SL-16815 Remove frame stalls from occlusion queries, bumpmap updates, and querying for available video memory.

14 files changed, 316 insertions, 314 deletions

diff --git a/indra/llrender/llgl.cpp b/indra/llrender/llgl.cpp
index 50a5972d28..639d1fba32 100644
--- a/indra/llrender/llgl.cpp
+++ b/indra/llrender/llgl.cpp
@@ -2150,95 +2150,6 @@ LLGLUserClipPlane::~LLGLUserClipPlane()
 	disable();
 }
 
-LLGLNamePool::LLGLNamePool()
-{
-}
-
-LLGLNamePool::~LLGLNamePool()
-{
-}
-
-void LLGLNamePool::upkeep()
-{
-	std::sort(mNameList.begin(), mNameList.end(), CompareUsed());
-}
-
-void LLGLNamePool::cleanup()
-{
-	for (name_list_t::iterator iter = mNameList.begin(); iter != mNameList.end(); ++iter)
-	{
-		releaseName(iter->name);
-	}
-
-	mNameList.clear();
-}
-
-GLuint LLGLNamePool::allocate()
-{
-    LL_PROFILE_ZONE_SCOPED_CATEGORY_PIPELINE;
-#if LL_GL_NAME_POOLING
-	for (name_list_t::iterator iter = mNameList.begin(); iter != mNameList.end(); ++iter)
-	{
-		if (!iter->used)
-		{
-			iter->used = TRUE;
-			return iter->name;
-		}
-	}
-
-	NameEntry entry;
-	entry.name = allocateName();
-	entry.used = TRUE;
-	mNameList.push_back(entry);
-
-	return entry.name;
-#else
-	return allocateName();
-#endif
-}
-
-void LLGLNamePool::release(GLuint name)
-{
-#if LL_GL_NAME_POOLING
-	for (name_list_t::iterator iter = mNameList.begin(); iter != mNameList.end(); ++iter)
-	{
-		if (iter->name == name)
-		{
-			if (iter->used)
-			{
-				iter->used = FALSE;
-				return;
-			}
-			else
-			{
-				LL_ERRS() << "Attempted to release a pooled name that is not in use!" << LL_ENDL;
-			}
-		}
-	}
-	LL_ERRS() << "Attempted to release a non pooled name!" << LL_ENDL;
-#else
-	releaseName(name);
-#endif
-}
-
-//static
-void LLGLNamePool::upkeepPools()
-{
-	for (auto& pool : instance_snapshot())
-	{
-		pool.upkeep();
-	}
-}
-
-//static
-void LLGLNamePool::cleanupPools()
-{
-	for (auto& pool : instance_snapshot())
-	{
-		pool.cleanup();
-	}
-}
-
 LLGLDepthTest::LLGLDepthTest(GLboolean depth_enabled, GLboolean write_enabled, GLenum depth_func)
 : mPrevDepthEnabled(sDepthEnabled), mPrevDepthFunc(sDepthFunc), mPrevWriteEnabled(sWriteEnabled)
 {
@@ -2460,3 +2371,4 @@ extern "C"
 }
 #endif
 
+
diff --git a/indra/llrender/llgl.h b/indra/llrender/llgl.h
index ec97eb0faa..52338364e6 100644
--- a/indra/llrender/llgl.h
+++ b/indra/llrender/llgl.h
@@ -367,51 +367,6 @@ public:
 };
 
 /*
-	Generic pooling scheme for things which use GL names (used for occlusion queries and vertex buffer objects).
-	Prevents thrashing of GL name caches by avoiding calls to glGenFoo and glDeleteFoo.
-*/
-class LLGLNamePool : public LLInstanceTracker<LLGLNamePool>
-{
-public:
-	typedef LLInstanceTracker<LLGLNamePool> tracker_t;
-
-	struct NameEntry
-	{
-		GLuint name;
-		BOOL used;
-	};
-
-	struct CompareUsed
-	{
-		bool operator()(const NameEntry& lhs, const NameEntry& rhs)
-		{
-			return lhs.used < rhs.used;  //FALSE entries first
-		}
-	};
-
-	typedef std::vector<NameEntry> name_list_t;
-	name_list_t mNameList;
-
-	LLGLNamePool();
-	virtual ~LLGLNamePool();
-	
-	void upkeep();
-	void cleanup();
-	
-	GLuint allocate();
-	void release(GLuint name);
-	
-	static void upkeepPools();
-	static void cleanupPools();
-
-protected:
-	typedef std::vector<LLGLNamePool*> pool_list_t;
-	
-	virtual GLuint allocateName() = 0;
-	virtual void releaseName(GLuint name) = 0;
-};
-
-/*
 	Interface for objects that need periodic GL updates applied to them.
 	Used to synchronize GL updates with GL thread.
 */
diff --git a/indra/llrender/llgltexture.cpp b/indra/llrender/llgltexture.cpp
index c7de8bc338..e012eb9a62 100644
--- a/indra/llrender/llgltexture.cpp
+++ b/indra/llrender/llgltexture.cpp
@@ -164,11 +164,11 @@ BOOL LLGLTexture::createGLTexture()
 	return mGLTexturep->createGLTexture() ;
 }
 
-BOOL LLGLTexture::createGLTexture(S32 discard_level, const LLImageRaw* imageraw, S32 usename, BOOL to_create, S32 category, bool defer_copy)
+BOOL LLGLTexture::createGLTexture(S32 discard_level, const LLImageRaw* imageraw, S32 usename, BOOL to_create, S32 category, bool defer_copy, LLGLuint* tex_name)
 {
-	llassert(mGLTexturep.notNull()) ;	
+	llassert(mGLTexturep.notNull());
 
-	BOOL ret = mGLTexturep->createGLTexture(discard_level, imageraw, usename, to_create, category, defer_copy) ;
+	BOOL ret = mGLTexturep->createGLTexture(discard_level, imageraw, usename, to_create, category, defer_copy, tex_name) ;
 
 	if(ret)
 	{
@@ -260,20 +260,20 @@ LLTexUnit::eTextureType LLGLTexture::getTarget(void) const
 	return mGLTexturep->getTarget() ;
 }
 
-BOOL LLGLTexture::setSubImage(const LLImageRaw* imageraw, S32 x_pos, S32 y_pos, S32 width, S32 height, bool use_new_name /* = false */)
+BOOL LLGLTexture::setSubImage(const LLImageRaw* imageraw, S32 x_pos, S32 y_pos, S32 width, S32 height, LLGLuint use_name)
 {
     LL_PROFILE_ZONE_SCOPED_CATEGORY_TEXTURE;
 	llassert(mGLTexturep.notNull()) ;
 
-	return mGLTexturep->setSubImage(imageraw, x_pos, y_pos, width, height, 0, use_new_name) ;
+	return mGLTexturep->setSubImage(imageraw, x_pos, y_pos, width, height, 0, use_name) ;
 }
 
-BOOL LLGLTexture::setSubImage(const U8* datap, S32 data_width, S32 data_height, S32 x_pos, S32 y_pos, S32 width, S32 height, bool use_new_name /* = false */)
+BOOL LLGLTexture::setSubImage(const U8* datap, S32 data_width, S32 data_height, S32 x_pos, S32 y_pos, S32 width, S32 height, LLGLuint use_name)
 {
     LL_PROFILE_ZONE_SCOPED_CATEGORY_TEXTURE;
 	llassert(mGLTexturep.notNull()) ;
 
-	return mGLTexturep->setSubImage(datap, data_width, data_height, x_pos, y_pos, width, height, 0, use_new_name) ;
+	return mGLTexturep->setSubImage(datap, data_width, data_height, x_pos, y_pos, width, height, 0, use_name) ;
 }
 
 void LLGLTexture::setGLTextureCreated (bool initialized)
diff --git a/indra/llrender/llgltexture.h b/indra/llrender/llgltexture.h
index 216e7f640d..8cfe7b62de 100644
--- a/indra/llrender/llgltexture.h
+++ b/indra/llrender/llgltexture.h
@@ -132,13 +132,14 @@ public:
     // to_create - set to FALSE to force gl texture to not be created
     // category - LLGLTexture category for this LLGLTexture
     // defer_copy - set to true to allocate GL texture but NOT initialize with imageraw data
-    BOOL       createGLTexture(S32 discard_level, const LLImageRaw* imageraw, S32 usename = 0, BOOL to_create = TRUE, S32 category = LLGLTexture::OTHER, bool defer_copy = false);
+    // tex_name - if not null, will be set to the GL name of the texture created
+    BOOL       createGLTexture(S32 discard_level, const LLImageRaw* imageraw, S32 usename = 0, BOOL to_create = TRUE, S32 category = LLGLTexture::OTHER, bool defer_copy = false, LLGLuint* tex_name = nullptr);
 
 	void       setFilteringOption(LLTexUnit::eTextureFilterOptions option);
 	void       setExplicitFormat(LLGLint internal_format, LLGLenum primary_format, LLGLenum type_format = 0, BOOL swap_bytes = FALSE);
 	void       setAddressMode(LLTexUnit::eTextureAddressMode mode);
-	BOOL       setSubImage(const LLImageRaw* imageraw, S32 x_pos, S32 y_pos, S32 width, S32 height, bool use_new_name = false);
-	BOOL       setSubImage(const U8* datap, S32 data_width, S32 data_height, S32 x_pos, S32 y_pos, S32 width, S32 height, bool use_new_name = false);
+	BOOL       setSubImage(const LLImageRaw* imageraw, S32 x_pos, S32 y_pos, S32 width, S32 height, LLGLuint use_name = 0);
+	BOOL       setSubImage(const U8* datap, S32 data_width, S32 data_height, S32 x_pos, S32 y_pos, S32 width, S32 height, LLGLuint use_name = 0);
 	void       setGLTextureCreated (bool initialized);
 	void       setCategory(S32 category) ;
     void       setTexName(LLGLuint); // for forcing w/ externally created textures only
diff --git a/indra/llrender/llimagegl.cpp b/indra/llrender/llimagegl.cpp
index 694c4a7b06..110074bdc0 100644
--- a/indra/llrender/llimagegl.cpp
+++ b/indra/llrender/llimagegl.cpp
@@ -40,6 +40,7 @@
 #include "llglslshader.h"
 #include "llrender.h"
 #include "llwindow.h"
+#include "llframetimer.h"
 
 #if !LL_IMAGEGL_THREAD_CHECK
 #define checkActiveThread()
@@ -1075,14 +1076,15 @@ void LLImageGL::postAddToAtlas()
 	stop_glerror();	
 }
 
-BOOL LLImageGL::setSubImage(const U8* datap, S32 data_width, S32 data_height, S32 x_pos, S32 y_pos, S32 width, S32 height, BOOL force_fast_update /* = FALSE */, bool use_new_name /* = false */)
+BOOL LLImageGL::setSubImage(const U8* datap, S32 data_width, S32 data_height, S32 x_pos, S32 y_pos, S32 width, S32 height, BOOL force_fast_update /* = FALSE */, LLGLuint use_name)
 {
     LL_PROFILE_ZONE_SCOPED_CATEGORY_TEXTURE;
 	if (!width || !height)
 	{
 		return TRUE;
 	}
-	if (0 == (use_new_name ? mNewTexName : mTexName))
+    LLGLuint tex_name = use_name != 0 ? use_name : mTexName;
+	if (0 == tex_name)
 	{
 		// *TODO: Re-enable warning?  Ran into thread locking issues? DK 2011-02-18
 		//LL_WARNS() << "Setting subimage on image without GL texture" << LL_ENDL;
@@ -1098,7 +1100,7 @@ BOOL LLImageGL::setSubImage(const U8* datap, S32 data_width, S32 data_height, S3
 	// HACK: allow the caller to explicitly force the fast path (i.e. using glTexSubImage2D here instead of calling setImage) even when updating the full texture.
 	if (!force_fast_update && x_pos == 0 && y_pos == 0 && width == getWidth() && height == getHeight() && data_width == width && data_height == height)
 	{
-		setImage(datap, FALSE, use_new_name ? mNewTexName : mTexName);
+		setImage(datap, FALSE, tex_name);
 	}
 	else
 	{
@@ -1150,7 +1152,7 @@ BOOL LLImageGL::setSubImage(const U8* datap, S32 data_width, S32 data_height, S3
 
 		datap += (y_pos * data_width + x_pos) * getComponents();
 		// Update the GL texture
-		BOOL res = gGL.getTexUnit(0)->bindManual(mBindTarget, use_new_name ? mNewTexName : mTexName);
+		BOOL res = gGL.getTexUnit(0)->bindManual(mBindTarget, tex_name);
 		if (!res) LL_ERRS() << "LLImageGL::setSubImage(): bindTexture failed" << LL_ENDL;
 		stop_glerror();
 
@@ -1171,10 +1173,10 @@ BOOL LLImageGL::setSubImage(const U8* datap, S32 data_width, S32 data_height, S3
 	return TRUE;
 }
 
-BOOL LLImageGL::setSubImage(const LLImageRaw* imageraw, S32 x_pos, S32 y_pos, S32 width, S32 height, BOOL force_fast_update /* = FALSE */, bool use_new_name /* = false */)
+BOOL LLImageGL::setSubImage(const LLImageRaw* imageraw, S32 x_pos, S32 y_pos, S32 width, S32 height, BOOL force_fast_update /* = FALSE */, LLGLuint use_name)
 {
     LL_PROFILE_ZONE_SCOPED_CATEGORY_TEXTURE;
-	return setSubImage(imageraw->getData(), imageraw->getWidth(), imageraw->getHeight(), x_pos, y_pos, width, height, force_fast_update, use_new_name);
+	return setSubImage(imageraw->getData(), imageraw->getWidth(), imageraw->getHeight(), x_pos, y_pos, width, height, force_fast_update, use_name);
 }
 
 // Copy sub image from frame buffer
@@ -1361,7 +1363,7 @@ BOOL LLImageGL::createGLTexture()
 	return TRUE ;
 }
 
-BOOL LLImageGL::createGLTexture(S32 discard_level, const LLImageRaw* imageraw, S32 usename/*=0*/, BOOL to_create, S32 category, bool defer_copy)
+BOOL LLImageGL::createGLTexture(S32 discard_level, const LLImageRaw* imageraw, S32 usename/*=0*/, BOOL to_create, S32 category, bool defer_copy, LLGLuint* tex_name)
 {
     LL_PROFILE_ZONE_SCOPED_CATEGORY_TEXTURE;
     checkActiveThread();
@@ -1474,15 +1476,17 @@ BOOL LLImageGL::createGLTexture(S32 discard_level, const LLImageRaw* imageraw, S
 
 	setCategory(category);
  	const U8* rawdata = imageraw->getData();
-	return createGLTexture(discard_level, rawdata, FALSE, usename, defer_copy);
+	return createGLTexture(discard_level, rawdata, FALSE, usename, defer_copy, tex_name);
 }
 
-BOOL LLImageGL::createGLTexture(S32 discard_level, const U8* data_in, BOOL data_hasmips, S32 usename, bool defer_copy)
+BOOL LLImageGL::createGLTexture(S32 discard_level, const U8* data_in, BOOL data_hasmips, S32 usename, bool defer_copy, LLGLuint* tex_name)
     // Call with void data, vmem is allocated but unitialized
 {
     LL_PROFILE_ZONE_SCOPED_CATEGORY_TEXTURE;
     checkActiveThread();
 
+    bool main_thread = on_main_thread();
+
     if (defer_copy)
     {
         data_in = nullptr;
@@ -1501,29 +1505,40 @@ BOOL LLImageGL::createGLTexture(S32 discard_level, const U8* data_in, BOOL data_
     }
     discard_level = llclamp(discard_level, 0, (S32)mMaxDiscardLevel);
 
-    if (!defer_copy // <--- hacky way to force creation of mNewTexName from media texture update
+    if (main_thread // <--- always force creation of mNewTexName when not on main thread
         && mTexName != 0 && discard_level == mCurrentDiscardLevel)
     {
+        LL_PROFILE_ZONE_NAMED("cglt - early setImage");
         // This will only be true if the size has not changed
+        if (tex_name != nullptr)
+        {
+            *tex_name = mTexName;
+        }
         return setImage(data_in, data_hasmips);
     }
 
     GLuint old_texname = mTexName;
-    
+    GLuint new_texname = 0;
     if (usename != 0)
     {
-        mNewTexName = usename;
+        llassert(main_thread);
+        new_texname = usename;
     }
     else
     {
-        LLImageGL::generateTextures(1, &mNewTexName);
+        LLImageGL::generateTextures(1, &new_texname);
         {
-            gGL.getTexUnit(0)->bind(this, false, false, mNewTexName);
+            gGL.getTexUnit(0)->bind(this, false, false, new_texname);
             glTexParameteri(LLTexUnit::getInternalType(mBindTarget), GL_TEXTURE_BASE_LEVEL, 0);
             glTexParameteri(LLTexUnit::getInternalType(mBindTarget), GL_TEXTURE_MAX_LEVEL, mMaxDiscardLevel - discard_level);
         }
     }
     
+    if (tex_name != nullptr)
+    {
+        *tex_name = new_texname;
+    }
+
     if (mUseMipMaps)
     {
         mAutoGenMips = gGLManager.mHasMipMapGeneration;
@@ -1531,9 +1546,12 @@ BOOL LLImageGL::createGLTexture(S32 discard_level, const U8* data_in, BOOL data_
 
     mCurrentDiscardLevel = discard_level;
 
-    if (!setImage(data_in, data_hasmips, mNewTexName))
     {
-        return FALSE;
+        LL_PROFILE_ZONE_NAMED("cglt - late setImage");
+        if (!setImage(data_in, data_hasmips, new_texname))
+        {
+            return FALSE;
+        }
     }
 
     // Set texture options to our defaults.
@@ -1550,11 +1568,11 @@ BOOL LLImageGL::createGLTexture(S32 discard_level, const U8* data_in, BOOL data_
     }
 
     //if we're on the image loading thread, be sure to delete old_texname and update mTexName on the main thread
-    if (!on_main_thread())
+    if (!main_thread)
     {
         if (!defer_copy)
         {
-            syncToMainThread();
+            syncToMainThread(new_texname);
         }
     }
     else 
@@ -1564,8 +1582,7 @@ BOOL LLImageGL::createGLTexture(S32 discard_level, const U8* data_in, BOOL data_
         {
             LLImageGL::deleteTextures(1, &old_texname);
         }
-        mTexName = mNewTexName;
-        mNewTexName = 0;
+        mTexName = new_texname;
     }
     
     mTextureMemory = (S32Bytes)getMipBytes(mCurrentDiscardLevel);
@@ -1579,13 +1596,36 @@ BOOL LLImageGL::createGLTexture(S32 discard_level, const U8* data_in, BOOL data_
     return TRUE;
 }
 
-void LLImageGL::syncToMainThread()
+void update_free_vram()
+{
+    LL_PROFILE_ZONE_SCOPED;
+
+    if (gGLManager.mHasATIMemInfo)
+    {
+        S32 meminfo[4];
+        glGetIntegerv(GL_TEXTURE_FREE_MEMORY_ATI, meminfo);
+        LLImageGLThread::sFreeVRAMMegabytes = meminfo[0];
+
+    }
+    else if (gGLManager.mHasNVXMemInfo)
+    {
+        S32 free_memory;
+        glGetIntegerv(GL_GPU_MEMORY_INFO_CURRENT_AVAILABLE_VIDMEM_NVX, &free_memory);
+        LLImageGLThread::sFreeVRAMMegabytes = free_memory / 1024;
+    }
+}
+
+void LLImageGL::syncToMainThread(LLGLuint new_tex_name)
 {
+    LL_PROFILE_ZONE_SCOPED;
     {
         LL_PROFILE_ZONE_NAMED("cglt - sync");
         if (gGLManager.mHasSync)
         {
             // post a sync to the main thread (will execute before tex name swap lambda below)
+            // glFlush calls here are partly superstitious and partly backed by observation
+            // on AMD hardware
+            glFlush();
             auto sync = glFenceSync(GL_SYNC_GPU_COMMANDS_COMPLETE, 0);
             glFlush();
             LL::WorkQueue::postMaybe(
@@ -1615,17 +1655,27 @@ void LLImageGL::syncToMainThread()
         [=]()
         {
             LL_PROFILE_ZONE_NAMED("cglt - delete callback");
-            if (mNewTexName != 0)
+            if (new_tex_name != 0)
             {
                 if (mTexName != 0)
                 {
                     LLImageGL::deleteTextures(1, &mTexName);
                 }
-                mTexName = mNewTexName;
-                mNewTexName = 0;
+                mTexName = new_tex_name;
                 unref();
             }
         });
+
+
+    //update free vram periodically
+    static LLFrameTimer timer;
+    
+    if (timer.getElapsedTimeF32() > 1.f) //call this once per second.
+    {
+        update_free_vram();
+        timer.reset();
+    }
+    
 }
 
 BOOL LLImageGL::readBackRaw(S32 discard_level, LLImageRaw* imageraw, bool compressed_ok) const
@@ -1756,14 +1806,6 @@ void LLImageGL::destroyGLTexture()
 		mTexName = 0;		
 		mGLTextureCreated = FALSE ;
 	}
-
-    // clean up any in-flight name change
-    if (0 != mNewTexName)
-    {
-        // Memory is transient, not tracked by sGlobalTextuerMemory
-        LLImageGL::deleteTextures(1, &mNewTexName);
-        mNewTexName = 0;
-    }
 }
 
 //force to invalidate the gl texture, most likely a sculpty texture
@@ -2306,6 +2348,8 @@ void LLImageGL::checkActiveThread()
 		glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAX_LEVEL,  nummips);
 */  
 
+std::atomic<S32> LLImageGLThread::sFreeVRAMMegabytes;
+
 LLImageGLThread::LLImageGLThread(LLWindow* window)
     // We want exactly one thread, but a very large capacity: we never want
     // anyone, especially inner-loop render code, to have to block on post()
@@ -2332,3 +2376,14 @@ void LLImageGLThread::run()
     gGL.shutdown();
     mWindow->destroySharedContext(mContext);
 }
+
+S32 LLImageGLThread::getFreeVRAMMegabytes()
+{
+    if (!sEnabled)
+    {
+        update_free_vram();
+    }
+
+    return sFreeVRAMMegabytes;
+}
+
diff --git a/indra/llrender/llimagegl.h b/indra/llrender/llimagegl.h
index 6c6e327e42..50706890f9 100644
--- a/indra/llrender/llimagegl.h
+++ b/indra/llrender/llimagegl.h
@@ -110,17 +110,17 @@ public:
     
 	BOOL createGLTexture() ;
 	BOOL createGLTexture(S32 discard_level, const LLImageRaw* imageraw, S32 usename = 0, BOOL to_create = TRUE,
-		S32 category = sMaxCategories-1, bool defer_copy = false);
-	BOOL createGLTexture(S32 discard_level, const U8* data, BOOL data_hasmips = FALSE, S32 usename = 0, bool defer_copy = false);
+		S32 category = sMaxCategories-1, bool defer_copy = false, LLGLuint* tex_name = nullptr);
+	BOOL createGLTexture(S32 discard_level, const U8* data, BOOL data_hasmips = FALSE, S32 usename = 0, bool defer_copy = false, LLGLuint* tex_name = nullptr);
 	void setImage(const LLImageRaw* imageraw);
 	BOOL setImage(const U8* data_in, BOOL data_hasmips = FALSE, S32 usename = 0);
-	BOOL setSubImage(const LLImageRaw* imageraw, S32 x_pos, S32 y_pos, S32 width, S32 height, BOOL force_fast_update = FALSE, bool use_new_name = false);
-	BOOL setSubImage(const U8* datap, S32 data_width, S32 data_height, S32 x_pos, S32 y_pos, S32 width, S32 height, BOOL force_fast_update = FALSE, bool use_new_name = false);
+	BOOL setSubImage(const LLImageRaw* imageraw, S32 x_pos, S32 y_pos, S32 width, S32 height, BOOL force_fast_update = FALSE, LLGLuint use_name = 0);
+	BOOL setSubImage(const U8* datap, S32 data_width, S32 data_height, S32 x_pos, S32 y_pos, S32 width, S32 height, BOOL force_fast_update = FALSE, LLGLuint use_name = 0);
 	BOOL setSubImageFromFrameBuffer(S32 fb_x, S32 fb_y, S32 x_pos, S32 y_pos, S32 width, S32 height);
 
     // wait for gl commands to finish on current thread and push
     // a lambda to main thread to swap mNewTexName and mTexName
-    void syncToMainThread();
+    void syncToMainThread(LLGLuint new_tex_name);
 
 	// Read back a raw image for this discard level, if it exists
 	BOOL readBackRaw(S32 discard_level, LLImageRaw* imageraw, bool compressed_ok) const;
@@ -224,7 +224,7 @@ private:
 
 	bool     mGLTextureCreated ;
 	LLGLuint mTexName;
-    LLGLuint mNewTexName = 0; // tex name set by background thread to be applied in main thread
+    //LLGLuint mNewTexName = 0; // tex name set by background thread to be applied in main thread
 	U16      mWidth;
 	U16      mHeight;
 	S8       mCurrentDiscardLevel;
@@ -324,6 +324,9 @@ public:
     // follows gSavedSettings "RenderGLMultiThreaded"
     static bool sEnabled;
 
+    // free video memory in megabytes
+    static std::atomic<S32> sFreeVRAMMegabytes;
+
     LLImageGLThread(LLWindow* window);
 
     // post a function to be executed on the LLImageGL background thread
@@ -335,6 +338,8 @@ public:
 
     void run() override;
 
+    static S32 getFreeVRAMMegabytes();
+
 private:
     LLWindow* mWindow;
     void* mContext = nullptr;
diff --git a/indra/llwindow/llwindowwin32.cpp b/indra/llwindow/llwindowwin32.cpp
index 8c792daac0..5ce51bb43e 100644
--- a/indra/llwindow/llwindowwin32.cpp
+++ b/indra/llwindow/llwindowwin32.cpp
@@ -3564,6 +3564,7 @@ void LLWindowWin32::swapBuffers()
 {
     LL_PROFILE_ZONE_SCOPED_CATEGORY_WIN32;
     ASSERT_MAIN_THREAD();
+    glFlush(); //superstitious flush for maybe frame stall removal?
 	SwapBuffers(mhDC);
 
     LL_PROFILER_GPU_COLLECT
diff --git a/indra/newview/lldrawpoolbump.cpp b/indra/newview/lldrawpoolbump.cpp
index 471b0e2c48..20287a7777 100644
--- a/indra/newview/lldrawpoolbump.cpp
+++ b/indra/newview/lldrawpoolbump.cpp
@@ -54,6 +54,8 @@
 
 // static
 LLStandardBumpmap gStandardBumpmapList[TEM_BUMPMAP_COUNT]; 
+LL::WorkQueue::weak_t LLBumpImageList::sMainQueue;
+LL::WorkQueue::weak_t LLBumpImageList::sTexUpdateQueue;
 
 // static
 U32 LLStandardBumpmap::sStandardBumpmapCount = 0;
@@ -761,6 +763,8 @@ void LLBumpImageList::init()
 	LLStandardBumpmap::init();
 
 	LLStandardBumpmap::restoreGL();
+    sMainQueue = LL::WorkQueue::getInstance("mainloop");
+    sTexUpdateQueue = LL::WorkQueue::getInstance("LLImageGL"); // Share work queue with tex loader.
 }
 
 void LLBumpImageList::clear()
@@ -909,10 +913,7 @@ LLViewerTexture* LLBumpImageList::getBrightnessDarknessImage(LLViewerFetchedText
 	}
 	else
 	{
-		LLPointer<LLImageRaw> raw = new LLImageRaw(1,1,1);
-		raw->clear(0x77, 0x77, 0xFF, 0xFF);
-
-		(*entries_list)[src_image->getID()] = LLViewerTextureManager::getLocalTexture( raw.get(), TRUE);
+		(*entries_list)[src_image->getID()] = LLViewerTextureManager::getLocalTexture( TRUE );
 		bump = (*entries_list)[src_image->getID()]; // In case callback was called immediately and replaced the image
 	}
 
@@ -1043,10 +1044,7 @@ void LLBumpImageList::onSourceLoaded( BOOL success, LLViewerTexture *src_vi, LLI
 							iter->second->getWidth() != src->getWidth() ||
 							iter->second->getHeight() != src->getHeight()) // bump not cached yet or has changed resolution
 			{ //make sure an entry exists for this image
-				LLPointer<LLImageRaw> raw = new LLImageRaw(1,1,1);
-				raw->clear(0x77, 0x77, 0xFF, 0xFF);
-
-				entries_list[src_vi->getID()] = LLViewerTextureManager::getLocalTexture( raw.get(), TRUE);
+				entries_list[src_vi->getID()] = LLViewerTextureManager::getLocalTexture(TRUE);
 				iter = entries_list.find(src_vi->getID());
 			}
 		}
@@ -1166,108 +1164,161 @@ void LLBumpImageList::onSourceLoaded( BOOL success, LLViewerTexture *src_vi, LLI
 			}
 
 			//---------------------------------------------------
-			// immediately assign bump to a global smart pointer in case some local smart pointer
+			// immediately assign bump to a smart pointer in case some local smart pointer
 			// accidentally releases it.
-			LLPointer<LLViewerTexture> bump = LLViewerTextureManager::getLocalTexture( TRUE );
+            LLPointer<LLViewerTexture> bump = iter->second;
 
 			if (!LLPipeline::sRenderDeferred)
 			{
 				bump->setExplicitFormat(GL_ALPHA8, GL_ALPHA);
-				bump->createGLTexture(0, dst_image);
+
+                auto tex_queue = LLImageGLThread::sEnabled ? sTexUpdateQueue.lock() : nullptr;
+
+                if (tex_queue)
+                { //dispatch creation to background thread
+                    LLImageRaw* dst_ptr = dst_image;
+                    LLViewerTexture* bump_ptr = bump;
+                    dst_ptr->ref();
+                    bump_ptr->ref();
+                    tex_queue->post(
+                        [=]()
+                        {
+                            LL_PROFILE_ZONE_NAMED("bil - create texture");
+                            bump_ptr->createGLTexture(0, dst_ptr);
+                            bump_ptr->unref();
+                            dst_ptr->unref();
+                        });
+                }
+                else
+                {
+                    bump->createGLTexture(0, dst_image);
+                }
 			}
 			else 
 			{ //convert to normal map
 				
 				//disable compression on normal maps to prevent errors below
 				bump->getGLTexture()->setAllowCompression(false);
+                bump->getGLTexture()->setUseMipMaps(TRUE);
 
-				{
-					bump->setExplicitFormat(GL_RGBA8, GL_ALPHA);
-					bump->createGLTexture(0, dst_image);
-				}
+                auto* bump_ptr = bump.get();
+                auto* dst_ptr = dst_image.get();
 
-				{
-					gPipeline.mScreen.bindTarget();
-					
-					LLGLDepthTest depth(GL_FALSE);
-					LLGLDisable cull(GL_CULL_FACE);
-					LLGLDisable blend(GL_BLEND);
-					gGL.setColorMask(TRUE, TRUE);
-					gNormalMapGenProgram.bind();
+                bump_ptr->ref();
+                dst_ptr->ref();
 
-					static LLStaticHashedString sNormScale("norm_scale");
-					static LLStaticHashedString sStepX("stepX");
-					static LLStaticHashedString sStepY("stepY");
+                bump_ptr->setExplicitFormat(GL_RGBA8, GL_ALPHA);
 
-					gNormalMapGenProgram.uniform1f(sNormScale, gSavedSettings.getF32("RenderNormalMapScale"));
-					gNormalMapGenProgram.uniform1f(sStepX, 1.f/bump->getWidth());
-					gNormalMapGenProgram.uniform1f(sStepY, 1.f/bump->getHeight());
+                auto create_texture = [bump_ptr, dst_ptr]()
+                {
+                    LL_PROFILE_ZONE_NAMED("bil - create texture deferred");
+                    bump_ptr->createGLTexture(0, dst_ptr);
+                };
 
-					LLVector2 v((F32) bump->getWidth()/gPipeline.mScreen.getWidth(),
-								(F32) bump->getHeight()/gPipeline.mScreen.getHeight());
+                auto gen_normal_map = [bump_ptr, dst_ptr]()
+                {
+                    LL_PROFILE_ZONE_NAMED("bil - generate normal map");
+                    gPipeline.mScreen.bindTarget();
 
-					gGL.getTexUnit(0)->bind(bump);
-					
-					S32 width = bump->getWidth();
-					S32 height = bump->getHeight();
+                    LLGLDepthTest depth(GL_FALSE);
+                    LLGLDisable cull(GL_CULL_FACE);
+                    LLGLDisable blend(GL_BLEND);
+                    gGL.setColorMask(TRUE, TRUE);
+                    gNormalMapGenProgram.bind();
 
-					S32 screen_width = gPipeline.mScreen.getWidth();
-					S32 screen_height = gPipeline.mScreen.getHeight();
+                    static LLStaticHashedString sNormScale("norm_scale");
+                    static LLStaticHashedString sStepX("stepX");
+                    static LLStaticHashedString sStepY("stepY");
 
-					glViewport(0, 0, screen_width, screen_height);
+                    gNormalMapGenProgram.uniform1f(sNormScale, gSavedSettings.getF32("RenderNormalMapScale"));
+                    gNormalMapGenProgram.uniform1f(sStepX, 1.f / bump_ptr->getWidth());
+                    gNormalMapGenProgram.uniform1f(sStepY, 1.f / bump_ptr->getHeight());
 
-					for (S32 left = 0; left < width; left += screen_width)
-					{
-						S32 right = left + screen_width;
-						right = llmin(right, width);
-						
-						F32 left_tc = (F32) left/ width;
-						F32 right_tc = (F32) right/width;
+                    LLVector2 v((F32)bump_ptr->getWidth() / gPipeline.mScreen.getWidth(),
+                        (F32)bump_ptr->getHeight() / gPipeline.mScreen.getHeight());
 
-						for (S32 bottom = 0; bottom < height; bottom += screen_height)
-						{
-							S32 top = bottom+screen_height;
-							top = llmin(top, height);
+                    gGL.getTexUnit(0)->bind(bump_ptr);
 
-							F32 bottom_tc = (F32) bottom/height;
-							F32 top_tc = (F32)(bottom+screen_height)/height;
-							top_tc = llmin(top_tc, 1.f);
+                    S32 width = bump_ptr->getWidth();
+                    S32 height = bump_ptr->getHeight();
 
-							F32 screen_right = (F32) (right-left)/screen_width;
-							F32 screen_top = (F32) (top-bottom)/screen_height;
+                    S32 screen_width = gPipeline.mScreen.getWidth();
+                    S32 screen_height = gPipeline.mScreen.getHeight();
 
-							gGL.begin(LLRender::TRIANGLE_STRIP);
-							gGL.texCoord2f(left_tc, bottom_tc);
-							gGL.vertex2f(0, 0);
+                    glViewport(0, 0, screen_width, screen_height);
 
-							gGL.texCoord2f(left_tc, top_tc);
-							gGL.vertex2f(0, screen_top);
+                    for (S32 left = 0; left < width; left += screen_width)
+                    {
+                        S32 right = left + screen_width;
+                        right = llmin(right, width);
 
-							gGL.texCoord2f(right_tc, bottom_tc);
-							gGL.vertex2f(screen_right, 0);
+                        F32 left_tc = (F32)left / width;
+                        F32 right_tc = (F32)right / width;
 
-							gGL.texCoord2f(right_tc, top_tc);
-							gGL.vertex2f(screen_right, screen_top);
+                        for (S32 bottom = 0; bottom < height; bottom += screen_height)
+                        {
+                            S32 top = bottom + screen_height;
+                            top = llmin(top, height);
 
-							gGL.end();
+                            F32 bottom_tc = (F32)bottom / height;
+                            F32 top_tc = (F32)(bottom + screen_height) / height;
+                            top_tc = llmin(top_tc, 1.f);
 
-							gGL.flush();
+                            F32 screen_right = (F32)(right - left) / screen_width;
+                            F32 screen_top = (F32)(top - bottom) / screen_height;
 
-							S32 w = right-left;
-							S32 h = top-bottom;
+                            gGL.begin(LLRender::TRIANGLE_STRIP);
+                            gGL.texCoord2f(left_tc, bottom_tc);
+                            gGL.vertex2f(0, 0);
 
-							glCopyTexSubImage2D(GL_TEXTURE_2D, 0, left, bottom, 0, 0, w, h);
-						}
-					}
+                            gGL.texCoord2f(left_tc, top_tc);
+                            gGL.vertex2f(0, screen_top);
 
-					glGenerateMipmap(GL_TEXTURE_2D);
+                            gGL.texCoord2f(right_tc, bottom_tc);
+                            gGL.vertex2f(screen_right, 0);
 
-					gPipeline.mScreen.flush();
+                            gGL.texCoord2f(right_tc, top_tc);
+                            gGL.vertex2f(screen_right, screen_top);
 
-					gNormalMapGenProgram.unbind();
-										
-					//generateNormalMapFromAlpha(dst_image, nrm_image);
-				}
+                            gGL.end();
+
+                            gGL.flush();
+
+                            S32 w = right - left;
+                            S32 h = top - bottom;
+
+                            glCopyTexSubImage2D(GL_TEXTURE_2D, 0, left, bottom, 0, 0, w, h);
+                        }
+                    }
+
+                    glGenerateMipmap(GL_TEXTURE_2D);
+
+                    gPipeline.mScreen.flush();
+
+                    gNormalMapGenProgram.unbind();
+
+                    //generateNormalMapFromAlpha(dst_image, nrm_image);
+
+                    bump_ptr->unref();
+                    dst_ptr->unref();
+                };
+
+                auto main_queue = LLImageGLThread::sEnabled ? sMainQueue.lock() : nullptr;
+
+                if (main_queue)
+                { //dispatch creation to background thread
+                    LLImageRaw* dst_ptr = dst_image;
+                    LLViewerTexture* bump_ptr = bump;
+                    dst_ptr->ref();
+                    bump_ptr->ref();
+
+                    main_queue->postTo(sTexUpdateQueue, create_texture, gen_normal_map);
+                }
+                else
+                {
+                    create_texture();
+                    gen_normal_map();
+                }
 			}
 		
 			iter->second = bump; // derefs (and deletes) old image
diff --git a/indra/newview/lldrawpoolbump.h b/indra/newview/lldrawpoolbump.h
index 38744a7d98..6e21859738 100644
--- a/indra/newview/lldrawpoolbump.h
+++ b/indra/newview/lldrawpoolbump.h
@@ -161,6 +161,8 @@ private:
 	typedef std::unordered_map<LLUUID, LLPointer<LLViewerTexture> > bump_image_map_t;
 	bump_image_map_t mBrightnessEntries;
 	bump_image_map_t mDarknessEntries;
+    static LL::WorkQueue::weak_t sMainQueue;
+    static LL::WorkQueue::weak_t sTexUpdateQueue;
 };
 
 extern LLBumpImageList gBumpImageList;
diff --git a/indra/newview/llviewerdisplay.cpp b/indra/newview/llviewerdisplay.cpp
index b5d3dc5d30..4fc1bdbec3 100644
--- a/indra/newview/llviewerdisplay.cpp
+++ b/indra/newview/llviewerdisplay.cpp
@@ -649,9 +649,6 @@ void display(BOOL rebuild, F32 zoom_factor, int subfield, BOOL for_snapshot)
 			gPipeline.toggleRenderType(LLPipeline::RENDER_TYPE_HUD_PARTICLES);
 		}
 
-		//upkeep gl name pools
-		LLGLNamePool::upkeepPools();
-		
 		stop_glerror();
 		display_update_camera();
 		stop_glerror();
diff --git a/indra/newview/llviewermedia.cpp b/indra/newview/llviewermedia.cpp
index bd60b9f1e2..42bd5d8367 100644
--- a/indra/newview/llviewermedia.cpp
+++ b/indra/newview/llviewermedia.cpp
@@ -2986,11 +2986,12 @@ void LLViewerMediaImpl::doMediaTexUpdate(LLViewerMediaTexture* media_tex, U8* da
     LLPointer<LLImageRaw> raw = new LLImageRaw(data, media_tex->getWidth(), media_tex->getHeight(), media_tex->getComponents(), true);
         
     // Allocate GL texture based on LLImageRaw but do NOT copy to GL
-    media_tex->createGLTexture(0, raw, 0, TRUE, LLGLTexture::OTHER, true);
+    LLGLuint tex_name = 0;
+    media_tex->createGLTexture(0, raw, 0, TRUE, LLGLTexture::OTHER, true, &tex_name);
 
     // copy just the subimage covered by the image raw to GL
-    media_tex->setSubImage(data, data_width, data_height, x_pos, y_pos, width, height, sync);
-    media_tex->getGLTexture()->syncToMainThread();
+    media_tex->setSubImage(data, data_width, data_height, x_pos, y_pos, width, height, tex_name);
+    media_tex->getGLTexture()->syncToMainThread(tex_name);
     
     // release the data pointer before freeing raw so LLImageRaw destructor doesn't
     // free memory at data pointer
diff --git a/indra/newview/llvieweroctree.cpp b/indra/newview/llvieweroctree.cpp
index 3cdef0ebff..65e9fa533d 100644
--- a/indra/newview/llvieweroctree.cpp
+++ b/indra/newview/llvieweroctree.cpp
@@ -787,42 +787,73 @@ void LLViewerOctreeGroup::checkStates()
 //occulsion culling functions and classes
 //-------------------------------------------------------------------------------------------
 std::set<U32> LLOcclusionCullingGroup::sPendingQueries;
-class LLOcclusionQueryPool : public LLGLNamePool
+
+static std::queue<GLuint> sFreeQueries;
+
+U32 LLOcclusionCullingGroup::getNewOcclusionQueryObjectName()
 {
-public:
-	LLOcclusionQueryPool()
-	{
-	}
+    LL_PROFILE_ZONE_SCOPED;
+    // TODO: refactor this to a general purpose name pool
+    static GLuint occlusion_queries[2][1024];
+    static std::atomic<S32> query_count[2];
+    static S32 query_page = -1;
+    
+    //reuse any query names that have been freed 
+    if (!sFreeQueries.empty())
+    {
+        GLuint ret = sFreeQueries.front();
+        sFreeQueries.pop();
+        return ret;
+    }
 
-protected:
+    // first call, immediately fill entire name pool
+    if (query_page == -1)
+    {
+        glGenQueriesARB(1024, occlusion_queries[0]);
+        glGenQueriesARB(1024, occlusion_queries[1]);
+        query_page = 0;
+        query_count[0] = 1024;
+        query_count[1] = 1024;
+    }
 
-	virtual GLuint allocateName()
-	{
-		GLuint ret = 0;
+    if (query_count[query_page] == 0) //this page is empty
+    {
+        //check the other page
+        query_page = (query_page + 1) % 2;
 
-		glGenQueriesARB(1, &ret);
-	
-		return ret;
-	}
+        if (query_count[query_page] == 0)
+        {
+            //the other page is also empty, generate immediately and return
+            GLuint ret;
+            glGenQueriesARB(1, &ret);
+            return ret;
+        }
+    }
 
-	virtual void releaseName(GLuint name)
-	{
-#if LL_TRACK_PENDING_OCCLUSION_QUERIES
-		LLOcclusionCullingGroup::sPendingQueries.erase(name);
-#endif
-		glDeleteQueriesARB(1, &name);
-	}
-};
+    GLuint ret = occlusion_queries[query_page][--query_count[query_page]];
 
-static LLOcclusionQueryPool sQueryPool;
-U32 LLOcclusionCullingGroup::getNewOcclusionQueryObjectName()
-{
-	return sQueryPool.allocate();
+    if (query_count[query_page] == 0)
+    { //exhausted this page, replenish on background thread
+        S32 page = query_page;
+        LL::WorkQueue::postMaybe(LL::WorkQueue::getInstance("LLImageGL"),
+            [=]()
+            {
+                LL_PROFILE_ZONE_NAMED("glGenQueries bg");
+                if (query_count[page] == 0) // <-- protect against redundant attempts to replenish
+                {
+                    glGenQueriesARB(1024, occlusion_queries[page]);
+                    query_count[page] = 1024;
+                    glFlush();
+                }
+            });
+    }
+
+    return ret;
 }
 
 void LLOcclusionCullingGroup::releaseOcclusionQueryObjectName(GLuint name)
 {
-	sQueryPool.release(name);
+    sFreeQueries.push(name);
 }
 
 //=====================================
@@ -1243,7 +1274,10 @@ void LLOcclusionCullingGroup::doOcclusion(LLCamera* camera, const LLVector4a* sh
 						//store which frame this query was issued on
 						mOcclusionIssued[LLViewerCamera::sCurCameraID] = gFrameCount;
 
-    					glBeginQueryARB(mode, mOcclusionQuery[LLViewerCamera::sCurCameraID]);					
+                        {
+                            LL_PROFILE_ZONE_NAMED("glBeginQuery");
+                            glBeginQueryARB(mode, mOcclusionQuery[LLViewerCamera::sCurCameraID]);
+                        }
 					
 						LLGLSLShader* shader = LLGLSLShader::sCurBoundShaderPtr;
 						llassert(shader);
@@ -1282,7 +1316,10 @@ void LLOcclusionCullingGroup::doOcclusion(LLCamera* camera, const LLVector4a* sh
 							}
 						}
 	
-						glEndQueryARB(mode);
+                        {
+                            LL_PROFILE_ZONE_NAMED("glEndQuery");
+                            glEndQueryARB(mode);
+                        }
 					}
 				}
 
diff --git a/indra/newview/llviewertexture.cpp b/indra/newview/llviewertexture.cpp
index 5fed46f437..fa2c070ea0 100644
--- a/indra/newview/llviewertexture.cpp
+++ b/indra/newview/llviewertexture.cpp
@@ -515,9 +515,10 @@ void LLViewerTexture::getGPUMemoryForTextures(S32Megabytes &gpu, S32Megabytes &p
 {
     LL_PROFILE_ZONE_SCOPED_CATEGORY_TEXTURE;
     static LLFrameTimer timer;
+
     static S32Megabytes gpu_res = S32Megabytes(S32_MAX);
     static S32Megabytes physical_res = S32Megabytes(S32_MAX);
-
+    
     if (timer.getElapsedTimeF32() < GPU_MEMORY_CHECK_WAIT_TIME) //call this once per second.
     {
         gpu = gpu_res;
@@ -527,22 +528,11 @@ void LLViewerTexture::getGPUMemoryForTextures(S32Megabytes &gpu, S32Megabytes &p
     timer.reset();
 
     {
-        if (gGLManager.mHasATIMemInfo)
-        {
-            S32 meminfo[4];
-            glGetIntegerv(GL_TEXTURE_FREE_MEMORY_ATI, meminfo);
-            gpu_res = (S32Megabytes)meminfo[0];
-
-            //check main memory, only works for windows.
-            LLMemory::updateMemoryInfo();
-            physical_res = LLMemory::getAvailableMemKB();
-        }
-        else if (gGLManager.mHasNVXMemInfo)
-        {
-            S32 free_memory;
-            glGetIntegerv(GL_GPU_MEMORY_INFO_CURRENT_AVAILABLE_VIDMEM_NVX, &free_memory);
-            gpu_res = (S32Megabytes)(free_memory / 1024);
-        }
+        gpu_res = (S32Megabytes) LLImageGLThread::getFreeVRAMMegabytes();
+        
+        //check main memory, only works for windows.
+        LLMemory::updateMemoryInfo();
+        physical_res = LLMemory::getAvailableMemKB();
 
         gpu = gpu_res;
         physical = physical_res;
diff --git a/indra/newview/pipeline.cpp b/indra/newview/pipeline.cpp
index 8d45e64bf8..ac242c2bf7 100644
--- a/indra/newview/pipeline.cpp
+++ b/indra/newview/pipeline.cpp
@@ -7267,11 +7267,6 @@ void LLPipeline::doResetVertexBuffers(bool forced)
 
 	SUBSYSTEM_CLEANUP(LLVertexBuffer);
 	
-	//delete all name pool caches
-	LLGLNamePool::cleanupPools();
-
-	
-
 	if (LLVertexBuffer::sGLCount > 0)
 	{
 		LL_WARNS() << "VBO wipe failed -- " << LLVertexBuffer::sGLCount << " buffers remaining." << LL_ENDL;