MAINT-616 Factor out calls to glGenFoo where possible, add setting to control synchronizing strategy WRT occlusion queries, add experimental transform feedback driven LoD update

27 files changed, 1277 insertions, 585 deletions

diff --git a/indra/llmath/llvolume.h b/indra/llmath/llvolume.h
index 76cf9de613..2e6f9e2f71 100644
--- a/indra/llmath/llvolume.h
+++ b/indra/llmath/llvolume.h
@@ -54,6 +54,7 @@ class LLVolumeTriangle;
 #include "llstrider.h"
 #include "v4coloru.h"
 #include "llrefcount.h"
+#include "llpointer.h"
 #include "llfile.h"
 
 //============================================================================
@@ -919,6 +920,10 @@ public:
 	LLVector2*  mTexCoords;
 	U16* mIndices;
 
+	//vertex buffer filled in by LLFace to cache this volume face geometry in vram 
+	// (declared as a LLPointer to LLRefCount to avoid dependency on LLVertexBuffer)
+	mutable LLPointer<LLRefCount> mVertexBuffer; 
+
 	std::vector<S32>	mEdge;
 
 	//list of skin weights for rigged volumes
diff --git a/indra/llrender/llcubemap.cpp b/indra/llrender/llcubemap.cpp
index 45a3b18179..32e4c0d18e 100644
--- a/indra/llrender/llcubemap.cpp
+++ b/indra/llrender/llcubemap.cpp
@@ -81,7 +81,7 @@ void LLCubeMap::initGL()
 		{
 			U32 texname = 0;
 			
-			LLImageGL::generateTextures(1, &texname);
+			LLImageGL::generateTextures(LLTexUnit::TT_CUBE_MAP, 1, &texname);
 
 			for (int i = 0; i < 6; i++)
 			{
diff --git a/indra/llrender/llgl.cpp b/indra/llrender/llgl.cpp
index 639d967853..3946c43929 100644
--- a/indra/llrender/llgl.cpp
+++ b/indra/llrender/llgl.cpp
@@ -249,6 +249,12 @@ PFNGLTEXIMAGE3DMULTISAMPLEPROC glTexImage3DMultisample = NULL;
 PFNGLGETMULTISAMPLEFVPROC glGetMultisamplefv = NULL;
 PFNGLSAMPLEMASKIPROC glSampleMaski = NULL;
 
+//transform feedback (4.0 core)
+PFNGLBEGINTRANSFORMFEEDBACKPROC glBeginTransformFeedback = NULL;
+PFNGLENDTRANSFORMFEEDBACKPROC glEndTransformFeedback = NULL;
+PFNGLTRANSFORMFEEDBACKVARYINGSPROC glTransformFeedbackVaryings = NULL;
+PFNGLBINDBUFFERRANGEPROC glBindBufferRange = NULL;
+
 //GL_ARB_debug_output
 PFNGLDEBUGMESSAGECONTROLARBPROC glDebugMessageControlARB = NULL;
 PFNGLDEBUGMESSAGEINSERTARBPROC glDebugMessageInsertARB = NULL;
@@ -421,6 +427,7 @@ LLGLManager::LLGLManager() :
 	mHasDrawBuffers(FALSE),
 	mHasTextureRectangle(FALSE),
 	mHasTextureMultisample(FALSE),
+	mHasTransformFeedback(FALSE),
 	mMaxSampleMaskWords(0),
 	mMaxColorTextureSamples(0),
 	mMaxDepthTextureSamples(0),
@@ -969,6 +976,7 @@ void LLGLManager::initExtensions()
 	mHasTextureRectangle = ExtensionExists("GL_ARB_texture_rectangle", gGLHExts.mSysExts);
 	mHasTextureMultisample = ExtensionExists("GL_ARB_texture_multisample", gGLHExts.mSysExts);
 	mHasDebugOutput = ExtensionExists("GL_ARB_debug_output", gGLHExts.mSysExts);
+	mHasTransformFeedback = mGLVersion >= 4.f ? TRUE : FALSE;
 #if !LL_DARWIN
 	mHasPointParameters = !mIsATI && ExtensionExists("GL_ARB_point_parameters", gGLHExts.mSysExts);
 #endif
@@ -1208,7 +1216,14 @@ void LLGLManager::initExtensions()
 		glTexImage3DMultisample = (PFNGLTEXIMAGE3DMULTISAMPLEPROC) GLH_EXT_GET_PROC_ADDRESS("glTexImage3DMultisample");
 		glGetMultisamplefv = (PFNGLGETMULTISAMPLEFVPROC) GLH_EXT_GET_PROC_ADDRESS("glGetMultisamplefv");
 		glSampleMaski = (PFNGLSAMPLEMASKIPROC) GLH_EXT_GET_PROC_ADDRESS("glSampleMaski");
-	}	
+	}
+	if (mHasTransformFeedback)
+	{
+		glBeginTransformFeedback = (PFNGLBEGINTRANSFORMFEEDBACKPROC) GLH_EXT_GET_PROC_ADDRESS("glBeginTransformFeedback");
+		glEndTransformFeedback = (PFNGLENDTRANSFORMFEEDBACKPROC) GLH_EXT_GET_PROC_ADDRESS("glEndTransformFeedback");
+		glTransformFeedbackVaryings = (PFNGLTRANSFORMFEEDBACKVARYINGSPROC) GLH_EXT_GET_PROC_ADDRESS("glTransformFeedbackVaryings");
+		glBindBufferRange = (PFNGLBINDBUFFERRANGEPROC) GLH_EXT_GET_PROC_ADDRESS("glBindBufferRange");
+	}
 	if (mHasDebugOutput)
 	{
 		glDebugMessageControlARB = (PFNGLDEBUGMESSAGECONTROLARBPROC) GLH_EXT_GET_PROC_ADDRESS("glDebugMessageControlARB");
@@ -2433,3 +2448,65 @@ LLGLSquashToFarClip::~LLGLSquashToFarClip()
 	gGL.matrixMode(LLRender::MM_MODELVIEW);
 }
 
+
+	
+LLGLSyncFence::LLGLSyncFence()
+{
+#ifdef GL_ARB_sync
+	mSync = 0;
+#endif
+}
+
+LLGLSyncFence::~LLGLSyncFence()
+{
+#ifdef GL_ARB_sync
+	if (mSync)
+	{
+		glDeleteSync(mSync);
+	}
+#endif
+}
+
+void LLGLSyncFence::placeFence()
+{
+#ifdef GL_ARB_sync
+	if (mSync)
+	{
+		glDeleteSync(mSync);
+	}
+	mSync = glFenceSync(GL_SYNC_GPU_COMMANDS_COMPLETE, 0);
+#endif
+}
+
+bool LLGLSyncFence::isCompleted()
+{
+	bool ret = true;
+#ifdef GL_ARB_sync
+	if (mSync)
+	{
+		GLenum status = glClientWaitSync(mSync, 0, 1);
+		if (status == GL_TIMEOUT_EXPIRED)
+		{
+			ret = false;
+		}
+	}
+#endif
+	return ret;
+}
+
+void LLGLSyncFence::wait()
+{
+#ifdef GL_ARB_sync
+	if (mSync)
+	{
+		while (glClientWaitSync(mSync, 0, FENCE_WAIT_TIME_NANOSECONDS) == GL_TIMEOUT_EXPIRED)
+		{ //track the number of times we've waited here
+			static S32 waits = 0;
+			waits++;
+		}
+	}
+#endif
+}
+
+
+
diff --git a/indra/llrender/llgl.h b/indra/llrender/llgl.h
index 9c3a47bd50..c26b75eff7 100644
--- a/indra/llrender/llgl.h
+++ b/indra/llrender/llgl.h
@@ -104,6 +104,7 @@ public:
 	BOOL mHasDepthClamp;
 	BOOL mHasTextureRectangle;
 	BOOL mHasTextureMultisample;
+	BOOL mHasTransformFeedback;
 	S32 mMaxSampleMaskWords;
 	S32 mMaxColorTextureSamples;
 	S32 mMaxDepthTextureSamples;
@@ -418,6 +419,31 @@ public:
 	virtual void updateGL() = 0;
 };
 
+const U32 FENCE_WAIT_TIME_NANOSECONDS = 10000;  //1 ms
+
+class LLGLFence
+{
+public:
+	virtual void placeFence() = 0;
+	virtual bool isCompleted() = 0;
+	virtual void wait() = 0;
+};
+
+class LLGLSyncFence : public LLGLFence
+{
+public:
+#ifdef GL_ARB_sync
+	GLsync mSync;
+#endif
+	
+	LLGLSyncFence();
+	virtual ~LLGLSyncFence();
+
+	void placeFence();
+	bool isCompleted();
+	void wait();
+};
+
 extern LLMatrix4 gGLObliqueProjectionInverse;
 
 #include "llglstates.h"
diff --git a/indra/llrender/llglheaders.h b/indra/llrender/llglheaders.h
index d61ec707f0..a0727b8686 100644
--- a/indra/llrender/llglheaders.h
+++ b/indra/llrender/llglheaders.h
@@ -528,6 +528,13 @@ extern PFNGLTEXIMAGE3DMULTISAMPLEPROC glTexImage3DMultisample;
 extern PFNGLGETMULTISAMPLEFVPROC glGetMultisamplefv;
 extern PFNGLSAMPLEMASKIPROC glSampleMaski;
 
+//transform feedback (4.0 core)
+extern PFNGLBEGINTRANSFORMFEEDBACKPROC glBeginTransformFeedback;
+extern PFNGLENDTRANSFORMFEEDBACKPROC glEndTransformFeedback;
+extern PFNGLTRANSFORMFEEDBACKVARYINGSPROC glTransformFeedbackVaryings;
+extern PFNGLBINDBUFFERRANGEPROC glBindBufferRange;
+
+
 #elif LL_WINDOWS
 //----------------------------------------------------------------------------
 // LL_WINDOWS
@@ -759,6 +766,12 @@ extern PFNGLTEXIMAGE3DMULTISAMPLEPROC glTexImage3DMultisample;
 extern PFNGLGETMULTISAMPLEFVPROC glGetMultisamplefv;
 extern PFNGLSAMPLEMASKIPROC glSampleMaski;
 
+//transform feedback (4.0 core)
+extern PFNGLBEGINTRANSFORMFEEDBACKPROC glBeginTransformFeedback;
+extern PFNGLENDTRANSFORMFEEDBACKPROC glEndTransformFeedback;
+extern PFNGLTRANSFORMFEEDBACKVARYINGSPROC glTransformFeedbackVaryings;
+extern PFNGLBINDBUFFERRANGEPROC glBindBufferRange;
+
 //GL_ARB_debug_output
 extern PFNGLDEBUGMESSAGECONTROLARBPROC glDebugMessageControlARB;
 extern PFNGLDEBUGMESSAGEINSERTARBPROC glDebugMessageInsertARB;
diff --git a/indra/llrender/llglslshader.cpp b/indra/llrender/llglslshader.cpp
index 4b7e639aed..149e8cc548 100644
--- a/indra/llrender/llglslshader.cpp
+++ b/indra/llrender/llglslshader.cpp
@@ -129,7 +129,9 @@ void LLGLSLShader::unload()
 }
 
 BOOL LLGLSLShader::createShader(vector<string> * attributes,
-								vector<string> * uniforms)
+								vector<string> * uniforms,
+								U32 varying_count,
+								const char** varyings)
 {
 	//reloading, reset matrix hash values
 	for (U32 i = 0; i < LLRender::NUM_MATRIX_MODES; ++i)
@@ -172,6 +174,11 @@ BOOL LLGLSLShader::createShader(vector<string> * attributes,
 		mFeatures.mIndexedTextureChannels = llmin(mFeatures.mIndexedTextureChannels, 1);
 	}
 
+	if (varying_count > 0 && varyings)
+	{
+		glTransformFeedbackVaryings(mProgramObject, varying_count, varyings, GL_INTERLEAVED_ATTRIBS);
+	}
+
 	// Map attributes and uniforms
 	if (success)
 	{
diff --git a/indra/llrender/llglslshader.h b/indra/llrender/llglslshader.h
index 7873fe3c4e..5c68cb46eb 100644
--- a/indra/llrender/llglslshader.h
+++ b/indra/llrender/llglslshader.h
@@ -76,7 +76,9 @@ public:
 
 	void unload();
 	BOOL createShader(std::vector<std::string> * attributes,
-						std::vector<std::string> * uniforms);
+						std::vector<std::string> * uniforms,
+						U32 varying_count = 0,
+						const char** varyings = NULL);
 	BOOL attachObject(std::string object);
 	void attachObject(GLhandleARB object);
 	void attachObjects(GLhandleARB* objects = NULL, S32 count = 0);
diff --git a/indra/llrender/llimagegl.cpp b/indra/llrender/llimagegl.cpp
index 17131c9d8a..3bdee6cade 100644
--- a/indra/llrender/llimagegl.cpp
+++ b/indra/llrender/llimagegl.cpp
@@ -51,7 +51,8 @@ S32 LLImageGL::sGlobalTextureMemoryInBytes		= 0;
 S32 LLImageGL::sBoundTextureMemoryInBytes		= 0;
 S32 LLImageGL::sCurBoundTextureMemory	= 0;
 S32 LLImageGL::sCount					= 0;
-std::list<U32> LLImageGL::sDeadTextureList;
+std::list<U32> LLImageGL::sDeadTextureList[LLTexUnit::TT_NONE];
+U32 LLImageGL::sCurTexName = 1;
 
 BOOL LLImageGL::sGlobalUseAnisotropic	= FALSE;
 F32 LLImageGL::sLastFrameTime			= 0.f;
@@ -1093,23 +1094,49 @@ BOOL LLImageGL::setSubImageFromFrameBuffer(S32 fb_x, S32 fb_y, S32 x_pos, S32 y_
 }
 
 // static
-void LLImageGL::generateTextures(S32 numTextures, U32 *textures)
+void LLImageGL::generateTextures(LLTexUnit::eTextureType type, S32 numTextures, U32 *textures)
 {
-	glGenTextures(numTextures, (GLuint*)textures);
+	for (S32 i = 0; i < numTextures; ++i)
+	{
+		if (!sDeadTextureList[type].empty())
+		{
+			textures[i] = sDeadTextureList[type].front();
+			sDeadTextureList[type].pop_front();
+		}
+		else
+		{
+			textures[i] = sCurTexName++;
+		}
+	}
 }
 
 // static
-void LLImageGL::deleteTextures(S32 numTextures, U32 *textures, bool immediate)
+void LLImageGL::deleteTextures(LLTexUnit::eTextureType type, S32 numTextures, U32 *textures, bool immediate)
 {
-	for (S32 i = 0; i < numTextures; i++)
-	{
-		sDeadTextureList.push_back(textures[i]);
-	}
+	for (S32 i = 0; i < numTextures; ++i)
+	{ //remove texture from VRAM by setting its size to zero
+		gGL.getTexUnit(0)->bindManual(type, textures[i]);
 
-	if (immediate)
+		if (type == LLTexUnit::TT_CUBE_MAP)
+		{
+			glTexImage2D(GL_TEXTURE_CUBE_MAP_POSITIVE_X, 0, GL_RGBA, 0, 0, 0, GL_RGBA, GL_UNSIGNED_BYTE, NULL);
+			glTexImage2D(GL_TEXTURE_CUBE_MAP_POSITIVE_Y, 0, GL_RGBA, 0, 0, 0, GL_RGBA, GL_UNSIGNED_BYTE, NULL);
+			glTexImage2D(GL_TEXTURE_CUBE_MAP_POSITIVE_Z, 0, GL_RGBA, 0, 0, 0, GL_RGBA, GL_UNSIGNED_BYTE, NULL);
+			glTexImage2D(GL_TEXTURE_CUBE_MAP_NEGATIVE_X, 0, GL_RGBA, 0, 0, 0, GL_RGBA, GL_UNSIGNED_BYTE, NULL);
+			glTexImage2D(GL_TEXTURE_CUBE_MAP_NEGATIVE_Y, 0, GL_RGBA, 0, 0, 0, GL_RGBA, GL_UNSIGNED_BYTE, NULL);
+			glTexImage2D(GL_TEXTURE_CUBE_MAP_NEGATIVE_Z, 0, GL_RGBA, 0, 0, 0, GL_RGBA, GL_UNSIGNED_BYTE, NULL);
+		}
+		else
+		{
+			glTexImage2D(LLTexUnit::getInternalType(type), 0, GL_RGBA, 0, 0, 0, GL_RGBA, GL_UNSIGNED_BYTE, NULL);
+		}
+		sDeadTextureList[type].push_back(textures[i]);
+	}
+	
+	/*if (immediate)
 	{
 		LLImageGL::deleteDeadTextures();
-	}
+	}*/
 }
 
 // static
@@ -1234,10 +1261,11 @@ BOOL LLImageGL::createGLTexture()
 
 	if(mTexName)
 	{
-		glDeleteTextures(1, (reinterpret_cast<GLuint*>(&mTexName))) ;
+		LLImageGL::deleteTextures(mBindTarget, 1, (reinterpret_cast<GLuint*>(&mTexName))) ;
 	}
 	
-	glGenTextures(1, (GLuint*)&mTexName);
+
+	LLImageGL::generateTextures(mBindTarget, 1, &mTexName);
 	stop_glerror();
 	if (!mTexName)
 	{
@@ -1350,7 +1378,7 @@ BOOL LLImageGL::createGLTexture(S32 discard_level, const U8* data_in, BOOL data_
 	}
 	else
 	{
-		LLImageGL::generateTextures(1, &mTexName);
+		LLImageGL::generateTextures(mBindTarget, 1, &mTexName);
 		stop_glerror();
 		{
 			llverify(gGL.getTexUnit(0)->bind(this));
@@ -1400,7 +1428,7 @@ BOOL LLImageGL::createGLTexture(S32 discard_level, const U8* data_in, BOOL data_
 			decTextureCounter(mTextureMemory, mComponents, mCategory) ;
 		}
 
-		LLImageGL::deleteTextures(1, &old_name);
+		LLImageGL::deleteTextures(mBindTarget,1, &old_name);
 
 		stop_glerror();
 	}
@@ -1533,7 +1561,7 @@ void LLImageGL::deleteDeadTextures()
 {
 	bool reset = false;
 
-	while (!sDeadTextureList.empty())
+	/*while (!sDeadTextureList.empty())
 	{
 		GLuint tex = sDeadTextureList.front();
 		sDeadTextureList.pop_front();
@@ -1555,7 +1583,7 @@ void LLImageGL::deleteDeadTextures()
 		
 		glDeleteTextures(1, &tex);
 		stop_glerror();
-	}
+	}*/
 
 	if (reset)
 	{
@@ -1577,7 +1605,7 @@ void LLImageGL::destroyGLTexture()
 			mTextureMemory = 0;
 		}
 		
-		LLImageGL::deleteTextures(1, &mTexName);			
+		LLImageGL::deleteTextures(mBindTarget, 1, &mTexName);			
 		mTexName = 0;
 		mCurrentDiscardLevel = -1 ; //invalidate mCurrentDiscardLevel.
 		mGLTextureCreated = FALSE ;
diff --git a/indra/llrender/llimagegl.h b/indra/llrender/llimagegl.h
index e23005fe29..8c9cea111e 100644
--- a/indra/llrender/llimagegl.h
+++ b/indra/llrender/llimagegl.h
@@ -45,8 +45,12 @@ class LLImageGL : public LLRefCount
 {
 	friend class LLTexUnit;
 public:
-	static std::list<U32> sDeadTextureList;
+	static U32 sCurTexName;
+	static std::list<U32> sDeadTextureList[LLTexUnit::TT_NONE];
 
+	// These 2 functions replace glGenTextures() and glDeleteTextures()
+	static void generateTextures(LLTexUnit::eTextureType type, S32 numTextures, U32 *textures);
+	static void deleteTextures(LLTexUnit::eTextureType type, S32 numTextures, U32 *textures, bool immediate = false);
 	static void deleteDeadTextures();
 
 	// Size calculation
@@ -96,10 +100,6 @@ public:
 	void setComponents(S32 ncomponents) { mComponents = (S8)ncomponents ;}
 	void setAllowCompression(bool allow) { mAllowCompression = allow; }
 
-	// These 3 functions currently wrap glGenTextures(), glDeleteTextures(), and glTexImage2D() 
-	// for tracking purposes and will be deprecated in the future
-	static void generateTextures(S32 numTextures, U32 *textures);
-	static void deleteTextures(S32 numTextures, U32 *textures, bool immediate = false);
 	static void setManualImage(U32 target, S32 miplevel, S32 intformat, S32 width, S32 height, U32 pixformat, U32 pixtype, const void *pixels, bool allow_compression = true);
 
 	BOOL createGLTexture() ;
diff --git a/indra/llrender/llrendertarget.cpp b/indra/llrender/llrendertarget.cpp
index 780f1dc484..f0dd6f3bd6 100644
--- a/indra/llrender/llrendertarget.cpp
+++ b/indra/llrender/llrendertarget.cpp
@@ -135,7 +135,7 @@ bool LLRenderTarget::addColorAttachment(U32 color_fmt)
 	}
 
 	U32 tex;
-	LLImageGL::generateTextures(1, &tex);
+	LLImageGL::generateTextures(mUsage, 1, &tex);
 	gGL.getTexUnit(0)->bindManual(mUsage, tex);
 
 	stop_glerror();
@@ -217,7 +217,7 @@ bool LLRenderTarget::allocateDepth()
 	}
 	else
 	{
-		LLImageGL::generateTextures(1, &mDepth);
+		LLImageGL::generateTextures(mUsage, 1, &mDepth);
 		gGL.getTexUnit(0)->bindManual(mUsage, mDepth);
 		
 		U32 internal_type = LLTexUnit::getInternalType(mUsage);
@@ -294,7 +294,7 @@ void LLRenderTarget::release()
 		}
 		else
 		{
-			LLImageGL::deleteTextures(1, &mDepth, true);
+			LLImageGL::deleteTextures(mUsage, 1, &mDepth, true);
 			stop_glerror();
 		}
 		mDepth = 0;
@@ -326,7 +326,7 @@ void LLRenderTarget::release()
 	if (mTex.size() > 0)
 	{
 		sBytesAllocated -= mResX*mResY*4*mTex.size();
-		LLImageGL::deleteTextures(mTex.size(), &mTex[0], true);
+		LLImageGL::deleteTextures(mUsage, mTex.size(), &mTex[0], true);
 		mTex.clear();
 	}
 	
diff --git a/indra/llrender/llvertexbuffer.cpp b/indra/llrender/llvertexbuffer.cpp
index 7b12304967..6a218e7734 100644
--- a/indra/llrender/llvertexbuffer.cpp
+++ b/indra/llrender/llvertexbuffer.cpp
@@ -93,6 +93,11 @@ LLVBOPool LLVertexBuffer::sDynamicIBOPool(GL_DYNAMIC_DRAW_ARB, GL_ELEMENT_ARRAY_
 
 U32 LLVBOPool::sBytesPooled = 0;
 U32 LLVBOPool::sIndexBytesPooled = 0;
+U32 LLVBOPool::sCurGLName = 1;
+
+std::list<U32> LLVertexBuffer::sAvailableVAOName;
+U32 LLVertexBuffer::sCurVAOName = 1;
+
 U32 LLVertexBuffer::sAllocatedIndexBytes = 0;
 U32 LLVertexBuffer::sIndexCount = 0;
 
@@ -117,59 +122,38 @@ bool LLVertexBuffer::sUseStreamDraw = true;
 bool LLVertexBuffer::sUseVAO = false;
 bool LLVertexBuffer::sPreferStreamDraw = false;
 
-const U32 FENCE_WAIT_TIME_NANOSECONDS = 10000;  //1 ms
 
-class LLGLSyncFence : public LLGLFence
+U32 LLVBOPool::genBuffer()
 {
-public:
-#ifdef GL_ARB_sync
-	GLsync mSync;
-#endif
-	
-	LLGLSyncFence()
-	{
-#ifdef GL_ARB_sync
-		mSync = 0;
-#endif
-	}
+	U32 ret = 0;
 
-	virtual ~LLGLSyncFence()
+	if (mGLNamePool.empty())
 	{
-#ifdef GL_ARB_sync
-		if (mSync)
-		{
-			glDeleteSync(mSync);
-		}
-#endif
+		ret = sCurGLName++;
 	}
-
-	void placeFence()
+	else
 	{
-#ifdef GL_ARB_sync
-		if (mSync)
-		{
-			glDeleteSync(mSync);
-		}
-		mSync = glFenceSync(GL_SYNC_GPU_COMMANDS_COMPLETE, 0);
-#endif
+		ret = mGLNamePool.front();
+		mGLNamePool.pop_front();
 	}
 
-	void wait()
-	{
-#ifdef GL_ARB_sync
-		if (mSync)
-		{
-			while (glClientWaitSync(mSync, 0, FENCE_WAIT_TIME_NANOSECONDS) == GL_TIMEOUT_EXPIRED)
-			{ //track the number of times we've waited here
-				static S32 waits = 0;
-				waits++;
-			}
-		}
-#endif
-	}
+	return ret;
+}
+
+void LLVBOPool::deleteBuffer(U32 name)
+{
+	LLVertexBuffer::unbind();
 
+	glBindBufferARB(mType, name);
+	glBufferDataARB(mType, 0, NULL, mUsage);
+
+	llassert(std::find(mGLNamePool.begin(), mGLNamePool.end(), name) == mGLNamePool.end());
+
+	mGLNamePool.push_back(name);
+
+	LLVertexBuffer::unbind();
+}
 
-};
 
 LLVBOPool::LLVBOPool(U32 vboUsage, U32 vboType)
 : mUsage(vboUsage), mType(vboType)
@@ -178,6 +162,9 @@ LLVBOPool::LLVBOPool(U32 vboUsage, U32 vboType)
 	std::fill(mMissCount.begin(), mMissCount.end(), 0);
 }
 
+static LLFastTimer::DeclareTimer FTM_VBO_GEN_BUFFER("gen buffers");
+static LLFastTimer::DeclareTimer FTM_VBO_BUFFER_DATA("glBufferData");
+
 
 volatile U8* LLVBOPool::allocate(U32& name, U32 size, bool for_seed)
 {
@@ -198,7 +185,10 @@ volatile U8* LLVBOPool::allocate(U32& name, U32 size, bool for_seed)
 	if (mFreeList[i].empty() || for_seed)
 	{
 		//make a new buffer
-		glGenBuffersARB(1, &name);
+		{
+			LLFastTimer t(FTM_VBO_GEN_BUFFER);
+			name = genBuffer();
+		}
 		glBindBufferARB(mType, name);
 
 		if (!for_seed && i < LL_VBO_POOL_SEED_COUNT)
@@ -222,6 +212,7 @@ volatile U8* LLVBOPool::allocate(U32& name, U32 size, bool for_seed)
 		}
 		else
 		{ //always use a true hint of static draw when allocating non-client-backed buffers
+			LLFastTimer t(FTM_VBO_BUFFER_DATA);
 			glBufferDataARB(mType, size, 0, GL_STATIC_DRAW_ARB);
 		}
 
@@ -324,7 +315,7 @@ void LLVBOPool::release(U32 name, volatile U8* buffer, U32 size)
 		mFreeList[i].push_back(rec);
 	}
 #else //no pooling
-	glDeleteBuffersARB(1, &name);
+	deleteBuffer(name);
 	ll_aligned_free_16((U8*) buffer);
 
 	if (mType == GL_ARRAY_BUFFER_ARB)
@@ -375,8 +366,8 @@ void LLVBOPool::cleanup()
 		{
 			Record& r = l.front();
 
-			glDeleteBuffersARB(1, &r.mGLName);
-
+			deleteBuffer(r.mGLName);
+			
 			if (r.mClientData)
 			{
 				ll_aligned_free_16((void*) r.mClientData);
@@ -434,6 +425,30 @@ U32 LLVertexBuffer::sGLMode[LLRender::NUM_MODES] =
 	GL_LINE_LOOP,
 };
 
+//static
+U32 LLVertexBuffer::getVAOName()
+{
+	U32 ret = 0;
+
+	if (!sAvailableVAOName.empty())
+	{
+		ret = sAvailableVAOName.front();
+		sAvailableVAOName.pop_front();
+	}
+	else
+	{
+		glGenVertexArrays(1, &ret);
+	}
+
+	return ret;		
+}
+
+//static
+void LLVertexBuffer::releaseVAOName(U32 name)
+{
+	sAvailableVAOName.push_back(name);
+}
+
 
 //static
 void LLVertexBuffer::seedPools()
@@ -1052,7 +1067,7 @@ LLVertexBuffer::~LLVertexBuffer()
 	if (mGLArray)
 	{
 #if GL_ARB_vertex_array_object
-		glDeleteVertexArrays(1, &mGLArray);
+		releaseVAOName(mGLArray);
 #endif
 	}
 
@@ -1337,7 +1352,7 @@ void LLVertexBuffer::allocateBuffer(S32 nverts, S32 nindices, bool create)
 		if (gGLManager.mHasVertexArrayObject && useVBOs() && (LLRender::sGLCoreProfile || sUseVAO))
 		{
 #if GL_ARB_vertex_array_object
-			glGenVertexArrays(1, &mGLArray);
+			mGLArray = getVAOName();
 #endif
 			setupVertexArray();
 		}
@@ -2207,6 +2222,14 @@ void LLVertexBuffer::flush()
 	}
 }
 
+// bind for transform feedback (quick 'n dirty)
+void LLVertexBuffer::bindForFeedback(U32 channel, U32 type, U32 index, U32 count)
+{
+	U32 offset = mOffsets[type] + sTypeSize[type]*index;
+	U32 size= (sTypeSize[type]*count);
+	glBindBufferRange(GL_TRANSFORM_FEEDBACK_BUFFER, channel, mGLBuffer, offset, size);
+}
+
 // Set for rendering
 void LLVertexBuffer::setBuffer(U32 data_mask)
 {
diff --git a/indra/llrender/llvertexbuffer.h b/indra/llrender/llvertexbuffer.h
index a64daa1a90..11fa4ab6a0 100644
--- a/indra/llrender/llvertexbuffer.h
+++ b/indra/llrender/llvertexbuffer.h
@@ -57,6 +57,8 @@ public:
 	static U32 sBytesPooled;
 	static U32 sIndexBytesPooled;
 	
+	static U32 sCurGLName;
+
 	LLVBOPool(U32 vboUsage, U32 vboType);
 		
 	const U32 mUsage;
@@ -74,6 +76,9 @@ public:
 	//destroy all records in mFreeList
 	void cleanup();
 
+	U32 genBuffer();
+	void deleteBuffer(U32 name);
+
 	class Record
 	{
 	public:
@@ -81,18 +86,14 @@ public:
 		volatile U8* mClientData;
 	};
 
+	std::list<U32> mGLNamePool;
+
 	typedef std::list<Record> record_list_t;
 	std::vector<record_list_t> mFreeList;
 	std::vector<U32> mMissCount;
 
 };
 
-class LLGLFence
-{
-public:
-	virtual void placeFence() = 0;
-	virtual void wait() = 0;
-};
 
 //============================================================================
 // base class 
@@ -127,15 +128,22 @@ public:
 	static LLVBOPool sStreamIBOPool;
 	static LLVBOPool sDynamicIBOPool;
 
+	static std::list<U32> sAvailableVAOName;
+	static U32 sCurVAOName;
+
 	static bool	sUseStreamDraw;
 	static bool sUseVAO;
 	static bool	sPreferStreamDraw;
 
 	static void seedPools();
 
+	static U32 getVAOName();
+	static void releaseVAOName(U32 name);
+
 	static void initClass(bool use_vbo, bool no_vbo_mapping);
 	static void cleanupClass();
 	static void setupClientArrays(U32 data_mask);
+	static void pushPositions(U32 mode, const LLVector4a* pos, U32 count);
 	static void drawArrays(U32 mode, const std::vector<LLVector3>& pos, const std::vector<LLVector3>& norm);
 	static void drawElements(U32 mode, const LLVector4a* pos, const LLVector2* tc, S32 num_indices, const U16* indicesp);
 
@@ -212,7 +220,6 @@ protected:
 	void 	destroyGLIndices();
 	void	updateNumVerts(S32 nverts);
 	void	updateNumIndices(S32 nindices); 
-	bool	useVBOs() const;
 	void	unmapBuffer();
 		
 public:
@@ -222,6 +229,8 @@ public:
 	volatile U8*		mapVertexBuffer(S32 type, S32 index, S32 count, bool map_range);
 	volatile U8*		mapIndexBuffer(S32 index, S32 count, bool map_range);
 
+	void bindForFeedback(U32 channel, U32 type, U32 index, U32 count);
+
 	// set for rendering
 	virtual void	setBuffer(U32 data_mask); 	// calls  setupVertexBuffer() if data_mask is not 0
 	void flush(); //flush pending data to GL memory
@@ -244,12 +253,14 @@ public:
 	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 getColorStrider(LLStrider<LLColor4U>& strider, S32 index=0, S32 count = -1, bool map_range = false);
+	bool getTextureIndexStrider(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);
 	bool getWeight4Strider(LLStrider<LLVector4>& strider, S32 index=0, S32 count = -1, bool map_range = false);
 	bool getClothWeightStrider(LLStrider<LLVector4>& strider, S32 index=0, S32 count = -1, bool map_range = false);
 	
 
+	bool useVBOs() const;
 	bool isEmpty() const					{ return mEmpty; }
 	bool isLocked() const					{ return mVertexLocked || mIndexLocked; }
 	S32 getNumVerts() const					{ return mNumVerts; }
diff --git a/indra/newview/app_settings/settings.xml b/indra/newview/app_settings/settings.xml
index bd109120cf..8cb456b4fd 100644
--- a/indra/newview/app_settings/settings.xml
+++ b/indra/newview/app_settings/settings.xml
@@ -9162,7 +9162,7 @@
     <key>RenderUseVAO</key>
     <map>
       <key>Comment</key>
-      <string>Use GL Vertex Array Objects</string>
+      <string>[EXPERIMENTAL] Use GL Vertex Array Objects</string>
       <key>Persist</key>
       <integer>1</integer>
       <key>Type</key>
@@ -9170,7 +9170,19 @@
       <key>Value</key>
       <integer>0</integer>
     </map>
-    <key>RenderVBOMappingDisable</key>
+  <key>RenderUseTransformFeedback</key>
+  <map>
+    <key>Comment</key>
+    <string>[EXPERIMENTAL] Use transform feedback shaders for LoD updates</string>
+    <key>Persist</key>
+    <integer>1</integer>
+    <key>Type</key>
+    <string>Boolean</string>
+    <key>Value</key>
+    <integer>0</integer>
+  </map>
+
+  <key>RenderVBOMappingDisable</key>
     <map>
       <key>Comment</key>
       <string>Disable VBO glMapBufferARB</string>
@@ -12140,6 +12152,17 @@
       <key>Value</key>
       <integer>1</integer>
     </map>
+  <key>RenderSynchronousOcclusion</key>
+  <map>
+    <key>Comment</key>
+    <string>Don't let occlusion queries get more than one frame behind (block until they complete).</string>
+    <key>Persist</key>
+    <integer>1</integer>
+    <key>Type</key>
+    <string>Boolean</string>
+    <key>Value</key>
+    <integer>1</integer>
+  </map>
     <key>RenderDelayVBUpdate</key>
     <map>
       <key>Comment</key>
diff --git a/indra/newview/lldrawpoolalpha.cpp b/indra/newview/lldrawpoolalpha.cpp
index 5b62dbc560..a56e71baa8 100644
--- a/indra/newview/lldrawpoolalpha.cpp
+++ b/indra/newview/lldrawpoolalpha.cpp
@@ -405,6 +405,12 @@ void LLDrawPoolAlpha::renderAlpha(U32 mask)
 			{
 				LLDrawInfo& params = **k;
 
+				if ((params.mVertexBuffer->getTypeMask() & mask) != mask)
+				{ //FIXME!
+					//llwarns << "Missing required components, skipping render batch." << llendl;
+					continue;
+				}
+
 				LLRenderPass::applyModelMatrix(params);
 
 				
diff --git a/indra/newview/llface.cpp b/indra/newview/llface.cpp
index 97b832520d..2824b1a32d 100644
--- a/indra/newview/llface.cpp
+++ b/indra/newview/llface.cpp
@@ -50,6 +50,8 @@
 #include "pipeline.h"
 #include "llviewerregion.h"
 #include "llviewerwindow.h"
+#include "llviewershadermgr.h"
+
 
 #define LL_MAX_INDICES_COUNT 1000000
 
@@ -57,7 +59,6 @@ BOOL LLFace::sSafeRenderSelect = TRUE; // FALSE
 
 #define DOTVEC(a,b) (a.mV[0]*b.mV[0] + a.mV[1]*b.mV[1] + a.mV[2]*b.mV[2])
 
-
 /*
 For each vertex, given:
 	B - binormal
@@ -1111,6 +1112,73 @@ bool LLFace::canRenderAsMask()
 }
 
 
+static LLFastTimer::DeclareTimer FTM_FACE_GEOM_VOLUME("Volume VB Cache");
+
+//static 
+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;
+	
+	if (vf.mWeights)
+	{
+		mask |= LLVertexBuffer::MAP_WEIGHT4;
+	}
+
+	LLVertexBuffer* buff = new LLVertexBuffer(mask, GL_STATIC_DRAW_ARB);
+	vf.mVertexBuffer = buff;
+
+	buff->allocateBuffer(vf.mNumVertices, 0, true);
+
+	LLStrider<LLVector4a> f_vert;
+	LLStrider<LLVector3> f_binorm;
+	LLStrider<LLVector3> f_norm;
+	LLStrider<LLVector2> f_tc;
+
+	buff->getBinormalStrider(f_binorm);
+	buff->getVertexStrider(f_vert);
+	buff->getNormalStrider(f_norm);
+	buff->getTexCoord0Strider(f_tc);
+
+	for (U32 i = 0; i < vf.mNumVertices; ++i)
+	{
+		*f_vert++ = vf.mPositions[i];
+		(*f_binorm++).set(vf.mBinormals[i].getF32ptr());
+		*f_tc++ = vf.mTexCoords[i];
+		(*f_norm++).set(vf.mNormals[i].getF32ptr());
+	}
+
+	if (vf.mWeights)
+	{
+		LLStrider<LLVector4> f_wght;
+		buff->getWeight4Strider(f_wght);
+		for (U32 i = 0; i < vf.mNumVertices; ++i)
+		{
+			(*f_wght++).set(vf.mWeights[i].getF32ptr());
+		}
+	}
+
+	buff->flush();
+}
+
+//helper function for pushing primitives for transform shaders and cleaning up
+//uninitialized data on the tail, plus tracking number of expected primitives
+void push_for_transform(LLVertexBuffer* buff, U32 source_count, U32 dest_count)
+{
+	if (source_count > 0 && dest_count >= source_count) //protect against possible U32 wrapping
+	{
+		//push source primitives
+		buff->drawArrays(LLRender::POINTS, 0, source_count);
+		U32 tail = dest_count-source_count;
+		for (U32 i = 0; i < tail; ++i)
+		{ //copy last source primitive into each element in tail
+			buff->drawArrays(LLRender::POINTS, source_count-1, 1);
+		}
+		gPipeline.mTransformFeedbackPrimitives += dest_count;
+	}
+}
+
 static LLFastTimer::DeclareTimer FTM_FACE_GET_GEOM("Face Geom");
 static LLFastTimer::DeclareTimer FTM_FACE_GEOM_POSITION("Position");
 static LLFastTimer::DeclareTimer FTM_FACE_GEOM_NORMAL("Normal");
@@ -1128,7 +1196,6 @@ static LLFastTimer::DeclareTimer FTM_FACE_TEX_DEFAULT("Default");
 static LLFastTimer::DeclareTimer FTM_FACE_TEX_QUICK("Quick");
 static LLFastTimer::DeclareTimer FTM_FACE_TEX_QUICK_NO_XFORM("No Xform");
 static LLFastTimer::DeclareTimer FTM_FACE_TEX_QUICK_XFORM("Xform");
-
 static LLFastTimer::DeclareTimer FTM_FACE_TEX_QUICK_PLANAR("Quick Planar");
 
 BOOL LLFace::getGeometryVolume(const LLVolume& volume,
@@ -1301,17 +1368,10 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume,
 	LLMatrix4a mat_normal;
 	mat_normal.loadu(mat_norm_in);
 	
-	//if it's not fullbright and has no normals, bake sunlight based on face normal
-	//bool bake_sunlight = !getTextureEntry()->getFullbright() &&
-	//  !mVertexBuffer->hasDataType(LLVertexBuffer::TYPE_NORMAL);
-
 	F32 r = 0, os = 0, ot = 0, ms = 0, mt = 0, cos_ang = 0, sin_ang = 0;
-
+	bool do_xform = false;
 	if (rebuild_tcoord)
 	{
-		LLFastTimer t(FTM_FACE_GEOM_TEXTURE);
-		bool do_xform;
-			
 		if (tep)
 		{
 			r  = tep->getRotation();
@@ -1340,599 +1400,757 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume,
 		{
 			do_xform = false;
 		}
+	}
+	
+	static LLCachedControl<bool> use_transform_feedback(gSavedSettings, "RenderUseTransformFeedback");
+
+	if (use_transform_feedback &&
+		gTransformPositionProgram.mProgramObject && //transform shaders are loaded
+		mVertexBuffer->useVBOs() && //target buffer is in VRAM
+		!rebuild_weights && //TODO: add support for weights
+		!volume.isUnique()) //source volume is NOT flexi
+	{ //use transform feedback to pack vertex buffer
+
+		LLVertexBuffer* buff = (LLVertexBuffer*) vf.mVertexBuffer.get();
+
+		if (vf.mVertexBuffer.isNull() || buff->getNumVerts() != vf.mNumVertices)
+		{
+			mVObjp->getVolume()->genBinormals(f);
+			LLFace::cacheFaceInVRAM(vf);
+			buff = (LLVertexBuffer*) vf.mVertexBuffer.get();
+		}		
+
+		LLGLSLShader* cur_shader = LLGLSLShader::sCurBoundShaderPtr;
+		
+		gGL.pushMatrix();
+		gGL.loadMatrix((GLfloat*) mat_vert_in.mMatrix);
+
+		if (rebuild_pos)
+		{
+			LLFastTimer t(FTM_FACE_GEOM_POSITION);
+			gTransformPositionProgram.bind();
+
+			mVertexBuffer->bindForFeedback(0, LLVertexBuffer::TYPE_VERTEX, mGeomIndex, mGeomCount);
+
+			U8 index = mTextureIndex < 255 ? mTextureIndex : 0;
+
+			S32 val = 0.f;
+			U8* vp = (U8*) &val;
+			vp[0] = index;
+			vp[1] = 0;
+			vp[2] = 0;
+			vp[3] = 0;
+			
+			gTransformPositionProgram.uniform1i("texture_index_in", val);
+			glBeginTransformFeedback(GL_POINTS);
+			buff->setBuffer(LLVertexBuffer::MAP_VERTEX);
+
+			push_for_transform(buff, vf.mNumVertices, mGeomCount);
+
+			glEndTransformFeedback();
+		}
+
+		if (rebuild_color)
+		{
+			LLFastTimer t(FTM_FACE_GEOM_COLOR);
+			gTransformColorProgram.bind();
+			
+			mVertexBuffer->bindForFeedback(0, LLVertexBuffer::TYPE_COLOR, mGeomIndex, mGeomCount);
+
+			S32 val = *((S32*) color.mV);
+
+			gTransformColorProgram.uniform1i("color_in", val);
+			glBeginTransformFeedback(GL_POINTS);
+			buff->setBuffer(LLVertexBuffer::MAP_VERTEX);
+			push_for_transform(buff, vf.mNumVertices, mGeomCount);
+			glEndTransformFeedback();
+		}
+
+		if (rebuild_emissive)
+		{
+			LLFastTimer t(FTM_FACE_GEOM_EMISSIVE);
+			gTransformColorProgram.bind();
+			
+			mVertexBuffer->bindForFeedback(0, LLVertexBuffer::TYPE_EMISSIVE, mGeomIndex, mGeomCount);
+
+			U8 glow = (U8) llclamp((S32) (getTextureEntry()->getGlow()*255), 0, 255);
+
+			S32 glow32 = glow |
+						 (glow << 8) |
+						 (glow << 16) |
+						 (glow << 24);
+
+			gTransformColorProgram.uniform1i("color_in", glow32);
+			glBeginTransformFeedback(GL_POINTS);
+			buff->setBuffer(LLVertexBuffer::MAP_VERTEX);
+			push_for_transform(buff, vf.mNumVertices, mGeomCount);
+			glEndTransformFeedback();
+		}
+
+		if (rebuild_normal)
+		{
+			LLFastTimer t(FTM_FACE_GEOM_NORMAL);
+			gTransformNormalProgram.bind();
+			
+			mVertexBuffer->bindForFeedback(0, LLVertexBuffer::TYPE_NORMAL, mGeomIndex, mGeomCount);
 						
-		//bump setup
-		LLVector4a binormal_dir( -sin_ang, cos_ang, 0.f );
-		LLVector4a bump_s_primary_light_ray(0.f, 0.f, 0.f);
-		LLVector4a bump_t_primary_light_ray(0.f, 0.f, 0.f);
+			glBeginTransformFeedback(GL_POINTS);
+			buff->setBuffer(LLVertexBuffer::MAP_NORMAL);
+			push_for_transform(buff, vf.mNumVertices, mGeomCount);
+			glEndTransformFeedback();
+		}
 
-		LLQuaternion bump_quat;
-		if (mDrawablep->isActive())
+		if (rebuild_binormal)
 		{
-			bump_quat = LLQuaternion(mDrawablep->getRenderMatrix());
+			LLFastTimer t(FTM_FACE_GEOM_BINORMAL);
+			gTransformBinormalProgram.bind();
+			
+			mVertexBuffer->bindForFeedback(0, LLVertexBuffer::TYPE_BINORMAL, mGeomIndex, mGeomCount);
+						
+			glBeginTransformFeedback(GL_POINTS);
+			buff->setBuffer(LLVertexBuffer::MAP_BINORMAL);
+			push_for_transform(buff, vf.mNumVertices, mGeomCount);
+			glEndTransformFeedback();
 		}
-		
-		if (bump_code)
+
+		if (rebuild_tcoord)
 		{
-			mVObjp->getVolume()->genBinormals(f);
-			F32 offset_multiple; 
-			switch( bump_code )
-			{
-				case BE_NO_BUMP:
-				offset_multiple = 0.f;
-				break;
-				case BE_BRIGHTNESS:
-				case BE_DARKNESS:
-				if( mTexture.notNull() && mTexture->hasGLTexture())
-				{
-					// Offset by approximately one texel
-					S32 cur_discard = mTexture->getDiscardLevel();
-					S32 max_size = llmax( mTexture->getWidth(), mTexture->getHeight() );
-					max_size <<= cur_discard;
-					const F32 ARTIFICIAL_OFFSET = 2.f;
-					offset_multiple = ARTIFICIAL_OFFSET / (F32)max_size;
-				}
-				else
-				{
-					offset_multiple = 1.f/256;
-				}
-				break;
+			LLFastTimer t(FTM_FACE_GEOM_TEXTURE);
+			gTransformTexCoordProgram.bind();
+			
+			mVertexBuffer->bindForFeedback(0, LLVertexBuffer::TYPE_TEXCOORD0, mGeomIndex, mGeomCount);
+						
+			glBeginTransformFeedback(GL_POINTS);
+			buff->setBuffer(LLVertexBuffer::MAP_TEXCOORD0);
+			push_for_transform(buff, vf.mNumVertices, mGeomCount);
+			glEndTransformFeedback();
 
-				default:  // Standard bumpmap textures.  Assumed to be 256x256
-				offset_multiple = 1.f / 256;
-				break;
-			}
+			bool do_bump = bump_code && mVertexBuffer->hasDataType(LLVertexBuffer::TYPE_TEXCOORD1);
 
-			F32 s_scale = 1.f;
-			F32 t_scale = 1.f;
-			if( tep )
+			if (do_bump)
 			{
-				tep->getScale( &s_scale, &t_scale );
-			}
-			// Use the nudged south when coming from above sun angle, such
-			// that emboss mapping always shows up on the upward faces of cubes when 
-			// it's noon (since a lot of builders build with the sun forced to noon).
-			LLVector3   sun_ray  = gSky.mVOSkyp->mBumpSunDir;
-			LLVector3   moon_ray = gSky.getMoonDirection();
-			LLVector3& primary_light_ray = (sun_ray.mV[VZ] > 0) ? sun_ray : moon_ray;
-
-			bump_s_primary_light_ray.load3((offset_multiple * s_scale * primary_light_ray).mV);
-			bump_t_primary_light_ray.load3((offset_multiple * t_scale * primary_light_ray).mV);
+				mVertexBuffer->bindForFeedback(0, LLVertexBuffer::TYPE_TEXCOORD1, mGeomIndex, mGeomCount);
+				glBeginTransformFeedback(GL_POINTS);
+				buff->setBuffer(LLVertexBuffer::MAP_TEXCOORD0);
+				push_for_transform(buff, vf.mNumVertices, mGeomCount);
+				glEndTransformFeedback();
+			}				
 		}
 
-		U8 texgen = getTextureEntry()->getTexGen();
-		if (rebuild_tcoord && texgen != LLTextureEntry::TEX_GEN_DEFAULT)
-		{ //planar texgen needs binormals
-			mVObjp->getVolume()->genBinormals(f);
+		glBindBufferARB(GL_TRANSFORM_FEEDBACK_BUFFER, 0);
+
+		gGL.popMatrix();
+
+		if (cur_shader)
+		{
+			cur_shader->bind();
 		}
+	}
+	else
+	{
+		//if it's not fullbright and has no normals, bake sunlight based on face normal
+		//bool bake_sunlight = !getTextureEntry()->getFullbright() &&
+		//  !mVertexBuffer->hasDataType(LLVertexBuffer::TYPE_NORMAL);
 
-		U8 tex_mode = 0;
-	
-		if (isState(TEXTURE_ANIM))
+		if (rebuild_tcoord)
 		{
-			LLVOVolume* vobj = (LLVOVolume*) (LLViewerObject*) mVObjp;	
-			tex_mode = vobj->mTexAnimMode;
+			LLFastTimer t(FTM_FACE_GEOM_TEXTURE);
+									
+			//bump setup
+			LLVector4a binormal_dir( -sin_ang, cos_ang, 0.f );
+			LLVector4a bump_s_primary_light_ray(0.f, 0.f, 0.f);
+			LLVector4a bump_t_primary_light_ray(0.f, 0.f, 0.f);
 
-			if (!tex_mode)
+			LLQuaternion bump_quat;
+			if (mDrawablep->isActive())
 			{
-				clearState(TEXTURE_ANIM);
+				bump_quat = LLQuaternion(mDrawablep->getRenderMatrix());
 			}
-			else
+		
+			if (bump_code)
 			{
-				os = ot = 0.f;
-				r = 0.f;
-				cos_ang = 1.f;
-				sin_ang = 0.f;
-				ms = mt = 1.f;
+				mVObjp->getVolume()->genBinormals(f);
+				F32 offset_multiple; 
+				switch( bump_code )
+				{
+					case BE_NO_BUMP:
+					offset_multiple = 0.f;
+					break;
+					case BE_BRIGHTNESS:
+					case BE_DARKNESS:
+					if( mTexture.notNull() && mTexture->hasGLTexture())
+					{
+						// Offset by approximately one texel
+						S32 cur_discard = mTexture->getDiscardLevel();
+						S32 max_size = llmax( mTexture->getWidth(), mTexture->getHeight() );
+						max_size <<= cur_discard;
+						const F32 ARTIFICIAL_OFFSET = 2.f;
+						offset_multiple = ARTIFICIAL_OFFSET / (F32)max_size;
+					}
+					else
+					{
+						offset_multiple = 1.f/256;
+					}
+					break;
 
-				do_xform = false;
+					default:  // Standard bumpmap textures.  Assumed to be 256x256
+					offset_multiple = 1.f / 256;
+					break;
+				}
+
+				F32 s_scale = 1.f;
+				F32 t_scale = 1.f;
+				if( tep )
+				{
+					tep->getScale( &s_scale, &t_scale );
+				}
+				// Use the nudged south when coming from above sun angle, such
+				// that emboss mapping always shows up on the upward faces of cubes when 
+				// it's noon (since a lot of builders build with the sun forced to noon).
+				LLVector3   sun_ray  = gSky.mVOSkyp->mBumpSunDir;
+				LLVector3   moon_ray = gSky.getMoonDirection();
+				LLVector3& primary_light_ray = (sun_ray.mV[VZ] > 0) ? sun_ray : moon_ray;
+
+				bump_s_primary_light_ray.load3((offset_multiple * s_scale * primary_light_ray).mV);
+				bump_t_primary_light_ray.load3((offset_multiple * t_scale * primary_light_ray).mV);
 			}
 
-			if (getVirtualSize() >= MIN_TEX_ANIM_SIZE)
-			{ //don't override texture transform during tc bake
-				tex_mode = 0;
+			U8 texgen = getTextureEntry()->getTexGen();
+			if (rebuild_tcoord && texgen != LLTextureEntry::TEX_GEN_DEFAULT)
+			{ //planar texgen needs binormals
+				mVObjp->getVolume()->genBinormals(f);
 			}
-		}
 
-		LLVector4a scalea;
-		scalea.load3(scale.mV);
+			U8 tex_mode = 0;
+	
+			if (isState(TEXTURE_ANIM))
+			{
+				LLVOVolume* vobj = (LLVOVolume*) (LLViewerObject*) mVObjp;	
+				tex_mode = vobj->mTexAnimMode;
+
+				if (!tex_mode)
+				{
+					clearState(TEXTURE_ANIM);
+				}
+				else
+				{
+					os = ot = 0.f;
+					r = 0.f;
+					cos_ang = 1.f;
+					sin_ang = 0.f;
+					ms = mt = 1.f;
 
-		bool do_bump = bump_code && mVertexBuffer->hasDataType(LLVertexBuffer::TYPE_TEXCOORD1);
-		bool do_tex_mat = tex_mode && mTextureMatrix;
+					do_xform = false;
+				}
 
-		if (!in_atlas && !do_bump)
-		{ //not in atlas or not bump mapped, might be able to do a cheap update
-			mVertexBuffer->getTexCoord0Strider(tex_coords, mGeomIndex, mGeomCount);
+				if (getVirtualSize() >= MIN_TEX_ANIM_SIZE)
+				{ //don't override texture transform during tc bake
+					tex_mode = 0;
+				}
+			}
 
-			if (texgen != LLTextureEntry::TEX_GEN_PLANAR)
-			{
-				LLFastTimer t(FTM_FACE_TEX_QUICK);
-				if (!do_tex_mat)
+			LLVector4a scalea;
+			scalea.load3(scale.mV);
+
+			bool do_bump = bump_code && mVertexBuffer->hasDataType(LLVertexBuffer::TYPE_TEXCOORD1);
+			bool do_tex_mat = tex_mode && mTextureMatrix;
+
+			if (!in_atlas && !do_bump)
+			{ //not in atlas or not bump mapped, might be able to do a cheap update
+				mVertexBuffer->getTexCoord0Strider(tex_coords, mGeomIndex, mGeomCount);
+
+				if (texgen != LLTextureEntry::TEX_GEN_PLANAR)
 				{
-					if (!do_xform)
+					LLFastTimer t(FTM_FACE_TEX_QUICK);
+					if (!do_tex_mat)
 					{
-						LLFastTimer t(FTM_FACE_TEX_QUICK_NO_XFORM);
-						LLVector4a::memcpyNonAliased16((F32*) tex_coords.get(), (F32*) vf.mTexCoords, num_vertices*2*sizeof(F32));
-					}
-					else
-					{
-						LLFastTimer t(FTM_FACE_TEX_QUICK_XFORM);
-						F32* dst = (F32*) tex_coords.get();
-						LLVector4a* src = (LLVector4a*) vf.mTexCoords;
+						if (!do_xform)
+						{
+							LLFastTimer t(FTM_FACE_TEX_QUICK_NO_XFORM);
+							LLVector4a::memcpyNonAliased16((F32*) tex_coords.get(), (F32*) vf.mTexCoords, num_vertices*2*sizeof(F32));
+						}
+						else
+						{
+							LLFastTimer t(FTM_FACE_TEX_QUICK_XFORM);
+							F32* dst = (F32*) tex_coords.get();
+							LLVector4a* src = (LLVector4a*) vf.mTexCoords;
 
-						LLVector4a trans;
-						trans.splat(-0.5f);
+							LLVector4a trans;
+							trans.splat(-0.5f);
 
-						LLVector4a rot0;
-						rot0.set(cos_ang, -sin_ang, cos_ang, -sin_ang);
+							LLVector4a rot0;
+							rot0.set(cos_ang, -sin_ang, cos_ang, -sin_ang);
 
-						LLVector4a rot1;
-						rot1.set(sin_ang, cos_ang, sin_ang, cos_ang);
+							LLVector4a rot1;
+							rot1.set(sin_ang, cos_ang, sin_ang, cos_ang);
 
-						LLVector4a scale;
-						scale.set(ms, mt, ms, mt);
+							LLVector4a scale;
+							scale.set(ms, mt, ms, mt);
 
-						LLVector4a offset;
-						offset.set(os+0.5f, ot+0.5f, os+0.5f, ot+0.5f);
+							LLVector4a offset;
+							offset.set(os+0.5f, ot+0.5f, os+0.5f, ot+0.5f);
 
-						LLVector4Logical mask;
-						mask.clear();
-						mask.setElement<2>();
-						mask.setElement<3>();
+							LLVector4Logical mask;
+							mask.clear();
+							mask.setElement<2>();
+							mask.setElement<3>();
 
-						U32 count = num_vertices/2 + num_vertices%2;
+							U32 count = num_vertices/2 + num_vertices%2;
 
-						for (S32 i = 0; i < count; i++)
+							for (S32 i = 0; i < count; i++)
+							{	
+								LLVector4a res = *src++;
+								xform4a(res, trans, mask, rot0, rot1, offset, scale);
+								res.store4a(dst);
+								dst += 4;
+							}
+						}
+					}
+					else
+					{ //do tex mat, no texgen, no atlas, no bump
+						for (S32 i = 0; i < num_vertices; i++)
 						{	
-							LLVector4a res = *src++;
-							xform4a(res, trans, mask, rot0, rot1, offset, scale);
-							res.store4a(dst);
-							dst += 4;
+							LLVector2 tc(vf.mTexCoords[i]);
+							//LLVector4a& norm = vf.mNormals[i];
+							//LLVector4a& center = *(vf.mCenter);
+
+							LLVector3 tmp(tc.mV[0], tc.mV[1], 0.f);
+							tmp = tmp * *mTextureMatrix;
+							tc.mV[0] = tmp.mV[0];
+							tc.mV[1] = tmp.mV[1];
+							*tex_coords++ = tc;	
 						}
 					}
 				}
 				else
-				{ //do tex mat, no texgen, no atlas, no bump
-					for (S32 i = 0; i < num_vertices; i++)
-					{	
-						LLVector2 tc(vf.mTexCoords[i]);
-						//LLVector4a& norm = vf.mNormals[i];
-						//LLVector4a& center = *(vf.mCenter);
-
-						LLVector3 tmp(tc.mV[0], tc.mV[1], 0.f);
-						tmp = tmp * *mTextureMatrix;
-						tc.mV[0] = tmp.mV[0];
-						tc.mV[1] = tmp.mV[1];
-						*tex_coords++ = tc;	
-					}
-				}
-			}
-			else
-			{ //no bump, no atlas, tex gen planar
-				LLFastTimer t(FTM_FACE_TEX_QUICK_PLANAR);
-				if (do_tex_mat)
-				{
-					for (S32 i = 0; i < num_vertices; i++)
-					{	
-						LLVector2 tc(vf.mTexCoords[i]);
-						LLVector4a& norm = vf.mNormals[i];
-						LLVector4a& center = *(vf.mCenter);
-						LLVector4a vec = vf.mPositions[i];	
-						vec.mul(scalea);
-						planarProjection(tc, norm, center, vec);
+				{ //no bump, no atlas, tex gen planar
+					LLFastTimer t(FTM_FACE_TEX_QUICK_PLANAR);
+					if (do_tex_mat)
+					{
+						for (S32 i = 0; i < num_vertices; i++)
+						{	
+							LLVector2 tc(vf.mTexCoords[i]);
+							LLVector4a& norm = vf.mNormals[i];
+							LLVector4a& center = *(vf.mCenter);
+							LLVector4a vec = vf.mPositions[i];	
+							vec.mul(scalea);
+							planarProjection(tc, norm, center, vec);
 						
-						LLVector3 tmp(tc.mV[0], tc.mV[1], 0.f);
-						tmp = tmp * *mTextureMatrix;
-						tc.mV[0] = tmp.mV[0];
-						tc.mV[1] = tmp.mV[1];
+							LLVector3 tmp(tc.mV[0], tc.mV[1], 0.f);
+							tmp = tmp * *mTextureMatrix;
+							tc.mV[0] = tmp.mV[0];
+							tc.mV[1] = tmp.mV[1];
 				
-						*tex_coords++ = tc;	
+							*tex_coords++ = tc;	
+						}
 					}
-				}
-				else
-				{
-					for (S32 i = 0; i < num_vertices; i++)
-					{	
-						LLVector2 tc(vf.mTexCoords[i]);
-						LLVector4a& norm = vf.mNormals[i];
-						LLVector4a& center = *(vf.mCenter);
-						LLVector4a vec = vf.mPositions[i];	
-						vec.mul(scalea);
-						planarProjection(tc, norm, center, vec);
+					else
+					{
+						for (S32 i = 0; i < num_vertices; i++)
+						{	
+							LLVector2 tc(vf.mTexCoords[i]);
+							LLVector4a& norm = vf.mNormals[i];
+							LLVector4a& center = *(vf.mCenter);
+							LLVector4a vec = vf.mPositions[i];	
+							vec.mul(scalea);
+							planarProjection(tc, norm, center, vec);
 						
-						xform(tc, cos_ang, sin_ang, os, ot, ms, mt);
+							xform(tc, cos_ang, sin_ang, os, ot, ms, mt);
 
-						*tex_coords++ = tc;	
+							*tex_coords++ = tc;	
+						}
 					}
 				}
-			}
 
-			if (map_range)
-			{
-				mVertexBuffer->flush();
+				if (map_range)
+				{
+					mVertexBuffer->flush();
+				}
 			}
-		}
-		else
-		{ //either bump mapped or in atlas, just do the whole expensive loop
-			LLFastTimer t(FTM_FACE_TEX_DEFAULT);
-			mVertexBuffer->getTexCoord0Strider(tex_coords, mGeomIndex, mGeomCount, map_range);
+			else
+			{ //either bump mapped or in atlas, just do the whole expensive loop
+				LLFastTimer t(FTM_FACE_TEX_DEFAULT);
+				mVertexBuffer->getTexCoord0Strider(tex_coords, mGeomIndex, mGeomCount, map_range);
 
-			std::vector<LLVector2> bump_tc;
+				std::vector<LLVector2> bump_tc;
 		
-			for (S32 i = 0; i < num_vertices; i++)
-			{	
-				LLVector2 tc(vf.mTexCoords[i]);
+				for (S32 i = 0; i < num_vertices; i++)
+				{	
+					LLVector2 tc(vf.mTexCoords[i]);
 			
-				LLVector4a& norm = vf.mNormals[i];
+					LLVector4a& norm = vf.mNormals[i];
 				
-				LLVector4a& center = *(vf.mCenter);
+					LLVector4a& center = *(vf.mCenter);
 		   
-				if (texgen != LLTextureEntry::TEX_GEN_DEFAULT)
-				{
-					LLVector4a vec = vf.mPositions[i];
+					if (texgen != LLTextureEntry::TEX_GEN_DEFAULT)
+					{
+						LLVector4a vec = vf.mPositions[i];
 				
-					vec.mul(scalea);
+						vec.mul(scalea);
+
+						switch (texgen)
+						{
+							case LLTextureEntry::TEX_GEN_PLANAR:
+								planarProjection(tc, norm, center, vec);
+								break;
+							case LLTextureEntry::TEX_GEN_SPHERICAL:
+								sphericalProjection(tc, norm, center, vec);
+								break;
+							case LLTextureEntry::TEX_GEN_CYLINDRICAL:
+								cylindricalProjection(tc, norm, center, vec);
+								break;
+							default:
+								break;
+						}		
+					}
 
-					switch (texgen)
+					if (tex_mode && mTextureMatrix)
 					{
-						case LLTextureEntry::TEX_GEN_PLANAR:
-							planarProjection(tc, norm, center, vec);
+						LLVector3 tmp(tc.mV[0], tc.mV[1], 0.f);
+						tmp = tmp * *mTextureMatrix;
+						tc.mV[0] = tmp.mV[0];
+						tc.mV[1] = tmp.mV[1];
+					}
+					else
+					{
+						xform(tc, cos_ang, sin_ang, os, ot, ms, mt);
+					}
+
+					if(in_atlas)
+					{
+						//
+						//manually calculate tex-coord per vertex for varying address modes.
+						//should be removed if shader can handle this.
+						//
+
+						S32 int_part = 0 ;
+						switch(mTexture->getAddressMode())
+						{
+						case LLTexUnit::TAM_CLAMP:
+							if(tc.mV[0] < 0.f)
+							{
+								tc.mV[0] = 0.f ;
+							}
+							else if(tc.mV[0] > 1.f)
+							{
+								tc.mV[0] = 1.f;
+							}
+
+							if(tc.mV[1] < 0.f)
+							{
+								tc.mV[1] = 0.f ;
+							}
+							else if(tc.mV[1] > 1.f)
+							{
+								tc.mV[1] = 1.f;
+							}
 							break;
-						case LLTextureEntry::TEX_GEN_SPHERICAL:
-							sphericalProjection(tc, norm, center, vec);
+						case LLTexUnit::TAM_MIRROR:
+							if(tc.mV[0] < 0.f)
+							{
+								tc.mV[0] = -tc.mV[0] ;
+							}
+							int_part = (S32)tc.mV[0] ;
+							if(int_part & 1) //odd number
+							{
+								tc.mV[0] = int_part + 1 - tc.mV[0] ;
+							}
+							else //even number
+							{
+								tc.mV[0] -= int_part ;
+							}
+
+							if(tc.mV[1] < 0.f)
+							{
+								tc.mV[1] = -tc.mV[1] ;
+							}
+							int_part = (S32)tc.mV[1] ;
+							if(int_part & 1) //odd number
+							{
+								tc.mV[1] = int_part + 1 - tc.mV[1] ;
+							}
+							else //even number
+							{
+								tc.mV[1] -= int_part ;
+							}
 							break;
-						case LLTextureEntry::TEX_GEN_CYLINDRICAL:
-							cylindricalProjection(tc, norm, center, vec);
+						case LLTexUnit::TAM_WRAP:
+							if(tc.mV[0] > 1.f)
+								tc.mV[0] -= (S32)(tc.mV[0] - 0.00001f) ;
+							else if(tc.mV[0] < -1.f)
+								tc.mV[0] -= (S32)(tc.mV[0] + 0.00001f) ;
+
+							if(tc.mV[1] > 1.f)
+								tc.mV[1] -= (S32)(tc.mV[1] - 0.00001f) ;
+							else if(tc.mV[1] < -1.f)
+								tc.mV[1] -= (S32)(tc.mV[1] + 0.00001f) ;
+
+							if(tc.mV[0] < 0.f)
+							{
+								tc.mV[0] = 1.0f + tc.mV[0] ;
+							}
+							if(tc.mV[1] < 0.f)
+							{
+								tc.mV[1] = 1.0f + tc.mV[1] ;
+							}
 							break;
 						default:
 							break;
-					}		
-				}
+						}
+				
+						tc.mV[0] = tcoord_xoffset + tcoord_xscale * tc.mV[0] ;
+						tc.mV[1] = tcoord_yoffset + tcoord_yscale * tc.mV[1] ;
+					}
+				
 
-				if (tex_mode && mTextureMatrix)
-				{
-					LLVector3 tmp(tc.mV[0], tc.mV[1], 0.f);
-					tmp = tmp * *mTextureMatrix;
-					tc.mV[0] = tmp.mV[0];
-					tc.mV[1] = tmp.mV[1];
+					*tex_coords++ = tc;
+					if (do_bump)
+					{
+						bump_tc.push_back(tc);
+					}
 				}
-				else
+
+				if (map_range)
 				{
-					xform(tc, cos_ang, sin_ang, os, ot, ms, mt);
+					mVertexBuffer->flush();
 				}
 
-				if(in_atlas)
+				if (do_bump)
 				{
-					//
-					//manually calculate tex-coord per vertex for varying address modes.
-					//should be removed if shader can handle this.
-					//
-
-					S32 int_part = 0 ;
-					switch(mTexture->getAddressMode())
+					mVertexBuffer->getTexCoord1Strider(tex_coords2, mGeomIndex, mGeomCount, map_range);
+		
+					for (S32 i = 0; i < num_vertices; i++)
 					{
-					case LLTexUnit::TAM_CLAMP:
-						if(tc.mV[0] < 0.f)
-						{
-							tc.mV[0] = 0.f ;
-						}
-						else if(tc.mV[0] > 1.f)
-						{
-							tc.mV[0] = 1.f;
-						}
-
-						if(tc.mV[1] < 0.f)
-						{
-							tc.mV[1] = 0.f ;
-						}
-						else if(tc.mV[1] > 1.f)
-						{
-							tc.mV[1] = 1.f;
-						}
-						break;
-					case LLTexUnit::TAM_MIRROR:
-						if(tc.mV[0] < 0.f)
-						{
-							tc.mV[0] = -tc.mV[0] ;
-						}
-						int_part = (S32)tc.mV[0] ;
-						if(int_part & 1) //odd number
-						{
-							tc.mV[0] = int_part + 1 - tc.mV[0] ;
-						}
-						else //even number
+						LLVector4a tangent;
+						tangent.setCross3(vf.mBinormals[i], vf.mNormals[i]);
+
+						LLMatrix4a tangent_to_object;
+						tangent_to_object.setRows(tangent, vf.mBinormals[i], vf.mNormals[i]);
+						LLVector4a t;
+						tangent_to_object.rotate(binormal_dir, t);
+						LLVector4a binormal;
+						mat_normal.rotate(t, binormal);
+						
+						//VECTORIZE THIS
+						if (mDrawablep->isActive())
 						{
-							tc.mV[0] -= int_part ;
+							LLVector3 t;
+							t.set(binormal.getF32ptr());
+							t *= bump_quat;
+							binormal.load3(t.mV);
 						}
 
-						if(tc.mV[1] < 0.f)
-						{
-							tc.mV[1] = -tc.mV[1] ;
-						}
-						int_part = (S32)tc.mV[1] ;
-						if(int_part & 1) //odd number
-						{
-							tc.mV[1] = int_part + 1 - tc.mV[1] ;
-						}
-						else //even number
-						{
-							tc.mV[1] -= int_part ;
-						}
-						break;
-					case LLTexUnit::TAM_WRAP:
-						if(tc.mV[0] > 1.f)
-							tc.mV[0] -= (S32)(tc.mV[0] - 0.00001f) ;
-						else if(tc.mV[0] < -1.f)
-							tc.mV[0] -= (S32)(tc.mV[0] + 0.00001f) ;
-
-						if(tc.mV[1] > 1.f)
-							tc.mV[1] -= (S32)(tc.mV[1] - 0.00001f) ;
-						else if(tc.mV[1] < -1.f)
-							tc.mV[1] -= (S32)(tc.mV[1] + 0.00001f) ;
-
-						if(tc.mV[0] < 0.f)
-						{
-							tc.mV[0] = 1.0f + tc.mV[0] ;
-						}
-						if(tc.mV[1] < 0.f)
-						{
-							tc.mV[1] = 1.0f + tc.mV[1] ;
-						}
-						break;
-					default:
-						break;
+						binormal.normalize3fast();
+						LLVector2 tc = bump_tc[i];
+						tc += LLVector2( bump_s_primary_light_ray.dot3(tangent).getF32(), bump_t_primary_light_ray.dot3(binormal).getF32() );
+					
+						*tex_coords2++ = tc;
 					}
-				
-					tc.mV[0] = tcoord_xoffset + tcoord_xscale * tc.mV[0] ;
-					tc.mV[1] = tcoord_yoffset + tcoord_yscale * tc.mV[1] ;
-				}
-				
-
-				*tex_coords++ = tc;
-				if (do_bump)
-				{
-					bump_tc.push_back(tc);
-				}
-			}
-
-			if (map_range)
-			{
-				mVertexBuffer->flush();
-			}
 
-			if (do_bump)
-			{
-				mVertexBuffer->getTexCoord1Strider(tex_coords2, mGeomIndex, mGeomCount, map_range);
-		
-				for (S32 i = 0; i < num_vertices; i++)
-				{
-					LLVector4a tangent;
-					tangent.setCross3(vf.mBinormals[i], vf.mNormals[i]);
-
-					LLMatrix4a tangent_to_object;
-					tangent_to_object.setRows(tangent, vf.mBinormals[i], vf.mNormals[i]);
-					LLVector4a t;
-					tangent_to_object.rotate(binormal_dir, t);
-					LLVector4a binormal;
-					mat_normal.rotate(t, binormal);
-						
-					//VECTORIZE THIS
-					if (mDrawablep->isActive())
+					if (map_range)
 					{
-						LLVector3 t;
-						t.set(binormal.getF32ptr());
-						t *= bump_quat;
-						binormal.load3(t.mV);
+						mVertexBuffer->flush();
 					}
-
-					binormal.normalize3fast();
-					LLVector2 tc = bump_tc[i];
-					tc += LLVector2( bump_s_primary_light_ray.dot3(tangent).getF32(), bump_t_primary_light_ray.dot3(binormal).getF32() );
-					
-					*tex_coords2++ = tc;
-				}
-
-				if (map_range)
-				{
-					mVertexBuffer->flush();
 				}
 			}
 		}
-	}
 
-	if (rebuild_pos)
-	{
-		LLFastTimer t(FTM_FACE_GEOM_POSITION);
-		llassert(num_vertices > 0);
+		if (rebuild_pos)
+		{
+			LLFastTimer t(FTM_FACE_GEOM_POSITION);
+			llassert(num_vertices > 0);
 		
-		mVertexBuffer->getVertexStrider(vert, mGeomIndex, mGeomCount, map_range);
+			mVertexBuffer->getVertexStrider(vert, mGeomIndex, mGeomCount, map_range);
 			
 
-		LLMatrix4a mat_vert;
-		mat_vert.loadu(mat_vert_in);
+			LLMatrix4a mat_vert;
+			mat_vert.loadu(mat_vert_in);
 
-		LLVector4a* src = vf.mPositions;
-		volatile F32* dst = (volatile F32*) vert.get();
+			LLVector4a* src = vf.mPositions;
+			volatile F32* dst = (volatile F32*) vert.get();
 
-		volatile F32* end = dst+num_vertices*4;
-		LLVector4a res;
+			volatile F32* end = dst+num_vertices*4;
+			LLVector4a res;
 
-		LLVector4a texIdx;
+			LLVector4a texIdx;
 
-		U8 index = mTextureIndex < 255 ? mTextureIndex : 0;
+			U8 index = mTextureIndex < 255 ? mTextureIndex : 0;
 
-		F32 val = 0.f;
-		U8* vp = (U8*) &val;
-		vp[0] = index;
-		vp[1] = 0;
-		vp[2] = 0;
-		vp[3] = 0;
+			F32 val = 0.f;
+			U8* vp = (U8*) &val;
+			vp[0] = index;
+			vp[1] = 0;
+			vp[2] = 0;
+			vp[3] = 0;
 
-		llassert(index <= LLGLSLShader::sIndexedTextureChannels-1);
+			llassert(index <= LLGLSLShader::sIndexedTextureChannels-1);
 
-		LLVector4Logical mask;
-		mask.clear();
-		mask.setElement<3>();
+			LLVector4Logical mask;
+			mask.clear();
+			mask.setElement<3>();
 		
-		texIdx.set(0,0,0,val);
-
-		{
-			LLFastTimer t(FTM_FACE_POSITION_STORE);
-			LLVector4a tmp;
+			texIdx.set(0,0,0,val);
 
-			do
-			{	
-				mat_vert.affineTransform(*src++, res);
-				tmp.setSelectWithMask(mask, texIdx, res);
-				tmp.store4a((F32*) dst);
-				dst += 4;
+			{
+				LLFastTimer t(FTM_FACE_POSITION_STORE);
+				LLVector4a tmp;
+
+				do
+				{	
+					mat_vert.affineTransform(*src++, res);
+					tmp.setSelectWithMask(mask, texIdx, res);
+					tmp.store4a((F32*) dst);
+					dst += 4;
+				}
+				while(dst < end);
 			}
-			while(dst < end);
-		}
 
-		{
-			LLFastTimer t(FTM_FACE_POSITION_PAD);
-			S32 aligned_pad_vertices = mGeomCount - num_vertices;
-			res.set(res[0], res[1], res[2], 0.f);
+			{
+				LLFastTimer t(FTM_FACE_POSITION_PAD);
+				S32 aligned_pad_vertices = mGeomCount - num_vertices;
+				res.set(res[0], res[1], res[2], 0.f);
+
+				while (aligned_pad_vertices > 0)
+				{
+					--aligned_pad_vertices;
+					res.store4a((F32*) dst);
+					dst += 4;
+				}
+			}
 
-			while (aligned_pad_vertices > 0)
+			if (map_range)
 			{
-				--aligned_pad_vertices;
-				res.store4a((F32*) dst);
-				dst += 4;
+				mVertexBuffer->flush();
 			}
 		}
 
-		if (map_range)
-		{
-			mVertexBuffer->flush();
-		}
-	}
 		
-	if (rebuild_normal)
-	{
-		LLFastTimer t(FTM_FACE_GEOM_NORMAL);
-		mVertexBuffer->getNormalStrider(norm, mGeomIndex, mGeomCount, map_range);
-		F32* normals = (F32*) norm.get();
+		if (rebuild_normal)
+		{
+			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;
-			mat_normal.rotate(vf.mNormals[i], normal);
-			normal.normalize3fast();
-			normal.store4a(normals);
-			normals += 4;
-		}
+			for (S32 i = 0; i < num_vertices; i++)
+			{	
+				LLVector4a normal;
+				mat_normal.rotate(vf.mNormals[i], normal);
+				normal.normalize3fast();
+				normal.store4a(normals);
+				normals += 4;
+			}
 
-		if (map_range)
-		{
-			mVertexBuffer->flush();
+			if (map_range)
+			{
+				mVertexBuffer->flush();
+			}
 		}
-	}
 		
-	if (rebuild_binormal)
-	{
-		LLFastTimer t(FTM_FACE_GEOM_BINORMAL);
-		mVertexBuffer->getBinormalStrider(binorm, mGeomIndex, mGeomCount, map_range);
-		F32* binormals = (F32*) binorm.get();
+		if (rebuild_binormal)
+		{
+			LLFastTimer t(FTM_FACE_GEOM_BINORMAL);
+			mVertexBuffer->getBinormalStrider(binorm, mGeomIndex, mGeomCount, map_range);
+			F32* binormals = (F32*) binorm.get();
 		
-		for (S32 i = 0; i < num_vertices; i++)
-		{	
-			LLVector4a binormal;
-			mat_normal.rotate(vf.mBinormals[i], binormal);
-			binormal.normalize3fast();
-			binormal.store4a(binormals);
-			binormals += 4;
-		}
+			for (S32 i = 0; i < num_vertices; i++)
+			{	
+				LLVector4a binormal;
+				mat_normal.rotate(vf.mBinormals[i], binormal);
+				binormal.normalize3fast();
+				binormal.store4a(binormals);
+				binormals += 4;
+			}
 
-		if (map_range)
-		{
-			mVertexBuffer->flush();
+			if (map_range)
+			{
+				mVertexBuffer->flush();
+			}
 		}
-	}
 	
-	if (rebuild_weights && vf.mWeights)
-	{
-		LLFastTimer t(FTM_FACE_GEOM_WEIGHTS);
-		mVertexBuffer->getWeight4Strider(wght, mGeomIndex, mGeomCount, map_range);
-		F32* weights = (F32*) wght.get();
-		LLVector4a::memcpyNonAliased16(weights, (F32*) vf.mWeights, num_vertices*4*sizeof(F32));
-		if (map_range)
+		if (rebuild_weights && vf.mWeights)
 		{
-			mVertexBuffer->flush();
+			LLFastTimer t(FTM_FACE_GEOM_WEIGHTS);
+			mVertexBuffer->getWeight4Strider(wght, mGeomIndex, mGeomCount, map_range);
+			F32* weights = (F32*) wght.get();
+			LLVector4a::memcpyNonAliased16(weights, (F32*) vf.mWeights, num_vertices*4*sizeof(F32));
+			if (map_range)
+			{
+				mVertexBuffer->flush();
+			}
 		}
-	}
 
-	if (rebuild_color && mVertexBuffer->hasDataType(LLVertexBuffer::TYPE_COLOR) )
-	{
-		LLFastTimer t(FTM_FACE_GEOM_COLOR);
-		mVertexBuffer->getColorStrider(colors, mGeomIndex, mGeomCount, map_range);
+		if (rebuild_color && mVertexBuffer->hasDataType(LLVertexBuffer::TYPE_COLOR) )
+		{
+			LLFastTimer t(FTM_FACE_GEOM_COLOR);
+			mVertexBuffer->getColorStrider(colors, mGeomIndex, mGeomCount, map_range);
 
-		LLVector4a src;
+			LLVector4a src;
 
-		U32 vec[4];
-		vec[0] = vec[1] = vec[2] = vec[3] = color.mAll;
+			U32 vec[4];
+			vec[0] = vec[1] = vec[2] = vec[3] = color.mAll;
 		
-		src.loadua((F32*) vec);
+			src.loadua((F32*) vec);
 
-		F32* dst = (F32*) colors.get();
-		S32 num_vecs = num_vertices/4;
-		if (num_vertices%4 > 0)
-		{
-			++num_vecs;
-		}
+			F32* dst = (F32*) colors.get();
+			S32 num_vecs = num_vertices/4;
+			if (num_vertices%4 > 0)
+			{
+				++num_vecs;
+			}
 
-		for (S32 i = 0; i < num_vecs; i++)
-		{	
-			src.store4a(dst);
-			dst += 4;
-		}
+			for (S32 i = 0; i < num_vecs; i++)
+			{	
+				src.store4a(dst);
+				dst += 4;
+			}
 
-		if (map_range)
-		{
-			mVertexBuffer->flush();
+			if (map_range)
+			{
+				mVertexBuffer->flush();
+			}
 		}
-	}
 
-	if (rebuild_emissive)
-	{
-		LLFastTimer t(FTM_FACE_GEOM_EMISSIVE);
-		LLStrider<LLColor4U> emissive;
-		mVertexBuffer->getEmissiveStrider(emissive, mGeomIndex, mGeomCount, map_range);
+		if (rebuild_emissive)
+		{
+			LLFastTimer t(FTM_FACE_GEOM_EMISSIVE);
+			LLStrider<LLColor4U> emissive;
+			mVertexBuffer->getEmissiveStrider(emissive, mGeomIndex, mGeomCount, map_range);
 
-		U8 glow = (U8) llclamp((S32) (getTextureEntry()->getGlow()*255), 0, 255);
+			U8 glow = (U8) llclamp((S32) (getTextureEntry()->getGlow()*255), 0, 255);
 
-		LLVector4a src;
+			LLVector4a src;
 
 		
-		U32 glow32 = glow |
-					 (glow << 8) |
-					 (glow << 16) |
-					 (glow << 24);
+			U32 glow32 = glow |
+						 (glow << 8) |
+						 (glow << 16) |
+						 (glow << 24);
 
-		U32 vec[4];
-		vec[0] = vec[1] = vec[2] = vec[3] = glow32;
+			U32 vec[4];
+			vec[0] = vec[1] = vec[2] = vec[3] = glow32;
 		
-		src.loadua((F32*) vec);
+			src.loadua((F32*) vec);
 
-		F32* dst = (F32*) emissive.get();
-		S32 num_vecs = num_vertices/4;
-		if (num_vertices%4 > 0)
-		{
-			++num_vecs;
-		}
+			F32* dst = (F32*) emissive.get();
+			S32 num_vecs = num_vertices/4;
+			if (num_vertices%4 > 0)
+			{
+				++num_vecs;
+			}
 
-		for (S32 i = 0; i < num_vecs; i++)
-		{	
-			src.store4a(dst);
-			dst += 4;
-		}
+			for (S32 i = 0; i < num_vecs; i++)
+			{	
+				src.store4a(dst);
+				dst += 4;
+			}
 
-		if (map_range)
-		{
-			mVertexBuffer->flush();
+			if (map_range)
+			{
+				mVertexBuffer->flush();
+			}
 		}
 	}
+
 	if (rebuild_tcoord)
 	{
 		mTexExtents[0].setVec(0,0);
diff --git a/indra/newview/llface.h b/indra/newview/llface.h
index 82e4ab61b7..c31f357693 100644
--- a/indra/newview/llface.h
+++ b/indra/newview/llface.h
@@ -83,6 +83,8 @@ public:
 
 	static void initClass();
 
+	static void cacheFaceInVRAM(const LLVolumeFace& vf);
+
 public:
 	LLFace(LLDrawable* drawablep, LLViewerObject* objp)   { init(drawablep, objp); }
 	~LLFace()  { destroy(); }
diff --git a/indra/newview/llfloatermodelpreview.cpp b/indra/newview/llfloatermodelpreview.cpp
index 7448f2bb2a..b362fa340c 100755
--- a/indra/newview/llfloatermodelpreview.cpp
+++ b/indra/newview/llfloatermodelpreview.cpp
@@ -540,7 +540,7 @@ LLFloaterModelPreview::~LLFloaterModelPreview()
 
 	if (mGLName)
 	{
-		LLImageGL::deleteTextures(1, &mGLName );
+		LLImageGL::deleteTextures(LLTexUnit::TT_TEXTURE, 1, &mGLName );
 	}
 
 	delete mStatusLock;
diff --git a/indra/newview/llspatialpartition.cpp b/indra/newview/llspatialpartition.cpp
index e886f83671..06604fba6f 100644
--- a/indra/newview/llspatialpartition.cpp
+++ b/indra/newview/llspatialpartition.cpp
@@ -85,12 +85,32 @@ static F32 sCurMaxTexPriority = 1.f;
 
 class LLOcclusionQueryPool : public LLGLNamePool
 {
+public:
+	LLOcclusionQueryPool()
+	{
+		mCurQuery = 1;
+	}
+
 protected:
+
+	std::list<GLuint> mAvailableName;
+	GLuint mCurQuery;
+		
 	virtual GLuint allocateName()
 	{
-		GLuint name;
-		glGenQueriesARB(1, &name);
-		return name;
+		GLuint ret = 0;
+
+		if (!mAvailableName.empty())
+		{
+			ret = mAvailableName.front();
+			mAvailableName.pop_front();
+		}
+		else
+		{
+			ret = mCurQuery++;
+		}
+
+		return ret;
 	}
 
 	virtual void releaseName(GLuint name)
@@ -98,7 +118,8 @@ protected:
 #if LL_TRACK_PENDING_OCCLUSION_QUERIES
 		LLSpatialGroup::sPendingQueries.erase(name);
 #endif
-		glDeleteQueriesARB(1, &name);
+		llassert(std::find(mAvailableName.begin(), mAvailableName.end(), name) == mAvailableName.end());
+		mAvailableName.push_back(name);
 	}
 };
 
@@ -687,6 +708,11 @@ void LLSpatialGroup::rebuildGeom()
 	if (!isDead())
 	{
 		mSpatialPartition->rebuildGeom(this);
+
+		if (isState(LLSpatialGroup::MESH_DIRTY))
+		{
+			gPipeline.markMeshDirty(this);
+		}
 	}
 }
 
@@ -1587,7 +1613,7 @@ BOOL LLSpatialGroup::rebound()
 }
 
 static LLFastTimer::DeclareTimer FTM_OCCLUSION_READBACK("Readback Occlusion");
-static LLFastTimer::DeclareTimer FTM_OCCLUSION_WAIT("Wait");
+static LLFastTimer::DeclareTimer FTM_OCCLUSION_WAIT("Occlusion Wait");
 
 void LLSpatialGroup::checkOcclusion()
 {
@@ -1607,7 +1633,9 @@ void LLSpatialGroup::checkOcclusion()
 			{
 				glGetQueryObjectuivARB(mOcclusionQuery[LLViewerCamera::sCurCameraID], GL_QUERY_RESULT_AVAILABLE_ARB, &available);
 
-				if (mOcclusionIssued[LLViewerCamera::sCurCameraID] < gFrameCount)
+				static LLCachedControl<bool> wait_for_query(gSavedSettings, "RenderSynchronousOcclusion");
+
+				if (wait_for_query && mOcclusionIssued[LLViewerCamera::sCurCameraID] < gFrameCount)
 				{ //query was issued last frame, wait until it's available
 					S32 max_loop = 1024;
 					LLFastTimer t(FTM_OCCLUSION_WAIT);
diff --git a/indra/newview/llviewerdisplay.cpp b/indra/newview/llviewerdisplay.cpp
index 001d8e184a..99ed04d1af 100644
--- a/indra/newview/llviewerdisplay.cpp
+++ b/indra/newview/llviewerdisplay.cpp
@@ -755,12 +755,12 @@ void display(BOOL rebuild, F32 zoom_factor, int subfield, BOOL for_snapshot)
 				gTextureList.updateImages(max_image_decode_time);
 			}
 
-			{
+			/*{
 				LLFastTimer t(FTM_IMAGE_UPDATE_DELETE);
 				//remove dead textures from GL
 				LLImageGL::deleteDeadTextures();
 				stop_glerror();
-			}
+			}*/
 		}
 
 		LLGLState::checkStates();
diff --git a/indra/newview/llviewershadermgr.cpp b/indra/newview/llviewershadermgr.cpp
index 0473e2b7c0..485e3af62d 100644
--- a/indra/newview/llviewershadermgr.cpp
+++ b/indra/newview/llviewershadermgr.cpp
@@ -63,6 +63,13 @@ bool				LLViewerShaderMgr::sSkipReload = false;
 
 LLVector4			gShinyOrigin;
 
+//transform shaders
+LLGLSLShader			gTransformPositionProgram;
+LLGLSLShader			gTransformTexCoordProgram;
+LLGLSLShader			gTransformNormalProgram;
+LLGLSLShader			gTransformColorProgram;
+LLGLSLShader			gTransformBinormalProgram;
+
 //utility shaders
 LLGLSLShader	gOcclusionProgram;
 LLGLSLShader	gCustomAlphaProgram;
@@ -438,7 +445,8 @@ void LLViewerShaderMgr::setShaders()
 		S32 wl_class = 2;
 		S32 water_class = 2;
 		S32 deferred_class = 0;
-		
+		S32 transform_class = gGLManager.mHasTransformFeedback ? 1 : 0;
+
 		if (LLFeatureManager::getInstance()->isFeatureAvailable("RenderDeferred") &&
 		    gSavedSettings.getBOOL("RenderDeferred") &&
 			gSavedSettings.getBOOL("RenderAvatarVP") &&
@@ -476,6 +484,7 @@ void LLViewerShaderMgr::setShaders()
 			gSky.mVOSkyp->forceSkyUpdate();
 		}
 
+		
 		// Load lighting shaders
 		mVertexShaderLevel[SHADER_LIGHTING] = light_class;
 		mVertexShaderLevel[SHADER_INTERFACE] = light_class;
@@ -485,6 +494,7 @@ void LLViewerShaderMgr::setShaders()
 		mVertexShaderLevel[SHADER_EFFECT] = effect_class;
 		mVertexShaderLevel[SHADER_WINDLIGHT] = wl_class;
 		mVertexShaderLevel[SHADER_DEFERRED] = deferred_class;
+		mVertexShaderLevel[SHADER_TRANSFORM] = transform_class;
 
 		BOOL loaded = loadBasicShaders();
 
@@ -518,6 +528,11 @@ void LLViewerShaderMgr::setShaders()
 			
 			if (loaded)
 			{
+				loaded = loadTransformShaders();
+			}
+
+			if (loaded)
+			{
 				// Load max avatar shaders to set the max level
 				mVertexShaderLevel[SHADER_AVATAR] = 3;
 				mMaxAvatarShaderLevel = 3;
@@ -733,6 +748,12 @@ void LLViewerShaderMgr::unloadShaders()
 	gDeferredSkinnedBumpProgram.unload();
 	gDeferredSkinnedAlphaProgram.unload();
 
+	gTransformPositionProgram.unload();
+	gTransformTexCoordProgram.unload();
+	gTransformNormalProgram.unload();
+	gTransformColorProgram.unload();
+	gTransformBinormalProgram.unload();
+
 	mVertexShaderLevel[SHADER_LIGHTING] = 0;
 	mVertexShaderLevel[SHADER_OBJECT] = 0;
 	mVertexShaderLevel[SHADER_AVATAR] = 0;
@@ -741,6 +762,7 @@ void LLViewerShaderMgr::unloadShaders()
 	mVertexShaderLevel[SHADER_INTERFACE] = 0;
 	mVertexShaderLevel[SHADER_EFFECT] = 0;
 	mVertexShaderLevel[SHADER_WINDLIGHT] = 0;
+	mVertexShaderLevel[SHADER_TRANSFORM] = 0;
 
 	gPipeline.mVertexShadersLoaded = 0;
 }
@@ -2763,6 +2785,95 @@ BOOL LLViewerShaderMgr::loadShadersWindLight()
 	return success;
 }
 
+BOOL LLViewerShaderMgr::loadTransformShaders()
+{
+	BOOL success = TRUE;
+	
+	if (mVertexShaderLevel[SHADER_TRANSFORM] < 1)
+	{
+		gTransformPositionProgram.unload();
+		gTransformTexCoordProgram.unload();
+		gTransformNormalProgram.unload();
+		gTransformColorProgram.unload();
+		gTransformBinormalProgram.unload();
+		return TRUE;
+	}
+
+	if (success)
+	{
+		gTransformPositionProgram.mName = "Position Transform Shader";
+		gTransformPositionProgram.mShaderFiles.clear();
+		gTransformPositionProgram.mShaderFiles.push_back(make_pair("transform/positionV.glsl", GL_VERTEX_SHADER_ARB));
+		gTransformPositionProgram.mShaderLevel = mVertexShaderLevel[SHADER_TRANSFORM];
+
+		const char* varyings[] = {
+			"position_out",
+			"texture_index_out",
+		};
+	
+		success = gTransformPositionProgram.createShader(NULL, NULL, 2, varyings);
+	}
+
+	if (success)
+	{
+		gTransformTexCoordProgram.mName = "TexCoord Transform Shader";
+		gTransformTexCoordProgram.mShaderFiles.clear();
+		gTransformTexCoordProgram.mShaderFiles.push_back(make_pair("transform/texcoordV.glsl", GL_VERTEX_SHADER_ARB));
+		gTransformTexCoordProgram.mShaderLevel = mVertexShaderLevel[SHADER_TRANSFORM];
+
+		const char* varyings[] = {
+			"texcoord_out",
+		};
+	
+		success = gTransformTexCoordProgram.createShader(NULL, NULL, 1, varyings);
+	}
+
+	if (success)
+	{
+		gTransformNormalProgram.mName = "Normal Transform Shader";
+		gTransformNormalProgram.mShaderFiles.clear();
+		gTransformNormalProgram.mShaderFiles.push_back(make_pair("transform/normalV.glsl", GL_VERTEX_SHADER_ARB));
+		gTransformNormalProgram.mShaderLevel = mVertexShaderLevel[SHADER_TRANSFORM];
+
+		const char* varyings[] = {
+			"normal_out",
+		};
+	
+		success = gTransformNormalProgram.createShader(NULL, NULL, 1, varyings);
+	}
+
+	if (success)
+	{
+		gTransformColorProgram.mName = "Color Transform Shader";
+		gTransformColorProgram.mShaderFiles.clear();
+		gTransformColorProgram.mShaderFiles.push_back(make_pair("transform/colorV.glsl", GL_VERTEX_SHADER_ARB));
+		gTransformColorProgram.mShaderLevel = mVertexShaderLevel[SHADER_TRANSFORM];
+
+		const char* varyings[] = {
+			"color_out",
+		};
+	
+		success = gTransformColorProgram.createShader(NULL, NULL, 1, varyings);
+	}
+
+	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];
+
+		const char* varyings[] = {
+			"binormal_out",
+		};
+	
+		success = gTransformBinormalProgram.createShader(NULL, NULL, 1, varyings);
+	}
+
+	
+	return success;
+}
+
 std::string LLViewerShaderMgr::getShaderDirPrefix(void)
 {
 	return gDirUtilp->getExpandedFilename(LL_PATH_APP_SETTINGS, "shaders/class");
diff --git a/indra/newview/llviewershadermgr.h b/indra/newview/llviewershadermgr.h
index d10aba85c7..03d686e07e 100644
--- a/indra/newview/llviewershadermgr.h
+++ b/indra/newview/llviewershadermgr.h
@@ -54,6 +54,7 @@ public:
 	BOOL loadShadersWater();
 	BOOL loadShadersInterface();
 	BOOL loadShadersWindLight();
+	BOOL loadTransformShaders();
 
 	std::vector<S32> mVertexShaderLevel;
 	S32	mMaxAvatarShaderLevel;
@@ -69,6 +70,7 @@ public:
 		SHADER_WINDLIGHT,
 		SHADER_WATER,
 		SHADER_DEFERRED,
+		SHADER_TRANSFORM,
 		SHADER_COUNT
 	};
 
@@ -209,6 +211,15 @@ inline bool operator != (LLViewerShaderMgr::shader_iter const & a, LLViewerShade
 
 extern LLVector4			gShinyOrigin;
 
+//transform shaders
+extern LLGLSLShader			gTransformPositionProgram;
+extern LLGLSLShader			gTransformTexCoordProgram;
+extern LLGLSLShader			gTransformNormalProgram;
+extern LLGLSLShader			gTransformColorProgram;
+extern LLGLSLShader			gTransformBinormalProgram;
+
+
+
 //utility shaders
 extern LLGLSLShader			gOcclusionProgram;
 extern LLGLSLShader			gCustomAlphaProgram;
diff --git a/indra/newview/llvoavatar.cpp b/indra/newview/llvoavatar.cpp
index c221c7fdd8..083ad622cd 100644
--- a/indra/newview/llvoavatar.cpp
+++ b/indra/newview/llvoavatar.cpp
@@ -896,7 +896,7 @@ void LLVOAvatar::deleteLayerSetCaches(bool clearAll)
 		}
 		if (mBakedTextureDatas[i].mMaskTexName)
 		{
-			glDeleteTextures(1, (GLuint*)&(mBakedTextureDatas[i].mMaskTexName));
+			LLImageGL::deleteTextures(LLTexUnit::TT_TEXTURE, 1, (GLuint*)&(mBakedTextureDatas[i].mMaskTexName));
 			mBakedTextureDatas[i].mMaskTexName = 0 ;
 		}
 	}
@@ -7394,7 +7394,7 @@ void LLVOAvatar::onBakedTextureMasksLoaded( BOOL success, LLViewerFetchedTexture
 			}
 
 			U32 gl_name;
-			LLImageGL::generateTextures(1, &gl_name );
+			LLImageGL::generateTextures(LLTexUnit::TT_TEXTURE, 1, &gl_name );
 			stop_glerror();
 
 			gGL.getTexUnit(0)->bindManual(LLTexUnit::TT_TEXTURE, gl_name);
@@ -7431,7 +7431,7 @@ void LLVOAvatar::onBakedTextureMasksLoaded( BOOL success, LLViewerFetchedTexture
 						maskData->mLastDiscardLevel = discard_level;
 						if (self->mBakedTextureDatas[baked_index].mMaskTexName)
 						{
-							LLImageGL::deleteTextures(1, &(self->mBakedTextureDatas[baked_index].mMaskTexName));
+							LLImageGL::deleteTextures(LLTexUnit::TT_TEXTURE, 1, &(self->mBakedTextureDatas[baked_index].mMaskTexName));
 						}
 						self->mBakedTextureDatas[baked_index].mMaskTexName = gl_name;
 						found_texture_id = true;
diff --git a/indra/newview/llvoavatarself.cpp b/indra/newview/llvoavatarself.cpp
index f063653cc5..b6d4d6b56f 100644
--- a/indra/newview/llvoavatarself.cpp
+++ b/indra/newview/llvoavatarself.cpp
@@ -2605,7 +2605,7 @@ void LLVOAvatarSelf::deleteScratchTextures()
 		 namep; 
 		 namep = sScratchTexNames.getNextData() )
 	{
-		LLImageGL::deleteTextures(1, (U32 *)namep );
+		LLImageGL::deleteTextures(LLTexUnit::TT_TEXTURE, 1, (U32 *)namep );
 		stop_glerror();
 	}
 
diff --git a/indra/newview/llvopartgroup.cpp b/indra/newview/llvopartgroup.cpp
index b6adc776cc..fca97987a2 100644
--- a/indra/newview/llvopartgroup.cpp
+++ b/indra/newview/llvopartgroup.cpp
@@ -125,7 +125,7 @@ S32 LLVOPartGroup::findAvailableVBSlot()
 void LLVOPartGroup::freeVBSlot(S32 idx)
 {
 	llassert(idx < LL_MAX_PARTICLE_COUNT && idx >= 0);
-	llassert(sVBSlotCursor > sVBSlotFree);
+	//llassert(sVBSlotCursor > sVBSlotFree);
 
 	if (sVBSlotCursor > sVBSlotFree)
 	{
diff --git a/indra/newview/llvovolume.cpp b/indra/newview/llvovolume.cpp
index 6d4565ec2f..6bb4e9e63b 100644
--- a/indra/newview/llvovolume.cpp
+++ b/indra/newview/llvovolume.cpp
@@ -1089,9 +1089,33 @@ BOOL LLVOVolume::setVolume(const LLVolumeParams &params_in, const S32 detail, bo
 			}
 		}
 
+
+		static LLCachedControl<bool> use_transform_feedback(gSavedSettings, "RenderUseTransformFeedback");
+
+		bool cache_in_vram = use_transform_feedback && gTransformPositionProgram.mProgramObject &&
+			(!mVolumeImpl || !mVolumeImpl->isVolumeUnique());
+
+		if (cache_in_vram)
+		{ //this volume might be used as source data for a transform object, put it in vram
+			LLVolume* volume = getVolume();
+			for (S32 i = 0; i < volume->getNumFaces(); ++i)
+			{
+				const LLVolumeFace& face = volume->getVolumeFace(i);
+				if (face.mVertexBuffer.notNull())
+				{ //already cached
+					break;
+				}
+				volume->genBinormals(i);
+				LLFace::cacheFaceInVRAM(face);
+			}
+		}
+		
+
 		return TRUE;
 	}
 
+
+
 	return FALSE;
 }
 
diff --git a/indra/newview/pipeline.cpp b/indra/newview/pipeline.cpp
index a691302cb9..a76a32f834 100644
--- a/indra/newview/pipeline.cpp
+++ b/indra/newview/pipeline.cpp
@@ -51,6 +51,10 @@
 // newview includes
 #include "llagent.h"
 #include "llagentcamera.h"
+#include "llappviewer.h"
+#include "lltexturecache.h"
+#include "lltexturefetch.h"
+#include "llimageworker.h"
 #include "lldrawable.h"
 #include "lldrawpoolalpha.h"
 #include "lldrawpoolavatar.h"
@@ -403,9 +407,11 @@ LLPipeline::LLPipeline() :
 	mInitialized(FALSE),
 	mVertexShadersEnabled(FALSE),
 	mVertexShadersLoaded(0),
+	mTransformFeedbackPrimitives(0),
 	mRenderDebugFeatureMask(0),
 	mRenderDebugMask(0),
 	mOldRenderDebugMask(0),
+	mMeshDirtyQueryObject(0),
 	mGroupQ1Locked(false),
 	mGroupQ2Locked(false),
 	mResetVertexBuffers(false),
@@ -693,6 +699,12 @@ void LLPipeline::destroyGL()
 	{
 		LLVertexBuffer::sEnableVBOs = FALSE;
 	}
+
+	if (mMeshDirtyQueryObject)
+	{
+		glDeleteQueriesARB(1, &mMeshDirtyQueryObject);
+		mMeshDirtyQueryObject = 0;
+	}
 }
 
 static LLFastTimer::DeclareTimer FTM_RESIZE_SCREEN_TEXTURE("Resize Screen Texture");
@@ -1028,19 +1040,19 @@ void LLPipeline::releaseGLBuffers()
 	
 	if (mNoiseMap)
 	{
-		LLImageGL::deleteTextures(1, &mNoiseMap);
+		LLImageGL::deleteTextures(LLTexUnit::TT_TEXTURE, 1, &mNoiseMap);
 		mNoiseMap = 0;
 	}
 
 	if (mTrueNoiseMap)
 	{
-		LLImageGL::deleteTextures(1, &mTrueNoiseMap);
+		LLImageGL::deleteTextures(LLTexUnit::TT_TEXTURE, 1, &mTrueNoiseMap);
 		mTrueNoiseMap = 0;
 	}
 
 	if (mLightFunc)
 	{
-		LLImageGL::deleteTextures(1, &mLightFunc);
+		LLImageGL::deleteTextures(LLTexUnit::TT_TEXTURE, 1, &mLightFunc);
 		mLightFunc = 0;
 	}
 
@@ -1131,7 +1143,7 @@ void LLPipeline::createGLBuffers()
 				noise[i].mV[2] = ll_frand()*scaler+1.f-scaler/2.f;
 			}
 
-			LLImageGL::generateTextures(1, &mNoiseMap);
+			LLImageGL::generateTextures(LLTexUnit::TT_TEXTURE, 1, &mNoiseMap);
 			
 			gGL.getTexUnit(0)->bindManual(LLTexUnit::TT_TEXTURE, mNoiseMap);
 			LLImageGL::setManualImage(LLTexUnit::getInternalType(LLTexUnit::TT_TEXTURE), 0, GL_RGB16F_ARB, noiseRes, noiseRes, GL_RGB, GL_FLOAT, noise, false);
@@ -1147,7 +1159,7 @@ void LLPipeline::createGLBuffers()
 				noise[i] = ll_frand()*2.0-1.0;
 			}
 
-			LLImageGL::generateTextures(1, &mTrueNoiseMap);
+			LLImageGL::generateTextures(LLTexUnit::TT_TEXTURE, 1, &mTrueNoiseMap);
 			gGL.getTexUnit(0)->bindManual(LLTexUnit::TT_TEXTURE, mTrueNoiseMap);
 			LLImageGL::setManualImage(LLTexUnit::getInternalType(LLTexUnit::TT_TEXTURE), 0, GL_RGB16F_ARB, noiseRes, noiseRes, GL_RGB,GL_FLOAT, noise, false);
 			gGL.getTexUnit(0)->setTextureFilteringOption(LLTexUnit::TFO_POINT);
@@ -1183,7 +1195,7 @@ void LLPipeline::createGLBuffers()
 				}
 			}
 
-			LLImageGL::generateTextures(1, &mLightFunc);
+			LLImageGL::generateTextures(LLTexUnit::TT_TEXTURE, 1, &mLightFunc);
 			gGL.getTexUnit(0)->bindManual(LLTexUnit::TT_TEXTURE, mLightFunc);
 			LLImageGL::setManualImage(LLTexUnit::getInternalType(LLTexUnit::TT_TEXTURE), 0, GL_R8, lightResX, lightResY, GL_RED, GL_UNSIGNED_BYTE, lg, false);
 			gGL.getTexUnit(0)->setTextureAddressMode(LLTexUnit::TAM_CLAMP);
@@ -2829,6 +2841,11 @@ void LLPipeline::processPartitionQ()
 	mPartitionQ.clear();
 }
 
+void LLPipeline::markMeshDirty(LLSpatialGroup* group)
+{
+	mMeshDirtyGroup.push_back(group);
+}
+
 void LLPipeline::markRebuild(LLSpatialGroup* group, BOOL priority)
 {
 	LLMemType mt(LLMemType::MTYPE_PIPELINE);
@@ -3437,15 +3454,43 @@ void LLPipeline::postSort(LLCamera& camera)
 			}
 		}
 	}
+	
+	//flush particle VB
+	LLVOPartGroup::sVB->flush();
+
+	/*bool use_transform_feedback = gTransformPositionProgram.mProgramObject && !mMeshDirtyGroup.empty();
+
+	if (use_transform_feedback)
+	{ //place a query around potential transform feedback code for synchronization
+		mTransformFeedbackPrimitives = 0;
+
+		if (!mMeshDirtyQueryObject)
+		{
+			glGenQueriesARB(1, &mMeshDirtyQueryObject);
+		}
+
 		
+		glBeginQueryARB(GL_TRANSFORM_FEEDBACK_PRIMITIVES_WRITTEN, mMeshDirtyQueryObject);
+	}*/
+
+	//pack vertex buffers for groups that chose to delay their updates
+	for (LLSpatialGroup::sg_vector_t::iterator iter = mMeshDirtyGroup.begin(); iter != mMeshDirtyGroup.end(); ++iter)
+	{
+		(*iter)->rebuildMesh();
+	}
+
+	/*if (use_transform_feedback)
+	{
+		glEndQueryARB(GL_TRANSFORM_FEEDBACK_PRIMITIVES_WRITTEN);
+	}*/
+	
+	mMeshDirtyGroup.clear();
+
 	if (!sShadowRender)
 	{
 		std::sort(sCull->beginAlphaGroups(), sCull->endAlphaGroups(), LLSpatialGroup::CompareDepthGreater());
 	}
 
-	//flush particle VB
-	LLVOPartGroup::sVB->flush();
-
 	llpushcallstacks ;
 	// only render if the flag is set. The flag is only set if we are in edit mode or the toggle is set in the menus
 	if (LLFloaterReg::instanceVisible("beacons") && !sShadowRender)
@@ -3532,6 +3577,33 @@ void LLPipeline::postSort(LLCamera& camera)
 		}
 	}
 
+	/*static LLFastTimer::DeclareTimer FTM_TRANSFORM_WAIT("Transform Fence");
+	static LLFastTimer::DeclareTimer FTM_TRANSFORM_DO_WORK("Transform Work");
+	if (use_transform_feedback)
+	{ //using transform feedback, wait for transform feedback to complete
+		LLFastTimer t(FTM_TRANSFORM_WAIT);
+
+		S32 done = 0;
+		//glGetQueryivARB(GL_TRANSFORM_FEEDBACK_PRIMITIVES_WRITTEN, GL_CURRENT_QUERY, &count);
+		
+		glGetQueryObjectivARB(mMeshDirtyQueryObject, GL_QUERY_RESULT_AVAILABLE, &done);
+		
+		while (!done)
+		{ 
+			{
+				LLFastTimer t(FTM_TRANSFORM_DO_WORK);
+				F32 max_time = llmin(gFrameIntervalSeconds*10.f, 1.f);
+				//do some useful work while we wait
+				LLAppViewer::getTextureCache()->update(max_time); // unpauses the texture cache thread
+				LLAppViewer::getImageDecodeThread()->update(max_time); // unpauses the image thread
+				LLAppViewer::getTextureFetch()->update(max_time); // unpauses the texture fetch thread
+			}
+			glGetQueryObjectivARB(mMeshDirtyQueryObject, GL_QUERY_RESULT_AVAILABLE, &done);
+		}
+
+		mTransformFeedbackPrimitives = 0;
+	}*/
+						
 	//LLSpatialGroup::sNoDelete = FALSE;
 	llpushcallstacks ;
 }
diff --git a/indra/newview/pipeline.h b/indra/newview/pipeline.h
index 252fe1346c..9eebe3831f 100644
--- a/indra/newview/pipeline.h
+++ b/indra/newview/pipeline.h
@@ -161,6 +161,7 @@ public:
 	void		markRebuild(LLSpatialGroup* group, BOOL priority = FALSE);
 	void        markRebuild(LLDrawable *drawablep, LLDrawable::EDrawableFlags flag = LLDrawable::REBUILD_ALL, BOOL priority = FALSE);
 	void		markPartitionMove(LLDrawable* drawablep);
+	void		markMeshDirty(LLSpatialGroup* group);
 
 	//get the object between start and end that's closest to start.
 	LLViewerObject* lineSegmentIntersectInWorld(const LLVector3& start, const LLVector3& end,
@@ -592,6 +593,7 @@ public:
 	BOOL					mVertexShadersEnabled;
 	S32						mVertexShadersLoaded; // 0 = no, 1 = yes, -1 = failed
 
+	U32						mTransformFeedbackPrimitives; //number of primitives expected to be generated by transform feedback
 protected:
 	BOOL					mRenderTypeEnabled[NUM_RENDER_TYPES];
 	std::stack<std::string> mRenderTypeEnableStack;
@@ -649,6 +651,9 @@ protected:
 	LLSpatialGroup::sg_vector_t		mGroupQ1; //priority
 	LLSpatialGroup::sg_vector_t		mGroupQ2; // non-priority
 
+	LLSpatialGroup::sg_vector_t		mMeshDirtyGroup; //groups that need rebuildMesh called
+	U32 mMeshDirtyQueryObject;
+
 	LLDrawable::drawable_list_t		mPartitionQ; //drawables that need to update their spatial partition radius 
 
 	bool mGroupQ2Locked;