Unstaged changes cleanup. Further vectorization. Change in binormal/bitangent calculation.

2013-10-09 14:47:06 -05:00
parent b473661cf4
commit f25eb07fab
51 changed files with 1987 additions and 1895 deletions
--- a/indra/llappearance/llpolymesh.cpp
+++ b/indra/llappearance/llpolymesh.cpp
@@ -1302,6 +1302,57 @@ BOOL LLPolyMesh::saveOBJ(LLFILE *fp)
 	return TRUE;
 }

+// Finds binormal based on three vertices with texture coordinates.
+// Fills in dummy values if the triangle has degenerate texture coordinates.
+void calc_binormal_from_triangle(LLVector4a& binormal,
+
+	const LLVector4a& pos0,
+	const LLVector2& tex0,
+	const LLVector4a& pos1,
+	const LLVector2& tex1,
+	const LLVector4a& pos2,
+	const LLVector2& tex2)
+{
+	LLVector4a rx0( pos0[VX], tex0.mV[VX], tex0.mV[VY] );
+	LLVector4a rx1( pos1[VX], tex1.mV[VX], tex1.mV[VY] );
+	LLVector4a rx2( pos2[VX], tex2.mV[VX], tex2.mV[VY] );
+	
+	LLVector4a ry0( pos0[VY], tex0.mV[VX], tex0.mV[VY] );
+	LLVector4a ry1( pos1[VY], tex1.mV[VX], tex1.mV[VY] );
+	LLVector4a ry2( pos2[VY], tex2.mV[VX], tex2.mV[VY] );
+
+	LLVector4a rz0( pos0[VZ], tex0.mV[VX], tex0.mV[VY] );
+	LLVector4a rz1( pos1[VZ], tex1.mV[VX], tex1.mV[VY] );
+	LLVector4a rz2( pos2[VZ], tex2.mV[VX], tex2.mV[VY] );
+	
+	LLVector4a lhs, rhs;
+
+	LLVector4a r0; 
+	lhs.setSub(rx0, rx1); rhs.setSub(rx0, rx2);
+	r0.setCross3(lhs, rhs);
+		
+	LLVector4a r1;
+	lhs.setSub(ry0, ry1); rhs.setSub(ry0, ry2);
+	r1.setCross3(lhs, rhs);
+
+	LLVector4a r2;
+	lhs.setSub(rz0, rz1); rhs.setSub(rz0, rz2);
+	r2.setCross3(lhs, rhs);
+
+	if( r0[VX] && r1[VX] && r2[VX] )
+	{
+		binormal.set(
+				-r0[VZ] / r0[VX],
+				-r1[VZ] / r1[VX],
+				-r2[VZ] / r2[VX]);
+		// binormal.normVec();
+	}
+	else
+	{
+		binormal.set( 0, 1 , 0 );
+	}
+}
+
 //-----------------------------------------------------------------------------
 // LLPolyMesh::loadOBJ()
 //-----------------------------------------------------------------------------
--- a/indra/llcommon/CMakeLists.txt
+++ b/indra/llcommon/CMakeLists.txt
@@ -116,6 +116,7 @@ set(llcommon_HEADER_FILES
    linden_common.h
    linked_lists.h
    llaccountingcost.h
+    llalignedarray.h
    llagentconstants.h
    llallocator.h
    llallocator_heap_profile.h
--- a/indra/llcommon/llalignedarray.h
+++ b/indra/llcommon/llalignedarray.h
@@ -0,0 +1,139 @@
+/** 
+ * @file llalignedarray.h
+ * @brief A static array which obeys alignment restrictions and mimics std::vector accessors.
+ *
+ * $LicenseInfo:firstyear=2002&license=viewerlgpl$
+ * Second Life Viewer Source Code
+ * Copyright (C) 2010, Linden Research, Inc.
+ * 
+ * This library is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation;
+ * version 2.1 of the License only.
+ * 
+ * This library is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ * 
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with this library; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA
+ * 
+ * Linden Research, Inc., 945 Battery Street, San Francisco, CA  94111  USA
+ * $/LicenseInfo$
+ */
+
+#ifndef LL_LLALIGNEDARRAY_H
+#define LL_LLALIGNEDARRAY_H
+
+#include "llmemory.h"
+
+template <class T, U32 alignment>
+class LLAlignedArray
+{
+public:
+	T* mArray;
+	U32 mElementCount;
+	U32 mCapacity;
+
+	LLAlignedArray();
+	~LLAlignedArray();
+
+	void push_back(const T& elem);
+	U32 size() const { return mElementCount; }
+	void resize(U32 size);
+	T* append(S32 N);
+	T& operator[](int idx);
+	const T& operator[](int idx) const;
+};
+
+template <class T, U32 alignment>
+LLAlignedArray<T, alignment>::LLAlignedArray()
+{
+	llassert(alignment >= 16);
+	mArray = NULL;
+	mElementCount = 0;
+	mCapacity = 0;
+}
+
+template <class T, U32 alignment>
+LLAlignedArray<T, alignment>::~LLAlignedArray()
+{
+	ll_aligned_free(mArray);
+	mArray = NULL;
+	mElementCount = 0;
+	mCapacity = 0;
+}
+
+template <class T, U32 alignment>
+void LLAlignedArray<T, alignment>::push_back(const T& elem)
+{
+	T* old_buf = NULL;
+	if (mCapacity <= mElementCount)
+	{
+		mCapacity++;
+		mCapacity *= 2;
+		T* new_buf = (T*) ll_aligned_malloc(mCapacity*sizeof(T), alignment);
+		if (mArray)
+		{
+			ll_memcpy_nonaliased_aligned_16((char*)new_buf, (char*)mArray, sizeof(T)*mElementCount);
+		}
+		old_buf = mArray;
+		mArray = new_buf;
+	}
+
+	mArray[mElementCount++] = elem;
+
+	//delete old array here to prevent error on a.push_back(a[0])
+	ll_aligned_free(old_buf);
+}
+
+template <class T, U32 alignment>
+void LLAlignedArray<T, alignment>::resize(U32 size)
+{
+	if (mCapacity < size)
+	{
+		mCapacity = size+mCapacity*2;
+		T* new_buf = mCapacity > 0 ? (T*) ll_aligned_malloc(mCapacity*sizeof(T), alignment) : NULL;
+		if (mArray)
+		{
+			ll_memcpy_nonaliased_aligned_16((char*) new_buf, (char*) mArray, sizeof(T)*mElementCount);
+			ll_aligned_free(mArray);
+		}
+
+		/*for (U32 i = mElementCount; i < mCapacity; ++i)
+		{
+			new(new_buf+i) T();
+		}*/
+		mArray = new_buf;
+	}
+
+	mElementCount = size;
+}
+
+
+template <class T, U32 alignment>
+T& LLAlignedArray<T, alignment>::operator[](int idx)
+{
+	llassert(idx < mElementCount);
+	return mArray[idx];
+}
+
+template <class T, U32 alignment>
+const T& LLAlignedArray<T, alignment>::operator[](int idx) const
+{
+	llassert(idx < mElementCount);
+	return mArray[idx];
+}
+
+template <class T, U32 alignment>
+T* LLAlignedArray<T, alignment>::append(S32 N)
+{
+	U32 sz = size();
+	resize(sz+N);
+	return &((*this)[sz]);
+}
+
+#endif
+
--- a/indra/llcommon/llmemory.h
+++ b/indra/llcommon/llmemory.h
@@ -42,27 +42,77 @@ class LLMutex ;
 #define LL_CHECK_MEMORY
 #endif

+LL_COMMON_API void ll_assert_aligned_func(uintptr_t ptr,U32 alignment);
+
+#ifdef SHOW_ASSERT
+#define ll_assert_aligned(ptr,alignment) ll_assert_aligned_func(reinterpret_cast<uintptr_t>(ptr),((U32)alignment))
+#else
+#define ll_assert_aligned(ptr,alignment)
+#endif
+
+#include <xmmintrin.h>
+
+template <typename T> T* LL_NEXT_ALIGNED_ADDRESS(T* address) 
+{ 
+	return reinterpret_cast<T*>(
+		(reinterpret_cast<uintptr_t>(address) + 0xF) & ~0xF);
+}
+
+template <typename T> T* LL_NEXT_ALIGNED_ADDRESS_64(T* address) 
+{ 
+	return reinterpret_cast<T*>(
+		(reinterpret_cast<uintptr_t>(address) + 0x3F) & ~0x3F);
+}
+
+#if LL_LINUX || LL_DARWIN
+
+#define			LL_ALIGN_PREFIX(x)
+#define			LL_ALIGN_POSTFIX(x)		__attribute__((aligned(x)))
+
+#elif LL_WINDOWS
+
+#define			LL_ALIGN_PREFIX(x)		__declspec(align(x))
+#define			LL_ALIGN_POSTFIX(x)
+
+#else
+#error "LL_ALIGN_PREFIX and LL_ALIGN_POSTFIX undefined"
+#endif
+
+#define LL_ALIGN_16(var) LL_ALIGN_PREFIX(16) var LL_ALIGN_POSTFIX(16)
+
 inline void* ll_aligned_malloc( size_t size, int align )
 {
+#if defined(LL_WINDOWS)
+	return _aligned_malloc(size, align);
+#else
 	void* mem = malloc( size + (align - 1) + sizeof(void*) );
 	char* aligned = ((char*)mem) + sizeof(void*);
 	aligned += align - ((uintptr_t)aligned & (align - 1));

 	((void**)aligned)[-1] = mem;
 	return aligned;
+#endif
 }

 inline void ll_aligned_free( void* ptr )
 {
-	free( ((void**)ptr)[-1] );
+#if defined(LL_WINDOWS)
+	_aligned_free(ptr);
+#else
+	if (ptr)
+	{
+		free( ((void**)ptr)[-1] );
+	}
+#endif
 }

+#if !LL_USE_TCMALLOC
 inline void* ll_aligned_malloc_16(size_t size) // returned hunk MUST be freed with ll_aligned_free_16().
 {
-#if (LL_DARWIN || LL_USE_TCMALLOC)
-	return malloc(size); // default osx malloc is 16 byte aligned.
-#elif LL_WINDOWS
+#if defined(LL_WINDOWS)
 	return _aligned_malloc(size, 16);
+#elif defined(LL_DARWIN)
+	return malloc(size); // default osx malloc is 16 byte aligned.
 #else
 	void *rtn;
 	if (LL_LIKELY(0 == posix_memalign(&rtn, 16, size)))
@@ -74,10 +124,10 @@ inline void* ll_aligned_malloc_16(size_t size) // returned hunk MUST be freed wi

 inline void ll_aligned_free_16(void *p)
 {
-#if (LL_DARWIN || LL_USE_TCMALLOC)
-	free(p);
-#elif LL_WINDOWS
+#if defined(LL_WINDOWS)
 	_aligned_free(p);
+#elif defined(LL_DARWIN)
+	return free(p);
 #else
 	free(p); // posix_memalign() is compatible with heap deallocator
 #endif
@@ -85,10 +135,10 @@ inline void ll_aligned_free_16(void *p)

 inline void* ll_aligned_realloc_16(void* ptr, size_t size, size_t old_size) // returned hunk MUST be freed with ll_aligned_free_16().
 {
-#if (LL_DARWIN || LL_USE_TCMALLOC)
-	return realloc(ptr,size); // default osx malloc is 16 byte aligned.
-#elif LL_WINDOWS
+#if defined(LL_WINDOWS)
 	return _aligned_realloc(ptr, size, 16);
+#elif defined(LL_DARWIN)
+	return realloc(ptr,size); // default osx malloc is 16 byte aligned.
 #else
 	//FIXME: memcpy is SLOW
 	void* ret = ll_aligned_malloc_16(size);
@@ -105,11 +155,18 @@ inline void* ll_aligned_realloc_16(void* ptr, size_t size, size_t old_size) // r
 #endif
 }

+#else // USE_TCMALLOC
+// ll_aligned_foo_16 are not needed with tcmalloc
+#define ll_aligned_malloc_16 malloc
+#define ll_aligned_realloc_16(a,b,c) realloc(a,b)
+#define ll_aligned_free_16 free
+#endif // USE_TCMALLOC
+
 inline void* ll_aligned_malloc_32(size_t size) // returned hunk MUST be freed with ll_aligned_free_32().
 {
-#if LL_WINDOWS
+#if defined(LL_WINDOWS)
 	return _aligned_malloc(size, 32);
-#elif LL_DARWIN
+#elif defined(LL_DARWIN)
 	return ll_aligned_malloc( size, 32 );
 #else
 	void *rtn;
@@ -122,15 +179,87 @@ inline void* ll_aligned_malloc_32(size_t size) // returned hunk MUST be freed wi

 inline void ll_aligned_free_32(void *p)
 {
-#if LL_WINDOWS
+#if defined(LL_WINDOWS)
 	_aligned_free(p);
-#elif LL_DARWIN
+#elif defined(LL_DARWIN)
 	ll_aligned_free( p );
 #else
 	free(p); // posix_memalign() is compatible with heap deallocator
 #endif
 }

+
+// Copy words 16-byte blocks from src to dst. Source and destination MUST NOT OVERLAP. 
+// Source and dest must be 16-byte aligned and size must be multiple of 16.
+//
+inline void ll_memcpy_nonaliased_aligned_16(char* __restrict dst, const char* __restrict src, size_t bytes)
+{
+	assert(src != NULL);
+	assert(dst != NULL);
+	assert(bytes > 0);
+	assert((bytes % sizeof(F32))== 0); 
+	ll_assert_aligned(src,16);
+	ll_assert_aligned(dst,16);
+	assert((src < dst) ? ((src + bytes) < dst) : ((dst + bytes) < src));
+	assert(bytes%16==0);
+
+	char* end = dst + bytes;
+
+	if (bytes > 64)
+	{
+
+		// Find start of 64b aligned area within block
+		//
+		void* begin_64 = LL_NEXT_ALIGNED_ADDRESS_64(dst);
+		
+		//at least 64 bytes before the end of the destination, switch to 16 byte copies
+		void* end_64 = end-64;
+	
+		// Prefetch the head of the 64b area now
+		//
+		_mm_prefetch((char*)begin_64, _MM_HINT_NTA);
+		_mm_prefetch((char*)begin_64 + 64, _MM_HINT_NTA);
+		_mm_prefetch((char*)begin_64 + 128, _MM_HINT_NTA);
+		_mm_prefetch((char*)begin_64 + 192, _MM_HINT_NTA);
+	
+		// Copy 16b chunks until we're 64b aligned
+		//
+		while (dst < begin_64)
+		{
+
+			_mm_store_ps((F32*)dst, _mm_load_ps((F32*)src));
+			dst += 16;
+			src += 16;
+		}
+	
+		// Copy 64b chunks up to your tail
+		//
+		// might be good to shmoo the 512b prefetch offset
+		// (characterize performance for various values)
+		//
+		while (dst < end_64)
+		{
+			_mm_prefetch((char*)src + 512, _MM_HINT_NTA);
+			_mm_prefetch((char*)dst + 512, _MM_HINT_NTA);
+			_mm_store_ps((F32*)dst, _mm_load_ps((F32*)src));
+			_mm_store_ps((F32*)(dst + 16), _mm_load_ps((F32*)(src + 16)));
+			_mm_store_ps((F32*)(dst + 32), _mm_load_ps((F32*)(src + 32)));
+			_mm_store_ps((F32*)(dst + 48), _mm_load_ps((F32*)(src + 48)));
+			dst += 64;
+			src += 64;
+		}
+	}
+
+	// Copy remainder 16b tail chunks (or ALL 16b chunks for sub-64b copies)
+	//
+	while (dst < end)
+	{
+		_mm_store_ps((F32*)dst, _mm_load_ps((F32*)src));
+		dst += 16;
+		src += 16;
+	}
+}
+
 #ifndef __DEBUG_PRIVATE_MEM__
 #define __DEBUG_PRIVATE_MEM__  0
 #endif
--- a/indra/llmath/llmatrix3a.inl
+++ b/indra/llmath/llmatrix3a.inl
@@ -60,7 +60,7 @@ inline void LLMatrix3a::setTranspose(const LLMatrix3a& src)
 	const LLQuad srcCol1 = src.mColumns[1];
 	const LLQuad unpacklo = _mm_unpacklo_ps( srcCol0, srcCol1 );
 	mColumns[0] = _mm_movelh_ps( unpacklo, src.mColumns[2] );
-	mColumns[1] = _mm_shuffle_ps( unpacklo, src.mColumns[2], _MM_SHUFFLE(0, 1, 3, 2) );
+	mColumns[1] = _mm_shuffle_ps( _mm_movehl_ps( srcCol0, unpacklo ), src.mColumns[2], _MM_SHUFFLE(0, 1, 1, 0) );
 	mColumns[2] = _mm_shuffle_ps( _mm_unpackhi_ps( srcCol0, srcCol1 ), src.mColumns[2], _MM_SHUFFLE(0, 2, 1, 0) );
 }

--- a/indra/llmath/llsimdmath.h
+++ b/indra/llmath/llsimdmath.h
@@ -39,34 +39,6 @@
 #include <stdint.h>
 #endif

-template <typename T> T* LL_NEXT_ALIGNED_ADDRESS(T* address) 
-{ 
-	return reinterpret_cast<T*>(
-		(reinterpret_cast<uintptr_t>(address) + 0xF) & ~0xF);
-}
-
-template <typename T> T* LL_NEXT_ALIGNED_ADDRESS_64(T* address) 
-{ 
-	return reinterpret_cast<T*>(
-		(reinterpret_cast<uintptr_t>(address) + 0x3F) & ~0x3F);
-}
-
-#if LL_LINUX || LL_DARWIN
-
-#define			LL_ALIGN_PREFIX(x)
-#define			LL_ALIGN_POSTFIX(x)		__attribute__((aligned(x)))
-
-#elif LL_WINDOWS
-
-#define			LL_ALIGN_PREFIX(x)		__declspec(align(x))
-#define			LL_ALIGN_POSTFIX(x)
-
-#else
-#error "LL_ALIGN_PREFIX and LL_ALIGN_POSTFIX undefined"
-#endif
-
-#define LL_ALIGN_16(var) LL_ALIGN_PREFIX(16) var LL_ALIGN_POSTFIX(16)
-
 #include <xmmintrin.h>
 #include <emmintrin.h>

--- a/indra/llmath/llvector4a.cpp
+++ b/indra/llmath/llvector4a.cpp
@@ -41,55 +41,7 @@ extern const LLVector4a LL_V4A_EPSILON = reinterpret_cast<const LLVector4a&> ( F

 /*static */void LLVector4a::memcpyNonAliased16(F32* __restrict dst, const F32* __restrict src, size_t bytes)
 {
-	assert(src != NULL);
-	assert(dst != NULL);
-	assert(bytes > 0);
-	assert((bytes % sizeof(F32))== 0); 
-	ll_assert_aligned(src,16);
-	ll_assert_aligned(dst,16);
-	assert(bytes%16==0);
-
-	F32* end = dst + (bytes / sizeof(F32) );
-
-	if (bytes > 64)
-	{
-		F32* begin_64 = LL_NEXT_ALIGNED_ADDRESS_64(dst);
-		
-		//at least 64 (16*4) bytes before the end of the destination, switch to 16 byte copies
-		F32* end_64 = end-16;
-		
-		_mm_prefetch((char*)begin_64, _MM_HINT_NTA);
-		_mm_prefetch((char*)begin_64 + 64, _MM_HINT_NTA);
-		_mm_prefetch((char*)begin_64 + 128, _MM_HINT_NTA);
-		_mm_prefetch((char*)begin_64 + 192, _MM_HINT_NTA);
-		
-		while (dst < begin_64)
-		{
-			copy4a(dst, src);
-			dst += 4;
-			src += 4;
-		}
-		
-		while (dst < end_64)
-		{
-			_mm_prefetch((char*)src + 512, _MM_HINT_NTA);
-			_mm_prefetch((char*)dst + 512, _MM_HINT_NTA);
-			copy4a(dst, src);
-			copy4a(dst+4, src+4);
-			copy4a(dst+8, src+8);
-			copy4a(dst+12, src+12);
-			
-			dst += 16;
-			src += 16;
-		}
-	}
-
-	while (dst < end)
-	{
-		copy4a(dst, src);
-		dst += 4;
-		src += 4;
-	}
+        ll_memcpy_nonaliased_aligned_16((char*)dst, (char*)src, bytes);
 }

 void LLVector4a::setRotated( const LLRotation& rot, const LLVector4a& vec )
--- a/indra/llmath/llvector4a.h
+++ b/indra/llmath/llvector4a.h
@@ -46,6 +46,7 @@ class LLRotation;
 // of this writing, July 08, 2010) about getting it implemented before you resort to
 // LLVector3/LLVector4. 
 /////////////////////////////////
+class LLVector4a;

 LL_ALIGN_PREFIX(16)
 class LLVector4a
@@ -235,6 +236,11 @@ public:
 	// Note that this does not consider zero length vectors!
 	inline void normalize3fast();

+	// Normalize this vector with respect to the x, y, and z components only. Accurate only to 10-12 bits of precision. W component is destroyed
+	// Same as above except substitutes default vector contents if the vector is non-finite or degenerate due to zero length.
+	//
+	inline void normalize3fast_checked(LLVector4a* d = 0);
+
 	// Return true if this vector is normalized with respect to x,y,z up to tolerance
 	inline LLBool32 isNormalized3( F32 tolerance = 1e-3 ) const;

--- a/indra/llmath/llvector4a.inl
+++ b/indra/llmath/llvector4a.inl
@@ -409,6 +409,26 @@ inline void LLVector4a::normalize3fast()
 	mQ = _mm_mul_ps( mQ, approxRsqrt );
 }

+inline void LLVector4a::normalize3fast_checked(LLVector4a* d)
+{
+	if (!isFinite3())
+	{
+		*this = d ? *d : LLVector4a(0,1,0,1);
+		return;
+	}
+
+	LLVector4a lenSqrd; lenSqrd.setAllDot3( *this, *this );
+
+	if (lenSqrd.getF32ptr()[0] <= FLT_EPSILON)
+	{
+		*this = d ? *d : LLVector4a(0,1,0,1);
+		return;
+	}
+
+	const LLQuad approxRsqrt = _mm_rsqrt_ps(lenSqrd.mQ);
+	mQ = _mm_mul_ps( mQ, approxRsqrt );
+}
+
 // Return true if this vector is normalized with respect to x,y,z up to tolerance
 inline LLBool32 LLVector4a::isNormalized3( F32 tolerance ) const
 {
@@ -460,21 +480,19 @@ inline void LLVector4a::setMax(const LLVector4a& lhs, const LLVector4a& rhs)
 	mQ = _mm_max_ps(lhs.mQ, rhs.mQ);
 }

-// Set this to  (c * lhs) + rhs * ( 1 - c)
+// Set this to  lhs + (rhs-lhs)*c
 inline void LLVector4a::setLerp(const LLVector4a& lhs, const LLVector4a& rhs, F32 c)
 {
-	LLVector4a a = lhs;
-	a.mul(c);
-	
-	LLVector4a b = rhs;
-	b.mul(1.f-c);
-	
-	setAdd(a, b);
+	LLVector4a t;
+	t.setSub(rhs,lhs);
+	t.mul(c);
+	setAdd(lhs, t);
 }

 inline LLBool32 LLVector4a::isFinite3() const
 {
 	static LL_ALIGN_16(const U32 nanOrInfMask[4]) = { 0x7f800000, 0x7f800000, 0x7f800000, 0x7f800000 };
+	ll_assert_aligned(nanOrInfMask,16);
 	const __m128i nanOrInfMaskV = *reinterpret_cast<const __m128i*> (nanOrInfMask);
 	const __m128i maskResult = _mm_and_si128( _mm_castps_si128(mQ), nanOrInfMaskV );
 	const LLVector4Logical equalityCheck = _mm_castsi128_ps(_mm_cmpeq_epi32( maskResult, nanOrInfMaskV ));
--- a/indra/llmath/llvolume.cpp
+++ b/indra/llmath/llvolume.cpp
--- a/indra/llmath/llvolume.h
+++ b/indra/llmath/llvolume.h
@@ -37,7 +37,6 @@ class LLPath;

 template <class T> class LLOctreeNode;

-class LLVector4a;
 class LLVolumeFace;
 class LLVolume;
 class LLVolumeTriangle;
@@ -50,12 +49,15 @@ class LLVolumeTriangle;
 #include "v3math.h"
 #include "v3dmath.h"
 #include "v4math.h"
+#include "llvector4a.h"
+#include "llmatrix4a.h"
 #include "llquaternion.h"
 #include "llstrider.h"
 #include "v4coloru.h"
 #include "llrefcount.h"
 #include "llpointer.h"
 #include "llfile.h"
+#include "llalignedarray.h"

 //============================================================================

@@ -709,16 +711,16 @@ public:
 		LLFaceID  mFaceID;
 	};
 	
-	std::vector<LLVector3> mProfile;	
-	std::vector<LLVector2> mNormals;
+	LLAlignedArray<LLVector4a, 64> mProfile;	
+	//LLAlignedArray<LLVector4a, 64> mNormals;
 	std::vector<Face>      mFaces;
-	std::vector<LLVector3> mEdgeNormals;
-	std::vector<LLVector3> mEdgeCenters;
+
+	//LLAlignedArray<LLVector4a, 64> mEdgeNormals;
+	//LLAlignedArray<LLVector4a, 64> mEdgeCenters;

 	friend std::ostream& operator<<(std::ostream &s, const LLProfile &profile);

 protected:
-	void genNormals(const LLProfileParams& params);
 	static S32 getNumNGonPoints(const LLProfileParams& params, S32 sides, F32 offset=0.0f, F32 bevel = 0.0f, F32 ang_scale = 1.f, S32 split = 0);
 	void genNGon(const LLProfileParams& params, S32 sides, F32 offset=0.0f, F32 bevel = 0.0f, F32 ang_scale = 1.f, S32 split = 0);

@@ -742,13 +744,29 @@ protected:
 class LLPath
 {
 public:
-	struct PathPt
+	class PathPt
 	{
-		LLVector3	 mPos;
-		LLVector2    mScale;
-		LLQuaternion mRot;
+	public:
+		LLMatrix4a   mRot;
+		LLVector4a	 mPos;
+		
+		LLVector4a   mScale;
 		F32			 mTexT;
-		PathPt() { mPos.setVec(0,0,0); mTexT = 0; mScale.setVec(0,0); mRot.loadIdentity(); }
+		F32 pad[3]; //for alignment
+		PathPt() 
+		{ 
+			mPos.clear(); 
+			mTexT = 0; 
+			mScale.clear(); 
+			mRot.setRows(LLVector4a(1,0,0,0),
+						LLVector4a(0,1,0,0),
+						LLVector4a(0,0,1,0));
+
+			//distinguished data in the pad for debugging
+			pad[0] = 3.14159f;
+			pad[1] = -3.14159f;
+			pad[2] = 0.585f;
+		}
 	};

 public:
@@ -780,7 +798,7 @@ public:
 	friend std::ostream& operator<<(std::ostream &s, const LLPath &path);

 public:
-	std::vector<PathPt> mPath;
+	LLAlignedArray<PathPt, 64> mPath;

 protected:
 	BOOL		  mOpen;
@@ -845,12 +863,12 @@ private:
 public:

 	BOOL create(LLVolume* volume, BOOL partial_build = FALSE);
-	void createBinormals();
+	void createTangents();
 	
 	void appendFace(const LLVolumeFace& face, LLMatrix4& transform, LLMatrix4& normal_tranform);

 	void resizeVertices(S32 num_verts);
-	void allocateBinormals(S32 num_verts);
+	void allocateTangents(S32 num_verts);
 	void allocateWeights(S32 num_verts);
 	void resizeIndices(S32 num_indices);
 	void fillFromLegacyData(std::vector<LLVolumeFace::VertexData>& v, std::vector<U16>& idx);
@@ -913,11 +931,12 @@ public:
 	LLVector2   mTexCoordExtents[2]; //minimum and maximum of texture coordinates of the face.

 	S32 mNumVertices;
+	S32 mNumAllocatedVertices;
 	S32 mNumIndices;

 	LLVector4a* mPositions;
 	LLVector4a* mNormals;
-	LLVector4a* mBinormals;
+	LLVector4a* mTangents;
 	LLVector2*  mTexCoords;
 	U16* mIndices;

@@ -931,9 +950,12 @@ public:
 	// format is mWeights[vertex_index].mV[influence] = <joint_index>.<weight>
 	// mWeights.size() should be empty or match mVertices.size()  
 	LLVector4a* mWeights;
- 
+
 	LLOctreeNode<LLVolumeTriangle>* mOctree;

+	//whether or not face has been cache optimized
+	BOOL mOptimized;
+
 private:
 	BOOL createUnCutCubeCap(LLVolume* volume, BOOL partial_build = FALSE);
 	BOOL createCap(LLVolume* volume, BOOL partial_build = FALSE);
@@ -945,15 +967,10 @@ class LLVolume : public LLRefCount
 	friend class LLVolumeLODGroup;

 protected:
-	LLVolume(const LLVolume&);  // Don't implement
 	~LLVolume(); // use unref

 public:
-	struct Point
-	{
-		LLVector3 mPos;
-	};
-
+		
 	struct FaceParams
 	{
 		LLFaceID mFaceID;
@@ -976,13 +993,13 @@ public:
 	const LLProfile& getProfile() const						{ return *mProfilep; }
 	LLPath& getPath() const									{ return *mPathp; }
 	void resizePath(S32 length);
-	const std::vector<Point>& getMesh() const				{ return mMesh; }
-	const LLVector3& getMeshPt(const U32 i) const			{ return mMesh[i].mPos; }
+	const LLAlignedArray<LLVector4a,64>&	getMesh() const				{ return mMesh; }
+	const LLVector4a& getMeshPt(const U32 i) const			{ return mMesh[i]; }

 	void setDirty() { mPathp->setDirty(); mProfilep->setDirty(); }

 	void regen();
-	void genBinormals(S32 face);
+	void genTangents(S32 face);

 	BOOL isConvex() const;
 	BOOL isCap(S32 face);
@@ -992,10 +1009,7 @@ public:
 	S32 getSculptLevel() const                              { return mSculptLevel; }
 	void setSculptLevel(S32 level)							{ mSculptLevel = level; }

-	S32 *getTriangleIndices(U32 &num_indices) const;
-
-	// returns number of triangle indeces required for path/profile mesh
-	S32 getNumTriangleIndices() const;
+	
 	static void getLoDTriangleCounts(const LLVolumeParams& params, S32* counts);

 	S32 getNumTriangles(S32* vcount = NULL) const;
@@ -1010,21 +1024,14 @@ public:
 	//get the face index of the face that intersects with the given line segment at the point 
 	//closest to start.  Moves end to the point of intersection.  Returns -1 if no intersection.
 	//Line segment must be in volume space.
-	S32 lineSegmentIntersect(const LLVector3& start, const LLVector3& end,
+	S32 lineSegmentIntersect(const LLVector4a& start, const LLVector4a& end,
 							 S32 face = -1,                          // which face to check, -1 = ALL_SIDES
-							 LLVector3* intersection = NULL,         // return the intersection point
+							 LLVector4a* intersection = NULL,         // return the intersection point
 							 LLVector2* tex_coord = NULL,            // return the texture coordinates of the intersection point
-							 LLVector3* normal = NULL,               // return the surface normal at the intersection point
-							 LLVector3* bi_normal = NULL             // return the surface bi-normal at the intersection point
+							 LLVector4a* normal = NULL,               // return the surface normal at the intersection point
+							 LLVector4a* tangent = NULL             // return the surface tangent at the intersection point
 		);

-	S32 lineSegmentIntersect(const LLVector4a& start, const LLVector4a& end, 
-								   S32 face = 1,
-								   LLVector3* intersection = NULL,
-								   LLVector2* tex_coord = NULL,
-								   LLVector3* normal = NULL,
-								   LLVector3* bi_normal = NULL);
-	
 	LLFaceID generateFaceMask();

 	BOOL isFaceMaskValid(LLFaceID face_mask);
@@ -1068,7 +1075,8 @@ public:
 	LLVolumeParams mParams;
 	LLPath *mPathp;
 	LLProfile *mProfilep;
-	std::vector<Point> mMesh;
+	LLAlignedArray<LLVector4a,64> mMesh;
+	
 	
 	BOOL mGenerateSingleFace;
 	typedef std::vector<LLVolumeFace> face_list_t;
@@ -1083,21 +1091,12 @@ public:

 std::ostream& operator<<(std::ostream &s, const LLVolumeParams &volume_params);

-void calc_binormal_from_triangle(
-		LLVector4a& binormal,
-		const LLVector4a& pos0,
-		const LLVector2& tex0,
-		const LLVector4a& pos1,
-		const LLVector2& tex1,
-		const LLVector4a& pos2,
-		const LLVector2& tex2);
-
 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);
+//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);

 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);
--- a/indra/llmath/llvolumeoctree.cpp
+++ b/indra/llmath/llvolumeoctree.cpp
@@ -94,14 +94,14 @@ void LLVolumeOctreeListener::handleChildAddition(const LLOctreeNode<LLVolumeTria

 LLOctreeTriangleRayIntersect::LLOctreeTriangleRayIntersect(const LLVector4a& start, const LLVector4a& dir, 
 							   const LLVolumeFace* face, F32* closest_t,
-							   LLVector3* intersection,LLVector2* tex_coord, LLVector3* normal, LLVector3* bi_normal)
+							   LLVector4a* intersection,LLVector2* tex_coord, LLVector4a* normal, LLVector4a* tangent)
   : mFace(face),
     mStart(start),
 	 mDir(dir),
 	 mIntersection(intersection),
 	 mTexCoord(tex_coord),
 	 mNormal(normal),
-	 mBinormal(bi_normal),
+	 mTangent(tangent),
 	 mClosestT(closest_t),
 	 mHitFace(false)
 {
@@ -112,13 +112,7 @@ void LLOctreeTriangleRayIntersect::traverse(const LLOctreeNode<LLVolumeTriangle>
 {
 	LLVolumeOctreeListener* vl = (LLVolumeOctreeListener*) node->getListener(0);

-	/*const F32* start = mStart.getF32();
-	const F32* end = mEnd.getF32();
-	const F32* center = vl->mBounds[0].getF32();
-	const F32* size = vl->mBounds[1].getF32();*/
-
-	//if (LLLineSegmentBoxIntersect(mStart, mEnd, vl->mBounds[0], vl->mBounds[1]))
-	if (LLLineSegmentBoxIntersect(mStart.getF32ptr(), mEnd.getF32ptr(), vl->mBounds[0].getF32ptr(), vl->mBounds[1].getF32ptr()))
+	if (LLLineSegmentBoxIntersect(mStart, mEnd, vl->mBounds[0], vl->mBounds[1]))
 	{
 		node->accept(this);
 		for (U32 i = 0; i < node->getChildCount(); ++i)
@@ -152,34 +146,60 @@ void LLOctreeTriangleRayIntersect::visit(const LLOctreeNode<LLVolumeTriangle>* n
 					LLVector4a intersect = mDir;
 					intersect.mul(*mClosestT);
 					intersect.add(mStart);
-					mIntersection->set(intersect.getF32ptr());
+					*mIntersection = intersect;
 				}

+				U32 idx0 = tri->mIndex[0];
+				U32 idx1 = tri->mIndex[1];
+				U32 idx2 = tri->mIndex[2];

 				if (mTexCoord != NULL)
 				{
 					LLVector2* tc = (LLVector2*) mFace->mTexCoords;
-					*mTexCoord = ((1.f - a - b)  * tc[tri->mIndex[0]] +
-						a              * tc[tri->mIndex[1]] +
-						b              * tc[tri->mIndex[2]]);
+					*mTexCoord = ((1.f - a - b)  * tc[idx0] +
+						a              * tc[idx1] +
+						b              * tc[idx2]);

 				}

 				if (mNormal != NULL)
 				{
-					LLVector4* norm = (LLVector4*) mFace->mNormals;
+					LLVector4a* norm = mFace->mNormals;
+								
+					LLVector4a n1,n2,n3;
+					n1 = norm[idx0];
+					n1.mul(1.f-a-b);
+								
+					n2 = norm[idx1];
+					n2.mul(a);
+								
+					n3 = norm[idx2];
+					n3.mul(b);

-					*mNormal    = ((1.f - a - b)  * LLVector3(norm[tri->mIndex[0]]) + 
-						a              * LLVector3(norm[tri->mIndex[1]]) +
-						b              * LLVector3(norm[tri->mIndex[2]]));
+					n1.add(n2);
+					n1.add(n3);
+								
+					*mNormal		= n1; 
 				}

-				if (mBinormal != NULL)
+				if (mTangent != NULL)
 				{
-					LLVector4* binormal = (LLVector4*) mFace->mBinormals;
-					*mBinormal = ((1.f - a - b)  * LLVector3(binormal[tri->mIndex[0]]) + 
-							a              * LLVector3(binormal[tri->mIndex[1]]) +
-							b              * LLVector3(binormal[tri->mIndex[2]]));
+					LLVector4a* tangents = mFace->mTangents;
+								
+					LLVector4a t1,t2,t3;
+					t1 = tangents[idx0];
+					t1.mul(1.f-a-b);
+								
+					t2 = tangents[idx1];
+					t2.mul(a);
+								
+					t3 = tangents[idx2];
+					t3.mul(b);
+
+					t1.add(t2);
+					t1.add(t3);
+								
+					*mTangent = t1; 
 				}
 			}
 		}
--- a/indra/llmath/llvolumeoctree.h
+++ b/indra/llmath/llvolumeoctree.h
@@ -134,16 +134,16 @@ public:
 	LLVector4a mStart;
 	LLVector4a mDir;
 	LLVector4a mEnd;
-	LLVector3* mIntersection;
+	LLVector4a* mIntersection;
 	LLVector2* mTexCoord;
-	LLVector3* mNormal;
-	LLVector3* mBinormal;
+	LLVector4a* mNormal;
+	LLVector4a* mTangent;
 	F32* mClosestT;
 	bool mHitFace;

 	LLOctreeTriangleRayIntersect(const LLVector4a& start, const LLVector4a& dir, 
 								   const LLVolumeFace* face, F32* closest_t,
-								   LLVector3* intersection,LLVector2* tex_coord, LLVector3* normal, LLVector3* bi_normal);
+								   LLVector4a* intersection,LLVector2* tex_coord, LLVector4a* normal, LLVector4a* tangent);

 	void traverse(const LLOctreeNode<LLVolumeTriangle>* node);

--- a/indra/llprimitive/llmodel.cpp
+++ b/indra/llprimitive/llmodel.cpp
@@ -263,13 +263,13 @@ LLModel::EModelStatus load_face_from_dom_triangles(std::vector<LLVolumeFace>& fa
 			LLVolumeFace& new_face = *face_list.rbegin();
 			if (!norm_source)
 			{
-				ll_aligned_free_16(new_face.mNormals);
+				//l_aligned_free_16(new_face.mNormals);
 				new_face.mNormals = NULL;
 			}

 			if (!tc_source)
 			{
-				ll_aligned_free_16(new_face.mTexCoords);
+				//ll_aligned_free_16(new_face.mTexCoords);
 				new_face.mTexCoords = NULL;
 			}

@@ -294,13 +294,13 @@ LLModel::EModelStatus load_face_from_dom_triangles(std::vector<LLVolumeFace>& fa
 		LLVolumeFace& new_face = *face_list.rbegin();
 		if (!norm_source)
 		{
-			ll_aligned_free_16(new_face.mNormals);
+			//ll_aligned_free_16(new_face.mNormals);
 			new_face.mNormals = NULL;
 		}

 		if (!tc_source)
 		{
-			ll_aligned_free_16(new_face.mTexCoords);
+			//ll_aligned_free_16(new_face.mTexCoords);
 			new_face.mTexCoords = NULL;
 		}
 	}
@@ -482,13 +482,13 @@ LLModel::EModelStatus load_face_from_dom_polylist(std::vector<LLVolumeFace>& fac
 				LLVolumeFace& new_face = *face_list.rbegin();
 				if (!norm_source)
 				{
-					ll_aligned_free_16(new_face.mNormals);
+					//ll_aligned_free_16(new_face.mNormals);
 					new_face.mNormals = NULL;
 				}

 				if (!tc_source)
 				{
-					ll_aligned_free_16(new_face.mTexCoords);
+					//ll_aligned_free_16(new_face.mTexCoords);
 					new_face.mTexCoords = NULL;
 				}

@@ -516,13 +516,13 @@ LLModel::EModelStatus load_face_from_dom_polylist(std::vector<LLVolumeFace>& fac
 		LLVolumeFace& new_face = *face_list.rbegin();
 		if (!norm_source)
 		{
-			ll_aligned_free_16(new_face.mNormals);
+			//ll_aligned_free_16(new_face.mNormals);
 			new_face.mNormals = NULL;
 		}

 		if (!tc_source)
 		{
-			ll_aligned_free_16(new_face.mTexCoords);
+			//ll_aligned_free_16(new_face.mTexCoords);
 			new_face.mTexCoords = NULL;
 		}
 	}
@@ -716,13 +716,13 @@ LLModel::EModelStatus load_face_from_dom_polygons(std::vector<LLVolumeFace>& fac
 		LLVolumeFace& new_face = *face_list.rbegin();
 		if (!n)
 		{
-			ll_aligned_free_16(new_face.mNormals);
+			//ll_aligned_free_16(new_face.mNormals);
 			new_face.mNormals = NULL;
 		}

 		if (!t)
 		{
-			ll_aligned_free_16(new_face.mTexCoords);
+			//ll_aligned_free_16(new_face.mTexCoords);
 			new_face.mTexCoords = NULL;
 		}
 	}
@@ -996,6 +996,43 @@ void LLModel::getNormalizedScaleTranslation(LLVector3& scale_out, LLVector3& tra
 	translation_out = mNormalizedTranslation;
 }

+LLVector3 LLModel::getTransformedCenter(const LLMatrix4& mat)
+{
+	LLVector3 ret;
+
+	if (!mVolumeFaces.empty())
+	{
+		LLMatrix4a m;
+		m.loadu(mat);
+
+		LLVector4a minv,maxv;
+
+		LLVector4a t;
+		m.affineTransform(mVolumeFaces[0].mPositions[0], t);
+		minv = maxv = t;
+
+		for (S32 i = 0; i < (S32)mVolumeFaces.size(); ++i)
+		{
+			LLVolumeFace& face = mVolumeFaces[i];
+
+			for (U32 j = 0; j < (U32)face.mNumVertices; ++j)
+			{
+				m.affineTransform(face.mPositions[j],t);
+				update_min_max(minv, maxv, t);
+			}
+		}
+
+		minv.add(maxv);
+		minv.mul(0.5f);
+
+		ret.set(minv.getF32ptr());
+	}
+
+	return ret;
+}
+
+
+
 void LLModel::setNumVolumeFaces(S32 count)
 {
 	mVolumeFaces.resize(count);
@@ -1022,7 +1059,7 @@ void LLModel::setVolumeFaceData(
 	}
 	else
 	{
-		ll_aligned_free_16(face.mNormals);
+		//ll_aligned_free_16(face.mNormals);
 		face.mNormals = NULL;
 	}

@@ -1033,7 +1070,7 @@ void LLModel::setVolumeFaceData(
 	}
 	else
 	{
-		ll_aligned_free_16(face.mTexCoords);
+		//ll_aligned_free_16(face.mTexCoords);
 		face.mTexCoords = NULL;
 	}

@@ -1232,7 +1269,7 @@ void LLModel::generateNormals(F32 angle_cutoff)
 		}
 		else
 		{
-			ll_aligned_free_16(new_face.mTexCoords);
+			//ll_aligned_free_16(new_face.mTexCoords);
 			new_face.mTexCoords = NULL;
 		}

--- a/indra/llprimitive/llmodel.h
+++ b/indra/llprimitive/llmodel.h
@@ -173,7 +173,8 @@ public:
 	void optimizeVolumeFaces();
 	void offsetMesh( const LLVector3& pivotPoint );
 	void getNormalizedScaleTranslation(LLVector3& scale_out, LLVector3& translation_out);
-	
+	LLVector3 getTransformedCenter(const LLMatrix4& mat);
+
 	//reorder face list based on mMaterialList in this and reference so 
 	//order matches that of reference (material ordering touchup)
 	bool matchMaterialOrder(LLModel* ref, int& refFaceCnt, int& modelFaceCnt );
--- a/indra/llrender/llshadermgr.cpp
+++ b/indra/llrender/llshadermgr.cpp
@@ -1035,7 +1035,7 @@ void LLShaderMgr::initAttribsAndUniforms()
 	mReservedAttribs.push_back("texcoord3");
 	mReservedAttribs.push_back("diffuse_color");
 	mReservedAttribs.push_back("emissive");
-	mReservedAttribs.push_back("binormal");
+	mReservedAttribs.push_back("tangent");
 	mReservedAttribs.push_back("weight");
 	mReservedAttribs.push_back("weight4");
 	mReservedAttribs.push_back("clothing");
--- a/indra/llrender/llvertexbuffer.cpp
+++ b/indra/llrender/llvertexbuffer.cpp
@@ -342,13 +342,32 @@ S32 LLVertexBuffer::sTypeSize[LLVertexBuffer::TYPE_MAX] =
 	sizeof(LLVector2), // TYPE_TEXCOORD3,
 	sizeof(LLColor4U), // TYPE_COLOR,
 	sizeof(LLColor4U), // TYPE_EMISSIVE, only alpha is used currently
-	sizeof(LLVector4), // TYPE_BINORMAL,
+	sizeof(LLVector4), // TYPE_TANGENT,
 	sizeof(F32),	   // TYPE_WEIGHT,
 	sizeof(LLVector4), // TYPE_WEIGHT4,
 	sizeof(LLVector4), // TYPE_CLOTHWEIGHT,
 	sizeof(LLVector4), // TYPE_TEXTURE_INDEX (actually exists as position.w), no extra data, but stride is 16 bytes
 };

+static std::string vb_type_name[] =
+{
+	"TYPE_VERTEX",
+	"TYPE_NORMAL",
+	"TYPE_TEXCOORD0",
+	"TYPE_TEXCOORD1",
+	"TYPE_TEXCOORD2",
+	"TYPE_TEXCOORD3",
+	"TYPE_COLOR",
+	"TYPE_EMISSIVE",
+	"TYPE_TANGENT",
+	"TYPE_WEIGHT",
+	"TYPE_WEIGHT4",
+	"TYPE_CLOTHWEIGHT",
+	"TYPE_TEXTURE_INDEX",
+	"TYPE_MAX",
+	"TYPE_INDEX",	
+};
+
 U32 LLVertexBuffer::sGLMode[LLRender::NUM_MODES] = 
 {
 	GL_TRIANGLES,
@@ -523,16 +542,16 @@ void LLVertexBuffer::setupClientArrays(U32 data_mask)
 				}
 			}

-			if (sLastMask & MAP_BINORMAL)
+			if (sLastMask & MAP_TANGENT)
 			{
-				if (!(data_mask & MAP_BINORMAL))
+				if (!(data_mask & MAP_TANGENT))
 				{
 					glClientActiveTextureARB(GL_TEXTURE2_ARB);
 					glDisableClientState(GL_TEXTURE_COORD_ARRAY);
 					glClientActiveTextureARB(GL_TEXTURE0_ARB);
 				}
 			}
-			else if (data_mask & MAP_BINORMAL)
+			else if (data_mask & MAP_TANGENT)
 			{
 				glClientActiveTextureARB(GL_TEXTURE2_ARB);
 				glEnableClientState(GL_TEXTURE_COORD_ARRAY);
@@ -1320,7 +1339,7 @@ void LLVertexBuffer::setupVertexArray()
 		2, //TYPE_TEXCOORD3,
 		4, //TYPE_COLOR,
 		4, //TYPE_EMISSIVE,
-		3, //TYPE_BINORMAL,
+		4, //TYPE_TANGENT,
 		1, //TYPE_WEIGHT,
 		4, //TYPE_WEIGHT4,
 		4, //TYPE_CLOTHWEIGHT,
@@ -1337,7 +1356,7 @@ void LLVertexBuffer::setupVertexArray()
 		GL_FLOAT, //TYPE_TEXCOORD3,
 		GL_UNSIGNED_BYTE, //TYPE_COLOR,
 		GL_UNSIGNED_BYTE, //TYPE_EMISSIVE,
-		GL_FLOAT,   //TYPE_BINORMAL,
+		GL_FLOAT,   //TYPE_TANGENT,
 		GL_FLOAT, //TYPE_WEIGHT,
 		GL_FLOAT, //TYPE_WEIGHT4,
 		GL_FLOAT, //TYPE_CLOTHWEIGHT,
@@ -1354,7 +1373,7 @@ void LLVertexBuffer::setupVertexArray()
 		false, //TYPE_TEXCOORD3,
 		false, //TYPE_COLOR,
 		false, //TYPE_EMISSIVE,
-		false, //TYPE_BINORMAL,
+		false, //TYPE_TANGENT,
 		false, //TYPE_WEIGHT,
 		false, //TYPE_WEIGHT4,
 		false, //TYPE_CLOTHWEIGHT,
@@ -1371,7 +1390,7 @@ void LLVertexBuffer::setupVertexArray()
 		GL_FALSE, //TYPE_TEXCOORD3,
 		GL_TRUE, //TYPE_COLOR,
 		GL_TRUE, //TYPE_EMISSIVE,
-		GL_FALSE,   //TYPE_BINORMAL,
+		GL_FALSE,   //TYPE_TANGENT,
 		GL_FALSE, //TYPE_WEIGHT,
 		GL_FALSE, //TYPE_WEIGHT4,
 		GL_FALSE, //TYPE_CLOTHWEIGHT,
@@ -2033,9 +2052,13 @@ bool LLVertexBuffer::getNormalStrider(LLStrider<LLVector3>& strider, S32 index,
 {
 	return VertexBufferStrider<LLVector3,TYPE_NORMAL>::get(*this, strider, index, count, map_range);
 }
-bool LLVertexBuffer::getBinormalStrider(LLStrider<LLVector3>& strider, S32 index, S32 count, bool map_range)
+bool LLVertexBuffer::getTangentStrider(LLStrider<LLVector3>& strider, S32 index, S32 count, bool map_range)
 {
-	return VertexBufferStrider<LLVector3,TYPE_BINORMAL>::get(*this, strider, index, count, map_range);
+	return VertexBufferStrider<LLVector3,TYPE_TANGENT>::get(*this, strider, index, count, map_range);
+}
+bool LLVertexBuffer::getTangentStrider(LLStrider<LLVector4a>& strider, S32 index, S32 count, bool map_range)
+{
+	return VertexBufferStrider<LLVector4a,TYPE_TANGENT>::get(*this, strider, index, count, map_range);
 }
 bool LLVertexBuffer::getColorStrider(LLStrider<LLColor4U>& strider, S32 index, S32 count, bool map_range)
 {
@@ -2308,6 +2331,14 @@ void LLVertexBuffer::setupVertexBuffer(U32 data_mask)

 	if (gDebugGL && ((data_mask & mTypeMask) != data_mask))
 	{
+		for (U32 i = 0; i < LLVertexBuffer::TYPE_MAX; ++i)
+		{
+			U32 mask = 1 << i;
+			if (mask & data_mask && !(mask & mTypeMask))
+			{ //bit set in data_mask, but not set in mTypeMask
+				llwarns << "Missing required component " << vb_type_name[i] << llendl;
+			}
+		}
 		llerrs << "LLVertexBuffer::setupVertexBuffer missing required components for supplied data mask." << llendl;
 	}

@@ -2337,11 +2368,11 @@ void LLVertexBuffer::setupVertexBuffer(U32 data_mask)
 			void* ptr = (void*)(base + mOffsets[TYPE_TEXCOORD1]);
 			glVertexAttribPointerARB(loc,2,GL_FLOAT, GL_FALSE, LLVertexBuffer::sTypeSize[TYPE_TEXCOORD1], ptr);
 		}
-		if (data_mask & MAP_BINORMAL)
+		if (data_mask & MAP_TANGENT)
 		{
-			S32 loc = TYPE_BINORMAL;
-			void* ptr = (void*)(base + mOffsets[TYPE_BINORMAL]);
-			glVertexAttribPointerARB(loc, 3,GL_FLOAT, GL_FALSE, LLVertexBuffer::sTypeSize[TYPE_BINORMAL], ptr);
+			S32 loc = TYPE_TANGENT;
+			void* ptr = (void*)(base + mOffsets[TYPE_TANGENT]);
+			glVertexAttribPointerARB(loc, 4,GL_FLOAT, GL_FALSE, LLVertexBuffer::sTypeSize[TYPE_TANGENT], ptr);
 		}
 		if (data_mask & MAP_TEXCOORD0)
 		{
@@ -2419,10 +2450,10 @@ void LLVertexBuffer::setupVertexBuffer(U32 data_mask)
 			glTexCoordPointer(2,GL_FLOAT, LLVertexBuffer::sTypeSize[TYPE_TEXCOORD1], (void*)(base + mOffsets[TYPE_TEXCOORD1]));
 			glClientActiveTextureARB(GL_TEXTURE0_ARB);
 		}
-		if (data_mask & MAP_BINORMAL)
+		if (data_mask & MAP_TANGENT)
 		{
 			glClientActiveTextureARB(GL_TEXTURE2_ARB);
-			glTexCoordPointer(3,GL_FLOAT, LLVertexBuffer::sTypeSize[TYPE_BINORMAL], (void*)(base + mOffsets[TYPE_BINORMAL]));
+			glTexCoordPointer(4,GL_FLOAT, LLVertexBuffer::sTypeSize[TYPE_TANGENT], (void*)(base + mOffsets[TYPE_TANGENT]));
 			glClientActiveTextureARB(GL_TEXTURE0_ARB);
 		}
 		if (data_mask & MAP_TEXCOORD0)
--- a/indra/llrender/llvertexbuffer.h
+++ b/indra/llrender/llvertexbuffer.h
@@ -173,7 +173,7 @@ public:
 		TYPE_TEXCOORD3,
 		TYPE_COLOR,
 		TYPE_EMISSIVE,
-		TYPE_BINORMAL,
+		TYPE_TANGENT,
 		TYPE_WEIGHT,
 		TYPE_WEIGHT4,
 		TYPE_CLOTHWEIGHT,
@@ -191,7 +191,7 @@ public:
 		MAP_COLOR = (1<<TYPE_COLOR),
 		MAP_EMISSIVE = (1<<TYPE_EMISSIVE),
 		// These use VertexAttribPointer and should possibly be made generic
-		MAP_BINORMAL = (1<<TYPE_BINORMAL),
+		MAP_TANGENT = (1<<TYPE_TANGENT),
 		MAP_WEIGHT = (1<<TYPE_WEIGHT),
 		MAP_WEIGHT4 = (1<<TYPE_WEIGHT4),
 		MAP_CLOTHWEIGHT = (1<<TYPE_CLOTHWEIGHT),
@@ -251,7 +251,8 @@ public:
 	bool getTexCoord0Strider(LLStrider<LLVector2>& strider, S32 index=0, S32 count = -1, bool map_range = false);
 	bool getTexCoord1Strider(LLStrider<LLVector2>& strider, S32 index=0, S32 count = -1, bool map_range = false);
 	bool getNormalStrider(LLStrider<LLVector3>& strider, S32 index=0, S32 count = -1, bool map_range = false);
-	bool getBinormalStrider(LLStrider<LLVector3>& strider, S32 index=0, S32 count = -1, bool map_range = false);
+	bool getTangentStrider(LLStrider<LLVector3>& strider, S32 index=0, S32 count = -1, bool map_range = false);
+	bool getTangentStrider(LLStrider<LLVector4a>& strider, S32 index=0, S32 count = -1, bool map_range = false);
 	bool getColorStrider(LLStrider<LLColor4U>& strider, S32 index=0, S32 count = -1, bool map_range = false);
 	bool getEmissiveStrider(LLStrider<LLColor4U>& strider, S32 index=0, S32 count = -1, bool map_range = false);
 	bool getWeightStrider(LLStrider<F32>& strider, S32 index=0, S32 count = -1, bool map_range = false);
--- a/indra/newview/app_settings/shaders/class1/deferred/bumpSkinnedV.glsl
+++ b/indra/newview/app_settings/shaders/class1/deferred/bumpSkinnedV.glsl
@@ -30,7 +30,7 @@ ATTRIBUTE vec3 position;
 ATTRIBUTE vec4 diffuse_color;
 ATTRIBUTE vec3 normal;
 ATTRIBUTE vec2 texcoord0;
-ATTRIBUTE vec3 binormal;
+ATTRIBUTE vec4 tangent;

 VARYING vec3 vary_mat0;
 VARYING vec3 vary_mat1;
@@ -52,8 +52,8 @@ void main()
 	
 	
 	vec3 n = normalize((mat * vec4(normal.xyz+position.xyz, 1.0)).xyz-pos.xyz);
-	vec3 b = normalize((mat * vec4(binormal.xyz+position.xyz, 1.0)).xyz-pos.xyz);
-	vec3 t = cross(b, n);
+	vec3 t = normalize((mat * vec4(tangent.xyz+position.xyz, 1.0)).xyz-pos.xyz);
+	vec3 b = cross(n, t) * tangent.w;
 	
 	vary_mat0 = vec3(t.x, b.x, n.x);
 	vary_mat1 = vec3(t.y, b.y, n.y);
--- a/indra/newview/app_settings/shaders/class1/deferred/bumpV.glsl
+++ b/indra/newview/app_settings/shaders/class1/deferred/bumpV.glsl
@@ -31,7 +31,7 @@ ATTRIBUTE vec3 position;
 ATTRIBUTE vec4 diffuse_color;
 ATTRIBUTE vec3 normal;
 ATTRIBUTE vec2 texcoord0;
-ATTRIBUTE vec3 binormal;
+ATTRIBUTE vec4 tangent;

 VARYING vec3 vary_mat0;
 VARYING vec3 vary_mat1;
@@ -46,8 +46,8 @@ void main()
 	vary_texcoord0 = (texture_matrix0 * vec4(texcoord0,0,1)).xy;
 	
 	vec3 n = normalize(normal_matrix * normal);
-	vec3 b = normalize(normal_matrix * binormal);
-	vec3 t = cross(b, n);
+	vec3 t = normalize(normal_matrix * tangent.xyz);
+	vec3 b = cross(n, t) * tangent.w;
 	
 	vary_mat0 = vec3(t.x, b.x, n.x);
 	vary_mat1 = vec3(t.y, b.y, n.y);
--- a/indra/newview/app_settings/shaders/class1/deferred/fxaaF.glsl
+++ b/indra/newview/app_settings/shaders/class1/deferred/fxaaF.glsl
@@ -2093,7 +2093,6 @@ uniform sampler2D diffuseMap;
 uniform vec2 rcp_screen_res;
 uniform vec4 rcp_frame_opt;
 uniform vec4 rcp_frame_opt2;
-uniform vec2 screen_res;
 VARYING vec2 vary_fragcoord;
 VARYING vec2 vary_tc;

--- a/indra/newview/app_settings/shaders/class1/transform/binormalV.glsl
+++ b/indra/newview/app_settings/shaders/class1/transform/binormalV.glsl
@@ -25,12 +25,12 @@

 uniform mat3 normal_matrix;

-ATTRIBUTE vec3 binormal;
+ATTRIBUTE vec4 tangent;

-VARYING vec4 binormal_out;
+VARYING vec4 tangent_out;

 void main()
 {
-	binormal_out = vec4(normal_matrix * binormal, 0.0);
+	tangent_out = vec4(normal_matrix * tangent.xyz, tangent.w);
 }

--- a/indra/newview/lldrawpoolavatar.h
+++ b/indra/newview/lldrawpoolavatar.h
@@ -190,7 +190,7 @@ public:
 		RIGGED_DEFERRED_BUMP_MASK = LLVertexBuffer::MAP_VERTEX | 
 							 LLVertexBuffer::MAP_NORMAL | 
 							 LLVertexBuffer::MAP_TEXCOORD0 |
-							 LLVertexBuffer::MAP_BINORMAL |
+							 LLVertexBuffer::MAP_TANGENT |
 							 LLVertexBuffer::MAP_COLOR |
 							 LLVertexBuffer::MAP_WEIGHT4,
 		RIGGED_DEFERRED_SIMPLE_MASK = LLVertexBuffer::MAP_VERTEX | 
--- a/indra/newview/lldrawpoolbump.cpp
+++ b/indra/newview/lldrawpoolbump.cpp
@@ -851,7 +851,7 @@ void LLDrawPoolBump::renderDeferred(S32 pass)
 	LLCullResult::drawinfo_iterator begin = gPipeline.beginRenderMap(type);
 	LLCullResult::drawinfo_iterator end = gPipeline.endRenderMap(type);

-	U32 mask = LLVertexBuffer::MAP_VERTEX | LLVertexBuffer::MAP_TEXCOORD0 | LLVertexBuffer::MAP_BINORMAL | LLVertexBuffer::MAP_NORMAL | LLVertexBuffer::MAP_COLOR;
+	U32 mask = LLVertexBuffer::MAP_VERTEX | LLVertexBuffer::MAP_TEXCOORD0 | LLVertexBuffer::MAP_TANGENT | LLVertexBuffer::MAP_NORMAL | LLVertexBuffer::MAP_COLOR;
 	
 	for (LLCullResult::drawinfo_iterator i = begin; i != end; ++i)	
 	{
--- a/indra/newview/llface.cpp
+++ b/indra/newview/llface.cpp
@@ -511,7 +511,7 @@ void LLFace::renderSelected(LLViewerTexture *imagep, const LLColor4& color)
 		{
 			LLGLEnable poly_offset(GL_POLYGON_OFFSET_FILL);
 			glPolygonOffset(-1.f,-1.f);
-			mVertexBuffer->setBuffer(LLVertexBuffer::MAP_VERTEX | LLVertexBuffer::MAP_TEXCOORD0);
+			mVertexBuffer->setBuffer(mVertexBuffer->getTypeMask());
 			mVertexBuffer->draw(LLRender::TRIANGLES, mIndicesCount, mIndicesIndex);
 		}

@@ -789,6 +789,12 @@ BOOL LLFace::genVolumeBBoxes(const LLVolume &volume, S32 f,
 			size.mul(scale);
 		}

+		// Catch potential badness from normalization before it happens
+		//
+		llassert(mat_normal.mMatrix[0].isFinite3() && (mat_normal.mMatrix[0].dot3(mat_normal.mMatrix[0]).getF32() > F_APPROXIMATELY_ZERO));
+		llassert(mat_normal.mMatrix[1].isFinite3() && (mat_normal.mMatrix[1].dot3(mat_normal.mMatrix[1]).getF32() > F_APPROXIMATELY_ZERO));
+		llassert(mat_normal.mMatrix[2].isFinite3() && (mat_normal.mMatrix[2].dot3(mat_normal.mMatrix[2]).getF32() > F_APPROXIMATELY_ZERO));
+
 		mat_normal.mMatrix[0].normalize3fast();
 		mat_normal.mMatrix[1].normalize3fast();
 		mat_normal.mMatrix[2].normalize3fast();
@@ -874,7 +880,7 @@ BOOL LLFace::genVolumeBBoxes(const LLVolume &volume, S32 f,
 // integrated with getGeometryVolume() for its texture coordinate
 // generation - but i'll leave that to someone more familiar
 // with the implications.
-LLVector2 LLFace::surfaceToTexture(LLVector2 surface_coord, LLVector3 position, LLVector3 normal)
+LLVector2 LLFace::surfaceToTexture(LLVector2 surface_coord, const LLVector4a& position, const LLVector4a& normal)
 {
 	LLVector2 tc = surface_coord;
 	
@@ -894,7 +900,9 @@ LLVector2 LLFace::surfaceToTexture(LLVector2 surface_coord, LLVector3 position,
 		LLVector4a& center = *(mDrawablep->getVOVolume()->getVolume()->getVolumeFace(mTEOffset).mCenter);
 		
 		LLVector4a volume_position;
-		volume_position.load3(mDrawablep->getVOVolume()->agentPositionToVolume(position).mV);
+		LLVector3 v_position(position.getF32ptr());
+
+		volume_position.load3(mDrawablep->getVOVolume()->agentPositionToVolume(v_position).mV);
 		
 		if (!mDrawablep->getVOVolume()->isVolumeGlobal())
 		{
@@ -904,7 +912,8 @@ LLVector2 LLFace::surfaceToTexture(LLVector2 surface_coord, LLVector3 position,
 		}
 		
 		LLVector4a volume_normal;
-		volume_normal.load3(mDrawablep->getVOVolume()->agentDirectionToVolume(normal).mV);
+		LLVector3 v_normal(normal.getF32ptr());
+		volume_normal.load3(mDrawablep->getVOVolume()->agentDirectionToVolume(v_normal).mV);
 		volume_normal.normalize3fast();
 		
 		if (texgen == LLTextureEntry::TEX_GEN_PLANAR)
@@ -937,7 +946,12 @@ void LLFace::getPlanarProjectedParams(LLQuaternion* face_rot, LLVector3* face_po
 	const LLMatrix4& vol_mat = getWorldMatrix();
 	const LLVolumeFace& vf = getViewerObject()->getVolume()->getVolumeFace(mTEOffset);
 	const LLVector4a& normal4a = vf.mNormals[0];
-	const LLVector4a& binormal4a = vf.mBinormals[0];
+	const LLVector4a& tangent = vf.mTangents[0];
+
+	LLVector4a binormal4a;
+	binormal4a.setCross3(normal4a, tangent);
+	binormal4a.mul(tangent.getF32ptr()[3]);
+
 	LLVector2 projected_binormal;
 	planarProjection(projected_binormal, normal4a, *vf.mCenter, binormal4a);
 	projected_binormal -= LLVector2(0.5f, 0.5f); // this normally happens in xform()
@@ -1067,7 +1081,7 @@ void LLFace::cacheFaceInVRAM(const LLVolumeFace& vf)
 {
 	LLFastTimer t(FTM_FACE_GEOM_VOLUME);
 	U32 mask = LLVertexBuffer::MAP_VERTEX | LLVertexBuffer::MAP_TEXCOORD0 |
-				LLVertexBuffer::MAP_BINORMAL | LLVertexBuffer::MAP_NORMAL;
+				LLVertexBuffer::MAP_TANGENT | LLVertexBuffer::MAP_NORMAL;
 	
 	if (vf.mWeights)
 	{
@@ -1080,11 +1094,11 @@ void LLFace::cacheFaceInVRAM(const LLVolumeFace& vf)
 	buff->allocateBuffer(vf.mNumVertices, 0, true);

 	LLStrider<LLVector4a> f_vert;
-	LLStrider<LLVector3> f_binorm;
+	LLStrider<LLVector4a> f_tangent;
 	LLStrider<LLVector3> f_norm;
 	LLStrider<LLVector2> f_tc;

-	buff->getBinormalStrider(f_binorm);
+	buff->getTangentStrider(f_tangent);
 	buff->getVertexStrider(f_vert);
 	buff->getNormalStrider(f_norm);
 	buff->getTexCoord0Strider(f_tc);
@@ -1092,7 +1106,7 @@ void LLFace::cacheFaceInVRAM(const LLVolumeFace& vf)
 	for (U32 i = 0; i < (U32)vf.mNumVertices; ++i)
 	{
 		*f_vert++ = vf.mPositions[i];
-		(*f_binorm++).set(vf.mBinormals[i].getF32ptr());
+		*f_tangent++ = vf.mTangents[i];
 		*f_tc++ = vf.mTexCoords[i];
 		(*f_norm++).set(vf.mNormals[i].getF32ptr());
 	}
@@ -1133,7 +1147,7 @@ static LLFastTimer::DeclareTimer FTM_FACE_GEOM_TEXTURE("Texture");
 static LLFastTimer::DeclareTimer FTM_FACE_GEOM_COLOR("Color");
 static LLFastTimer::DeclareTimer FTM_FACE_GEOM_EMISSIVE("Emissive");
 static LLFastTimer::DeclareTimer FTM_FACE_GEOM_WEIGHTS("Weights");
-static LLFastTimer::DeclareTimer FTM_FACE_GEOM_BINORMAL("Binormal");
+static LLFastTimer::DeclareTimer FTM_FACE_GEOM_TANGENT("Binormal");

 static LLFastTimer::DeclareTimer FTM_FACE_GEOM_FEEDBACK("Face Feedback");
 static LLFastTimer::DeclareTimer FTM_FACE_GEOM_FEEDBACK_POSITION("Feedback Position");
@@ -1207,7 +1221,7 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume,
 	LLStrider<LLVector2> tex_coords1;
 	LLStrider<LLVector3> norm;
 	LLStrider<LLColor4U> colors;
-	LLStrider<LLVector3> binorm;
+	LLStrider<LLVector3> tangent;
 	LLStrider<U16> indicesp;
 	LLStrider<LLVector4> wght;

@@ -1229,7 +1243,7 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume,
 	bool rebuild_emissive = rebuild_color && mVertexBuffer->hasDataType(LLVertexBuffer::TYPE_EMISSIVE);
 	bool rebuild_tcoord = full_rebuild || mDrawablep->isState(LLDrawable::REBUILD_TCOORD);
 	bool rebuild_normal = rebuild_pos && mVertexBuffer->hasDataType(LLVertexBuffer::TYPE_NORMAL);
-	bool rebuild_binormal = rebuild_pos && mVertexBuffer->hasDataType(LLVertexBuffer::TYPE_BINORMAL);
+	bool rebuild_tangent = rebuild_pos && mVertexBuffer->hasDataType(LLVertexBuffer::TYPE_TANGENT);
 	bool rebuild_weights = rebuild_pos && mVertexBuffer->hasDataType(LLVertexBuffer::TYPE_WEIGHT4);

 	const LLTextureEntry *tep = mVObjp->getTE(f);
@@ -1374,7 +1388,7 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume,

 		if (vf.mVertexBuffer.isNull() || buff->getNumVerts() != vf.mNumVertices)
 		{
-			mVObjp->getVolume()->genBinormals(f);
+			mVObjp->getVolume()->genTangents(f);
 			LLFace::cacheFaceInVRAM(vf);
 			buff = (LLVertexBuffer*) vf.mVertexBuffer.get();
 		}		
@@ -1459,15 +1473,15 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume,
 			glEndTransformFeedback();
 		}

-		if (rebuild_binormal)
+		if (rebuild_tangent)
 		{
-			LLFastTimer t(FTM_FACE_GEOM_FEEDBACK_BINORMAL);
-			gTransformBinormalProgram.bind();
+			LLFastTimer t(FTM_FACE_GEOM_TANGENT);
+			gTransformTangentProgram.bind();
 			
-			mVertexBuffer->bindForFeedback(0, LLVertexBuffer::TYPE_BINORMAL, mGeomIndex, mGeomCount);
+			mVertexBuffer->bindForFeedback(0, LLVertexBuffer::TYPE_TANGENT, mGeomIndex, mGeomCount);
 						
 			glBeginTransformFeedback(GL_POINTS);
-			buff->setBuffer(LLVertexBuffer::MAP_BINORMAL);
+			buff->setBuffer(LLVertexBuffer::MAP_TANGENT);
 			push_for_transform(buff, vf.mNumVertices, mGeomCount);
 			glEndTransformFeedback();
 		}
@@ -1529,7 +1543,7 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume,
 		
 			if (bump_code)
 			{
-				mVObjp->getVolume()->genBinormals(f);
+				mVObjp->getVolume()->genTangents(f);
 				F32 offset_multiple; 
 				switch( bump_code )
 				{
@@ -1578,7 +1592,7 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume,
 			U8 texgen = getTextureEntry()->getTexGen();
 			if (rebuild_tcoord && texgen != LLTextureEntry::TEX_GEN_DEFAULT)
 			{ //planar texgen needs binormals
-				mVObjp->getVolume()->genBinormals(f);
+				mVObjp->getVolume()->genTangents(f);
 			}

 			U8 tex_mode = 0;
@@ -1789,11 +1803,14 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume,
 		
 					for (S32 i = 0; i < num_vertices; i++)
 					{
-						LLVector4a tangent;
-						tangent.setCross3(vf.mBinormals[i], vf.mNormals[i]);
+						LLVector4a tangent = vf.mTangents[i];

+						LLVector4a binorm;
+						binorm.setCross3(vf.mNormals[i], tangent);
+						binorm.mul(tangent.getF32ptr()[3]);
+						
 						LLMatrix4a tangent_to_object;
-						tangent_to_object.setRows(tangent, vf.mBinormals[i], vf.mNormals[i]);
+						tangent_to_object.setRows(tangent, binorm, vf.mNormals[i]);
 						LLVector4a t;
 						tangent_to_object.rotate(binormal_dir, t);
 						LLVector4a binormal;
@@ -1899,7 +1916,6 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume,
 			LLFastTimer t(FTM_FACE_GEOM_NORMAL);
 			mVertexBuffer->getNormalStrider(norm, mGeomIndex, mGeomCount, map_range);
 			F32* normals = (F32*) norm.get();
-	
 			for (S32 i = 0; i < num_vertices; i++)
 			{	
 				LLVector4a normal;
@@ -1915,19 +1931,27 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume,
 			}
 		}
 		
-		if (rebuild_binormal)
+		if (rebuild_tangent)
 		{
-			LLFastTimer t(FTM_FACE_GEOM_BINORMAL);
-			mVertexBuffer->getBinormalStrider(binorm, mGeomIndex, mGeomCount, map_range);
-			F32* binormals = (F32*) binorm.get();
-		
+			LLFastTimer t(FTM_FACE_GEOM_TANGENT);
+			mVertexBuffer->getTangentStrider(tangent, mGeomIndex, mGeomCount, map_range);
+			F32* tangents = (F32*) tangent.get();
+			
+			mVObjp->getVolume()->genTangents(f);
+			
+			LLVector4Logical mask;
+			mask.clear();
+			mask.setElement<3>();
+
 			for (S32 i = 0; i < num_vertices; i++)
-			{	
-				LLVector4a binormal;
-				mat_normal.rotate(vf.mBinormals[i], binormal);
-				binormal.normalize3fast();
-				binormal.store4a(binormals);
-				binormals += 4;
+			{
+				LLVector4a tangent_out;
+				mat_normal.rotate(vf.mTangents[i], tangent_out);
+				tangent_out.normalize3fast();
+				tangent_out.setSelectWithMask(mask, vf.mTangents[i], tangent_out);
+				tangent_out.store4a(tangents);
+				
+				tangents += 4;
 			}

 			if (map_range)
--- a/indra/newview/llface.h
+++ b/indra/newview/llface.h
@@ -121,7 +121,7 @@ public:
 	LLXformMatrix*	getXform()			const	{ return mXform; }
 	BOOL			hasGeometry()		const	{ return mGeomCount > 0; }
 	LLVector3		getPositionAgent()	const;
-	LLVector2       surfaceToTexture(LLVector2 surface_coord, LLVector3 position, LLVector3 normal);
+	LLVector2       surfaceToTexture(LLVector2 surface_coord, const LLVector4a& position, const LLVector4a& normal);
 	void 			getPlanarProjectedParams(LLQuaternion* face_rot, LLVector3* face_pos, F32* scale) const;
 	bool			calcAlignedPlanarTE(const LLFace* align_to, LLVector2* st_offset,
 										LLVector2* st_scale, F32* st_rot) const;
--- a/indra/newview/llflexibleobject.cpp
+++ b/indra/newview/llflexibleobject.cpp
@@ -293,6 +293,9 @@ void LLVolumeImplFlexible::onSetVolume(const LLVolumeParams &volume_params, cons

 void LLVolumeImplFlexible::updateRenderRes()
 {
+	if (!mAttributes)
+		return;
+
 	LLDrawable* drawablep = mVO->mDrawable;

 	S32 new_res = mAttributes->getSimulateLOD();
@@ -424,7 +427,7 @@ void LLVolumeImplFlexible::doFlexibleUpdate()
 		}
 	}

-	if(!mInitialized)
+	if(!mInitialized || !mAttributes)
 	{
 		//the object is not visible
 		return ;
@@ -682,15 +685,17 @@ void LLVolumeImplFlexible::doFlexibleUpdate()
 		new_point = &path->mPath[i];
 		LLVector3 pos = newSection[i].mPosition * rel_xform;
 		LLQuaternion rot = mSection[i].mAxisRotation * newSection[i].mRotation * delta_rot;
-		
-		if (!mUpdated || (new_point->mPos-pos).magVec()/mVO->mDrawable->mDistanceWRTCamera > 0.001f)
+	
+		LLVector3 np(new_point->mPos.getF32ptr());
+
+		if (!mUpdated || (np-pos).magVec()/mVO->mDrawable->mDistanceWRTCamera > 0.001f)
 		{
-			new_point->mPos = newSection[i].mPosition * rel_xform;
+			new_point->mPos.load3((newSection[i].mPosition * rel_xform).mV);
 			mUpdated = FALSE;
 		}

-		new_point->mRot = rot;
-		new_point->mScale = newSection[i].mScale;
+		new_point->mRot.loadu(LLMatrix3(rot));
+		new_point->mScale.set(newSection[i].mScale.mV[0], newSection[i].mScale.mV[1], 0,1);
 		new_point->mTexT = ((F32)i)/(num_render_sections);
 	}

--- a/indra/newview/llhudicon.cpp
+++ b/indra/newview/llhudicon.cpp
@@ -208,7 +208,7 @@ void LLHUDIcon::render()
 	renderIcon(FALSE);
 }

-BOOL LLHUDIcon::lineSegmentIntersect(const LLVector3& start, const LLVector3& end, LLVector3* intersection)
+BOOL LLHUDIcon::lineSegmentIntersect(const LLVector4a& start, const LLVector4a& end, LLVector4a* intersection)
 {
 	if (mHidden)
 		return FALSE;
@@ -281,23 +281,18 @@ BOOL LLHUDIcon::lineSegmentIntersect(const LLVector3& start, const LLVector3& en
 	LLVector4a upper_right;
 	upper_right.setAdd(lower_right, y_scalea);

-	LLVector4a enda;
-	enda.load3(end.mV);
-	LLVector4a starta;
-	starta.load3(start.mV);
 	LLVector4a dir;
-	dir.setSub(enda, starta);
+	dir.setSub(end, start);

 	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 (LLTriangleRayIntersect(upper_right, upper_left, lower_right, start, dir, a,b,t) ||
+		LLTriangleRayIntersect(upper_left, lower_left, lower_right, start, dir, a,b,t))
 	{
 		if (intersection)
 		{
 			dir.mul(t);
-			starta.add(dir);
-			*intersection = LLVector3(starta.getF32ptr());
+			intersection->setAdd(start, dir);
 		}
 		return TRUE;
 	}
@@ -337,12 +332,12 @@ LLHUDIcon* LLHUDIcon::handlePick(S32 pick_id)
 }

 //static
-LLHUDIcon* LLHUDIcon::lineSegmentIntersectAll(const LLVector3& start, const LLVector3& end, LLVector3* intersection)
+LLHUDIcon* LLHUDIcon::lineSegmentIntersectAll(const LLVector4a& start, const LLVector4a& end, LLVector4a* intersection)
 {
 	icon_instance_t::iterator icon_it;

-	LLVector3 local_end = end;
-	LLVector3 position;
+	LLVector4a local_end = end;
+	LLVector4a position;

 	LLHUDIcon* ret = NULL;
 	for(icon_it = sIconInstances.begin(); icon_it != sIconInstances.end(); ++icon_it)
--- a/indra/newview/llhudicon.h
+++ b/indra/newview/llhudicon.h
@@ -68,7 +68,7 @@ public:

 	static S32 generatePickIDs(S32 start_id, S32 step_size);
 	static LLHUDIcon* handlePick(S32 pick_id);
-	static LLHUDIcon* lineSegmentIntersectAll(const LLVector3& start, const LLVector3& end, LLVector3* intersection);
+	static LLHUDIcon* lineSegmentIntersectAll(const LLVector4a& start, const LLVector4a& end, LLVector4a* intersection);

 	static void updateAll();
 	static void cleanupDeadIcons();
@@ -79,7 +79,7 @@ public:
 	BOOL getHidden() const { return mHidden; }
 	void setHidden( BOOL hide ) { mHidden = hide; }

-	BOOL lineSegmentIntersect(const LLVector3& start, const LLVector3& end, LLVector3* intersection);
+	BOOL lineSegmentIntersect(const LLVector4a& start, const LLVector4a& end, LLVector4a* intersection);

 protected:
 	LLHUDIcon(const U8 type);
--- a/indra/newview/llhudnametag.cpp
+++ b/indra/newview/llhudnametag.cpp
@@ -120,7 +120,7 @@ LLHUDNameTag::~LLHUDNameTag()
 }


-BOOL LLHUDNameTag::lineSegmentIntersect(const LLVector3& start, const LLVector3& end, LLVector3& intersection, BOOL debug_render)
+BOOL LLHUDNameTag::lineSegmentIntersect(const LLVector4a& start, const LLVector4a& end, LLVector4a& intersection, BOOL debug_render)
 {
 	if (!mVisible || mHidden)
 	{
@@ -218,16 +218,23 @@ BOOL LLHUDNameTag::lineSegmentIntersect(const LLVector3& start, const LLVector3&
 			gGL.vertex3fv(v[2].mV);
 			gGL.end();
 		}
-
-		LLVector3 dir = end-start;
+		LLVector4a dir;
+		dir.setSub(end,start);
 		F32 a, b, t;

-		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) )
+		LLVector4a v0,v1,v2,v3;
+		v0.load3(v[0].mV);
+		v1.load3(v[1].mV);
+		v2.load3(v[2].mV);
+		v3.load3(v[3].mV);
+
+		if (LLTriangleRayIntersect(v0, v1, v2, start, dir, a, b, t) ||
+			LLTriangleRayIntersect(v2, v3, v0, start, dir, a, b, t) )
 		{
 			if (t <= 1.f)
 			{
-				intersection = start + dir*t;
+				dir.mul(t);
+				intersection.setAdd(start, dir);
 				return TRUE;
 			}
 		}
--- a/indra/newview/llhudnametag.h
+++ b/indra/newview/llhudnametag.h
@@ -126,7 +126,7 @@ public:
 	void setHidden( BOOL hide ) { mHidden = hide; }
 	void shift(const LLVector3& offset);

-	BOOL lineSegmentIntersect(const LLVector3& start, const LLVector3& end, LLVector3& intersection, BOOL debug_render = FALSE);
+	BOOL lineSegmentIntersect(const LLVector4a& start, const LLVector4a& end, LLVector4a& intersection, BOOL debug_render = FALSE);

 	static void shiftAll(const LLVector3& offset);
 	static void addPickable(std::set<LLViewerObject*> &pick_list);
--- a/indra/newview/llspatialpartition.cpp
+++ b/indra/newview/llspatialpartition.cpp
@@ -2944,9 +2944,9 @@ void renderNormals(LLDrawable* drawablep)
 				gGL.vertex3fv(face.mPositions[j].getF32ptr());
 				gGL.vertex3fv(p.getF32ptr());
 				
-				if (face.mBinormals)
+				if (face.mTangents)
 				{
-					n.setMul(face.mBinormals[j], scale);
+					n.setMul(face.mTangents[j], scale);
 					p.setAdd(face.mPositions[j], n);
 				
 					gGL.diffuseColor4f(0,1,1,1);
@@ -3694,11 +3694,17 @@ void renderRaycast(LLDrawable* drawablep)
 					gGL.translatef(trans.mV[0], trans.mV[1], trans.mV[2]);					
 					gGL.multMatrix((F32*) vobj->getRelativeXform().mMatrix);

-					LLVector3 start, end;
+					LLVector4a start, end;
 					if (transform)
 					{
-						start = vobj->agentPositionToVolume(gDebugRaycastStart);
-						end = vobj->agentPositionToVolume(gDebugRaycastEnd);
+						LLVector3 v_start(gDebugRaycastStart.getF32ptr());
+						LLVector3 v_end(gDebugRaycastEnd.getF32ptr());
+
+						v_start = vobj->agentPositionToVolume(v_start);
+						v_end = vobj->agentPositionToVolume(v_end);
+
+						start.load3(v_start.mV);
+						end.load3(v_end.mV);
 					}
 					else
 					{
@@ -3706,11 +3712,8 @@ void renderRaycast(LLDrawable* drawablep)
 						end = gDebugRaycastEnd;
 					}

-					LLVector4a starta, enda;
-					starta.load3(start.mV);
-					enda.load3(end.mV);
 					LLVector4a dir;
-					dir.setSub(enda, starta);
+					dir.setSub(end, start);

 					gGL.flush();
 					glPolygonMode(GL_FRONT_AND_BACK, GL_LINE);				
@@ -3733,10 +3736,11 @@ void renderRaycast(LLDrawable* drawablep)
 							((LLVolumeFace*) &face)->createOctree(); 
 						}

-						LLRenderOctreeRaycast render(starta, dir, &t);
+						LLRenderOctreeRaycast render(start, dir, &t);
 					
 						render.traverse(face.mOctree);
 					}
+
 					gGL.popMatrix();		
 					glPolygonMode(GL_FRONT_AND_BACK, GL_FILL);
 				}
@@ -3757,10 +3761,18 @@ void renderRaycast(LLDrawable* drawablep)
 			// draw intersection point
 			gGL.pushMatrix();
 			gGL.loadMatrix(gGLModelView);
-			LLVector3 translate = gDebugRaycastIntersection;
+			LLVector3 translate(gDebugRaycastIntersection.getF32ptr());
 			gGL.translatef(translate.mV[0], translate.mV[1], translate.mV[2]);
 			LLCoordFrame orient;
-			orient.lookDir(gDebugRaycastNormal, gDebugRaycastBinormal);
+			LLVector4a debug_binormal;
+			
+			debug_binormal.setCross3(gDebugRaycastNormal, gDebugRaycastTangent);
+			debug_binormal.mul(gDebugRaycastTangent.getF32ptr()[3]);
+
+			LLVector3 normal(gDebugRaycastNormal.getF32ptr());
+			LLVector3 binormal(debug_binormal.getF32ptr());
+						
+			orient.lookDir(normal, binormal);
 			LLMatrix4 rotation;
 			orient.getRotMatrixToParent(rotation);
 			gGL.multMatrix((float*)rotation.mMatrix);
@@ -4261,28 +4273,30 @@ BOOL LLSpatialPartition::isVisible(const LLVector3& v)
 	return TRUE;
 }

+LL_ALIGN_PREFIX(16)
 class LLOctreeIntersect : public LLSpatialGroup::OctreeTraveler
 {
 public:
-	LLVector3 mStart;
-	LLVector3 mEnd;
+	LL_ALIGN_16(LLVector4a mStart);
+	LL_ALIGN_16(LLVector4a mEnd);
+
 	S32       *mFaceHit;
-	LLVector3 *mIntersection;
+	LLVector4a *mIntersection;
 	LLVector2 *mTexCoord;
-	LLVector3 *mNormal;
-	LLVector3 *mBinormal;
+	LLVector4a *mNormal;
+	LLVector4a *mTangent;
 	LLDrawable* mHit;
 	BOOL mPickTransparent;

-	LLOctreeIntersect(LLVector3 start, LLVector3 end, BOOL pick_transparent,
-					  S32* face_hit, LLVector3* intersection, LLVector2* tex_coord, LLVector3* normal, LLVector3* binormal)
+	LLOctreeIntersect(const LLVector4a& start, const LLVector4a& end, BOOL pick_transparent,
+					  S32* face_hit, LLVector4a* intersection, LLVector2* tex_coord, LLVector4a* normal, LLVector4a* tangent)
 		: mStart(start),
 		  mEnd(end),
 		  mFaceHit(face_hit),
 		  mIntersection(intersection),
 		  mTexCoord(tex_coord),
 		  mNormal(normal),
-		  mBinormal(binormal),
+		  mTangent(tangent),
 		  mHit(NULL),
 		  mPickTransparent(pick_transparent)
 	{
@@ -4315,23 +4329,22 @@ public:
 			size = group->mBounds[1];
 			center = group->mBounds[0];
 			
-			LLVector3 local_start = mStart;
-			LLVector3 local_end   = mEnd;
+			LLVector4a local_start = mStart;
+			LLVector4a local_end   = mEnd;

 			if (group->mSpatialPartition->isBridge())
 			{
 				LLMatrix4 local_matrix = group->mSpatialPartition->asBridge()->mDrawable->getRenderMatrix();
 				local_matrix.invert();
-				
-				local_start = mStart * local_matrix;
-				local_end   = mEnd   * local_matrix;
+
+				LLMatrix4a local_matrix4a;
+				local_matrix4a.loadu(local_matrix);
+
+				local_matrix4a.affineTransform(mStart, local_start);
+				local_matrix4a.affineTransform(mEnd, local_end);
 			}

-			LLVector4a start, end;
-			start.load3(local_start.mV);
-			end.load3(local_end.mV);
-
-			if (LLLineSegmentBoxIntersect(start, end, center, size))
+			if (LLLineSegmentBoxIntersect(local_start, local_end, center, size))
 			{
 				check(child);
 			}
@@ -4362,14 +4375,14 @@ public:

 			if (vobj)
 			{
-				LLVector3 intersection;
+				LLVector4a intersection;
 				bool skip_check = false;
 				if (vobj->isAvatar())
 				{
 					LLVOAvatar* avatar = (LLVOAvatar*) vobj;
 					if (avatar->isSelf() && gFloaterTools->getVisible())
 					{
-						LLViewerObject* hit = avatar->lineSegmentIntersectRiggedAttachments(mStart, mEnd, -1, mPickTransparent, mFaceHit, &intersection, mTexCoord, mNormal, mBinormal);
+						LLViewerObject* hit = avatar->lineSegmentIntersectRiggedAttachments(mStart, mEnd, -1, mPickTransparent, mFaceHit, &intersection, mTexCoord, mNormal, mTangent);
 						if (hit)
 						{
 							mEnd = intersection;
@@ -4389,7 +4402,7 @@ public:
 					}
 				}

-				if (!skip_check && vobj->lineSegmentIntersect(mStart, mEnd, -1, mPickTransparent, mFaceHit, &intersection, mTexCoord, mNormal, mBinormal))
+				if (!skip_check && vobj->lineSegmentIntersect(mStart, mEnd, -1, mPickTransparent, mFaceHit, &intersection, mTexCoord, mNormal, mTangent))
 				{
 					mEnd = intersection;  // shorten ray so we only find CLOSER hits
 					if (mIntersection)
@@ -4404,19 +4417,19 @@ public:
 				
 		return false;
 	}
-};
+} LL_ALIGN_POSTFIX(16);

-LLDrawable* LLSpatialPartition::lineSegmentIntersect(const LLVector3& start, const LLVector3& end,
+LLDrawable* LLSpatialPartition::lineSegmentIntersect(const LLVector4a& start, const LLVector4a& end,
 													 BOOL pick_transparent,													
 													 S32* face_hit,                   // return the face hit
-													 LLVector3* intersection,         // return the intersection point
+													 LLVector4a* intersection,         // return the intersection point
 													 LLVector2* tex_coord,            // return the texture coordinates of the intersection point
-													 LLVector3* normal,               // return the surface normal at the intersection point
-													 LLVector3* bi_normal             // return the surface bi-normal at the intersection point
+													 LLVector4a* normal,               // return the surface normal at the intersection point
+													 LLVector4a* tangent			// return the surface tangent at the intersection point
 	)

 {
-	LLOctreeIntersect intersect(start, end, pick_transparent, face_hit, intersection, tex_coord, normal, bi_normal);
+	LLOctreeIntersect intersect(start, end, pick_transparent, face_hit, intersection, tex_coord, normal, tangent);
 	LLDrawable* drawable = intersect.check(mOctree);

 	return drawable;
--- a/indra/newview/llspatialpartition.h
+++ b/indra/newview/llspatialpartition.h
@@ -457,13 +457,13 @@ public:
 	LLSpatialGroup *put(LLDrawable *drawablep, BOOL was_visible = FALSE);
 	BOOL remove(LLDrawable *drawablep, LLSpatialGroup *curp);
 	
-	LLDrawable* lineSegmentIntersect(const LLVector3& start, const LLVector3& end,
+	LLDrawable* lineSegmentIntersect(const LLVector4a& start, const LLVector4a& end,
 									 BOOL pick_transparent, 
 									 S32* face_hit,                          // return the face hit
-									 LLVector3* intersection = NULL,         // return the intersection point
+									 LLVector4a* intersection = NULL,         // return the intersection point
 									 LLVector2* tex_coord = NULL,            // return the texture coordinates of the intersection point
-									 LLVector3* normal = NULL,               // return the surface normal at the intersection point
-									 LLVector3* bi_normal = NULL             // return the surface bi-normal at the intersection point
+									 LLVector4a* normal = NULL,               // return the surface normal at the intersection point
+									 LLVector4a* tangent = NULL             // return the surface tangent at the intersection point
 		);
 	
 	
@@ -695,7 +695,7 @@ class LLVolumeGeometryManager: public LLGeometryManager
 	virtual void rebuildGeom(LLSpatialGroup* group);
 	virtual void rebuildMesh(LLSpatialGroup* group);
 	virtual void getGeometry(LLSpatialGroup* group);
-	void genDrawInfo(LLSpatialGroup* group, U32 mask, std::vector<LLFace*>& faces, BOOL distance_sort = FALSE, BOOL batch_textures = FALSE);
+	void genDrawInfo(LLSpatialGroup* group, U32 mask, LLFace** faces, U32 face_count, BOOL distance_sort = FALSE, BOOL batch_textures = FALSE, BOOL no_materials = FALSE);
 	void registerFace(LLSpatialGroup* group, LLFace* facep, U32 type);
 };

--- a/indra/newview/llviewerobject.cpp
+++ b/indra/newview/llviewerobject.cpp
@@ -3904,19 +3904,19 @@ LLViewerObject* LLViewerObject::getRootEdit() const
 }


-BOOL LLViewerObject::lineSegmentIntersect(const LLVector3& start, const LLVector3& end,
+BOOL LLViewerObject::lineSegmentIntersect(const LLVector4a& start, const LLVector4a& end,
 										  S32 face,
 										  BOOL pick_transparent,
 										  S32* face_hit,
-										  LLVector3* intersection,
+										  LLVector4a* intersection,
 										  LLVector2* tex_coord,
-										  LLVector3* normal,
-										  LLVector3* bi_normal)
+										  LLVector4a* normal,
+										  LLVector4a* tangent)
 {
 	return false;
 }

-BOOL LLViewerObject::lineSegmentBoundingBox(const LLVector3& start, const LLVector3& end)
+BOOL LLViewerObject::lineSegmentBoundingBox(const LLVector4a& start, const LLVector4a& end)
 {
 	if (mDrawable.isNull() || mDrawable->isDead())
 	{
@@ -3933,11 +3933,7 @@ BOOL LLViewerObject::lineSegmentBoundingBox(const LLVector3& start, const LLVect
 	size.setSub(ext[1], ext[0]);
 	size.mul(0.5f);

-	LLVector4a starta, enda;
-	starta.load3(start.mV);
-	enda.load3(end.mV);
-
-	return LLLineSegmentBoxIntersect(starta, enda, center, size);
+	return LLLineSegmentBoxIntersect(start, end, center, size);
 }

 U8 LLViewerObject::getMediaType() const
--- a/indra/newview/llviewerobject.h
+++ b/indra/newview/llviewerobject.h
@@ -262,17 +262,17 @@ public:

 	//detect if given line segment (in agent space) intersects with this viewer object.
 	//returns TRUE if intersection detected and returns information about intersection
-	virtual BOOL lineSegmentIntersect(const LLVector3& start, const LLVector3& end,
+	virtual BOOL lineSegmentIntersect(const LLVector4a& start, const LLVector4a& end,
 									  S32 face = -1,                          // which face to check, -1 = ALL_SIDES
 									  BOOL pick_transparent = FALSE,
 									  S32* face_hit = NULL,                   // which face was hit
-									  LLVector3* intersection = NULL,         // return the intersection point
+									  LLVector4a* intersection = NULL,         // return the intersection point
 									  LLVector2* tex_coord = NULL,            // return the texture coordinates of the intersection point
-									  LLVector3* normal = NULL,               // return the surface normal at the intersection point
-									  LLVector3* bi_normal = NULL             // return the surface bi-normal at the intersection point
+									  LLVector4a* normal = NULL,               // return the surface normal at the intersection point
+									  LLVector4a* tangent = NULL             // return the surface tangent at the intersection point
 		);
 	
-	virtual BOOL lineSegmentBoundingBox(const LLVector3& start, const LLVector3& end);
+	virtual BOOL lineSegmentBoundingBox(const LLVector4a& start, const LLVector4a& end);

 	virtual const LLVector3d getPositionGlobal() const;
 	virtual const LLVector3 &getPositionRegion() const;
--- a/indra/newview/llviewershadermgr.cpp
+++ b/indra/newview/llviewershadermgr.cpp
@@ -77,7 +77,7 @@ LLGLSLShader			gTransformPositionProgram(LLViewerShaderMgr::SHADER_TRANSFORM);
 LLGLSLShader			gTransformTexCoordProgram(LLViewerShaderMgr::SHADER_TRANSFORM);
 LLGLSLShader			gTransformNormalProgram(LLViewerShaderMgr::SHADER_TRANSFORM);
 LLGLSLShader			gTransformColorProgram(LLViewerShaderMgr::SHADER_TRANSFORM);
-LLGLSLShader			gTransformBinormalProgram(LLViewerShaderMgr::SHADER_TRANSFORM);
+LLGLSLShader			gTransformTangentProgram(LLViewerShaderMgr::SHADER_TRANSFORM);

 //utility shaders
 LLGLSLShader	gOcclusionProgram(LLViewerShaderMgr::SHADER_INTERFACE);
@@ -2747,16 +2747,16 @@ BOOL LLViewerShaderMgr::loadTransformShaders()

 	if (success)
 	{
-		gTransformBinormalProgram.mName = "Binormal Transform Shader";
-		gTransformBinormalProgram.mShaderFiles.clear();
-		gTransformBinormalProgram.mShaderFiles.push_back(make_pair("transform/binormalV.glsl", GL_VERTEX_SHADER_ARB));
-		gTransformBinormalProgram.mShaderLevel = mVertexShaderLevel[SHADER_TRANSFORM];
+		gTransformTangentProgram.mName = "Binormal Transform Shader";
+		gTransformTangentProgram.mShaderFiles.clear();
+		gTransformTangentProgram.mShaderFiles.push_back(make_pair("transform/binormalV.glsl", GL_VERTEX_SHADER_ARB));
+		gTransformTangentProgram.mShaderLevel = mVertexShaderLevel[SHADER_TRANSFORM];

 		const char* varyings[] = {
-			"binormal_out",
+			"tangent_out",
 		};
 	
-		success = gTransformBinormalProgram.createShader(NULL, NULL, 1, varyings);
+		success = gTransformTangentProgram.createShader(NULL, NULL, 1, varyings);
 	}

 	
--- a/indra/newview/llviewershadermgr.h
+++ b/indra/newview/llviewershadermgr.h
@@ -234,7 +234,7 @@ extern LLGLSLShader			gTransformPositionProgram;
 extern LLGLSLShader			gTransformTexCoordProgram;
 extern LLGLSLShader			gTransformNormalProgram;
 extern LLGLSLShader			gTransformColorProgram;
-extern LLGLSLShader			gTransformBinormalProgram;
+extern LLGLSLShader			gTransformTangentProgram;
 //utility shaders
 extern LLGLSLShader			gOcclusionProgram;
 extern LLGLSLShader			gOcclusionCubeProgram;
@@ -244,7 +244,6 @@ extern LLGLSLShader			gSplatTextureRectProgram;
 extern LLGLSLShader			gGlowCombineFXAAProgram;
 extern LLGLSLShader			gDebugProgram;
 extern LLGLSLShader			gClipProgram;
-extern LLGLSLShader			gAlphaMaskProgram;

 //output tex0[tc0] + tex1[tc1]
 extern LLGLSLShader			gTwoTextureAddProgram;
--- a/indra/newview/llviewerwindow.cpp
+++ b/indra/newview/llviewerwindow.cpp
@@ -218,13 +218,13 @@ BOOL			gShowOverlayTitle = FALSE;
 BOOL			gPickTransparent = TRUE;

 LLViewerObject*  gDebugRaycastObject = NULL;
-LLVector3       gDebugRaycastIntersection;
-LLVector2       gDebugRaycastTexCoord;
-LLVector3       gDebugRaycastNormal;
-LLVector3       gDebugRaycastBinormal;
-S32				gDebugRaycastFaceHit;
-LLVector3		gDebugRaycastStart;
-LLVector3		gDebugRaycastEnd;
+LLVector4a       gDebugRaycastIntersection;
+LLVector2        gDebugRaycastTexCoord;
+LLVector4a       gDebugRaycastNormal;
+LLVector4a       gDebugRaycastTangent;
+S32				 gDebugRaycastFaceHit;
+LLVector4a		 gDebugRaycastStart;
+LLVector4a		 gDebugRaycastEnd;

 // HUD display lines in lower right
 BOOL				gDisplayWindInfo = FALSE;
@@ -3038,7 +3038,7 @@ void LLViewerWindow::updateUI()
 											  &gDebugRaycastIntersection,
 											  &gDebugRaycastTexCoord,
 											  &gDebugRaycastNormal,
-											  &gDebugRaycastBinormal,
+											  &gDebugRaycastTangent,
 											  &gDebugRaycastStart,
 											  &gDebugRaycastEnd);
 	}
@@ -3982,7 +3982,7 @@ LLPickInfo LLViewerWindow::pickImmediate(S32 x, S32 y_from_bot,  BOOL pick_trans
 }

 LLHUDIcon* LLViewerWindow::cursorIntersectIcon(S32 mouse_x, S32 mouse_y, F32 depth,
-										   LLVector3* intersection)
+										   LLVector4a* intersection)
 {
 	S32 x = mouse_x;
 	S32 y = mouse_y;
@@ -3999,9 +3999,11 @@ LLHUDIcon* LLViewerWindow::cursorIntersectIcon(S32 mouse_x, S32 mouse_y, F32 dep
 	LLVector3 mouse_world_start = mouse_point_global;
 	LLVector3 mouse_world_end   = mouse_point_global + mouse_direction_global * depth;

-	return LLHUDIcon::lineSegmentIntersectAll(mouse_world_start, mouse_world_end, intersection);
-
+	LLVector4a start, end;
+	start.load3(mouse_world_start.mV);
+	end.load3(mouse_world_end.mV);
 	
+	return LLHUDIcon::lineSegmentIntersectAll(start, end, intersection);
 }

 LLViewerObject* LLViewerWindow::cursorIntersect(S32 mouse_x, S32 mouse_y, F32 depth,
@@ -4009,12 +4011,12 @@ LLViewerObject* LLViewerWindow::cursorIntersect(S32 mouse_x, S32 mouse_y, F32 de
 												S32 this_face,
 												BOOL pick_transparent,
 												S32* face_hit,
-												LLVector3 *intersection,
+												LLVector4a *intersection,
 												LLVector2 *uv,
-												LLVector3 *normal,
-												LLVector3 *binormal,
-												LLVector3* start,
-												LLVector3* end)
+												LLVector4a *normal,
+												LLVector4a *tangent,
+												LLVector4a* start,
+												LLVector4a* end)
 {
 	S32 x = mouse_x;
 	S32 y = mouse_y;
@@ -4049,17 +4051,27 @@ LLViewerObject* LLViewerWindow::cursorIntersect(S32 mouse_x, S32 mouse_y, F32 de
 	if (!LLViewerJoystick::getInstance()->getOverrideCamera())
 	{ //always set raycast intersection to mouse_world_end unless
 		//flycam is on (for DoF effect)
-		gDebugRaycastIntersection = mouse_world_end;
+		gDebugRaycastIntersection.load3(mouse_world_end.mV);
 	}

+	LLVector4a mw_start;
+	mw_start.load3(mouse_world_start.mV);
+	LLVector4a mw_end;
+	mw_end.load3(mouse_world_end.mV);
+
+	LLVector4a mh_start;
+	mh_start.load3(mouse_hud_start.mV);
+	LLVector4a mh_end;
+	mh_end.load3(mouse_hud_end.mV);
+
 	if (start)
 	{
-		*start = mouse_world_start;
+		*start = mw_start;
 	}

 	if (end)
 	{
-		*end = mouse_world_end;
+		*end = mw_end;
 	}

 	LLViewerObject* found = NULL;
@@ -4068,25 +4080,25 @@ LLViewerObject* LLViewerWindow::cursorIntersect(S32 mouse_x, S32 mouse_y, F32 de
 	{
 		if (this_object->isHUDAttachment()) // is a HUD object?
 		{
-			if (this_object->lineSegmentIntersect(mouse_hud_start, mouse_hud_end, this_face, pick_transparent,
-												  face_hit, intersection, uv, normal, binormal))
+			if (this_object->lineSegmentIntersect(mh_start, mh_end, this_face, pick_transparent,
+												  face_hit, intersection, uv, normal, tangent))
 			{
 				found = this_object;
 			}
-			}
+		}
 		else // is a world object
 		{
-			if (this_object->lineSegmentIntersect(mouse_world_start, mouse_world_end, this_face, pick_transparent,
-												  face_hit, intersection, uv, normal, binormal))
+			if (this_object->lineSegmentIntersect(mw_start, mw_end, this_face, pick_transparent,
+												  face_hit, intersection, uv, normal, tangent))
 			{
 				found = this_object;
 			}
-			}
-			}
+		}
+	}
 	else // check ALL objects
 	{
-		found = gPipeline.lineSegmentIntersectInHUD(mouse_hud_start, mouse_hud_end, pick_transparent,
-													face_hit, intersection, uv, normal, binormal);
+		found = gPipeline.lineSegmentIntersectInHUD(mh_start, mh_end, pick_transparent,
+													face_hit, intersection, uv, normal, tangent);

 // [RLVa:KB] - Checked: 2009-12-28 (RLVa-1.1.0k) | Modified: RLVa-1.1.0k
 		if ( (rlv_handler_t::isEnabled()) && (LLToolCamera::getInstance()->hasMouseCapture()) && (gKeyboard->currentMask(TRUE) & MASK_ALT) )
@@ -4097,8 +4109,8 @@ LLViewerObject* LLViewerWindow::cursorIntersect(S32 mouse_x, S32 mouse_y, F32 de

 		if (!found) // if not found in HUD, look in world:
 		{
-			found = gPipeline.lineSegmentIntersectInWorld(mouse_world_start, mouse_world_end, pick_transparent,
-														  face_hit, intersection, uv, normal, binormal);
+			found = gPipeline.lineSegmentIntersectInWorld(mw_start, mw_end, pick_transparent,
+														  face_hit, intersection, uv, normal, tangent);

 // [RLVa:KB] - Checked: 2010-01-02 (RLVa-1.1.0l) | Added: RLVa-1.1.0l
 #ifdef RLV_EXTENSION_CMD_INTERACT
@@ -4118,6 +4130,10 @@ LLViewerObject* LLViewerWindow::cursorIntersect(S32 mouse_x, S32 mouse_y, F32 de
 			}
 #endif // RLV_EXTENSION_CMD_INTERACT
 // [/RLVa:KB]
+			if (found && !pick_transparent)
+			{
+				gDebugRaycastIntersection = *intersection;
+			}
 		}
 	}

@@ -5815,6 +5831,7 @@ LLPickInfo::LLPickInfo()
 	  mXYCoords(-1, -1),
 	  mIntersection(),
 	  mNormal(),
+	  mTangent(),
 	  mBinormal(),
 	  mHUDIcon(NULL),
 	  mPickTransparent(FALSE)
@@ -5836,6 +5853,7 @@ LLPickInfo::LLPickInfo(const LLCoordGL& mouse_pos,
 	  mSTCoords(-1.f, -1.f),
 	  mXYCoords(-1, -1),
 	  mNormal(),
+	  mTangent(),
 	  mBinormal(),
 	  mHUDIcon(NULL),
 	  mPickTransparent(pick_transparent)
@@ -5846,19 +5864,26 @@ void LLPickInfo::fetchResults()
 {

 	S32 face_hit = -1;
-	LLVector3 intersection, normal, binormal;
+	LLVector4a intersection, normal;
+	LLVector4a tangent;
+
 	LLVector2 uv;

 	LLHUDIcon* hit_icon = gViewerWindow->cursorIntersectIcon(mMousePt.mX, mMousePt.mY, 512.f, &intersection);
 	
+	LLVector4a origin;
+	origin.load3(LLViewerCamera::getInstance()->getOrigin().mV);
 	F32 icon_dist = 0.f;
 	if (hit_icon)
 	{
-		icon_dist = (LLViewerCamera::getInstance()->getOrigin()-intersection).magVec();
+		LLVector4a delta;
+		delta.setSub(intersection, origin);
+		icon_dist = delta.getLength3().getF32();
 	}
+
 	LLViewerObject* hit_object = gViewerWindow->cursorIntersect(mMousePt.mX, mMousePt.mY, 512.f,
 									NULL, -1, mPickTransparent, &face_hit,
-									&intersection, &uv, &normal, &binormal);
+									&intersection, &uv, &normal, &tangent);
 	
 	mPickPt = mMousePt;

@@ -5868,9 +5893,13 @@ void LLPickInfo::fetchResults()
 	
 	LLViewerObject* objectp = hit_object;

+
+	LLVector4a delta;
+	delta.setSub(origin, intersection);
+
 	if (hit_icon && 
 		(!objectp || 
-		icon_dist < (LLViewerCamera::getInstance()->getOrigin()-intersection).magVec()))
+		icon_dist < delta.getLength3().getF32()))
 	{
 		// was this name referring to a hud icon?
 		mHUDIcon = hit_icon;
@@ -5907,11 +5936,16 @@ void LLPickInfo::fetchResults()
 			{
 				mPickType = PICK_OBJECT;
 			}
-			mObjectOffset = gAgentCamera.calcFocusOffset(objectp, intersection, mPickPt.mX, mPickPt.mY);
+
+			LLVector3 v_intersection(intersection.getF32ptr());
+
+			mObjectOffset = gAgentCamera.calcFocusOffset(objectp, v_intersection, mPickPt.mX, mPickPt.mY);
 			mObjectID = objectp->mID;
 			mObjectFace = (te_offset == NO_FACE) ? -1 : (S32)te_offset;

-			mPosGlobal = gAgent.getPosGlobalFromAgent(intersection);
+			
+
+			mPosGlobal = gAgent.getPosGlobalFromAgent(v_intersection);
 			
 			if (mWantSurfaceInfo)
 			{
@@ -5955,7 +5989,16 @@ void LLPickInfo::getSurfaceInfo()
 	mIntersection = LLVector3(0,0,0);
 	mNormal       = LLVector3(0,0,0);
 	mBinormal     = LLVector3(0,0,0);
+	mTangent	  = LLVector4(0,0,0,0);
 	
+	LLVector4a tangent;
+	LLVector4a intersection;
+	LLVector4a normal;
+
+	tangent.clear();
+	normal.clear();
+	intersection.clear();
+
 	LLViewerObject* objectp = getObject();

 	if (objectp)
@@ -5963,10 +6006,10 @@ void LLPickInfo::getSurfaceInfo()
 		if (gViewerWindow->cursorIntersect(llround((F32)mMousePt.mX), llround((F32)mMousePt.mY), 1024.f,
 										   objectp, -1, mPickTransparent,
 										   &mObjectFace,
-										   &mIntersection,
+										   &intersection,
 										   &mSTCoords,
-										   &mNormal,
-										   &mBinormal))
+										   &normal,
+										   &tangent))
 		{
 			// if we succeeded with the intersect above, compute the texture coordinates:

@@ -5975,10 +6018,26 @@ void LLPickInfo::getSurfaceInfo()
 				LLFace* facep = objectp->mDrawable->getFace(mObjectFace);
 				if (facep)
 				{
-					mUVCoords = facep->surfaceToTexture(mSTCoords, mIntersection, mNormal);
+					mUVCoords = facep->surfaceToTexture(mSTCoords, intersection, normal);
 				}
 			}

+			mIntersection.set(intersection.getF32ptr());
+			mNormal.set(normal.getF32ptr());
+			mTangent.set(tangent.getF32ptr());
+
+			//extrapoloate binormal from normal and tangent
+			
+			LLVector4a binormal;
+			binormal.setCross3(normal, tangent);
+			binormal.mul(tangent.getF32ptr()[3]);
+
+			mBinormal.set(binormal.getF32ptr());
+
+			mBinormal.normalize();
+			mNormal.normalize();
+			mTangent.normalize();
+
 			// and XY coords:
 			updateXYCoords();
 			
--- a/indra/newview/llviewerwindow.h
+++ b/indra/newview/llviewerwindow.h
@@ -115,6 +115,7 @@ public:
 	LLVector2       mSTCoords;
 	LLCoordScreen	mXYCoords;
 	LLVector3		mNormal;
+	LLVector4		mTangent;
 	LLVector3		mBinormal;
 	BOOL			mPickTransparent;
 	void		    getSurfaceInfo();
@@ -347,19 +348,19 @@ public:
 	static void     hoverPickCallback(const LLPickInfo& pick_info);
 	
 	LLHUDIcon* cursorIntersectIcon(S32 mouse_x, S32 mouse_y, F32 depth,
-										   LLVector3* intersection);
+										   LLVector4a* intersection);

 	LLViewerObject* cursorIntersect(S32 mouse_x = -1, S32 mouse_y = -1, F32 depth = 512.f,
 									LLViewerObject *this_object = NULL,
 									S32 this_face = -1,
 									BOOL pick_transparent = FALSE,
 									S32* face_hit = NULL,
-									LLVector3 *intersection = NULL,
+									LLVector4a *intersection = NULL,
 									LLVector2 *uv = NULL,
-									LLVector3 *normal = NULL,
-									LLVector3 *binormal = NULL,
-									LLVector3* start = NULL,
-									LLVector3* end = NULL);
+									LLVector4a *normal = NULL,
+									LLVector4a *tangent = NULL,
+									LLVector4a* start = NULL,
+									LLVector4a* end = NULL);
 	
 	
 	// Returns a pointer to the last object hit
@@ -509,13 +510,13 @@ extern LLFrameTimer		gAwayTimer;				// tracks time before setting the avatar awa
 extern LLFrameTimer		gAwayTriggerTimer;		// how long the avatar has been away

 extern LLViewerObject*  gDebugRaycastObject;
-extern LLVector3        gDebugRaycastIntersection;
+extern LLVector4a       gDebugRaycastIntersection;
 extern LLVector2        gDebugRaycastTexCoord;
-extern LLVector3        gDebugRaycastNormal;
-extern LLVector3        gDebugRaycastBinormal;
+extern LLVector4a       gDebugRaycastNormal;
+extern LLVector4a       gDebugRaycastTangent;
 extern S32				gDebugRaycastFaceHit;
-extern LLVector3		gDebugRaycastStart;
-extern LLVector3		gDebugRaycastEnd;
+extern LLVector4a		gDebugRaycastStart;
+extern LLVector4a		gDebugRaycastEnd;

 extern BOOL			gDisplayCameraPos;
 extern BOOL			gDisplayWindInfo;
--- a/indra/newview/llvoavatar.cpp
+++ b/indra/newview/llvoavatar.cpp
@@ -1729,19 +1729,20 @@ void LLVOAvatar::renderCollisionVolumes()

 	if (mNameText.notNull())
 	{
-		LLVector3 unused;
-		mNameText->lineSegmentIntersect(LLVector3(0,0,0), LLVector3(0,0,1), unused, TRUE);
+		LLVector4a unused;
+	
+		mNameText->lineSegmentIntersect(unused, unused, unused, TRUE);
 	}
 }

-BOOL LLVOAvatar::lineSegmentIntersect(const LLVector3& start, const LLVector3& end,
+BOOL LLVOAvatar::lineSegmentIntersect(const LLVector4a& start, const LLVector4a& end,
 									  S32 face,
 									  BOOL pick_transparent,
 									  S32* face_hit,
-									  LLVector3* intersection,
+									  LLVector4a* intersection,
 									  LLVector2* tex_coord,
-									  LLVector3* normal,
-									  LLVector3* bi_normal)
+									  LLVector4a* normal,
+									  LLVector4a* tangent)
 {
 	if ((isSelf() && !gAgent.needsRenderAvatar()) || !LLPipeline::sPickAvatar)
 	{
@@ -1758,8 +1759,8 @@ BOOL LLVOAvatar::lineSegmentIntersect(const LLVector3& start, const LLVector3& e
 			glh::matrix4f inverse = mat.inverse();
 			glh::matrix4f norm_mat = inverse.transpose();

-			glh::vec3f p1(start.mV);
-			glh::vec3f p2(end.mV);
+			glh::vec3f p1(start.getF32ptr());
+			glh::vec3f p2(end.getF32ptr());

 			inverse.mult_matrix_vec(p1);
 			inverse.mult_matrix_vec(p2);
@@ -1778,12 +1779,12 @@ BOOL LLVOAvatar::lineSegmentIntersect(const LLVector3& start, const LLVector3& e

 				if (intersection)
 				{
-					*intersection = LLVector3(res_pos.v);
+					intersection->load3(res_pos.v);
 				}

 				if (normal)
 				{
-					*normal = LLVector3(res_norm.v);
+					normal->load3(res_norm.v);
 				}

 				return TRUE;
@@ -1817,7 +1818,8 @@ BOOL LLVOAvatar::lineSegmentIntersect(const LLVector3& start, const LLVector3& e
 		}
 	}
 	
-	LLVector3 position;
+	
+	LLVector4a position;
 	if (mNameText.notNull() && mNameText->lineSegmentIntersect(start, end, position))
 	{
 		if (intersection)
@@ -1831,14 +1833,14 @@ BOOL LLVOAvatar::lineSegmentIntersect(const LLVector3& start, const LLVector3& e
 	return FALSE;
 }

-LLViewerObject* LLVOAvatar::lineSegmentIntersectRiggedAttachments(const LLVector3& start, const LLVector3& end,
+LLViewerObject* LLVOAvatar::lineSegmentIntersectRiggedAttachments(const LLVector4a& start, const LLVector4a& end,
 									  S32 face,
 									  BOOL pick_transparent,
 									  S32* face_hit,
-									  LLVector3* intersection,
+									  LLVector4a* intersection,
 									  LLVector2* tex_coord,
-									  LLVector3* normal,
-									  LLVector3* bi_normal)
+									  LLVector4a* normal,
+									  LLVector4a* tangent)
 {
 	static const LLCachedControl<bool> allow_mesh_picking("SGAllowRiggedMeshSelection");

@@ -1851,8 +1853,8 @@ LLViewerObject* LLVOAvatar::lineSegmentIntersectRiggedAttachments(const LLVector

 	if (lineSegmentBoundingBox(start, end))
 	{
-		LLVector3 local_end = end;
-		LLVector3 local_intersection;
+		LLVector4a local_end = end;
+		LLVector4a local_intersection;

 		for (attachment_map_t::iterator iter = mAttachmentPoints.begin(); 
 			iter != mAttachmentPoints.end();
@@ -1866,7 +1868,7 @@ LLViewerObject* LLVOAvatar::lineSegmentIntersectRiggedAttachments(const LLVector
 			{
 				LLViewerObject* attached_object = (*attachment_iter);
 					
-				if (attached_object->lineSegmentIntersect(start, local_end, face, pick_transparent, face_hit, &local_intersection, tex_coord, normal, bi_normal))
+				if (attached_object->lineSegmentIntersect(start, local_end, face, pick_transparent, face_hit, &local_intersection, tex_coord, normal, tangent))
 				{
 					local_end = local_intersection;
 					if (intersection)
--- a/indra/newview/llvoavatar.h
+++ b/indra/newview/llvoavatar.h
@@ -201,22 +201,22 @@ public:
 	/*virtual*/ void   	 	 	updateRegion(LLViewerRegion *regionp);
 	/*virtual*/ void   	 	 	updateSpatialExtents(LLVector4a& newMin, LLVector4a &newMax);
 	/*virtual*/ void   	 	 	getSpatialExtents(LLVector4a& newMin, LLVector4a& newMax);
-	/*virtual*/ BOOL   	 	 	lineSegmentIntersect(const LLVector3& start, const LLVector3& end,
+	/*virtual*/ BOOL   	 	 	lineSegmentIntersect(const LLVector4a& start, const LLVector4a& end,
 												 S32 face = -1,                    // which face to check, -1 = ALL_SIDES
 												 BOOL pick_transparent = FALSE,
 												 S32* face_hit = NULL,             // which face was hit
-												 LLVector3* intersection = NULL,   // return the intersection point
+												 LLVector4a* intersection = NULL,   // return the intersection point
 												 LLVector2* tex_coord = NULL,      // return the texture coordinates of the intersection point
-												 LLVector3* normal = NULL,         // return the surface normal at the intersection point
-												 LLVector3* bi_normal = NULL);     // return the surface bi-normal at the intersection point
-	LLViewerObject*	lineSegmentIntersectRiggedAttachments(const LLVector3& start, const LLVector3& end,
+												 LLVector4a* normal = NULL,         // return the surface normal at the intersection point
+												 LLVector4a* tangent = NULL);     // return the surface tangent at the intersection point
+	LLViewerObject*	lineSegmentIntersectRiggedAttachments(const LLVector4a& start, const LLVector4a& end,
 												 S32 face = -1,                    // which face to check, -1 = ALL_SIDES
 												 BOOL pick_transparent = FALSE,
 												 S32* face_hit = NULL,             // which face was hit
-												 LLVector3* intersection = NULL,   // return the intersection point
+												 LLVector4a* intersection = NULL,   // return the intersection point
 												 LLVector2* tex_coord = NULL,      // return the texture coordinates of the intersection point
-												 LLVector3* normal = NULL,         // return the surface normal at the intersection point
-												 LLVector3* bi_normal = NULL);     // return the surface bi-normal at the intersection point
+												 LLVector4a* normal = NULL,         // return the surface normal at the intersection point
+												 LLVector4a* tangent = NULL);     // return the surface tangent at the intersection point

 	//--------------------------------------------------------------------
 	// LLCharacter interface and related
--- a/indra/newview/llvograss.cpp
+++ b/indra/newview/llvograss.cpp
@@ -789,8 +789,8 @@ void LLVOGrass::updateDrawable(BOOL force_damped)
 }

 // virtual 
-BOOL LLVOGrass::lineSegmentIntersect(const LLVector3& start, const LLVector3& end, S32 face, BOOL pick_transparent, S32 *face_hitp,
-									  LLVector3* intersection,LLVector2* tex_coord, LLVector3* normal, LLVector3* bi_normal)
+BOOL LLVOGrass::lineSegmentIntersect(const LLVector4a& start, const LLVector4a& end, S32 face, BOOL pick_transparent, S32 *face_hitp,
+									  LLVector4a* intersection,LLVector2* tex_coord, LLVector4a* normal, LLVector4a* tangent)
 	
 {
 	BOOL ret = FALSE;
@@ -801,7 +801,8 @@ BOOL LLVOGrass::lineSegmentIntersect(const LLVector3& start, const LLVector3& en
 		return FALSE;
 	}

-	LLVector3 dir = end-start;
+	LLVector4a dir;
+	dir.setSub(end, start);

 	mPatch = mRegionp->getLand().resolvePatchRegion(getPositionRegion());
 	
@@ -869,23 +870,31 @@ 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))
+		LLVector4a v0a,v1a,v2a,v3a;
+
+		v0a.load3(v[0].mV);
+		v1a.load3(v[1].mV);
+		v2a.load3(v[2].mV);
+		v3a.load3(v[3].mV);
+
+		
+		if (LLTriangleRayIntersect(v0a, v1a, v2a, start, dir, a, b, t))
 		{
 			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(v1a, v3a, v2a, start, dir, a, b, t))
 		{
 			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(v2a, v1a, v0a, start, dir, a, b, t))
 		{
 			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(v2a, v3a, v1a, start, dir, a, b, t))
 		{
 			normal1 = -normal1;
 			hit = TRUE;
@@ -908,7 +917,8 @@ BOOL LLVOGrass::lineSegmentIntersect(const LLVector3& start, const LLVector3& en
 					closest_t = t;
 					if (intersection != NULL)
 					{
-						*intersection = start+dir*closest_t;
+						dir.mul(closest_t);
+						intersection->setAdd(start, dir);
 					}

 					if (tex_coord != NULL)
@@ -918,7 +928,7 @@ BOOL LLVOGrass::lineSegmentIntersect(const LLVector3& start, const LLVector3& en

 					if (normal != NULL)
 					{
-						*normal    = normal1;
+						normal->load3(normal1.mV);
 					}
 					ret = TRUE;
 				}
--- a/indra/newview/llvograss.h
+++ b/indra/newview/llvograss.h
@@ -81,14 +81,14 @@ public:
 	/*virtual*/ BOOL    isActive() const; // Whether this object needs to do an idleUpdate.
 	/*virtual*/ void idleUpdate(LLAgent &agent, LLWorld &world, const F64 &time);

-	/*virtual*/ BOOL lineSegmentIntersect(const LLVector3& start, const LLVector3& end, 
+	/*virtual*/ BOOL lineSegmentIntersect(const LLVector4a& start, const LLVector4a& end, 
 										  S32 face = -1,                        // which face to check, -1 = ALL_SIDES
 										  BOOL pick_transparent = FALSE,
 										  S32* face_hit = NULL,                 // which face was hit
-										  LLVector3* intersection = NULL,       // return the intersection point
+										  LLVector4a* intersection = NULL,       // return the intersection point
 										  LLVector2* tex_coord = NULL,          // return the texture coordinates of the intersection point
-										  LLVector3* normal = NULL,             // return the surface normal at the intersection point
-										  LLVector3* bi_normal = NULL           // return the surface bi-normal at the intersection point
+										  LLVector4a* normal = NULL,             // return the surface normal at the intersection point
+										  LLVector4a* tangent = NULL           // return the surface tangent at the intersection point
 		);

 	static S32 sMaxGrassSpecies;
--- a/indra/newview/llvosurfacepatch.cpp
+++ b/indra/newview/llvosurfacepatch.cpp
@@ -103,10 +103,10 @@ public:
 			glTexCoordPointer(2,GL_FLOAT, LLVertexBuffer::sTypeSize[TYPE_TEXCOORD1], (void*)(base + mOffsets[TYPE_TEXCOORD1]));
 			glClientActiveTextureARB(GL_TEXTURE0_ARB);
 		}
-		if (data_mask & MAP_BINORMAL)
+		if (data_mask & MAP_TANGENT)
 		{
 			glClientActiveTextureARB(GL_TEXTURE2_ARB);
-			glTexCoordPointer(3,GL_FLOAT, LLVertexBuffer::sTypeSize[TYPE_BINORMAL], (void*)(base + mOffsets[TYPE_BINORMAL]));
+			glTexCoordPointer(3,GL_FLOAT, LLVertexBuffer::sTypeSize[TYPE_TANGENT], (void*)(base + mOffsets[TYPE_TANGENT]));
 			glClientActiveTextureARB(GL_TEXTURE0_ARB);
 		}
 		if (data_mask & MAP_TEXCOORD0)
@@ -942,8 +942,8 @@ void LLVOSurfacePatch::getGeomSizesEast(const S32 stride, const S32 east_stride,
 	}
 }

-BOOL LLVOSurfacePatch::lineSegmentIntersect(const LLVector3& start, const LLVector3& end, S32 face, BOOL pick_transparent, S32 *face_hitp,
-									  LLVector3* intersection,LLVector2* tex_coord, LLVector3* normal, LLVector3* bi_normal)
+BOOL LLVOSurfacePatch::lineSegmentIntersect(const LLVector4a& start, const LLVector4a& end, S32 face, BOOL pick_transparent, S32 *face_hitp,
+									  LLVector4a* intersection,LLVector2* tex_coord, LLVector4a* normal, LLVector4a* tangent)
 	
 {

@@ -952,7 +952,9 @@ BOOL LLVOSurfacePatch::lineSegmentIntersect(const LLVector3& start, const LLVect
 		return FALSE;
 	}

-	LLVector3 delta = end-start;
+	LLVector4a da;
+	da.setSub(end, start);
+	LLVector3 delta(da.getF32ptr());
 		
 	LLVector3 pdelta = delta;
 	pdelta.mV[2] = 0;
@@ -961,7 +963,9 @@ BOOL LLVOSurfacePatch::lineSegmentIntersect(const LLVector3& start, const LLVect
 	
 	F32 tdelta = 1.f/plength;

-	LLVector3 origin = start - mRegionp->getOriginAgent();
+	LLVector3 v_start(start.getF32ptr());
+
+	LLVector3 origin = v_start - mRegionp->getOriginAgent();

 	if (mRegionp->getLandHeightRegion(origin) > origin.mV[2])
 	{
@@ -1016,12 +1020,12 @@ BOOL LLVOSurfacePatch::lineSegmentIntersect(const LLVector3& start, const LLVect
 					{
 						sample.mV[2] = mRegionp->getLandHeightRegion(sample);
 					}
-					*intersection = sample + mRegionp->getOriginAgent();
+					intersection->load3((sample + mRegionp->getOriginAgent()).mV);
 				}

 				if (normal)
 				{
-					*normal = mRegionp->getLand().resolveNormalGlobal(mRegionp->getPosGlobalFromRegion(sample));
+					normal->load3((mRegionp->getLand().resolveNormalGlobal(mRegionp->getPosGlobalFromRegion(sample))).mV);
 				}

 				return TRUE;
--- a/indra/newview/llvosurfacepatch.h
+++ b/indra/newview/llvosurfacepatch.h
@@ -85,14 +85,14 @@ public:
 	void dirtyPatch();
 	void dirtyGeom();

-	/*virtual*/ BOOL lineSegmentIntersect(const LLVector3& start, const LLVector3& end, 
+	/*virtual*/ BOOL lineSegmentIntersect(const LLVector4a& start, const LLVector4a& end, 
 										  S32 face = -1,                        // which face to check, -1 = ALL_SIDES
 										  BOOL pick_transparent = FALSE,
 										  S32* face_hit = NULL,                 // which face was hit
-										  LLVector3* intersection = NULL,       // return the intersection point
+										  LLVector4a* intersection = NULL,       // return the intersection point
 										  LLVector2* tex_coord = NULL,          // return the texture coordinates of the intersection point
-										  LLVector3* normal = NULL,             // return the surface normal at the intersection point
-										  LLVector3* bi_normal = NULL           // return the surface bi-normal at the intersection point
+										  LLVector4a* normal = NULL,             // return the surface normal at the intersection point
+										  LLVector4a* tangent = NULL           // return the surface tangent at the intersection point
 		);

 	BOOL			mDirtiedPatch;
--- a/indra/newview/llvotree.cpp
+++ b/indra/newview/llvotree.cpp
@@ -1253,8 +1253,8 @@ void LLVOTree::updateSpatialExtents(LLVector4a& newMin, LLVector4a& newMax)
 	mDrawable->setPositionGroup(pos);
 }

-BOOL LLVOTree::lineSegmentIntersect(const LLVector3& start, const LLVector3& end, S32 face, BOOL pick_transparent, S32 *face_hitp,
-									  LLVector3* intersection,LLVector2* tex_coord, LLVector3* normal, LLVector3* bi_normal)
+BOOL LLVOTree::lineSegmentIntersect(const LLVector4a& start, const LLVector4a& end, S32 face, BOOL pick_transparent, S32 *face_hitp,
+									  LLVector4a* intersection,LLVector2* tex_coord, LLVector4a* normal, LLVector4a* tangent)
 	
 {

@@ -1283,16 +1283,19 @@ BOOL LLVOTree::lineSegmentIntersect(const LLVector3& start, const LLVector3& end

 	LLVector3 pos, norm;
 		
-	if (linesegment_tetrahedron(start, end, center, size, quat, pos, norm))
+	LLVector3 start3(start.getF32ptr());
+	LLVector3 end3(end.getF32ptr());
+
+	if (linesegment_tetrahedron(start3, end3, center, size, quat, pos, norm))
 	{
 		if (intersection)
 		{
-			*intersection = pos;
+			intersection->load3(pos.mV);
 		}

 		if (normal)
 		{
-			*normal = norm;
+			normal->load3(norm.mV);
 		}
 		return TRUE;
 	}
--- a/indra/newview/llvotree.h
+++ b/indra/newview/llvotree.h
@@ -115,14 +115,14 @@ public:
 	U32 drawBranchPipeline(LLMatrix4& matrix, U16* indicesp, S32 trunk_LOD, S32 stop_level, U16 depth, U16 trunk_depth,  F32 scale, F32 twist, F32 droop,  F32 branches, F32 alpha);
 

-	 /*virtual*/ BOOL lineSegmentIntersect(const LLVector3& start, const LLVector3& end, 
+	 /*virtual*/ BOOL lineSegmentIntersect(const LLVector4a& start, const LLVector4a& end, 
 										  S32 face = -1,                        // which face to check, -1 = ALL_SIDES
 										  BOOL pick_transparent = FALSE,
 										  S32* face_hit = NULL,                 // which face was hit
-										  LLVector3* intersection = NULL,       // return the intersection point
+										  LLVector4a* intersection = NULL,       // return the intersection point
 										  LLVector2* tex_coord = NULL,          // return the texture coordinates of the intersection point
-										  LLVector3* normal = NULL,             // return the surface normal at the intersection point
-										  LLVector3* bi_normal = NULL           // return the surface bi-normal at the intersection point
+										  LLVector4a* normal = NULL,             // return the surface normal at the intersection point
+										  LLVector4a* tangent = NULL           // return the surface tangent at the intersection point
 		);

 	static S32 sMaxTreeSpecies;
--- a/indra/newview/llvovolume.cpp
+++ b/indra/newview/llvovolume.cpp
@@ -1053,13 +1053,17 @@ BOOL LLVOVolume::setVolume(const LLVolumeParams &params_in, const S32 detail, bo
 				{ //already cached
 					break;
 				}
-				volume->genBinormals(i);
+				volume->genTangents(i);
 				LLFace::cacheFaceInVRAM(face);
 			}
 		}
 		
+
 		return TRUE;
 	}
+
+
+
 	return FALSE;
 }

@@ -1218,7 +1222,7 @@ BOOL LLVOVolume::calcLOD()
 	else
 	{
 		distance = mDrawable->mDistanceWRTCamera;
-		radius = getVolume()->mLODScaleBias.scaledVec(getScale()).length();
+		radius = getVolume() ? getVolume()->mLODScaleBias.scaledVec(getScale()).length() : getScale().length();
 	}
 	

@@ -3589,8 +3593,8 @@ LLVector3 LLVOVolume::volumeDirectionToAgent(const LLVector3& dir) const
 }


-BOOL LLVOVolume::lineSegmentIntersect(const LLVector3& start, const LLVector3& end, S32 face, BOOL pick_transparent, S32 *face_hitp,
-									  LLVector3* intersection,LLVector2* tex_coord, LLVector3* normal, LLVector3* bi_normal)
+BOOL LLVOVolume::lineSegmentIntersect(const LLVector4a& start, const LLVector4a& end, S32 face, BOOL pick_transparent, S32 *face_hitp,
+									  LLVector4a* intersection,LLVector2* tex_coord, LLVector4a* normal, LLVector4a* tangent)
 	
 {
 	if (!mbCanSelect ||
@@ -3631,23 +3635,25 @@ BOOL LLVOVolume::lineSegmentIntersect(const LLVector3& start, const LLVector3& e
 	
 	if (volume)
 	{	
-		LLVector3 v_start, v_end, v_dir;
-	
+		LLVector4a local_start = start;
+		LLVector4a local_end = end;
+		
 		if (transform)
 		{
-			v_start = agentPositionToVolume(start);
-			v_end = agentPositionToVolume(end);
-		}
-		else
-		{
-			v_start = start;
-			v_end = end;
-		}
+			LLVector3 v_start(start.getF32ptr());
+			LLVector3 v_end(end.getF32ptr());
 		
-		LLVector3 p;
-		LLVector3 n;
+			v_start = agentPositionToVolume(v_start);
+			v_end = agentPositionToVolume(v_end);
+
+			local_start.load3(v_start.mV);
+			local_end.load3(v_end.mV);
+		}
+				
+		LLVector4a p;
+		LLVector4a n;
 		LLVector2 tc;
-		LLVector3 bn;
+		LLVector4a tn;

 		if (intersection != NULL)
 		{
@@ -3664,9 +3670,9 @@ BOOL LLVOVolume::lineSegmentIntersect(const LLVector3& start, const LLVector3& e
 			n = *normal;
 		}

-		if (bi_normal != NULL)
+		if (tangent != NULL)
 		{
-			bn = *bi_normal;
+			tn = *tangent;
 		}

 		S32 face_hit = -1;
@@ -3692,8 +3698,8 @@ BOOL LLVOVolume::lineSegmentIntersect(const LLVector3& start, const LLVector3& e
 				continue;
 			}

-			face_hit = volume->lineSegmentIntersect(v_start, v_end, i,
-													&p, &tc, &n, &bn);
+			face_hit = volume->lineSegmentIntersect(local_start, local_end, i,
+													&p, &tc, &n, &tn);
 			
 			if (face_hit >= 0 && mDrawable->getNumFaces() > face_hit)
 			{
@@ -3702,7 +3708,7 @@ BOOL LLVOVolume::lineSegmentIntersect(const LLVector3& start, const LLVector3& e
 				if (face &&
 					(pick_transparent || !face->getTexture() || !face->getTexture()->hasGLTexture() || face->getTexture()->getMask(face->surfaceToTexture(tc, p, n))))
 				{
-					v_end = p;
+					local_end = p;
 					if (face_hitp != NULL)
 					{
 						*face_hitp = face_hit;
@@ -3712,7 +3718,9 @@ BOOL LLVOVolume::lineSegmentIntersect(const LLVector3& start, const LLVector3& e
 					{
 						if (transform)
 						{
-							*intersection = volumePositionToAgent(p);  // must map back to agent space
+							LLVector3 v_p(p.getF32ptr());
+
+							intersection->load3(volumePositionToAgent(v_p).mV);  // must map back to agent space
 						}
 						else
 						{
@@ -3724,27 +3732,36 @@ BOOL LLVOVolume::lineSegmentIntersect(const LLVector3& start, const LLVector3& e
 					{
 						if (transform)
 						{
-							*normal = volumeDirectionToAgent(n);
+							LLVector3 v_n(n.getF32ptr());
+							normal->load3(volumeDirectionToAgent(v_n).mV);
 						}
 						else
 						{
 							*normal = n;
 						}
-
-						(*normal).normVec();
+						(*normal).normalize3fast();
 					}

-					if (bi_normal != NULL)
+					if (tangent != NULL)
 					{
 						if (transform)
 						{
-							*bi_normal = volumeDirectionToAgent(bn);
+							LLVector3 v_tn(tn.getF32ptr());
+
+							LLVector4a trans_tangent;
+							trans_tangent.load3(volumeDirectionToAgent(v_tn).mV);
+
+							LLVector4Logical mask;
+							mask.clear();
+							mask.setElement<3>();
+
+							tangent->setSelectWithMask(mask, tn, trans_tangent);
 						}
 						else
 						{
-							*bi_normal = bn;
+							*tangent = tn;
 						}
-						(*bi_normal).normVec();
+						(*tangent).normalize3fast();
 					}

 					if (tex_coord != NULL)
@@ -4262,11 +4279,18 @@ void LLVolumeGeometryManager::rebuildGeom(LLSpatialGroup* group)

 	mFaceList.clear();

-	std::vector<LLFace*> fullbright_faces;
-	std::vector<LLFace*> bump_faces;
-	std::vector<LLFace*> simple_faces;
+	const U32 MAX_FACE_COUNT = 4096;
+	
+	static LLFace** fullbright_faces = (LLFace**) ll_aligned_malloc(MAX_FACE_COUNT*sizeof(LLFace*),64);
+	static LLFace** bump_faces = (LLFace**) ll_aligned_malloc(MAX_FACE_COUNT*sizeof(LLFace*),64);
+	static LLFace** simple_faces = (LLFace**) ll_aligned_malloc(MAX_FACE_COUNT*sizeof(LLFace*),64);
+	static LLFace** alpha_faces = (LLFace**) ll_aligned_malloc(MAX_FACE_COUNT*sizeof(LLFace*),64);
+	
+	U32 fullbright_count = 0;
+	U32 bump_count = 0;
+	U32 simple_count = 0;
+	U32 alpha_count = 0;

-	std::vector<LLFace*> alpha_faces;
 	U32 useage = group->mSpatialPartition->mBufferUsage;

 	static const LLCachedControl<S32> render_max_vbo_size("RenderMaxVBOSize", 512);
@@ -4623,7 +4647,10 @@ void LLVolumeGeometryManager::rebuildGeom(LLSpatialGroup* group)
 					{
 						if (facep->canRenderAsMask())
 						{ //can be treated as alpha mask
-							simple_faces.push_back(facep);
+							if (simple_count < MAX_FACE_COUNT)
+							{
+								simple_faces[simple_count++] = facep;
+							}
 						}
 						else
 						{
@@ -4631,7 +4658,10 @@ void LLVolumeGeometryManager::rebuildGeom(LLSpatialGroup* group)
 							{ //only treat as alpha in the pipeline if < 100% transparent
 								drawablep->setState(LLDrawable::HAS_ALPHA);
 							}
-							alpha_faces.push_back(facep);
+							if (alpha_count < MAX_FACE_COUNT)
+							{
+								alpha_faces[alpha_count++] = facep;
+							}
 						}
 					}
 					else
@@ -4645,34 +4675,52 @@ void LLVolumeGeometryManager::rebuildGeom(LLSpatialGroup* group)
 							&& LLPipeline::sRenderBump)
 						{
 							if (te->getBumpmap())
-							{ //needs normal + binormal
-								bump_faces.push_back(facep);
+							{ //needs normal + tangent
+								if (bump_count < MAX_FACE_COUNT)
+								{
+									bump_faces[bump_count++] = facep;
+								}
 							}
 							else if (te->getShiny() || !te->getFullbright())
 							{ //needs normal
-								simple_faces.push_back(facep);
+								if (simple_count < MAX_FACE_COUNT)
+								{
+									simple_faces[simple_count++] = facep;
+								}
 							}
 							else 
 							{ //doesn't need normal
 								facep->setState(LLFace::FULLBRIGHT);
-								fullbright_faces.push_back(facep);
+								if (fullbright_count < MAX_FACE_COUNT)
+								{
+									fullbright_faces[fullbright_count++] = facep;
+								}
 							}
 						}
 						else
 						{
 							if (te->getBumpmap() && LLPipeline::sRenderBump)
-							{ //needs normal + binormal
-								bump_faces.push_back(facep);
+							{ //needs normal + tangent
+								if (bump_count < MAX_FACE_COUNT)
+								{
+									bump_faces[bump_count++] = facep;
+								}
 							}
 							else if ((te->getShiny() && LLPipeline::sRenderBump) ||
 								!(te->getFullbright() || bake_sunlight))
 							{ //needs normal
-								simple_faces.push_back(facep);
+								if (simple_count < MAX_FACE_COUNT)
+								{
+									simple_faces[simple_count++] = facep;
+								}
 							}
 							else 
 							{ //doesn't need normal
 								facep->setState(LLFace::FULLBRIGHT);
-								fullbright_faces.push_back(facep);
+								if (fullbright_count < MAX_FACE_COUNT)
+								{
+									fullbright_faces[fullbright_count++] = facep;
+								}
 							}
 						}
 					}
@@ -4714,18 +4762,18 @@ void LLVolumeGeometryManager::rebuildGeom(LLSpatialGroup* group)

 	if (batch_textures)
 	{
-		bump_mask |= LLVertexBuffer::MAP_BINORMAL;
-		genDrawInfo(group, simple_mask | LLVertexBuffer::MAP_TEXTURE_INDEX, simple_faces, FALSE, TRUE);
-		genDrawInfo(group, fullbright_mask | LLVertexBuffer::MAP_TEXTURE_INDEX, fullbright_faces, FALSE, TRUE);
-		genDrawInfo(group, bump_mask | LLVertexBuffer::MAP_TEXTURE_INDEX, bump_faces, FALSE, TRUE);
-		genDrawInfo(group, alpha_mask | LLVertexBuffer::MAP_TEXTURE_INDEX, alpha_faces, TRUE, TRUE);
+		bump_mask |= LLVertexBuffer::MAP_TANGENT;
+		genDrawInfo(group, simple_mask | LLVertexBuffer::MAP_TEXTURE_INDEX, simple_faces, simple_count, FALSE, TRUE);
+		genDrawInfo(group, fullbright_mask | LLVertexBuffer::MAP_TEXTURE_INDEX, fullbright_faces, fullbright_count, FALSE, TRUE);
+		genDrawInfo(group, bump_mask | LLVertexBuffer::MAP_TEXTURE_INDEX, bump_faces, bump_count, FALSE, TRUE);
+		genDrawInfo(group, alpha_mask | LLVertexBuffer::MAP_TEXTURE_INDEX, alpha_faces, alpha_count, TRUE, TRUE);
 	}
 	else
 	{
-		genDrawInfo(group, simple_mask, simple_faces);
-		genDrawInfo(group, fullbright_mask, fullbright_faces);
-		genDrawInfo(group, bump_mask, bump_faces, FALSE, TRUE);
-		genDrawInfo(group, alpha_mask, alpha_faces, TRUE);
+		genDrawInfo(group, simple_mask, simple_faces, simple_count);
+		genDrawInfo(group, fullbright_mask, fullbright_faces, fullbright_count);
+		genDrawInfo(group, bump_mask, bump_faces, bump_count, FALSE, TRUE);
+		genDrawInfo(group, alpha_mask, alpha_faces, alpha_count, TRUE);
 	}
 	

@@ -4767,13 +4815,17 @@ void LLVolumeGeometryManager::rebuildMesh(LLSpatialGroup* group)
 	{
 		LLFastTimer ftm(FTM_REBUILD_VOLUME_VB);
 		LLFastTimer t(FTM_REBUILD_VOLUME_GEN_DRAW_INFO); //make sure getgeometryvolume shows up in the right place in timers
-		S32 num_mapped_vertex_buffer = LLVertexBuffer::sMappedCount ;

 		group->mBuilt = 1.f;
 		
-		std::set<LLVertexBuffer*> mapped_buffers;
-
 		OctreeGuard guard(group->mOctreeNode);
+		S32 num_mapped_vertex_buffer = LLVertexBuffer::sMappedCount ;
+
+		const U32 MAX_BUFFER_COUNT = 4096;
+		static LLVertexBuffer* locked_buffer[MAX_BUFFER_COUNT];
+		
+		U32 buffer_count = 0;
+
 		for (LLSpatialGroup::element_iter drawable_iter = group->getDataBegin(); drawable_iter != group->getDataEnd(); ++drawable_iter)
 		{
 			LLDrawable* drawablep = *drawable_iter;
@@ -4807,9 +4859,9 @@ void LLVolumeGeometryManager::rebuildMesh(LLSpatialGroup* group)
 							}


-							if (buff->isLocked())
+							if (buff->isLocked() && buffer_count < MAX_BUFFER_COUNT)
 							{
-								mapped_buffers.insert(buff);
+								locked_buffer[buffer_count++] = buff;
 							}
 						}
 					}
@@ -4825,7 +4877,7 @@ void LLVolumeGeometryManager::rebuildMesh(LLSpatialGroup* group)
 			}
 		}
 		
-		for (std::set<LLVertexBuffer*>::iterator iter = mapped_buffers.begin(); iter != mapped_buffers.end(); ++iter)
+		for (LLVertexBuffer** iter = locked_buffer, ** end_iter = locked_buffer+buffer_count; iter != end_iter; ++iter)
 		{
 			(*iter)->flush();
 		}
@@ -4904,7 +4956,7 @@ static LLFastTimer::DeclareTimer FTM_GEN_DRAW_INFO_RESIZE_VB("Resize VB");



-void LLVolumeGeometryManager::genDrawInfo(LLSpatialGroup* group, U32 mask, std::vector<LLFace*>& faces, BOOL distance_sort, BOOL batch_textures)
+void LLVolumeGeometryManager::genDrawInfo(LLSpatialGroup* group, U32 mask, LLFace** faces, U32 face_count, BOOL distance_sort, BOOL batch_textures, BOOL no_materials)
 {
 	LLFastTimer t(FTM_REBUILD_VOLUME_GEN_DRAW_INFO);

@@ -4931,17 +4983,18 @@ void LLVolumeGeometryManager::genDrawInfo(LLSpatialGroup* group, U32 mask, std::
 		if (!distance_sort)
 		{
 			//sort faces by things that break batches
-			std::sort(faces.begin(), faces.end(), CompareBatchBreakerModified());
+			std::sort(faces, faces+face_count, CompareBatchBreakerModified());
 		}
 		else
 		{
 			//sort faces by distance
-			std::sort(faces.begin(), faces.end(), LLFace::CompareDistanceGreater());
+			std::sort(faces, faces+face_count, LLFace::CompareDistanceGreater());
 		}
 	}
 				
 	bool hud_group = group->isHUDGroup() ;
-	std::vector<LLFace*>::iterator face_iter = faces.begin();
+	LLFace** face_iter = faces;
+	LLFace** end_faces = faces+face_count;
 	
 	LLSpatialGroup::buffer_map_t buffer_map;

@@ -4973,7 +5026,7 @@ void LLVolumeGeometryManager::genDrawInfo(LLSpatialGroup* group, U32 mask, std::
 	//NEVER use more than 16 texture index channels (workaround for prevalent driver bug)
 	texture_index_channels = llmin(texture_index_channels, 16);

-	while (face_iter != faces.end())
+	while (face_iter != end_faces)
 	{
 		//pull off next face
 		LLFace* facep = *face_iter;
@@ -4999,11 +5052,15 @@ void LLVolumeGeometryManager::genDrawInfo(LLSpatialGroup* group, U32 mask, std::
 		U32 index_count = facep->getIndicesCount();
 		U32 geom_count = facep->getGeomCount();

+
 		//sum up vertices needed for this render batch
-		std::vector<LLFace*>::iterator i = face_iter;
+		LLFace** i = face_iter;
 		++i;
 		
-		std::vector<LLViewerTexture*> texture_list;
+		const U32 MAX_TEXTURE_COUNT = 32;
+		static LLViewerTexture* texture_list[MAX_TEXTURE_COUNT];
+		
+		U32 texture_count = 0;

 		{
 			LLFastTimer t(FTM_GEN_DRAW_INFO_FACE_SIZE);
@@ -5011,11 +5068,15 @@ void LLVolumeGeometryManager::genDrawInfo(LLSpatialGroup* group, U32 mask, std::
 			{
 				U8 cur_tex = 0;
 				facep->setTextureIndex(cur_tex);
-				texture_list.push_back(tex);
+				if (texture_count < MAX_TEXTURE_COUNT)
+				{
+					texture_list[texture_count++] = tex;
+				}

 				if (can_batch_texture(facep))
-				{
-					while (i != faces.end())
+				{ //populate texture_list with any textures that can be batched
+				  //move i to the next unbatchable face
+					while (i != end_faces)
 					{
 						facep = *i;
 						if (!can_batch_texture(facep))
@@ -5028,7 +5089,7 @@ void LLVolumeGeometryManager::genDrawInfo(LLSpatialGroup* group, U32 mask, std::
 							if (distance_sort)
 							{ //textures might be out of order, see if texture exists in current batch
 								bool found = false;
-								for (U32 tex_idx = 0; tex_idx < texture_list.size(); ++tex_idx)
+								for (U32 tex_idx = 0; tex_idx < texture_count; ++tex_idx)
 								{
 									if (facep->getTexture() == texture_list[tex_idx])
 									{
@@ -5040,7 +5101,7 @@ void LLVolumeGeometryManager::genDrawInfo(LLSpatialGroup* group, U32 mask, std::

 								if (!found)
 								{
-									cur_tex = texture_list.size();
+									cur_tex = texture_count;
 								}
 							}
 							else
@@ -5055,7 +5116,10 @@ void LLVolumeGeometryManager::genDrawInfo(LLSpatialGroup* group, U32 mask, std::

 							tex = facep->getTexture();

-							texture_list.push_back(tex);
+							if (texture_count < MAX_TEXTURE_COUNT)
+							{
+								texture_list[texture_count++] = tex;
+							}
 						}

 						if (geom_count + facep->getGeomCount() > max_vertices)
@@ -5074,7 +5138,7 @@ void LLVolumeGeometryManager::genDrawInfo(LLSpatialGroup* group, U32 mask, std::
 			}
 			else
 			{
-				while (i != faces.end() && 
+				while (i != end_faces && 
 					(LLPipeline::sTextureBindTest || (distance_sort || (*i)->getTexture() == tex)))
 				{
 					facep = *i;
--- a/indra/newview/llvovolume.h
+++ b/indra/newview/llvovolume.h
@@ -142,14 +142,14 @@ public:

 	/*virtual*/ U32		getTriangleCount(S32* vcount = NULL) const;
 	/*virtual*/ U32		getHighLODTriangleCount();
-	/*virtual*/ BOOL lineSegmentIntersect(const LLVector3& start, const LLVector3& end, 
+	/*virtual*/ BOOL lineSegmentIntersect(const LLVector4a& start, const LLVector4a& end, 
 										  S32 face = -1,                        // which face to check, -1 = ALL_SIDES
 										  BOOL pick_transparent = FALSE,
 										  S32* face_hit = NULL,                 // which face was hit
-										  LLVector3* intersection = NULL,       // return the intersection point
+										  LLVector4a* intersection = NULL,       // return the intersection point
 										  LLVector2* tex_coord = NULL,          // return the texture coordinates of the intersection point
-										  LLVector3* normal = NULL,             // return the surface normal at the intersection point
-										  LLVector3* bi_normal = NULL           // return the surface bi-normal at the intersection point
+										  LLVector4a* normal = NULL,             // return the surface normal at the intersection point
+										  LLVector4a* tangent = NULL           // return the surface tangent at the intersection point
 		);
 	
 				LLVector3 agentPositionToVolume(const LLVector3& pos) const;
--- a/indra/newview/pipeline.cpp
+++ b/indra/newview/pipeline.cpp
@@ -6167,20 +6167,20 @@ BOOL LLPipeline::getRenderHighlights(void*)
 	return sRenderHighlight;
 }

-LLViewerObject* LLPipeline::lineSegmentIntersectInWorld(const LLVector3& start, const LLVector3& end,
+LLViewerObject* LLPipeline::lineSegmentIntersectInWorld(const LLVector4a& start, const LLVector4a& end,
 														BOOL pick_transparent,												
 														S32* face_hit,
-														LLVector3* intersection,         // return the intersection point
+														LLVector4a* intersection,         // return the intersection point
 														LLVector2* tex_coord,            // return the texture coordinates of the intersection point
-														LLVector3* normal,               // return the surface normal at the intersection point
-														LLVector3* bi_normal             // return the surface bi-normal at the intersection point
+														LLVector4a* normal,               // return the surface normal at the intersection point
+														LLVector4a* tangent             // return the surface tangent at the intersection point
 	)
 {
 	LLDrawable* drawable = NULL;

-	LLVector3 local_end = end;
+	LLVector4a local_end = end;

-	LLVector3 position;
+	LLVector4a position;

 	sPickAvatar = FALSE; //LLToolMgr::getInstance()->inBuildMode() ? FALSE : TRUE;
 	
@@ -6200,7 +6200,7 @@ LLViewerObject* LLPipeline::lineSegmentIntersectInWorld(const LLVector3& start,
 				LLSpatialPartition* part = region->getSpatialPartition(j);
 				if (part && hasRenderType(part->mDrawableType))
 				{
-					LLDrawable* hit = part->lineSegmentIntersect(start, local_end, pick_transparent, face_hit, &position, tex_coord, normal, bi_normal);
+					LLDrawable* hit = part->lineSegmentIntersect(start, local_end, pick_transparent, face_hit, &position, tex_coord, normal, tangent);
 					if (hit)
 					{
 						drawable = hit;
@@ -6215,8 +6215,8 @@ LLViewerObject* LLPipeline::lineSegmentIntersectInWorld(const LLVector3& start,
 	{
 		//save hit info in case we need to restore
 		//due to attachment override
-		LLVector3 local_normal;
-		LLVector3 local_binormal;
+		LLVector4a local_normal;
+		LLVector4a local_tangent;
 		LLVector2 local_texcoord;
 		S32 local_face_hit = -1;

@@ -6228,14 +6228,22 @@ LLViewerObject* LLPipeline::lineSegmentIntersectInWorld(const LLVector3& start,
 		{
 			local_texcoord = *tex_coord;
 		}
-		if (bi_normal)
+		if (tangent)
 		{
-			local_binormal = *bi_normal;
+			local_tangent = *tangent;
+		}
+		else
+		{
+			local_tangent.clear();
 		}
 		if (normal)
 		{
 			local_normal = *normal;
 		}
+		else
+		{
+			local_normal.clear();
+		}
 				
 		const F32 ATTACHMENT_OVERRIDE_DIST = 0.1f;

@@ -6249,12 +6257,15 @@ LLViewerObject* LLPipeline::lineSegmentIntersectInWorld(const LLVector3& start,
 			LLSpatialPartition* part = region->getSpatialPartition(LLViewerRegion::PARTITION_BRIDGE);
 			if (part && hasRenderType(part->mDrawableType))
 			{
-				LLDrawable* hit = part->lineSegmentIntersect(start, local_end, pick_transparent, face_hit, &position, tex_coord, normal, bi_normal);
+				LLDrawable* hit = part->lineSegmentIntersect(start, local_end, pick_transparent, face_hit, &position, tex_coord, normal, tangent);
 				if (hit)
 				{
+					LLVector4a delta;
+					delta.setSub(position, local_end);
+
 					if (!drawable || 
 						!drawable->getVObj()->isAttachment() ||
-						(position-local_end).magVec() > ATTACHMENT_OVERRIDE_DIST)
+						delta.getLength3().getF32() > ATTACHMENT_OVERRIDE_DIST)
 					{ //avatar overrides if previously hit drawable is not an attachment or 
 					  //attachment is far enough away from detected intersection
 						drawable = hit;
@@ -6272,9 +6283,9 @@ LLViewerObject* LLPipeline::lineSegmentIntersectInWorld(const LLVector3& start,
 						{
 							*tex_coord = local_texcoord;
 						}
-						if (bi_normal)
+						if (tangent)
 						{
-							*bi_normal = local_binormal;
+							*tangent = local_tangent;
 						}
 						if (normal)
 						{
@@ -6308,13 +6319,13 @@ LLViewerObject* LLPipeline::lineSegmentIntersectInWorld(const LLVector3& start,
 	return drawable ? drawable->getVObj().get() : NULL;
 }

-LLViewerObject* LLPipeline::lineSegmentIntersectInHUD(const LLVector3& start, const LLVector3& end,
+LLViewerObject* LLPipeline::lineSegmentIntersectInHUD(const LLVector4a& start, const LLVector4a& end,
 													  BOOL pick_transparent,													
 													  S32* face_hit,
-													  LLVector3* intersection,         // return the intersection point
+													  LLVector4a* intersection,         // return the intersection point
 													  LLVector2* tex_coord,            // return the texture coordinates of the intersection point
-													  LLVector3* normal,               // return the surface normal at the intersection point
-													  LLVector3* bi_normal             // return the surface bi-normal at the intersection point
+													  LLVector4a* normal,               // return the surface normal at the intersection point
+													  LLVector4a* tangent				// return the surface tangent at the intersection point
 	)
 {
 	LLDrawable* drawable = NULL;
@@ -6334,7 +6345,7 @@ LLViewerObject* LLPipeline::lineSegmentIntersectInHUD(const LLVector3& start, co
 		LLSpatialPartition* part = region->getSpatialPartition(LLViewerRegion::PARTITION_HUD);
 		if (part)
 		{
-			LLDrawable* hit = part->lineSegmentIntersect(start, end, pick_transparent, face_hit, intersection, tex_coord, normal, bi_normal);
+			LLDrawable* hit = part->lineSegmentIntersect(start, end, pick_transparent, face_hit, intersection, tex_coord, normal, tangent);
 			if (hit)
 			{
 				drawable = hit;
@@ -6815,13 +6826,18 @@ void LLPipeline::renderBloom(BOOL for_snapshot, F32 zoom_factor, int subfield, b
 			{
 				if (LLViewerJoystick::getInstance()->getOverrideCamera())
 				{ //focus on point under cursor
-					focus_point = gDebugRaycastIntersection;
+					focus_point.set(gDebugRaycastIntersection.getF32ptr());
 				}
 				else if (gAgentCamera.cameraMouselook())
 				{ //focus on point under mouselook crosshairs
+					LLVector4a result;
+					result.clear();
+
 					gViewerWindow->cursorIntersect(-1, -1, 512.f, NULL, -1, FALSE,
-												  NULL,
-												  &focus_point);
+													NULL,
+													&result);
+
+					focus_point.set(result.getF32ptr());
 				}
 				else if(gAgent.getRegion())
 				{
--- a/indra/newview/pipeline.h
+++ b/indra/newview/pipeline.h
@@ -189,21 +189,21 @@ public:
 	void		markMeshDirty(LLSpatialGroup* group);

 	//get the object between start and end that's closest to start.
-	LLViewerObject* lineSegmentIntersectInWorld(const LLVector3& start, const LLVector3& end,
+	LLViewerObject* lineSegmentIntersectInWorld(const LLVector4a& start, const LLVector4a& end,
 												BOOL pick_transparent,
 												S32* face_hit,                          // return the face hit
-												LLVector3* intersection = NULL,         // return the intersection point
+												LLVector4a* intersection = NULL,         // return the intersection point
 												LLVector2* tex_coord = NULL,            // return the texture coordinates of the intersection point
-												LLVector3* normal = NULL,               // return the surface normal at the intersection point
-												LLVector3* bi_normal = NULL             // return the surface bi-normal at the intersection point  
+												LLVector4a* normal = NULL,               // return the surface normal at the intersection point
+												LLVector4a* tangent = NULL             // return the surface tangent at the intersection point  
 		);
-	LLViewerObject* lineSegmentIntersectInHUD(const LLVector3& start, const LLVector3& end,
+	LLViewerObject* lineSegmentIntersectInHUD(const LLVector4a& start, const LLVector4a& end,
 											  BOOL pick_transparent,
 											  S32* face_hit,                          // return the face hit
-											  LLVector3* intersection = NULL,         // return the intersection point
+											  LLVector4a* intersection = NULL,         // return the intersection point
 											  LLVector2* tex_coord = NULL,            // return the texture coordinates of the intersection point
-											  LLVector3* normal = NULL,               // return the surface normal at the intersection point
-											  LLVector3* bi_normal = NULL             // return the surface bi-normal at the intersection point
+											  LLVector4a* normal = NULL,               // return the surface normal at the intersection point
+											  LLVector4a* tangent = NULL             // return the surface tangent at the intersection point
 		);

 	// Something about these textures has changed.  Dirty them.