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

9 files changed, 233 insertions, 228 deletions

diff --git a/indra/llrender/llgl.cpp b/indra/llrender/llgl.cpp
index 197bc2b422..628a8d6131 100644
--- a/indra/llrender/llgl.cpp
+++ b/indra/llrender/llgl.cpp
@@ -94,6 +94,10 @@ void APIENTRY gl_debug_callback(GLenum source,
 	llwarns << "Severity: " << std::hex << severity << llendl;
 	llwarns << "Message: " << message << llendl;
 	llwarns << "-----------------------" << llendl;
+	if (severity == GL_DEBUG_SEVERITY_HIGH_ARB)
+	{
+		llerrs << "Halting on GL Error" << llendl;
+	}
 }
 #endif
 
@@ -572,6 +576,15 @@ bool LLGLManager::initGL()
 #endif
 	}
 
+	if (mGLVersion >= 2.1f && LLImageGL::sCompressTextures)
+	{ //use texture compression
+		glHint(GL_TEXTURE_COMPRESSION_HINT, GL_NICEST);
+	}
+	else
+	{ //GL version is < 3.0, always disable texture compression
+		LLImageGL::sCompressTextures = false;
+	}
+	
 	// Trailing space necessary to keep "nVidia Corpor_ati_on" cards
 	// from being recognized as ATI.
 	if (mGLVendor.substr(0,4) == "ATI ")
@@ -592,11 +605,8 @@ bool LLGLManager::initGL()
 #endif // LL_WINDOWS
 
 #if (LL_WINDOWS || LL_LINUX) && !LL_MESA_HEADLESS
-		// release 7277 is a point at which we verify that ATI OpenGL
-		// drivers get pretty stable with SL, ~Catalyst 8.2,
-		// for both Win32 and Linux.
-		if (mDriverVersionRelease < 7277 &&
-		    mDriverVersionRelease != 0) // 0 == Undetectable driver version - these get to pretend to be new ATI drivers, though that decision may be revisited.
+		// count any pre OpenGL 3.0 implementation as an old driver
+		if (mGLVersion < 3.f) 
 		{
 			mATIOldDriver = TRUE;
 		}
@@ -735,6 +745,11 @@ bool LLGLManager::initGL()
 	}
 #endif
 
+	if (mIsIntel && mGLVersion <= 3.f)
+	{ //never try to use framebuffer objects on older intel drivers (crashy)
+		mHasFramebufferObject = FALSE;
+	}
+
 	if (mHasFramebufferObject)
 	{
 		glGetIntegerv(GL_MAX_SAMPLES, &mMaxSamples);
@@ -1897,7 +1912,7 @@ void LLGLState::checkClientArrays(const std::string& msg, U32 data_mask)
 	glClientActiveTextureARB(GL_TEXTURE0_ARB);
 	gGL.getTexUnit(0)->activate();
 
-	if (gGLManager.mHasVertexShader)
+	if (gGLManager.mHasVertexShader && LLGLSLShader::sNoFixedFunction)
 	{	//make sure vertex attribs are all disabled
 		GLint count;
 		glGetIntegerv(GL_MAX_VERTEX_ATTRIBS_ARB, &count);
diff --git a/indra/llrender/llimagegl.cpp b/indra/llrender/llimagegl.cpp
index 78591ddd38..c04a3f6b41 100644
--- a/indra/llrender/llimagegl.cpp
+++ b/indra/llrender/llimagegl.cpp
@@ -43,7 +43,6 @@
 const F32 MIN_TEXTURE_LIFETIME = 10.f;
 
 //statics
-LLGLuint LLImageGL::sCurrentBoundTextures[MAX_GL_TEXTURE_UNITS] = { 0 };
 
 U32 LLImageGL::sUniqueCount				= 0;
 U32 LLImageGL::sBindCount				= 0;
@@ -57,6 +56,7 @@ BOOL LLImageGL::sGlobalUseAnisotropic	= FALSE;
 F32 LLImageGL::sLastFrameTime			= 0.f;
 BOOL LLImageGL::sAllowReadBackRaw       = FALSE ;
 LLImageGL* LLImageGL::sDefaultGLTexture = NULL ;
+bool LLImageGL::sCompressTextures = false;
 
 std::set<LLImageGL*> LLImageGL::sImageList;
 
@@ -65,19 +65,10 @@ std::set<LLImageGL*> LLImageGL::sImageList;
 //****************************************************************************************************
 //-----------------------
 //debug use
-BOOL gAuditTexture = FALSE ;
-#define MAX_TEXTURE_LOG_SIZE 22 //2048 * 2048
-std::vector<S32> LLImageGL::sTextureLoadedCounter(MAX_TEXTURE_LOG_SIZE + 1) ;
-std::vector<S32> LLImageGL::sTextureBoundCounter(MAX_TEXTURE_LOG_SIZE + 1) ;
-std::vector<S32> LLImageGL::sTextureCurBoundCounter(MAX_TEXTURE_LOG_SIZE + 1) ;
 S32 LLImageGL::sCurTexSizeBar = -1 ;
 S32 LLImageGL::sCurTexPickSize = -1 ;
-LLPointer<LLImageGL> LLImageGL::sHighlightTexturep = NULL;
-S32 LLImageGL::sMaxCatagories = 1 ;
+S32 LLImageGL::sMaxCategories = 1 ;
 
-std::vector<S32> LLImageGL::sTextureMemByCategory;
-std::vector<S32> LLImageGL::sTextureMemByCategoryBound ;
-std::vector<S32> LLImageGL::sTextureCurMemByCategoryBound ;
 //------------------------
 //****************************************************************************************************
 //End for texture auditing use only
@@ -175,49 +166,11 @@ BOOL is_little_endian()
 //static 
 void LLImageGL::initClass(S32 num_catagories) 
 {
-	sMaxCatagories = num_catagories ;
-
-	sTextureMemByCategory.resize(sMaxCatagories);
-	sTextureMemByCategoryBound.resize(sMaxCatagories) ;
-	sTextureCurMemByCategoryBound.resize(sMaxCatagories) ;
 }
 
 //static 
 void LLImageGL::cleanupClass() 
 {	
-	sTextureMemByCategory.clear() ;
-	sTextureMemByCategoryBound.clear() ;
-	sTextureCurMemByCategoryBound.clear() ;
-}
-
-//static 
-void LLImageGL::setHighlightTexture(S32 category) 
-{
-	const S32 dim = 128;
-	sHighlightTexturep = new LLImageGL() ;
-	LLPointer<LLImageRaw> image_raw = new LLImageRaw(dim,dim,3);
-	U8* data = image_raw->getData();
-	for (S32 i = 0; i<dim; i++)
-	{
-		for (S32 j = 0; j<dim; j++)
-		{
-			const S32 border = 2;
-			if (i<border || j<border || i>=(dim-border) || j>=(dim-border))
-			{
-				*data++ = 0xff;
-				*data++ = 0xff;
-				*data++ = 0xff;
-			}
-			else
-			{
-				*data++ = 0xff;
-				*data++ = 0xff;
-				*data++ = 0x00;
-			}
-		}
-	}
-	sHighlightTexturep->createGLTexture(0, image_raw, 0, TRUE, category);
-	image_raw = NULL;
 }
 
 //static
@@ -285,31 +238,11 @@ void LLImageGL::updateStats(F32 current_time)
 	sLastFrameTime = current_time;
 	sBoundTextureMemoryInBytes = sCurBoundTextureMemory;
 	sCurBoundTextureMemory = 0;
-
-	if(gAuditTexture)
-	{
-		for(U32 i = 0 ; i < sTextureCurBoundCounter.size() ; i++)
-		{
-			sTextureBoundCounter[i] = sTextureCurBoundCounter[i] ;
-			sTextureCurBoundCounter[i] = 0 ;
-		}
-		for(U32 i = 0 ; i < sTextureCurMemByCategoryBound.size() ; i++)
-		{
-			sTextureMemByCategoryBound[i] = sTextureCurMemByCategoryBound[i] ;
-			sTextureCurMemByCategoryBound[i] = 0 ;
-		}
-	}
 }
 
 //static
 S32 LLImageGL::updateBoundTexMem(const S32 mem, const S32 ncomponents, S32 category)
 {
-	if(gAuditTexture && ncomponents > 0 && category > -1)
-	{
-		sTextureCurBoundCounter[getTextureCounterIndex(mem / ncomponents)]++ ;
-		sTextureCurMemByCategoryBound[category] += mem ;
-	}
-	
 	LLImageGL::sCurBoundTextureMemory += mem ;
 	return LLImageGL::sCurBoundTextureMemory;
 }
@@ -477,6 +410,8 @@ void LLImageGL::init(BOOL usemipmaps)
 	mDiscardLevelInAtlas = -1 ;
 	mTexelsInAtlas = 0 ;
 	mTexelsInGLTexture = 0 ;
+
+	mAllowCompression = true;
 	
 	mTarget = GL_TEXTURE_2D;
 	mBindTarget = LLTexUnit::TT_TEXTURE;
@@ -705,7 +640,7 @@ void LLImageGL::setImage(const U8* data_in, BOOL data_hasmips)
 						stop_glerror();
 					}
 						
-					LLImageGL::setManualImage(mTarget, gl_level, mFormatInternal, w, h, mFormatPrimary, GL_UNSIGNED_BYTE, (GLvoid*)data_in);
+					LLImageGL::setManualImage(mTarget, gl_level, mFormatInternal, w, h, mFormatPrimary, GL_UNSIGNED_BYTE, (GLvoid*)data_in, mAllowCompression);
 					if (gl_level == 0)
 					{
 						analyzeAlpha(data_in, w, h);
@@ -747,7 +682,7 @@ void LLImageGL::setImage(const U8* data_in, BOOL data_hasmips)
 					LLImageGL::setManualImage(mTarget, 0, mFormatInternal,
 								 w, h, 
 								 mFormatPrimary, mFormatType,
-								 data_in);
+								 data_in, mAllowCompression);
 					analyzeAlpha(data_in, w, h);
 					stop_glerror();
 
@@ -805,7 +740,7 @@ void LLImageGL::setImage(const U8* data_in, BOOL data_hasmips)
 							stop_glerror();
 						}
 
-						LLImageGL::setManualImage(mTarget, m, mFormatInternal, w, h, mFormatPrimary, mFormatType, cur_mip_data);
+						LLImageGL::setManualImage(mTarget, m, mFormatInternal, w, h, mFormatPrimary, mFormatType, cur_mip_data, mAllowCompression);
 						if (m == 0)
 						{
 							analyzeAlpha(data_in, w, h);
@@ -863,7 +798,7 @@ void LLImageGL::setImage(const U8* data_in, BOOL data_hasmips)
 			}
 
 			LLImageGL::setManualImage(mTarget, 0, mFormatInternal, w, h,
-						 mFormatPrimary, mFormatType, (GLvoid *)data_in);
+						 mFormatPrimary, mFormatType, (GLvoid *)data_in, mAllowCompression);
 			analyzeAlpha(data_in, w, h);
 			
 			updatePickMask(w, h, data_in);
@@ -1110,7 +1045,7 @@ void LLImageGL::deleteTextures(S32 numTextures, U32 *textures, bool immediate)
 }
 
 // static
-void LLImageGL::setManualImage(U32 target, S32 miplevel, S32 intformat, S32 width, S32 height, U32 pixformat, U32 pixtype, const void *pixels)
+void LLImageGL::setManualImage(U32 target, S32 miplevel, S32 intformat, S32 width, S32 height, U32 pixformat, U32 pixtype, const void *pixels, bool allow_compression)
 {
 	bool use_scratch = false;
 	U32* scratch = NULL;
@@ -1173,6 +1108,36 @@ void LLImageGL::setManualImage(U32 target, S32 miplevel, S32 intformat, S32 widt
 		}
 	}
 
+	if (LLImageGL::sCompressTextures && allow_compression)
+	{
+		switch (intformat)
+		{
+			case GL_RGB: 
+			case GL_RGB8:
+				intformat = GL_COMPRESSED_RGB; 
+				break;
+			case GL_RGBA:
+			case GL_RGBA8:
+				intformat = GL_COMPRESSED_RGBA; 
+				break;
+			case GL_LUMINANCE:
+			case GL_LUMINANCE8:
+				intformat = GL_COMPRESSED_LUMINANCE;
+				break;
+			case GL_LUMINANCE_ALPHA:
+			case GL_LUMINANCE8_ALPHA8:
+				intformat = GL_COMPRESSED_LUMINANCE_ALPHA;
+				break;
+			case GL_ALPHA:
+			case GL_ALPHA8:
+				intformat = GL_COMPRESSED_ALPHA;
+				break;
+			default:
+				llwarns << "Could not compress format: " << std::hex << intformat << llendl;
+				break;
+		}
+	}
+
 	stop_glerror();
 	glTexImage2D(target, miplevel, intformat, width, height, 0, pixformat, pixtype, use_scratch ? scratch : pixels);
 	stop_glerror();
@@ -1284,7 +1249,7 @@ BOOL LLImageGL::createGLTexture(S32 discard_level, const LLImageRaw* imageraw, S
 		return TRUE ;
 	}
 
-	setCategory(category) ;
+	setCategory(category);
  	const U8* rawdata = imageraw->getData();
 	return createGLTexture(discard_level, rawdata, FALSE, usename);
 }
@@ -1362,11 +1327,6 @@ BOOL LLImageGL::createGLTexture(S32 discard_level, const U8* data_in, BOOL data_
 	{
 		sGlobalTextureMemoryInBytes -= mTextureMemory;
 
-		if(gAuditTexture)
-		{
-			decTextureCounter(mTextureMemory, mComponents, mCategory) ;
-		}
-
 		LLImageGL::deleteTextures(1, &old_name);
 
 		stop_glerror();
@@ -1376,10 +1336,6 @@ BOOL LLImageGL::createGLTexture(S32 discard_level, const U8* data_in, BOOL data_
 	sGlobalTextureMemoryInBytes += mTextureMemory;
 	mTexelsInGLTexture = getWidth() * getHeight() ;
 
-	if(gAuditTexture)
-	{
-		incTextureCounter(mTextureMemory, mComponents, mCategory) ;
-	}
 	// mark this as bound at this point, so we don't throw it out immediately
 	mLastBindTime = sLastFrameTime;
 	return TRUE;
@@ -1536,22 +1492,29 @@ void LLImageGL::destroyGLTexture()
 	{
 		if(mTextureMemory)
 		{
-			if(gAuditTexture)
-			{
-				decTextureCounter(mTextureMemory, mComponents, mCategory) ;
-			}
 			sGlobalTextureMemoryInBytes -= mTextureMemory;
 			mTextureMemory = 0;
 		}
 		
 		LLImageGL::deleteTextures(1, &mTexName);			
-		mTexName = 0;
 		mCurrentDiscardLevel = -1 ; //invalidate mCurrentDiscardLevel.
+		mTexName = 0;		
 		mGLTextureCreated = FALSE ;
-	}
+	}	
 }
 
-
+//force to invalidate the gl texture, most likely a sculpty texture
+void LLImageGL::forceToInvalidateGLTexture()
+{
+	if (mTexName != 0)
+	{
+		destroyGLTexture();
+	}
+	else
+	{
+		mCurrentDiscardLevel = -1 ; //invalidate mCurrentDiscardLevel.
+	}
+}
 
 //----------------------------------------------------------------------------
 
@@ -1969,70 +1932,6 @@ BOOL LLImageGL::getMask(const LLVector2 &tc)
 	return res;
 }
 
-void LLImageGL::setCategory(S32 category) 
-{
-#if 0 //turn this off temporarily because it is not in use now.
-	if(!gAuditTexture)
-	{
-		return ;
-	}
-	if(mCategory != category)
-	{		
-		if(mCategory > -1)
-		{
-			sTextureMemByCategory[mCategory] -= mTextureMemory ;
-		}
-		if(category > -1 && category < sMaxCatagories)
-		{
-			sTextureMemByCategory[category] += mTextureMemory ;		
-			mCategory = category;
-		}
-		else
-		{
-			mCategory = -1 ;
-		}
-	}
-#endif
-}
-
-//for debug use 
-//val is a "power of two" number
-S32 LLImageGL::getTextureCounterIndex(U32 val) 
-{
-	//index range is [0, MAX_TEXTURE_LOG_SIZE].
-	if(val < 2)
-	{
-		return 0 ;
-	}
-	else if(val >= (1 << MAX_TEXTURE_LOG_SIZE))
-	{
-		return MAX_TEXTURE_LOG_SIZE ;
-	}
-	else
-	{
-		S32 ret = 0 ;
-		while(val >>= 1)
-		{
-			++ret;
-		}
-		return ret ;
-	}
-}
-
-//static
-void LLImageGL::incTextureCounter(U32 val, S32 ncomponents, S32 category) 
-{
-	sTextureLoadedCounter[getTextureCounterIndex(val)]++ ;
-	sTextureMemByCategory[category] += (S32)val * ncomponents ;
-}
-
-//static
-void LLImageGL::decTextureCounter(U32 val, S32 ncomponents, S32 category) 
-{
-	sTextureLoadedCounter[getTextureCounterIndex(val)]-- ;
-	sTextureMemByCategory[category] += (S32)val * ncomponents ;
-}
-
 void LLImageGL::setCurTexSizebar(S32 index, BOOL set_pick_size)
 {
 	sCurTexSizeBar = index ;
diff --git a/indra/llrender/llimagegl.h b/indra/llrender/llimagegl.h
index 2cfb15b0d9..f34b9fa91a 100644
--- a/indra/llrender/llimagegl.h
+++ b/indra/llrender/llimagegl.h
@@ -94,16 +94,17 @@ public:
 	
 	void setSize(S32 width, S32 height, S32 ncomponents);
 	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);
+	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() ;
-	BOOL createGLTexture(S32 discard_level, const LLImageRaw* imageraw, S32 usename = 0, BOOL to_create = TRUE, 
-		S32 category = sMaxCatagories - 1);
+	BOOL createGLTexture(S32 discard_level, const LLImageRaw* imageraw, S32 usename = 0, BOOL to_create = TRUE,
+		S32 category = sMaxCategories-1);
 	BOOL createGLTexture(S32 discard_level, const U8* data, BOOL data_hasmips = FALSE, S32 usename = 0);
 	void setImage(const LLImageRaw* imageraw);
 	void setImage(const U8* data_in, BOOL data_hasmips = FALSE);
@@ -114,6 +115,7 @@ public:
 	// Read back a raw image for this discard level, if it exists
 	BOOL readBackRaw(S32 discard_level, LLImageRaw* imageraw, bool compressed_ok) const;
 	void destroyGLTexture();
+	void forceToInvalidateGLTexture();
 
 	void setExplicitFormat(LLGLint internal_format, LLGLenum primary_format, LLGLenum type_format = 0, BOOL swap_bytes = FALSE);
 	void setComponents(S8 ncomponents) { mComponents = ncomponents; }
@@ -209,6 +211,8 @@ private:
 	U32      mTexelsInAtlas ;
 	U32      mTexelsInGLTexture;
 
+	bool mAllowCompression;
+
 protected:
 	LLGLenum mTarget;		// Normally GL_TEXTURE2D, sometimes something else (ex. cube maps)
 	LLTexUnit::eTextureType mBindTarget;	// Normally TT_TEXTURE, sometimes something else (ex. cube maps)
@@ -234,8 +238,6 @@ public:
 	static S32 sCount;
 	
 	static F32 sLastFrameTime;
-	
-	static LLGLuint sCurrentBoundTextures[MAX_GL_TEXTURE_UNITS]; // Currently bound texture ID
 
 	// Global memory statistics
 	static S32 sGlobalTextureMemoryInBytes;		// Tracks main memory texmem
@@ -246,7 +248,7 @@ public:
 	static BOOL sGlobalUseAnisotropic;
 	static LLImageGL* sDefaultGLTexture ;	
 	static BOOL sAutomatedTest;
-
+	static bool sCompressTextures;			//use GL texture compression
 #if DEBUG_MISS
 	BOOL mMissed; // Missed on last bind?
 	BOOL getMissed() const { return mMissed; };
@@ -257,9 +259,10 @@ public:
 public:
 	static void initClass(S32 num_catagories) ;
 	static void cleanupClass() ;
-private:
-	static S32 sMaxCatagories ;
 
+private:
+	static S32 sMaxCategories;
+	
 	//the flag to allow to call readBackRaw(...).
 	//can be removed if we do not use that function at all.
 	static BOOL sAllowReadBackRaw ;
@@ -269,39 +272,22 @@ private:
 //****************************************************************************************************
 private:
 	S32 mCategory ;
-public:		
-	void setCategory(S32 category) ;
-	S32  getCategory()const {return mCategory ;}
-
+public:
+	void setCategory(S32 category) {mCategory = category;}
+	S32  getCategory()const {return mCategory;}
+	
 	//for debug use: show texture size distribution 
 	//----------------------------------------
-	static LLPointer<LLImageGL> sHighlightTexturep; //default texture to replace normal textures
-	static std::vector<S32> sTextureLoadedCounter ;
-	static std::vector<S32> sTextureBoundCounter ;
-	static std::vector<S32> sTextureCurBoundCounter ;
 	static S32 sCurTexSizeBar ;
 	static S32 sCurTexPickSize ;
 
-	static void setHighlightTexture(S32 category) ;
-	static S32 getTextureCounterIndex(U32 val) ;
-	static void incTextureCounter(U32 val, S32 ncomponents, S32 category) ;
-	static void decTextureCounter(U32 val, S32 ncomponents, S32 category) ;
 	static void setCurTexSizebar(S32 index, BOOL set_pick_size = TRUE) ;
 	static void resetCurTexSizebar();
-	//----------------------------------------
 
-	//for debug use: show texture category distribution 
-	//----------------------------------------		
-	
-	static std::vector<S32> sTextureMemByCategory;
-	static std::vector<S32> sTextureMemByCategoryBound ;
-	static std::vector<S32> sTextureCurMemByCategoryBound ;
-	//----------------------------------------	
 //****************************************************************************************************
 //End of definitions for texture auditing use only
 //****************************************************************************************************
 
 };
 
-extern BOOL gAuditTexture;
 #endif // LL_LLIMAGEGL_H
diff --git a/indra/llrender/llrender.cpp b/indra/llrender/llrender.cpp
index b0ddacbb05..93bac4c779 100644
--- a/indra/llrender/llrender.cpp
+++ b/indra/llrender/llrender.cpp
@@ -246,14 +246,6 @@ bool LLTexUnit::bind(LLTexture* texture, bool for_rendering, bool forceBind)
 	}
 
 	//in audit, replace the selected texture by the default one.
-	if(gAuditTexture && for_rendering && LLImageGL::sCurTexPickSize > 0)
-	{
-		if(texture->getWidth() * texture->getHeight() == LLImageGL::sCurTexPickSize)
-		{
-			gl_tex->updateBindStats(gl_tex->mTextureMemory);
-			return bind(LLImageGL::sHighlightTexturep.get());
-		}
-	}
 	if ((mCurrTexture != gl_tex->getTexName()) || forceBind)
 	{
 		activate();
diff --git a/indra/llrender/llrendertarget.cpp b/indra/llrender/llrendertarget.cpp
index ef2a7395da..780f1dc484 100644
--- a/indra/llrender/llrendertarget.cpp
+++ b/indra/llrender/llrendertarget.cpp
@@ -143,7 +143,7 @@ bool LLRenderTarget::addColorAttachment(U32 color_fmt)
 
 	{
 		clear_glerror();
-		LLImageGL::setManualImage(LLTexUnit::getInternalType(mUsage), 0, color_fmt, mResX, mResY, GL_RGBA, GL_UNSIGNED_BYTE, NULL);
+		LLImageGL::setManualImage(LLTexUnit::getInternalType(mUsage), 0, color_fmt, mResX, mResY, GL_RGBA, GL_UNSIGNED_BYTE, NULL, false);
 		if (glGetError() != GL_NO_ERROR)
 		{
 			llwarns << "Could not allocate color buffer for render target." << llendl;
@@ -223,7 +223,7 @@ bool LLRenderTarget::allocateDepth()
 		U32 internal_type = LLTexUnit::getInternalType(mUsage);
 		stop_glerror();
 		clear_glerror();
-		LLImageGL::setManualImage(internal_type, 0, GL_DEPTH_COMPONENT24, mResX, mResY, GL_DEPTH_COMPONENT, GL_UNSIGNED_INT, NULL);
+		LLImageGL::setManualImage(internal_type, 0, GL_DEPTH_COMPONENT24, mResX, mResY, GL_DEPTH_COMPONENT, GL_UNSIGNED_INT, NULL, false);
 		gGL.getTexUnit(0)->setTextureFilteringOption(LLTexUnit::TFO_POINT);
 	}
 
diff --git a/indra/llrender/llshadermgr.cpp b/indra/llrender/llshadermgr.cpp
index 7d384450e6..5a6e6cab3e 100644
--- a/indra/llrender/llshadermgr.cpp
+++ b/indra/llrender/llshadermgr.cpp
@@ -1062,8 +1062,9 @@ void LLShaderMgr::initAttribsAndUniforms()
 	mReservedUniforms.push_back("proj_shadow_res");
 	mReservedUniforms.push_back("depth_cutoff");
 	mReservedUniforms.push_back("norm_cutoff");
+	mReservedUniforms.push_back("shadow_target_width");
 	
-	llassert(mReservedUniforms.size() == LLShaderMgr::DEFERRED_NORM_CUTOFF+1);
+	llassert(mReservedUniforms.size() == LLShaderMgr::DEFERRED_SHADOW_TARGET_WIDTH+1);
 
 	mReservedUniforms.push_back("tc_scale");
 	mReservedUniforms.push_back("rcp_screen_res");
diff --git a/indra/llrender/llshadermgr.h b/indra/llrender/llshadermgr.h
index e28bda6de2..f792faa8f0 100644
--- a/indra/llrender/llshadermgr.h
+++ b/indra/llrender/llshadermgr.h
@@ -130,6 +130,7 @@ public:
 		DEFERRED_PROJ_SHADOW_RES,
 		DEFERRED_DEPTH_CUTOFF,
 		DEFERRED_NORM_CUTOFF,
+		DEFERRED_SHADOW_TARGET_WIDTH,
 
 		FXAA_TC_SCALE,
 		FXAA_RCP_SCREEN_RES,
diff --git a/indra/llrender/llvertexbuffer.cpp b/indra/llrender/llvertexbuffer.cpp
index 8b5503229f..823c6b9dc5 100644
--- a/indra/llrender/llvertexbuffer.cpp
+++ b/indra/llrender/llvertexbuffer.cpp
@@ -38,6 +38,12 @@
 #include "llglslshader.h"
 #include "llmemory.h"
 
+#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)
@@ -49,6 +55,35 @@ U32 nhpo2(U32 v)
 	return r;
 }
 
+//which power of 2 is i?
+//assumes i is a power of 2 > 0
+U32 wpo2(U32 i)
+{
+	llassert(i > 0);
+	llassert(nhpo2(i) == i);
+
+	U32 r = 0;
+
+	while (i >>= 1) ++r;
+
+	return r;
+}
+
+
+const U32 LL_VBO_BLOCK_SIZE = 2048;
+
+U32 vbo_block_size(U32 size)
+{ //what block size will fit size?
+	U32 mod = size % LL_VBO_BLOCK_SIZE;
+	return mod == 0 ? size : size + (LL_VBO_BLOCK_SIZE-mod);
+}
+
+U32 vbo_block_index(U32 size)
+{
+	return vbo_block_size(size)/LL_VBO_BLOCK_SIZE;
+}
+
+
 
 //============================================================================
 
@@ -57,7 +92,11 @@ LLVBOPool LLVertexBuffer::sStreamVBOPool(GL_STREAM_DRAW_ARB, GL_ARRAY_BUFFER_ARB
 LLVBOPool LLVertexBuffer::sDynamicVBOPool(GL_DYNAMIC_DRAW_ARB, GL_ARRAY_BUFFER_ARB);
 LLVBOPool LLVertexBuffer::sStreamIBOPool(GL_STREAM_DRAW_ARB, GL_ELEMENT_ARRAY_BUFFER_ARB);
 LLVBOPool LLVertexBuffer::sDynamicIBOPool(GL_DYNAMIC_DRAW_ARB, GL_ELEMENT_ARRAY_BUFFER_ARB);
+
 U32 LLVBOPool::sBytesPooled = 0;
+U32 LLVBOPool::sIndexBytesPooled = 0;
+U32 LLVertexBuffer::sAllocatedIndexBytes = 0;
+U32 LLVertexBuffer::sIndexCount = 0;
 
 LLPrivateMemoryPool* LLVertexBuffer::sPrivatePoolp = NULL;
 U32 LLVertexBuffer::sBindCount = 0;
@@ -74,6 +113,7 @@ U32 LLVertexBuffer::sLastMask = 0;
 bool LLVertexBuffer::sVBOActive = false;
 bool LLVertexBuffer::sIBOActive = false;
 U32 LLVertexBuffer::sAllocatedBytes = 0;
+U32 LLVertexBuffer::sVertexCount = 0;
 bool LLVertexBuffer::sMapped = false;
 bool LLVertexBuffer::sUseStreamDraw = true;
 bool LLVertexBuffer::sUseVAO = false;
@@ -134,39 +174,35 @@ public:
 };
 
 
-//which power of 2 is i?
-//assumes i is a power of 2 > 0
-U32 wpo2(U32 i)
-{
-	llassert(i > 0);
-	llassert(nhpo2(i) == i);
-
-	U32 r = 0;
-
-	while (i >>= 1) ++r;
-
-	return r;
-}
-
 volatile U8* LLVBOPool::allocate(U32& name, U32 size)
 {
-	llassert(nhpo2(size) == size);
+	llassert(vbo_block_size(size) == size);
+	
+	volatile U8* ret = NULL;
+
+#if LL_VBO_POOLING
 
-	U32 i = wpo2(size);
+	U32 i = vbo_block_index(size);
 
 	if (mFreeList.size() <= i)
 	{
 		mFreeList.resize(i+1);
 	}
 
-	volatile U8* ret = NULL;
-
 	if (mFreeList[i].empty())
 	{
 		//make a new buffer
 		glGenBuffersARB(1, &name);
 		glBindBufferARB(mType, name);
-		LLVertexBuffer::sAllocatedBytes += size;
+
+		if (mType == GL_ARRAY_BUFFER_ARB)
+		{
+			LLVertexBuffer::sAllocatedBytes += size;
+		}
+		else
+		{
+			LLVertexBuffer::sAllocatedIndexBytes += size;
+		}
 
 		if (LLVertexBuffer::sDisableVBOMapping || mUsage != GL_DYNAMIC_DRAW_ARB)
 		{
@@ -185,19 +221,55 @@ volatile U8* LLVBOPool::allocate(U32& name, U32 size)
 		name = mFreeList[i].front().mGLName;
 		ret = mFreeList[i].front().mClientData;
 
-		sBytesPooled -= size;
+		if (mType == GL_ARRAY_BUFFER_ARB)
+		{
+			sBytesPooled -= size;
+		}
+		else
+		{
+			sIndexBytesPooled -= 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;
 }
 
 void LLVBOPool::release(U32 name, volatile U8* buffer, U32 size)
 {
-	llassert(nhpo2(size) == size);
+	llassert(vbo_block_size(size) == size);
 
-	U32 i = wpo2(size);
+#if LL_VBO_POOLING
+
+	U32 i = vbo_block_index(size);
 
 	llassert(mFreeList.size() > i);
 
@@ -211,9 +283,29 @@ void LLVBOPool::release(U32 name, volatile U8* buffer, U32 size)
 	}
 	else
 	{
-		sBytesPooled += size;
+		if (mType == GL_ARRAY_BUFFER_ARB)
+		{
+			sBytesPooled += size;
+		}
+		else
+		{
+			sIndexBytesPooled += size;
+		}
 		mFreeList[i].push_back(rec);
 	}
+#else //no pooling
+	glDeleteBuffersARB(1, &name);
+	ll_aligned_free_16((U8*) buffer);
+
+	if (mType == GL_ARRAY_BUFFER_ARB)
+	{
+		LLVertexBuffer::sAllocatedBytes -= size;
+	}
+	else
+	{
+		LLVertexBuffer::sAllocatedIndexBytes -= size;
+	}
+#endif
 }
 
 void LLVBOPool::cleanup()
@@ -237,8 +329,16 @@ void LLVBOPool::cleanup()
 
 			l.pop_front();
 
-			LLVertexBuffer::sAllocatedBytes -= size;
-			sBytesPooled -= size;
+			if (mType == GL_ARRAY_BUFFER_ARB)
+			{
+				sBytesPooled -= size;
+				LLVertexBuffer::sAllocatedBytes -= size;
+			}
+			else
+			{
+				sIndexBytesPooled -= size;
+				LLVertexBuffer::sAllocatedIndexBytes -= size;
+			}
 		}
 
 		size *= 2;
@@ -898,6 +998,9 @@ LLVertexBuffer::~LLVertexBuffer()
 	
 	mFence = NULL;
 
+	sVertexCount -= mNumVerts;
+	sIndexCount -= mNumIndices;
+
 	llassert_always(!mMappedData && !mMappedIndexData);
 };
 
@@ -929,7 +1032,7 @@ void LLVertexBuffer::waitFence() const
 
 void LLVertexBuffer::genBuffer(U32 size)
 {
-	mSize = nhpo2(size);
+	mSize = vbo_block_size(size);
 
 	if (mUsage == GL_STREAM_DRAW_ARB)
 	{
@@ -945,7 +1048,7 @@ void LLVertexBuffer::genBuffer(U32 size)
 
 void LLVertexBuffer::genIndices(U32 size)
 {
-	mIndicesSize = nhpo2(size);
+	mIndicesSize = vbo_block_size(size);
 
 	if (mUsage == GL_STREAM_DRAW_ARB)
 	{
@@ -1121,7 +1224,9 @@ void LLVertexBuffer::updateNumVerts(S32 nverts)
 		createGLBuffer(needed_size);
 	}
 
+	sVertexCount -= mNumVerts;
 	mNumVerts = nverts;
+	sVertexCount += mNumVerts;
 }
 
 void LLVertexBuffer::updateNumIndices(S32 nindices)
@@ -1137,7 +1242,9 @@ void LLVertexBuffer::updateNumIndices(S32 nindices)
 		createGLIndices(needed_size);
 	}
 
+	sIndexCount -= mNumIndices;
 	mNumIndices = nindices;
+	sIndexCount += mNumIndices;
 }
 
 void LLVertexBuffer::allocateBuffer(S32 nverts, S32 nindices, bool create)
diff --git a/indra/llrender/llvertexbuffer.h b/indra/llrender/llvertexbuffer.h
index d859199663..7477dec3ad 100644
--- a/indra/llrender/llvertexbuffer.h
+++ b/indra/llrender/llvertexbuffer.h
@@ -55,6 +55,7 @@ class LLVBOPool
 {
 public:
 	static U32 sBytesPooled;
+	static U32 sIndexBytesPooled;
 	
 	LLVBOPool(U32 vboUsage, U32 vboType)
 		: mUsage(vboUsage)
@@ -332,6 +333,9 @@ public:
 	static bool sIBOActive;
 	static U32 sLastMask;
 	static U32 sAllocatedBytes;
+	static U32 sAllocatedIndexBytes;
+	static U32 sVertexCount;
+	static U32 sIndexCount;
 	static U32 sBindCount;
 	static U32 sSetCount;
 };