{
+		return LLTriangleRayIntersectTwoSided(vert0a, vert1a, vert2a, origa, dira, 
+				intersection_a, intersection_b, intersection_t);
+	}
+	else
+	{
+		return LLTriangleRayIntersect(vert0a, vert1a, vert2a, origa, dira, 
+				intersection_a, intersection_b, intersection_t);
+	}
 }
 
+class LLVolumeOctreeRebound : public LLOctreeTravelerDepthFirst<LLVolumeTriangle>
+{
+public:
+	const LLVolumeFace* mFace;
+
+	LLVolumeOctreeRebound(const LLVolumeFace* face)
+	{
+		mFace = face;
+	}
+
+	virtual void visit(const LLOctreeNode<LLVolumeTriangle>* branch)
+	{
+		LLVolumeOctreeListener* node = (LLVolumeOctreeListener*) branch->getListener(0);
+
+		LLVector4a& min = node->mExtents[0];
+		LLVector4a& max = node->mExtents[1];
+
+		if (branch->getElementCount() != 0)
+		{
+			const LLVolumeTriangle* tri = *(branch->getData().begin());
+						
+			min = *(tri->mV[0]);
+			max = *(tri->mV[0]);
+			
+			for (LLOctreeNode<LLVolumeTriangle>::const_element_iter iter = 
+				branch->getData().begin(); iter != branch->getData().end(); ++iter)
+			{
+				//stretch by triangles in node
+				tri = *iter;
+				
+				min.setMin(*tri->mV[0]);
+				min.setMin(*tri->mV[1]);
+				min.setMin(*tri->mV[2]);
+
+				max.setMax(*tri->mV[0]);
+				max.setMax(*tri->mV[1]);
+				max.setMax(*tri->mV[2]);
+			}
+
+			for (S32 i = 0; i < branch->getChildCount(); ++i)
+			{  //stretch by child extents
+				LLVolumeOctreeListener* child = (LLVolumeOctreeListener*) branch->getChild(i)->getListener(0);
+				min.setMin(child->mExtents[0]);
+				max.setMax(child->mExtents[1]);
+			}
+		}
+		else if (branch->getChildCount() != 0)
+		{
+			LLVolumeOctreeListener* child = (LLVolumeOctreeListener*) branch->getChild(0)->getListener(0);
+
+			min = child->mExtents[0];
+			max = child->mExtents[1];
+
+			for (S32 i = 1; i < branch->getChildCount(); ++i)
+			{  //stretch by child extents
+				child = (LLVolumeOctreeListener*) branch->getChild(i)->getListener(0);
+				min.setMin(child->mExtents[0]);
+				max.setMax(child->mExtents[1]);
+			}
+		}
+		else
+		{
+			llerrs << "WTF? Empty leaf" << llendl;
+		}
+		
+		node->mBounds[0].setAdd(min, max);
+		node->mBounds[0].mul(0.5f);
+
+		node->mBounds[1].setSub(max,min);
+		node->mBounds[1].mul(0.5f);
+	}
+};
+
 
 //-------------------------------------------------------------------
 // statics
@@ -4244,6 +4355,7 @@ S32 LLVolume::lineSegmentIntersect(const LLVector3& start, const LLVector3& end,
 
 }
 
+
 S32 LLVolume::lineSegmentIntersect(const LLVector4a& start, const LLVector4a& end, 
 								   S32 face,
 								   LLVector3* intersection,LLVector2* tex_coord, LLVector3* normal, LLVector3* bi_normal)
@@ -4282,72 +4394,25 @@ S32 LLVolume::lineSegmentIntersect(const LLVector4a& start, const LLVector4a& en
 		LLVector4a box_size;
 		box_size.setSub(face.mExtents[1], face.mExtents[0]);
 
-        if (LLLineSegmentBoxIntersect(start.getF32(), end.getF32(), box_center.getF32(), box_size.getF32()))
+        if (LLLineSegmentBoxIntersect(start, end, box_center, box_size))
 		{
 			if (bi_normal != NULL) // if the caller wants binormals, we may need to generate them
 			{
 				genBinormals(i);
 			}
-			
-			LLVector4a* p = (LLVector4a*) face.mPositions;
 
-			for (U32 tri = 0; tri < face.mNumIndices/3; tri++) 
+			if (!face.mOctree)
 			{
-				S32 index1 = face.mIndices[tri*3+0];
-				S32 index2 = face.mIndices[tri*3+1];
-				S32 index3 = face.mIndices[tri*3+2];
-
-				F32 a, b, t;
+				face.createOctree();
+			}
 			
-				if (LLTriangleRayIntersect(p[index1],
-					p[index2],
-					p[index3],
-					start, dir, &a, &b, &t, FALSE))
-				{
-					if ((t >= 0.f) &&      // if hit is after start
-						(t <= 1.f) &&      // and before end
-						(t < closest_t))   // and this hit is closer
-					{
-						closest_t = t;
-						hit_face = i;
-
-						if (intersection != NULL)
-						{
-							LLVector4a intersect = dir;
-							intersect.mul(closest_t);
-							intersect.add(start);
-							intersection->set(intersect.getF32());
-						}
-
-
-						if (tex_coord != NULL)
-						{
-							LLVector2* tc = (LLVector2*) face.mTexCoords;
-							*tex_coord = ((1.f - a - b)  * tc[index1] +
-								a              * tc[index2] +
-								b              * tc[index3]);
-
-						}
-
-						if (normal != NULL)
-						{
-							LLVector4* norm = (LLVector4*) face.mNormals;
-
-							*normal    = ((1.f - a - b)  * LLVector3(norm[index1]) + 
-								a              * LLVector3(norm[index2]) +
-								b              * LLVector3(norm[index3]));
-						}
-
-						if (bi_normal != NULL)
-						{
-							LLVector4* binormal = (LLVector4*) face.mBinormals;
-							*bi_normal = ((1.f - a - b)  * LLVector3(binormal[index1]) + 
-									a              * LLVector3(binormal[index2]) +
-									b              * LLVector3(binormal[index3]));
-						}
+			LLVector4a* p = (LLVector4a*) face.mPositions;
 
-					}
-				}
+			LLOctreeTriangleRayIntersect intersect(start, dir, &face, &closest_t, intersection, tex_coord, normal, bi_normal);
+			intersect.traverse(face.mOctree);
+			if (intersect.mHitFace)
+			{
+				hit_face = i;
 			}
 		}		
 	}
@@ -5128,13 +5193,29 @@ LLVolumeFace::LLVolumeFace() :
 	mBinormals(NULL),
 	mTexCoords(NULL),
 	mIndices(NULL),
-	mWeights(NULL)
+	mWeights(NULL),
+	mOctree(NULL)
 {
 	mExtents = (LLVector4a*) _mm_malloc(48, 16);
 	mCenter = mExtents+2;
 }
 
 LLVolumeFace::LLVolumeFace(const LLVolumeFace& src)
+:	mID(0),
+	mTypeMask(0),
+	mBeginS(0),
+	mBeginT(0),
+	mNumS(0),
+	mNumT(0),
+	mNumVertices(0),
+	mNumIndices(0),
+	mPositions(NULL),
+	mNormals(NULL),
+	mBinormals(NULL),
+	mTexCoords(NULL),
+	mIndices(NULL),
+	mWeights(NULL),
+	mOctree(NULL)
 { 
 	mExtents = (LLVector4a*) _mm_malloc(48, 16);
 	mCenter = mExtents+2;
@@ -5157,13 +5238,9 @@ LLVolumeFace& LLVolumeFace::operator=(const LLVolumeFace& src)
 
 	mNumVertices = 0;
 	mNumIndices = 0;
-	mPositions = NULL;
-	mNormals = NULL;
-	mBinormals = NULL;
-	mTexCoords = NULL;
-	mWeights = NULL;
-	mIndices = NULL;
 
+	freeData();
+	
 	LLVector4a::memcpyNonAliased16((F32*) mExtents, (F32*) src.mExtents, 12);
 
 	resizeVertices(src.mNumVertices);
@@ -5179,6 +5256,7 @@ LLVolumeFace& LLVolumeFace::operator=(const LLVolumeFace& src)
 		LLVector4a::memcpyNonAliased16((F32*) mNormals, (F32*) src.mNormals, vert_size);
 		LLVector4a::memcpyNonAliased16((F32*) mTexCoords, (F32*) src.mTexCoords, vert_size);
 
+
 		if (src.mBinormals)
 		{
 			allocateBinormals(src.mNumVertices);
@@ -5217,17 +5295,37 @@ LLVolumeFace& LLVolumeFace::operator=(const LLVolumeFace& src)
 
 LLVolumeFace::~LLVolumeFace()
 {
+	_mm_free(mExtents);
+	mExtents = NULL;
+
+	freeData();
+}
+
+void LLVolumeFace::freeData()
+{
 	_mm_free(mPositions);
+	mPositions = NULL;
 	_mm_free(mNormals);
+	mNormals = NULL;
 	_mm_free(mTexCoords);
+	mTexCoords = NULL;
 	_mm_free(mIndices);
+	mIndices = NULL;
 	_mm_free(mBinormals);
+	mBinormals = NULL;
 	_mm_free(mWeights);
-	_mm_free(mExtents);
+	mWeights = NULL;
+
+	delete mOctree;
+	mOctree = NULL;
 }
 
 BOOL LLVolumeFace::create(LLVolume* volume, BOOL partial_build)
 {
+	//tree for this face is no longer valid
+	delete mOctree;
+	mOctree = NULL;
+
 	if (mTypeMask & CAP_MASK)
 	{
 		return createCap(volume, partial_build);
@@ -5250,6 +5348,18 @@ void LLVolumeFace::getVertexData(U16 index, LLVolumeFace::VertexData& cv)
 	cv.mTexCoord = mTexCoords[index];
 }
 
+bool LLVolumeFace::VertexMapData::operator==(const LLVolumeFace::VertexData& rhs) const
+{
+	return getPosition().equal3(rhs.getPosition()) &&
+		mTexCoord == rhs.mTexCoord &&
+		getNormal().equal3(rhs.getNormal());
+}
+
+bool LLVolumeFace::VertexMapData::ComparePosition::operator()(const LLVector4a& a, const LLVector4a& b) const
+{
+	return a.less3(b);			
+}
+
 void LLVolumeFace::optimize(F32 angle_cutoff)
 {
 	LLVolumeFace new_face;
@@ -5305,6 +5415,60 @@ void LLVolumeFace::optimize(F32 angle_cutoff)
 	swapData(new_face);
 }
 
+
+void LLVolumeFace::createOctree()
+{
+	LLVector4a center;
+	LLVector4a size;
+	center.splat(0.f);
+	size.splat(1.f);
+
+	mOctree = new LLOctreeRoot<LLVolumeTriangle>(center, size, NULL);
+	new LLVolumeOctreeListener(mOctree);
+
+	for (U32 i = 0; i < mNumIndices; i+= 3)
+	{
+		LLPointer<LLVolumeTriangle> tri = new LLVolumeTriangle();
+				
+		const LLVector4a& v0 = mPositions[mIndices[i]];
+		const LLVector4a& v1 = mPositions[mIndices[i+1]];
+		const LLVector4a& v2 = mPositions[mIndices[i+2]];
+
+		tri->mV[0] = &v0;
+		tri->mV[1] = &v1;
+		tri->mV[2] = &v2;
+
+		tri->mIndex[0] = mIndices[i];
+		tri->mIndex[1] = mIndices[i+1];
+		tri->mIndex[2] = mIndices[i+2];
+
+		LLVector4a min = v0;
+		min.setMin(v1);
+		min.setMin(v2);
+
+		LLVector4a max = v0;
+		max.setMax(v1);
+		max.setMax(v2);
+
+		LLVector4a center;
+		center.setAdd(min, max);
+		center.mul(0.5f);
+
+		*tri->mPositionGroup = center;
+
+		LLVector4a size;
+		size.setSub(max,min);
+		
+		tri->mRadius = size.length3() * 0.5f;
+		
+		mOctree->insert(tri);
+	}
+
+	LLVolumeOctreeRebound rebound(this);
+	rebound.traverse(mOctree);
+}
+
+
 void LLVolumeFace::swapData(LLVolumeFace& rhs)
 {
 	llswap(rhs.mPositions, mPositions);
diff --git a/indra/llmath/llvolume.h b/indra/llmath/llvolume.h
index 7c63266aab..c49d1c650d 100644
--- a/indra/llmath/llvolume.h
+++ b/indra/llmath/llvolume.h
@@ -40,8 +40,13 @@ class LLPathParams;
 class LLVolumeParams;
 class LLProfile;
 class LLPath;
+
+template <class T> class LLOctreeNode;
+
+class LLVector4a;
 class LLVolumeFace;
 class LLVolume;
+class LLVolumeTriangle;
 
 #include "lldarray.h"
 #include "lluuid.h"
@@ -49,15 +54,14 @@ class LLVolume;
 //#include "vmath.h"
 #include "v2math.h"
 #include "v3math.h"
+#include "v3dmath.h"
 #include "v4math.h"
-#include "llvector4a.h"
 #include "llquaternion.h"
 #include "llstrider.h"
 #include "v4coloru.h"
 #include "llrefcount.h"
 #include "llfile.h"
 
-
 //============================================================================
 
 const S32 MIN_DETAIL_FACES = 6;
@@ -830,6 +834,9 @@ public:
 	LLVolumeFace& operator=(const LLVolumeFace& rhs);
 
 	~LLVolumeFace();
+private:
+	void freeData();
+public:
 
 	BOOL create(LLVolume* volume, BOOL partial_build = FALSE);
 	void createBinormals();
@@ -855,26 +862,19 @@ public:
 	public:
 		U16 mIndex;
 
-		bool operator==(const LLVolumeFace::VertexData& rhs) const
-		{
-			return getPosition().equal3(rhs.getPosition()) &&
-				mTexCoord == rhs.mTexCoord &&
-				getNormal().equal3(rhs.getNormal());
-		}
+		bool operator==(const LLVolumeFace::VertexData& rhs) const;
 
 		struct ComparePosition
 		{
-			bool operator()(const LLVector4a& a, const LLVector4a& b) const
-			{
-				return a.less3(b);			
-			}
+			bool operator()(const LLVector4a& a, const LLVector4a& b) const;
 		};
 
 		typedef std::map<LLVector4a, std::vector<VertexMapData>, VertexMapData::ComparePosition > PointMap;
 	};
 
 	void optimize(F32 angle_cutoff = 2.f);
-	
+	void createOctree();
+
 	enum
 	{
 		SINGLE_MASK =	0x0001,
@@ -919,6 +919,8 @@ public:
 	// mWeights.size() should be empty or match mVertices.size()  
 	LLVector4a* mWeights;
 
+	LLOctreeNode<LLVolumeTriangle>* mOctree;
+
 private:
 	BOOL createUnCutCubeCap(LLVolume* volume, BOOL partial_build = FALSE);
 	BOOL createCap(LLVolume* volume, BOOL partial_build = FALSE);
@@ -1082,12 +1084,15 @@ void calc_binormal_from_triangle(
 
 BOOL LLLineSegmentBoxIntersect(const F32* start, const F32* end, const F32* center, const F32* size);
 BOOL LLLineSegmentBoxIntersect(const LLVector3& start, const LLVector3& end, const LLVector3& center, const LLVector3& size);
+BOOL LLLineSegmentBoxIntersect(const LLVector4a& start, const LLVector4a& end, const LLVector4a& center, const LLVector4a& size);
 
 BOOL LLTriangleRayIntersect(const LLVector3& vert0, const LLVector3& vert1, const LLVector3& vert2, const LLVector3& orig, const LLVector3& dir,
-							F32* intersection_a, F32* intersection_b, F32* intersection_t, BOOL two_sided);
+							F32& intersection_a, F32& intersection_b, F32& intersection_t, BOOL two_sided);
 
 BOOL LLTriangleRayIntersect(const LLVector4a& vert0, const LLVector4a& vert1, const LLVector4a& vert2, const LLVector4a& orig, const LLVector4a& dir,
-							F32* intersection_a, F32* intersection_b, F32* intersection_t, BOOL two_sided);
+							F32& intersection_a, F32& intersection_b, F32& intersection_t);
+BOOL LLTriangleRayIntersectTwoSided(const LLVector4a& vert0, const LLVector4a& vert1, const LLVector4a& vert2, const LLVector4a& orig, const LLVector4a& dir,
+							F32& intersection_a, F32& intersection_b, F32& intersection_t);
 	
 	
 
diff --git a/indra/llrender/llrender.cpp b/indra/llrender/llrender.cpp
index 2472339ec4..3f70ccacb1 100644
--- a/indra/llrender/llrender.cpp
+++ b/indra/llrender/llrender.cpp
@@ -39,6 +39,7 @@
 #include "llimagegl.h"
 #include "llrendertarget.h"
 #include "lltexture.h"
+#include "llvector4a.h"
 
 LLRender gGL;
 
@@ -53,6 +54,7 @@ U32 LLRender::sUICalls = 0;
 U32 LLRender::sUIVerts = 0;
 
 static const U32 LL_NUM_TEXTURE_LAYERS = 16; 
+static const U32 LL_MAX_UI_STACK_DEPTH = 32;
 
 static GLenum sGLTextureType[] =
 {
@@ -757,14 +759,27 @@ LLRender::LLRender()
     mCount(0),
     mMode(LLRender::TRIANGLES),
     mCurrTextureUnitIndex(0),
-    mMaxAnisotropy(0.f) 
+    mMaxAnisotropy(0.f),
+	mUIStackDepth(0)
 {
 	mBuffer = new LLVertexBuffer(immediate_mask, 0);
 	mBuffer->allocateBuffer(4096, 0, TRUE);
-	mBuffer->getVertexStrider(mVerticesp);
-	mBuffer->getTexCoord0Strider(mTexcoordsp);
-	mBuffer->getColorStrider(mColorsp);
-	
+
+	LLStrider<LLVector3> vert;
+	LLStrider<LLVector2> tc;
+	LLStrider<LLColor4U> color;
+
+	mBuffer->getVertexStrider(vert);
+	mBuffer->getTexCoord0Strider(tc);
+	mBuffer->getColorStrider(color);
+
+	mVerticesp = (LLVector4a*) vert.get();
+	mTexcoordsp = tc.get();
+	mColorsp = color.get();
+
+	mUIOffset = (LLVector4a*) _mm_malloc(LL_MAX_UI_STACK_DEPTH*sizeof(LLVector4a), 16);
+	mUIScale = (LLVector4a*) _mm_malloc(LL_MAX_UI_STACK_DEPTH*sizeof(LLVector4a), 16);
+		
 	mTexUnits.reserve(LL_NUM_TEXTURE_LAYERS);
 	for (U32 i = 0; i < LL_NUM_TEXTURE_LAYERS; i++)
 	{
@@ -800,6 +815,11 @@ void LLRender::shutdown()
 	mTexUnits.clear();
 	delete mDummyTexUnit;
 	mDummyTexUnit = NULL;
+
+	_mm_free(mUIOffset);
+	mUIOffset = NULL;
+	_mm_free(mUIScale);
+	mUIScale = NULL;
 }
 
 void LLRender::refreshState(void)
@@ -848,84 +868,83 @@ void LLRender::popMatrix()
 
 void LLRender::translateUI(F32 x, F32 y, F32 z)
 {
-	if (mUIOffset.empty())
+	if (mUIStackDepth == 0)
 	{
 		llerrs << "Need to push a UI translation frame before offsetting" << llendl;
 	}
 
-	mUIOffset.front().mV[0] += x;
-	mUIOffset.front().mV[1] += y;
-	mUIOffset.front().mV[2] += z;
+	LLVector4a trans(x,y,z);
+	mUIOffset[mUIStackDepth-1].add(trans);
 }
 
 void LLRender::scaleUI(F32 x, F32 y, F32 z)
 {
-	if (mUIScale.empty())
+	if (mUIStackDepth == 0)
 	{
 		llerrs << "Need to push a UI transformation frame before scaling." << llendl;
 	}
 
-	mUIScale.front().scaleVec(LLVector3(x,y,z));
+	LLVector4a scale(x,y,z);
+	mUIScale[mUIStackDepth-1].mul(scale);
 }
 
 void LLRender::pushUIMatrix()
 {
-	if (mUIOffset.empty())
+	if (mUIStackDepth == 0)
 	{
-		mUIOffset.push_front(LLVector3(0,0,0));
+		mUIOffset[0].clear();
+		mUIScale[0].splat(1.f);
 	}
-	else
+	else if (mUIStackDepth < LL_MAX_UI_STACK_DEPTH)
 	{
-		mUIOffset.push_front(mUIOffset.front());
-	}
-	
-	if (mUIScale.empty())
-	{
-		mUIScale.push_front(LLVector3(1,1,1));
+		mUIOffset[mUIStackDepth] = mUIOffset[mUIStackDepth-1];
+		mUIScale[mUIStackDepth] = mUIScale[mUIStackDepth-1];
 	}
 	else
 	{
-		mUIScale.push_front(mUIScale.front());
+		llerrs << "Blown UI matrix stack." << llendl;
 	}
+	
+	++mUIStackDepth;
+	
 }
 
 void LLRender::popUIMatrix()
 {
-	if (mUIOffset.empty())
+	if (mUIStackDepth == 0)
 	{
 		llerrs << "UI offset stack blown." << llendl;
 	}
-	mUIOffset.pop_front();
-	mUIScale.pop_front();
+	--mUIStackDepth;
 }
 
 LLVector3 LLRender::getUITranslation()
 {
-	if (mUIOffset.empty())
+	if (mUIStackDepth == 0)
 	{
 		llerrs << "UI offset stack empty." << llendl;
 	}
-	return mUIOffset.front();
+	return LLVector3(mUIOffset[mUIStackDepth-1].getF32());
 }
 
 LLVector3 LLRender::getUIScale()
 {
-	if (mUIScale.empty())
+	if (mUIStackDepth == 0)
 	{
 		llerrs << "UI scale stack empty." << llendl;
 	}
-	return mUIScale.front();
+	return LLVector3(mUIScale[mUIStackDepth-1].getF32());
 }
 
 
 void LLRender::loadUIIdentity()
 {
-	if (mUIOffset.empty())
+	if (mUIStackDepth == 0)
 	{
 		llerrs << "Need to push UI translation frame before clearing offset." << llendl;
 	}
-	mUIOffset.front().setVec(0,0,0);
-	mUIScale.front().setVec(1,1,1);
+	mUIOffset[mUIStackDepth-1].clear();
+	mUIScale[mUIStackDepth-1].splat(1.f);
 }
 
 void LLRender::setColorMask(bool writeColor, bool writeAlpha)
@@ -1154,7 +1173,7 @@ void LLRender::flush()
 		}
 #endif
 				
-		if (!mUIOffset.empty())
+		if (mUIStackDepth > 0)
 		{
 			sUICalls++;
 			sUIVerts += mCount;
@@ -1206,24 +1225,22 @@ void LLRender::vertex3f(const GLfloat& x, const GLfloat& y, const GLfloat& z)
 		return;
 	}
 
-	if (mUIOffset.empty())
+	LLVector3& v = reinterpret_cast<LLVector3&>(mVerticesp[mCount]);
+	v.set(x,y,z);
+	if (mUIStackDepth != 0)
 	{
-		mVerticesp[mCount] = LLVector3(x,y,z);
-	}
-	else
-	{
-		LLVector3 vert = (LLVector3(x,y,z)+mUIOffset.front()).scaledVec(mUIScale.front());
-		mVerticesp[mCount] = vert;
+		v += reinterpret_cast<LLVector3&>(mUIOffset[mUIStackDepth-1]);
+		v.scaleVec(reinterpret_cast<LLVector3&>(mUIScale[mUIStackDepth-1]));
 	}
 
 	mCount++;
 	if (mCount < 4096)
 	{
-		mVerticesp[mCount] = mVerticesp[mCount-1];
 		mColorsp[mCount] = mColorsp[mCount-1];
 		mTexcoordsp[mCount] = mTexcoordsp[mCount-1];
 	}
 }
+
 void LLRender::vertex2i(const GLint& x, const GLint& y)
 {
 	vertex3f((GLfloat) x, (GLfloat) y, 0);	
diff --git a/indra/llrender/llrender.h b/indra/llrender/llrender.h
index 3cda4b5770..2bacf16dc6 100644
--- a/indra/llrender/llrender.h
+++ b/indra/llrender/llrender.h
@@ -52,6 +52,7 @@ class LLCubeMap;
 class LLImageGL;
 class LLRenderTarget;
 class LLTexture ;
+class LLVector4a;
 
 class LLTexUnit
 {
@@ -360,9 +361,10 @@ private:
 	F32				mCurrAlphaFuncVal;
 
 	LLPointer<LLVertexBuffer>	mBuffer;
-	LLStrider<LLVector3>		mVerticesp;
-	LLStrider<LLVector2>		mTexcoordsp;
-	LLStrider<LLColor4U>		mColorsp;
+	LLVector4a*					mVerticesp;
+	LLVector2*					mTexcoordsp;
+	LLColor4U*					mColorsp;
+
 	std::vector<LLTexUnit*>		mTexUnits;
 	LLTexUnit*			mDummyTexUnit;
 
@@ -372,9 +374,10 @@ private:
 	eBlendFactor mCurrBlendAlphaDFactor;
 	F32				mMaxAnisotropy;
 
-	std::list<LLVector3> mUIOffset;
-	std::list<LLVector3> mUIScale;
+	LLVector4a* mUIOffset;
+	LLVector4a* mUIScale;
 
+	U32 mUIStackDepth;
 };
 
 extern F64 gGLModelView[16];
diff --git a/indra/llrender/llvertexbuffer.h b/indra/llrender/llvertexbuffer.h
index 47146a5ec4..715309b64a 100644
--- a/indra/llrender/llvertexbuffer.h
+++ b/indra/llrender/llvertexbuffer.h
@@ -79,6 +79,17 @@ protected:
 class LLVertexBuffer : public LLRefCount
 {
 public:
+	LLVertexBuffer(const LLVertexBuffer& rhs)
+	{
+		*this = rhs;
+	}
+
+	const LLVertexBuffer& operator=(const LLVertexBuffer& rhs)
+	{
+		llerrs << "Illegal operation!" << llendl;
+		return *this;
+	}
+
 	static LLVBOPool sStreamVBOPool;
 	static LLVBOPool sDynamicVBOPool;
 	static LLVBOPool sStreamIBOPool;
diff --git a/indra/llui/lllineeditor.cpp b/indra/llui/lllineeditor.cpp
index 45f9de8e8d..c0cc294c02 100644
--- a/indra/llui/lllineeditor.cpp
+++ b/indra/llui/lllineeditor.cpp
@@ -377,7 +377,14 @@ void LLLineEditor::setText(const LLStringExplicit &new_text)
 	setCursor(llmin((S32)mText.length(), getCursor()));
 
 	// Set current history line to end of history.
-	mCurrentHistoryLine = mLineHistory.end() - 1;
+	if (mLineHistory.empty())
+	{
+		mCurrentHistoryLine = mLineHistory.end();
+	}
+	else
+	{
+		mCurrentHistoryLine = mLineHistory.end() - 1;
+	}
 
 	mPrevText = mText;
 }
diff --git a/indra/llui/llview.cpp b/indra/llui/llview.cpp
index e67f0ec3fc..459041f67d 100644
--- a/indra/llui/llview.cpp
+++ b/indra/llui/llview.cpp
@@ -123,7 +123,8 @@ LLView::Params::Params()
 }
 
 LLView::LLView(const LLView::Params& p)
-:	mName(p.name),
+:	mVisible(p.visible),
+	mName(p.name),
 	mParentView(NULL),
 	mReshapeFlags(FOLLOWS_NONE),
 	mFromXUI(p.from_xui),
@@ -132,7 +133,6 @@ LLView::LLView(const LLView::Params& p)
 	mNextInsertionOrdinal(0),
 	mHoverCursor(getCursorFromString(p.hover_cursor)),
 	mEnabled(p.enabled),
-	mVisible(p.visible),
 	mMouseOpaque(p.mouse_opaque),
 	mSoundFlags(p.sound_flags),
 	mUseBoundingRect(p.use_bounding_rect),
@@ -1309,7 +1309,13 @@ void LLView::drawChildren()
 {
 	if (!mChildList.empty())
 	{
-		LLRect rootRect = getRootView()->getRect();
+		static const LLRect* rootRect = NULL;
+		
+		if (!mParentView)
+		{
+			rootRect = &mRect;
+		}
+
 		LLRect screenRect;
 
 		++sDepth;
@@ -1319,13 +1325,15 @@ void LLView::drawChildren()
 			child_list_reverse_iter_t child = child_iter++;
 			LLView *viewp = *child;
 
-			if (viewp->getVisible() && viewp->getRect().isValid())
+			if (!viewp->getVisible() || !viewp->getRect().isValid())
+			{
+				continue;
+			}
+
 			{
-				// check for bad data
-				llassert_always(viewp->getVisible() == TRUE);
 				// Only draw views that are within the root view
 				localRectToScreen(viewp->getRect(),&screenRect);
-				if ( rootRect.overlaps(screenRect)  && LLUI::sDirtyRect.overlaps(screenRect))
+				if ( rootRect->overlaps(screenRect)  && LLUI::sDirtyRect.overlaps(screenRect))
 				{
 					LLUI::pushMatrix();
 					{
diff --git a/indra/newview/lldrawable.cpp b/indra/newview/lldrawable.cpp
index 03eee12707..04e433dcfd 100644
--- a/indra/newview/lldrawable.cpp
+++ b/indra/newview/lldrawable.cpp
@@ -41,6 +41,7 @@
 #include "llcriticaldamp.h"
 #include "llface.h"
 #include "lllightconstants.h"
+#include "llmatrix4a.h"
 #include "llsky.h"
 #include "llsurfacepatch.h"
 #include "llviewercamera.h"
@@ -91,8 +92,12 @@ void LLDrawable::incrementVisible()
 	sCurVisible++;
 	sCurPixelAngle = (F32) gViewerWindow->getWindowHeightRaw()/LLViewerCamera::getInstance()->getView();
 }
+
 void LLDrawable::init()
 {
+	mExtents = (LLVector4a*) _mm_malloc(sizeof(LLVector4a)*3, 32);
+	mPositionGroup = mExtents + 2;
+
 	// mXform
 	mParent = NULL;
 	mRenderType = 0;
@@ -121,6 +126,11 @@ void LLDrawable::initClass()
 
 void LLDrawable::destroy()
 {
+	if (gDebugGL)
+	{
+		gPipeline.checkReferences(this);
+	}
+
 	if (isDead())
 	{
 		sNumZombieDrawables--;
@@ -139,6 +149,9 @@ void LLDrawable::destroy()
 	{
 		llinfos << "- Zombie drawables: " << sNumZombieDrawables << llendl;
 	}*/	
+
+	_mm_free(mExtents);
+	mExtents = mPositionGroup = NULL;
 }
 
 void LLDrawable::markDead()
@@ -714,12 +727,14 @@ void LLDrawable::updateDistance(LLCamera& camera, bool force_update)
 					LLFace* facep = getFace(i);
 					if (force_update || facep->getPoolType() == LLDrawPool::POOL_ALPHA)
 					{
-						LLVector3 box = (facep->mExtents[1] - facep->mExtents[0]) * 0.25f;
+						LLVector4a box;
+						box.setSub(facep->mExtents[1], facep->mExtents[0]);
+						box.mul(0.25f);
 						LLVector3 v = (facep->mCenterLocal-camera.getOrigin());
 						const LLVector3& at = camera.getAtAxis();
 						for (U32 j = 0; j < 3; j++)
 						{
-							v.mV[j] -= box.mV[j] * at.mV[j];
+							v.mV[j] -= box[j] * at.mV[j];
 						}
 						facep->mDistance = v * camera.getAtAxis();
 					}
@@ -728,7 +743,7 @@ void LLDrawable::updateDistance(LLCamera& camera, bool force_update)
 		}
 		else
 		{
-			pos = LLVector3(getPositionGroup());
+			pos = LLVector3(getPositionGroup().getF32());
 		}
 
 		pos -= camera.getOrigin();	
@@ -777,7 +792,7 @@ BOOL LLDrawable::updateGeometry(BOOL priority)
 	return res;
 }
 
-void LLDrawable::shiftPos(const LLVector3 &shift_vector)
+void LLDrawable::shiftPos(const LLVector4a &shift_vector)
 {
 	if (isDead())
 	{
@@ -809,9 +824,9 @@ void LLDrawable::shiftPos(const LLVector3 &shift_vector)
 		for (S32 i = 0; i < getNumFaces(); i++)
 		{
 			LLFace *facep = getFace(i);
-			facep->mCenterAgent += shift_vector;
-			facep->mExtents[0] += shift_vector;
-			facep->mExtents[1] += shift_vector;
+			facep->mCenterAgent += LLVector3(shift_vector.getF32());
+			facep->mExtents[0].add(shift_vector);
+			facep->mExtents[1].add(shift_vector);
 			
 			if (!volume && facep->hasGeometry())
 			{
@@ -820,9 +835,9 @@ void LLDrawable::shiftPos(const LLVector3 &shift_vector)
 			}
 		}
 		
-		mExtents[0] += shift_vector;
-		mExtents[1] += shift_vector;
-		mPositionGroup += LLVector3d(shift_vector);
+		mExtents[0].add(shift_vector);
+		mExtents[1].add(shift_vector);
+		mPositionGroup->add(shift_vector);
 	}
 	else if (mSpatialBridge)
 	{
@@ -830,9 +845,9 @@ void LLDrawable::shiftPos(const LLVector3 &shift_vector)
 	}
 	else if (isAvatar())
 	{
-		mExtents[0] += shift_vector;
-		mExtents[1] += shift_vector;
-		mPositionGroup += LLVector3d(shift_vector);
+		mExtents[0].add(shift_vector);
+		mExtents[1].add(shift_vector);
+		mPositionGroup->add(shift_vector);
 	}
 	
 	mVObjp->onShift(shift_vector);
@@ -844,21 +859,26 @@ const LLVector3& LLDrawable::getBounds(LLVector3& min, LLVector3& max) const
 	return mXform.getPositionW();
 }
 
-const LLVector3* LLDrawable::getSpatialExtents() const
+const LLVector4a* LLDrawable::getSpatialExtents() const
 {
 	return mExtents;
 }
 
-void LLDrawable::setSpatialExtents(LLVector3 min, LLVector3 max)
+void LLDrawable::setSpatialExtents(const LLVector3& min, const LLVector3& max)
+{ 
+	mExtents[0].load3(min.mV); 
+	mExtents[1].load3(max.mV);
+}
+
+void LLDrawable::setSpatialExtents(const LLVector4a& min, const LLVector4a& max)
 { 
-	LLVector3 size = max - min;
 	mExtents[0] = min; 
-	mExtents[1] = max; 
+	mExtents[1] = max;
 }
 
-void LLDrawable::setPositionGroup(const LLVector3d& pos)
+void LLDrawable::setPositionGroup(const LLVector4a& pos)
 {
-	mPositionGroup.setVec(pos);
+	*mPositionGroup = pos;
 }
 
 void LLDrawable::updateSpatialExtents()
@@ -872,7 +892,7 @@ void LLDrawable::updateSpatialExtents()
 	
 	if (mSpatialBridge.notNull())
 	{
-		mPositionGroup.setVec(0,0,0);
+		mPositionGroup->splat(0.f);
 	}
 }
 
@@ -1083,59 +1103,72 @@ void LLSpatialBridge::updateSpatialExtents()
 		root->rebound();
 	}
 	
-	LLXformMatrix* mat = mDrawable->getXform();
-	
-	LLVector3 offset = root->mBounds[0];
-	LLVector3 size = root->mBounds[1];
+	LLVector4a offset;
+	LLVector4a size = root->mBounds[1];
 		
-	LLVector3 center = LLVector3(0,0,0) * mat->getWorldMatrix();
-	LLQuaternion rotation = LLQuaternion(mat->getWorldMatrix());
+	//VECTORIZE THIS
+	LLMatrix4a mat;
+	mat.loadu(mDrawable->getXform()->getWorldMatrix());
+
+	LLVector4a t;
+	t.splat(0.f);
+
+	LLVector4a center;
+	mat.affineTransform(t, center);
 	
-	offset *= rotation;
-	center += offset;
+	mat.rotate(root->mBounds[0], offset);
+	center.add(offset);
 	
-	LLVector3 v[4];
+	LLVector4a v[4];
+
 	//get 4 corners of bounding box
-	v[0] = (size * rotation);
-	v[1] = (LLVector3(-size.mV[0], -size.mV[1], size.mV[2]) * rotation);
-	v[2] = (LLVector3(size.mV[0], -size.mV[1], -size.mV[2]) * rotation);
-	v[3] = (LLVector3(-size.mV[0], size.mV[1], -size.mV[2]) * rotation);
+	mat.rotate(size,v[0]);
 
-	LLVector3& newMin = mExtents[0];
-	LLVector3& newMax = mExtents[1];
+	LLVector4a scale;
+	
+	scale.set(-1.f, -1.f, 1.f);
+	scale.mul(size);
+	mat.rotate(scale, v[1]);
+	
+	scale.set(1.f, -1.f, -1.f);
+	scale.mul(size);
+	mat.rotate(scale, v[2]);
+	
+	scale.set(-1.f, 1.f, -1.f);
+	scale.mul(size);
+	mat.rotate(scale, v[3]);
+
+	
+	LLVector4a& newMin = mExtents[0];
+	LLVector4a& newMax = mExtents[1];
 	
 	newMin = newMax = center;
 	
 	for (U32 i = 0; i < 4; i++)
 	{
-		for (U32 j = 0; j < 3; j++)
-		{
-			F32 delta = fabsf(v[i].mV[j]);
-			F32 min = center.mV[j] - delta;
-			F32 max = center.mV[j] + delta;
-			
-			if (min < newMin.mV[j])
-			{
-				newMin.mV[j] = min;
-			}
-			
-			if (max > newMax.mV[j])
-			{
-				newMax.mV[j] = max;
-			}
-		}
-	}
+		LLVector4a delta;
+		delta.setAbs(v[i]);
+		LLVector4a min;
+		min.setSub(center, delta);
+		LLVector4a max;
+		max.setAdd(center, delta);
 
-	LLVector3 diagonal = newMax - newMin;
-	mRadius = diagonal.magVec() * 0.5f;
+		newMin.setMin(min);
+		newMax.setMax(max);
+	}
+	
+	LLVector4a diagonal;
+	diagonal.setSub(newMax, newMin);
+	mRadius = diagonal.length3() * 0.5f;
 	
-	mPositionGroup.setVec((newMin + newMax) * 0.5f);
+	mPositionGroup->setAdd(newMin,newMax);
+	mPositionGroup->mul(0.5f);
 	updateBinRadius();
 }
 
 void LLSpatialBridge::updateBinRadius()
 {
-	mBinRadius = llmin((F32) mOctree->getSize().mdV[0]*0.5f, 256.f);
+	mBinRadius = llmin( mOctree->getSize()[0]*0.5f, 256.f);
 }
 
 LLCamera LLSpatialBridge::transformCamera(LLCamera& camera)
@@ -1276,8 +1309,12 @@ void LLSpatialBridge::setVisible(LLCamera& camera_in, std::vector<LLDrawable*>*
 	LLSpatialGroup* group = (LLSpatialGroup*) mOctree->getListener(0);
 	group->rebound();
 	
-	LLVector3 center = (mExtents[0] + mExtents[1]) * 0.5f;
-	LLVector3 size = (mExtents[1]-mExtents[0]) * 0.5f;
+	LLVector4a center;
+	center.setAdd(mExtents[0], mExtents[1]);
+	center.mul(0.5f);
+	LLVector4a size;
+	size.setSub(mExtents[1], mExtents[0]);
+	size.mul(0.5f);
 
 	if ((LLPipeline::sShadowRender && camera_in.AABBInFrustum(center, size)) ||
 		LLPipeline::sImpostorRender ||
@@ -1389,11 +1426,11 @@ BOOL LLSpatialBridge::updateMove()
 	return TRUE;
 }
 
-void LLSpatialBridge::shiftPos(const LLVector3& vec)
+void LLSpatialBridge::shiftPos(const LLVector4a& vec)
 {
-	mExtents[0] += vec;
-	mExtents[1] += vec;
-	mPositionGroup += LLVector3d(vec);
+	mExtents[0].add(vec);
+	mExtents[1].add(vec);
+	mPositionGroup->add(vec);
 }
 
 void LLSpatialBridge::cleanupReferences()
@@ -1511,7 +1548,7 @@ F32 LLHUDBridge::calcPixelArea(LLSpatialGroup* group, LLCamera& camera)
 }
 
 
-void LLHUDBridge::shiftPos(const LLVector3& vec)
+void LLHUDBridge::shiftPos(const LLVector4a& vec)
 {
 	//don't shift hud bridges on region crossing
 }
diff --git a/indra/newview/lldrawable.h b/indra/newview/lldrawable.h
index c3c6cbe12f..811ff1801b 100644
--- a/indra/newview/lldrawable.h
+++ b/indra/newview/lldrawable.h
@@ -41,6 +41,7 @@
 #include "v4math.h"
 #include "m4math.h"
 #include "v4coloru.h"
+#include "llvector4a.h"
 #include "llquaternion.h"
 #include "xform.h"
 #include "llmemtype.h"
@@ -66,6 +67,17 @@ const U32 SILHOUETTE_HIGHLIGHT = 0;
 class LLDrawable : public LLRefCount
 {
 public:
+	LLDrawable(const LLDrawable& rhs)
+	{
+		*this = rhs;
+	}
+
+	const LLDrawable& operator=(const LLDrawable& rhs)
+	{
+		llerrs << "Illegal operation!" << llendl;
+		return *this;
+	}
+
 	static void initClass();
 
 	LLDrawable()				{ init(); }
@@ -94,14 +106,14 @@ public:
 	const LLVector3&	  getPosition() const			{ return mXform.getPosition(); }
 	const LLVector3&      getWorldPosition() const		{ return mXform.getPositionW(); }
 	const LLVector3		  getPositionAgent() const;
-	const LLVector3d&	  getPositionGroup() const		{ return mPositionGroup; }
+	const LLVector4a&	  getPositionGroup() const		{ return *mPositionGroup; }
 	const LLVector3&	  getScale() const				{ return mCurrentScale; }
 	void				  setScale(const LLVector3& scale) { mCurrentScale = scale; }
 	const LLQuaternion&   getWorldRotation() const		{ return mXform.getWorldRotation(); }
 	const LLQuaternion&   getRotation() const			{ return mXform.getRotation(); }
 	F32			          getIntensity() const			{ return llmin(mXform.getScale().mV[0], 4.f); }
 	S32					  getLOD() const				{ return mVObjp ? mVObjp->getLOD() : 1; }
-	F64					  getBinRadius() const			{ return mBinRadius; }
+	F32					  getBinRadius() const			{ return mBinRadius; }
 	void  getMinMax(LLVector3& min,LLVector3& max) const { mXform.getMinMax(min,max); }
 	LLXformMatrix*		getXform() { return &mXform; }
 
@@ -155,7 +167,7 @@ public:
 		
 	void updateSpecialHoverCursor(BOOL enabled);
 
-	virtual void shiftPos(const LLVector3 &shift_vector);
+	virtual void shiftPos(const LLVector4a &shift_vector);
 
 	S32 getGeneration() const					{ return mGeneration; }
 
@@ -173,11 +185,12 @@ public:
 	const LLVector3& getBounds(LLVector3& min, LLVector3& max) const;
 	virtual void updateSpatialExtents();
 	virtual void updateBinRadius();
-	const LLVector3* getSpatialExtents() const;
-	void setSpatialExtents(LLVector3 min, LLVector3 max);
-	void setPositionGroup(const LLVector3d& pos);
-	void setPositionGroup(const LLVector3& pos) { setPositionGroup(LLVector3d(pos)); }
+	const LLVector4a* getSpatialExtents() const;
+	void setSpatialExtents(const LLVector3& min, const LLVector3& max);
+	void setSpatialExtents(const LLVector4a& min, const LLVector4a& max);
 
+	void setPositionGroup(const LLVector4a& pos);
+	
 	void setRenderType(S32 type) 				{ mRenderType = type; }
 	BOOL isRenderType(S32 type) 				{ return mRenderType == type; }
 	S32  getRenderType()						{ return mRenderType; }
@@ -288,6 +301,9 @@ public:
 private:
 	typedef std::vector<LLFace*> face_list_t;
 	
+	LLVector4a*		mExtents;
+	LLVector4a*		mPositionGroup;
+		
 	U32				mState;
 	S32				mRenderType;
 	LLPointer<LLViewerObject> mVObjp;
@@ -297,9 +313,7 @@ private:
 	
 	mutable U32		mVisible;
 	F32				mRadius;
-	LLVector3		mExtents[2];
-	LLVector3d		mPositionGroup;
-	F64				mBinRadius;
+	F32				mBinRadius;
 	S32				mGeneration;
 	
 	LLVector3		mCurrentScale;
diff --git a/indra/newview/llface.cpp b/indra/newview/llface.cpp
index 7a0aa9244d..b8407a6f5b 100644
--- a/indra/newview/llface.cpp
+++ b/indra/newview/llface.cpp
@@ -152,6 +152,8 @@ void cylindricalProjection(LLVector2 &tc, const LLVector4a& normal, const LLVect
 
 void LLFace::init(LLDrawable* drawablep, LLViewerObject* objp)
 {
+	mExtents = (LLVector4a*) _mm_malloc(sizeof(LLVector4a)*2, 16);
+
 	mLastUpdateTime = gFrameTimeSeconds;
 	mLastMoveTime = 0.f;
 	mVSize = 0.f;
@@ -206,6 +208,12 @@ static LLFastTimer::DeclareTimer FTM_FACE_DEREF("Deref");
 void LLFace::destroy()
 {
 	LLFastTimer t(FTM_DESTROY_FACE);
+
+	if (gDebugGL)
+	{
+		gPipeline.checkReferences(this);
+	}
+
 	if(mTexture.notNull())
 	{
 		LLFastTimer t(FTM_DESTROY_TEXTURE);
@@ -260,6 +268,9 @@ void LLFace::destroy()
 		mDrawablep = NULL;
 		mVObjp = NULL;
 	}
+
+	_mm_free(mExtents);
+	mExtents = NULL;
 }
 
 
@@ -725,13 +736,20 @@ static void xform(LLVector2 &tex_coord, F32 cosAng, F32 sinAng, F32 offS, F32 of
 
 
 BOOL LLFace::genVolumeBBoxes(const LLVolume &volume, S32 f,
-								const LLMatrix4& mat_vert, const LLMatrix3& mat_normal, BOOL global_volume)
+								const LLMatrix4& mat_vert_in, const LLMatrix3& mat_normal_in, BOOL global_volume)
 {
 	LLMemType mt1(LLMemType::MTYPE_DRAWABLE);
 
 	//get bounding box
 	if (mDrawablep->isState(LLDrawable::REBUILD_VOLUME | LLDrawable::REBUILD_POSITION))
 	{
+		//VECTORIZE THIS
+		LLMatrix4a mat_vert;
+		mat_vert.loadu(mat_vert_in);
+
+		LLMatrix4a mat_normal;
+		mat_normal.loadu(mat_normal_in);
+
 		//if (mDrawablep->isState(LLDrawable::REBUILD_VOLUME))
 		//{ //vertex buffer no longer valid
 		//	mVertexBuffer = NULL;
@@ -739,82 +757,96 @@ BOOL LLFace::genVolumeBBoxes(const LLVolume &volume, S32 f,
 		//}
 
 		//VECTORIZE THIS
-		LLVector3 min,max;
+		LLVector4a min,max;
 	
 		if (f >= volume.getNumVolumeFaces())
 		{
-			min = LLVector3(-1,-1,-1);
-			max = LLVector3(1,1,1);
-		}
-		else
-		{
-			const LLVolumeFace &face = volume.getVolumeFace(f);
-			min.set(face.mExtents[0].getF32());
-			max.set(face.mExtents[1].getF32());
+			llwarns << "Generating bounding box for invalid face index!" << llendl;
+			f = 0;
 		}
 
+		const LLVolumeFace &face = volume.getVolumeFace(f);
+		min = face.mExtents[0];
+		max = face.mExtents[1];
+		
+
 		//min, max are in volume space, convert to drawable render space
-		LLVector3 center = ((min + max) * 0.5f)*mat_vert;
-		LLVector3 size = ((max-min) * 0.5f);
+		LLVector4a center;
+		LLVector4a t;
+		t.setAdd(min, max);
+		t.mul(0.5f);
+		mat_vert.affineTransform(t, center);
+		LLVector4a size;
+		size.setSub(max, min);
+		size.mul(0.5f);
+
 		if (!global_volume)
 		{
-			size.scaleVec(mDrawablep->getVObj()->getScale());
+			//VECTORIZE THIS
+			LLVector4a scale;
+			scale.load3(mDrawablep->getVObj()->getScale().mV);
+			size.mul(scale);
 		}
 
-		LLMatrix3 mat = mat_normal;
-		LLVector3 x = mat.getFwdRow();
-		LLVector3 y = mat.getLeftRow();
-		LLVector3 z = mat.getUpRow();
-		x.normVec();
-		y.normVec();
-		z.normVec();
+		mat_normal.mMatrix[0].normalize3fast();
+		mat_normal.mMatrix[1].normalize3fast();
+		mat_normal.mMatrix[2].normalize3fast();
+		
+		LLVector4a v[4];
 
-		mat.setRows(x,y,z);
+		//get 4 corners of bounding box
+		mat_normal.rotate(size,v[0]);
 
-		LLQuaternion rotation = LLQuaternion(mat);
+		//VECTORIZE THIS
+		LLVector4a scale;
 		
-		LLVector3 v[4];
-		//get 4 corners of bounding box
-		v[0] = (size * rotation);
-		v[1] = (LLVector3(-size.mV[0], -size.mV[1], size.mV[2]) * rotation);
-		v[2] = (LLVector3(size.mV[0], -size.mV[1], -size.mV[2]) * rotation);
-		v[3] = (LLVector3(-size.mV[0], size.mV[1], -size.mV[2]) * rotation);
+		scale.set(-1.f, -1.f, 1.f);
+		scale.mul(size);
+		mat_normal.rotate(scale, v[1]);
+		
+		scale.set(1.f, -1.f, -1.f);
+		scale.mul(size);
+		mat_normal.rotate(scale, v[2]);
+		
+		scale.set(-1.f, 1.f, -1.f);
+		scale.mul(size);
+		mat_normal.rotate(scale, v[3]);
 
-		LLVector3& newMin = mExtents[0];
-		LLVector3& newMax = mExtents[1];
+		LLVector4a& newMin = mExtents[0];
+		LLVector4a& newMax = mExtents[1];
 		
 		newMin = newMax = center;
 		
 		for (U32 i = 0; i < 4; i++)
 		{
-			for (U32 j = 0; j < 3; j++)
-			{
-				F32 delta = fabsf(v[i].mV[j]);
-				F32 min = center.mV[j] - delta;
-				F32 max = center.mV[j] + delta;
-				
-				if (min < newMin.mV[j])
-				{
-					newMin.mV[j] = min;
-				}
-				
-				if (max > newMax.mV[j])
-				{
-					newMax.mV[j] = max;
-				}
-			}
+			LLVector4a delta;
+			delta.setAbs(v[i]);
+			LLVector4a min;
+			min.setSub(center, delta);
+			LLVector4a max;
+			max.setAdd(center, delta);
+
+			newMin.setMin(min);
+			newMax.setMax(max);
 		}
 
 		if (!mDrawablep->isActive())
 		{
-			LLVector3 offset = mDrawablep->getRegion()->getOriginAgent();
-			newMin += offset;
-			newMax += offset;
+			LLVector4a offset;
+			offset.load3(mDrawablep->getRegion()->getOriginAgent().mV);
+			newMin.add(offset);
+			newMax.add(offset);
 		}
 
-		mCenterLocal = (newMin+newMax)*0.5f;
-		LLVector3 tmp = (newMin - newMax) ;
-		mBoundingSphereRadius = tmp.length() * 0.5f ;
+		t.setAdd(newMin, newMax);
+		t.mul(0.5f);
+
+		//VECTORIZE THIS
+		mCenterLocal.set(t.getF32());
+		
+		t.setSub(newMax,newMin);
+		t.mul(0.5f);
+		mBoundingSphereRadius = t.length3();
 
 		updateCenterAgent();
 	}
@@ -1073,8 +1105,6 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume,
 		mVertexBuffer->getColorStrider(colors, mGeomIndex);
 	}
 
-	F32 r = 0, os = 0, ot = 0, ms = 0, mt = 0, cos_ang = 0, sin_ang = 0;
-	
 	BOOL is_static = mDrawablep->isStatic();
 	BOOL is_global = is_static;
 
@@ -1089,57 +1119,6 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume,
 		clearState(GLOBAL);
 	}
 
-	LLVector2 tmin, tmax;
-	
-	if (rebuild_tcoord)
-	{
-		if (tep)
-		{
-			r  = tep->getRotation();
-			os = tep->mOffsetS;
-			ot = tep->mOffsetT;
-			ms = tep->mScaleS;
-			mt = tep->mScaleT;
-			cos_ang = cos(r);
-			sin_ang = sin(r);
-		}
-		else
-		{
-			cos_ang = 1.0f;
-			sin_ang = 0.0f;
-			os = 0.0f;
-			ot = 0.0f;
-			ms = 1.0f;
-			mt = 1.0f;
-		}
-	}
-
-	U8 tex_mode = 0;
-	
-	if (isState(TEXTURE_ANIM))
-	{
-		LLVOVolume* vobj = (LLVOVolume*) (LLViewerObject*) mVObjp;	
-		tex_mode = vobj->mTexAnimMode;
-
-		if (!tex_mode)
-		{
-			clearState(TEXTURE_ANIM);
-		}
-		else
-		{
-			os = ot = 0.f;
-			r = 0.f;
-			cos_ang = 1.f;
-			sin_ang = 0.f;
-			ms = mt = 1.f;
-		}
-
-		if (getVirtualSize() >= MIN_TEX_ANIM_SIZE)
-		{ //don't override texture transform during tc bake
-			tex_mode = 0;
-		}
-	}
-
 	LLColor4U color = tep->getColor();
 
 	if (rebuild_color)
@@ -1183,242 +1162,382 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume,
 		}
 	}
 	
+	LLMatrix4a mat_normal;
+	mat_normal.loadu(mat_norm_in);
 	
-	//bump setup
-	LLVector4a binormal_dir( -sin_ang, cos_ang, 0.f );
-	LLVector4a bump_s_primary_light_ray(0.f, 0.f, 0.f);
-	LLVector4a bump_t_primary_light_ray(0.f, 0.f, 0.f);
+	//if it's not fullbright and has no normals, bake sunlight based on face normal
+	//bool bake_sunlight = !getTextureEntry()->getFullbright() &&
+	//  !mVertexBuffer->hasDataType(LLVertexBuffer::TYPE_NORMAL);
 
-	LLQuaternion bump_quat;
-	if (mDrawablep->isActive())
-	{
-		bump_quat = LLQuaternion(mDrawablep->getRenderMatrix());
-	}
-	
-	if (bump_code)
+	F32 r = 0, os = 0, ot = 0, ms = 0, mt = 0, cos_ang = 0, sin_ang = 0;
+
+	if (rebuild_tcoord)
 	{
-		mVObjp->getVolume()->genBinormals(f);
-		F32 offset_multiple; 
-		switch( bump_code )
+		bool do_xform;
+			
+		if (tep)
 		{
-			case BE_NO_BUMP:
-			offset_multiple = 0.f;
-			break;
-			case BE_BRIGHTNESS:
-			case BE_DARKNESS:
-			if( mTexture.notNull() && mTexture->hasGLTexture())
+			r  = tep->getRotation();
+			os = tep->mOffsetS;
+			ot = tep->mOffsetT;
+			ms = tep->mScaleS;
+			mt = tep->mScaleT;
+			cos_ang = cos(r);
+			sin_ang = sin(r);
+
+			if (cos_ang != 1.f || 
+				sin_ang != 0.f ||
+				os != 0.f ||
+				ot != 0.f ||
+				ms != 1.f ||
+				mt != 1.f)
 			{
-				// Offset by approximately one texel
-				S32 cur_discard = mTexture->getDiscardLevel();
-				S32 max_size = llmax( mTexture->getWidth(), mTexture->getHeight() );
-				max_size <<= cur_discard;
-				const F32 ARTIFICIAL_OFFSET = 2.f;
-				offset_multiple = ARTIFICIAL_OFFSET / (F32)max_size;
+				do_xform = true;
 			}
 			else
 			{
-				offset_multiple = 1.f/256;
-			}
-			break;
-
-			default:  // Standard bumpmap textures.  Assumed to be 256x256
-			offset_multiple = 1.f / 256;
-			break;
+				do_xform = false;
+			}	
 		}
-
-		F32 s_scale = 1.f;
-		F32 t_scale = 1.f;
-		if( tep )
+		else
 		{
-			tep->getScale( &s_scale, &t_scale );
+			do_xform = false;
 		}
-		// Use the nudged south when coming from above sun angle, such
-		// that emboss mapping always shows up on the upward faces of cubes when 
-		// it's noon (since a lot of builders build with the sun forced to noon).
-		LLVector3   sun_ray  = gSky.mVOSkyp->mBumpSunDir;
-		LLVector3   moon_ray = gSky.getMoonDirection();
-		LLVector3& primary_light_ray = (sun_ray.mV[VZ] > 0) ? sun_ray : moon_ray;
+						
+		//bump setup
+		LLVector4a binormal_dir( -sin_ang, cos_ang, 0.f );
+		LLVector4a bump_s_primary_light_ray(0.f, 0.f, 0.f);
+		LLVector4a bump_t_primary_light_ray(0.f, 0.f, 0.f);
 
-		bump_s_primary_light_ray.load3((offset_multiple * s_scale * primary_light_ray).mV);
-		bump_t_primary_light_ray.load3((offset_multiple * t_scale * primary_light_ray).mV);
-	}
+		LLQuaternion bump_quat;
+		if (mDrawablep->isActive())
+		{
+			bump_quat = LLQuaternion(mDrawablep->getRenderMatrix());
+		}
 		
-	U8 texgen = getTextureEntry()->getTexGen();
-	if (rebuild_tcoord && texgen != LLTextureEntry::TEX_GEN_DEFAULT)
-	{ //planar texgen needs binormals
-		mVObjp->getVolume()->genBinormals(f);
-	}
-
-	LLMatrix4a mat_normal;
-	mat_normal.loadu(mat_norm_in);
-	
-	//if it's not fullbright and has no normals, bake sunlight based on face normal
-	//bool bake_sunlight = !getTextureEntry()->getFullbright() &&
-	//  !mVertexBuffer->hasDataType(LLVertexBuffer::TYPE_NORMAL);
-
-	if (rebuild_tcoord)
-	{
-		LLVector4a scalea;
-		scalea.load3(scale.mV);
-
+		if (bump_code)
+		{
+			mVObjp->getVolume()->genBinormals(f);
+			F32 offset_multiple; 
+			switch( bump_code )
+			{
+				case BE_NO_BUMP:
+				offset_multiple = 0.f;
+				break;
+				case BE_BRIGHTNESS:
+				case BE_DARKNESS:
+				if( mTexture.notNull() && mTexture->hasGLTexture())
+				{
+					// Offset by approximately one texel
+					S32 cur_discard = mTexture->getDiscardLevel();
+					S32 max_size = llmax( mTexture->getWidth(), mTexture->getHeight() );
+					max_size <<= cur_discard;
+					const F32 ARTIFICIAL_OFFSET = 2.f;
+					offset_multiple = ARTIFICIAL_OFFSET / (F32)max_size;
+				}
+				else
+				{
+					offset_multiple = 1.f/256;
+				}
+				break;
 
-		for (S32 i = 0; i < num_vertices; i++)
-		{	
-			LLVector2 tc(vf.mTexCoords[i]);
-		
-			LLVector4a& norm = vf.mNormals[i];
-			
-			LLVector4a& center = *(vf.mCenter);
+				default:  // Standard bumpmap textures.  Assumed to be 256x256
+				offset_multiple = 1.f / 256;
+				break;
+			}
 
-			if (texgen != LLTextureEntry::TEX_GEN_DEFAULT)
+			F32 s_scale = 1.f;
+			F32 t_scale = 1.f;
+			if( tep )
 			{
-				LLVector4a vec = vf.mPositions[i];
-			
-				vec.mul(scalea);
-
-				switch (texgen)
-				{
-					case LLTextureEntry::TEX_GEN_PLANAR:
-						planarProjection(tc, norm, center, vec);
-						break;
-					case LLTextureEntry::TEX_GEN_SPHERICAL:
-						sphericalProjection(tc, norm, center, vec);
-						break;
-					case LLTextureEntry::TEX_GEN_CYLINDRICAL:
-						cylindricalProjection(tc, norm, center, vec);
-						break;
-					default:
-						break;
-				}		
+				tep->getScale( &s_scale, &t_scale );
 			}
+			// Use the nudged south when coming from above sun angle, such
+			// that emboss mapping always shows up on the upward faces of cubes when 
+			// it's noon (since a lot of builders build with the sun forced to noon).
+			LLVector3   sun_ray  = gSky.mVOSkyp->mBumpSunDir;
+			LLVector3   moon_ray = gSky.getMoonDirection();
+			LLVector3& primary_light_ray = (sun_ray.mV[VZ] > 0) ? sun_ray : moon_ray;
+
+			bump_s_primary_light_ray.load3((offset_multiple * s_scale * primary_light_ray).mV);
+			bump_t_primary_light_ray.load3((offset_multiple * t_scale * primary_light_ray).mV);
+		}
 
-			if (tex_mode && mTextureMatrix)
+		U8 texgen = getTextureEntry()->getTexGen();
+		if (rebuild_tcoord && texgen != LLTextureEntry::TEX_GEN_DEFAULT)
+		{ //planar texgen needs binormals
+			mVObjp->getVolume()->genBinormals(f);
+		}
+
+		U8 tex_mode = 0;
+	
+		if (isState(TEXTURE_ANIM))
+		{
+			LLVOVolume* vobj = (LLVOVolume*) (LLViewerObject*) mVObjp;	
+			tex_mode = vobj->mTexAnimMode;
+
+			if (!tex_mode)
 			{
-				LLVector3 tmp(tc.mV[0], tc.mV[1], 0.f);
-				tmp = tmp * *mTextureMatrix;
-				tc.mV[0] = tmp.mV[0];
-				tc.mV[1] = tmp.mV[1];
+				clearState(TEXTURE_ANIM);
 			}
 			else
 			{
-				xform(tc, cos_ang, sin_ang, os, ot, ms, mt);
+				os = ot = 0.f;
+				r = 0.f;
+				cos_ang = 1.f;
+				sin_ang = 0.f;
+				ms = mt = 1.f;
+
+				do_xform = false;
 			}
 
-			if(in_atlas)
-			{
-				//
-				//manually calculate tex-coord per vertex for varying address modes.
-				//should be removed if shader can handle this.
-				//
+			if (getVirtualSize() >= MIN_TEX_ANIM_SIZE)
+			{ //don't override texture transform during tc bake
+				tex_mode = 0;
+			}
+		}
 
-				S32 int_part = 0 ;
-				switch(mTexture->getAddressMode())
-				{
-				case LLTexUnit::TAM_CLAMP:
-					if(tc.mV[0] < 0.f)
-					{
-						tc.mV[0] = 0.f ;
-					}
-					else if(tc.mV[0] > 1.f)
-					{
-						tc.mV[0] = 1.f;
-					}
+		LLVector4a scalea;
+		scalea.load3(scale.mV);
 
-					if(tc.mV[1] < 0.f)
-					{
-						tc.mV[1] = 0.f ;
-					}
-					else if(tc.mV[1] > 1.f)
-					{
-						tc.mV[1] = 1.f;
-					}
-					break;
-				case LLTexUnit::TAM_MIRROR:
-					if(tc.mV[0] < 0.f)
-					{
-						tc.mV[0] = -tc.mV[0] ;
-					}
-					int_part = (S32)tc.mV[0] ;
-					if(int_part & 1) //odd number
-					{
-						tc.mV[0] = int_part + 1 - tc.mV[0] ;
-					}
-					else //even number
-					{
-						tc.mV[0] -= int_part ;
-					}
+		bool do_bump = bump_code && mVertexBuffer->hasDataType(LLVertexBuffer::TYPE_TEXCOORD1);
+		bool do_tex_mat = tex_mode && mTextureMatrix;
 
-					if(tc.mV[1] < 0.f)
+		if (!in_atlas && !do_bump)
+		{ //not in atlas or not bump mapped, might be able to do a cheap update
+			if (texgen != LLTextureEntry::TEX_GEN_PLANAR)
+			{
+				if (!do_tex_mat)
+				{
+					if (!do_xform)
 					{
-						tc.mV[1] = -tc.mV[1] ;
+						LLVector4a::memcpyNonAliased16((F32*) tex_coords.get(), (F32*) vf.mTexCoords, num_vertices*2);
 					}
-					int_part = (S32)tc.mV[1] ;
-					if(int_part & 1) //odd number
+					else
 					{
-						tc.mV[1] = int_part + 1 - tc.mV[1] ;
+						for (S32 i = 0; i < num_vertices; i++)
+						{	
+							LLVector2 tc(vf.mTexCoords[i]);
+							xform(tc, cos_ang, sin_ang, os, ot, ms, mt);
+							*tex_coords++ = tc;	
+						}
 					}
-					else //even number
-					{
-						tc.mV[1] -= int_part ;
+				}
+				else
+				{ //do tex mat, no texgen, no atlas, no bump
+					for (S32 i = 0; i < num_vertices; i++)
+					{	
+						LLVector2 tc(vf.mTexCoords[i]);
+						LLVector4a& norm = vf.mNormals[i];
+						LLVector4a& center = *(vf.mCenter);
+
+						LLVector3 tmp(tc.mV[0], tc.mV[1], 0.f);
+						tmp = tmp * *mTextureMatrix;
+						tc.mV[0] = tmp.mV[0];
+						tc.mV[1] = tmp.mV[1];
+						*tex_coords++ = tc;	
 					}
-					break;
-				case LLTexUnit::TAM_WRAP:
-					if(tc.mV[0] > 1.f)
-						tc.mV[0] -= (S32)(tc.mV[0] - 0.00001f) ;
-					else if(tc.mV[0] < -1.f)
-						tc.mV[0] -= (S32)(tc.mV[0] + 0.00001f) ;
-
-					if(tc.mV[1] > 1.f)
-						tc.mV[1] -= (S32)(tc.mV[1] - 0.00001f) ;
-					else if(tc.mV[1] < -1.f)
-						tc.mV[1] -= (S32)(tc.mV[1] + 0.00001f) ;
-
-					if(tc.mV[0] < 0.f)
-					{
-						tc.mV[0] = 1.0f + tc.mV[0] ;
+				}
+			}
+			else
+			{ //no bump, no atlas, tex gen planar
+				if (do_tex_mat)
+				{
+					for (S32 i = 0; i < num_vertices; i++)
+					{	
+						LLVector2 tc(vf.mTexCoords[i]);
+						LLVector4a& norm = vf.mNormals[i];
+						LLVector4a& center = *(vf.mCenter);
+						LLVector4a vec = vf.mPositions[i];	
+						vec.mul(scalea);
+						planarProjection(tc, norm, center, vec);
+						
+						LLVector3 tmp(tc.mV[0], tc.mV[1], 0.f);
+						tmp = tmp * *mTextureMatrix;
+						tc.mV[0] = tmp.mV[0];
+						tc.mV[1] = tmp.mV[1];
+				
+						*tex_coords++ = tc;	
 					}
-					if(tc.mV[1] < 0.f)
-					{
-						tc.mV[1] = 1.0f + tc.mV[1] ;
+				}
+				else
+				{
+					for (S32 i = 0; i < num_vertices; i++)
+					{	
+						LLVector2 tc(vf.mTexCoords[i]);
+						LLVector4a& norm = vf.mNormals[i];
+						LLVector4a& center = *(vf.mCenter);
+						LLVector4a vec = vf.mPositions[i];	
+						vec.mul(scalea);
+						planarProjection(tc, norm, center, vec);
+						
+						xform(tc, cos_ang, sin_ang, os, ot, ms, mt);
+
+						*tex_coords++ = tc;	
 					}
-					break;
-				default:
-					break;
 				}
-			
-				tc.mV[0] = tcoord_xoffset + tcoord_xscale * tc.mV[0] ;
-				tc.mV[1] = tcoord_yoffset + tcoord_yscale * tc.mV[1] ;
 			}
+		}
+		else
+		{ //either bump mapped or in atlas, just do the whole expensive loop
+			for (S32 i = 0; i < num_vertices; i++)
+			{	
+				LLVector2 tc(vf.mTexCoords[i]);
 			
+				LLVector4a& norm = vf.mNormals[i];
+				
+				LLVector4a& center = *(vf.mCenter);
 
-			*tex_coords++ = tc;
-			
-			if (bump_code && mVertexBuffer->hasDataType(LLVertexBuffer::TYPE_TEXCOORD1))
-			{
-				LLVector4a tangent;
-				tangent.setCross3(vf.mBinormals[i], vf.mNormals[i]);
-
-				LLMatrix4a tangent_to_object;
-				tangent_to_object.setRows(tangent, vf.mBinormals[i], vf.mNormals[i]);
-				LLVector4a t;
-				tangent_to_object.rotate(binormal_dir, t);
-				LLVector4a binormal;
-				mat_normal.rotate(t, binormal);
-					
-				//VECTORIZE THIS
-				if (mDrawablep->isActive())
+				if (texgen != LLTextureEntry::TEX_GEN_DEFAULT)
 				{
-					LLVector3 t;
-					t.set(binormal.getF32());
-					t *= bump_quat;
-					binormal.load3(t.mV);
+					LLVector4a vec = vf.mPositions[i];
+				
+					vec.mul(scalea);
+
+					switch (texgen)
+					{
+						case LLTextureEntry::TEX_GEN_PLANAR:
+							planarProjection(tc, norm, center, vec);
+							break;
+						case LLTextureEntry::TEX_GEN_SPHERICAL:
+							sphericalProjection(tc, norm, center, vec);
+							break;
+						case LLTextureEntry::TEX_GEN_CYLINDRICAL:
+							cylindricalProjection(tc, norm, center, vec);
+							break;
+						default:
+							break;
+					}		
 				}
 
-				binormal.normalize3fast();
-				tc += LLVector2( bump_s_primary_light_ray.dot3(tangent), bump_t_primary_light_ray.dot3(binormal) );
+				if (tex_mode && mTextureMatrix)
+				{
+					LLVector3 tmp(tc.mV[0], tc.mV[1], 0.f);
+					tmp = tmp * *mTextureMatrix;
+					tc.mV[0] = tmp.mV[0];
+					tc.mV[1] = tmp.mV[1];
+				}
+				else
+				{
+					xform(tc, cos_ang, sin_ang, os, ot, ms, mt);
+				}
+
+				if(in_atlas)
+				{
+					//
+					//manually calculate tex-coord per vertex for varying address modes.
+					//should be removed if shader can handle this.
+					//
+
+					S32 int_part = 0 ;
+					switch(mTexture->getAddressMode())
+					{
+					case LLTexUnit::TAM_CLAMP:
+						if(tc.mV[0] < 0.f)
+						{
+							tc.mV[0] = 0.f ;
+						}
+						else if(tc.mV[0] > 1.f)
+						{
+							tc.mV[0] = 1.f;
+						}
+
+						if(tc.mV[1] < 0.f)
+						{
+							tc.mV[1] = 0.f ;
+						}
+						else if(tc.mV[1] > 1.f)
+						{
+							tc.mV[1] = 1.f;
+						}
+						break;
+					case LLTexUnit::TAM_MIRROR:
+						if(tc.mV[0] < 0.f)
+						{
+							tc.mV[0] = -tc.mV[0] ;
+						}
+						int_part = (S32)tc.mV[0] ;
+						if(int_part & 1) //odd number
+						{
+							tc.mV[0] = int_part + 1 - tc.mV[0] ;
+						}
+						else //even number
+						{
+							tc.mV[0] -= int_part ;
+						}
+
+						if(tc.mV[1] < 0.f)
+						{
+							tc.mV[1] = -tc.mV[1] ;
+						}
+						int_part = (S32)tc.mV[1] ;
+						if(int_part & 1) //odd number
+						{
+							tc.mV[1] = int_part + 1 - tc.mV[1] ;
+						}
+						else //even number
+						{
+							tc.mV[1] -= int_part ;
+						}
+						break;
+					case LLTexUnit::TAM_WRAP:
+						if(tc.mV[0] > 1.f)
+							tc.mV[0] -= (S32)(tc.mV[0] - 0.00001f) ;
+						else if(tc.mV[0] < -1.f)
+							tc.mV[0] -= (S32)(tc.mV[0] + 0.00001f) ;
+
+						if(tc.mV[1] > 1.f)
+							tc.mV[1] -= (S32)(tc.mV[1] - 0.00001f) ;
+						else if(tc.mV[1] < -1.f)
+							tc.mV[1] -= (S32)(tc.mV[1] + 0.00001f) ;
+
+						if(tc.mV[0] < 0.f)
+						{
+							tc.mV[0] = 1.0f + tc.mV[0] ;
+						}
+						if(tc.mV[1] < 0.f)
+						{
+							tc.mV[1] = 1.0f + tc.mV[1] ;
+						}
+						break;
+					default:
+						break;
+					}
 				
-				*tex_coords2++ = tc;
-			}	
+					tc.mV[0] = tcoord_xoffset + tcoord_xscale * tc.mV[0] ;
+					tc.mV[1] = tcoord_yoffset + tcoord_yscale * tc.mV[1] ;
+				}
+				
+
+				*tex_coords++ = tc;
+				
+				if (bump_code && mVertexBuffer->hasDataType(LLVertexBuffer::TYPE_TEXCOORD1))
+				{
+					LLVector4a tangent;
+					tangent.setCross3(vf.mBinormals[i], vf.mNormals[i]);
+
+					LLMatrix4a tangent_to_object;
+					tangent_to_object.setRows(tangent, vf.mBinormals[i], vf.mNormals[i]);
+					LLVector4a t;
+					tangent_to_object.rotate(binormal_dir, t);
+					LLVector4a binormal;
+					mat_normal.rotate(t, binormal);
+						
+					//VECTORIZE THIS
+					if (mDrawablep->isActive())
+					{
+						LLVector3 t;
+						t.set(binormal.getF32());
+						t *= bump_quat;
+						binormal.load3(t.mV);
+					}
+
+					binormal.normalize3fast();
+					tc += LLVector2( bump_s_primary_light_ray.dot3(tangent), bump_t_primary_light_ray.dot3(binormal) );
+					
+					*tex_coords2++ = tc;
+				}	
+			}
 		}
 	}
 
@@ -1560,20 +1679,31 @@ F32 LLFace::getTextureVirtualSize()
 
 BOOL LLFace::calcPixelArea(F32& cos_angle_to_view_dir, F32& radius)
 {
+	//VECTORIZE THIS
 	//get area of circle around face
-	LLVector3 center = getPositionAgent();
-	LLVector3 size = (mExtents[1] - mExtents[0]) * 0.5f;	
+	LLVector4a center;
+	center.load3(getPositionAgent().mV);
+	LLVector4a size;
+	size.setSub(mExtents[1], mExtents[0]);
+	size.mul(0.5f);
+
 	LLViewerCamera* camera = LLViewerCamera::getInstance();
 
-	F32 size_squared = size.lengthSquared() ;
-	LLVector3 lookAt = center - camera->getOrigin();
-	F32 dist = lookAt.normVec() ;	
+	F32 size_squared = size.dot3(size);
+	LLVector4a lookAt;
+	LLVector4a t;
+	t.load3(camera->getOrigin().mV);
+	lookAt.setSub(center, t);
+	F32 dist = lookAt.length3();
+	lookAt.normalize3fast() ;	
 
 	//get area of circle around node
 	F32 app_angle = atanf(fsqrtf(size_squared) / dist);
 	radius = app_angle*LLDrawable::sCurPixelAngle;
 	mPixelArea = radius*radius * 3.14159f;
-	cos_angle_to_view_dir = lookAt * camera->getXAxis() ;
+	LLVector4a x_axis;
+	x_axis.load3(camera->getXAxis().mV);
+	cos_angle_to_view_dir = lookAt.dot3(x_axis);
 
 	//if has media, check if the face is out of the view frustum.	
 	if(hasMedia())
@@ -1589,7 +1719,10 @@ BOOL LLFace::calcPixelArea(F32& cos_angle_to_view_dir, F32& radius)
 		}
 		else
 		{		
-			if(dist * dist * (lookAt - camera->getXAxis()).lengthSquared() < size_squared)
+			LLVector4a d;
+			d.setSub(lookAt, x_axis);
+
+			if(dist * dist * d.dot3(d) < size_squared)
 			{
 				cos_angle_to_view_dir = 1.0f ;
 			}
diff --git a/indra/newview/llface.h b/indra/newview/llface.h
index 48909d7895..0cd472a2fd 100644
--- a/indra/newview/llface.h
+++ b/indra/newview/llface.h
@@ -65,6 +65,17 @@ class LLFace
 {
 public:
 
+	LLFace(const LLFace& rhs)
+	{
+		*this = rhs;
+	}
+
+	const LLFace& operator=(const LLFace& rhs)
+	{
+		llerrs << "Illegal operation!" << llendl;
+		return *this;
+	}
+
 	enum EMasks
 	{
 		LIGHT			= 0x0001,
@@ -221,7 +232,9 @@ public:
 	
 	LLVector3		mCenterLocal;
 	LLVector3		mCenterAgent;
-	LLVector3		mExtents[2];
+	
+	LLVector4a*		mExtents;
+	
 	LLVector2		mTexExtents[2];
 	F32				mDistance;
 	LLPointer<LLVertexBuffer> mVertexBuffer;
diff --git a/indra/newview/llflexibleobject.cpp b/indra/newview/llflexibleobject.cpp
index 561965d021..8be4e34748 100644
--- a/indra/newview/llflexibleobject.cpp
+++ b/indra/newview/llflexibleobject.cpp
@@ -97,11 +97,13 @@ void LLVolumeImplFlexible::onParameterChanged(U16 param_type, LLNetworkData *dat
 	}
 }
 
-void LLVolumeImplFlexible::onShift(const LLVector3 &shift_vector)
+void LLVolumeImplFlexible::onShift(const LLVector4a &shift_vector)
 {	
+	//VECTORIZE THIS
+	LLVector3 shift(shift_vector.getF32());
 	for (int section = 0; section < (1<<FLEXIBLE_OBJECT_MAX_SECTIONS)+1; ++section)
 	{
-		mSection[section].mPosition += shift_vector;	
+		mSection[section].mPosition += shift;	
 	}
 }
 
diff --git a/indra/newview/llflexibleobject.h b/indra/newview/llflexibleobject.h
index 811ae24df2..bdfbded82d 100644
--- a/indra/newview/llflexibleobject.h
+++ b/indra/newview/llflexibleobject.h
@@ -90,7 +90,7 @@ class LLVolumeImplFlexible : public LLVolumeInterface
 		void onSetVolume(const LLVolumeParams &volume_params, const S32 detail);
 		void onSetScale(const LLVector3 &scale, BOOL damped);
 		void onParameterChanged(U16 param_type, LLNetworkData *data, BOOL in_use, bool local_origin);
-		void onShift(const LLVector3 &shift_vector);
+		void onShift(const LLVector4a &shift_vector);
 		bool isVolumeUnique() const { return true; }
 		bool isVolumeGlobal() const { return true; }
 		bool isActive() const { return true; }
diff --git a/indra/newview/llhudicon.cpp b/indra/newview/llhudicon.cpp
index 3c5a4de7f8..63040904df 100644
--- a/indra/newview/llhudicon.cpp
+++ b/indra/newview/llhudicon.cpp
@@ -286,7 +286,6 @@ BOOL LLHUDIcon::lineSegmentIntersect(const LLVector3& start, const LLVector3& en
 	LLVector4a upper_right;
 	upper_right.setAdd(lower_right, y_scalea);
 
-	F32 t = 0.f;
 	LLVector4a enda;
 	enda.load3(end.mV);
 	LLVector4a starta;
@@ -294,8 +293,10 @@ BOOL LLHUDIcon::lineSegmentIntersect(const LLVector3& start, const LLVector3& en
 	LLVector4a dir;
 	dir.setSub(enda, starta);
 
-	if (LLTriangleRayIntersect(upper_right, upper_left, lower_right, starta, dir, NULL, NULL, &t, FALSE) ||
-		LLTriangleRayIntersect(upper_left, lower_left, lower_right, starta, dir, NULL, NULL, &t, FALSE))
+	F32 a,b,t;
+
+	if (LLTriangleRayIntersect(upper_right, upper_left, lower_right, starta, dir, a,b,t) ||
+		LLTriangleRayIntersect(upper_left, lower_left, lower_right, starta, dir, a,b,t))
 	{
 		if (intersection)
 		{
diff --git a/indra/newview/llhudtext.cpp b/indra/newview/llhudtext.cpp
index 9ed5d13831..7f9eddc837 100644
--- a/indra/newview/llhudtext.cpp
+++ b/indra/newview/llhudtext.cpp
@@ -207,10 +207,11 @@ BOOL LLHUDText::lineSegmentIntersect(const LLVector3& start, const LLVector3& en
 		}
 
 		LLVector3 dir = end-start;
-		F32 t = 0.f;
+		F32 a,b,t;
 
-		if (LLTriangleRayIntersect(v[0], v[1], v[2], start, dir, NULL, NULL, &t, FALSE) ||
-			LLTriangleRayIntersect(v[2], v[3], v[0], start, dir, NULL, NULL, &t, FALSE) )
+
+		if (LLTriangleRayIntersect(v[0], v[1], v[2], start, dir, a, b, t, FALSE) ||
+			LLTriangleRayIntersect(v[2], v[3], v[0], start, dir, a, b, t, FALSE) )
 		{
 			if (t <= 1.f)
 			{
diff --git a/indra/newview/llpanelprimmediacontrols.cpp b/indra/newview/llpanelprimmediacontrols.cpp
index a5804aa04e..98fbebbc5d 100644
--- a/indra/newview/llpanelprimmediacontrols.cpp
+++ b/indra/newview/llpanelprimmediacontrols.cpp
@@ -65,6 +65,7 @@
 #include "llweb.h"
 #include "llwindow.h"
 #include "llfloatertools.h"  // to enable hide if build tools are up
+#include "llvector4a.h"
 
 // Functions pulled from pipeline.cpp
 glh::matrix4f glh_get_current_modelview();
diff --git a/indra/newview/llselectmgr.cpp b/indra/newview/llselectmgr.cpp
index e66be1023d..d6cff71991 100644
--- a/indra/newview/llselectmgr.cpp
+++ b/indra/newview/llselectmgr.cpp
@@ -1100,8 +1100,8 @@ void LLSelectMgr::getGrid(LLVector3& origin, LLQuaternion &rotation, LLVector3 &
 		mGridRotation = first_grid_object->getRenderRotation();
 		LLVector3 first_grid_obj_pos = first_grid_object->getRenderPosition();
 
-		LLVector3 min_extents(F32_MAX, F32_MAX, F32_MAX);
-		LLVector3 max_extents(-F32_MAX, -F32_MAX, -F32_MAX);
+		LLVector4a min_extents(F32_MAX);
+		LLVector4a max_extents(-F32_MAX);
 		BOOL grid_changed = FALSE;
 		for (LLObjectSelection::iterator iter = mGridObjects.begin();
 			 iter != mGridObjects.end(); ++iter)
@@ -1110,7 +1110,7 @@ void LLSelectMgr::getGrid(LLVector3& origin, LLQuaternion &rotation, LLVector3 &
 			LLDrawable* drawable = object->mDrawable;
 			if (drawable)
 			{
-				const LLVector3* ext = drawable->getSpatialExtents();
+				const LLVector4a* ext = drawable->getSpatialExtents();
 				update_min_max(min_extents, max_extents, ext[0]);
 				update_min_max(min_extents, max_extents, ext[1]);
 				grid_changed = TRUE;
@@ -1118,13 +1118,19 @@ void LLSelectMgr::getGrid(LLVector3& origin, LLQuaternion &rotation, LLVector3 &
 		}
 		if (grid_changed)
 		{
-			mGridOrigin = lerp(min_extents, max_extents, 0.5f);
+			LLVector4a center, size;
+			center.setAdd(min_extents, max_extents);
+			center.mul(0.5f);
+			size.setSub(max_extents, min_extents);
+			size.mul(0.5f);
+
+			mGridOrigin.set(center.getF32());
 			LLDrawable* drawable = first_grid_object->mDrawable;
 			if (drawable && drawable->isActive())
 			{
 				mGridOrigin = mGridOrigin * first_grid_object->getRenderMatrix();
 			}
-			mGridScale = (max_extents - min_extents) * 0.5f;
+			mGridScale.set(size.getF32());
 		}
 	}
 	else // GRID_MODE_WORLD or just plain default
diff --git a/indra/newview/llspatialpartition.cpp b/indra/newview/llspatialpartition.cpp
index 470c332b42..3cf0138303 100644
--- a/indra/newview/llspatialpartition.cpp
+++ b/indra/newview/llspatialpartition.cpp
@@ -38,6 +38,7 @@
 #include "llviewerobjectlist.h"
 #include "llvovolume.h"
 #include "llvolume.h"
+#include "llvolumeoctree.h"
 #include "llviewercamera.h"
 #include "llface.h"
 #include "llviewercontrol.h"
@@ -101,23 +102,6 @@ void sg_assert(BOOL expr)
 #endif
 }
 
-#if LL_DEBUG
-void validate_drawable(LLDrawable* drawablep)
-{
-	F64 rad = drawablep->getBinRadius();
-	const LLVector3* ext = drawablep->getSpatialExtents();
-
-	if (rad < 0 || rad > 4096 ||
-		(ext[1]-ext[0]).magVec() > 4096)
-	{
-		llwarns << "Invalid drawable found in octree." << llendl;
-	}
-}
-#else
-#define validate_drawable(x)
-#endif
-
-
 S32 AABBSphereIntersect(const LLVector3& min, const LLVector3& max, const LLVector3 &origin, const F32 &rad)
 {
 	return AABBSphereIntersectR2(min, max, origin, rad*rad);
@@ -157,6 +141,55 @@ S32 AABBSphereIntersectR2(const LLVector3& min, const LLVector3& max, const LLVe
 }
 
 
+S32 AABBSphereIntersect(const LLVector4a& min, const LLVector4a& max, const LLVector3 &origin, const F32 &rad)
+{
+	return AABBSphereIntersectR2(min, max, origin, rad*rad);
+}
+
+S32 AABBSphereIntersectR2(const LLVector4a& min, const LLVector4a& max, const LLVector3 &origin, const F32 &r)
+{
+	F32 d = 0.f;
+	F32 t;
+	
+	LLVector4a origina;
+	origina.load3(origin.mV);
+
+	LLVector4a v;
+	v.setSub(min, origina);
+	
+	if (v.dot3(v) < r)
+	{
+		v.setSub(max, origina);
+		if (v.dot3(v) < r)
+		{
+			return 2;
+		}
+	}
+
+
+	for (U32 i = 0; i < 3; i++)
+	{
+		if (origin.mV[i] < min[i])
+		{
+			t = min[i] - origin.mV[i];
+			d += t*t;
+		}
+		else if (origin.mV[i] > max[i])
+		{
+			t = origin.mV[i] - max[i];
+			d += t*t;
+		}
+
+		if (d > r)
+		{
+			return 0;
+		}
+	}
+
+	return 1;
+}
+
+
 typedef enum
 {
 	b000 = 0x00,
@@ -192,24 +225,13 @@ static U8 sOcclusionIndices[] =
 		b000, b110, b100, b101, b001, b011, b010, b110,
 };
 
-U8* get_box_fan_indices(LLCamera* camera, const LLVector3& center)
+U8* get_box_fan_indices(LLCamera* camera, const LLVector4a& center)
 {
-	LLVector3 d = center - camera->getOrigin();
-
-	U8 cypher = 0;
-	if (d.mV[0] > 0)
-	{
-		cypher |= b100;
-	}
-	if (d.mV[1] > 0)
-	{
-		cypher |= b010;
-	}
-	if (d.mV[2] > 0)
-	{
-		cypher |= b001;
-	}
+	LLVector4a origin;
+	origin.load3(camera->getOrigin().mV);
 
+	S32 cypher = center.greaterThan4(origin).getComparisonMask() & 0x7;
+	
 	return sOcclusionIndices+cypher*8;
 }
 						
@@ -217,33 +239,49 @@ void LLSpatialGroup::buildOcclusion()
 {
 	if (!mOcclusionVerts)
 	{
-		mOcclusionVerts = new F32[8*3];
+		mOcclusionVerts = (LLVector4a*) _mm_malloc(sizeof(LLVector4a)*8, 16);
 	}
 
-	LLVector3 r = mBounds[1] + LLVector3(SG_OCCLUSION_FUDGE, SG_OCCLUSION_FUDGE, SG_OCCLUSION_FUDGE);
+	LLVector4a fudge;
+	fudge.splat(SG_OCCLUSION_FUDGE);
 
-	for (U32 k = 0; k < 3; k++)
-	{
-		r.mV[k] = llmin(mBounds[1].mV[k]+0.25f, r.mV[k]);
-	}
+	LLVector4a r;
+	r.setAdd(mBounds[1], fudge);
+
+	LLVector4a r2;
+	r2.splat(0.25f);
+	r2.add(mBounds[1]);
+
+	r.setMin(r2);
 
-	F32* v = mOcclusionVerts;
-	F32* c = mBounds[0].mV;
-	F32* s = r.mV;
+	LLVector4a* v = mOcclusionVerts;
+	const LLVector4a& c = mBounds[0];
+	const LLVector4a& s = r;
 	
+	static const LLVector4a octant[] =
+	{
+		LLVector4a(-1.f, -1.f, -1.f),
+		LLVector4a(-1.f, -1.f, 1.f),
+		LLVector4a(-1.f, 1.f, -1.f),
+		LLVector4a(-1.f, 1.f, 1.f),
+
+		LLVector4a(1.f, -1.f, -1.f),
+		LLVector4a(1.f, -1.f, 1.f),
+		LLVector4a(1.f, 1.f, -1.f),
+		LLVector4a(1.f, 1.f, 1.f),
+	};
+
 	//vertex positions are encoded so the 3 bits of their vertex index 
 	//correspond to their axis facing, with bit position 3,2,1 matching
 	//axis facing x,y,z, bit set meaning positive facing, bit clear 
 	//meaning negative facing
-	v[0] = c[0]-s[0]; v[1]  = c[1]-s[1]; v[2]  = c[2]-s[2];  // 0 - 0000 
-	v[3] = c[0]-s[0]; v[4]  = c[1]-s[1]; v[5]  = c[2]+s[2];  // 1 - 0001
-	v[6] = c[0]-s[0]; v[7]  = c[1]+s[1]; v[8]  = c[2]-s[2];  // 2 - 0010
-	v[9] = c[0]-s[0]; v[10] = c[1]+s[1]; v[11] = c[2]+s[2];  // 3 - 0011
-																					   
-	v[12] = c[0]+s[0]; v[13] = c[1]-s[1]; v[14] = c[2]-s[2]; // 4 - 0100
-	v[15] = c[0]+s[0]; v[16] = c[1]-s[1]; v[17] = c[2]+s[2]; // 5 - 0101
-	v[18] = c[0]+s[0]; v[19] = c[1]+s[1]; v[20] = c[2]-s[2]; // 6 - 0110
-	v[21] = c[0]+s[0]; v[22] = c[1]+s[1]; v[23] = c[2]+s[2]; // 7 - 0111
+	
+	for (S32 i = 0; i < 8; ++i)
+	{
+		v[i] = s;
+		v[i].mul(octant[i]);
+		v[i].add(c);
+	}
 
 	clearState(LLSpatialGroup::OCCLUSION_DIRTY);
 }
@@ -287,6 +325,11 @@ LLSpatialGroup::~LLSpatialGroup()
 		llerrs << "Illegal deletion of LLSpatialGroup!" << llendl;
 	}*/
 
+	if (gDebugGL)
+	{
+		gPipeline.checkReferences(this);
+	}
+
 	if (isState(DEAD))
 	{
 		sZombieGroups--;
@@ -299,11 +342,13 @@ LLSpatialGroup::~LLSpatialGroup()
 		sQueryPool.release(mOcclusionQuery[LLViewerCamera::sCurCameraID]);
 	}
 
-	delete [] mOcclusionVerts;
+	_mm_free(mOcclusionVerts);
 
 	LLMemType mt(LLMemType::MTYPE_SPACE_PARTITION);
 	clearDrawMap();
 	clearAtlasList() ;
+
+	_mm_free(mBounds);
 }
 
 BOOL LLSpatialGroup::hasAtlas(LLTextureAtlas* atlasp)
@@ -455,8 +500,10 @@ void LLSpatialGroup::validate()
 	sg_assert(!isState(DIRTY));
 	sg_assert(!isDead());
 
-	LLVector3 myMin = mBounds[0] - mBounds[1];
-	LLVector3 myMax = mBounds[0] + mBounds[1];
+	LLVector4a myMin;
+	myMin.setSub(mBounds[0], mBounds[1]);
+	LLVector4a myMax;
+	myMax.setAdd(mBounds[0], mBounds[1]);
 
 	validateDrawMap();
 
@@ -488,16 +535,18 @@ void LLSpatialGroup::validate()
 		group->validate();
 		
 		//ensure all children are enclosed in this node
-		LLVector3 center = group->mBounds[0];
-		LLVector3 size = group->mBounds[1];
+		LLVector4a center = group->mBounds[0];
+		LLVector4a size = group->mBounds[1];
 		
-		LLVector3 min = center - size;
-		LLVector3 max = center + size;
+		LLVector4a min;
+		min.setSub(center, size);
+		LLVector4a max;
+		max.setAdd(center, size);
 		
 		for (U32 j = 0; j < 3; j++)
 		{
-			sg_assert(min.mV[j] >= myMin.mV[j]-0.02f);
-			sg_assert(max.mV[j] <= myMax.mV[j]+0.02f);
+			sg_assert(min[j] >= myMin[j]-0.02f);
+			sg_assert(max[j] <= myMax[j]+0.02f);
 		}
 	}
 
@@ -507,8 +556,8 @@ void LLSpatialGroup::validate()
 void LLSpatialGroup::checkStates()
 {
 #if LL_OCTREE_PARANOIA_CHECK
-	LLOctreeStateCheck checker;
-	checker.traverse(mOctreeNode);
+	//LLOctreeStateCheck checker;
+	//checker.traverse(mOctreeNode);
 #endif
 }
 
@@ -533,19 +582,17 @@ BOOL LLSpatialGroup::updateInGroup(LLDrawable *drawablep, BOOL immediate)
 	LLMemType mt(LLMemType::MTYPE_SPACE_PARTITION);
 		
 	drawablep->updateSpatialExtents();
-	validate_drawable(drawablep);
 
 	OctreeNode* parent = mOctreeNode->getOctParent();
 	
 	if (mOctreeNode->isInside(drawablep->getPositionGroup()) && 
 		(mOctreeNode->contains(drawablep) ||
-		 (drawablep->getBinRadius() > mOctreeNode->getSize().mdV[0] &&
+		 (drawablep->getBinRadius() > mOctreeNode->getSize()[0] &&
 				parent && parent->getElementCount() >= LL_OCTREE_MAX_CAPACITY)))
 	{
 		unbound();
 		setState(OBJECT_DIRTY);
 		//setState(GEOM_DIRTY);
-		validate_drawable(drawablep);
 		return TRUE;
 	}
 		
@@ -563,7 +610,6 @@ BOOL LLSpatialGroup::addObject(LLDrawable *drawablep, BOOL add_all, BOOL from_oc
 	else
 	{
 		drawablep->setSpatialGroup(this);
-		validate_drawable(drawablep);
 		setState(OBJECT_DIRTY | GEOM_DIRTY);
 		setOcclusionState(LLSpatialGroup::DISCARD_QUERY, LLSpatialGroup::STATE_MODE_ALL_CAMERAS);
 		gPipeline.markRebuild(this, TRUE);
@@ -664,7 +710,7 @@ void LLSpatialPartition::rebuildMesh(LLSpatialGroup* group)
 
 }
 
-BOOL LLSpatialGroup::boundObjects(BOOL empty, LLVector3& minOut, LLVector3& maxOut)
+BOOL LLSpatialGroup::boundObjects(BOOL empty, LLVector4a& minOut, LLVector4a& maxOut)
 {	
 	const OctreeNode* node = mOctreeNode;
 
@@ -677,8 +723,8 @@ BOOL LLSpatialGroup::boundObjects(BOOL empty, LLVector3& minOut, LLVector3& maxO
 		return FALSE;
 	}
 
-	LLVector3& newMin = mObjectExtents[0];
-	LLVector3& newMax = mObjectExtents[1];
+	LLVector4a& newMin = mObjectExtents[0];
+	LLVector4a& newMax = mObjectExtents[1];
 	
 	if (isState(OBJECT_DIRTY))
 	{ //calculate new bounding box
@@ -687,10 +733,10 @@ BOOL LLSpatialGroup::boundObjects(BOOL empty, LLVector3& minOut, LLVector3& maxO
 		//initialize bounding box to first element
 		OctreeNode::const_element_iter i = node->getData().begin();
 		LLDrawable* drawablep = *i;
-		const LLVector3* minMax = drawablep->getSpatialExtents();
+		const LLVector4a* minMax = drawablep->getSpatialExtents();
 
-		newMin.setVec(minMax[0]);
-		newMax.setVec(minMax[1]);
+		newMin = minMax[0];
+		newMax = minMax[1];
 
 		for (++i; i != node->getData().end(); ++i)
 		{
@@ -714,8 +760,10 @@ BOOL LLSpatialGroup::boundObjects(BOOL empty, LLVector3& minOut, LLVector3& maxO
 			}*/
 		}
 		
-		mObjectBounds[0] = (newMin + newMax) * 0.5f;
-		mObjectBounds[1] = (newMax - newMin) * 0.5f;
+		mObjectBounds[0].setAdd(newMin, newMax);
+		mObjectBounds[0].mul(0.5f);
+		mObjectBounds[1].setSub(newMax, newMin);
+		mObjectBounds[1].mul(0.5f);
 	}
 	
 	if (empty)
@@ -725,17 +773,8 @@ BOOL LLSpatialGroup::boundObjects(BOOL empty, LLVector3& minOut, LLVector3& maxO
 	}
 	else
 	{
-		for (U32 i = 0; i < 3; i++)
-		{
-			if (newMin.mV[i] < minOut.mV[i])
-			{
-				minOut.mV[i] = newMin.mV[i];
-			}
-			if (newMax.mV[i] > maxOut.mV[i])
-			{
-				maxOut.mV[i] = newMax.mV[i];
-			}
-		}
+		minOut.setMin(newMin);
+		maxOut.setMax(newMax);
 	}
 		
 	return TRUE;
@@ -826,18 +865,19 @@ BOOL LLSpatialGroup::removeObject(LLDrawable *drawablep, BOOL from_octree)
 	return TRUE;
 }
 
-void LLSpatialGroup::shift(const LLVector3 &offset)
+void LLSpatialGroup::shift(const LLVector4a &offset)
 {
 	LLMemType mt(LLMemType::MTYPE_SPACE_PARTITION);
-	LLVector3d offsetd(offset);
-	mOctreeNode->setCenter(mOctreeNode->getCenter()+offsetd);
+	LLVector4a t = mOctreeNode->getCenter();
+	t.add(offset);	
+	mOctreeNode->setCenter(t);
 	mOctreeNode->updateMinMax();
-	mBounds[0] += offset;
-	mExtents[0] += offset;
-	mExtents[1] += offset;
-	mObjectBounds[0] += offset;
-	mObjectExtents[0] += offset;
-	mObjectExtents[1] += offset;
+	mBounds[0].add(offset);
+	mExtents[0].add(offset);
+	mExtents[1].add(offset);
+	mObjectBounds[0].add(offset);
+	mObjectExtents[0].add(offset);
+	mObjectExtents[1].add(offset);
 
 	//if (!mSpatialPartition->mRenderByGroup)
 	{
@@ -849,10 +889,7 @@ void LLSpatialGroup::shift(const LLVector3 &offset)
 	{
 		for (U32 i = 0; i < 8; i++)
 		{
-			F32* v = mOcclusionVerts+i*3;
-			v[0] += offset.mV[0];
-			v[1] += offset.mV[1];
-			v[2] += offset.mV[2];
+			mOcclusionVerts[i].add(offset);
 		}
 	}
 }
@@ -1118,8 +1155,6 @@ LLSpatialGroup::LLSpatialGroup(OctreeNode* node, LLSpatialPartition* part) :
 	mDepth(0.f),
 	mLastUpdateDistance(-1.f), 
 	mLastUpdateTime(gFrameTimeSeconds),
-	mViewAngle(0.f),
-	mLastUpdateViewAngle(-1.f),
 	mAtlasList(4),
 	mCurUpdatingTime(0),
 	mCurUpdatingSlotp(NULL),
@@ -1128,13 +1163,25 @@ LLSpatialGroup::LLSpatialGroup(OctreeNode* node, LLSpatialPartition* part) :
 	sNodeCount++;
 	LLMemType mt(LLMemType::MTYPE_SPACE_PARTITION);
 
+	mBounds = (LLVector4a*) _mm_malloc(sizeof(LLVector4a) * V4_COUNT, 16);
+	mExtents = mBounds + EXTENTS;
+	mObjectBounds = mBounds + OBJECT_BOUNDS;
+	mObjectExtents = mBounds + OBJECT_EXTENTS;
+	mViewAngle = mBounds+VIEW_ANGLE;
+	mLastUpdateViewAngle = mBounds+LAST_VIEW_ANGLE;
+
+	mViewAngle->splat(0.f);
+	mLastUpdateViewAngle->splat(-1.f);
+	mExtents[0] = mExtents[1] = mObjectBounds[0] = mObjectBounds[0] = mObjectBounds[1] = 
+		mObjectExtents[0] = mObjectExtents[1] = *mViewAngle;
+
 	sg_assert(mOctreeNode->getListenerCount() == 0);
 	mOctreeNode->addListener(this);
 	setState(SG_INITIAL_STATE_MASK);
 	gPipeline.markRebuild(this, TRUE);
 
-	mBounds[0] = LLVector3(node->getCenter());
-	mBounds[1] = LLVector3(node->getSize());
+	mBounds[0] = node->getCenter();
+	mBounds[1] = node->getSize();
 
 	part->mLODSeed = (part->mLODSeed+1)%part->mLODPeriod;
 	mLODHash = part->mLODSeed;
@@ -1171,8 +1218,8 @@ void LLSpatialGroup::updateDistance(LLCamera &camera)
 #endif
 	if (!getData().empty())
 	{
-		mRadius = mSpatialPartition->mRenderByGroup ? mObjectBounds[1].magVec() :
-						(F32) mOctreeNode->getSize().magVec();
+		mRadius = mSpatialPartition->mRenderByGroup ? mObjectBounds[1].length3() :
+						(F32) mOctreeNode->getSize().length3();
 		mDistance = mSpatialPartition->calcDistance(this, camera);
 		mPixelArea = mSpatialPartition->calcPixelArea(this, camera);
 	}
@@ -1180,27 +1227,34 @@ void LLSpatialGroup::updateDistance(LLCamera &camera)
 
 F32 LLSpatialPartition::calcDistance(LLSpatialGroup* group, LLCamera& camera)
 {
-	LLVector3 eye = group->mObjectBounds[0] - camera.getOrigin();
+	LLVector4a eye;
+	LLVector4a origin;
+	origin.load3(camera.getOrigin().mV);
+
+	eye.setSub(group->mObjectBounds[0], origin);
 
 	F32 dist = 0.f;
 
 	if (group->mDrawMap.find(LLRenderPass::PASS_ALPHA) != group->mDrawMap.end())
 	{
-		LLVector3 v = eye;
-		dist = eye.normVec();
+		LLVector4a v = eye;
+
+		dist = eye.length3();
+		eye.normalize3fast();
 
 		if (!group->isState(LLSpatialGroup::ALPHA_DIRTY))
 		{
 			if (!group->mSpatialPartition->isBridge())
 			{
-				LLVector3 view_angle = LLVector3(eye * LLVector3(1,0,0),
-												 eye * LLVector3(0,1,0),
-												 eye * LLVector3(0,0,1));
+				LLVector4a view_angle = eye;
+
+				LLVector4a diff;
+				diff.setSub(view_angle, *group->mLastUpdateViewAngle);
 
-				if ((view_angle-group->mLastUpdateViewAngle).magVec() > 0.64f)
+				if (diff.length3() > 0.64f)
 				{
-					group->mViewAngle = view_angle;
-					group->mLastUpdateViewAngle = view_angle;
+					*group->mViewAngle = view_angle;
+					*group->mLastUpdateViewAngle = view_angle;
 					//for occasional alpha sorting within the group
 					//NOTE: If there is a trivial way to detect that alpha sorting here would not change the render order,
 					//not setting this node to dirty would be a very good thing
@@ -1214,17 +1268,20 @@ F32 LLSpatialPartition::calcDistance(LLSpatialGroup* group, LLCamera& camera)
 
 		LLVector3 at = camera.getAtAxis();
 
-		//front of bounding box
-		for (U32 i = 0; i < 3; i++)
-		{
-			v.mV[i] -= group->mObjectBounds[1].mV[i]*0.25f * at.mV[i];
-		}
+		LLVector4a ata;
+		ata.load3(at.mV);
 
-		group->mDepth = v * at;
+		LLVector4a t = ata;
+		//front of bounding box
+		t.mul(0.25f);
+		t.mul(group->mObjectBounds[1]);
+		v.sub(t);
+		
+		group->mDepth = v.dot3(ata);
 	}
 	else
 	{
-		dist = eye.magVec();
+		dist = eye.length3();
 	}
 
 	if (dist < 16.f)
@@ -1377,7 +1434,7 @@ void LLSpatialGroup::destroyGL()
 		}
 	}
 
-	delete [] mOcclusionVerts;
+	_mm_free(mOcclusionVerts);
 	mOcclusionVerts = NULL;
 
 	for (LLSpatialGroup::element_iter i = getData().begin(); i != getData().end(); ++i)
@@ -1420,8 +1477,8 @@ BOOL LLSpatialGroup::rebound()
 	}
 	else
 	{
-		LLVector3& newMin = mExtents[0];
-		LLVector3& newMax = mExtents[1];
+		LLVector4a& newMin = mExtents[0];
+		LLVector4a& newMax = mExtents[1];
 		LLSpatialGroup* group = (LLSpatialGroup*) mOctreeNode->getChild(0)->getListener(0);
 		group->clearState(SKIP_FRUSTUM_CHECK);
 		group->rebound();
@@ -1435,26 +1492,19 @@ BOOL LLSpatialGroup::rebound()
 			group = (LLSpatialGroup*) mOctreeNode->getChild(i)->getListener(0);
 			group->clearState(SKIP_FRUSTUM_CHECK);
 			group->rebound();
-			const LLVector3& max = group->mExtents[1];
-			const LLVector3& min = group->mExtents[0];
+			const LLVector4a& max = group->mExtents[1];
+			const LLVector4a& min = group->mExtents[0];
 
-			for (U32 j = 0; j < 3; j++)
-			{
-				if (max.mV[j] > newMax.mV[j])
-				{
-					newMax.mV[j] = max.mV[j];
-				}
-				if (min.mV[j] < newMin.mV[j])
-				{
-					newMin.mV[j] = min.mV[j];
-				}
-			}
+			newMax.setMax(max);
+			newMin.setMin(min);
 		}
 
 		boundObjects(FALSE, newMin, newMax);
 		
-		mBounds[0] = (newMin + newMax)*0.5f;
-		mBounds[1] = (newMax - newMin)*0.5f;
+		mBounds[0].setAdd(newMin, newMax);
+		mBounds[0].mul(0.5f);
+		mBounds[1].setSub(newMax, newMin);
+		mBounds[1].mul(0.5f);
 	}
 	
 	setState(OCCLUSION_DIRTY);
@@ -1539,7 +1589,7 @@ void LLSpatialGroup::doOcclusion(LLCamera* camera)
 				}
 
 				glBeginQueryARB(GL_SAMPLES_PASSED_ARB, mOcclusionQuery[LLViewerCamera::sCurCameraID]);					
-				glVertexPointer(3, GL_FLOAT, 0, mOcclusionVerts);
+				glVertexPointer(3, GL_FLOAT, 16, mOcclusionVerts);
 				if (camera->getOrigin().isExactlyZero())
 				{ //origin is invalid, draw entire box
 					glDrawRangeElements(GL_TRIANGLE_FAN, 0, 7, 8,
@@ -1580,8 +1630,11 @@ LLSpatialPartition::LLSpatialPartition(U32 data_mask, BOOL render_by_group, U32
 
 	LLGLNamePool::registerPool(&sQueryPool);
 
-	mOctree = new LLSpatialGroup::OctreeRoot(LLVector3d(0,0,0), 
-											LLVector3d(1,1,1), 
+	LLVector4a center, size;
+	center.splat(0.f);
+	size.splat(1.f);
+
+	mOctree = new LLSpatialGroup::OctreeRoot(center,size,
 											NULL);
 	new LLSpatialGroup(mOctree, this);
 }
@@ -1601,7 +1654,6 @@ LLSpatialGroup *LLSpatialPartition::put(LLDrawable *drawablep, BOOL was_visible)
 	LLMemType mt(LLMemType::MTYPE_SPACE_PARTITION);
 		
 	drawablep->updateSpatialExtents();
-	validate_drawable(drawablep);
 
 	//keep drawable from being garbage collected
 	LLPointer<LLDrawable> ptr = drawablep;
@@ -1685,16 +1737,16 @@ void LLSpatialPartition::move(LLDrawable *drawablep, LLSpatialGroup *curp, BOOL
 class LLSpatialShift : public LLSpatialGroup::OctreeTraveler
 {
 public:
-	LLSpatialShift(LLVector3 offset) : mOffset(offset) { }
+	const LLVector4a& mOffset;
+
+	LLSpatialShift(const LLVector4a& offset) : mOffset(offset) { }
 	virtual void visit(const LLSpatialGroup::OctreeNode* branch) 
 	{ 
 		((LLSpatialGroup*) branch->getListener(0))->shift(mOffset); 
 	}
-		
-	LLVector3 mOffset;
 };
 
-void LLSpatialPartition::shift(const LLVector3 &offset)
+void LLSpatialPartition::shift(const LLVector4a &offset)
 { //shift octree node bounding boxes by offset
 	LLMemType mt(LLMemType::MTYPE_SPACE_PARTITION);
 	LLSpatialShift shifter(offset);
@@ -1856,7 +1908,7 @@ public:
 class LLOctreeCullVisExtents: public LLOctreeCullShadow
 {
 public:
-	LLOctreeCullVisExtents(LLCamera* camera, LLVector3& min, LLVector3& max)
+	LLOctreeCullVisExtents(LLCamera* camera, LLVector4a& min, LLVector4a& max)
 		: LLOctreeCullShadow(camera), mMin(min), mMax(max), mEmpty(TRUE) { }
 
 	virtual bool earlyFail(LLSpatialGroup* group)
@@ -1923,8 +1975,8 @@ public:
 	}
 
 	BOOL mEmpty;
-	LLVector3& mMin;
-	LLVector3& mMax;
+	LLVector4a& mMin;
+	LLVector4a& mMax;
 };
 
 class LLOctreeCullDetectVisible: public LLOctreeCullShadow
@@ -2028,6 +2080,11 @@ void drawBox(const LLVector3& c, const LLVector3& r)
 	gGL.end();	
 }
 
+void drawBox(const LLVector4a& c, const LLVector4a& r)
+{
+	drawBox(reinterpret_cast<const LLVector3&>(c), reinterpret_cast<const LLVector3&>(r));
+}
+
 void drawBoxOutline(const LLVector3& pos, const LLVector3& size)
 {
 	LLVector3 v1 = size.scaledVec(LLVector3( 1, 1,1));
@@ -2074,6 +2131,11 @@ void drawBoxOutline(const LLVector3& pos, const LLVector3& size)
 	gGL.end();
 }
 
+void drawBoxOutline(const LLVector4a& pos, const LLVector4a& size)
+{
+	drawBoxOutline(reinterpret_cast<const LLVector3&>(pos), reinterpret_cast<const LLVector3&>(size));
+}
+
 class LLOctreeDirty : public LLOctreeTraveler<LLDrawable>
 {
 public:
@@ -2117,14 +2179,21 @@ BOOL LLSpatialPartition::isOcclusionEnabled()
 
 BOOL LLSpatialPartition::getVisibleExtents(LLCamera& camera, LLVector3& visMin, LLVector3& visMax)
 {
+	LLVector4a visMina, visMaxa;
+	visMina.load3(visMin.mV);
+	visMaxa.load3(visMax.mV);
+
 	{
 		LLFastTimer ftm(FTM_CULL_REBOUND);		
 		LLSpatialGroup* group = (LLSpatialGroup*) mOctree->getListener(0);
 		group->rebound();
 	}
 
-	LLOctreeCullVisExtents vis(&camera, visMin, visMax);
+	LLOctreeCullVisExtents vis(&camera, visMina, visMaxa);
 	vis.traverse(mOctree);
+
+	visMin.set(visMina.getF32());
+	visMax.set(visMina.getF32());
 	return vis.mEmpty;
 }
 
@@ -2187,25 +2256,36 @@ BOOL earlyFail(LLCamera* camera, LLSpatialGroup* group)
 	}
 
 	const F32 vel = SG_OCCLUSION_FUDGE*2.f;
-	LLVector3 c = group->mBounds[0];
-	LLVector3 r = group->mBounds[1] + LLVector3(vel,vel,vel);
-    
+	LLVector4a fudge;
+	fudge.splat(vel);
+
+	const LLVector4a& c = group->mBounds[0];
+	LLVector4a r;
+	r.setAdd(group->mBounds[1], fudge);
+
 	/*if (r.magVecSquared() > 1024.0*1024.0)
 	{
 		return TRUE;
 	}*/
 
-	LLVector3 e = camera->getOrigin();
+	LLVector4a e;
+	e.load3(camera->getOrigin().mV);
 	
-	LLVector3 min = c - r;
-	LLVector3 max = c + r;
+	LLVector4a min;
+	min.setSub(c,r);
+	LLVector4a max;
+	max.setAdd(c,r);
 	
-	for (U32 j = 0; j < 3; j++)
+	S32 lt = e.lessThan4(min).getComparisonMask() & 0x7;
+	if (lt)
 	{
-		if (e.mV[j] < min.mV[j] || e.mV[j] > max.mV[j])
-		{
-			return FALSE;
-		}
+		return FALSE;
+	}
+
+	S32 gt = e.greaterThan4(max).getComparisonMask() & 0x7;
+	if (gt)
+	{
+		return FALSE;
 	}
 
 	return TRUE;
@@ -2410,7 +2490,13 @@ void renderOctree(LLSpatialGroup* group)
 	}
 
 	gGL.color4fv(col.mV);
-	drawBox(group->mObjectBounds[0], group->mObjectBounds[1]*1.01f+LLVector3(0.001f, 0.001f, 0.001f));
+	LLVector4a fudge;
+	fudge.splat(0.001f);
+	LLVector4a size = group->mObjectBounds[1];
+	size.mul(1.01f);
+	size.add(fudge);
+
+	drawBox(group->mObjectBounds[0], fudge);
 	
 	gGL.setSceneBlendType(LLRender::BT_ALPHA);
 
@@ -2441,8 +2527,12 @@ void renderOctree(LLSpatialGroup* group)
 				for (LLSpatialGroup::drawmap_elem_t::iterator j = i->second.begin(); j != i->second.end(); ++j)
 				{
 					LLDrawInfo* draw_info = *j;
-					LLVector3 center = (draw_info->mExtents[1] + draw_info->mExtents[0])*0.5f;
-					LLVector3 size = (draw_info->mExtents[1] - draw_info->mExtents[0])*0.5f;
+					LLVector4a center;
+					center.setAdd(draw_info->mExtents[1], draw_info->mExtents[0]);
+					center.mul(0.5f);
+					LLVector4a size;
+					size.setSub(draw_info->mExtents[1], draw_info->mExtents[0]);
+					size.mul(0.5f);
 					drawBoxOutline(center, size);
 				}
 			}
@@ -2492,7 +2582,7 @@ void renderVisibility(LLSpatialGroup* group, LLCamera* camera)
 		else if (camera && group->mOcclusionVerts)
 		{
 			LLVertexBuffer::unbind();
-			glVertexPointer(3, GL_FLOAT, 0, group->mOcclusionVerts);
+			glVertexPointer(3, GL_FLOAT, 16, group->mOcclusionVerts);
 
 			glColor4f(1.0f, 0.f, 0.f, 0.5f);
 			glDrawRangeElements(GL_TRIANGLE_FAN, 0, 7, 8, GL_UNSIGNED_BYTE, get_box_fan_indices(camera, group->mBounds[0]));
@@ -2571,8 +2661,8 @@ void renderBoundingBox(LLDrawable* drawable, BOOL set_color = TRUE)
 		}
 	}
 
-	const LLVector3* ext;
-	LLVector3 pos, size;
+	const LLVector4a* ext;
+	LLVector4a pos, size;
 
 	//render face bounding boxes
 	for (S32 i = 0; i < drawable->getNumFaces(); i++)
@@ -2581,20 +2671,21 @@ void renderBoundingBox(LLDrawable* drawable, BOOL set_color = TRUE)
 
 		ext = facep->mExtents;
 
-		if (ext[0].isExactlyZero() && ext[1].isExactlyZero())
-		{
-			continue;
-		}
-		pos = (ext[0] + ext[1]) * 0.5f;
-		size = (ext[1] - ext[0]) * 0.5f;
+		pos.setAdd(ext[0], ext[1]);
+		pos.mul(0.5f);
+		size.setSub(ext[1], ext[0]);
+		size.mul(0.5f);
+		
 		drawBoxOutline(pos,size);
 	}
 
 	//render drawable bounding box
 	ext = drawable->getSpatialExtents();
 
-	pos = (ext[0] + ext[1]) * 0.5f;
-	size = (ext[1] - ext[0]) * 0.5f;
+	pos.setAdd(ext[0], ext[1]);
+	pos.mul(0.5f);
+	size.setSub(ext[1], ext[0]);
+	size.mul(0.5f);
 	
 	LLViewerObject* vobj = drawable->getVObj();
 	if (vobj && vobj->onActiveList())
@@ -2650,8 +2741,13 @@ void renderTexturePriority(LLDrawable* drawable)
 		//	gGL.color4f(1,0,1,1);
 		//}
 		
-		LLVector3 center = (facep->mExtents[1]+facep->mExtents[0])*0.5f;
-		LLVector3 size = (facep->mExtents[1]-facep->mExtents[0])*0.5f + LLVector3(0.01f, 0.01f, 0.01f);
+		LLVector4a center;
+		center.setAdd(facep->mExtents[1],facep->mExtents[0]);
+		center.mul(0.5f);
+		LLVector4a size;
+		size.setSub(facep->mExtents[1],facep->mExtents[0]);
+		size.mul(0.5f);
+		size.add(LLVector4a(0.01f));
 		drawBox(center, size);
 		
 		/*S32 boost = imagep->getBoostLevel();
@@ -2675,7 +2771,6 @@ void renderPoints(LLDrawable* drawablep)
 	{
 		gGL.begin(LLRender::POINTS);
 		gGL.color3f(1,1,1);
-		LLVector3 center(drawablep->getPositionGroup());
 		for (S32 i = 0; i < drawablep->getNumFaces(); i++)
 		{
 			gGL.vertex3fv(drawablep->getFace(i)->mCenterLocal.mV);
@@ -2707,8 +2802,12 @@ void renderShadowFrusta(LLDrawInfo* params)
 	LLGLEnable blend(GL_BLEND);
 	gGL.setSceneBlendType(LLRender::BT_ADD);
 
-	LLVector3 center = (params->mExtents[1]+params->mExtents[0])*0.5f;
-	LLVector3 size = (params->mExtents[1]-params->mExtents[0])*0.5f;
+	LLVector4a center;
+	center.setAdd(params->mExtents[1], params->mExtents[0]);
+	center.mul(0.5f);
+	LLVector4a size;
+	size.setSub(params->mExtents[1],params->mExtents[0]);
+	size.mul(0.5f);
 
 	if (gPipeline.mShadowCamera[4].AABBInFrustum(center, size))
 	{
@@ -2752,10 +2851,14 @@ void renderLights(LLDrawable* drawablep)
 			pushVerts(drawablep->getFace(i), LLVertexBuffer::MAP_VERTEX);
 		}
 
-		const LLVector3* ext = drawablep->getSpatialExtents();
+		const LLVector4a* ext = drawablep->getSpatialExtents();
 
-		LLVector3 pos = (ext[0] + ext[1]) * 0.5f;
-		LLVector3 size = (ext[1] - ext[0]) * 0.5f;
+		LLVector4a pos;
+		pos.setAdd(ext[0], ext[1]);
+		pos.mul(0.5f);
+		LLVector4a size;
+		size.setSub(ext[1], ext[0]);
+		size.mul(0.5f);
 
 		{
 			LLGLDepthTest depth(GL_FALSE, GL_TRUE);
@@ -2765,10 +2868,32 @@ void renderLights(LLDrawable* drawablep)
 
 		gGL.color4f(1,1,0,1);
 		F32 rad = drawablep->getVOVolume()->getLightRadius();
-		drawBoxOutline(pos, LLVector3(rad,rad,rad));
+		drawBoxOutline(pos, LLVector4a(rad));
 	}
 }
 
+class LLRenderOctreeRaycast : public LLOctreeTriangleRayIntersect
+{
+public:
+	
+	LLRenderOctreeRaycast(const LLVector3& start, const LLVector3& end)
+	{
+		mStart.load3(start.mV);
+		mEnd.load3(end.mV);
+		mDir.setSub(mEnd, mStart);
+	}
+
+	void visit(const LLOctreeNode<LLVolumeTriangle>* branch)
+	{
+		LLVolumeOctreeListener* vl = (LLVolumeOctreeListener*) branch->getListener(0);
+
+		LLVector3 center, size;
+		center.set(vl->mBounds[0].getF32());
+		size.set(vl->mBounds[1].getF32());
+
+		drawBoxOutline(center, size);
+	}
+};
 
 void renderRaycast(LLDrawable* drawablep)
 {
@@ -2787,6 +2912,27 @@ void renderRaycast(LLDrawable* drawablep)
 			glPolygonMode(GL_FRONT_AND_BACK, GL_LINE);
 			pushVerts(drawablep->getFace(gDebugRaycastFaceHit), LLVertexBuffer::MAP_VERTEX);
 			glPolygonMode(GL_FRONT_AND_BACK, GL_FILL);
+
+			LLVOVolume* vobj = drawablep->getVOVolume();
+			LLVolume* volume = vobj->getVolume();
+			if (volume && volume->getNumVolumeFaces() > gDebugRaycastFaceHit)
+			{
+				const LLVolumeFace& face = volume->getVolumeFace(gDebugRaycastFaceHit);
+				if (!face.mOctree)
+				{
+					((LLVolumeFace*) &face)->createOctree(); 
+				}
+
+				gGL.pushMatrix();
+				glMultMatrixf((F32*) vobj->getRelativeXform().mMatrix);
+				LLVector3 start, end;
+				start = vobj->agentPositionToVolume(gDebugRaycastStart);
+				end = vobj->agentPositionToVolume(gDebugRaycastEnd);
+
+				LLRenderOctreeRaycast render(start, end);
+				render.traverse(face.mOctree);
+				gGL.popMatrix();		
+			}
 		}
 		else if (drawablep->isAvatar())
 		{
@@ -2815,10 +2961,14 @@ void renderRaycast(LLDrawable* drawablep)
 		glPopMatrix();
 
 		// draw bounding box of prim
-		const LLVector3* ext = drawablep->getSpatialExtents();
+		const LLVector4a* ext = drawablep->getSpatialExtents();
 
-		LLVector3 pos = (ext[0] + ext[1]) * 0.5f;
-		LLVector3 size = (ext[1] - ext[0]) * 0.5f;
+		LLVector4a pos;
+		pos.setAdd(ext[0], ext[1]);
+		pos.mul(0.5f);
+		LLVector4a size;
+		size.setSub(ext[1], ext[0]);
+		size.mul(0.5f);
 
 		LLGLDepthTest depth(GL_FALSE, GL_TRUE);
 		gGL.color4f(0,0.5f,0.5f,1);
@@ -2905,8 +3055,8 @@ public:
 			return;
 		}
 
-		LLVector3 nodeCenter = group->mBounds[0];
-		LLVector3 octCenter = LLVector3(group->mOctreeNode->getCenter());
+		LLVector4a nodeCenter = group->mBounds[0];
+		LLVector4a octCenter = group->mOctreeNode->getCenter();
 
 		group->rebuildGeom();
 		group->rebuildMesh();
@@ -2935,8 +3085,14 @@ public:
 				if (drawable->isState(LLDrawable::IN_REBUILD_Q2))
 				{
 					gGL.color4f(0.6f, 0.6f, 0.1f, 1.f);
-					const LLVector3* ext = drawable->getSpatialExtents();
-					drawBoxOutline((ext[0]+ext[1])*0.5f, (ext[1]-ext[0])*0.5f);
+					const LLVector4a* ext = drawable->getSpatialExtents();
+					LLVector4a center;
+					center.setAdd(ext[0], ext[1]);
+					center.mul(0.5f);
+					LLVector4a size;
+					size.setSub(ext[1], ext[0]);
+					size.mul(0.5f);
+					drawBoxOutline(center, size);
 				}
 			}	
 
@@ -3167,7 +3323,11 @@ void LLSpatialPartition::renderDebug()
 void LLSpatialGroup::drawObjectBox(LLColor4 col)
 {
 	gGL.color4fv(col.mV);
-	drawBox(mObjectBounds[0], mObjectBounds[1]*1.01f+LLVector3(0.001f, 0.001f, 0.001f));
+	LLVector4a size;
+	size = mObjectBounds[0];
+	size.mul(1.01f);
+	size.add(LLVector4a(0.001f));
+	drawBox(mObjectBounds[0], size);
 }
 
 
@@ -3227,8 +3387,8 @@ public:
 
 			LLSpatialGroup* group = (LLSpatialGroup*) child->getListener(0);
 			
-			LLVector3 size;
-			LLVector3 center;
+			LLVector4a size;
+			LLVector4a center;
 			
 			size = group->mBounds[1];
 			center = group->mBounds[0];
@@ -3245,7 +3405,11 @@ public:
 				local_end   = mEnd   * local_matrix;
 			}
 
-			if (LLLineSegmentBoxIntersect(local_start, local_end, center, size))
+			LLVector4a start, end;
+			start.load3(local_start.mV);
+			end.load3(local_end.mV);
+
+			if (LLLineSegmentBoxIntersect(start, end, center, size))
 			{
 				check(child);
 			}
@@ -3336,6 +3500,7 @@ LLDrawInfo::LLDrawInfo(U16 start, U16 end, U32 count, U32 offset,
 	mDrawMode(LLRender::TRIANGLES)
 {
 	mVertexBuffer->validateRange(mStart, mEnd, mCount, mOffset);
+	mExtents = (LLVector4a*) _mm_malloc(sizeof(LLVector4a)*2, 16);
 
 	mDebugColor = (rand() << 16) + rand();
 }
@@ -3351,6 +3516,13 @@ LLDrawInfo::~LLDrawInfo()
 	{
 		mFace->setDrawInfo(NULL);
 	}
+
+	if (gDebugGL)
+	{
+		gPipeline.checkReferences(this);
+	}
+
+	_mm_free(mExtents);
 }
 
 void LLDrawInfo::validate()
diff --git a/indra/newview/llspatialpartition.h b/indra/newview/llspatialpartition.h
index 9b252d1035..119945113a 100644
--- a/indra/newview/llspatialpartition.h
+++ b/indra/newview/llspatialpartition.h
@@ -45,7 +45,7 @@
 #include "lldrawpool.h"
 #include "llface.h"
 #include "llviewercamera.h"
-
+#include "llvector4a.h"
 #include <queue>
 
 #define SG_STATE_INHERIT_MASK (OCCLUDED)
@@ -57,12 +57,15 @@ class LLSpatialGroup;
 class LLTextureAtlas;
 class LLTextureAtlasSlot;
 
+S32 AABBSphereIntersect(const LLVector4a& min, const LLVector4a& max, const LLVector3 &origin, const F32 &rad);
+S32 AABBSphereIntersectR2(const LLVector4a& min, const LLVector4a& max, const LLVector3 &origin, const F32 &radius_squared);
+
 S32 AABBSphereIntersect(const LLVector3& min, const LLVector3& max, const LLVector3 &origin, const F32 &rad);
 S32 AABBSphereIntersectR2(const LLVector3& min, const LLVector3& max, const LLVector3 &origin, const F32 &radius_squared);
 void pushVerts(LLFace* face, U32 mask);
 
 // get index buffer for binary encoded axis vertex buffer given a box at center being viewed by given camera
-U8* get_box_fan_indices(LLCamera* camera, const LLVector3& center);
+U8* get_box_fan_indices(LLCamera* camera, const LLVector4a& center);
 
 class LLDrawInfo : public LLRefCount 
 {
@@ -70,6 +73,18 @@ protected:
 	~LLDrawInfo();	
 	
 public:
+
+	LLDrawInfo(const LLDrawInfo& rhs)
+	{
+		*this = rhs;
+	}
+
+	const LLDrawInfo& operator=(const LLDrawInfo& rhs)
+	{
+		llerrs << "Illegal operation!" << llendl;
+		return *this;
+	}
+
 	LLDrawInfo(U16 start, U16 end, U32 count, U32 offset, 
 				LLViewerTexture* image, LLVertexBuffer* buffer, 
 				BOOL fullbright = FALSE, U8 bump = 0, BOOL particle = FALSE, F32 part_size = 0);
@@ -77,6 +92,8 @@ public:
 
 	void validate();
 
+	LLVector4a* mExtents;
+	
 	LLPointer<LLVertexBuffer> mVertexBuffer;
 	LLPointer<LLViewerTexture>     mTexture;
 	LLColor4U mGlowColor;
@@ -95,7 +112,6 @@ public:
 	LLSpatialGroup* mGroup;
 	LLFace* mFace; //associated face
 	F32 mDistance;
-	LLVector3 mExtents[2];
 	U32 mDrawMode;
 
 	struct CompareTexture
@@ -158,11 +174,24 @@ public:
 	};
 };
 
+LL_ALIGN_PREFIX(64)
 class LLSpatialGroup : public LLOctreeListener<LLDrawable>
 {
 	friend class LLSpatialPartition;
 	friend class LLOctreeStateCheck;
 public:
+
+	LLSpatialGroup(const LLSpatialGroup& rhs)
+	{
+		*this = rhs;
+	}
+
+	const LLSpatialGroup& operator=(const LLSpatialGroup& rhs)
+	{
+		llerrs << "Illegal operation!" << llendl;
+		return *this;
+	}
+
 	static U32 sNodeCount;
 	static BOOL sNoDelete; //deletion of spatial groups and draw info not allowed if TRUE
 
@@ -273,8 +302,8 @@ public:
 	BOOL isVisible() const;
 	BOOL isRecentlyVisible() const;
 	void setVisible();
-	void shift(const LLVector3 &offset);
-	BOOL boundObjects(BOOL empty, LLVector3& newMin, LLVector3& newMax);
+	void shift(const LLVector4a &offset);
+	BOOL boundObjects(BOOL empty, LLVector4a& newMin, LLVector4a& newMax);
 	void unbound();
 	BOOL rebound();
 	void buildOcclusion(); //rebuild mOcclusionVerts
@@ -322,6 +351,27 @@ public:
 	void addAtlas(LLTextureAtlas* atlasp, S8 recursive_level = 3) ;
 	void removeAtlas(LLTextureAtlas* atlasp, BOOL remove_group = TRUE, S8 recursive_level = 3) ;
 	void clearAtlasList() ;
+
+public:
+
+	typedef enum
+	{
+		BOUNDS = 0,
+		EXTENTS = 2,
+		OBJECT_BOUNDS = 4,
+		OBJECT_EXTENTS = 6,
+		VIEW_ANGLE = 8,
+		LAST_VIEW_ANGLE = 9,
+		V4_COUNT = 10
+	} eV4Index;
+
+	LLVector4a* mBounds; // bounding box (center, size) of this node and all its children (tight fit to objects)
+	LLVector4a* mExtents; // extents (min, max) of this node and all its children
+	LLVector4a* mObjectExtents; // extents (min, max) of objects in this node
+	LLVector4a* mObjectBounds; // bounding box (center, size) of objects in this node
+	LLVector4a* mViewAngle;
+	LLVector4a* mLastUpdateViewAngle;
+		
 private:
 	U32                     mCurUpdatingTime ;
 	//do not make the below two to use LLPointer
@@ -349,14 +399,9 @@ public:
 	F32 mBuilt;
 	OctreeNode* mOctreeNode;
 	LLSpatialPartition* mSpatialPartition;
-	LLVector3 mBounds[2]; // bounding box (center, size) of this node and all its children (tight fit to objects)
-	LLVector3 mExtents[2]; // extents (min, max) of this node and all its children
 	
-	LLVector3 mObjectExtents[2]; // extents (min, max) of objects in this node
-	LLVector3 mObjectBounds[2]; // bounding box (center, size) of objects in this node
-
 	LLPointer<LLVertexBuffer> mVertexBuffer;
-	F32*					mOcclusionVerts;
+	LLVector4a*				mOcclusionVerts;
 	GLuint					mOcclusionQuery[LLViewerCamera::NUM_CAMERAS];
 
 	U32 mBufferUsage;
@@ -367,13 +412,10 @@ public:
 	F32 mDepth;
 	F32 mLastUpdateDistance;
 	F32 mLastUpdateTime;
-			
-	LLVector3 mViewAngle;
-	LLVector3 mLastUpdateViewAngle;
 	
 	F32 mPixelArea;
 	F32 mRadius;
-};
+} LL_ALIGN_POSTFIX(64);
 
 class LLGeometryManager
 {
@@ -409,7 +451,7 @@ public:
 	
 	// If the drawable moves, move it here.
 	virtual void move(LLDrawable *drawablep, LLSpatialGroup *curp, BOOL immediate = FALSE);
-	virtual void shift(const LLVector3 &offset);
+	virtual void shift(const LLVector4a &offset);
 
 	virtual F32 calcDistance(LLSpatialGroup* group, LLCamera& camera);
 	virtual F32 calcPixelArea(LLSpatialGroup* group, LLCamera& camera);
@@ -467,7 +509,7 @@ public:
 	virtual void makeActive();
 	virtual void move(LLDrawable *drawablep, LLSpatialGroup *curp, BOOL immediate = FALSE);
 	virtual BOOL updateMove();
-	virtual void shiftPos(const LLVector3& vec);
+	virtual void shiftPos(const LLVector4a& vec);
 	virtual void cleanupReferences();
 	virtual LLSpatialPartition* asPartition()		{ return this; }
 	virtual LLSpatialBridge* asBridge()				{ return this; }
@@ -658,7 +700,7 @@ class LLHUDBridge : public LLVolumeBridge
 {
 public:
 	LLHUDBridge(LLDrawable* drawablep);
-	virtual void shiftPos(const LLVector3& vec);
+	virtual void shiftPos(const LLVector4a& vec);
 	virtual F32 calcPixelArea(LLSpatialGroup* group, LLCamera& camera);
 };
 
@@ -675,7 +717,7 @@ class LLHUDPartition : public LLBridgePartition
 {
 public:
 	LLHUDPartition();
-	virtual void shift(const LLVector3 &offset);
+	virtual void shift(const LLVector4a &offset);
 };
 
 extern const F32 SG_BOX_SIDE;
diff --git a/indra/newview/llsurfacepatch.cpp b/indra/newview/llsurfacepatch.cpp
index 48e4a6ccc7..06431d428f 100644
--- a/indra/newview/llsurfacepatch.cpp
+++ b/indra/newview/llsurfacepatch.cpp
@@ -860,8 +860,10 @@ void LLSurfacePatch::updateVisibility()
 	F32 stride_per_distance = DEFAULT_DELTA_ANGLE / mSurfacep->getMetersPerGrid();
 	U32 grids_per_patch_edge = mSurfacep->getGridsPerPatchEdge();
 
-	LLVector3 center = mCenterRegion + mSurfacep->getOriginAgent();
-	LLVector3 radius = LLVector3(mRadius, mRadius, mRadius);
+	LLVector4a center;
+	center.load3( (mCenterRegion + mSurfacep->getOriginAgent()).mV);
+	LLVector4a radius;
+	radius.splat(mRadius);
 
 	// sphere in frustum on global coordinates
 	if (LLViewerCamera::getInstance()->AABBInFrustumNoFarClip(center, radius))
diff --git a/indra/newview/llviewerdisplay.cpp b/indra/newview/llviewerdisplay.cpp
index 1490f8153c..d31b0f51fd 100644
--- a/indra/newview/llviewerdisplay.cpp
+++ b/indra/newview/llviewerdisplay.cpp
@@ -900,9 +900,11 @@ void display(BOOL rebuild, F32 zoom_factor, int subfield, BOOL for_snapshot)
 			render_ui();
 		}
 
-		gPipeline.rebuildGroups();
-
+		
 		LLSpatialGroup::sNoDelete = FALSE;
+		gPipeline.clearReferences();
+
+		gPipeline.rebuildGroups();
 	}
 
 	LLAppViewer::instance()->pingMainloopTimeout("Display:FrameStats");
@@ -1000,6 +1002,7 @@ void render_hud_attachments()
 		gPipeline.renderGeom(hud_cam);
 
 		LLSpatialGroup::sNoDelete = FALSE;
+		gPipeline.clearReferences();
 
 		render_hud_elements();
 
diff --git a/indra/newview/llviewerobject.cpp b/indra/newview/llviewerobject.cpp
index 3aecd0175d..0ed2d1da09 100644
--- a/indra/newview/llviewerobject.cpp
+++ b/indra/newview/llviewerobject.cpp
@@ -2863,21 +2863,26 @@ void LLViewerObject::setScale(const LLVector3 &scale, BOOL damped)
 	}
 }
 
-void LLViewerObject::updateSpatialExtents(LLVector3& newMin, LLVector3 &newMax)
-{
-	LLVector3 center = getRenderPosition();
-	LLVector3 size = getScale();
-	newMin.setVec(center-size);
-	newMax.setVec(center+size);
-	mDrawable->setPositionGroup((newMin + newMax) * 0.5f);
+void LLViewerObject::updateSpatialExtents(LLVector4a& newMin, LLVector4a &newMax)
+{
+	LLVector4a center;
+	center.load3(getRenderPosition().mV);
+	LLVector4a size;
+	size.load3(getScale().mV);
+	newMin.setSub(center, size);
+	newMax.setAdd(center, size);
+	
+	mDrawable->setPositionGroup(center);
 }
 
 F32 LLViewerObject::getBinRadius()
 {
 	if (mDrawable.notNull())
 	{
-		const LLVector3* ext = mDrawable->getSpatialExtents();
-		return (ext[1]-ext[0]).magVec();
+		const LLVector4a* ext = mDrawable->getSpatialExtents();
+		LLVector4a diff;
+		diff.setSub(ext[1], ext[0]);
+		return diff.length3();
 	}
 	
 	return getScale().magVec();
@@ -3469,12 +3474,21 @@ BOOL LLViewerObject::lineSegmentBoundingBox(const LLVector3& start, const LLVect
 		return FALSE;
 	}
 
-	const LLVector3* ext = mDrawable->getSpatialExtents();
+	const LLVector4a* ext = mDrawable->getSpatialExtents();
+
+	//VECTORIZE THIS
+	LLVector4a center;
+	center.setAdd(ext[1], ext[0]);
+	center.mul(0.5f);
+	LLVector4a size;
+	size.setSub(ext[1], ext[0]);
+	size.mul(0.5f);
 
-	LLVector3 center = (ext[1]+ext[0])*0.5f;
-	LLVector3 size = (ext[1]-ext[0])*0.5f;
+	LLVector4a starta, enda;
+	starta.load3(start.mV);
+	enda.load3(end.mV);
 
-	return LLLineSegmentBoxIntersect(start, end, center, size);
+	return LLLineSegmentBoxIntersect(starta, enda, center, size);
 }
 
 U8 LLViewerObject::getMediaType() const
diff --git a/indra/newview/llviewerobject.h b/indra/newview/llviewerobject.h
index 0fd0cbfa60..6ebd1cbe21 100644
--- a/indra/newview/llviewerobject.h
+++ b/indra/newview/llviewerobject.h
@@ -373,7 +373,7 @@ public:
 
 	void markForUpdate(BOOL priority);
 	void updateVolume(const LLVolumeParams& volume_params);
-	virtual	void updateSpatialExtents(LLVector3& min, LLVector3& max);
+	virtual	void updateSpatialExtents(LLVector4a& min, LLVector4a& max);
 	virtual F32 getBinRadius();
 	
 	LLBBox				getBoundingBoxAgent() const;
@@ -386,7 +386,7 @@ public:
 	void clearDrawableState(U32 state, BOOL recursive = TRUE);
 
 	// Called when the drawable shifts
-	virtual void onShift(const LLVector3 &shift_vector)	{ }
+	virtual void onShift(const LLVector4a &shift_vector)	{ }
 		
 	//////////////////////////////////////
 	//
diff --git a/indra/newview/llviewerpartsim.cpp b/indra/newview/llviewerpartsim.cpp
index 6b480ccf8e..41848e8b7a 100644
--- a/indra/newview/llviewerpartsim.cpp
+++ b/indra/newview/llviewerpartsim.cpp
@@ -161,8 +161,8 @@ LLViewerPartGroup::LLViewerPartGroup(const LLVector3 &center_agent, const F32 bo
 
 	if (group != NULL)
 	{
-		LLVector3 center(group->mOctreeNode->getCenter());
-		LLVector3 size(group->mOctreeNode->getSize());
+		LLVector3 center(group->mOctreeNode->getCenter().getF32());
+		LLVector3 size(group->mOctreeNode->getSize().getF32());
 		size += LLVector3(0.01f, 0.01f, 0.01f);
 		mMinObjPos = center - size;
 		mMaxObjPos = center + size;
diff --git a/indra/newview/llviewerwindow.cpp b/indra/newview/llviewerwindow.cpp
index 2422995288..0564f02ce5 100644
--- a/indra/newview/llviewerwindow.cpp
+++ b/indra/newview/llviewerwindow.cpp
@@ -231,6 +231,8 @@ LLVector2       gDebugRaycastTexCoord;
 LLVector3       gDebugRaycastNormal;
 LLVector3       gDebugRaycastBinormal;
 S32				gDebugRaycastFaceHit;
+LLVector3		gDebugRaycastStart;
+LLVector3		gDebugRaycastEnd;
 
 // HUD display lines in lower right
 BOOL				gDisplayWindInfo = FALSE;
@@ -2529,7 +2531,9 @@ void LLViewerWindow::updateUI()
 											  &gDebugRaycastIntersection,
 											  &gDebugRaycastTexCoord,
 											  &gDebugRaycastNormal,
-											  &gDebugRaycastBinormal);
+											  &gDebugRaycastBinormal,
+											  &gDebugRaycastStart,
+											  &gDebugRaycastEnd);
 	}
 
 	updateMouseDelta();
@@ -3445,7 +3449,9 @@ LLViewerObject* LLViewerWindow::cursorIntersect(S32 mouse_x, S32 mouse_y, F32 de
 												LLVector3 *intersection,
 												LLVector2 *uv,
 												LLVector3 *normal,
-												LLVector3 *binormal)
+												LLVector3 *binormal,
+												LLVector3* start,
+												LLVector3* end)
 {
 	S32 x = mouse_x;
 	S32 y = mouse_y;
@@ -3477,7 +3483,16 @@ LLViewerObject* LLViewerWindow::cursorIntersect(S32 mouse_x, S32 mouse_y, F32 de
 	LLVector3 mouse_world_start = mouse_point_global;
 	LLVector3 mouse_world_end   = mouse_point_global + mouse_direction_global * depth;
 
-	
+	if (start)
+	{
+		*start = mouse_world_start;
+	}
+
+	if (end)
+	{
+		*end = mouse_world_end;
+	}
+
 	LLViewerObject* found = NULL;
 
 	if (this_object)  // check only this object
diff --git a/indra/newview/llviewerwindow.h b/indra/newview/llviewerwindow.h
index 410445d97f..156a1ff8ad 100644
--- a/indra/newview/llviewerwindow.h
+++ b/indra/newview/llviewerwindow.h
@@ -361,7 +361,9 @@ public:
 									LLVector3 *intersection = NULL,
 									LLVector2 *uv = NULL,
 									LLVector3 *normal = NULL,
-									LLVector3 *binormal = NULL);
+									LLVector3 *binormal = NULL,
+									LLVector3* start = NULL,
+									LLVector3* end = NULL);
 	
 	
 	// Returns a pointer to the last object hit
@@ -507,6 +509,8 @@ extern LLVector2        gDebugRaycastTexCoord;
 extern LLVector3        gDebugRaycastNormal;
 extern LLVector3        gDebugRaycastBinormal;
 extern S32				gDebugRaycastFaceHit;
+extern LLVector3		gDebugRaycastStart;
+extern LLVector3		gDebugRaycastEnd;
 
 extern S32 CHAT_BAR_HEIGHT; 
 
diff --git a/indra/newview/llvoavatar.cpp b/indra/newview/llvoavatar.cpp
index ac109771dd..b097461822 100644
--- a/indra/newview/llvoavatar.cpp
+++ b/indra/newview/llvoavatar.cpp
@@ -1286,41 +1286,46 @@ void LLVOAvatar::updateDrawable(BOOL force_damped)
 	clearChanged(SHIFTED);
 }
 
-void LLVOAvatar::onShift(const LLVector3& shift_vector)
+void LLVOAvatar::onShift(const LLVector4a& shift_vector)
 {
-	mLastAnimExtents[0] += shift_vector;
-	mLastAnimExtents[1] += shift_vector;
+	const LLVector3& shift = reinterpret_cast<const LLVector3&>(shift_vector);
+	mLastAnimExtents[0] += shift;
+	mLastAnimExtents[1] += shift;
 	mNeedsImpostorUpdate = TRUE;
 	mNeedsAnimUpdate = TRUE;
 }
 
-void LLVOAvatar::updateSpatialExtents(LLVector3& newMin, LLVector3 &newMax)
+void LLVOAvatar::updateSpatialExtents(LLVector4a& newMin, LLVector4a &newMax)
 {
 	if (isImpostor() && !needsImpostorUpdate())
 	{
 		LLVector3 delta = getRenderPosition() -
-			((LLVector3(mDrawable->getPositionGroup())-mImpostorOffset));
+			((LLVector3(mDrawable->getPositionGroup().getF32())-mImpostorOffset));
 		
-		newMin = mLastAnimExtents[0] + delta;
-		newMax = mLastAnimExtents[1] + delta;
+		newMin.load3( (mLastAnimExtents[0] + delta).mV);
+		newMax.load3( (mLastAnimExtents[1] + delta).mV);
 	}
 	else
 	{
 		getSpatialExtents(newMin,newMax);
-		mLastAnimExtents[0] = newMin;
-		mLastAnimExtents[1] = newMax;
-		LLVector3 pos_group = (newMin+newMax)*0.5f;
-		mImpostorOffset = pos_group-getRenderPosition();
+		mLastAnimExtents[0].set(newMin.getF32());
+		mLastAnimExtents[1].set(newMax.getF32());
+		LLVector4a pos_group;
+		pos_group.setAdd(newMin,newMax);
+		pos_group.mul(0.5f);
+		mImpostorOffset = LLVector3(pos_group.getF32())-getRenderPosition();
 		mDrawable->setPositionGroup(pos_group);
 	}
 }
 
-void LLVOAvatar::getSpatialExtents(LLVector3& newMin, LLVector3& newMax)
+void LLVOAvatar::getSpatialExtents(LLVector4a& newMin, LLVector4a& newMax)
 {
-	LLVector3 buffer(0.25f, 0.25f, 0.25f);
-	LLVector3 pos = getRenderPosition();
-	newMin = pos - buffer;
-	newMax = pos + buffer;
+	LLVector4a buffer(0.25f);
+	LLVector4a pos;
+	pos.load3(getRenderPosition().mV);
+	newMin.setSub(pos, buffer);
+	newMax.setAdd(pos, buffer);
+
 	float max_attachment_span = DEFAULT_MAX_PRIM_SCALE * 5.0f;
 	
 	//stretch bounding box by joint positions
@@ -1329,12 +1334,20 @@ void LLVOAvatar::getSpatialExtents(LLVector3& newMin, LLVector3& newMax)
 		LLPolyMesh* mesh = i->second;
 		for (S32 joint_num = 0; joint_num < mesh->mJointRenderData.count(); joint_num++)
 		{
-			update_min_max(newMin, newMax, 
-						   mesh->mJointRenderData[joint_num]->mWorldMatrix->getTranslation());
+			LLVector4a trans;
+			trans.load3( mesh->mJointRenderData[joint_num]->mWorldMatrix->getTranslation().mV);
+			update_min_max(newMin, newMax, trans);
 		}
 	}
 
-	mPixelArea = LLPipeline::calcPixelArea((newMin+newMax)*0.5f, (newMax-newMin)*0.5f, *LLViewerCamera::getInstance());
+	LLVector4a center, size;
+	center.setAdd(newMin, newMax);
+	center.mul(0.5f);
+
+	size.setSub(newMax,newMin);
+	size.mul(0.5f);
+
+	mPixelArea = LLPipeline::calcPixelArea(center, size, *LLViewerCamera::getInstance());
 
 	//stretch bounding box by attachments
 	for (attachment_map_t::iterator iter = mAttachmentPoints.begin(); 
@@ -1361,15 +1374,17 @@ void LLVOAvatar::getSpatialExtents(LLVector3& newMin, LLVector3& newMax)
 					LLSpatialBridge* bridge = drawable->getSpatialBridge();
 					if (bridge)
 					{
-						const LLVector3* ext = bridge->getSpatialExtents();
-						LLVector3 distance = (ext[1] - ext[0]);
+						const LLVector4a* ext = bridge->getSpatialExtents();
+						LLVector4a distance;
+						distance.setSub(ext[1], ext[0]);
+						LLVector4a max_span(max_attachment_span);
+
+						S32 lt = distance.lessThan4(max_span).getComparisonMask() & 0x7;
 						
 						// Only add the prim to spatial extents calculations if it isn't a megaprim.
 						// max_attachment_span calculated at the start of the function 
 						// (currently 5 times our max prim size) 
-						if (distance.mV[0] < max_attachment_span 
-							&& distance.mV[1] < max_attachment_span
-							&& distance.mV[2] < max_attachment_span)
+						if (lt == 0x7)
 						{
 							update_min_max(newMin,newMax,ext[0]);
 							update_min_max(newMin,newMax,ext[1]);
@@ -1381,8 +1396,9 @@ void LLVOAvatar::getSpatialExtents(LLVector3& newMin, LLVector3& newMax)
 	}
 
 	//pad bounding box	
-	newMin -= buffer;
-	newMax += buffer;
+
+	newMin.sub(buffer);
+	newMax.add(buffer);
 }
 
 //-----------------------------------------------------------------------------
@@ -2371,7 +2387,7 @@ void LLVOAvatar::idleUpdateMisc(bool detailed_update)
 
 	if (isImpostor() && !mNeedsImpostorUpdate)
 	{
-		LLVector3 ext[2];
+		LLVector4a ext[2];
 		F32 distance;
 		LLVector3 angle;
 
@@ -2400,12 +2416,22 @@ void LLVOAvatar::idleUpdateMisc(bool detailed_update)
 			}
 			else
 			{
+				//VECTORIZE THIS
 				getSpatialExtents(ext[0], ext[1]);
-				if ((ext[1]-mImpostorExtents[1]).length() > 0.05f ||
-					(ext[0]-mImpostorExtents[0]).length() > 0.05f)
+				LLVector4a diff;
+				diff.setSub(ext[1], mImpostorExtents[1]);
+				if (diff.length3() > 0.05f)
 				{
 					mNeedsImpostorUpdate = TRUE;
 				}
+				else
+				{
+					diff.setSub(ext[0], mImpostorExtents[0]);
+					if (diff.length3() > 0.05f)
+					{
+						mNeedsImpostorUpdate = TRUE;
+					}
+				}
 			}
 		}
 	}
@@ -5151,9 +5177,13 @@ void LLVOAvatar::setPixelAreaAndAngle(LLAgent &agent)
 		return;
 	}
 
-	const LLVector3* ext = mDrawable->getSpatialExtents();
-	LLVector3 center = (ext[1] + ext[0]) * 0.5f;
-	LLVector3 size = (ext[1]-ext[0])*0.5f;
+	const LLVector4a* ext = mDrawable->getSpatialExtents();
+	LLVector4a center;
+	center.setAdd(ext[1], ext[0]);
+	center.mul(0.5f);
+	LLVector4a size;
+	size.setSub(ext[1], ext[0]);
+	size.mul(0.5f);
 
 	mImpostorPixelArea = LLPipeline::calcPixelArea(center, size, *LLViewerCamera::getInstance());
 
@@ -5165,7 +5195,7 @@ void LLVOAvatar::setPixelAreaAndAngle(LLAgent &agent)
 	}
 	else
 	{
-		F32 radius = size.length();
+		F32 radius = size.length3();
 		mAppAngle = (F32) atan2( radius, range) * RAD_TO_DEG;
 	}
 
@@ -7546,9 +7576,9 @@ void LLVOAvatar::cacheImpostorValues()
 	getImpostorValues(mImpostorExtents, mImpostorAngle, mImpostorDistance);
 }
 
-void LLVOAvatar::getImpostorValues(LLVector3* extents, LLVector3& angle, F32& distance) const
+void LLVOAvatar::getImpostorValues(LLVector4a* extents, LLVector3& angle, F32& distance) const
 {
-	const LLVector3* ext = mDrawable->getSpatialExtents();
+	const LLVector4a* ext = mDrawable->getSpatialExtents();
 	extents[0] = ext[0];
 	extents[1] = ext[1];
 
diff --git a/indra/newview/llvoavatar.h b/indra/newview/llvoavatar.h
index a851b7a150..71c3ed1cc2 100644
--- a/indra/newview/llvoavatar.h
+++ b/indra/newview/llvoavatar.h
@@ -127,7 +127,7 @@ public:
 	virtual BOOL   	 	 	isActive() const; // Whether this object needs to do an idleUpdate.
 	virtual void   	 	 	updateTextures();
 	virtual S32    	 	 	setTETexture(const U8 te, const LLUUID& uuid); // If setting a baked texture, need to request it from a non-local sim.
-	virtual void   	 	 	onShift(const LLVector3& shift_vector);
+	virtual void   	 	 	onShift(const LLVector4a& shift_vector);
 	virtual U32    	 	 	getPartitionType() const;
 	virtual const  	 	 	LLVector3 getRenderPosition() const;
 	virtual void   	 	 	updateDrawable(BOOL force_damped);
@@ -135,8 +135,8 @@ public:
 	virtual BOOL   	 	 	updateGeometry(LLDrawable *drawable);
 	virtual void   	 	 	setPixelAreaAndAngle(LLAgent &agent);
 	virtual void   	 	 	updateRegion(LLViewerRegion *regionp);
-	virtual void   	 	 	updateSpatialExtents(LLVector3& newMin, LLVector3 &newMax);
-	virtual void   	 	 	getSpatialExtents(LLVector3& newMin, LLVector3& newMax);
+	virtual void   	 	 	updateSpatialExtents(LLVector4a& newMin, LLVector4a &newMax);
+	virtual void   	 	 	getSpatialExtents(LLVector4a& newMin, LLVector4a& newMax);
 	virtual BOOL   	 	 	lineSegmentIntersect(const LLVector3& start, const LLVector3& end,
 													 S32 face = -1,                    // which face to check, -1 = ALL_SIDES
 													 BOOL pick_transparent = FALSE,
@@ -391,7 +391,7 @@ public:
 	BOOL 	    needsImpostorUpdate() const;
 	const LLVector3& getImpostorOffset() const;
 	const LLVector2& getImpostorDim() const;
-	void 		getImpostorValues(LLVector3* extents, LLVector3& angle, F32& distance) const;
+	void 		getImpostorValues(LLVector4a* extents, LLVector3& angle, F32& distance) const;
 	void 		cacheImpostorValues();
 	void 		setImpostorDim(const LLVector2& dim);
 	static void	resetImpostors();
@@ -402,7 +402,7 @@ private:
 	LLVector3	mImpostorOffset;
 	LLVector2	mImpostorDim;
 	BOOL		mNeedsAnimUpdate;
-	LLVector3	mImpostorExtents[2];
+	LL_ALIGN_16(LLVector4a	mImpostorExtents[2]);
 	LLVector3	mImpostorAngle;
 	F32			mImpostorDistance;
 	F32			mImpostorPixelArea;
diff --git a/indra/newview/llvograss.cpp b/indra/newview/llvograss.cpp
index 91c9b762c5..fe1e36cbe8 100644
--- a/indra/newview/llvograss.cpp
+++ b/indra/newview/llvograss.cpp
@@ -647,23 +647,23 @@ BOOL LLVOGrass::lineSegmentIntersect(const LLVector3& start, const LLVector3& en
 
 		U32 idx0 = 0,idx1 = 0,idx2 = 0;
 
-		if (LLTriangleRayIntersect(v[0], v[1], v[2], start, dir, &a, &b, &t, FALSE))
+		if (LLTriangleRayIntersect(v[0], v[1], v[2], start, dir, a, b, t, FALSE))
 		{
 			hit = TRUE;
 			idx0 = 0; idx1 = 1; idx2 = 2;
 		}
-		else if (LLTriangleRayIntersect(v[1], v[3], v[2], start, dir, &a, &b, &t, FALSE))
+		else if (LLTriangleRayIntersect(v[1], v[3], v[2], start, dir, a, b, t, FALSE))
 		{
 			hit = TRUE;
 			idx0 = 1; idx1 = 3; idx2 = 2;
 		}
-		else if (LLTriangleRayIntersect(v[2], v[1], v[0], start, dir, &a, &b, &t, FALSE))
+		else if (LLTriangleRayIntersect(v[2], v[1], v[0], start, dir, a, b, t, FALSE))
 		{
 			normal1 = -normal1;
 			hit = TRUE;
 			idx0 = 2; idx1 = 1; idx2 = 0;
 		}
-		else if (LLTriangleRayIntersect(v[2], v[3], v[1], start, dir, &a, &b, &t, FALSE))
+		else if (LLTriangleRayIntersect(v[2], v[3], v[1], start, dir, a, b, t, FALSE))
 		{
 			normal1 = -normal1;
 			hit = TRUE;
diff --git a/indra/newview/llvopartgroup.cpp b/indra/newview/llvopartgroup.cpp
index 3ba4ecad0c..b5fd8182c6 100644
--- a/indra/newview/llvopartgroup.cpp
+++ b/indra/newview/llvopartgroup.cpp
@@ -79,12 +79,14 @@ F32 LLVOPartGroup::getBinRadius()
 	return mScale.mV[0]*2.f;
 }
 
-void LLVOPartGroup::updateSpatialExtents(LLVector3& newMin, LLVector3& newMax)
+void LLVOPartGroup::updateSpatialExtents(LLVector4a& newMin, LLVector4a& newMax)
 {		
 	const LLVector3& pos_agent = getPositionAgent();
-	newMin = pos_agent - mScale;
-	newMax = pos_agent + mScale;
-	mDrawable->setPositionGroup(pos_agent);
+	newMin.load3( (pos_agent - mScale).mV);
+	newMax.load3( (pos_agent + mScale).mV);
+	LLVector4a pos;
+	pos.load3(pos_agent.mV);
+	mDrawable->setPositionGroup(pos);
 }
 
 BOOL LLVOPartGroup::idleUpdate(LLAgent &agent, LLWorld &world, const F64 &time)
diff --git a/indra/newview/llvopartgroup.h b/indra/newview/llvopartgroup.h
index 18583b4be9..771ae1c1eb 100644
--- a/indra/newview/llvopartgroup.h
+++ b/indra/newview/llvopartgroup.h
@@ -57,7 +57,7 @@ public:
 	BOOL idleUpdate(LLAgent &agent, LLWorld &world, const F64 &time);
 
 	virtual F32 getBinRadius();
-	virtual void updateSpatialExtents(LLVector3& newMin, LLVector3& newMax);
+	virtual void updateSpatialExtents(LLVector4a& newMin, LLVector4a& newMax);
 	virtual U32 getPartitionType() const;
 	
 	/*virtual*/ void setPixelAreaAndAngle(LLAgent &agent);
diff --git a/indra/newview/llvosurfacepatch.cpp b/indra/newview/llvosurfacepatch.cpp
index eef62ddf1a..02e7e7e60f 100644
--- a/indra/newview/llvosurfacepatch.cpp
+++ b/indra/newview/llvosurfacepatch.cpp
@@ -995,7 +995,13 @@ BOOL LLVOSurfacePatch::lineSegmentIntersect(const LLVector3& start, const LLVect
 
 	//step one meter at a time until intersection point found
 
-	const LLVector3* ext = mDrawable->getSpatialExtents();
+	//VECTORIZE THIS
+	const LLVector4a* exta = mDrawable->getSpatialExtents();
+
+	LLVector3 ext[2];
+	ext[0].set(exta[0].getF32());
+	ext[1].set(exta[1].getF32());
+
 	F32 rad = (delta*tdelta).magVecSquared();
 
 	F32 t = 0.f;
@@ -1057,13 +1063,16 @@ BOOL LLVOSurfacePatch::lineSegmentIntersect(const LLVector3& start, const LLVect
 	return FALSE;
 }
 
-void LLVOSurfacePatch::updateSpatialExtents(LLVector3& newMin, LLVector3 &newMax)
+void LLVOSurfacePatch::updateSpatialExtents(LLVector4a& newMin, LLVector4a &newMax)
 {
 	LLVector3 posAgent = getPositionAgent();
 	LLVector3 scale = getScale();
-	newMin = posAgent-scale*0.5f; // Changing to 2.f makes the culling a -little- better, but still wrong
-	newMax = posAgent+scale*0.5f;
-	mDrawable->setPositionGroup((newMin+newMax)*0.5f);
+	newMin.load3( (posAgent-scale*0.5f).mV); // Changing to 2.f makes the culling a -little- better, but still wrong
+	newMax.load3( (posAgent+scale*0.5f).mV);
+	LLVector4a pos;
+	pos.setAdd(newMin,newMax);
+	pos.mul(0.5f);
+	mDrawable->setPositionGroup(pos);
 }
 
 U32 LLVOSurfacePatch::getPartitionType() const
diff --git a/indra/newview/llvosurfacepatch.h b/indra/newview/llvosurfacepatch.h
index 10a5888526..15442e1947 100644
--- a/indra/newview/llvosurfacepatch.h
+++ b/indra/newview/llvosurfacepatch.h
@@ -78,7 +78,7 @@ public:
 	/*virtual*/ void updateTextures();
 	/*virtual*/ void setPixelAreaAndAngle(LLAgent &agent); // generate accurate apparent angle and area
 
-	/*virtual*/ void updateSpatialExtents(LLVector3& newMin, LLVector3& newMax);
+	/*virtual*/ void updateSpatialExtents(LLVector4a& newMin, LLVector4a& newMax);
 	/*virtual*/ BOOL isActive() const; // Whether this object needs to do an idleUpdate.
 
 	void setPatch(LLSurfacePatch *patchp);
diff --git a/indra/newview/llvotree.cpp b/indra/newview/llvotree.cpp
index b89c0cd638..d564643161 100644
--- a/indra/newview/llvotree.cpp
+++ b/indra/newview/llvotree.cpp
@@ -1238,7 +1238,7 @@ void LLVOTree::updateRadius()
 	mDrawable->setRadius(32.0f);
 }
 
-void LLVOTree::updateSpatialExtents(LLVector3& newMin, LLVector3& newMax)
+void LLVOTree::updateSpatialExtents(LLVector4a& newMin, LLVector4a& newMax)
 {
 	F32 radius = getScale().length()*0.05f;
 	LLVector3 center = getRenderPosition();
@@ -1248,9 +1248,11 @@ void LLVOTree::updateSpatialExtents(LLVector3& newMin, LLVector3& newMax)
 
 	center += LLVector3(0, 0, size.mV[2]) * getRotation();
 	
-	newMin.set(center-size);
-	newMax.set(center+size);
-	mDrawable->setPositionGroup(center);
+	newMin.load3((center-size).mV);
+	newMax.load3((center+size).mV);
+	LLVector4a pos;
+	pos.load3(center.mV);
+	mDrawable->setPositionGroup(pos);
 }
 
 BOOL LLVOTree::lineSegmentIntersect(const LLVector3& start, const LLVector3& end, S32 face, BOOL pick_transparent, S32 *face_hitp,
@@ -1263,8 +1265,13 @@ BOOL LLVOTree::lineSegmentIntersect(const LLVector3& start, const LLVector3& end
 		return FALSE;
 	}
 
-	const LLVector3* ext = mDrawable->getSpatialExtents();
+	const LLVector4a* exta = mDrawable->getSpatialExtents();
 
+	//VECTORIZE THIS
+	LLVector3 ext[2];
+	ext[0].set(exta[0].getF32());
+	ext[1].set(exta[1].getF32());
+	
 	LLVector3 center = (ext[1]+ext[0])*0.5f;
 	LLVector3 size = (ext[1]-ext[0]);
 
diff --git a/indra/newview/llvotree.h b/indra/newview/llvotree.h
index feac9e0675..2ce1b03d26 100644
--- a/indra/newview/llvotree.h
+++ b/indra/newview/llvotree.h
@@ -73,7 +73,7 @@ public:
 
 	/*virtual*/ LLDrawable* createDrawable(LLPipeline *pipeline);
 	/*virtual*/ BOOL		updateGeometry(LLDrawable *drawable);
-	/*virtual*/ void		updateSpatialExtents(LLVector3 &min, LLVector3 &max);
+	/*virtual*/ void		updateSpatialExtents(LLVector4a &min, LLVector4a &max);
 
 	virtual U32 getPartitionType() const;
 
diff --git a/indra/newview/llvovolume.cpp b/indra/newview/llvovolume.cpp
index a9f3abeef8..db9e0b88e1 100644
--- a/indra/newview/llvovolume.cpp
+++ b/indra/newview/llvovolume.cpp
@@ -701,7 +701,7 @@ void LLVOVolume::updateTextureVirtualSize()
 		const LLTextureEntry *te = face->getTextureEntry();
 		LLViewerTexture *imagep = face->getTexture();
 		if (!imagep || !te ||			
-			face->mExtents[0] == face->mExtents[1])
+			face->mExtents[0].equal3(face->mExtents[1]))
 		{
 			continue;
 		}
@@ -1332,7 +1332,7 @@ BOOL LLVOVolume::genBBoxes(BOOL force_global)
 {
 	BOOL res = TRUE;
 
-	LLVector3 min,max;
+	LLVector4a min,max;
 
 	BOOL rebuild = mDrawable->isState(LLDrawable::REBUILD_VOLUME | LLDrawable::REBUILD_POSITION);
 
@@ -1356,17 +1356,8 @@ BOOL LLVOVolume::genBBoxes(BOOL force_global)
 			}
 			else
 			{
-				for (U32 i = 0; i < 3; i++)
-				{
-					if (face->mExtents[0].mV[i] < min.mV[i])
-					{
-						min.mV[i] = face->mExtents[0].mV[i];
-					}
-					if (face->mExtents[1].mV[i] > max.mV[i])
-					{
-						max.mV[i] = face->mExtents[1].mV[i];
-					}
-				}
+				min.setMin(face->mExtents[0]);
+				max.setMax(face->mExtents[1]);
 			}
 		}
 	}
@@ -1374,7 +1365,9 @@ BOOL LLVOVolume::genBBoxes(BOOL force_global)
 	if (rebuild)
 	{
 		mDrawable->setSpatialExtents(min,max);
-		mDrawable->setPositionGroup((min+max)*0.5f);	
+		min.add(max);
+		min.mul(0.5f);
+		mDrawable->setPositionGroup(min);	
 	}
 
 	updateRadius();
@@ -3007,7 +3000,7 @@ void LLVOVolume::setSelected(BOOL sel)
 	}
 }
 
-void LLVOVolume::updateSpatialExtents(LLVector3& newMin, LLVector3& newMax)
+void LLVOVolume::updateSpatialExtents(LLVector4a& newMin, LLVector4a& newMax)
 {		
 }
 
@@ -3039,7 +3032,7 @@ F32 LLVOVolume::getBinRadius()
 		}
 	}
 
-	const LLVector3* ext = mDrawable->getSpatialExtents();
+	const LLVector4a* ext = mDrawable->getSpatialExtents();
 	
 	BOOL shrink_wrap = mDrawable->isAnimating();
 	BOOL alpha_wrap = FALSE;
@@ -3071,7 +3064,10 @@ F32 LLVOVolume::getBinRadius()
 	}
 	else if (shrink_wrap)
 	{
-		radius = (ext[1]-ext[0]).length()*0.5f;
+		LLVector4a rad;
+		rad.setSub(ext[1], ext[0]);
+		
+		radius = rad.length3()*0.5f;
 	}
 	else if (mDrawable->isStatic())
 	{
@@ -3107,7 +3103,7 @@ const LLVector3 LLVOVolume::getPivotPositionAgent() const
 	return LLViewerObject::getPivotPositionAgent();
 }
 
-void LLVOVolume::onShift(const LLVector3 &shift_vector)
+void LLVOVolume::onShift(const LLVector4a &shift_vector)
 {
 	if (mVolumeImpl)
 	{
@@ -3610,7 +3606,6 @@ void LLVolumeGeometryManager::rebuildGeom(LLSpatialGroup* group)
 							}
 						}
 					}
-
 				}
 
 				continue;
@@ -4217,7 +4212,7 @@ LLHUDPartition::LLHUDPartition()
 	mLODPeriod = 1;
 }
 
-void LLHUDPartition::shift(const LLVector3 &offset)
+void LLHUDPartition::shift(const LLVector4a &offset)
 {
 	//HUD objects don't shift with region crossing.  That would be silly.
 }
diff --git a/indra/newview/llvovolume.h b/indra/newview/llvovolume.h
index 2776988a12..d5606034d0 100644
--- a/indra/newview/llvovolume.h
+++ b/indra/newview/llvovolume.h
@@ -66,7 +66,7 @@ public:
 	virtual void onSetVolume(const LLVolumeParams &volume_params, const S32 detail) = 0;
 	virtual void onSetScale(const LLVector3 &scale, BOOL damped) = 0;
 	virtual void onParameterChanged(U16 param_type, LLNetworkData *data, BOOL in_use, bool local_origin) = 0;
-	virtual void onShift(const LLVector3 &shift_vector) = 0;
+	virtual void onShift(const LLVector4a &shift_vector) = 0;
 	virtual bool isVolumeUnique() const = 0; // Do we need a unique LLVolume instance?
 	virtual bool isVolumeGlobal() const = 0; // Are we in global space?
 	virtual bool isActive() const = 0; // Is this object currently active?
@@ -145,7 +145,7 @@ public:
 
 				void	markForUpdate(BOOL priority)			{ LLViewerObject::markForUpdate(priority); mVolumeChanged = TRUE; }
 
-	/*virtual*/ void	onShift(const LLVector3 &shift_vector); // Called when the drawable shifts
+	/*virtual*/ void	onShift(const LLVector4a &shift_vector); // Called when the drawable shifts
 
 	/*virtual*/ void	parameterChanged(U16 param_type, bool local_origin);
 	/*virtual*/ void	parameterChanged(U16 param_type, LLNetworkData* data, BOOL in_use, bool local_origin);
@@ -201,7 +201,7 @@ public:
 				void	regenFaces();
 				BOOL	genBBoxes(BOOL force_global);
 				void	preRebuild();
-	virtual		void	updateSpatialExtents(LLVector3& min, LLVector3& max);
+	virtual		void	updateSpatialExtents(LLVector4a& min, LLVector4a& max);
 	virtual		F32		getBinRadius();
 	
 	virtual U32 getPartitionType() const;
diff --git a/indra/newview/llvowater.cpp b/indra/newview/llvowater.cpp
index a8c4625f6e..7c1b22d432 100644
--- a/indra/newview/llvowater.cpp
+++ b/indra/newview/llvowater.cpp
@@ -258,15 +258,21 @@ void LLVOWater::setIsEdgePatch(const BOOL edge_patch)
 	mIsEdgePatch = edge_patch;
 }
 
-void LLVOWater::updateSpatialExtents(LLVector3 &newMin, LLVector3& newMax)
+void LLVOWater::updateSpatialExtents(LLVector4a &newMin, LLVector4a& newMax)
 {
-	LLVector3 pos = getPositionAgent();
-	LLVector3 scale = getScale();
-
-	newMin = pos - scale * 0.5f;
-	newMax = pos + scale * 0.5f;
+	LLVector4a pos;
+	pos.load3(getPositionAgent().mV);
+	LLVector4a scale;
+	scale.load3(getScale().mV);
+	scale.mul(0.5f);
+
+	newMin.setSub(pos, scale);
+	newMax.setAdd(pos, scale);
+	
+	pos.setAdd(newMin,newMax);
+	pos.mul(0.5f);
 
-	mDrawable->setPositionGroup((newMin + newMax) * 0.5f);
+	mDrawable->setPositionGroup(pos);
 }
 
 U32 LLVOWater::getPartitionType() const
diff --git a/indra/newview/llvowater.h b/indra/newview/llvowater.h
index 3cc031e589..a868afe58b 100644
--- a/indra/newview/llvowater.h
+++ b/indra/newview/llvowater.h
@@ -66,7 +66,7 @@ public:
 	/*virtual*/ BOOL idleUpdate(LLAgent &agent, LLWorld &world, const F64 &time);
 	/*virtual*/ LLDrawable* createDrawable(LLPipeline *pipeline);
 	/*virtual*/ BOOL        updateGeometry(LLDrawable *drawable);
-	/*virtual*/ void		updateSpatialExtents(LLVector3& newMin, LLVector3& newMax);
+	/*virtual*/ void		updateSpatialExtents(LLVector4a& newMin, LLVector4a& newMax);
 
 	/*virtual*/ void updateTextures();
 	/*virtual*/ void setPixelAreaAndAngle(LLAgent &agent); // generate accurate apparent angle and area
diff --git a/indra/newview/pipeline.cpp b/indra/newview/pipeline.cpp
index 91c3805d3b..01027e5be6 100644
--- a/indra/newview/pipeline.cpp
+++ b/indra/newview/pipeline.cpp
@@ -1508,11 +1508,214 @@ F32 LLPipeline::calcPixelArea(LLVector3 center, LLVector3 size, LLCamera &camera
 	return radius*radius * F_PI;
 }
 
+//static
+F32 LLPipeline::calcPixelArea(const LLVector4a& center, const LLVector4a& size, LLCamera &camera)
+{
+	LLVector4a origin;
+	origin.load3(camera.getOrigin().mV);
+
+	LLVector4a lookAt;
+	lookAt.setSub(center, origin);
+	F32 dist = lookAt.length3();
+
+	//ramp down distance for nearby objects
+	//shrink dist by dist/16.
+	if (dist < 16.f)
+	{
+		dist /= 16.f;
+		dist *= dist;
+		dist *= 16.f;
+	}
+
+	//get area of circle around node
+	F32 app_angle = atanf(size.length3()/dist);
+	F32 radius = app_angle*LLDrawable::sCurPixelAngle;
+	return radius*radius * F_PI;
+}
+
 void LLPipeline::grabReferences(LLCullResult& result)
 {
 	sCull = &result;
 }
 
+void LLPipeline::clearReferences()
+{
+	sCull = NULL;
+}
+
+void check_references(LLSpatialGroup* group, LLDrawable* drawable)
+{
+	for (LLSpatialGroup::element_iter i = group->getData().begin(); i != group->getData().end(); ++i)
+	{
+		if (drawable == *i)
+		{
+			llerrs << "LLDrawable deleted while actively reference by LLPipeline." << llendl;
+		}
+	}			
+}
+
+void check_references(LLDrawable* drawable, LLFace* face)
+{
+	for (S32 i = 0; i < drawable->getNumFaces(); ++i)
+	{
+		if (drawable->getFace(i) == face)
+		{
+			llerrs << "LLFace deleted while actively referenced by LLPipeline." << llendl;
+		}
+	}
+}
+
+void check_references(LLSpatialGroup* group, LLFace* face)
+{
+	for (LLSpatialGroup::element_iter i = group->getData().begin(); i != group->getData().end(); ++i)
+	{
+		LLDrawable* drawable = *i;
+		check_references(drawable, face);
+	}			
+}
+
+void LLPipeline::checkReferences(LLFace* face)
+{
+#if 0
+	if (sCull)
+	{
+		for (LLCullResult::sg_list_t::iterator iter = sCull->beginVisibleGroups(); iter != sCull->endVisibleGroups(); ++iter)
+		{
+			LLSpatialGroup* group = *iter;
+			check_references(group, face);
+		}
+
+		for (LLCullResult::sg_list_t::iterator iter = sCull->beginAlphaGroups(); iter != sCull->endAlphaGroups(); ++iter)
+		{
+			LLSpatialGroup* group = *iter;
+			check_references(group, face);
+		}
+
+		for (LLCullResult::sg_list_t::iterator iter = sCull->beginDrawableGroups(); iter != sCull->endDrawableGroups(); ++iter)
+		{
+			LLSpatialGroup* group = *iter;
+			check_references(group, face);
+		}
+
+		for (LLCullResult::drawable_list_t::iterator iter = sCull->beginVisibleList(); iter != sCull->endVisibleList(); ++iter)
+		{
+			LLDrawable* drawable = *iter;
+			check_references(drawable, face);	
+		}
+	}
+#endif
+}
+
+void LLPipeline::checkReferences(LLDrawable* drawable)
+{
+#if 0
+	if (sCull)
+	{
+		for (LLCullResult::sg_list_t::iterator iter = sCull->beginVisibleGroups(); iter != sCull->endVisibleGroups(); ++iter)
+		{
+			LLSpatialGroup* group = *iter;
+			check_references(group, drawable);
+		}
+
+		for (LLCullResult::sg_list_t::iterator iter = sCull->beginAlphaGroups(); iter != sCull->endAlphaGroups(); ++iter)
+		{
+			LLSpatialGroup* group = *iter;
+			check_references(group, drawable);
+		}
+
+		for (LLCullResult::sg_list_t::iterator iter = sCull->beginDrawableGroups(); iter != sCull->endDrawableGroups(); ++iter)
+		{
+			LLSpatialGroup* group = *iter;
+			check_references(group, drawable);
+		}
+
+		for (LLCullResult::drawable_list_t::iterator iter = sCull->beginVisibleList(); iter != sCull->endVisibleList(); ++iter)
+		{
+			if (drawable == *iter)
+			{
+				llerrs << "LLDrawable deleted while actively referenced by LLPipeline." << llendl;
+			}
+		}
+	}
+#endif
+}
+
+void check_references(LLSpatialGroup* group, LLDrawInfo* draw_info)
+{
+	for (LLSpatialGroup::draw_map_t::iterator i = group->mDrawMap.begin(); i != group->mDrawMap.end(); ++i)
+	{
+		LLSpatialGroup::drawmap_elem_t& draw_vec = i->second;
+		for (LLSpatialGroup::drawmap_elem_t::iterator j = draw_vec.begin(); j != draw_vec.end(); ++j)
+		{
+			LLDrawInfo* params = *j;
+			if (params == draw_info)
+			{
+				llerrs << "LLDrawInfo deleted while actively referenced by LLPipeline." << llendl;
+			}
+		}
+	}
+}
+
+
+void LLPipeline::checkReferences(LLDrawInfo* draw_info)
+{
+#if 0
+	if (sCull)
+	{
+		for (LLCullResult::sg_list_t::iterator iter = sCull->beginVisibleGroups(); iter != sCull->endVisibleGroups(); ++iter)
+		{
+			LLSpatialGroup* group = *iter;
+			check_references(group, draw_info);
+		}
+
+		for (LLCullResult::sg_list_t::iterator iter = sCull->beginAlphaGroups(); iter != sCull->endAlphaGroups(); ++iter)
+		{
+			LLSpatialGroup* group = *iter;
+			check_references(group, draw_info);
+		}
+
+		for (LLCullResult::sg_list_t::iterator iter = sCull->beginDrawableGroups(); iter != sCull->endDrawableGroups(); ++iter)
+		{
+			LLSpatialGroup* group = *iter;
+			check_references(group, draw_info);
+		}
+	}
+#endif
+}
+
+void LLPipeline::checkReferences(LLSpatialGroup* group)
+{
+#if 0
+	if (sCull)
+	{
+		for (LLCullResult::sg_list_t::iterator iter = sCull->beginVisibleGroups(); iter != sCull->endVisibleGroups(); ++iter)
+		{
+			if (group == *iter)
+			{
+				llerrs << "LLSpatialGroup deleted while actively referenced by LLPipeline." << llendl;
+			}
+		}
+
+		for (LLCullResult::sg_list_t::iterator iter = sCull->beginAlphaGroups(); iter != sCull->endAlphaGroups(); ++iter)
+		{
+			if (group == *iter)
+			{
+				llerrs << "LLSpatialGroup deleted while actively referenced by LLPipeline." << llendl;
+			}
+		}
+
+		for (LLCullResult::sg_list_t::iterator iter = sCull->beginDrawableGroups(); iter != sCull->endDrawableGroups(); ++iter)
+		{
+			if (group == *iter)
+			{
+				llerrs << "LLSpatialGroup deleted while actively referenced by LLPipeline." << llendl;
+			}
+		}
+	}
+#endif
+}
+
+
 BOOL LLPipeline::visibleObjectsInFrustum(LLCamera& camera)
 {
 	for (LLWorld::region_list_t::const_iterator iter = LLWorld::getInstance()->getRegionList().begin(); 
@@ -1714,7 +1917,7 @@ void LLPipeline::markNotCulled(LLSpatialGroup* group, LLCamera& camera)
 	}
 
 	if (sMinRenderSize > 0.f && 
-			llmax(llmax(group->mBounds[1].mV[0], group->mBounds[1].mV[1]), group->mBounds[1].mV[2]) < sMinRenderSize)
+			llmax(llmax(group->mBounds[1][0], group->mBounds[1][1]), group->mBounds[1][2]) < sMinRenderSize)
 	{
 		return;
 	}
@@ -2100,6 +2303,9 @@ void LLPipeline::shiftObjects(const LLVector3 &offset)
 	glClear(GL_DEPTH_BUFFER_BIT);
 	gDepthDirty = TRUE;
 		
+	LLVector4a offseta;
+	offseta.load3(offset.mV);
+
 	for (LLDrawable::drawable_vector_t::iterator iter = mShiftList.begin();
 		 iter != mShiftList.end(); iter++)
 	{
@@ -2108,7 +2314,7 @@ void LLPipeline::shiftObjects(const LLVector3 &offset)
 		{
 			continue;
 		}	
-		drawablep->shiftPos(offset);	
+		drawablep->shiftPos(offseta);	
 		drawablep->clearState(LLDrawable::ON_SHIFT_LIST);
 	}
 	mShiftList.resize(0);
@@ -2122,7 +2328,7 @@ void LLPipeline::shiftObjects(const LLVector3 &offset)
 			LLSpatialPartition* part = region->getSpatialPartition(i);
 			if (part)
 			{
-				part->shift(offset);
+				part->shift(offseta);
 			}
 		}
 	}
@@ -2659,8 +2865,10 @@ void LLPipeline::postSort(LLCamera& camera)
 			{
 				if (sMinRenderSize > 0.f)
 				{
-					LLVector3 bounds = (*k)->mExtents[1]-(*k)->mExtents[0];
-					if (llmax(llmax(bounds.mV[0], bounds.mV[1]), bounds.mV[2]) > sMinRenderSize)
+					LLVector4a bounds;
+					bounds.setSub((*k)->mExtents[1],(*k)->mExtents[0]);
+
+					if (llmax(llmax(bounds[0], bounds[1]), bounds[2]) > sMinRenderSize)
 					{
 						sCull->pushDrawInfo(j->first, *k);
 					}
@@ -6770,8 +6978,9 @@ void LLPipeline::renderDeferredLighting()
 					}
 
 
-					LLVector3 center = drawablep->getPositionAgent();
-					F32* c = center.mV;
+					LLVector4a center;
+					center.load3(drawablep->getPositionAgent().mV);
+					const F32* c = center.getF32();
 					F32 s = volume->getLightRadius()*1.5f;
 
 					LLColor3 col = volume->getLightColor();
@@ -6787,7 +6996,9 @@ void LLPipeline::renderDeferredLighting()
 						continue;
 					}
 
-					if (camera->AABBInFrustumNoFarClip(center, LLVector3(s,s,s)) == 0)
+					LLVector4a sa;
+					sa.splat(s);
+					if (camera->AABBInFrustumNoFarClip(center, sa) == 0)
 					{
 						continue;
 					}
@@ -6865,8 +7076,9 @@ void LLPipeline::renderDeferredLighting()
 
 					LLVOVolume* volume = drawablep->getVOVolume();
 
-					LLVector3 center = drawablep->getPositionAgent();
-					F32* c = center.mV;
+					LLVector4a center;
+					center.load3(drawablep->getPositionAgent().mV);
+					const F32* c = center.getF32();
 					F32 s = volume->getLightRadius()*1.5f;
 
 					sVisibleLightCount++;
@@ -8952,7 +9164,7 @@ void LLPipeline::generateImpostor(LLVOAvatar* avatar)
 
 	stateSort(*LLViewerCamera::getInstance(), result);
 	
-	const LLVector3* ext = avatar->mDrawable->getSpatialExtents();
+	const LLVector4a* ext = avatar->mDrawable->getSpatialExtents();
 	LLVector3 pos(avatar->getRenderPosition()+avatar->getImpostorOffset());
 
 	LLCamera camera = *viewer_camera;
@@ -8961,18 +9173,23 @@ void LLPipeline::generateImpostor(LLVOAvatar* avatar)
 	
 	LLVector2 tdim;
 
-	LLVector3 half_height = (ext[1]-ext[0])*0.5f;
 
-	LLVector3 left = camera.getLeftAxis();
-	left *= left;
-	left.normalize();
+	LLVector4a half_height;
+	half_height.setSub(ext[1], ext[0]);
+	half_height.mul(0.5f);
+
+	LLVector4a left;
+	left.load3(camera.getLeftAxis().mV);
+	left.mul(left);
+	left.normalize3fast();
 
-	LLVector3 up = camera.getUpAxis();
-	up *= up;
-	up.normalize();
+	LLVector4a up;
+	up.load3(camera.getUpAxis().mV);
+	up.mul(up);
+	up.normalize3fast();
 
-	tdim.mV[0] = fabsf(half_height * left);
-	tdim.mV[1] = fabsf(half_height * up);
+	tdim.mV[0] = fabsf(half_height.dot3(left));
+	tdim.mV[1] = fabsf(half_height.dot3(up));
 
 	glMatrixMode(GL_PROJECTION);
 	glPushMatrix();
diff --git a/indra/newview/pipeline.h b/indra/newview/pipeline.h
index c9384f5ba2..52f943cd1d 100644
--- a/indra/newview/pipeline.h
+++ b/indra/newview/pipeline.h
@@ -217,6 +217,7 @@ public:
 
 	//calculate pixel area of given box from vantage point of given camera
 	static F32 calcPixelArea(LLVector3 center, LLVector3 size, LLCamera& camera);
+	static F32 calcPixelArea(const LLVector4a& center, const LLVector4a& size, LLCamera &camera);
 
 	void stateSort(LLCamera& camera, LLCullResult& result);
 	void stateSort(LLSpatialGroup* group, LLCamera& camera);
@@ -229,6 +230,14 @@ public:
 	void renderGroups(LLRenderPass* pass, U32 type, U32 mask, BOOL texture);
 
 	void grabReferences(LLCullResult& result);
+	void clearReferences();
+
+	//check references will assert that there are no references in sCullResult to the provided data
+	void checkReferences(LLFace* face);
+	void checkReferences(LLDrawable* drawable);
+	void checkReferences(LLDrawInfo* draw_info);
+	void checkReferences(LLSpatialGroup* group);
+
 
 	void renderGeom(LLCamera& camera, BOOL forceVBOUpdate = FALSE);
 	void renderGeomDeferred(LLCamera& camera);