SL-16166 Optimization pass on LLRenderPass::pushBatch

14 files changed, 247 insertions, 54 deletions

diff --git a/indra/llrender/llgltexture.h b/indra/llrender/llgltexture.h
index 071912c2c2..028457c510 100644
--- a/indra/llrender/llgltexture.h
+++ b/indra/llrender/llgltexture.h
@@ -176,7 +176,7 @@ private:
 protected:
 	void setTexelsPerImage();
 
-	//note: do not make this function public.
+public:
 	/*virtual*/ LLImageGL* getGLTexture() const ;
 
 protected:
diff --git a/indra/llrender/llrender.cpp b/indra/llrender/llrender.cpp
index b6711e44e3..34c5b77944 100644
--- a/indra/llrender/llrender.cpp
+++ b/indra/llrender/llrender.cpp
@@ -169,6 +169,7 @@ void LLTexUnit::refreshState(void)
 
 void LLTexUnit::activate(void)
 {
+    LL_PROFILE_ZONE_SCOPED;
 	if (mIndex < 0) return;
 
 	if ((S32)gGL.mCurrTextureUnitIndex != mIndex || gGL.mDirty)
@@ -229,8 +230,20 @@ void LLTexUnit::disable(void)
 	}
 }
 
+void LLTexUnit::bindFast(LLTexture* texture)
+{
+    LLImageGL* gl_tex = texture->getGLTexture();
+
+    glActiveTextureARB(GL_TEXTURE0_ARB + mIndex);
+    gGL.mCurrTextureUnitIndex = mIndex;
+    mCurrTexture = gl_tex->getTexName();
+    glBindTexture(sGLTextureType[gl_tex->getTarget()], mCurrTexture);
+    mHasMipMaps = gl_tex->mHasMipMaps;
+}
+
 bool LLTexUnit::bind(LLTexture* texture, bool for_rendering, bool forceBind)
 {
+    LL_PROFILE_ZONE_SCOPED;
 	stop_glerror();
 	if (mIndex >= 0)
 	{
@@ -1243,8 +1256,6 @@ void LLRender::syncLightState()
 
 void LLRender::syncMatrices()
 {
-	stop_glerror();
-
 	static const U32 name[] = 
 	{
 		LLShaderMgr::MODELVIEW_MATRIX,
@@ -1415,8 +1426,6 @@ void LLRender::syncMatrices()
 			}
 		}
 	}
-
-	stop_glerror();
 }
 
 void LLRender::translatef(const GLfloat& x, const GLfloat& y, const GLfloat& z)
@@ -1925,6 +1934,7 @@ void LLRender::end()
 }
 void LLRender::flush()
 {
+    LL_PROFILE_ZONE_SCOPED;
 	if (mCount > 0)
 	{
 		if (!mUIOffset.empty())
diff --git a/indra/llrender/llrender.h b/indra/llrender/llrender.h
index c08c2d6881..658cdda504 100644
--- a/indra/llrender/llrender.h
+++ b/indra/llrender/llrender.h
@@ -161,6 +161,17 @@ public:
 	bool bind(LLImageGL* texture, bool for_rendering = false, bool forceBind = false);
     bool bind(LLTexture* texture, bool for_rendering = false, bool forceBind = false);
 
+    // bind implementation for inner loops
+    // makes the following assumptions:
+    //  - No need for gGL.flush() 
+    //  - texture is not null
+    //  - gl_tex->getTexName() is not zero
+    //  - This texture is not being bound redundantly
+    //  - USE_SRGB_DECODE is disabled
+    //  - mTexOptionsDirty is false
+    //  - 
+    void bindFast(LLTexture* texture);
+
 	// Binds a cubemap to this texture unit 
 	// (automatically enables the texture unit for cubemaps)
 	bool bind(LLCubeMap* cubeMap);
diff --git a/indra/llrender/lltexture.h b/indra/llrender/lltexture.h
index 41481fb8a7..256d85ce5a 100644
--- a/indra/llrender/lltexture.h
+++ b/indra/llrender/lltexture.h
@@ -67,11 +67,9 @@ public:
 	virtual S32	       getWidth(S32 discard_level = -1) const;
 	virtual S32	       getHeight(S32 discard_level = -1) const;
 	virtual bool       isActiveFetching();
+    virtual LLImageGL* getGLTexture() const;
 
 private:
-	//note: do not make this function public.
-	virtual LLImageGL* getGLTexture() const;
-
 	virtual void updateBindStatsForTester();
 };
 #endif
diff --git a/indra/llrender/llvertexbuffer.cpp b/indra/llrender/llvertexbuffer.cpp
index 0449ac392c..7b9388123f 100644
--- a/indra/llrender/llvertexbuffer.cpp
+++ b/indra/llrender/llvertexbuffer.cpp
@@ -787,6 +787,18 @@ void LLVertexBuffer::drawRange(U32 mode, U32 start, U32 end, U32 count, U32 indi
 	placeFence();
 }
 
+void LLVertexBuffer::drawRangeFast(U32 mode, U32 start, U32 end, U32 count, U32 indices_offset) const
+{
+    mMappable = false;
+    gGL.syncMatrices();
+
+    U16* idx = ((U16*)(U8*)mAlignedIndexOffset) + indices_offset;
+
+    LL_PROFILER_GPU_ZONEC("gl.DrawRangeElements", 0xFFFF00)
+        glDrawRangeElements(sGLMode[mode], start, end, count, GL_UNSIGNED_SHORT,
+            idx);
+}
+
 void LLVertexBuffer::draw(U32 mode, U32 count, U32 indices_offset) const
 {
 	llassert(!LLGLSLShader::sNoFixedFunction || LLGLSLShader::sCurBoundShaderPtr != NULL);
@@ -2272,6 +2284,21 @@ bool LLVertexBuffer::bindGLBuffer(bool force_bind)
 	return ret;
 }
 
+bool LLVertexBuffer::bindGLBufferFast()
+{
+    if (mGLBuffer != sGLRenderBuffer || !sVBOActive)
+    {
+        glBindBufferARB(GL_ARRAY_BUFFER_ARB, mGLBuffer);
+        sGLRenderBuffer = mGLBuffer;
+        sBindCount++;
+        sVBOActive = true;
+
+        return true;
+    }
+
+    return false;
+}
+
 static LLTrace::BlockTimerStatHandle FTM_BIND_GL_INDICES("Bind Indices");
 
 bool LLVertexBuffer::bindGLIndices(bool force_bind)
@@ -2297,6 +2324,21 @@ bool LLVertexBuffer::bindGLIndices(bool force_bind)
 	return ret;
 }
 
+bool LLVertexBuffer::bindGLIndicesFast()
+{
+    if (mGLIndices != sGLRenderIndices || !sIBOActive)
+    {
+        glBindBufferARB(GL_ELEMENT_ARRAY_BUFFER_ARB, mGLIndices);
+        sGLRenderIndices = mGLIndices;
+        sBindCount++;
+        sIBOActive = true;
+        
+        return true;
+    }
+
+    return false;
+}
+
 void LLVertexBuffer::flush()
 {
 	if (useVBOs())
@@ -2487,6 +2529,26 @@ void LLVertexBuffer::setBuffer(U32 data_mask)
 	}
 }
 
+void LLVertexBuffer::setBufferFast(U32 data_mask)
+{
+    //set up pointers if the data mask is different ...
+    bool setup = (sLastMask != data_mask);
+
+    
+    const bool bindBuffer = bindGLBufferFast();
+    const bool bindIndices = bindGLIndicesFast();
+
+    setup = setup || bindBuffer || bindIndices;
+
+    setupClientArrays(data_mask);
+  
+    if (data_mask && setup)
+    {
+        setupVertexBuffer(data_mask); // subclass specific setup (virtual function)
+        sSetCount++;
+    }
+}
+
 // virtual (default)
 void LLVertexBuffer::setupVertexBuffer(U32 data_mask)
 {
@@ -2644,6 +2706,99 @@ void LLVertexBuffer::setupVertexBuffer(U32 data_mask)
 	llglassertok();
 }
 
+void LLVertexBuffer::setupVertexBufferFast(U32 data_mask)
+{
+    U8* base = (U8*)mAlignedOffset;
+
+    if (data_mask & MAP_NORMAL)
+    {
+        S32 loc = TYPE_NORMAL;
+        void* ptr = (void*)(base + mOffsets[TYPE_NORMAL]);
+        glVertexAttribPointerARB(loc, 3, GL_FLOAT, GL_FALSE, LLVertexBuffer::sTypeSize[TYPE_NORMAL], ptr);
+    }
+    if (data_mask & MAP_TEXCOORD3)
+    {
+        S32 loc = TYPE_TEXCOORD3;
+        void* ptr = (void*)(base + mOffsets[TYPE_TEXCOORD3]);
+        glVertexAttribPointerARB(loc, 2, GL_FLOAT, GL_FALSE, LLVertexBuffer::sTypeSize[TYPE_TEXCOORD3], ptr);
+    }
+    if (data_mask & MAP_TEXCOORD2)
+    {
+        S32 loc = TYPE_TEXCOORD2;
+        void* ptr = (void*)(base + mOffsets[TYPE_TEXCOORD2]);
+        glVertexAttribPointerARB(loc, 2, GL_FLOAT, GL_FALSE, LLVertexBuffer::sTypeSize[TYPE_TEXCOORD2], ptr);
+    }
+    if (data_mask & MAP_TEXCOORD1)
+    {
+        S32 loc = TYPE_TEXCOORD1;
+        void* ptr = (void*)(base + mOffsets[TYPE_TEXCOORD1]);
+        glVertexAttribPointerARB(loc, 2, GL_FLOAT, GL_FALSE, LLVertexBuffer::sTypeSize[TYPE_TEXCOORD1], ptr);
+    }
+    if (data_mask & MAP_TANGENT)
+    {
+        S32 loc = TYPE_TANGENT;
+        void* ptr = (void*)(base + mOffsets[TYPE_TANGENT]);
+        glVertexAttribPointerARB(loc, 4, GL_FLOAT, GL_FALSE, LLVertexBuffer::sTypeSize[TYPE_TANGENT], ptr);
+    }
+    if (data_mask & MAP_TEXCOORD0)
+    {
+        S32 loc = TYPE_TEXCOORD0;
+        void* ptr = (void*)(base + mOffsets[TYPE_TEXCOORD0]);
+        glVertexAttribPointerARB(loc, 2, GL_FLOAT, GL_FALSE, LLVertexBuffer::sTypeSize[TYPE_TEXCOORD0], ptr);
+    }
+    if (data_mask & MAP_COLOR)
+    {
+        S32 loc = TYPE_COLOR;
+        //bind emissive instead of color pointer if emissive is present
+        void* ptr = (data_mask & MAP_EMISSIVE) ? (void*)(base + mOffsets[TYPE_EMISSIVE]) : (void*)(base + mOffsets[TYPE_COLOR]);
+        glVertexAttribPointerARB(loc, 4, GL_UNSIGNED_BYTE, GL_TRUE, LLVertexBuffer::sTypeSize[TYPE_COLOR], ptr);
+    }
+    if (data_mask & MAP_EMISSIVE)
+    {
+        S32 loc = TYPE_EMISSIVE;
+        void* ptr = (void*)(base + mOffsets[TYPE_EMISSIVE]);
+        glVertexAttribPointerARB(loc, 4, GL_UNSIGNED_BYTE, GL_TRUE, LLVertexBuffer::sTypeSize[TYPE_EMISSIVE], ptr);
+
+        if (!(data_mask & MAP_COLOR))
+        { //map emissive to color channel when color is not also being bound to avoid unnecessary shader swaps
+            loc = TYPE_COLOR;
+            glVertexAttribPointerARB(loc, 4, GL_UNSIGNED_BYTE, GL_TRUE, LLVertexBuffer::sTypeSize[TYPE_EMISSIVE], ptr);
+        }
+    }
+    if (data_mask & MAP_WEIGHT)
+    {
+        S32 loc = TYPE_WEIGHT;
+        void* ptr = (void*)(base + mOffsets[TYPE_WEIGHT]);
+        glVertexAttribPointerARB(loc, 1, GL_FLOAT, GL_FALSE, LLVertexBuffer::sTypeSize[TYPE_WEIGHT], ptr);
+    }
+    if (data_mask & MAP_WEIGHT4)
+    {
+        S32 loc = TYPE_WEIGHT4;
+        void* ptr = (void*)(base + mOffsets[TYPE_WEIGHT4]);
+        glVertexAttribPointerARB(loc, 4, GL_FLOAT, GL_FALSE, LLVertexBuffer::sTypeSize[TYPE_WEIGHT4], ptr);
+    }
+    if (data_mask & MAP_CLOTHWEIGHT)
+    {
+        S32 loc = TYPE_CLOTHWEIGHT;
+        void* ptr = (void*)(base + mOffsets[TYPE_CLOTHWEIGHT]);
+        glVertexAttribPointerARB(loc, 4, GL_FLOAT, GL_TRUE, LLVertexBuffer::sTypeSize[TYPE_CLOTHWEIGHT], ptr);
+    }
+    if (data_mask & MAP_TEXTURE_INDEX)
+    {
+#if !LL_DARWIN
+        S32 loc = TYPE_TEXTURE_INDEX;
+        void* ptr = (void*)(base + mOffsets[TYPE_VERTEX] + 12);
+        glVertexAttribIPointer(loc, 1, GL_UNSIGNED_INT, LLVertexBuffer::sTypeSize[TYPE_VERTEX], ptr);
+#endif
+    }
+    if (data_mask & MAP_VERTEX)
+    {
+        S32 loc = TYPE_VERTEX;
+        void* ptr = (void*)(base + mOffsets[TYPE_VERTEX]);
+        glVertexAttribPointerARB(loc, 3, GL_FLOAT, GL_FALSE, LLVertexBuffer::sTypeSize[TYPE_VERTEX], ptr);
+    }
+}
+
 LLVertexBuffer::MappedRegion::MappedRegion(S32 type, S32 index, S32 count)
 : mType(type), mIndex(index), mCount(count)
 { 
diff --git a/indra/llrender/llvertexbuffer.h b/indra/llrender/llvertexbuffer.h
index 1d60970df4..51ed85510e 100644
--- a/indra/llrender/llvertexbuffer.h
+++ b/indra/llrender/llvertexbuffer.h
@@ -210,13 +210,17 @@ protected:
 
 	virtual ~LLVertexBuffer(); // use unref()
 
-	virtual void setupVertexBuffer(U32 data_mask); // pure virtual, called from mapBuffer()
+	virtual void setupVertexBuffer(U32 data_mask);
+    void setupVertexBufferFast(U32 data_mask);
+
 	void setupVertexArray();
 	
 	void	genBuffer(U32 size);
 	void	genIndices(U32 size);
 	bool	bindGLBuffer(bool force_bind = false);
+    bool	bindGLBufferFast();
 	bool	bindGLIndices(bool force_bind = false);
+    bool    bindGLIndicesFast();
 	bool	bindGLArray();
 	void	releaseBuffer();
 	void	releaseIndices();
@@ -239,6 +243,8 @@ public:
 
 	// set for rendering
 	virtual void	setBuffer(U32 data_mask); 	// calls  setupVertexBuffer() if data_mask is not 0
+    void	setBufferFast(U32 data_mask); 	// calls setupVertexBufferFast(), assumes data_mask is not 0 among other assumptions
+
 	void flush(); //flush pending data to GL memory
 	// allocate buffer
 	bool	allocateBuffer(S32 nverts, S32 nindices, bool create);
@@ -290,6 +296,9 @@ public:
 	void drawArrays(U32 mode, U32 offset, U32 count) const;
 	void drawRange(U32 mode, U32 start, U32 end, U32 count, U32 indices_offset) const;
 
+    //implementation for inner loops that does no safety checking
+    void drawRangeFast(U32 mode, U32 start, U32 end, U32 count, U32 indices_offset) const;
+
 	//for debugging, validate data in given range is valid
 	void validateRange(U32 start, U32 end, U32 count, U32 offset) const;
 
diff --git a/indra/newview/lldrawpool.cpp b/indra/newview/lldrawpool.cpp
index d583a692f9..01dd307187 100644
--- a/indra/newview/lldrawpool.cpp
+++ b/indra/newview/lldrawpool.cpp
@@ -404,6 +404,7 @@ void LLRenderPass::renderTexture(U32 type, U32 mask, BOOL batch_textures)
 
 void LLRenderPass::pushBatches(U32 type, U32 mask, BOOL texture, BOOL batch_textures)
 {
+    LL_PROFILE_ZONE_SCOPED;
 	for (LLCullResult::drawinfo_iterator i = gPipeline.beginRenderMap(type); i != gPipeline.endRenderMap(type); ++i)	
 	{
 		LLDrawInfo* pparams = *i;
@@ -452,6 +453,7 @@ void LLRenderPass::applyModelMatrix(const LLDrawInfo& params)
 
 void LLRenderPass::pushBatch(LLDrawInfo& params, U32 mask, BOOL texture, BOOL batch_textures)
 {
+    LL_PROFILE_ZONE_SCOPED;
     if (!params.mCount)
     {
         return;
@@ -469,7 +471,7 @@ void LLRenderPass::pushBatch(LLDrawInfo& params, U32 mask, BOOL texture, BOOL ba
 			{
 				if (params.mTextureList[i].notNull())
 				{
-					gGL.getTexUnit(i)->bind(params.mTextureList[i], TRUE);
+					gGL.getTexUnit(i)->bindFast(params.mTextureList[i]);
 				}
 			}
 		}
@@ -477,8 +479,7 @@ void LLRenderPass::pushBatch(LLDrawInfo& params, U32 mask, BOOL texture, BOOL ba
 		{ //not batching textures or batch has only 1 texture -- might need a texture matrix
 			if (params.mTexture.notNull())
 			{
-				params.mTexture->addTextureStats(params.mVSize);
-				gGL.getTexUnit(0)->bind(params.mTexture, TRUE) ;
+				gGL.getTexUnit(0)->bindFast(params.mTexture);
 				if (params.mTextureMatrix)
 				{
 					tex_setup = true;
@@ -495,19 +496,17 @@ void LLRenderPass::pushBatch(LLDrawInfo& params, U32 mask, BOOL texture, BOOL ba
 		}
 	}
 	
-	if (params.mVertexBuffer.notNull())
-	{
-		if (params.mGroup)
-		{
-			params.mGroup->rebuildMesh();
-		}
+    {
+        if (params.mGroup)
+        {
+            params.mGroup->rebuildMesh();
+        }
 
-		LLGLEnableFunc stencil_test(GL_STENCIL_TEST, params.mSelected, &LLGLCommonFunc::selected_stencil_test);
-	
-		params.mVertexBuffer->setBuffer(mask);
-		params.mVertexBuffer->drawRange(params.mDrawMode, params.mStart, params.mEnd, params.mCount, params.mOffset);
-		gPipeline.addTrianglesDrawn(params.mCount, params.mDrawMode);
-	}
+        LLGLEnableFunc stencil_test(GL_STENCIL_TEST, params.mSelected, &LLGLCommonFunc::selected_stencil_test);
+
+        params.mVertexBuffer->setBufferFast(mask);
+        params.mVertexBuffer->drawRangeFast(params.mDrawMode, params.mStart, params.mEnd, params.mCount, params.mOffset);
+    }
 
 	if (tex_setup)
 	{
diff --git a/indra/newview/lldrawpoolmaterials.cpp b/indra/newview/lldrawpoolmaterials.cpp
index 05b0c1f1a9..d45e387e10 100644
--- a/indra/newview/lldrawpoolmaterials.cpp
+++ b/indra/newview/lldrawpoolmaterials.cpp
@@ -106,6 +106,7 @@ void LLDrawPoolMaterials::endDeferredPass(S32 pass)
 
 void LLDrawPoolMaterials::renderDeferred(S32 pass)
 {
+    LL_PROFILE_ZONE_SCOPED;
 	static const U32 type_list[] = 
 	{
 		LLRenderPass::PASS_MATERIAL,
diff --git a/indra/newview/lldrawpoolsimple.cpp b/indra/newview/lldrawpoolsimple.cpp
index 74e6665a96..ba8549904b 100644
--- a/indra/newview/lldrawpoolsimple.cpp
+++ b/indra/newview/lldrawpoolsimple.cpp
@@ -471,6 +471,7 @@ void LLDrawPoolSimple::endDeferredPass(S32 pass)
 
 void LLDrawPoolSimple::renderDeferred(S32 pass)
 {
+    LL_PROFILE_ZONE_SCOPED;
 	LLGLDisable blend(GL_BLEND);
 	LLGLDisable alpha_test(GL_ALPHA_TEST);
 
diff --git a/indra/newview/lldrawpooltree.cpp b/indra/newview/lldrawpooltree.cpp
index 0d5195bdbf..a1ff020068 100644
--- a/indra/newview/lldrawpooltree.cpp
+++ b/indra/newview/lldrawpooltree.cpp
@@ -153,6 +153,7 @@ void LLDrawPoolTree::beginDeferredPass(S32 pass)
 
 void LLDrawPoolTree::renderDeferred(S32 pass)
 {
+    LL_PROFILE_ZONE_SCOPED;
 	render(pass);
 }
 
diff --git a/indra/newview/llspatialpartition.h b/indra/newview/llspatialpartition.h
index 919f386d29..6ef82fac9c 100644
--- a/indra/newview/llspatialpartition.h
+++ b/indra/newview/llspatialpartition.h
@@ -41,6 +41,7 @@
 #include "llviewercamera.h"
 #include "llvector4a.h"
 #include <queue>
+#include <unordered_map>
 
 #define SG_STATE_INHERIT_MASK (OCCLUDED)
 #define SG_INITIAL_STATE_MASK (DIRTY | GEOM_DIRTY)
@@ -216,10 +217,10 @@ public:
 	typedef std::vector<LLPointer<LLSpatialGroup> > sg_vector_t;
 	typedef std::vector<LLPointer<LLSpatialBridge> > bridge_list_t;
 	typedef std::vector<LLPointer<LLDrawInfo> > drawmap_elem_t; 
-	typedef std::map<U32, drawmap_elem_t > draw_map_t;	
+	typedef std::unordered_map<U32, drawmap_elem_t > draw_map_t;	
 	typedef std::vector<LLPointer<LLVertexBuffer> > buffer_list_t;
-	typedef std::map<LLFace*, buffer_list_t> buffer_texture_map_t;
-	typedef std::map<U32, buffer_texture_map_t> buffer_map_t;
+	typedef std::unordered_map<LLFace*, buffer_list_t> buffer_texture_map_t;
+	typedef std::unordered_map<U32, buffer_texture_map_t> buffer_map_t;
 
 	struct CompareDistanceGreater
 	{
diff --git a/indra/newview/llviewertexture.cpp b/indra/newview/llviewertexture.cpp
index 274f53a160..949e71a4c9 100644
--- a/indra/newview/llviewertexture.cpp
+++ b/indra/newview/llviewertexture.cpp
@@ -845,14 +845,14 @@ void LLViewerTexture::addTextureStats(F32 virtual_size, BOOL needs_gltexture) co
 	{
 		//flag to reset the values because the old values are used.
 		resetMaxVirtualSizeResetCounter();
-		mMaxVirtualSize = virtual_size;		
-		mAdditionalDecodePriority = 0.f;	
+		mMaxVirtualSize = virtual_size;
+		mAdditionalDecodePriority = 0.f;
 		mNeedsGLTexture = needs_gltexture;
 	}
 	else if (virtual_size > mMaxVirtualSize)
 	{
 		mMaxVirtualSize = virtual_size;
-	}	
+	}
 }
 
 void LLViewerTexture::resetTextureStats()
diff --git a/indra/newview/pipeline.cpp b/indra/newview/pipeline.cpp
index c0b469af81..5ef3819de4 100644
--- a/indra/newview/pipeline.cpp
+++ b/indra/newview/pipeline.cpp
@@ -358,7 +358,6 @@ bool	LLPipeline::sRenderAttachedLights = true;
 bool	LLPipeline::sRenderAttachedParticles = true;
 bool	LLPipeline::sRenderDeferred = false;
 S32		LLPipeline::sVisibleLightCount = 0;
-F32		LLPipeline::sMinRenderSize = 0.f;
 bool	LLPipeline::sRenderingHUDs;
 F32     LLPipeline::sDistortionWaterClipPlaneMargin = 1.0125f;
 
@@ -2550,13 +2549,6 @@ void LLPipeline::markNotCulled(LLSpatialGroup* group, LLCamera& camera)
 		return;
 	}
 
-	const LLVector4a* bounds = group->getBounds();
-	if (sMinRenderSize > 0.f && 
-			llmax(llmax(bounds[1][0], bounds[1][1]), bounds[1][2]) < sMinRenderSize)
-	{
-		return;
-	}
-
 	assertInitialized();
 	
 	if (!group->getSpatialPartition()->mRenderByGroup)
@@ -3480,7 +3472,6 @@ void LLPipeline::stateSort(LLSpatialGroup* group, LLCamera& camera)
 			group->mLastUpdateDistance = group->mDistance;
 		}
 	}
-
 }
 
 void LLPipeline::stateSort(LLSpatialBridge* bridge, LLCamera& camera, BOOL fov_changed)
@@ -3787,6 +3778,27 @@ void renderSoundHighlights(LLDrawable* drawablep)
 }
 }
 
+void LLPipeline::touchTextures(LLDrawInfo* info)
+{
+    LL_PROFILE_ZONE_SCOPED;
+    for (auto& tex : info->mTextureList)
+    {
+        if (tex.notNull())
+        {
+            LLImageGL* gl_tex = tex->getGLTexture();
+            if (gl_tex && gl_tex->updateBindStats(gl_tex->mTextureMemory))
+            {
+                tex->setActive();
+            }
+        }
+    }
+
+    if (info->mTexture.notNull())
+    {
+        info->mTexture->addTextureStats(info->mVSize);
+    }
+}
+
 void LLPipeline::postSort(LLCamera& camera)
 {
 	LL_RECORD_BLOCK_TIME(FTM_STATESORT_POSTSORT);
@@ -3839,20 +3851,14 @@ void LLPipeline::postSort(LLCamera& camera)
 			
 			for (LLSpatialGroup::drawmap_elem_t::iterator k = src_vec.begin(); k != src_vec.end(); ++k)
 			{
-				if (sMinRenderSize > 0.f)
-				{
-					LLVector4a bounds;
-					bounds.setSub((*k)->mExtents[1],(*k)->mExtents[0]);
-
-					if (llmax(llmax(bounds[0], bounds[1]), bounds[2]) > sMinRenderSize)
-					{
-						sCull->pushDrawInfo(j->first, *k);
-					}
-				}
-				else
-				{
-					sCull->pushDrawInfo(j->first, *k);
-				}
+                LLDrawInfo* info = *k;
+				
+				sCull->pushDrawInfo(j->first, info);
+                if (!sShadowRender && !sReflectionRender)
+                {
+                    touchTextures(info);
+                    addTrianglesDrawn(info->mCount, info->mDrawMode);
+                }
 			}
 		}
 
diff --git a/indra/newview/pipeline.h b/indra/newview/pipeline.h
index 5605d26410..8ffbddca21 100644
--- a/indra/newview/pipeline.h
+++ b/indra/newview/pipeline.h
@@ -265,6 +265,8 @@ public:
 	void stateSort(LLSpatialBridge* bridge, LLCamera& camera, BOOL fov_changed = FALSE);
 	void stateSort(LLDrawable* drawablep, LLCamera& camera);
 	void postSort(LLCamera& camera);
+    //update stats for textures in given DrawInfo
+    void touchTextures(LLDrawInfo* info);
 	void forAllVisibleDrawables(void (*func)(LLDrawable*));
 
 	void renderObjects(U32 type, U32 mask, bool texture = true, bool batch_texture = false);
@@ -596,7 +598,6 @@ public:
 	static bool				sRenderAttachedParticles;
 	static bool				sRenderDeferred;
 	static S32				sVisibleLightCount;
-	static F32				sMinRenderSize;
 	static bool				sRenderingHUDs;
     static F32              sDistortionWaterClipPlaneMargin;