12 files changed, 254 insertions, 273 deletions

diff --git a/indra/llrender/llglslshader.cpp b/indra/llrender/llglslshader.cpp
index 79979657f1..25e4a88f28 100644
--- a/indra/llrender/llglslshader.cpp
+++ b/indra/llrender/llglslshader.cpp
@@ -54,6 +54,8 @@ using std::string;
 GLuint LLGLSLShader::sCurBoundShader = 0;
 LLGLSLShader* LLGLSLShader::sCurBoundShaderPtr = NULL;
 S32 LLGLSLShader::sIndexedTextureChannels = 0;
+U32 LLGLSLShader::sMaxGLTFMaterials = 0;
+U32 LLGLSLShader::sMaxGLTFNodes = 0;
 bool LLGLSLShader::sProfileEnabled = false;
 std::set<LLGLSLShader*> LLGLSLShader::sInstances;
 LLGLSLShader::defines_map_t LLGLSLShader::sGlobalDefines;
@@ -978,7 +980,9 @@ bool LLGLSLShader::mapUniforms()
     const char* ubo_names[] =
     {
         "ReflectionProbes", // UB_REFLECTION_PROBES
-        "GLTFJoints", // UB_GLTF_JOINTS
+        "GLTFJoints",       // UB_GLTF_JOINTS
+        "GLTFNodes",        // UB_GLTF_NODES
+        "GLTFMaterials",    // UB_GLTF_MATERIALS
     };
 
     llassert(LL_ARRAY_SIZE(ubo_names) == NUM_UNIFORM_BLOCKS);
@@ -1099,7 +1103,8 @@ S32 LLGLSLShader::bindTexture(S32 uniform, LLTexture* texture, LLTexUnit::eTextu
 
     if (uniform < 0 || uniform >= (S32)mTexture.size())
     {
-        LL_SHADER_UNIFORM_ERRS() << "Uniform out of range: " << uniform << LL_ENDL;
+        LL_WARNS_ONCE("Shader") << "Uniform index out of bounds. Size: " << (S32)mUniform.size() << " index: " << uniform << LL_ENDL;
+        llassert(false);
         return -1;
     }
 
@@ -1120,6 +1125,8 @@ S32 LLGLSLShader::bindTexture(S32 uniform, LLRenderTarget* texture, bool depth,
 
     if (uniform < 0 || uniform >= (S32)mTexture.size())
     {
+        LL_WARNS_ONCE("Shader") << "Uniform index out of bounds. Size: " << (S32)mUniform.size() << " index: " << uniform << LL_ENDL;
+        llassert(false);
         return -1;
     }
 
@@ -1167,7 +1174,8 @@ S32 LLGLSLShader::unbindTexture(S32 uniform, LLTexUnit::eTextureType mode)
 
     if (uniform < 0 || uniform >= (S32)mTexture.size())
     {
-        LL_SHADER_UNIFORM_ERRS() << "Uniform out of range: " << uniform << LL_ENDL;
+        LL_WARNS_ONCE("Shader") << "Uniform index out of bounds. Size: " << (S32)mUniform.size() << " index: " << uniform << LL_ENDL;
+        llassert(false);
         return -1;
     }
 
@@ -1192,7 +1200,8 @@ S32 LLGLSLShader::enableTexture(S32 uniform, LLTexUnit::eTextureType mode, LLTex
 
     if (uniform < 0 || uniform >= (S32)mTexture.size())
     {
-        LL_SHADER_UNIFORM_ERRS() << "Uniform out of range: " << uniform << LL_ENDL;
+        LL_WARNS_ONCE("Shader") << "Uniform index out of bounds. Size: " << (S32)mUniform.size() << " index: " << uniform << LL_ENDL;
+        llassert(false);
         return -1;
     }
 
@@ -1213,7 +1222,8 @@ S32 LLGLSLShader::disableTexture(S32 uniform, LLTexUnit::eTextureType mode, LLTe
 
     if (uniform < 0 || uniform >= (S32)mTexture.size())
     {
-        LL_SHADER_UNIFORM_ERRS() << "Uniform out of range: " << uniform << LL_ENDL;
+        LL_WARNS_ONCE("Shader") << "Uniform index out of bounds. Size: " << (S32)mUniform.size() << " index: " << uniform << LL_ENDL;
+        llassert(false);
         return -1;
     }
     S32 index = mTexture[uniform];
@@ -1244,7 +1254,8 @@ void LLGLSLShader::uniform1i(U32 index, GLint x)
     {
         if (mUniform.size() <= index)
         {
-            LL_SHADER_UNIFORM_ERRS() << "Uniform index out of bounds." << LL_ENDL;
+            LL_WARNS_ONCE("Shader") << "Uniform index out of bounds. Size: " << (S32)mUniform.size() << " index: " << index << LL_ENDL;
+            llassert(false);
             return;
         }
 
@@ -1269,7 +1280,8 @@ void LLGLSLShader::uniform1f(U32 index, GLfloat x)
     {
         if (mUniform.size() <= index)
         {
-            LL_SHADER_UNIFORM_ERRS() << "Uniform index out of bounds." << LL_ENDL;
+            LL_WARNS_ONCE("Shader") << "Uniform index out of bounds. Size: " << (S32)mUniform.size() << " index: " << index << LL_ENDL;
+            llassert(false);
             return;
         }
 
@@ -1304,7 +1316,8 @@ void LLGLSLShader::uniform2f(U32 index, GLfloat x, GLfloat y)
     {
         if (mUniform.size() <= index)
         {
-            LL_SHADER_UNIFORM_ERRS() << "Uniform index out of bounds." << LL_ENDL;
+            LL_WARNS_ONCE("Shader") << "Uniform index out of bounds. Size: " << (S32)mUniform.size() << " index: " << index << LL_ENDL;
+            llassert(false);
             return;
         }
 
@@ -1330,7 +1343,8 @@ void LLGLSLShader::uniform3f(U32 index, GLfloat x, GLfloat y, GLfloat z)
     {
         if (mUniform.size() <= index)
         {
-            LL_SHADER_UNIFORM_ERRS() << "Uniform index out of bounds." << LL_ENDL;
+            LL_WARNS_ONCE("Shader") << "Uniform index out of bounds. Size: " << (S32)mUniform.size() << " index: " << index << LL_ENDL;
+            llassert(false);
             return;
         }
 
@@ -1356,7 +1370,8 @@ void LLGLSLShader::uniform4f(U32 index, GLfloat x, GLfloat y, GLfloat z, GLfloat
     {
         if (mUniform.size() <= index)
         {
-            LL_SHADER_UNIFORM_ERRS() << "Uniform index out of bounds." << LL_ENDL;
+            LL_WARNS_ONCE("Shader") << "Uniform index out of bounds. Size: " << (S32)mUniform.size() << " index: " << index << LL_ENDL;
+            llassert(false);
             return;
         }
 
@@ -1382,7 +1397,8 @@ void LLGLSLShader::uniform1iv(U32 index, U32 count, const GLint* v)
     {
         if (mUniform.size() <= index)
         {
-            LL_SHADER_UNIFORM_ERRS() << "Uniform index out of bounds." << LL_ENDL;
+            LL_WARNS_ONCE("Shader") << "Uniform index out of bounds. Size: " << (S32)mUniform.size() << " index: " << index << LL_ENDL;
+            llassert(false);
             return;
         }
 
@@ -1408,7 +1424,8 @@ void LLGLSLShader::uniform4iv(U32 index, U32 count, const GLint* v)
     {
         if (mUniform.size() <= index)
         {
-            LL_SHADER_UNIFORM_ERRS() << "Uniform index out of bounds." << LL_ENDL;
+            LL_WARNS_ONCE("Shader") << "Uniform index out of bounds. Size: " << (S32)mUniform.size() << " index: " << index << LL_ENDL;
+            llassert(false);
             return;
         }
 
@@ -1435,7 +1452,8 @@ void LLGLSLShader::uniform1fv(U32 index, U32 count, const GLfloat* v)
     {
         if (mUniform.size() <= index)
         {
-            LL_SHADER_UNIFORM_ERRS() << "Uniform index out of bounds." << LL_ENDL;
+            LL_WARNS_ONCE("Shader") << "Uniform index out of bounds. Size: " << (S32)mUniform.size() << " index: " << index << LL_ENDL;
+            llassert(false);
             return;
         }
 
@@ -1461,7 +1479,8 @@ void LLGLSLShader::uniform2fv(U32 index, U32 count, const GLfloat* v)
     {
         if (mUniform.size() <= index)
         {
-            LL_SHADER_UNIFORM_ERRS() << "Uniform index out of bounds." << LL_ENDL;
+            LL_WARNS_ONCE("Shader") << "Uniform index out of bounds. Size: " << (S32)mUniform.size() << " index: " << index << LL_ENDL;
+            llassert(false);
             return;
         }
 
@@ -1487,7 +1506,8 @@ void LLGLSLShader::uniform3fv(U32 index, U32 count, const GLfloat* v)
     {
         if (mUniform.size() <= index)
         {
-            LL_SHADER_UNIFORM_ERRS() << "Uniform index out of bounds." << LL_ENDL;
+            LL_WARNS_ONCE("Shader") << "Uniform index out of bounds. Size: " << (S32)mUniform.size() << " index: " << index << LL_ENDL;
+            llassert(false);
             return;
         }
 
@@ -1513,7 +1533,8 @@ void LLGLSLShader::uniform4fv(U32 index, U32 count, const GLfloat* v)
     {
         if (mUniform.size() <= index)
         {
-            LL_SHADER_UNIFORM_ERRS() << "Uniform index out of bounds." << LL_ENDL;
+            LL_WARNS_ONCE("Shader") << "Uniform index out of bounds. Size: " << (S32)mUniform.size() << " index: " << index << LL_ENDL;
+            llassert(false);
             return;
         }
 
@@ -1540,7 +1561,8 @@ void LLGLSLShader::uniformMatrix2fv(U32 index, U32 count, GLboolean transpose, c
     {
         if (mUniform.size() <= index)
         {
-            LL_SHADER_UNIFORM_ERRS() << "Uniform index out of bounds." << LL_ENDL;
+            LL_WARNS_ONCE("Shader") << "Uniform index out of bounds. Size: " << (S32)mUniform.size() << " index: " << index << LL_ENDL;
+            llassert(false);
             return;
         }
 
@@ -1560,7 +1582,8 @@ void LLGLSLShader::uniformMatrix3fv(U32 index, U32 count, GLboolean transpose, c
     {
         if (mUniform.size() <= index)
         {
-            LL_SHADER_UNIFORM_ERRS() << "Uniform index out of bounds." << LL_ENDL;
+            LL_WARNS_ONCE("Shader") << "Uniform index out of bounds. Size: " << (S32)mUniform.size() << " index: " << index << LL_ENDL;
+            llassert(false);
             return;
         }
 
@@ -1580,7 +1603,8 @@ void LLGLSLShader::uniformMatrix3x4fv(U32 index, U32 count, GLboolean transpose,
     {
         if (mUniform.size() <= index)
         {
-            LL_SHADER_UNIFORM_ERRS() << "Uniform index out of bounds." << LL_ENDL;
+            LL_WARNS_ONCE("Shader") << "Uniform index out of bounds. Size: " << (S32)mUniform.size() << " index: " << index << LL_ENDL;
+            llassert(false);
             return;
         }
 
@@ -1600,7 +1624,8 @@ void LLGLSLShader::uniformMatrix4fv(U32 index, U32 count, GLboolean transpose, c
     {
         if (mUniform.size() <= index)
         {
-            LL_SHADER_UNIFORM_ERRS() << "Uniform index out of bounds." << LL_ENDL;
+            LL_WARNS_ONCE("Shader") << "Uniform index out of bounds. Size: " << (S32)mUniform.size() << " index: " << index << LL_ENDL;
+            llassert(false);
             return;
         }
 
diff --git a/indra/llrender/llglslshader.h b/indra/llrender/llglslshader.h
index cc2ba0fcff..86e5625dca 100644
--- a/indra/llrender/llglslshader.h
+++ b/indra/llrender/llglslshader.h
@@ -147,8 +147,10 @@ public:
 
     enum UniformBlock : GLuint
     {
-        UB_REFLECTION_PROBES,
-        UB_GLTF_JOINTS,
+        UB_REFLECTION_PROBES,   // "ReflectionProbes"
+        UB_GLTF_JOINTS,         // "GLTFJoints"
+        UB_GLTF_NODES,          // "GLTFNodes"
+        UB_GLTF_MATERIALS,      // "GLTFMaterials"
         NUM_UNIFORM_BLOCKS
     };
 
@@ -163,6 +165,9 @@ public:
     static LLGLSLShader* sCurBoundShaderPtr;
     static S32 sIndexedTextureChannels;
 
+    static U32 sMaxGLTFMaterials;
+    static U32 sMaxGLTFNodes;
+
     static void initProfile();
     static void finishProfile(bool emit_report = true);
 
diff --git a/indra/llrender/llgltexture.cpp b/indra/llrender/llgltexture.cpp
index e614f45986..4dcca5a726 100644
--- a/indra/llrender/llgltexture.cpp
+++ b/indra/llrender/llgltexture.cpp
@@ -351,20 +351,6 @@ void LLGLTexture::forceUpdateBindStats(void) const
     return mGLTexturep->forceUpdateBindStats() ;
 }
 
-U32 LLGLTexture::getTexelsInAtlas() const
-{
-    llassert(mGLTexturep.notNull()) ;
-
-    return mGLTexturep->getTexelsInAtlas() ;
-}
-
-U32 LLGLTexture::getTexelsInGLTexture() const
-{
-    llassert(mGLTexturep.notNull()) ;
-
-    return mGLTexturep->getTexelsInGLTexture() ;
-}
-
 bool LLGLTexture::isGLTextureCreated() const
 {
     llassert(mGLTexturep.notNull()) ;
@@ -372,13 +358,6 @@ bool LLGLTexture::isGLTextureCreated() const
     return mGLTexturep->isGLTextureCreated() ;
 }
 
-S32  LLGLTexture::getDiscardLevelInAtlas() const
-{
-    llassert(mGLTexturep.notNull()) ;
-
-    return mGLTexturep->getDiscardLevelInAtlas() ;
-}
-
 void LLGLTexture::destroyGLTexture()
 {
     if(mGLTexturep.notNull() && mGLTexturep->getHasGLTexture())
diff --git a/indra/llrender/llgltexture.h b/indra/llrender/llgltexture.h
index 0901243f8f..a7de20dc5c 100644
--- a/indra/llrender/llgltexture.h
+++ b/indra/llrender/llgltexture.h
@@ -51,10 +51,10 @@ public:
         BOOST_NONE          = 0,
         BOOST_AVATAR        ,
         BOOST_AVATAR_BAKED  ,
-        BOOST_SCULPTED      ,
         BOOST_TERRAIN       , // Needed for minimap generation for now. Lower than BOOST_HIGH so the texture stats don't get forced, i.e. texture stats are manually managed by minimap/terrain instead.
 
         BOOST_HIGH          = 10,
+        BOOST_SCULPTED      ,
         BOOST_BUMP          ,
         BOOST_UNUSED_1      , // Placeholder to avoid disrupting habits around texture debug
         BOOST_SELECTED      ,
@@ -75,7 +75,6 @@ public:
         AVATAR_SCRATCH_TEX,
         DYNAMIC_TEX,
         MEDIA,
-        ATLAS,
         OTHER,
         MAX_GL_IMAGE_CATEGORY
     };
@@ -156,10 +155,7 @@ public:
     bool       isJustBound()const ;
     void       forceUpdateBindStats(void) const;
 
-    U32        getTexelsInAtlas() const ;
-    U32        getTexelsInGLTexture() const ;
     bool       isGLTextureCreated() const ;
-    S32        getDiscardLevelInAtlas() const ;
     LLGLTextureState getTextureState() const { return mTextureState; }
 
     //---------------------------------------------------------------------------------------------
diff --git a/indra/llrender/llimagegl.cpp b/indra/llrender/llimagegl.cpp
index 7e5cd628c1..956bcef352 100644
--- a/indra/llrender/llimagegl.cpp
+++ b/indra/llrender/llimagegl.cpp
@@ -41,6 +41,7 @@
 #include "llrender.h"
 #include "llwindow.h"
 #include "llframetimer.h"
+#include <unordered_set>
 
 extern LL_COMMON_API bool on_main_thread();
 
@@ -130,10 +131,9 @@ S32 LLImageGL::sCount                   = 0;
 
 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;
+std::unordered_set<LLImageGL*> LLImageGL::sImageList;
 
 
 bool LLImageGLThread::sEnabledTextures = false;
@@ -150,6 +150,9 @@ S32 LLImageGL::sMaxCategories = 1 ;
 
 //optimization for when we don't need to calculate mIsMask
 bool LLImageGL::sSkipAnalyzeAlpha;
+U32  LLImageGL::sScratchPBO = 0;
+U32  LLImageGL::sScratchPBOSize = 0;
+
 
 //------------------------
 //****************************************************************************************************
@@ -159,20 +162,6 @@ bool LLImageGL::sSkipAnalyzeAlpha;
 //**************************************************************************************
 //below are functions for debug use
 //do not delete them even though they are not currently being used.
-void check_all_images()
-{
-    for (std::set<LLImageGL*>::iterator iter = LLImageGL::sImageList.begin();
-         iter != LLImageGL::sImageList.end(); iter++)
-    {
-        LLImageGL* glimage = *iter;
-        if (glimage->getTexName() && glimage->isGLTextureCreated())
-        {
-            gGL.getTexUnit(0)->bind(glimage) ;
-            glimage->checkTexSize() ;
-            gGL.getTexUnit(0)->unbind(glimage->getTarget()) ;
-        }
-    }
-}
 
 void LLImageGL::checkTexSize(bool forced) const
 {
@@ -252,6 +241,11 @@ void LLImageGL::initClass(LLWindow* window, S32 num_catagories, bool skip_analyz
     LL_PROFILE_ZONE_SCOPED_CATEGORY_TEXTURE;
     sSkipAnalyzeAlpha = skip_analyze_alpha;
 
+    if (sScratchPBO == 0)
+    {
+        glGenBuffers(1, &sScratchPBO);
+    }
+
     if (thread_texture_loads || thread_media_updates)
     {
         LLImageGLThread::createInstance(window);
@@ -265,6 +259,12 @@ void LLImageGL::cleanupClass()
 {
     LL_PROFILE_ZONE_SCOPED_CATEGORY_TEXTURE;
     LLImageGLThread::deleteSingleton();
+    if (sScratchPBO != 0)
+    {
+        glDeleteBuffers(1, &sScratchPBO);
+        sScratchPBO = 0;
+        sScratchPBOSize = 0;
+    }
 }
 
 
@@ -360,66 +360,19 @@ void LLImageGL::updateStats(F32 current_time)
 //----------------------------------------------------------------------------
 
 //static
-void LLImageGL::destroyGL(bool save_state)
+void LLImageGL::destroyGL()
 {
     for (S32 stage = 0; stage < gGLManager.mNumTextureImageUnits; stage++)
     {
         gGL.getTexUnit(stage)->unbind(LLTexUnit::TT_TEXTURE);
     }
-
-    sAllowReadBackRaw = true ;
-    for (std::set<LLImageGL*>::iterator iter = sImageList.begin();
-         iter != sImageList.end(); iter++)
-    {
-        LLImageGL* glimage = *iter;
-        if (glimage->mTexName)
-        {
-            if (save_state && glimage->isGLTextureCreated() && glimage->mComponents)
-            {
-                glimage->mSaveData = new LLImageRaw;
-                if(!glimage->readBackRaw(glimage->mCurrentDiscardLevel, glimage->mSaveData, false)) //necessary, keep it.
-                {
-                    glimage->mSaveData = NULL ;
-                }
-            }
-
-            glimage->destroyGLTexture();
-            stop_glerror();
-        }
-    }
-    sAllowReadBackRaw = false ;
-}
-
-//static
-void LLImageGL::restoreGL()
-{
-    for (std::set<LLImageGL*>::iterator iter = sImageList.begin();
-         iter != sImageList.end(); iter++)
-    {
-        LLImageGL* glimage = *iter;
-        if(glimage->getTexName())
-        {
-            LL_ERRS() << "tex name is not 0." << LL_ENDL ;
-        }
-        if (glimage->mSaveData.notNull())
-        {
-            if (glimage->getComponents() && glimage->mSaveData->getComponents())
-            {
-                glimage->createGLTexture(glimage->mCurrentDiscardLevel, glimage->mSaveData, 0, true, glimage->getCategory());
-                stop_glerror();
-            }
-            glimage->mSaveData = NULL; // deletes data
-        }
-    }
 }
 
 //static
 void LLImageGL::dirtyTexOptions()
 {
-    for (std::set<LLImageGL*>::iterator iter = sImageList.begin();
-         iter != sImageList.end(); iter++)
+    for (auto& glimage : sImageList)
     {
-        LLImageGL* glimage = *iter;
         glimage->mTexOptionsDirty = true;
         stop_glerror();
     }
@@ -542,10 +495,6 @@ void LLImageGL::init(bool usemipmaps)
     mHeight = 0;
     mCurrentDiscardLevel = -1;
 
-    mDiscardLevelInAtlas = -1 ;
-    mTexelsInAtlas = 0 ;
-    mTexelsInGLTexture = 0 ;
-
     mAllowCompression = true;
 
     mTarget = GL_TEXTURE_2D;
@@ -622,9 +571,6 @@ bool LLImageGL::setSize(S32 width, S32 height, S32 ncomponents, S32 discard_leve
             return false;
         }
 
-        // pickmask validity depends on old image size, delete it
-        freePickMask();
-
         mWidth = width;
         mHeight = height;
         mComponents = ncomponents;
@@ -1025,98 +971,6 @@ bool LLImageGL::setImage(const U8* data_in, bool data_hasmips /* = false */, S32
     return true;
 }
 
-bool LLImageGL::preAddToAtlas(S32 discard_level, const LLImageRaw* raw_image)
-{
-    //not compatible with core GL profile
-    llassert(!LLRender::sGLCoreProfile);
-
-    if (gGLManager.mIsDisabled)
-    {
-        LL_WARNS() << "Trying to create a texture while GL is disabled!" << LL_ENDL;
-        return false;
-    }
-    llassert(gGLManager.mInited);
-    stop_glerror();
-
-    if (discard_level < 0)
-    {
-        llassert(mCurrentDiscardLevel >= 0);
-        discard_level = mCurrentDiscardLevel;
-    }
-
-    // Actual image width/height = raw image width/height * 2^discard_level
-    S32 w = raw_image->getWidth() << discard_level;
-    S32 h = raw_image->getHeight() << discard_level;
-
-    // setSize may call destroyGLTexture if the size does not match
-    if (!setSize(w, h, raw_image->getComponents(), discard_level))
-    {
-        LL_WARNS() << "Trying to create a texture with incorrect dimensions!" << LL_ENDL;
-        return false;
-    }
-
-    if (!mHasExplicitFormat)
-    {
-        switch (mComponents)
-        {
-            case 1:
-                // Use luminance alpha (for fonts)
-                mFormatInternal = GL_LUMINANCE8;
-                mFormatPrimary  = GL_LUMINANCE;
-                mFormatType     = GL_UNSIGNED_BYTE;
-                break;
-            case 2:
-                // Use luminance alpha (for fonts)
-                mFormatInternal = GL_LUMINANCE8_ALPHA8;
-                mFormatPrimary  = GL_LUMINANCE_ALPHA;
-                mFormatType     = GL_UNSIGNED_BYTE;
-                break;
-            case 3:
-                mFormatInternal = GL_RGB8;
-                mFormatPrimary = GL_RGB;
-                mFormatType    = GL_UNSIGNED_BYTE;
-                break;
-            case 4:
-                mFormatInternal = GL_RGBA8;
-                mFormatPrimary = GL_RGBA;
-                mFormatType    = GL_UNSIGNED_BYTE;
-                break;
-            default:
-                LL_ERRS() << "Bad number of components for texture: " << (U32) getComponents() << LL_ENDL;
-        }
-    }
-
-    mCurrentDiscardLevel = discard_level;
-    mDiscardLevelInAtlas = discard_level;
-    mTexelsInAtlas = raw_image->getWidth() * raw_image->getHeight() ;
-    mLastBindTime = sLastFrameTime;
-    mGLTextureCreated = false ;
-
-    glPixelStorei(GL_UNPACK_ROW_LENGTH, raw_image->getWidth());
-    stop_glerror();
-
-    if(mFormatSwapBytes)
-    {
-        glPixelStorei(GL_UNPACK_SWAP_BYTES, 1);
-        stop_glerror();
-    }
-
-    return true ;
-}
-
-void LLImageGL::postAddToAtlas()
-{
-    if(mFormatSwapBytes)
-    {
-        glPixelStorei(GL_UNPACK_SWAP_BYTES, 0);
-        stop_glerror();
-    }
-
-    glPixelStorei(GL_UNPACK_ROW_LENGTH, 0);
-    gGL.getTexUnit(0)->setTextureFilteringOption(mFilterOption);
-    stop_glerror();
-}
-
 U32 type_width_from_pixtype(U32 pixtype)
 {
     U32 type_width = 0;
@@ -1752,7 +1606,6 @@ bool LLImageGL::createGLTexture(S32 discard_level, const U8* data_in, bool data_
 
 
     mTextureMemory = (S64Bytes)getMipBytes(mCurrentDiscardLevel);
-    mTexelsInGLTexture = getWidth() * getHeight();
 
     // mark this as bound at this point, so we don't throw it out immediately
     mLastBindTime = sLastFrameTime;
@@ -1830,8 +1683,7 @@ void LLImageGL::syncTexName(LLGLuint texname)
 
 bool LLImageGL::readBackRaw(S32 discard_level, LLImageRaw* imageraw, bool compressed_ok) const
 {
-    llassert_always(sAllowReadBackRaw) ;
-    //LL_ERRS() << "should not call this function!" << LL_ENDL ;
+    LL_PROFILE_ZONE_SCOPED_CATEGORY_TEXTURE;
 
     if (discard_level < 0)
     {
@@ -2297,6 +2149,8 @@ void LLImageGL::analyzeAlpha(const void* data_in, U32 w, U32 h)
 //----------------------------------------------------------------------------
 U32 LLImageGL::createPickMask(S32 pWidth, S32 pHeight)
 {
+    LL_PROFILE_ZONE_SCOPED_CATEGORY_TEXTURE;
+    freePickMask();
     U32 pick_width = pWidth/2 + 1;
     U32 pick_height = pHeight/2 + 1;
 
@@ -2314,7 +2168,6 @@ U32 LLImageGL::createPickMask(S32 pWidth, S32 pHeight)
 //----------------------------------------------------------------------------
 void LLImageGL::freePickMask()
 {
-    // pickmask validity depends on old image size, delete it
     if (mPickMask != NULL)
     {
         delete [] mPickMask;
@@ -2352,16 +2205,16 @@ void LLImageGL::updatePickMask(S32 width, S32 height, const U8* data_in)
         return ;
     }
 
-    freePickMask();
-
     if (mFormatType != GL_UNSIGNED_BYTE ||
         ((mFormatPrimary != GL_RGBA)
       && (mFormatPrimary != GL_SRGB_ALPHA)))
     {
         //cannot generate a pick mask for this texture
+        freePickMask();
         return;
     }
 
+
 #ifdef SHOW_ASSERT
     const U32 pickSize = createPickMask(width, height);
 #else // SHOW_ASSERT
@@ -2460,6 +2313,73 @@ void LLImageGL::resetCurTexSizebar()
     sCurTexSizeBar = -1 ;
     sCurTexPickSize = -1 ;
 }
+
+bool LLImageGL::scaleDown(S32 desired_discard)
+{
+    LL_PROFILE_ZONE_SCOPED_CATEGORY_TEXTURE;
+
+    if (mTarget != GL_TEXTURE_2D)
+    {
+        return false;
+    }
+
+    desired_discard = llmin(desired_discard, mMaxDiscardLevel);
+
+    if (desired_discard <= mCurrentDiscardLevel)
+    {
+        return false;
+    }
+
+    S32 mip = desired_discard - mCurrentDiscardLevel;
+
+    S32 desired_width = getWidth(desired_discard);
+    S32 desired_height = getHeight(desired_discard);
+
+    U64 size = getBytes(desired_discard);
+    llassert(size <= 2048*2048*4); // we shouldn't be using this method to downscale huge textures, but it'll work
+    gGL.getTexUnit(0)->bind(this);
+
+
+    if (sScratchPBO == 0)
+    {
+        glGenBuffers(1, &sScratchPBO);
+        sScratchPBOSize = 0;
+    }
+
+    glBindBuffer(GL_PIXEL_PACK_BUFFER, sScratchPBO);
+
+    if (size > sScratchPBOSize)
+    {
+        glBufferData(GL_PIXEL_PACK_BUFFER, size, NULL, GL_STREAM_COPY);
+        sScratchPBOSize = size;
+    }
+
+    glGetTexImage(mTarget, mip, mFormatPrimary, mFormatType, nullptr);
+
+    free_tex_image(mTexName);
+
+    glBindBuffer(GL_PIXEL_PACK_BUFFER, 0);
+
+    glBindBuffer(GL_PIXEL_UNPACK_BUFFER, sScratchPBO);
+    glTexImage2D(mTarget, 0, mFormatPrimary, desired_width, desired_height, 0, mFormatPrimary, mFormatType, nullptr);
+    glBindBuffer(GL_PIXEL_UNPACK_BUFFER, 0);
+
+    alloc_tex_image(desired_width, desired_height, mFormatPrimary);
+
+    if (mHasMipMaps)
+    {
+        LL_PROFILE_ZONE_NAMED_CATEGORY_TEXTURE("scaleDown - glGenerateMipmap");
+        glGenerateMipmap(mTarget);
+    }
+
+    gGL.getTexUnit(0)->unbind(LLTexUnit::TT_TEXTURE);
+
+    mCurrentDiscardLevel = desired_discard;
+
+    return true;
+}
+
+
 //----------------------------------------------------------------------------
 #if LL_IMAGEGL_THREAD_CHECK
 void LLImageGL::checkActiveThread()
diff --git a/indra/llrender/llimagegl.h b/indra/llrender/llimagegl.h
index 5c7a5ce821..3892e9c014 100644
--- a/indra/llrender/llimagegl.h
+++ b/indra/llrender/llimagegl.h
@@ -39,6 +39,7 @@
 #include "llrender.h"
 #include "threadpool.h"
 #include "workqueue.h"
+#include <unordered_set>
 
 #define LL_IMAGEGL_THREAD_CHECK 0 //set to 1 to enable thread debugging for ImageGL
 
@@ -83,9 +84,8 @@ public:
     // needs to be called every frame
     static void updateStats(F32 current_time);
 
-    // Save off / restore GL textures
-    static void destroyGL(bool save_state = true);
-    static void restoreGL();
+    // cleanup GL state
+    static void destroyGL();
     static void dirtyTexOptions();
 
     static bool checkSize(S32 width, S32 height);
@@ -148,6 +148,10 @@ public:
     S32  getDiscardLevel() const        { return mCurrentDiscardLevel; }
     S32  getMaxDiscardLevel() const     { return mMaxDiscardLevel; }
 
+    // override the current discard level
+    // should only be used for local textures where you know exactly what you're doing
+    void setDiscardLevel(S32 level) { mCurrentDiscardLevel = level; }
+
     S32  getCurrentWidth() const { return mWidth ;}
     S32  getCurrentHeight() const { return mHeight ;}
     S32  getWidth(S32 discard_level = -1) const;
@@ -194,26 +198,26 @@ public:
     void setFilteringOption(LLTexUnit::eTextureFilterOptions option);
     LLTexUnit::eTextureFilterOptions getFilteringOption(void) const { return mFilterOption; }
 
-    LLGLenum getTexTarget()const { return mTarget ;}
-    S8       getDiscardLevelInAtlas()const {return mDiscardLevelInAtlas;}
-    U32      getTexelsInAtlas()const { return mTexelsInAtlas ;}
-    U32      getTexelsInGLTexture()const {return mTexelsInGLTexture;}
-
+    LLGLenum getTexTarget()const { return mTarget; }
 
     void init(bool usemipmaps);
     virtual void cleanup(); // Clean up the LLImageGL so it can be reinitialized.  Be careful when using this in derived class destructors
 
     void setNeedsAlphaAndPickMask(bool need_mask);
 
-    bool preAddToAtlas(S32 discard_level, const LLImageRaw* raw_image);
-    void postAddToAtlas() ;
-
 #if LL_IMAGEGL_THREAD_CHECK
     // thread debugging
     std::thread::id mActiveThread;
     void checkActiveThread();
 #endif
 
+    // scale down to the desired discard level using GPU
+    // returns true if texture was scaled down
+    // desired discard will be clamped to max discard
+    // if desired discard is less than or equal to current discard, no scaling will occur
+    // only works for GL_TEXTURE_2D target
+    bool scaleDown(S32 desired_discard);
+
 public:
     // Various GL/Rendering options
     S64Bytes mTextureMemory;
@@ -240,15 +244,10 @@ private:
 
     bool     mGLTextureCreated ;
     LLGLuint mTexName;
-    //LLGLuint mNewTexName = 0; // tex name set by background thread to be applied in main thread
     U16      mWidth;
     U16      mHeight;
     S8       mCurrentDiscardLevel;
 
-    S8       mDiscardLevelInAtlas;
-    U32      mTexelsInAtlas ;
-    U32      mTexelsInGLTexture;
-
     bool mAllowCompression;
 
 protected:
@@ -275,7 +274,7 @@ protected:
 
     // STATICS
 public:
-    static std::set<LLImageGL*> sImageList;
+    static std::unordered_set<LLImageGL*> sImageList;
     static S32 sCount;
 
     static F32 sLastFrameTime;
@@ -301,6 +300,8 @@ public:
 private:
     static S32 sMaxCategories;
     static bool sSkipAnalyzeAlpha;
+    static U32 sScratchPBO;
+    static U32 sScratchPBOSize;
 
     //the flag to allow to call readBackRaw(...).
     //can be removed if we do not use that function at all.
diff --git a/indra/llrender/llrender.cpp b/indra/llrender/llrender.cpp
index cfefde3acc..a0209fab43 100644
--- a/indra/llrender/llrender.cpp
+++ b/indra/llrender/llrender.cpp
@@ -991,6 +991,8 @@ void LLRender::syncLightState()
 void LLRender::syncMatrices()
 {
     STOP_GLERROR;
+    LL_PROFILE_ZONE_SCOPED_CATEGORY_DISPLAY;
+
     static const U32 name[] =
     {
         LLShaderMgr::MODELVIEW_MATRIX,
diff --git a/indra/llrender/llrender.h b/indra/llrender/llrender.h
index ebdc9e751d..be9f3895e7 100644
--- a/indra/llrender/llrender.h
+++ b/indra/llrender/llrender.h
@@ -548,6 +548,5 @@ glh::matrix4f gl_perspective(GLfloat fovy, GLfloat aspect, GLfloat zNear, GLfloa
 glh::matrix4f gl_lookat(LLVector3 eye, LLVector3 center, LLVector3 up);
 
 #define LL_SHADER_LOADING_WARNS(...) LL_WARNS()
-#define LL_SHADER_UNIFORM_ERRS(...)  LL_ERRS("Shader")
 
 #endif
diff --git a/indra/llrender/llshadermgr.cpp b/indra/llrender/llshadermgr.cpp
index 574cf55a0e..a8e9f20b40 100644
--- a/indra/llrender/llshadermgr.cpp
+++ b/indra/llrender/llshadermgr.cpp
@@ -1185,6 +1185,8 @@ void LLShaderMgr::initAttribsAndUniforms()
     mReservedUniforms.push_back("normal_texcoord"); // (GLTF)
     mReservedUniforms.push_back("metallic_roughness_texcoord"); // (GLTF)
     mReservedUniforms.push_back("occlusion_texcoord"); // (GLTF)
+    mReservedUniforms.push_back("gltf_node_id"); // (GLTF)
+    mReservedUniforms.push_back("gltf_material_id"); // (GLTF)
 
     mReservedUniforms.push_back("terrain_texture_transforms"); // (GLTF)
 
diff --git a/indra/llrender/llshadermgr.h b/indra/llrender/llshadermgr.h
index c00aff3a9a..fe6137c448 100644
--- a/indra/llrender/llshadermgr.h
+++ b/indra/llrender/llshadermgr.h
@@ -63,6 +63,8 @@ public:
         NORMAL_TEXCOORD,                      //  "normal_texcoord" (GLTF)
         METALLIC_ROUGHNESS_TEXCOORD,          //  "metallic_roughness_texcoord" (GLTF)
         OCCLUSION_TEXCOORD,                   //  "occlusion_texcoord" (GLTF)
+        GLTF_NODE_ID,                         //  "gltf_node_id" (GLTF)
+        GLTF_MATERIAL_ID,                     //  "gltf_material_id" (GLTF)
 
         TERRAIN_TEXTURE_TRANSFORMS,           //  "terrain_texture_transforms" (GLTF)
 
diff --git a/indra/llrender/llvertexbuffer.cpp b/indra/llrender/llvertexbuffer.cpp
index f82ec30242..2eb7c21f77 100644
--- a/indra/llrender/llvertexbuffer.cpp
+++ b/indra/llrender/llvertexbuffer.cpp
@@ -806,6 +806,13 @@ void LLVertexBuffer::drawRange(U32 mode, U32 start, U32 end, U32 count, U32 indi
     STOP_GLERROR;
 }
 
+void LLVertexBuffer::drawRangeFast(U32 mode, U32 start, U32 end, U32 count, U32 indices_offset) const
+{
+    glDrawRangeElements(sGLMode[mode], start, end, count, mIndicesType,
+        (GLvoid*)(indices_offset * (size_t)mIndicesStride));
+}
+
+
 void LLVertexBuffer::draw(U32 mode, U32 count, U32 indices_offset) const
 {
     drawRange(mode, 0, mNumVerts-1, count, indices_offset);
@@ -1062,12 +1069,6 @@ bool LLVertexBuffer::updateNumVerts(U32 nverts)
 
     bool success = true;
 
-    if (nverts > 65536)
-    {
-        LL_WARNS() << "Vertex buffer overflow!" << LL_ENDL;
-        nverts = 65536;
-    }
-
     U32 needed_size = calcOffsets(mTypeMask, mOffsets, nverts);
 
     if (needed_size != mSize)
@@ -1210,7 +1211,7 @@ U8* LLVertexBuffer::mapIndexBuffer(U32 index, S32 count)
 //  end -- last byte to copy (NOT last byte + 1)
 //  data -- data to be flushed
 //  dst -- mMappedData or mMappedIndexData
-static void flush_vbo(GLenum target, U32 start, U32 end, void* data, U8* dst)
+void LLVertexBuffer::flush_vbo(GLenum target, U32 start, U32 end, void* data, U8* dst)
 {
 #if LL_DARWIN
     LL_PROFILE_ZONE_NAMED_CATEGORY_VERTEX("vb memcpy");
@@ -1218,6 +1219,8 @@ static void flush_vbo(GLenum target, U32 start, U32 end, void* data, U8* dst)
     // copy into mapped buffer
     memcpy(dst+start, data, end-start+1);
 #else
+    llassert(target == GL_ARRAY_BUFFER ? sGLRenderBuffer == mGLBuffer : sGLRenderIndices == mGLIndices);
+
     // skip mapped data and stream to GPU via glBufferSubData
     if (end != 0)
     {
@@ -1627,81 +1630,51 @@ void LLVertexBuffer::setupVertexBuffer()
 
 void LLVertexBuffer::setPositionData(const LLVector4a* data)
 {
-#if !LL_DARWIN
-    llassert(sGLRenderBuffer == mGLBuffer);
-#endif
     flush_vbo(GL_ARRAY_BUFFER, 0, sizeof(LLVector4a) * getNumVerts()-1, (U8*) data, mMappedData);
 }
 
 void LLVertexBuffer::setTexCoord0Data(const LLVector2* data)
 {
-#if !LL_DARWIN
-    llassert(sGLRenderBuffer == mGLBuffer);
-#endif
     flush_vbo(GL_ARRAY_BUFFER, mOffsets[TYPE_TEXCOORD0], mOffsets[TYPE_TEXCOORD0] + sTypeSize[TYPE_TEXCOORD0] * getNumVerts() - 1, (U8*)data, mMappedData);
 }
 
 void LLVertexBuffer::setTexCoord1Data(const LLVector2* data)
 {
-#if !LL_DARWIN
-    llassert(sGLRenderBuffer == mGLBuffer);
-#endif
     flush_vbo(GL_ARRAY_BUFFER, mOffsets[TYPE_TEXCOORD1], mOffsets[TYPE_TEXCOORD1] + sTypeSize[TYPE_TEXCOORD1] * getNumVerts() - 1, (U8*)data, mMappedData);
 }
 
 void LLVertexBuffer::setColorData(const LLColor4U* data)
 {
-#if !LL_DARWIN
-    llassert(sGLRenderBuffer == mGLBuffer);
-#endif
     flush_vbo(GL_ARRAY_BUFFER, mOffsets[TYPE_COLOR], mOffsets[TYPE_COLOR] + sTypeSize[TYPE_COLOR] * getNumVerts() - 1, (U8*) data, mMappedData);
 }
 
 void LLVertexBuffer::setNormalData(const LLVector4a* data)
 {
-#if !LL_DARWIN
-    llassert(sGLRenderBuffer == mGLBuffer);
-#endif
     flush_vbo(GL_ARRAY_BUFFER, mOffsets[TYPE_NORMAL], mOffsets[TYPE_NORMAL] + sTypeSize[TYPE_NORMAL] * getNumVerts() - 1, (U8*) data, mMappedData);
 }
 
 void LLVertexBuffer::setTangentData(const LLVector4a* data)
 {
-#if !LL_DARWIN
-    llassert(sGLRenderBuffer == mGLBuffer);
-#endif
     flush_vbo(GL_ARRAY_BUFFER, mOffsets[TYPE_TANGENT], mOffsets[TYPE_TANGENT] + sTypeSize[TYPE_TANGENT] * getNumVerts() - 1, (U8*) data, mMappedData);
 }
 
 void LLVertexBuffer::setWeight4Data(const LLVector4a* data)
 {
-#if !LL_DARWIN
-    llassert(sGLRenderBuffer == mGLBuffer);
-#endif
     flush_vbo(GL_ARRAY_BUFFER, mOffsets[TYPE_WEIGHT4], mOffsets[TYPE_WEIGHT4] + sTypeSize[TYPE_WEIGHT4] * getNumVerts() - 1, (U8*) data, mMappedData);
 }
 
 void LLVertexBuffer::setJointData(const U64* data)
 {
-#if !LL_DARWIN
-    llassert(sGLRenderBuffer == mGLBuffer);
-#endif
     flush_vbo(GL_ARRAY_BUFFER, mOffsets[TYPE_JOINT], mOffsets[TYPE_JOINT] + sTypeSize[TYPE_JOINT] * getNumVerts() - 1, (U8*) data, mMappedData);
 }
 
 void LLVertexBuffer::setIndexData(const U16* data)
 {
-#if !LL_DARWIN
-    llassert(sGLRenderIndices == mGLIndices);
-#endif
     flush_vbo(GL_ELEMENT_ARRAY_BUFFER, 0, sizeof(U16) * getNumIndices() - 1, (U8*) data, mMappedIndexData);
 }
 
 void LLVertexBuffer::setIndexData(const U32* data)
 {
-#if !LL_DARWIN
-    llassert(sGLRenderIndices == mGLIndices);
-#endif
     if (mIndicesType != GL_UNSIGNED_INT)
     { // HACK -- vertex buffers are initialized as 16-bit indices, but can be switched to 32-bit indices
         mIndicesType = GL_UNSIGNED_INT;
@@ -1711,3 +1684,62 @@ void LLVertexBuffer::setIndexData(const U32* data)
     flush_vbo(GL_ELEMENT_ARRAY_BUFFER, 0, sizeof(U32) * getNumIndices() - 1, (U8*)data, mMappedIndexData);
 }
 
+void LLVertexBuffer::setPositionData(const LLVector4a* data, U32 offset, U32 count)
+{
+    flush_vbo(GL_ARRAY_BUFFER, offset * sizeof(LLVector4a), (offset + count) * sizeof(LLVector4a) - 1, (U8*)data, mMappedData);
+}
+
+void LLVertexBuffer::setNormalData(const LLVector4a* data, U32 offset, U32 count)
+{
+    flush_vbo(GL_ARRAY_BUFFER, mOffsets[TYPE_NORMAL] + offset * sTypeSize[TYPE_NORMAL], mOffsets[TYPE_NORMAL] + (offset + count) * sTypeSize[TYPE_NORMAL] - 1, (U8*)data, mMappedData);
+}
+
+void LLVertexBuffer::setTexCoord0Data(const LLVector2* data, U32 offset, U32 count)
+{
+    flush_vbo(GL_ARRAY_BUFFER, mOffsets[TYPE_TEXCOORD0] + offset * sTypeSize[TYPE_TEXCOORD0], mOffsets[TYPE_TEXCOORD0] + (offset + count) * sTypeSize[TYPE_TEXCOORD0] - 1, (U8*)data, mMappedData);
+}
+
+void LLVertexBuffer::setTexCoord1Data(const LLVector2* data, U32 offset, U32 count)
+{
+    flush_vbo(GL_ARRAY_BUFFER, mOffsets[TYPE_TEXCOORD1] + offset * sTypeSize[TYPE_TEXCOORD1], mOffsets[TYPE_TEXCOORD1] + (offset + count) * sTypeSize[TYPE_TEXCOORD1] - 1, (U8*)data, mMappedData);
+}
+
+void LLVertexBuffer::setColorData(const LLColor4U* data, U32 offset, U32 count)
+{
+    flush_vbo(GL_ARRAY_BUFFER, mOffsets[TYPE_COLOR] + offset * sTypeSize[TYPE_COLOR], mOffsets[TYPE_COLOR] + (offset + count) * sTypeSize[TYPE_COLOR] - 1, (U8*)data, mMappedData);
+}
+
+void LLVertexBuffer::setTangentData(const LLVector4a* data, U32 offset, U32 count)
+{
+    flush_vbo(GL_ARRAY_BUFFER, mOffsets[TYPE_TANGENT] + offset * sTypeSize[TYPE_TANGENT], mOffsets[TYPE_TANGENT] + (offset + count) * sTypeSize[TYPE_TANGENT] - 1, (U8*)data, mMappedData);
+}
+
+void LLVertexBuffer::setWeight4Data(const LLVector4a* data, U32 offset, U32 count)
+{
+    flush_vbo(GL_ARRAY_BUFFER, mOffsets[TYPE_WEIGHT4] + offset * sTypeSize[TYPE_WEIGHT4], mOffsets[TYPE_WEIGHT4] + (offset + count) * sTypeSize[TYPE_WEIGHT4] - 1, (U8*)data, mMappedData);
+}
+
+void LLVertexBuffer::setJointData(const U64* data, U32 offset, U32 count)
+{
+    flush_vbo(GL_ARRAY_BUFFER, mOffsets[TYPE_JOINT] + offset * sTypeSize[TYPE_JOINT], mOffsets[TYPE_JOINT] + (offset + count) * sTypeSize[TYPE_JOINT] - 1, (U8*)data, mMappedData);
+}
+
+void LLVertexBuffer::setIndexData(const U16* data, U32 offset, U32 count)
+{
+    flush_vbo(GL_ELEMENT_ARRAY_BUFFER, offset * sizeof(U16), (offset + count) * sizeof(U16) - 1, (U8*)data, mMappedIndexData);
+}
+
+void LLVertexBuffer::setIndexData(const U32* data, U32 offset, U32 count)
+{
+    if (mIndicesType != GL_UNSIGNED_INT)
+    { // HACK -- vertex buffers are initialized as 16-bit indices, but can be switched to 32-bit indices
+        mIndicesType = GL_UNSIGNED_INT;
+        mIndicesStride = 4;
+        mNumIndices /= 2;
+    }
+    flush_vbo(GL_ELEMENT_ARRAY_BUFFER, offset * sizeof(U32), (offset + count) * sizeof(U32) - 1, (U8*)data, mMappedIndexData);
+}
+
+
+
+
diff --git a/indra/llrender/llvertexbuffer.h b/indra/llrender/llvertexbuffer.h
index 601096abf9..49500e28ce 100644
--- a/indra/llrender/llvertexbuffer.h
+++ b/indra/llrender/llvertexbuffer.h
@@ -202,6 +202,18 @@ public:
     void setIndexData(const U16* data);
     void setIndexData(const U32* data);
 
+    void setPositionData(const LLVector4a* data, U32 offset, U32 count);
+    void setNormalData(const LLVector4a* data, U32 offset, U32 count);
+    void setTangentData(const LLVector4a* data, U32 offset, U32 count);
+    void setWeight4Data(const LLVector4a* data, U32 offset, U32 count);
+    void setJointData(const U64* data, U32 offset, U32 count);
+    void setTexCoord0Data(const LLVector2* data, U32 offset, U32 count);
+    void setTexCoord1Data(const LLVector2* data, U32 offset, U32 count);
+    void setColorData(const LLColor4U* data, U32 offset, U32 count);
+    void setIndexData(const U16* data, U32 offset, U32 count);
+    void setIndexData(const U32* data, U32 offset, U32 count);
+
+
     U32 getNumVerts() const                 { return mNumVerts; }
     U32 getNumIndices() const               { return mNumIndices; }
 
@@ -219,6 +231,10 @@ public:
     void drawArrays(U32 mode, U32 offset, U32 count) const;
     void drawRange(U32 mode, U32 start, U32 end, U32 count, U32 indices_offset) const;
 
+    // draw without syncing matrices.  If you're positive there have been no matrix
+    // since the last call to syncMatrices, this is much faster than drawRange
+    void drawRangeFast(U32 mode, U32 start, U32 end, U32 count, U32 indices_offset) const;
+
     //for debugging, validate data in given range is valid
     bool validateRange(U32 start, U32 end, U32 count, U32 offset) const;
 
@@ -256,6 +272,8 @@ private:
     friend class LLNavShapeVBOManager;
     friend class LLNavMeshVBOManager;
 
+    void flush_vbo(GLenum target, U32 start, U32 end, void* data, U8* dst);
+
     LLVertexBuffer(U32 typemask, U32 usage)
         : LLVertexBuffer(typemask)
     {}