Merge branch 'DRTVWR-559' into marchcat/587-v-pbr-merge

# Conflicts:
#	indra/llcommon/CMakeLists.txt
#	indra/newview/llspatialpartition.cpp
#	indra/newview/llviewergenericmessage.cpp
#	indra/newview/llvoavatar.cpp

1 files changed, 198 insertions, 133 deletions

diff --git a/indra/llrender/llrender.cpp b/indra/llrender/llrender.cpp
index 72cca1f2a2..ee66122774 100644
--- a/indra/llrender/llrender.cpp
+++ b/indra/llrender/llrender.cpp
@@ -35,6 +35,7 @@
 #include "llrendertarget.h"
 #include "lltexture.h"
 #include "llshadermgr.h"
+#include "hbxxh.h"
 
 #if LL_WINDOWS
 extern void APIENTRY gl_debug_callback(GLenum source,
@@ -54,8 +55,14 @@ F32	gGLModelView[16];
 F32	gGLLastModelView[16];
 F32 gGLLastProjection[16];
 F32 gGLProjection[16];
+
+// transform from last frame's camera space to this frame's camera space (and inverse)
+F32 gGLDeltaModelView[16];
+F32 gGLInverseDeltaModelView[16];
+
 S32	gGLViewport[4];
 
+
 U32 LLRender::sUICalls = 0;
 U32 LLRender::sUIVerts = 0;
 U32 LLTexUnit::sWhiteTexture = 0;
@@ -63,14 +70,20 @@ bool LLRender::sGLCoreProfile = false;
 bool LLRender::sNsightDebugSupport = false;
 LLVector2 LLRender::sUIGLScaleFactor = LLVector2(1.f, 1.f);
 
-static const U32 LL_NUM_TEXTURE_LAYERS = 32; 
-static const U32 LL_NUM_LIGHT_UNITS = 8;
+struct LLVBCache
+{
+    LLPointer<LLVertexBuffer> vb;
+    std::chrono::steady_clock::time_point touched;
+};
+
+static std::unordered_map<U64, LLVBCache> sVBCache;
 
 static const GLenum sGLTextureType[] =
 {
 	GL_TEXTURE_2D,
-	GL_TEXTURE_RECTANGLE_ARB,
-	GL_TEXTURE_CUBE_MAP_ARB,
+	GL_TEXTURE_RECTANGLE,
+	GL_TEXTURE_CUBE_MAP,
+    GL_TEXTURE_CUBE_MAP_ARRAY,
 	GL_TEXTURE_2D_MULTISAMPLE,
     GL_TEXTURE_3D
 };
@@ -122,7 +135,7 @@ void LLTexUnit::refreshState(void)
 
 	gGL.flush();
 	
-	glActiveTextureARB(GL_TEXTURE0_ARB + mIndex);
+	glActiveTexture(GL_TEXTURE0 + mIndex);
 
 	if (mCurrTexType != TT_NONE)
 	{
@@ -143,7 +156,7 @@ void LLTexUnit::activate(void)
 	if ((S32)gGL.mCurrTextureUnitIndex != mIndex || gGL.mDirty)
 	{
 		gGL.flush();
-		glActiveTextureARB(GL_TEXTURE0_ARB + mIndex);
+		glActiveTexture(GL_TEXTURE0 + mIndex);
 		gGL.mCurrTextureUnitIndex = mIndex;
 	}
 }
@@ -154,13 +167,10 @@ void LLTexUnit::enable(eTextureType type)
 
 	if ( (mCurrTexType != type || gGL.mDirty) && (type != TT_NONE) )
 	{
-		stop_glerror();
 		activate();
-		stop_glerror();
 		if (mCurrTexType != TT_NONE && !gGL.mDirty)
 		{
 			disable(); // Force a disable of a previous texture type if it's enabled.
-			stop_glerror();
 		}
 		mCurrTexType = type;
 
@@ -174,11 +184,7 @@ void LLTexUnit::disable(void)
 
 	if (mCurrTexType != TT_NONE)
 	{
-		activate();
 		unbind(mCurrTexType);
-		gGL.flush();
-        setTextureColorSpace(TCS_LINEAR);
-		
 		mCurrTexType = TT_NONE;
 	}
 }
@@ -186,8 +192,8 @@ void LLTexUnit::disable(void)
 void LLTexUnit::bindFast(LLTexture* texture)
 {
     LLImageGL* gl_tex = texture->getGLTexture();
-
-    glActiveTextureARB(GL_TEXTURE0_ARB + mIndex);
+    texture->setActive();
+    glActiveTexture(GL_TEXTURE0 + mIndex);
     gGL.mCurrTextureUnitIndex = mIndex;
     mCurrTexture = gl_tex->getTexName();
     if (!mCurrTexture)
@@ -223,7 +229,7 @@ bool LLTexUnit::bind(LLTexture* texture, bool for_rendering, bool forceBind)
 					enable(gl_tex->getTarget());
 					mCurrTexture = gl_tex->getTexName();
 					glBindTexture(sGLTextureType[gl_tex->getTarget()], mCurrTexture);
-					if(gl_tex->updateBindStats(gl_tex->mTextureMemory))
+					if(gl_tex->updateBindStats())
 					{
 						texture->setActive() ;
 						texture->updateBindStatsForTester() ;
@@ -302,7 +308,7 @@ bool LLTexUnit::bind(LLImageGL* texture, bool for_rendering, bool forceBind, S32
 		mCurrTexture = texname;
 		glBindTexture(sGLTextureType[texture->getTarget()], mCurrTexture);
 		stop_glerror();
-		texture->updateBindStats(texture->mTextureMemory);		
+        texture->updateBindStats();
 		mHasMipMaps = texture->mHasMipMaps;
 		if (texture->mTexOptionsDirty)
 		{
@@ -334,14 +340,14 @@ bool LLTexUnit::bind(LLCubeMap* cubeMap)
 
 	if (mCurrTexture != cubeMap->mImages[0]->getTexName())
 	{
-		if (gGLManager.mHasCubeMap && LLCubeMap::sUseCubeMaps)
+		if (LLCubeMap::sUseCubeMaps)
 		{
 			activate();
 			enable(LLTexUnit::TT_CUBE_MAP);
             mCurrTexture = cubeMap->mImages[0]->getTexName();
-			glBindTexture(GL_TEXTURE_CUBE_MAP_ARB, mCurrTexture);
+			glBindTexture(GL_TEXTURE_CUBE_MAP, mCurrTexture);
 			mHasMipMaps = cubeMap->mImages[0]->mHasMipMaps;
-			cubeMap->mImages[0]->updateBindStats(cubeMap->mImages[0]->mTextureMemory);
+			cubeMap->mImages[0]->updateBindStats();
 			if (cubeMap->mImages[0]->mTexOptionsDirty)
 			{
 				cubeMap->mImages[0]->mTexOptionsDirty = false;
@@ -369,10 +375,7 @@ bool LLTexUnit::bind(LLRenderTarget* renderTarget, bool bindDepth)
 
 	if (bindDepth)
 	{
-		if (renderTarget->hasStencil())
-		{
-			LL_ERRS() << "Cannot bind a render buffer for sampling.  Allocate render target without a stencil buffer if sampling of depth buffer is required." << LL_ENDL;
-		}
+        llassert(renderTarget->getDepth()); // target MUST have a depth buffer attachment
 
 		bindManual(renderTarget->getUsage(), renderTarget->getDepth());
 	}
@@ -469,7 +472,7 @@ void LLTexUnit::setTextureAddressMode(eTextureAddressMode mode)
 	glTexParameteri (sGLTextureType[mCurrTexType], GL_TEXTURE_WRAP_T, sGLAddressMode[mode]);
 	if (mCurrTexType == TT_CUBE_MAP)
 	{
-		glTexParameteri (GL_TEXTURE_CUBE_MAP_ARB, GL_TEXTURE_WRAP_R, sGLAddressMode[mode]);
+		glTexParameteri (GL_TEXTURE_CUBE_MAP, GL_TEXTURE_WRAP_R, sGLAddressMode[mode]);
 	}
 }
 
@@ -515,22 +518,15 @@ void LLTexUnit::setTextureFilteringOption(LLTexUnit::eTextureFilterOptions optio
 		}
 	}
 
-	if (gGLManager.mHasAnisotropic)
+	if (gGLManager.mGLVersion >= 4.59f)
 	{
 		if (LLImageGL::sGlobalUseAnisotropic && option == TFO_ANISOTROPIC)
 		{
-			if (gGL.mMaxAnisotropy < 1.f)
-			{
-				glGetFloatv(GL_MAX_TEXTURE_MAX_ANISOTROPY_EXT, &gGL.mMaxAnisotropy);
-
-				LL_INFOS() << "gGL.mMaxAnisotropy: " << gGL.mMaxAnisotropy << LL_ENDL ;
-				gGL.mMaxAnisotropy = llmax(1.f, gGL.mMaxAnisotropy) ;
-			}
-			glTexParameterf(sGLTextureType[mCurrTexType], GL_TEXTURE_MAX_ANISOTROPY_EXT, gGL.mMaxAnisotropy);
+			glTexParameterf(sGLTextureType[mCurrTexType], GL_TEXTURE_MAX_ANISOTROPY, gGLManager.mMaxAnisotropy);
 		}
 		else
 		{
-			glTexParameterf(sGLTextureType[mCurrTexType], GL_TEXTURE_MAX_ANISOTROPY_EXT, 1.f);
+			glTexParameterf(sGLTextureType[mCurrTexType], GL_TEXTURE_MAX_ANISOTROPY, 1.f);
 		}
 	}
 }
@@ -544,7 +540,7 @@ GLint LLTexUnit::getTextureSource(eTextureBlendSrc src)
 		case TBS_PREV_ALPHA:
 		case TBS_ONE_MINUS_PREV_COLOR:
 		case TBS_ONE_MINUS_PREV_ALPHA:
-			return GL_PREVIOUS_ARB;
+			return GL_PREVIOUS;
 
 		// All four cases should return the same value.
 		case TBS_TEX_COLOR:
@@ -558,18 +554,18 @@ GLint LLTexUnit::getTextureSource(eTextureBlendSrc src)
 		case TBS_VERT_ALPHA:
 		case TBS_ONE_MINUS_VERT_COLOR:
 		case TBS_ONE_MINUS_VERT_ALPHA:
-			return GL_PRIMARY_COLOR_ARB;
+			return GL_PRIMARY_COLOR;
 
 		// All four cases should return the same value.
 		case TBS_CONST_COLOR:
 		case TBS_CONST_ALPHA:
 		case TBS_ONE_MINUS_CONST_COLOR:
 		case TBS_ONE_MINUS_CONST_ALPHA:
-			return GL_CONSTANT_ARB;
+			return GL_CONSTANT;
 
 		default:
 			LL_WARNS() << "Unknown eTextureBlendSrc: " << src << ".  Using Vertex Color instead." << LL_ENDL;
-			return GL_PRIMARY_COLOR_ARB;
+			return GL_PRIMARY_COLOR;
 	}
 }
 
@@ -638,10 +634,10 @@ void LLTexUnit::debugTextureUnit(void)
 	if (mIndex < 0) return;
 
 	GLint activeTexture;
-	glGetIntegerv(GL_ACTIVE_TEXTURE_ARB, &activeTexture);
-	if ((GL_TEXTURE0_ARB + mIndex) != activeTexture)
+	glGetIntegerv(GL_ACTIVE_TEXTURE, &activeTexture);
+	if ((GL_TEXTURE0 + mIndex) != activeTexture)
 	{
-		U32 set_unit = (activeTexture - GL_TEXTURE0_ARB);
+		U32 set_unit = (activeTexture - GL_TEXTURE0);
 		LL_WARNS() << "Incorrect Texture Unit!  Expected: " << set_unit << " Actual: " << mIndex << LL_ENDL;
 	}
 }
@@ -649,29 +645,6 @@ void LLTexUnit::debugTextureUnit(void)
 void LLTexUnit::setTextureColorSpace(eTextureColorSpace space)
 {
     mTexColorSpace = space;
-
-#if USE_SRGB_DECODE
-    if (gGLManager.mHasTexturesRGBDecode)
-    {
-        if (space == TCS_SRGB)
-        {
-            glTexParameteri(sGLTextureType[mCurrTexType], GL_TEXTURE_SRGB_DECODE_EXT, GL_DECODE_EXT);
-        }
-        else
-        {
-            glTexParameteri(sGLTextureType[mCurrTexType], GL_TEXTURE_SRGB_DECODE_EXT, GL_SKIP_DECODE_EXT);
-        }
-
-        if (gDebugGL)
-        {
-            assert_glerror();
-        }
-    }
-    else
-    {
-        glTexParameteri(sGLTextureType[mCurrTexType], GL_TEXTURE_SRGB_DECODE_EXT, GL_SKIP_DECODE_EXT);
-    }
-#endif
 }
 
 LLLightState::LLLightState(S32 index)
@@ -734,6 +707,24 @@ void LLLightState::setSunPrimary(bool v)
     }
 }
 
+void LLLightState::setSize(F32 v)
+{
+    if (mSize != v)
+    {
+        ++gGL.mLightHash;
+        mSize = v;
+    }
+}
+
+void LLLightState::setFalloff(F32 v)
+{
+    if (mFalloff != v)
+    {
+        ++gGL.mLightHash;
+        mFalloff = v;
+    }
+}
+
 void LLLightState::setAmbient(const LLColor4& ambient)
 {
 	if (mAmbient != ambient)
@@ -827,19 +818,16 @@ LLRender::LLRender()
     mCount(0),
 	mQuadCycle(0),
     mMode(LLRender::TRIANGLES),
-    mCurrTextureUnitIndex(0),
-    mMaxAnisotropy(0.f) 
+    mCurrTextureUnitIndex(0)
 {	
-	mTexUnits.reserve(LL_NUM_TEXTURE_LAYERS);
 	for (U32 i = 0; i < LL_NUM_TEXTURE_LAYERS; i++)
 	{
-		mTexUnits.push_back(new LLTexUnit(i));
+        mTexUnits[i].mIndex = i;
 	}
-	mDummyTexUnit = new LLTexUnit(-1);
 
 	for (U32 i = 0; i < LL_NUM_LIGHT_UNITS; ++i)
 	{
-		mLightState.push_back(new LLLightState(i));
+        mLightState[i].mIndex = i;
 	}
 
 	for (U32 i = 0; i < 4; i++)
@@ -874,9 +862,9 @@ void LLRender::init(bool needs_vertex_buffer)
 #if LL_WINDOWS
     if (gGLManager.mHasDebugOutput && gDebugGL)
     { //setup debug output callback
-        //glDebugMessageControlARB(GL_DONT_CARE, GL_DONT_CARE, GL_DEBUG_SEVERITY_LOW_ARB, 0, NULL, GL_TRUE);
-        glDebugMessageCallbackARB((GLDEBUGPROCARB) gl_debug_callback, NULL);
-        glEnable(GL_DEBUG_OUTPUT_SYNCHRONOUS_ARB);
+        //glDebugMessageControl(GL_DONT_CARE, GL_DONT_CARE, GL_DEBUG_SEVERITY_LOW_ARB, 0, NULL, GL_TRUE);
+        glDebugMessageCallback((GLDEBUGPROC) gl_debug_callback, NULL);
+        glEnable(GL_DEBUG_OUTPUT_SYNCHRONOUS);
     }
 #endif
 
@@ -888,14 +876,14 @@ void LLRender::init(bool needs_vertex_buffer)
 
     glCullFace(GL_BACK);
 
-	if (sGLCoreProfile && !LLVertexBuffer::sUseVAO)
-	{ //bind a dummy vertex array object so we're core profile compliant
-#ifdef GL_ARB_vertex_array_object
-		U32 ret;
-		glGenVertexArrays(1, &ret);
-		glBindVertexArray(ret);
-#endif
-	}
+    // necessary for reflection maps
+    glEnable(GL_TEXTURE_CUBE_MAP_SEAMLESS);
+
+    { //bind a dummy vertex array object so we're core profile compliant
+        U32 ret;
+        glGenVertexArrays(1, &ret);
+        glBindVertexArray(ret);
+    }
 
     if (needs_vertex_buffer)
     {
@@ -907,8 +895,8 @@ void LLRender::initVertexBuffer()
 {
     llassert_always(mBuffer.isNull());
     stop_glerror();
-    mBuffer = new LLVertexBuffer(immediate_mask, 0);
-    mBuffer->allocateBuffer(4096, 0, TRUE);
+    mBuffer = new LLVertexBuffer(immediate_mask);
+    mBuffer->allocateBuffer(4096, 0);
     mBuffer->getVertexStrider(mVerticesp);
     mBuffer->getTexCoord0Strider(mTexcoordsp);
     mBuffer->getColorStrider(mColorsp);
@@ -922,19 +910,6 @@ void LLRender::resetVertexBuffer()
 
 void LLRender::shutdown()
 {
-	for (U32 i = 0; i < mTexUnits.size(); i++)
-	{
-		delete mTexUnits[i];
-	}
-	mTexUnits.clear();
-	delete mDummyTexUnit;
-	mDummyTexUnit = NULL;
-
-	for (U32 i = 0; i < mLightState.size(); ++i)
-	{
-		delete mLightState[i];
-	}
-	mLightState.clear();
     resetVertexBuffer();
 }
 
@@ -946,10 +921,10 @@ void LLRender::refreshState(void)
 
 	for (U32 i = 0; i < mTexUnits.size(); i++)
 	{
-		mTexUnits[i]->refreshState();
+		mTexUnits[i].refreshState();
 	}
 	
-	mTexUnits[active_unit]->activate();
+	mTexUnits[active_unit].activate();
 
 	setColorMask(mCurrColorMask[0], mCurrColorMask[1], mCurrColorMask[2], mCurrColorMask[3]);
 	
@@ -977,10 +952,11 @@ void LLRender::syncLightState()
         LLVector3 diffuse[LL_NUM_LIGHT_UNITS];
         LLVector3 diffuse_b[LL_NUM_LIGHT_UNITS];
         bool      sun_primary[LL_NUM_LIGHT_UNITS];
+        LLVector2 size[LL_NUM_LIGHT_UNITS];
 
         for (U32 i = 0; i < LL_NUM_LIGHT_UNITS; i++)
         {
-            LLLightState *light = mLightState[i];
+            LLLightState *light = &mLightState[i];
 
             position[i]  = light->mPosition;
             direction[i] = light->mSpotDirection;
@@ -988,17 +964,19 @@ void LLRender::syncLightState()
             diffuse[i].set(light->mDiffuse.mV);
             diffuse_b[i].set(light->mDiffuseB.mV);
             sun_primary[i] = light->mSunIsPrimary;
+            size[i].set(light->mSize, light->mFalloff);
         }
 
         shader->uniform4fv(LLShaderMgr::LIGHT_POSITION, LL_NUM_LIGHT_UNITS, position[0].mV);
         shader->uniform3fv(LLShaderMgr::LIGHT_DIRECTION, LL_NUM_LIGHT_UNITS, direction[0].mV);
         shader->uniform4fv(LLShaderMgr::LIGHT_ATTENUATION, LL_NUM_LIGHT_UNITS, attenuation[0].mV);
+        shader->uniform2fv(LLShaderMgr::LIGHT_DEFERRED_ATTENUATION, LL_NUM_LIGHT_UNITS, size[0].mV);
         shader->uniform3fv(LLShaderMgr::LIGHT_DIFFUSE, LL_NUM_LIGHT_UNITS, diffuse[0].mV);
-        shader->uniform4fv(LLShaderMgr::LIGHT_AMBIENT, 1, mAmbientLightColor.mV);
+        shader->uniform3fv(LLShaderMgr::LIGHT_AMBIENT, 1, mAmbientLightColor.mV);
         shader->uniform1i(LLShaderMgr::SUN_UP_FACTOR, sun_primary[0] ? 1 : 0);
-        shader->uniform4fv(LLShaderMgr::AMBIENT, 1, mAmbientLightColor.mV);
-        shader->uniform4fv(LLShaderMgr::SUNLIGHT_COLOR, 1, diffuse[0].mV);
-        shader->uniform4fv(LLShaderMgr::MOONLIGHT_COLOR, 1, diffuse_b[0].mV);
+        //shader->uniform3fv(LLShaderMgr::AMBIENT, 1, mAmbientLightColor.mV);
+        //shader->uniform3fv(LLShaderMgr::SUNLIGHT_COLOR, 1, diffuse[0].mV);
+        //shader->uniform3fv(LLShaderMgr::MOONLIGHT_COLOR, 1, diffuse_b[0].mV);
     }
 }
 
@@ -1095,13 +1073,20 @@ void LLRender::syncMatrices()
 		{ //update projection matrix, normal, and MVP
 			glh::matrix4f& mat = mMatrix[MM_PROJECTION][mMatIdx[MM_PROJECTION]];
 
-            // it would be nice to have this automatically track the state of the proj matrix
-            // but certain render paths (deferred lighting) require it to be mismatched *sigh*
-            //if (shader->getUniformLocation(LLShaderMgr::INVERSE_PROJECTION_MATRIX))
-            //{
-	        //    glh::matrix4f inv_proj = mat.inverse();
-	        //    shader->uniformMatrix4fv(LLShaderMgr::INVERSE_PROJECTION_MATRIX, 1, FALSE, inv_proj.m);
-            //}
+            // GZ: This was previously disabled seemingly due to a bug involving the deferred renderer's regular pushing and popping of mats.
+			// We're reenabling this and cleaning up the code around that - that would've been the appropriate course initially.
+			// Anything beyond the standard proj and inv proj mats are special cases.  Please setup special uniforms accordingly in the future.
+            if (shader->getUniformLocation(LLShaderMgr::INVERSE_PROJECTION_MATRIX))
+            {
+	            glh::matrix4f inv_proj = mat.inverse();
+	            shader->uniformMatrix4fv(LLShaderMgr::INVERSE_PROJECTION_MATRIX, 1, FALSE, inv_proj.m);
+            }
+
+			// Used by some full screen effects - such as full screen lights, glow, etc.
+            if (shader->getUniformLocation(LLShaderMgr::IDENTITY_MATRIX))
+            {
+                shader->uniformMatrix4fv(LLShaderMgr::IDENTITY_MATRIX, 1, GL_FALSE, glh::matrix4f::identity().m);
+            }
 
 			shader->uniformMatrix4fv(name[MM_PROJECTION], 1, GL_FALSE, mat.m);
 			shader->mMatHash[MM_PROJECTION] = mMatHash[MM_PROJECTION];
@@ -1479,12 +1464,7 @@ void LLRender::blendFunc(eBlendFactor color_sfactor, eBlendFactor color_dfactor,
 	llassert(color_dfactor < BF_UNDEF);
 	llassert(alpha_sfactor < BF_UNDEF);
 	llassert(alpha_dfactor < BF_UNDEF);
-	if (!gGLManager.mHasBlendFuncSeparate)
-	{
-		LL_WARNS_ONCE("render") << "no glBlendFuncSeparateEXT(), using color-only blend func" << LL_ENDL;
-		blendFunc(color_sfactor, color_dfactor);
-		return;
-	}
+	
 	if (mCurrBlendColorSFactor != color_sfactor || mCurrBlendColorDFactor != color_dfactor ||
 	    mCurrBlendAlphaSFactor != alpha_sfactor || mCurrBlendAlphaDFactor != alpha_dfactor)
 	{
@@ -1493,8 +1473,9 @@ void LLRender::blendFunc(eBlendFactor color_sfactor, eBlendFactor color_dfactor,
 		mCurrBlendColorDFactor = color_dfactor;
 		mCurrBlendAlphaDFactor = alpha_dfactor;
 		flush();
-		glBlendFuncSeparateEXT(sGLBlendFactor[color_sfactor], sGLBlendFactor[color_dfactor],
-				       sGLBlendFactor[alpha_sfactor], sGLBlendFactor[alpha_dfactor]);
+        
+        glBlendFuncSeparate(sGLBlendFactor[color_sfactor], sGLBlendFactor[color_dfactor],
+                           sGLBlendFactor[alpha_sfactor], sGLBlendFactor[alpha_dfactor]);
 	}
 }
 
@@ -1502,12 +1483,12 @@ LLTexUnit* LLRender::getTexUnit(U32 index)
 {
 	if (index < mTexUnits.size())
 	{
-		return mTexUnits[index];
+		return &mTexUnits[index];
 	}
 	else 
 	{
 		LL_DEBUGS() << "Non-existing texture unit layer requested: " << index << LL_ENDL;
-		return mDummyTexUnit;
+		return &mDummyTexUnit;
 	}
 }
 
@@ -1515,7 +1496,7 @@ LLLightState* LLRender::getLight(U32 index)
 {
 	if (index < mLightState.size())
 	{
-		return mLightState[index];
+		return &mLightState[index];
 	}
 	
 	return NULL;
@@ -1599,6 +1580,7 @@ void LLRender::flush()
 	if (mCount > 0)
 	{
         LL_PROFILE_ZONE_SCOPED_CATEGORY_PIPELINE;
+        llassert(LLGLSLShader::sCurBoundShaderPtr != nullptr);
 		if (!mUIOffset.empty())
 		{
 			sUICalls++;
@@ -1639,24 +1621,108 @@ void LLRender::flush()
 
         if (mBuffer)
         {
-            if (mBuffer->useVBOs() && !mBuffer->isLocked())
-            { //hack to only flush the part of the buffer that was updated (relies on stream draw using buffersubdata)
-                mBuffer->getVertexStrider(mVerticesp, 0, count);
-                mBuffer->getTexCoord0Strider(mTexcoordsp, 0, count);
-                mBuffer->getColorStrider(mColorsp, 0, count);
+
+            HBXXH64 hash;
+            U32 attribute_mask = LLGLSLShader::sCurBoundShaderPtr->mAttributeMask;
+
+            {
+                LL_PROFILE_ZONE_NAMED_CATEGORY_VERTEX("vb cache hash");
+
+                hash.update((U8*)mVerticesp.get(), count * sizeof(LLVector4a));
+                if (attribute_mask & LLVertexBuffer::MAP_TEXCOORD0)
+                {
+                    hash.update((U8*)mTexcoordsp.get(), count * sizeof(LLVector2));
+                }
+
+                if (attribute_mask & LLVertexBuffer::MAP_COLOR)
+                {
+                    hash.update((U8*)mColorsp.get(), count * sizeof(LLColor4U));
+                }
+
+                hash.finalize();
+            }
+            
+            
+            U64 vhash = hash.digest();
+
+            // check the VB cache before making a new vertex buffer
+            // This is a giant hack to deal with (mostly) our terrible UI rendering code
+            // that was built on top of OpenGL immediate mode.  Huge performance wins
+            // can be had by not uploading geometry to VRAM unless absolutely necessary.
+            // Most of our usage of the "immediate mode" style draw calls is actually
+            // sending the same geometry over and over again.
+            // To leverage this, we maintain a running hash of the vertex stream being
+            // built up before a flush, and then check that hash against a VB 
+            // cache just before creating a vertex buffer in VRAM
+            std::unordered_map<U64, LLVBCache>::iterator cache = sVBCache.find(vhash);
+
+            LLPointer<LLVertexBuffer> vb;
+
+            if (cache != sVBCache.end())
+            {
+                LL_PROFILE_ZONE_NAMED_CATEGORY_VERTEX("vb cache hit");
+                // cache hit, just use the cached buffer
+                vb = cache->second.vb;
+                cache->second.touched = std::chrono::steady_clock::now();
+            }
+            else
+            {
+                LL_PROFILE_ZONE_NAMED_CATEGORY_VERTEX("vb cache miss");
+                vb = new LLVertexBuffer(attribute_mask);
+                vb->allocateBuffer(count, 0);
+
+                vb->setBuffer();
+
+                vb->setPositionData((LLVector4a*) mVerticesp.get());
+
+                if (attribute_mask & LLVertexBuffer::MAP_TEXCOORD0)
+                {
+                    vb->setTexCoordData(mTexcoordsp.get());
+                }
+
+                if (attribute_mask & LLVertexBuffer::MAP_COLOR)
+                {
+                    vb->setColorData(mColorsp.get());
+                }
+
+                vb->unbind();
+
+                sVBCache[vhash] = { vb , std::chrono::steady_clock::now() };
+
+                static U32 miss_count = 0;
+                miss_count++;
+                if (miss_count > 1024)
+                {
+                    LL_PROFILE_ZONE_NAMED_CATEGORY_VERTEX("vb cache clean");
+                    miss_count = 0;
+                    auto now = std::chrono::steady_clock::now();
+
+                    using namespace std::chrono_literals;
+                    // every 1024 misses, clean the cache of any VBs that haven't been touched in the last second
+                    for (std::unordered_map<U64, LLVBCache>::iterator iter = sVBCache.begin(); iter != sVBCache.end(); )
+                    {
+                        if (now - iter->second.touched > 1s)
+                        {
+                            iter = sVBCache.erase(iter);
+                        }
+                        else
+                        {
+                            ++iter;
+                        }
+                    }
+                }
             }
 
-            mBuffer->flush();
-            mBuffer->setBuffer(immediate_mask);
+            vb->setBuffer();
 
             if (mMode == LLRender::QUADS && sGLCoreProfile)
             {
-                mBuffer->drawArrays(LLRender::TRIANGLES, 0, count);
+                vb->drawArrays(LLRender::TRIANGLES, 0, count);
                 mQuadCycle = 1;
             }
             else
             {
-                mBuffer->drawArrays(mMode, 0, count);
+                vb->drawArrays(mMode, 0, count);
             }
         }
         else
@@ -1948,8 +2014,7 @@ void LLRender::texCoord2fv(const GLfloat* tc)
 
 void LLRender::color4ub(const GLubyte& r, const GLubyte& g, const GLubyte& b, const GLubyte& a)
 {
-	if (!LLGLSLShader::sCurBoundShaderPtr ||
-		LLGLSLShader::sCurBoundShaderPtr->mAttributeMask & LLVertexBuffer::MAP_COLOR)
+	if (!LLGLSLShader::sCurBoundShaderPtr || LLGLSLShader::sCurBoundShaderPtr->mAttributeMask & LLVertexBuffer::MAP_COLOR)
 	{
 		mColorsp[mCount] = LLColor4U(r,g,b,a);
 	}