Pull and merge from ssh://hg@bitbucket.org/lindenlab/viewer-release.

17 files changed, 494 insertions, 194 deletions

diff --git a/indra/llrender/llcubemap.cpp b/indra/llrender/llcubemap.cpp
index 45a3b18179..362452d837 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, GL_RGB8, 1, &texname);
 
 			for (int i = 0; i < 6; i++)
 			{
diff --git a/indra/llrender/llfontgl.cpp b/indra/llrender/llfontgl.cpp
index fccbf37a8d..4dc2fcd714 100644
--- a/indra/llrender/llfontgl.cpp
+++ b/indra/llrender/llfontgl.cpp
@@ -422,6 +422,16 @@ S32 LLFontGL::renderUTF8(const std::string &text, S32 begin_offset, S32 x, S32 y
 }
 
 // font metrics - override for LLFontFreetype that returns units of virtual pixels
+F32 LLFontGL::getAscenderHeight() const
+{ 
+	return mFontFreetype->getAscenderHeight() / sScaleY;
+}
+
+F32 LLFontGL::getDescenderHeight() const
+{ 
+	return mFontFreetype->getDescenderHeight() / sScaleY;
+}
+
 S32 LLFontGL::getLineHeight() const
 { 
 	return llceil(mFontFreetype->getAscenderHeight() / sScaleY) + llceil(mFontFreetype->getDescenderHeight() / sScaleY);
diff --git a/indra/llrender/llfontgl.h b/indra/llrender/llfontgl.h
index 74bdbb43e7..5ed5d2c4eb 100644
--- a/indra/llrender/llfontgl.h
+++ b/indra/llrender/llfontgl.h
@@ -115,6 +115,8 @@ public:
 	S32 renderUTF8(const std::string &text, S32 begin_offset, S32 x, S32 y, const LLColor4 &color, HAlign halign, VAlign valign, U8 style = NORMAL, ShadowType shadow = NO_SHADOW) const;
 
 	// font metrics - override for LLFontFreetype that returns units of virtual pixels
+	F32 getAscenderHeight() const;
+	F32 getDescenderHeight() const;
 	S32 getLineHeight() const;
 
 	S32 getWidth(const std::string& utf8text) const;
diff --git a/indra/llrender/llgl.cpp b/indra/llrender/llgl.cpp
index 628a8d6131..0b56b3889c 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),
@@ -558,7 +565,8 @@ bool LLGLManager::initGL()
 	parse_gl_version( &mDriverVersionMajor, 
 		&mDriverVersionMinor, 
 		&mDriverVersionRelease, 
-		&mDriverVersionVendorString );
+		&mDriverVersionVendorString,
+		&mGLVersionString);
 
 	mGLVersion = mDriverVersionMajor + mDriverVersionMinor * .1f;
 
@@ -938,7 +946,6 @@ void LLGLManager::initExtensions()
 	mHasMultitexture = glh_init_extensions("GL_ARB_multitexture");
 	mHasATIMemInfo = ExtensionExists("GL_ATI_meminfo", gGLHExts.mSysExts);
 	mHasNVXMemInfo = ExtensionExists("GL_NVX_gpu_memory_info", gGLHExts.mSysExts);
-	mHasMipMapGeneration = glh_init_extensions("GL_SGIS_generate_mipmap");
 	mHasSeparateSpecularColor = glh_init_extensions("GL_EXT_separate_specular_color");
 	mHasAnisotropic = glh_init_extensions("GL_EXT_texture_filter_anisotropic");
 	glh_init_extensions("GL_ARB_texture_cube_map");
@@ -963,11 +970,14 @@ void LLGLManager::initExtensions()
 							ExtensionExists("GL_EXT_packed_depth_stencil", gGLHExts.mSysExts);
 #endif
 	
+	mHasMipMapGeneration = mHasFramebufferObject || mGLVersion >= 1.4f;
+
 	mHasDrawBuffers = ExtensionExists("GL_ARB_draw_buffers", gGLHExts.mSysExts);
 	mHasBlendFuncSeparate = ExtensionExists("GL_EXT_blend_func_separate", gGLHExts.mSysExts);
 	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
@@ -1207,7 +1217,14 @@ 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");
@@ -1964,6 +1981,7 @@ LLGLState::LLGLState(LLGLenum state, S32 enabled) :
 			case GL_COLOR_MATERIAL:
 			case GL_FOG:
 			case GL_LINE_STIPPLE:
+			case GL_POLYGON_STIPPLE:
 				mState = 0;
 				break;
 		}
@@ -2052,7 +2070,7 @@ void LLGLManager::initGLStates()
 
 ////////////////////////////////////////////////////////////////////////////////
 
-void parse_gl_version( S32* major, S32* minor, S32* release, std::string* vendor_specific )
+void parse_gl_version( S32* major, S32* minor, S32* release, std::string* vendor_specific, std::string* version_string )
 {
 	// GL_VERSION returns a null-terminated string with the format: 
 	// <major>.<minor>[.<release>] [<vendor specific>]
@@ -2068,6 +2086,8 @@ void parse_gl_version( S32* major, S32* minor, S32* release, std::string* vendor
 		return;
 	}
 
+	version_string->assign(version);
+
 	std::string ver_copy( version );
 	S32 len = (S32)strlen( version );	/* Flawfinder: ignore */
 	S32 i = 0;
@@ -2429,3 +2449,65 @@ 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 5a33c98708..964495a3ab 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;
@@ -141,6 +142,7 @@ public:
 	S32 mGLSLVersionMajor;
 	S32 mGLSLVersionMinor;
 	std::string mDriverVersionVendorString;
+	std::string mGLVersionString;
 
 	S32 mVRAM; // VRAM in MB
 	S32 mGLMaxVertexRange;
@@ -417,13 +419,38 @@ public:
 	virtual void updateGL() = 0;
 };
 
+const U32 FENCE_WAIT_TIME_NANOSECONDS = 1000;  //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"
 
 void init_glstates();
 
-void parse_gl_version( S32* major, S32* minor, S32* release, std::string* vendor_specific );
+void parse_gl_version( S32* major, S32* minor, S32* release, std::string* vendor_specific, std::string* version_string );
 
 extern BOOL gClothRipple;
 extern BOOL gHeadlessClient;
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..7cbf39096e 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 +174,13 @@ BOOL LLGLSLShader::createShader(vector<string> * attributes,
 		mFeatures.mIndexedTextureChannels = llmin(mFeatures.mIndexedTextureChannels, 1);
 	}
 
+#ifdef GL_INTERLEAVED_ATTRIBS
+	if (varying_count > 0 && varyings)
+	{
+		glTransformFeedbackVaryings(mProgramObject, varying_count, varyings, GL_INTERLEAVED_ATTRIBS);
+	}
+#endif
+
 	// 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 c04a3f6b41..659d3ca409 100644
--- a/indra/llrender/llimagegl.cpp
+++ b/indra/llrender/llimagegl.cpp
@@ -42,6 +42,10 @@
 //----------------------------------------------------------------------------
 const F32 MIN_TEXTURE_LIFETIME = 10.f;
 
+//which power of 2 is i?
+//assumes i is a power of 2 > 0
+U32 wpo2(U32 i);
+
 //statics
 
 U32 LLImageGL::sUniqueCount				= 0;
@@ -50,7 +54,8 @@ S32 LLImageGL::sGlobalTextureMemoryInBytes		= 0;
 S32 LLImageGL::sBoundTextureMemoryInBytes		= 0;
 S32 LLImageGL::sCurBoundTextureMemory	= 0;
 S32 LLImageGL::sCount					= 0;
-std::list<U32> LLImageGL::sDeadTextureList;
+LLImageGL::dead_texturelist_t LLImageGL::sDeadTextureList[LLTexUnit::TT_NONE];
+U32 LLImageGL::sCurTexName = 1;
 
 BOOL LLImageGL::sGlobalUseAnisotropic	= FALSE;
 F32 LLImageGL::sLastFrameTime			= 0.f;
@@ -416,6 +421,7 @@ void LLImageGL::init(BOOL usemipmaps)
 	mTarget = GL_TEXTURE_2D;
 	mBindTarget = LLTexUnit::TT_TEXTURE;
 	mHasMipMaps = false;
+	mMipLevels = -1;
 
 	mIsResident = 0;
 
@@ -606,8 +612,24 @@ void LLImageGL::setImage(const U8* data_in, BOOL data_hasmips)
 		is_compressed = true;
 	}
 
+	
+	
+	if (mUseMipMaps)
+	{
+		//set has mip maps to true before binding image so tex parameters get set properly
+		gGL.getTexUnit(0)->unbind(mBindTarget);
+		mHasMipMaps = true;
+		mTexOptionsDirty = true;
+		setFilteringOption(LLTexUnit::TFO_ANISOTROPIC);
+	}
+	else
+	{
+		mHasMipMaps = false;
+	}
+	
 	llverify(gGL.getTexUnit(0)->bind(this));
 	
+	
 	if (mUseMipMaps)
 	{
 		if (data_hasmips)
@@ -620,6 +642,9 @@ void LLImageGL::setImage(const U8* data_in, BOOL data_hasmips)
 				S32 w = getWidth(d);
 				S32 h = getHeight(d);
 				S32 gl_level = d-mCurrentDiscardLevel;
+
+				mMipLevels = llmax(mMipLevels, gl_level);
+
 				if (d > mCurrentDiscardLevel)
 				{
 					data_in -= dataFormatBytes(mFormatPrimary, w, h); // see above comment
@@ -662,10 +687,6 @@ void LLImageGL::setImage(const U8* data_in, BOOL data_hasmips)
 		{
 			if (mAutoGenMips)
 			{
-				if (!gGLManager.mHasFramebufferObject)
-				{
-					glTexParameteri(LLTexUnit::getInternalType(mBindTarget), GL_GENERATE_MIPMAP_SGIS, TRUE);
-				}
 				stop_glerror();
 				{
 // 					LLFastTimer t2(FTM_TEMP4);
@@ -679,6 +700,11 @@ void LLImageGL::setImage(const U8* data_in, BOOL data_hasmips)
 					S32 w = getWidth(mCurrentDiscardLevel);
 					S32 h = getHeight(mCurrentDiscardLevel);
 
+					mMipLevels = wpo2(llmax(w, h));
+
+					//use legacy mipmap generation mode
+					glTexParameteri(mTarget, GL_GENERATE_MIPMAP, GL_TRUE);
+					
 					LLImageGL::setManualImage(mTarget, 0, mFormatInternal,
 								 w, h, 
 								 mFormatPrimary, mFormatType,
@@ -694,16 +720,10 @@ void LLImageGL::setImage(const U8* data_in, BOOL data_hasmips)
 						stop_glerror();
 					}
 				}
-
-				if (gGLManager.mHasFramebufferObject)
-				{
-					glGenerateMipmap(LLTexUnit::getInternalType(mBindTarget));
-				}
 			}
 			else
 			{
 				// Create mips by hand
-				// about 30% faster than autogen on ATI 9800, 50% slower on nVidia 4800
 				// ~4x faster than gluBuild2DMipmaps
 				S32 width = getWidth(mCurrentDiscardLevel);
 				S32 height = getHeight(mCurrentDiscardLevel);
@@ -713,6 +733,9 @@ void LLImageGL::setImage(const U8* data_in, BOOL data_hasmips)
 				const U8* cur_mip_data = 0;
 				S32 prev_mip_size = 0;
 				S32 cur_mip_size = 0;
+				
+				mMipLevels = nummips;
+
 				for (int m=0; m<nummips; m++)
 				{
 					if (m==0)
@@ -777,10 +800,10 @@ void LLImageGL::setImage(const U8* data_in, BOOL data_hasmips)
 		{
 			llerrs << "Compressed Image has mipmaps but data does not (can not auto generate compressed mips)" << llendl;
 		}
-		mHasMipMaps = true;
 	}
 	else
 	{
+		mMipLevels = 0;
 		S32 w = getWidth();
 		S32 h = getHeight();
 		if (is_compressed)
@@ -812,7 +835,6 @@ void LLImageGL::setImage(const U8* data_in, BOOL data_hasmips)
 			}
 
 		}
-		mHasMipMaps = false;
 	}
 	stop_glerror();
 	mGLTextureCreated = true;
@@ -1025,23 +1047,65 @@ 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, U32 format, S32 numTextures, U32 *textures)
 {
-	glGenTextures(numTextures, (GLuint*)textures);
+	bool empty = true;
+
+	dead_texturelist_t::iterator iter = sDeadTextureList[type].find(format);
+	
+	if (iter != sDeadTextureList[type].end())
+	{
+		empty = iter->second.empty();
+	}
+	
+	for (S32 i = 0; i < numTextures; ++i)
+	{
+		if (!empty)
+		{
+			textures[i] = iter->second.front();
+			iter->second.pop_front();
+			empty = iter->second.empty();
+		}
+		else
+		{
+			textures[i] = sCurTexName++;
+		}
+	}
 }
 
 // static
-void LLImageGL::deleteTextures(S32 numTextures, U32 *textures, bool immediate)
+void LLImageGL::deleteTextures(LLTexUnit::eTextureType type, U32 format, S32 mip_levels, S32 numTextures, U32 *textures, bool immediate)
 {
-	for (S32 i = 0; i < numTextures; i++)
+	if (gGLManager.mInited)
 	{
-		sDeadTextureList.push_back(textures[i]);
-	}
+		if (format == 0 ||  type == LLTexUnit::TT_CUBE_MAP || mip_levels == -1)
+		{ //unknown internal format or unknown number of mip levels, not safe to reuse
+			glDeleteTextures(numTextures, textures);
+		}
+		else
+		{
+			for (S32 i = 0; i < numTextures; ++i)
+			{ //remove texture from VRAM by setting its size to zero
+				for (S32 j = 0; j <= mip_levels; j++)
+				{
+					gGL.getTexUnit(0)->bindManual(type, textures[i]);
+
+					glTexImage2D(LLTexUnit::getInternalType(type), j, format, 0, 0, 0, GL_RGBA, GL_UNSIGNED_BYTE, NULL);
+				}
+
+				llassert(std::find(sDeadTextureList[type][format].begin(),
+								   sDeadTextureList[type][format].end(), textures[i]) == 
+								   sDeadTextureList[type][format].end());
 
-	if (immediate)
+				sDeadTextureList[type][format].push_back(textures[i]);
+			}	
+		}
+	}
+	
+	/*if (immediate)
 	{
 		LLImageGL::deleteDeadTextures();
-	}
+	}*/
 }
 
 // static
@@ -1166,10 +1230,11 @@ BOOL LLImageGL::createGLTexture()
 
 	if(mTexName)
 	{
-		glDeleteTextures(1, (reinterpret_cast<GLuint*>(&mTexName))) ;
+		LLImageGL::deleteTextures(mBindTarget, mFormatInternal, mMipLevels, 1, (reinterpret_cast<GLuint*>(&mTexName))) ;
 	}
 	
-	glGenTextures(1, (GLuint*)&mTexName);
+
+	LLImageGL::generateTextures(mBindTarget, mFormatInternal, 1, &mTexName);
 	stop_glerror();
 	if (!mTexName)
 	{
@@ -1282,7 +1347,7 @@ BOOL LLImageGL::createGLTexture(S32 discard_level, const U8* data_in, BOOL data_
 	}
 	else
 	{
-		LLImageGL::generateTextures(1, &mTexName);
+		LLImageGL::generateTextures(mBindTarget, mFormatInternal, 1, &mTexName);
 		stop_glerror();
 		{
 			llverify(gGL.getTexUnit(0)->bind(this));
@@ -1327,7 +1392,7 @@ BOOL LLImageGL::createGLTexture(S32 discard_level, const U8* data_in, BOOL data_
 	{
 		sGlobalTextureMemoryInBytes -= mTextureMemory;
 
-		LLImageGL::deleteTextures(1, &old_name);
+		LLImageGL::deleteTextures(mBindTarget, mFormatInternal, mMipLevels, 1, &old_name);
 
 		stop_glerror();
 	}
@@ -1456,7 +1521,7 @@ void LLImageGL::deleteDeadTextures()
 {
 	bool reset = false;
 
-	while (!sDeadTextureList.empty())
+	/*while (!sDeadTextureList.empty())
 	{
 		GLuint tex = sDeadTextureList.front();
 		sDeadTextureList.pop_front();
@@ -1478,7 +1543,7 @@ void LLImageGL::deleteDeadTextures()
 		
 		glDeleteTextures(1, &tex);
 		stop_glerror();
-	}
+	}*/
 
 	if (reset)
 	{
@@ -1496,7 +1561,7 @@ void LLImageGL::destroyGLTexture()
 			mTextureMemory = 0;
 		}
 		
-		LLImageGL::deleteTextures(1, &mTexName);			
+		LLImageGL::deleteTextures(mBindTarget,  mFormatInternal, mMipLevels, 1, &mTexName);			
 		mCurrentDiscardLevel = -1 ; //invalidate mCurrentDiscardLevel.
 		mTexName = 0;		
 		mGLTextureCreated = FALSE ;
diff --git a/indra/llrender/llimagegl.h b/indra/llrender/llimagegl.h
index f34b9fa91a..e118c28c1b 100644
--- a/indra/llrender/llimagegl.h
+++ b/indra/llrender/llimagegl.h
@@ -45,8 +45,16 @@ class LLImageGL : public LLRefCount
 {
 	friend class LLTexUnit;
 public:
-	static std::list<U32> sDeadTextureList;
+	static U32 sCurTexName;
 
+	//previously used but now available texture names
+	// sDeadTextureList[<usage>][<internal format>]
+	typedef std::map<U32, std::list<U32> > dead_texturelist_t;
+	static dead_texturelist_t sDeadTextureList[LLTexUnit::TT_NONE];
+
+	// These 2 functions replace glGenTextures() and glDeleteTextures()
+	static void generateTextures(LLTexUnit::eTextureType type, U32 format, S32 numTextures, U32 *textures);
+	static void deleteTextures(LLTexUnit::eTextureType type, U32 format, S32 mip_levels, S32 numTextures, U32 *textures, bool immediate = false);
 	static void deleteDeadTextures();
 
 	// Size calculation
@@ -96,10 +104,6 @@ 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() ;
@@ -217,7 +221,8 @@ protected:
 	LLGLenum mTarget;		// Normally GL_TEXTURE2D, sometimes something else (ex. cube maps)
 	LLTexUnit::eTextureType mBindTarget;	// Normally TT_TEXTURE, sometimes something else (ex. cube maps)
 	bool mHasMipMaps;
-	
+	S32 mMipLevels;
+
 	LLGLboolean mIsResident;
 	
 	S8 mComponents;
diff --git a/indra/llrender/llrender.cpp b/indra/llrender/llrender.cpp
index 93bac4c779..348c1eb1b7 100644
--- a/indra/llrender/llrender.cpp
+++ b/indra/llrender/llrender.cpp
@@ -408,12 +408,14 @@ void LLTexUnit::unbind(eTextureType type)
 
 	if (mIndex < 0) return;
 
+	//always flush and activate for consistency 
+	//   some code paths assume unbind always flushes and sets the active texture
+	gGL.flush();
+	activate();
+
 	// Disabled caching of binding state.
 	if (mCurrTexType == type)
 	{
-		gGL.flush();
-
-		activate();
 		mCurrTexture = 0;
 		if (LLGLSLShader::sNoFixedFunction && type == LLTexUnit::TT_TEXTURE)
 		{
@@ -464,11 +466,25 @@ void LLTexUnit::setTextureFilteringOption(LLTexUnit::eTextureFilterOptions optio
 	} 
 	else if (option >= TFO_BILINEAR)
 	{
-		glTexParameteri(sGLTextureType[mCurrTexType], GL_TEXTURE_MIN_FILTER, GL_LINEAR);
+		if (mHasMipMaps)
+		{
+			glTexParameteri(sGLTextureType[mCurrTexType], GL_TEXTURE_MIN_FILTER, GL_LINEAR_MIPMAP_NEAREST);
+		}
+		else
+		{
+			glTexParameteri(sGLTextureType[mCurrTexType], GL_TEXTURE_MIN_FILTER, GL_LINEAR);
+		}
 	}
 	else
 	{
-		glTexParameteri(sGLTextureType[mCurrTexType], GL_TEXTURE_MIN_FILTER, GL_NEAREST);
+		if (mHasMipMaps)
+		{
+			glTexParameteri(sGLTextureType[mCurrTexType], GL_TEXTURE_MIN_FILTER, GL_NEAREST_MIPMAP_NEAREST);
+		}
+		else
+		{
+			glTexParameteri(sGLTextureType[mCurrTexType], GL_TEXTURE_MIN_FILTER, GL_NEAREST);
+		}
 	}
 
 	if (gGLManager.mHasAnisotropic)
diff --git a/indra/llrender/llrendertarget.cpp b/indra/llrender/llrendertarget.cpp
index 780f1dc484..99f0da330c 100644
--- a/indra/llrender/llrendertarget.cpp
+++ b/indra/llrender/llrendertarget.cpp
@@ -55,7 +55,6 @@ bool LLRenderTarget::sUseFBO = false;
 LLRenderTarget::LLRenderTarget() :
 	mResX(0),
 	mResY(0),
-	mTex(0),
 	mFBO(0),
 	mDepth(0),
 	mStencil(0),
@@ -135,7 +134,7 @@ bool LLRenderTarget::addColorAttachment(U32 color_fmt)
 	}
 
 	U32 tex;
-	LLImageGL::generateTextures(1, &tex);
+	LLImageGL::generateTextures(mUsage, color_fmt, 1, &tex);
 	gGL.getTexUnit(0)->bindManual(mUsage, tex);
 
 	stop_glerror();
@@ -193,6 +192,7 @@ bool LLRenderTarget::addColorAttachment(U32 color_fmt)
 	}
 
 	mTex.push_back(tex);
+	mInternalFormat.push_back(color_fmt);
 
 	if (gDebugGL)
 	{ //bind and unbind to validate target
@@ -217,7 +217,7 @@ bool LLRenderTarget::allocateDepth()
 	}
 	else
 	{
-		LLImageGL::generateTextures(1, &mDepth);
+		LLImageGL::generateTextures(mUsage, GL_DEPTH_COMPONENT24, 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, 0, 0, 1, &mDepth, true);
 			stop_glerror();
 		}
 		mDepth = 0;
@@ -326,8 +326,9 @@ void LLRenderTarget::release()
 	if (mTex.size() > 0)
 	{
 		sBytesAllocated -= mResX*mResY*4*mTex.size();
-		LLImageGL::deleteTextures(mTex.size(), &mTex[0], true);
+		LLImageGL::deleteTextures(mUsage, mInternalFormat[0], 0, mTex.size(), &mTex[0], true);
 		mTex.clear();
+		mInternalFormat.clear();
 	}
 	
 	mResX = mResY = 0;
diff --git a/indra/llrender/llrendertarget.h b/indra/llrender/llrendertarget.h
index 2735ab21c5..8360458840 100644
--- a/indra/llrender/llrendertarget.h
+++ b/indra/llrender/llrendertarget.h
@@ -142,6 +142,7 @@ protected:
 	U32 mResX;
 	U32 mResY;
 	std::vector<U32> mTex;
+	std::vector<U32> mInternalFormat;
 	U32 mFBO;
 	U32 mDepth;
 	bool mStencil;
diff --git a/indra/llrender/llshadermgr.cpp b/indra/llrender/llshadermgr.cpp
index 5a6e6cab3e..a9248d4d73 100644
--- a/indra/llrender/llshadermgr.cpp
+++ b/indra/llrender/llshadermgr.cpp
@@ -702,7 +702,7 @@ GLhandleARB LLShaderMgr::loadShaderFile(const std::string& filename, S32 & shade
 
 		if (texture_index_channels > 1)
 		{
-			text[count++] = strdup("VARYING_FLAT ivec4 vary_texture_index;\n");
+			text[count++] = strdup("VARYING_FLAT int vary_texture_index;\n");
 		}
 
 		text[count++] = strdup("vec4 diffuseLookup(vec2 texcoord)\n");
@@ -716,20 +716,33 @@ GLhandleARB LLShaderMgr::loadShaderFile(const std::string& filename, S32 & shade
 		}
 		else if (major_version > 1 || minor_version >= 30)
 		{  //switches are supported in GLSL 1.30 and later
-			text[count++] = strdup("\tvec4 ret = vec4(1,0,1,1);\n");
-			text[count++] = strdup("\tswitch (vary_texture_index.r)\n");
-			text[count++] = strdup("\t{\n");
-		
-			//switch body
-			for (S32 i = 0; i < texture_index_channels; ++i)
-			{
-				std::string case_str = llformat("\t\tcase %d: ret = texture2D(tex%d, texcoord); break;\n", i, i);
-				text[count++] = strdup(case_str.c_str());
+			if (gGLManager.mIsNVIDIA)
+			{ //switches are unreliable on some NVIDIA drivers
+				for (U32 i = 0; i < texture_index_channels; ++i)
+				{
+					std::string if_string = llformat("\t%sif (vary_texture_index == %d) { return texture2D(tex%d, texcoord); }\n", i > 0 ? "else " : "", i, i); 
+					text[count++] = strdup(if_string.c_str());
+				}
+				text[count++] = strdup("\treturn vec4(1,0,1,1);\n");
+				text[count++] = strdup("}\n");
 			}
+			else
+			{
+				text[count++] = strdup("\tvec4 ret = vec4(1,0,1,1);\n");
+				text[count++] = strdup("\tswitch (vary_texture_index)\n");
+				text[count++] = strdup("\t{\n");
+		
+				//switch body
+				for (S32 i = 0; i < texture_index_channels; ++i)
+				{
+					std::string case_str = llformat("\t\tcase %d: return texture2D(tex%d, texcoord);\n", i, i);
+					text[count++] = strdup(case_str.c_str());
+				}
 
-			text[count++] = strdup("\t}\n");
-			text[count++] = strdup("\treturn ret;\n");
-			text[count++] = strdup("}\n");
+				text[count++] = strdup("\t}\n");
+				text[count++] = strdup("\treturn ret;\n");
+				text[count++] = strdup("}\n");
+			}
 		}
 		else
 		{ //should never get here.  Indexed texture rendering requires GLSL 1.30 or later 
@@ -1026,6 +1039,9 @@ void LLShaderMgr::initAttribsAndUniforms()
 	mReservedUniforms.push_back("size");
 	mReservedUniforms.push_back("falloff");
 
+	mReservedUniforms.push_back("box_center");
+	mReservedUniforms.push_back("box_size");
+
 
 	mReservedUniforms.push_back("minLuminance");
 	mReservedUniforms.push_back("maxExtractAlpha");
diff --git a/indra/llrender/llshadermgr.h b/indra/llrender/llshadermgr.h
index f792faa8f0..7a16b7c20f 100644
--- a/indra/llrender/llshadermgr.h
+++ b/indra/llrender/llshadermgr.h
@@ -97,6 +97,8 @@ public:
 		LIGHT_CENTER,
 		LIGHT_SIZE,
 		LIGHT_FALLOFF,
+		BOX_CENTER,
+		BOX_SIZE,
 
 		GLOW_MIN_LUMINANCE,
 		GLOW_MAX_EXTRACT_ALPHA,
diff --git a/indra/llrender/llvertexbuffer.cpp b/indra/llrender/llvertexbuffer.cpp
index 823c6b9dc5..80752231d7 100644
--- a/indra/llrender/llvertexbuffer.cpp
+++ b/indra/llrender/llvertexbuffer.cpp
@@ -41,9 +41,7 @@
 #if LL_DARWIN
 #define LL_VBO_POOLING 1
 #else
-#define LL_VBO_POOLING 0
 #endif
-
 //Next Highest Power Of Two
 //helper function, returns first number > v that is a power of 2, or v if v is already a power of 2
 U32 nhpo2(U32 v)
@@ -71,6 +69,7 @@ U32 wpo2(U32 i)
 
 
 const U32 LL_VBO_BLOCK_SIZE = 2048;
+const U32 LL_VBO_POOL_MAX_SEED_SIZE = 256*1024;
 
 U32 vbo_block_size(U32 size)
 { //what block size will fit size?
@@ -83,6 +82,7 @@ U32 vbo_block_index(U32 size)
 	return vbo_block_size(size)/LL_VBO_BLOCK_SIZE;
 }
 
+const U32 LL_VBO_POOL_SEED_COUNT = vbo_block_index(LL_VBO_POOL_MAX_SEED_SIZE);
 
 
 //============================================================================
@@ -95,6 +95,11 @@ 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;
 
@@ -119,69 +124,55 @@ 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()
+	return ret;
+}
+
+void LLVBOPool::deleteBuffer(U32 name)
+{
+	if (gGLManager.mInited)
 	{
-#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
-	}
+		LLVertexBuffer::unbind();
+
+		glBindBufferARB(mType, name);
+		glBufferDataARB(mType, 0, NULL, mUsage);
 
+		llassert(std::find(mGLNamePool.begin(), mGLNamePool.end(), name) == mGLNamePool.end());
+
+		mGLNamePool.push_back(name);
+
+		glBindBufferARB(mType, 0);
+	}
+}
 
-};
 
+LLVBOPool::LLVBOPool(U32 vboUsage, U32 vboType)
+: mUsage(vboUsage), mType(vboType)
+{
+	mMissCount.resize(LL_VBO_POOL_SEED_COUNT);
+	std::fill(mMissCount.begin(), mMissCount.end(), 0);
+}
 
-volatile U8* LLVBOPool::allocate(U32& name, U32 size)
+volatile U8* LLVBOPool::allocate(U32& name, U32 size, bool for_seed)
 {
 	llassert(vbo_block_size(size) == size);
 	
 	volatile U8* ret = NULL;
 
-#if LL_VBO_POOLING
-
 	U32 i = vbo_block_index(size);
 
 	if (mFreeList.size() <= i)
@@ -189,12 +180,18 @@ volatile U8* LLVBOPool::allocate(U32& name, U32 size)
 		mFreeList.resize(i+1);
 	}
 
-	if (mFreeList[i].empty())
+	if (mFreeList[i].empty() || for_seed)
 	{
 		//make a new buffer
-		glGenBuffersARB(1, &name);
+		name = genBuffer();
+		
 		glBindBufferARB(mType, name);
 
+		if (!for_seed && i < LL_VBO_POOL_SEED_COUNT)
+		{ //record this miss
+			mMissCount[i]++;	
+		}
+
 		if (mType == GL_ARRAY_BUFFER_ARB)
 		{
 			LLVertexBuffer::sAllocatedBytes += size;
@@ -215,6 +212,25 @@ volatile U8* LLVBOPool::allocate(U32& name, U32 size)
 		}
 
 		glBindBufferARB(mType, 0);
+
+		if (for_seed)
+		{ //put into pool for future use
+			llassert(mFreeList.size() > i);
+
+			Record rec;
+			rec.mGLName = name;
+			rec.mClientData = ret;
+	
+			if (mType == GL_ARRAY_BUFFER_ARB)
+			{
+				sBytesPooled += size;
+			}
+			else
+			{
+				sIndexBytesPooled += size;
+			}
+			mFreeList[i].push_back(rec);
+		}
 	}
 	else
 	{
@@ -232,33 +248,6 @@ volatile U8* LLVBOPool::allocate(U32& name, U32 size)
 
 		mFreeList[i].pop_front();
 	}
-#else //no pooling
-
-	glGenBuffersARB(1, &name);
-	glBindBufferARB(mType, name);
-
-	if (mType == GL_ARRAY_BUFFER_ARB)
-	{
-		LLVertexBuffer::sAllocatedBytes += size;
-	}
-	else
-	{
-		LLVertexBuffer::sAllocatedIndexBytes += size;
-	}
-
-	if (LLVertexBuffer::sDisableVBOMapping || mUsage != GL_DYNAMIC_DRAW_ARB)
-	{
-		glBufferDataARB(mType, size, 0, mUsage);
-		ret = (U8*) ll_aligned_malloc_16(size);
-	}
-	else
-	{ //always use a true hint of static draw when allocating non-client-backed buffers
-		glBufferDataARB(mType, size, 0, GL_STATIC_DRAW_ARB);
-	}
-
-	glBindBufferARB(mType, 0);
-
-#endif
 
 	return ret;
 }
@@ -267,50 +256,47 @@ void LLVBOPool::release(U32 name, volatile U8* buffer, U32 size)
 {
 	llassert(vbo_block_size(size) == size);
 
-#if LL_VBO_POOLING
-
-	U32 i = vbo_block_index(size);
-
-	llassert(mFreeList.size() > i);
+	deleteBuffer(name);
+	ll_aligned_free_16((U8*) buffer);
 
-	Record rec;
-	rec.mGLName = name;
-	rec.mClientData = buffer;
-	
-	if (buffer == NULL)
+	if (mType == GL_ARRAY_BUFFER_ARB)
 	{
-		glDeleteBuffersARB(1, &rec.mGLName);
+		LLVertexBuffer::sAllocatedBytes -= size;
 	}
 	else
 	{
-		if (mType == GL_ARRAY_BUFFER_ARB)
-		{
-			sBytesPooled += size;
-		}
-		else
-		{
-			sIndexBytesPooled += size;
-		}
-		mFreeList[i].push_back(rec);
+		LLVertexBuffer::sAllocatedIndexBytes -= size;
 	}
-#else //no pooling
-	glDeleteBuffersARB(1, &name);
-	ll_aligned_free_16((U8*) buffer);
+}
 
-	if (mType == GL_ARRAY_BUFFER_ARB)
+void LLVBOPool::seedPool()
+{
+	U32 dummy_name = 0;
+
+	if (mFreeList.size() < LL_VBO_POOL_SEED_COUNT)
 	{
-		LLVertexBuffer::sAllocatedBytes -= size;
+		mFreeList.resize(LL_VBO_POOL_SEED_COUNT);
 	}
-	else
+
+	for (U32 i = 0; i < LL_VBO_POOL_SEED_COUNT; i++)
 	{
-		LLVertexBuffer::sAllocatedIndexBytes -= size;
+		if (mMissCount[i] > mFreeList[i].size())
+		{ 
+			U32 size = i*LL_VBO_BLOCK_SIZE;
+		
+			S32 count = mMissCount[i] - mFreeList[i].size();
+			for (U32 j = 0; j < count; ++j)
+			{
+				allocate(dummy_name, size, true);
+			}
+		}
 	}
-#endif
 }
 
+
 void LLVBOPool::cleanup()
 {
-	U32 size = 1;
+	U32 size = LL_VBO_BLOCK_SIZE;
 
 	for (U32 i = 0; i < mFreeList.size(); ++i)
 	{
@@ -320,8 +306,8 @@ void LLVBOPool::cleanup()
 		{
 			Record& r = l.front();
 
-			glDeleteBuffersARB(1, &r.mGLName);
-
+			deleteBuffer(r.mGLName);
+			
 			if (r.mClientData)
 			{
 				ll_aligned_free_16((void*) r.mClientData);
@@ -341,8 +327,11 @@ void LLVBOPool::cleanup()
 			}
 		}
 
-		size *= 2;
+		size += LL_VBO_BLOCK_SIZE;
 	}
+
+	//reset miss counts
+	std::fill(mMissCount.begin(), mMissCount.end(), 0);
 }
 
 
@@ -376,6 +365,41 @@ 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
+	{
+#ifdef GL_ARB_vertex_array_object
+		glGenVertexArrays(1, &ret);
+#endif
+	}
+
+	return ret;		
+}
+
+//static
+void LLVertexBuffer::releaseVAOName(U32 name)
+{
+	sAvailableVAOName.push_back(name);
+}
+
+
+//static
+void LLVertexBuffer::seedPools()
+{
+	sStreamVBOPool.seedPool();
+	sDynamicVBOPool.seedPool();
+	sStreamIBOPool.seedPool();
+	sDynamicIBOPool.seedPool();
+}
 
 //static
 void LLVertexBuffer::setupClientArrays(U32 data_mask)
@@ -985,7 +1009,7 @@ LLVertexBuffer::~LLVertexBuffer()
 	if (mGLArray)
 	{
 #if GL_ARB_vertex_array_object
-		glDeleteVertexArrays(1, &mGLArray);
+		releaseVAOName(mGLArray);
 #endif
 	}
 
@@ -1211,10 +1235,10 @@ void LLVertexBuffer::updateNumVerts(S32 nverts)
 
 	llassert(nverts >= 0);
 
-	if (nverts >= 65535)
+	if (nverts > 65536)
 	{
 		llwarns << "Vertex buffer overflow!" << llendl;
-		nverts = 65535;
+		nverts = 65536;
 	}
 
 	U32 needed_size = calcOffsets(mTypeMask, mOffsets, nverts);
@@ -1270,7 +1294,7 @@ 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();
 		}
@@ -1306,7 +1330,7 @@ void LLVertexBuffer::setupVertexArray()
 		1, //TYPE_WEIGHT,
 		4, //TYPE_WEIGHT4,
 		4, //TYPE_CLOTHWEIGHT,
-		4, //TYPE_TEXTURE_INDEX
+		1, //TYPE_TEXTURE_INDEX
 	};
 
 	U32 attrib_type[] =
@@ -1323,7 +1347,7 @@ void LLVertexBuffer::setupVertexArray()
 		GL_FLOAT, //TYPE_WEIGHT,
 		GL_FLOAT, //TYPE_WEIGHT4,
 		GL_FLOAT, //TYPE_CLOTHWEIGHT,
-		GL_UNSIGNED_BYTE, //TYPE_TEXTURE_INDEX
+		GL_UNSIGNED_INT, //TYPE_TEXTURE_INDEX
 	};
 
 	bool attrib_integer[] = 
@@ -2140,6 +2164,16 @@ void LLVertexBuffer::flush()
 	}
 }
 
+// bind for transform feedback (quick 'n dirty)
+void LLVertexBuffer::bindForFeedback(U32 channel, U32 type, U32 index, U32 count)
+{
+#ifdef GL_TRANSFORM_FEEDBACK_BUFFER
+	U32 offset = mOffsets[type] + sTypeSize[type]*index;
+	U32 size= (sTypeSize[type]*count);
+	glBindBufferRange(GL_TRANSFORM_FEEDBACK_BUFFER, channel, mGLBuffer, offset, size);
+#endif
+}
+
 // Set for rendering
 void LLVertexBuffer::setBuffer(U32 data_mask)
 {
@@ -2291,10 +2325,10 @@ void LLVertexBuffer::setupVertexBuffer(U32 data_mask)
 	stop_glerror();
 	volatile U8* base = useVBOs() ? (U8*) mAlignedOffset : mMappedData;
 
-	/*if ((data_mask & mTypeMask) != data_mask)
+	if (gDebugGL && ((data_mask & mTypeMask) != data_mask))
 	{
 		llerrs << "LLVertexBuffer::setupVertexBuffer missing required components for supplied data mask." << llendl;
-	}*/
+	}
 
 	if (LLGLSLShader::sNoFixedFunction)
 	{
@@ -2370,7 +2404,7 @@ void LLVertexBuffer::setupVertexBuffer(U32 data_mask)
 #if !LL_DARWIN
 			S32 loc = TYPE_TEXTURE_INDEX;
 			void *ptr = (void*) (base + mOffsets[TYPE_VERTEX] + 12);
-			glVertexAttribIPointer(loc, 4, GL_UNSIGNED_BYTE, LLVertexBuffer::sTypeSize[TYPE_VERTEX], ptr);
+			glVertexAttribIPointer(loc, 1, GL_UNSIGNED_INT, LLVertexBuffer::sTypeSize[TYPE_VERTEX], ptr);
 #endif
 		}
 		if (data_mask & MAP_VERTEX)
diff --git a/indra/llrender/llvertexbuffer.h b/indra/llrender/llvertexbuffer.h
index 7477dec3ad..11fa4ab6a0 100644
--- a/indra/llrender/llvertexbuffer.h
+++ b/indra/llrender/llvertexbuffer.h
@@ -57,23 +57,28 @@ public:
 	static U32 sBytesPooled;
 	static U32 sIndexBytesPooled;
 	
-	LLVBOPool(U32 vboUsage, U32 vboType)
-		: mUsage(vboUsage)
-		, mType(vboType)
-	{}
+	static U32 sCurGLName;
 
+	LLVBOPool(U32 vboUsage, U32 vboType);
+		
 	const U32 mUsage;
 	const U32 mType;
 
 	//size MUST be a power of 2
-	volatile U8* allocate(U32& name, U32 size);
+	volatile U8* allocate(U32& name, U32 size, bool for_seed = false);
 	
 	//size MUST be the size provided to allocate that returned the given name
 	void release(U32 name, volatile U8* buffer, U32 size);
 	
+	//batch allocate buffers to be provided to the application on demand
+	void seedPool();
+
 	//destroy all records in mFreeList
 	void cleanup();
 
+	U32 genBuffer();
+	void deleteBuffer(U32 name);
+
 	class Record
 	{
 	public:
@@ -81,17 +86,15 @@ 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 
 class LLPrivateMemoryPool;
@@ -125,13 +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);
 
@@ -208,7 +220,6 @@ protected:
 	void 	destroyGLIndices();
 	void	updateNumVerts(S32 nverts);
 	void	updateNumIndices(S32 nindices); 
-	bool	useVBOs() const;
 	void	unmapBuffer();
 		
 public:
@@ -218,6 +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
@@ -240,12 +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; }