SH-1682 Use GL_ARB_map_buffer_range to mitigate impact of mapping larger vertex buffer objects. Limit number of textures per batch to 6 (prevents frame stalls on NVIDIA).

9 files changed, 387 insertions, 167 deletions

diff --git a/indra/llrender/llgl.cpp b/indra/llrender/llgl.cpp
index a8d7df3bc8..a3aed4dd8a 100644
--- a/indra/llrender/llgl.cpp
+++ b/indra/llrender/llgl.cpp
@@ -127,6 +127,11 @@ PFNGLUNMAPBUFFERARBPROC				glUnmapBufferARB = NULL;
 PFNGLGETBUFFERPARAMETERIVARBPROC	glGetBufferParameterivARB = NULL;
 PFNGLGETBUFFERPOINTERVARBPROC		glGetBufferPointervARB = NULL;
 
+// GL_ARB_map_buffer_range
+PFNGLMAPBUFFERRANGEPROC			glMapBufferRange;
+PFNGLFLUSHMAPPEDBUFFERRANGEPROC	glFlushMappedBufferRange;
+
+
 // vertex object prototypes
 PFNGLNEWOBJECTBUFFERATIPROC			glNewObjectBufferATI = NULL;
 PFNGLISOBJECTBUFFERATIPROC			glIsObjectBufferATI = NULL;
@@ -331,6 +336,7 @@ LLGLManager::LLGLManager() :
 	mHasBlendFuncSeparate(FALSE),
 
 	mHasVertexBufferObject(FALSE),
+	mHasMapBufferRange(FALSE),
 	mHasPBuffer(FALSE),
 	mHasShaderObjects(FALSE),
 	mHasVertexShader(FALSE),
@@ -761,6 +767,7 @@ void LLGLManager::initExtensions()
 	mHasOcclusionQuery = ExtensionExists("GL_ARB_occlusion_query", gGLHExts.mSysExts);
 	mHasOcclusionQuery2 = ExtensionExists("GL_ARB_occlusion_query2", gGLHExts.mSysExts);
 	mHasVertexBufferObject = ExtensionExists("GL_ARB_vertex_buffer_object", gGLHExts.mSysExts);
+	mHasMapBufferRange = ExtensionExists("GL_ARB_map_buffer_range", gGLHExts.mSysExts);
 	mHasDepthClamp = ExtensionExists("GL_ARB_depth_clamp", gGLHExts.mSysExts) || ExtensionExists("GL_NV_depth_clamp", gGLHExts.mSysExts);
 	// mask out FBO support when packed_depth_stencil isn't there 'cause we need it for LLRenderTarget -Brad
 #ifdef GL_ARB_framebuffer_object
@@ -955,6 +962,11 @@ void LLGLManager::initExtensions()
 			mHasVertexBufferObject = FALSE;
 		}
 	}
+	if (mHasMapBufferRange)
+	{
+		glMapBufferRange = (PFNGLMAPBUFFERRANGEPROC) GLH_EXT_GET_PROC_ADDRESS("glMapBufferRange");
+		glFlushMappedBufferRange = (PFNGLFLUSHMAPPEDBUFFERRANGEPROC) GLH_EXT_GET_PROC_ADDRESS("glFlushMappedBufferRange");
+	}
 	if (mHasFramebufferObject)
 	{
 		llinfos << "initExtensions() FramebufferObject-related procs..." << llendl;
@@ -1411,10 +1423,6 @@ void LLGLState::checkTextureChannels(const std::string& msg)
 		}
 	}
 
-	GLint maxTextureUnits = 0;
-	glGetIntegerv(GL_MAX_TEXTURE_UNITS_ARB, &maxTextureUnits);
-	stop_glerror();
-
 	static const char* label[] =
 	{
 		"GL_TEXTURE_2D",
@@ -1449,69 +1457,91 @@ void LLGLState::checkTextureChannels(const std::string& msg)
 	glh::matrix4f identity;
 	identity.identity();
 
-	for (GLint i = 1; i < maxTextureUnits; i++)
+	for (GLint i = 1; i < gGLManager.mNumTextureUnits; i++)
 	{
 		gGL.getTexUnit(i)->activate();
-		glClientActiveTextureARB(GL_TEXTURE0_ARB+i);
-		stop_glerror();
-		glGetIntegerv(GL_TEXTURE_STACK_DEPTH, &stackDepth);
-		stop_glerror();
 
-		if (stackDepth != 1)
+		if (i < gGLManager.mNumTextureUnits)
 		{
-			error = TRUE;
-			LL_WARNS("RenderState") << "Texture matrix stack corrupted." << LL_ENDL;
+			glClientActiveTextureARB(GL_TEXTURE0_ARB+i);
+			stop_glerror();
+			glGetIntegerv(GL_TEXTURE_STACK_DEPTH, &stackDepth);
+			stop_glerror();
 
-			if (gDebugSession)
+			if (stackDepth != 1)
 			{
-				gFailLog << "Texture matrix stack corrupted." << std::endl;
+				error = TRUE;
+				LL_WARNS("RenderState") << "Texture matrix stack corrupted." << LL_ENDL;
+
+				if (gDebugSession)
+				{
+					gFailLog << "Texture matrix stack corrupted." << std::endl;
+				}
 			}
-		}
 
-		glGetFloatv(GL_TEXTURE_MATRIX, (GLfloat*) mat.m);
-		stop_glerror();
+			glGetFloatv(GL_TEXTURE_MATRIX, (GLfloat*) mat.m);
+			stop_glerror();
 
-		if (mat != identity)
-		{
-			error = TRUE;
-			LL_WARNS("RenderState") << "Texture matrix in channel " << i << " corrupt." << LL_ENDL;
-			if (gDebugSession)
+			if (mat != identity)
 			{
-				gFailLog << "Texture matrix in channel " << i << " corrupt." << std::endl;
+				error = TRUE;
+				LL_WARNS("RenderState") << "Texture matrix in channel " << i << " corrupt." << LL_ENDL;
+				if (gDebugSession)
+				{
+					gFailLog << "Texture matrix in channel " << i << " corrupt." << std::endl;
+				}
 			}
-		}
 				
-		for (S32 j = (i == 0 ? 1 : 0); 
-			j < 9; j++)
-		{
-			if (j == 8 && !gGLManager.mHasTextureRectangle ||
-				j == 9 && !gGLManager.mHasTextureMultisample)
+			for (S32 j = (i == 0 ? 1 : 0); 
+				j < 9; j++)
 			{
-				continue;
-			}
+				if (j == 8 && !gGLManager.mHasTextureRectangle ||
+					j == 9 && !gGLManager.mHasTextureMultisample)
+				{
+					continue;
+				}
 				
-			if (glIsEnabled(value[j]))
+				if (glIsEnabled(value[j]))
+				{
+					error = TRUE;
+					LL_WARNS("RenderState") << "Texture channel " << i << " still has " << label[j] << " enabled." << LL_ENDL;
+					if (gDebugSession)
+					{
+						gFailLog << "Texture channel " << i << " still has " << label[j] << " enabled." << std::endl;
+					}
+				}
+				stop_glerror();
+			}
+
+			glGetFloatv(GL_TEXTURE_MATRIX, mat.m);
+			stop_glerror();
+
+			if (mat != identity)
 			{
 				error = TRUE;
-				LL_WARNS("RenderState") << "Texture channel " << i << " still has " << label[j] << " enabled." << LL_ENDL;
+				LL_WARNS("RenderState") << "Texture matrix " << i << " is not identity." << LL_ENDL;
 				if (gDebugSession)
 				{
-					gFailLog << "Texture channel " << i << " still has " << label[j] << " enabled." << std::endl;
+					gFailLog << "Texture matrix " << i << " is not identity." << std::endl;
 				}
 			}
-			stop_glerror();
 		}
 
-		glGetFloatv(GL_TEXTURE_MATRIX, mat.m);
-		stop_glerror();
-
-		if (mat != identity)
 		{
-			error = TRUE;
-			LL_WARNS("RenderState") << "Texture matrix " << i << " is not identity." << LL_ENDL;
-			if (gDebugSession)
+			GLint tex = 0;
+			stop_glerror();
+			glGetIntegerv(GL_TEXTURE_BINDING_2D, &tex);
+			stop_glerror();
+
+			if (tex != 0)
 			{
-				gFailLog << "Texture matrix " << i << " is not identity." << std::endl;
+				error = TRUE;
+				LL_WARNS("RenderState") << "Texture channel " << i << " still has texture " << tex << " bound." << llendl;
+
+				if (gDebugSession)
+				{
+					gFailLog << "Texture channel " << i << " still has texture " << tex << " bound." << std::endl;
+				}
 			}
 		}
 	}
diff --git a/indra/llrender/llgl.h b/indra/llrender/llgl.h
index 420922cf06..d1bee00161 100644
--- a/indra/llrender/llgl.h
+++ b/indra/llrender/llgl.h
@@ -88,6 +88,7 @@ public:
 		
 	// ARB Extensions
 	BOOL mHasVertexBufferObject;
+	BOOL mHasMapBufferRange;
 	BOOL mHasPBuffer;
 	BOOL mHasShaderObjects;
 	BOOL mHasVertexShader;
diff --git a/indra/llrender/llglheaders.h b/indra/llrender/llglheaders.h
index 825d304d35..94f7a08c92 100644
--- a/indra/llrender/llglheaders.h
+++ b/indra/llrender/llglheaders.h
@@ -68,6 +68,10 @@ extern PFNGLUNMAPBUFFERARBPROC		glUnmapBufferARB;
 extern PFNGLGETBUFFERPARAMETERIVARBPROC	glGetBufferParameterivARB;
 extern PFNGLGETBUFFERPOINTERVARBPROC	glGetBufferPointervARB;
 
+// GL_ARB_map_buffer_range
+extern PFNGLMAPBUFFERRANGEPROC			glMapBufferRange;
+extern PFNGLFLUSHMAPPEDBUFFERRANGEPROC	glFlushMappedBufferRange;
+
 // GL_ATI_vertex_array_object
 extern PFNGLNEWOBJECTBUFFERATIPROC			glNewObjectBufferATI;
 extern PFNGLISOBJECTBUFFERATIPROC			glIsObjectBufferATI;
@@ -306,6 +310,10 @@ extern PFNGLUNMAPBUFFERARBPROC		glUnmapBufferARB;
 extern PFNGLGETBUFFERPARAMETERIVARBPROC	glGetBufferParameterivARB;
 extern PFNGLGETBUFFERPOINTERVARBPROC	glGetBufferPointervARB;
 
+// GL_ARB_map_buffer_range
+extern PFNGLMAPBUFFERRANGEPROC			glMapBufferRange;
+extern PFNGLFLUSHMAPPEDBUFFERRANGEPROC	glFlushMappedBufferRange;
+
 // GL_ATI_vertex_array_object
 extern PFNGLNEWOBJECTBUFFERATIPROC			glNewObjectBufferATI;
 extern PFNGLISOBJECTBUFFERATIPROC			glIsObjectBufferATI;
@@ -511,6 +519,10 @@ extern PFNGLUNMAPBUFFERARBPROC		glUnmapBufferARB;
 extern PFNGLGETBUFFERPARAMETERIVARBPROC	glGetBufferParameterivARB;
 extern PFNGLGETBUFFERPOINTERVARBPROC	glGetBufferPointervARB;
 
+// GL_ARB_map_buffer_range
+extern PFNGLMAPBUFFERRANGEPROC			glMapBufferRange;
+extern PFNGLFLUSHMAPPEDBUFFERRANGEPROC	glFlushMappedBufferRange;
+
 // GL_ATI_vertex_array_object
 extern PFNGLNEWOBJECTBUFFERATIPROC			glNewObjectBufferATI;
 extern PFNGLISOBJECTBUFFERATIPROC			glIsObjectBufferATI;
diff --git a/indra/llrender/llglslshader.cpp b/indra/llrender/llglslshader.cpp
index 2dab757828..5473f23d86 100644
--- a/indra/llrender/llglslshader.cpp
+++ b/indra/llrender/llglslshader.cpp
@@ -48,6 +48,8 @@ using std::pair;
 using std::make_pair;
 using std::string;
 
+U32 LLGLSLShader::sCurBoundShader = 0;
+
 BOOL shouldChange(const LLVector4& v1, const LLVector4& v2)
 {
 	return v1 != v2;
@@ -367,7 +369,7 @@ void LLGLSLShader::bind()
 	if (gGLManager.mHasShaderObjects)
 	{
 		glUseProgramObjectARB(mProgramObject);
-
+		sCurBoundShader = mProgramObject;
 		if (mUniformsDirty)
 		{
 			LLShaderMgr::instance()->updateShaderUniforms(this);
@@ -390,6 +392,7 @@ void LLGLSLShader::unbind()
 			}
 		}
 		glUseProgramObjectARB(0);
+		sCurBoundShader = 0;
 		stop_glerror();
 	}
 }
@@ -397,6 +400,7 @@ void LLGLSLShader::unbind()
 void LLGLSLShader::bindNoShader(void)
 {
 	glUseProgramObjectARB(0);
+	sCurBoundShader = 0;
 }
 
 S32 LLGLSLShader::enableTexture(S32 uniform, LLTexUnit::eTextureType mode)
diff --git a/indra/llrender/llglslshader.h b/indra/llrender/llglslshader.h
index 51be56b295..4922eb6d67 100644
--- a/indra/llrender/llglslshader.h
+++ b/indra/llrender/llglslshader.h
@@ -66,6 +66,8 @@ public:
 	
 	LLGLSLShader();
 
+	static GLhandleARB sCurBoundShader;
+
 	void unload();
 	BOOL createShader(std::vector<std::string> * attributes,
 						std::vector<std::string> * uniforms);
diff --git a/indra/llrender/llimagegl.cpp b/indra/llrender/llimagegl.cpp
index 3a4139bace..60a5962234 100644
--- a/indra/llrender/llimagegl.cpp
+++ b/indra/llrender/llimagegl.cpp
@@ -1418,11 +1418,13 @@ void LLImageGL::deleteDeadTextures()
 	{
 		GLuint tex = sDeadTextureList.front();
 		sDeadTextureList.pop_front();
-		for (int i = 0; i < gGLManager.mNumTextureUnits; i++)
+		for (int i = 0; i < gGLManager.mNumTextureImageUnits; i++)
 		{
-			if (sCurrentBoundTextures[i] == tex)
+			LLTexUnit* tex_unit = gGL.getTexUnit(i);
+
+			if (tex_unit->getCurrTexture() == tex)
 			{
-				gGL.getTexUnit(i)->unbind(gGL.getTexUnit(i)->getCurrType());
+				tex_unit->unbind(tex_unit->getCurrType());
 				stop_glerror();
 			}
 		}
diff --git a/indra/llrender/llrender.cpp b/indra/llrender/llrender.cpp
index fdfcfe7fab..e91ceb873e 100644
--- a/indra/llrender/llrender.cpp
+++ b/indra/llrender/llrender.cpp
@@ -30,6 +30,7 @@
 
 #include "llvertexbuffer.h"
 #include "llcubemap.h"
+#include "llglslshader.h"
 #include "llimagegl.h"
 #include "llrendertarget.h"
 #include "lltexture.h"
@@ -183,7 +184,8 @@ void LLTexUnit::enable(eTextureType type)
 		mCurrTexType = type;
 
 		gGL.flush();
-		if (type != LLTexUnit::TT_MULTISAMPLE_TEXTURE &&
+		if (LLGLSLShader::sCurBoundShader == 0 &&
+			type != LLTexUnit::TT_MULTISAMPLE_TEXTURE &&
 			mIndex < gGLManager.mNumTextureUnits)
 		{
 			glEnable(sGLTextureType[type]);
@@ -200,12 +202,10 @@ void LLTexUnit::disable(void)
 		activate();
 		unbind(mCurrTexType);
 		gGL.flush();
-		if (mCurrTexType != LLTexUnit::TT_MULTISAMPLE_TEXTURE &&
+		if (LLGLSLShader::sCurBoundShader == 0 && mCurrTexType != LLTexUnit::TT_MULTISAMPLE_TEXTURE &&
 			mIndex < gGLManager.mNumTextureUnits)
 		{
-			stop_glerror();
 			glDisable(sGLTextureType[mCurrTexType]);
-			stop_glerror();
 		}
 		
 		mCurrTexType = TT_NONE;
@@ -295,7 +295,7 @@ bool LLTexUnit::bind(LLImageGL* texture, bool for_rendering, bool forceBind)
 		glBindTexture(sGLTextureType[texture->getTarget()], mCurrTexture);
 		texture->updateBindStats(texture->mTextureMemory);		
 		mHasMipMaps = texture->mHasMipMaps;
-		if (texture->mTexOptionsDirty)
+		if (mIndex == 0 && texture->mTexOptionsDirty)
 		{
 			texture->mTexOptionsDirty = false;
 			setTextureAddressMode(texture->mAddressMode);
diff --git a/indra/llrender/llvertexbuffer.cpp b/indra/llrender/llvertexbuffer.cpp
index f715a8e9ba..27cc88462a 100644
--- a/indra/llrender/llvertexbuffer.cpp
+++ b/indra/llrender/llvertexbuffer.cpp
@@ -934,8 +934,26 @@ void LLVertexBuffer::allocateClientIndexBuffer()
 	}
 }
 
+bool expand_region(LLVertexBuffer::MappedRegion& region, S32 index, S32 count)
+{
+	S32 end = index+count;
+	S32 region_end = region.mIndex+region.mCount;
+	
+	if (end < region.mIndex ||
+		index > region_end)
+	{ //gap exists, do not merge
+		return false;
+	}
+
+	S32 new_end = llmax(end, region_end);
+	S32 new_index = llmin(index, region.mIndex);
+	region.mIndex = new_index;
+	region.mCount = new_end-new_index;
+	return true;
+}
+
 // Map for data access
-U8* LLVertexBuffer::mapVertexBuffer(S32 type, S32 access)
+U8* LLVertexBuffer::mapVertexBuffer(S32 type, S32 index, S32 count, bool map_range)
 {
 	LLMemType mt2(LLMemType::MTYPE_VERTEX_MAP_BUFFER);
 	if (mFinal)
@@ -947,8 +965,44 @@ U8* LLVertexBuffer::mapVertexBuffer(S32 type, S32 access)
 		llerrs << "LLVertexBuffer::mapVertexBuffer() called on unallocated buffer." << llendl;
 	}
 		
-	if (!mVertexLocked && useVBOs())
+	if (useVBOs())
 	{
+
+		if (!sDisableVBOMapping && gGLManager.mHasMapBufferRange)
+		{
+			if (count == -1)
+			{
+				count = mNumVerts;
+			}
+
+			bool mapped = false;
+			//see if range is already mapped
+			for (U32 i = 0; i < mMappedVertexRegions.size(); ++i)
+			{
+				MappedRegion& region = mMappedVertexRegions[i];
+				if (region.mType == type)
+				{
+					if (expand_region(region, index, count))
+					{
+						mapped = true;
+					}
+				}
+			}
+
+			if (!mapped)
+			{
+				//not already mapped, map new region
+				MappedRegion region(type, map_range ? -1 : index, count);
+				mMappedVertexRegions.push_back(region);
+			}
+		}
+
+		if (mVertexLocked && map_range)
+		{
+			llerrs << "Attempted to map a specific range of a buffer that was already mapped." << llendl;
+		}
+
+		if (!mVertexLocked)
 		{
 			LLMemType mt_v(LLMemType::MTYPE_VERTEX_MAP_BUFFER_VERTICES);
 			setBuffer(0, type);
@@ -957,61 +1011,91 @@ U8* LLVertexBuffer::mapVertexBuffer(S32 type, S32 access)
 
 			if(sDisableVBOMapping)
 			{
+				map_range = false;
 				allocateClientVertexBuffer() ;
 			}
 			else
 			{
-				U8* src = (U8*) glMapBufferARB(GL_ARRAY_BUFFER_ARB, GL_WRITE_ONLY_ARB);
+				U8* src = NULL;
+				if (gGLManager.mHasMapBufferRange)
+				{
+					if (map_range)
+					{
+						S32 offset = mOffsets[type] + sTypeSize[type]*index;
+						S32 length = (sTypeSize[type]*count+0xF) & ~0xF;
+						src = (U8*) glMapBufferRange(GL_ARRAY_BUFFER_ARB, offset, length, GL_MAP_WRITE_BIT | GL_MAP_FLUSH_EXPLICIT_BIT | GL_MAP_INVALIDATE_RANGE_BIT);
+					}
+					else
+					{
+						src = (U8*) glMapBufferRange(GL_ARRAY_BUFFER_ARB, 0, mSize, GL_MAP_WRITE_BIT | GL_MAP_FLUSH_EXPLICIT_BIT);
+					}
+				}
+				else
+				{
+					map_range = false;
+					src = (U8*) glMapBufferARB(GL_ARRAY_BUFFER_ARB, GL_WRITE_ONLY_ARB);
+				}
+
 				mMappedData = LL_NEXT_ALIGNED_ADDRESS<U8>(src);
 				mAlignedOffset = mMappedData - src;
 			
 				stop_glerror();
 			}
-		}
-		
-		
-		if (!mMappedData)
-		{
-			log_glerror();
-
-			//check the availability of memory
-			U32 avail_phy_mem, avail_vir_mem;
-			LLMemoryInfo::getAvailableMemoryKB(avail_phy_mem, avail_vir_mem) ;
-			llinfos << "Available physical mwmory(KB): " << avail_phy_mem << llendl ; 
-			llinfos << "Available virtual memory(KB): " << avail_vir_mem << llendl;
-
-			if(!sDisableVBOMapping)
-			{			
-				//--------------------
-				//print out more debug info before crash
-				llinfos << "vertex buffer size: (num verts : num indices) = " << getNumVerts() << " : " << getNumIndices() << llendl ;
-				GLint size ;
-				glGetBufferParameterivARB(GL_ARRAY_BUFFER_ARB, GL_BUFFER_SIZE_ARB, &size) ;
-				llinfos << "GL_ARRAY_BUFFER_ARB size is " << size << llendl ;
-				//--------------------
+				
+			if (!mMappedData)
+			{
+				log_glerror();
+
+				//check the availability of memory
+				U32 avail_phy_mem, avail_vir_mem;
+				LLMemoryInfo::getAvailableMemoryKB(avail_phy_mem, avail_vir_mem) ;
+				llinfos << "Available physical mwmory(KB): " << avail_phy_mem << llendl ; 
+				llinfos << "Available virtual memory(KB): " << avail_vir_mem << llendl;
+
+				if(!sDisableVBOMapping)
+				{			
+					//--------------------
+					//print out more debug info before crash
+					llinfos << "vertex buffer size: (num verts : num indices) = " << getNumVerts() << " : " << getNumIndices() << llendl ;
+					GLint size ;
+					glGetBufferParameterivARB(GL_ARRAY_BUFFER_ARB, GL_BUFFER_SIZE_ARB, &size) ;
+					llinfos << "GL_ARRAY_BUFFER_ARB size is " << size << llendl ;
+					//--------------------
+
+					GLint buff;
+					glGetIntegerv(GL_ARRAY_BUFFER_BINDING_ARB, &buff);
+					if ((GLuint)buff != mGLBuffer)
+					{
+						llerrs << "Invalid GL vertex buffer bound: " << buff << llendl;
+					}
 
-				GLint buff;
-				glGetIntegerv(GL_ARRAY_BUFFER_BINDING_ARB, &buff);
-				if ((GLuint)buff != mGLBuffer)
+							
+					llerrs << "glMapBuffer returned NULL (no vertex data)" << llendl;
+				}
+				else
 				{
-					llerrs << "Invalid GL vertex buffer bound: " << buff << llendl;
+					llerrs << "memory allocation for vertex data failed." << llendl ;
 				}
-
-							
-				llerrs << "glMapBuffer returned NULL (no vertex data)" << llendl;
-			}
-			else
-			{
-				llerrs << "memory allocation for vertex data failed." << llendl ;
 			}
+			sMappedCount++;
 		}
-		sMappedCount++;
+	}
+	else
+	{
+		map_range = false;
 	}
 	
-	return mMappedData;
+	if (map_range)
+	{
+		return mMappedData;
+	}
+	else
+	{
+		return mMappedData+mOffsets[type]+sTypeSize[type]*index;
+	}
 }
 
-U8* LLVertexBuffer::mapIndexBuffer(S32 access)
+U8* LLVertexBuffer::mapIndexBuffer(S32 index, S32 count, bool map_range)
 {
 	LLMemType mt2(LLMemType::MTYPE_VERTEX_MAP_BUFFER);
 	if (mFinal)
@@ -1023,8 +1107,40 @@ U8* LLVertexBuffer::mapIndexBuffer(S32 access)
 		llerrs << "LLVertexBuffer::mapIndexBuffer() called on unallocated buffer." << llendl;
 	}
 
-	if (!mIndexLocked && useVBOs())
+	if (useVBOs())
 	{
+		if (!sDisableVBOMapping && gGLManager.mHasMapBufferRange)
+		{
+			if (count == -1)
+			{
+				count = mNumIndices;
+			}
+
+			bool mapped = false;
+			//see if range is already mapped
+			for (U32 i = 0; i < mMappedIndexRegions.size(); ++i)
+			{
+				MappedRegion& region = mMappedIndexRegions[i];
+				if (expand_region(region, index, count))
+				{
+					mapped = true;
+				}
+			}
+
+			if (!mapped)
+			{
+				//not already mapped, map new region
+				MappedRegion region(TYPE_INDEX, map_range ? -1 : index, count);
+				mMappedIndexRegions.push_back(region);
+			}
+		}
+
+		if (mIndexLocked && map_range)
+		{
+			llerrs << "Attempted to map a specific range of a buffer that was already mapped." << llendl;
+		}
+
+		if (!mIndexLocked)
 		{
 			LLMemType mt_v(LLMemType::MTYPE_VERTEX_MAP_BUFFER_INDICES);
 
@@ -1034,12 +1150,32 @@ U8* LLVertexBuffer::mapIndexBuffer(S32 access)
 
 			if(sDisableVBOMapping)
 			{
+				map_range = false;
 				allocateClientIndexBuffer() ;
 			}
 			else
 			{
-				U8* src = (U8*) glMapBufferARB(GL_ELEMENT_ARRAY_BUFFER_ARB, GL_WRITE_ONLY_ARB);
-				mMappedIndexData = LL_NEXT_ALIGNED_ADDRESS<U8>(src);
+				U8* src = NULL;
+				if (gGLManager.mHasMapBufferRange)
+				{
+					if (map_range)
+					{
+						S32 offset = sizeof(U16)*index;
+						S32 length = sizeof(U16)*count;
+						src = (U8*) glMapBufferRange(GL_ELEMENT_ARRAY_BUFFER_ARB, offset, length, GL_MAP_WRITE_BIT | GL_MAP_FLUSH_EXPLICIT_BIT | GL_MAP_INVALIDATE_RANGE_BIT);
+					}
+					else
+					{
+						src = (U8*) glMapBufferRange(GL_ELEMENT_ARRAY_BUFFER_ARB, 0, sizeof(U16)*mNumIndices, GL_MAP_WRITE_BIT | GL_MAP_FLUSH_EXPLICIT_BIT);
+					}
+				}
+				else
+				{
+					map_range = false;
+					src = (U8*) glMapBufferARB(GL_ELEMENT_ARRAY_BUFFER_ARB, GL_WRITE_ONLY_ARB);
+				}
+
+				mMappedIndexData = src; //LL_NEXT_ALIGNED_ADDRESS<U8>(src);
 				mAlignedIndexOffset = mMappedIndexData - src;
 				stop_glerror();
 			}
@@ -1068,19 +1204,31 @@ U8* LLVertexBuffer::mapIndexBuffer(S32 access)
 
 		sMappedCount++;
 	}
+	else
+	{
+		map_range = false;
+	}
 
-	return mMappedIndexData ;
+	if (map_range)
+	{
+		return mMappedIndexData;
+	}
+	else
+	{
+		return mMappedIndexData + sizeof(U16)*index;
+	}
 }
 
 void LLVertexBuffer::unmapBuffer(S32 type)
 {
 	LLMemType mt2(LLMemType::MTYPE_VERTEX_UNMAP_BUFFER);
-	if (!useVBOs())
+	if (!useVBOs() || type == -2)
 	{
 		return ; //nothing to unmap
 	}
 
 	bool updated_all = false ;
+
 	if (mMappedData && mVertexLocked && type != TYPE_INDEX)
 	{
 		updated_all = (mIndexLocked && type < 0) ; //both vertex and index buffers done updating
@@ -1093,6 +1241,23 @@ void LLVertexBuffer::unmapBuffer(S32 type)
 		}
 		else
 		{
+			if (gGLManager.mHasMapBufferRange)
+			{
+				if (!mMappedVertexRegions.empty())
+				{
+					stop_glerror();
+					for (U32 i = 0; i < mMappedVertexRegions.size(); ++i)
+					{
+						const MappedRegion& region = mMappedVertexRegions[i];
+						S32 offset = region.mIndex >= 0 ? mOffsets[region.mType]+sTypeSize[region.mType]*region.mIndex : 0;
+						S32 length = sTypeSize[region.mType]*region.mCount;
+						glFlushMappedBufferRange(GL_ARRAY_BUFFER_ARB, offset, length);
+						stop_glerror();
+					}
+
+					mMappedVertexRegions.clear();
+				}
+			}
 			stop_glerror();
 			glUnmapBufferARB(GL_ARRAY_BUFFER_ARB);
 			stop_glerror();
@@ -1103,8 +1268,8 @@ void LLVertexBuffer::unmapBuffer(S32 type)
 		mVertexLocked = FALSE ;
 		sMappedCount--;
 	}
-
-	if(mMappedIndexData && mIndexLocked && (type < 0 || type == TYPE_INDEX))
+	
+	if (mMappedIndexData && mIndexLocked && (type < 0 || type == TYPE_INDEX))
 	{
 		if(sDisableVBOMapping)
 		{
@@ -1114,6 +1279,23 @@ void LLVertexBuffer::unmapBuffer(S32 type)
 		}
 		else
 		{
+
+			if (gGLManager.mHasMapBufferRange)
+			{
+				if (!mMappedIndexRegions.empty())
+				{
+					for (U32 i = 0; i < mMappedIndexRegions.size(); ++i)
+					{
+						const MappedRegion& region = mMappedIndexRegions[i];
+						S32 offset = region.mIndex >= 0 ? sizeof(U16)*region.mIndex : 0;
+						S32 length = sizeof(U16)*region.mCount;
+						glFlushMappedBufferRange(GL_ELEMENT_ARRAY_BUFFER_ARB, offset, length);
+						stop_glerror();
+					}
+
+					mMappedIndexRegions.clear();
+				}
+			}
 			stop_glerror();
 			glUnmapBufferARB(GL_ELEMENT_ARRAY_BUFFER_ARB);
 			stop_glerror();
@@ -1152,19 +1334,21 @@ template <class T,S32 type> struct VertexBufferStrider
 	typedef LLStrider<T> strider_t;
 	static bool get(LLVertexBuffer& vbo, 
 					strider_t& strider, 
-					S32 index)
+					S32 index, S32 count, bool map_range)
 	{
 		if (type == LLVertexBuffer::TYPE_INDEX)
 		{
 			S32 stride = sizeof(T);
 
-			if (vbo.mapIndexBuffer() == NULL)
+			U8* ptr = vbo.mapIndexBuffer(index, count, map_range);
+
+			if (ptr == NULL)
 			{
 				llwarns << "mapIndexBuffer failed!" << llendl;
 				return FALSE;
 			}
 
-			strider = (T*)(vbo.getMappedIndices() + index*stride);
+			strider = (T*)ptr;
 			strider.setStride(0);
 			return TRUE;
 		}
@@ -1172,13 +1356,15 @@ template <class T,S32 type> struct VertexBufferStrider
 		{
 			S32 stride = LLVertexBuffer::sTypeSize[type];
 
-			if (vbo.mapVertexBuffer(type) == NULL)
+			U8* ptr = vbo.mapVertexBuffer(type, index, count, map_range);
+
+			if (ptr == NULL)
 			{
 				llwarns << "mapVertexBuffer failed!" << llendl;
 				return FALSE;
 			}
 
-			strider = (T*)(vbo.getMappedData() + vbo.getOffset(type)+index*stride);
+			strider = (T*)ptr;
 			strider.setStride(stride);
 			return TRUE;
 		}
@@ -1190,55 +1376,48 @@ template <class T,S32 type> struct VertexBufferStrider
 	}
 };
 
-bool LLVertexBuffer::getVertexStrider(LLStrider<LLVector3>& strider, S32 index)
+bool LLVertexBuffer::getVertexStrider(LLStrider<LLVector3>& strider, S32 index, S32 count, bool map_range)
 {
-	return VertexBufferStrider<LLVector3,TYPE_VERTEX>::get(*this, strider, index);
+	return VertexBufferStrider<LLVector3,TYPE_VERTEX>::get(*this, strider, index, count, map_range);
 }
-bool LLVertexBuffer::getIndexStrider(LLStrider<U16>& strider, S32 index)
+bool LLVertexBuffer::getIndexStrider(LLStrider<U16>& strider, S32 index, S32 count, bool map_range)
 {
-	return VertexBufferStrider<U16,TYPE_INDEX>::get(*this, strider, index);
+	return VertexBufferStrider<U16,TYPE_INDEX>::get(*this, strider, index, count, map_range);
 }
-bool LLVertexBuffer::getTexCoord0Strider(LLStrider<LLVector2>& strider, S32 index)
+bool LLVertexBuffer::getTexCoord0Strider(LLStrider<LLVector2>& strider, S32 index, S32 count, bool map_range)
 {
-	return VertexBufferStrider<LLVector2,TYPE_TEXCOORD0>::get(*this, strider, index);
+	return VertexBufferStrider<LLVector2,TYPE_TEXCOORD0>::get(*this, strider, index, count, map_range);
 }
-bool LLVertexBuffer::getTexCoord1Strider(LLStrider<LLVector2>& strider, S32 index)
+bool LLVertexBuffer::getTexCoord1Strider(LLStrider<LLVector2>& strider, S32 index, S32 count, bool map_range)
 {
-	return VertexBufferStrider<LLVector2,TYPE_TEXCOORD1>::get(*this, strider, index);
+	return VertexBufferStrider<LLVector2,TYPE_TEXCOORD1>::get(*this, strider, index, count, map_range);
 }
-/*bool LLVertexBuffer::getTexCoord2Strider(LLStrider<LLVector2>& strider, S32 index)
-{
-	return VertexBufferStrider<LLVector2,TYPE_TEXCOORD2>::get(*this, strider, index);
-}
-bool LLVertexBuffer::getTexCoord3Strider(LLStrider<LLVector2>& strider, S32 index)
-{
-	return VertexBufferStrider<LLVector2,TYPE_TEXCOORD3>::get(*this, strider, index);
-}*/
-bool LLVertexBuffer::getNormalStrider(LLStrider<LLVector3>& strider, S32 index)
+
+bool LLVertexBuffer::getNormalStrider(LLStrider<LLVector3>& strider, S32 index, S32 count, bool map_range)
 {
-	return VertexBufferStrider<LLVector3,TYPE_NORMAL>::get(*this, strider, index);
+	return VertexBufferStrider<LLVector3,TYPE_NORMAL>::get(*this, strider, index, count, map_range);
 }
-bool LLVertexBuffer::getBinormalStrider(LLStrider<LLVector3>& strider, S32 index)
+bool LLVertexBuffer::getBinormalStrider(LLStrider<LLVector3>& strider, S32 index, S32 count, bool map_range)
 {
-	return VertexBufferStrider<LLVector3,TYPE_BINORMAL>::get(*this, strider, index);
+	return VertexBufferStrider<LLVector3,TYPE_BINORMAL>::get(*this, strider, index, count, map_range);
 }
-bool LLVertexBuffer::getColorStrider(LLStrider<LLColor4U>& strider, S32 index)
+bool LLVertexBuffer::getColorStrider(LLStrider<LLColor4U>& strider, S32 index, S32 count, bool map_range)
 {
-	return VertexBufferStrider<LLColor4U,TYPE_COLOR>::get(*this, strider, index);
+	return VertexBufferStrider<LLColor4U,TYPE_COLOR>::get(*this, strider, index, count, map_range);
 }
-bool LLVertexBuffer::getWeightStrider(LLStrider<F32>& strider, S32 index)
+bool LLVertexBuffer::getWeightStrider(LLStrider<F32>& strider, S32 index, S32 count, bool map_range)
 {
-	return VertexBufferStrider<F32,TYPE_WEIGHT>::get(*this, strider, index);
+	return VertexBufferStrider<F32,TYPE_WEIGHT>::get(*this, strider, index, count, map_range);
 }
 
-bool LLVertexBuffer::getWeight4Strider(LLStrider<LLVector4>& strider, S32 index)
+bool LLVertexBuffer::getWeight4Strider(LLStrider<LLVector4>& strider, S32 index, S32 count, bool map_range)
 {
-	return VertexBufferStrider<LLVector4,TYPE_WEIGHT4>::get(*this, strider, index);
+	return VertexBufferStrider<LLVector4,TYPE_WEIGHT4>::get(*this, strider, index, count, map_range);
 }
 
-bool LLVertexBuffer::getClothWeightStrider(LLStrider<LLVector4>& strider, S32 index)
+bool LLVertexBuffer::getClothWeightStrider(LLStrider<LLVector4>& strider, S32 index, S32 count, bool map_range)
 {
-	return VertexBufferStrider<LLVector4,TYPE_CLOTHWEIGHT>::get(*this, strider, index);
+	return VertexBufferStrider<LLVector4,TYPE_CLOTHWEIGHT>::get(*this, strider, index, count, map_range);
 }
 
 //----------------------------------------------------------------------------
@@ -1510,11 +1689,3 @@ void LLVertexBuffer::setupVertexBuffer(U32 data_mask) const
 	llglassertok();
 }
 
-void LLVertexBuffer::markDirty(U32 vert_index, U32 vert_count, U32 indices_index, U32 indices_count)
-{
-	// TODO: use GL_APPLE_flush_buffer_range here
-	/*if (useVBOs() && !mFilthy)
-	{
-	
-	}*/
-}
diff --git a/indra/llrender/llvertexbuffer.h b/indra/llrender/llvertexbuffer.h
index 0c4b241537..aa5df305a6 100644
--- a/indra/llrender/llvertexbuffer.h
+++ b/indra/llrender/llvertexbuffer.h
@@ -77,6 +77,18 @@ protected:
 class LLVertexBuffer : public LLRefCount
 {
 public:
+	class MappedRegion
+	{
+	public:
+		S32 mType;
+		S32 mIndex;
+		S32 mCount;
+		
+		MappedRegion(S32 type, S32 index, S32 count)
+			: mType(type), mIndex(index), mCount(count)
+		{ }	
+	};
+
 	LLVertexBuffer(const LLVertexBuffer& rhs)
 	{
 		*this = rhs;
@@ -177,8 +189,8 @@ public:
 	LLVertexBuffer(U32 typemask, S32 usage);
 	
 	// map for data access
-	U8*		mapVertexBuffer(S32 type = -1, S32 access = -1);
-	U8*		mapIndexBuffer(S32 access = -1);
+	U8*		mapVertexBuffer(S32 type, S32 index, S32 count, bool map_range);
+	U8*		mapIndexBuffer(S32 index, S32 count, bool map_range);
 
 	// set for rendering
 	virtual void	setBuffer(U32 data_mask, S32 type = -1); 	// calls  setupVertexBuffer() if data_mask is not 0
@@ -193,16 +205,16 @@ public:
 	//   vb->getNormalStrider(norms);
 	//   setVertsNorms(verts, norms);
 	//   vb->unmapBuffer();
-	bool getVertexStrider(LLStrider<LLVector3>& strider, S32 index=0);
-	bool getIndexStrider(LLStrider<U16>& strider, S32 index=0);
-	bool getTexCoord0Strider(LLStrider<LLVector2>& strider, S32 index=0);
-	bool getTexCoord1Strider(LLStrider<LLVector2>& strider, S32 index=0);
-	bool getNormalStrider(LLStrider<LLVector3>& strider, S32 index=0);
-	bool getBinormalStrider(LLStrider<LLVector3>& strider, S32 index=0);
-	bool getColorStrider(LLStrider<LLColor4U>& strider, S32 index=0);
-	bool getWeightStrider(LLStrider<F32>& strider, S32 index=0);
-	bool getWeight4Strider(LLStrider<LLVector4>& strider, S32 index=0);
-	bool getClothWeightStrider(LLStrider<LLVector4>& strider, S32 index=0);
+	bool getVertexStrider(LLStrider<LLVector3>& strider, S32 index=0, S32 count = -1, bool map_range = false);
+	bool getIndexStrider(LLStrider<U16>& strider, S32 index=0, S32 count = -1, bool map_range = false);
+	bool getTexCoord0Strider(LLStrider<LLVector2>& strider, S32 index=0, S32 count = -1, bool map_range = false);
+	bool getTexCoord1Strider(LLStrider<LLVector2>& strider, S32 index=0, S32 count = -1, bool map_range = false);
+	bool getNormalStrider(LLStrider<LLVector3>& strider, S32 index=0, S32 count = -1, bool map_range = false);
+	bool getBinormalStrider(LLStrider<LLVector3>& strider, S32 index=0, S32 count = -1, bool map_range = false);
+	bool getColorStrider(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 isEmpty() const					{ return mEmpty; }
 	BOOL isLocked() const					{ return mVertexLocked || mIndexLocked; }
@@ -222,8 +234,6 @@ public:
 	S32 getOffset(S32 type) const			{ return mOffsets[type]; }
 	S32 getUsage() const					{ return mUsage; }
 
-	void markDirty(U32 vert_index, U32 vert_count, U32 indices_index, U32 indices_count);
-
 	void draw(U32 mode, U32 count, U32 indices_offset) const;
 	void drawArrays(U32 mode, U32 offset, U32 count) const;
 	void drawRange(U32 mode, U32 start, U32 end, U32 count, U32 indices_offset) const;
@@ -257,20 +267,8 @@ protected:
 	BOOL	mDynamicSize;	// if TRUE, buffer has been resized at least once (and should be padded)
 	S32		mOffsets[TYPE_MAX];
 
-	class DirtyRegion
-	{
-	public:
-		U32 mIndex;
-		U32 mCount;
-		U32 mIndicesIndex;
-		U32 mIndicesCount;
-
-		DirtyRegion(U32 vi, U32 vc, U32 ii, U32 ic)
-			: mIndex(vi), mCount(vc), mIndicesIndex(ii), mIndicesCount(ic)
-		{ }	
-	};
-
-	std::vector<DirtyRegion> mDirtyRegions; //vector of dirty regions to rebuild
+	std::vector<MappedRegion> mMappedVertexRegions;
+	std::vector<MappedRegion> mMappedIndexRegions;
 
 public:
 	static S32 sCount;