Merge branch 'DRTVWR-546' of ssh://bitbucket.org/lindenlab/viewer into DRTVWR-546

69 files changed, 1147 insertions, 517 deletions

diff --git a/indra/cmake/Copy3rdPartyLibs.cmake b/indra/cmake/Copy3rdPartyLibs.cmake
index 46ddb9d15b..75601ddfcb 100644
--- a/indra/cmake/Copy3rdPartyLibs.cmake
+++ b/indra/cmake/Copy3rdPartyLibs.cmake
@@ -97,6 +97,8 @@ if(WINDOWS)
         set(MSVC_VER 120)
     elseif (MSVC_VERSION GREATER_EQUAL 1910 AND MSVC_VERSION LESS 1920) # Visual Studio 2017
         set(MSVC_VER 140)
+    elseif (MSVC_VERSION GREATER_EQUAL 1920 AND MSVC_VERSION LESS 1930) # Visual Studio 2019
+        set(MSVC_VER 140)
     else (MSVC80)
         MESSAGE(WARNING "New MSVC_VERSION ${MSVC_VERSION} of MSVC: adapt Copy3rdPartyLibs.cmake")
     endif (MSVC80)
diff --git a/indra/llcharacter/lljoint.cpp b/indra/llcharacter/lljoint.cpp
index dee642310e..d72282ab42 100644
--- a/indra/llcharacter/lljoint.cpp
+++ b/indra/llcharacter/lljoint.cpp
@@ -922,6 +922,13 @@ const LLMatrix4 &LLJoint::getWorldMatrix()
 	return mXform.getWorldMatrix();
 }
 
+const LLMatrix4a& LLJoint::getWorldMatrix4a()
+{
+    updateWorldMatrixParent();
+
+    return mWorldMatrix;
+}
+
 
 //--------------------------------------------------------------------
 // setWorldMatrix()
@@ -1003,6 +1010,7 @@ void LLJoint::updateWorldMatrix()
 	{
 		sNumUpdates++;
 		mXform.updateMatrix(FALSE);
+        mWorldMatrix.loadu(mXform.getWorldMatrix());
 		mDirtyFlags = 0x0;
 	}
 }
diff --git a/indra/llcharacter/lljoint.h b/indra/llcharacter/lljoint.h
index 1b646b641f..ba821667c7 100644
--- a/indra/llcharacter/lljoint.h
+++ b/indra/llcharacter/lljoint.h
@@ -38,6 +38,7 @@
 #include "m4math.h"
 #include "llquaternion.h"
 #include "xform.h"
+#include "llmatrix4a.h"
 
 const S32 LL_CHARACTER_MAX_JOINTS_PER_MESH = 15;
 // Need to set this to count of animate-able joints,
@@ -123,6 +124,7 @@ protected:
 
 	// explicit transformation members
 	LLXformMatrix		mXform;
+    LLMatrix4a          mWorldMatrix;
 
     LLVector3       mDefaultPosition;
     LLVector3       mDefaultScale;
@@ -259,6 +261,8 @@ public:
 	const LLMatrix4 &getWorldMatrix();
 	void setWorldMatrix( const LLMatrix4& mat );
 
+    const LLMatrix4a& getWorldMatrix4a();
+
 	void updateWorldMatrixChildren();
 	void updateWorldMatrixParent();
 
diff --git a/indra/llcommon/linden_common.h b/indra/llcommon/linden_common.h
index 45ac43910c..a228fd22be 100644
--- a/indra/llcommon/linden_common.h
+++ b/indra/llcommon/linden_common.h
@@ -27,6 +27,14 @@
 #ifndef LL_LINDEN_COMMON_H
 #define LL_LINDEN_COMMON_H
 
+#include "llprofiler.h"
+#if TRACY_ENABLE && !defined(LL_PROFILER_ENABLE_TRACY_OPENGL)  // hooks for memory profiling
+void *tracy_aligned_malloc(size_t size, size_t alignment);
+void  tracy_aligned_free(void *memblock);
+#define _aligned_malloc(X, Y) tracy_aligned_malloc((X), (Y))
+#define _aligned_free(X)      tracy_aligned_free((X))
+#endif
+
 // *NOTE:  Please keep includes here to a minimum!
 //
 // Files included here are included in every library .cpp file and
@@ -60,6 +68,4 @@
 #include "llerror.h"
 #include "llfile.h"
 
-#include "llprofiler.h" // must be before fast timer; needed due to LLThreads potentially needing access to tracy
-
 #endif
diff --git a/indra/llcommon/llcommon.cpp b/indra/llcommon/llcommon.cpp
index 96be913d17..da61e7539a 100644
--- a/indra/llcommon/llcommon.cpp
+++ b/indra/llcommon/llcommon.cpp
@@ -33,6 +33,48 @@
 #include "lltracethreadrecorder.h"
 #include "llcleanup.h"
 
+#if (TRACY_ENABLE)
+// Override new/delet for tracy memory profiling
+void *operator new(size_t size)
+{
+    auto ptr = (malloc) (size);
+    if (!ptr)
+    {
+        throw std::bad_alloc();
+        return nullptr;
+    }
+    TracyAlloc(ptr, size);
+    return ptr;
+}
+
+void operator delete(void *ptr) noexcept
+{
+    TracyFree(ptr);
+    (free)(ptr);
+}
+
+// C-style malloc/free can't be so easily overridden, so we define tracy versions and use
+// a pre-processor #define in linden_common.h to redirect to them. The parens around the native
+// functions below prevents recursive substitution by the preprocessor.
+//
+// Unaligned mallocs are rare in LL code but hooking them causes problems in 3p lib code (looking at
+// you, Havok), so we'll only capture the aligned version.
+
+void *tracy_aligned_malloc(size_t size, size_t alignment)
+{
+    auto ptr = (_aligned_malloc) (size, alignment);
+    if (ptr) TracyAlloc(ptr, size);
+    return ptr;
+}
+
+void tracy_aligned_free(void *memblock)
+{
+    TracyFree(memblock);
+    (_aligned_free)(memblock);
+}
+
+#endif
+
 //static
 BOOL LLCommon::sAprInitialized = FALSE;
 
diff --git a/indra/llcommon/llerror.cpp b/indra/llcommon/llerror.cpp
index 8355df9045..f7af181927 100644
--- a/indra/llcommon/llerror.cpp
+++ b/indra/llcommon/llerror.cpp
@@ -109,6 +109,7 @@ namespace {
 		virtual void recordMessage(LLError::ELevel level,
 									const std::string& message) override
 		{
+            LL_PROFILE_ZONE_SCOPED
 			int syslogPriority = LOG_CRIT;
 			switch (level) {
 				case LLError::LEVEL_DEBUG:	syslogPriority = LOG_DEBUG;	break;
@@ -166,6 +167,7 @@ namespace {
         virtual void recordMessage(LLError::ELevel level,
                                     const std::string& message) override
         {
+            LL_PROFILE_ZONE_SCOPED
             if (LLError::getAlwaysFlush())
             {
                 mFile << message << std::endl;
@@ -208,6 +210,7 @@ namespace {
 		virtual void recordMessage(LLError::ELevel level,
 					   const std::string& message) override
 		{
+            LL_PROFILE_ZONE_SCOPED
             static std::string s_ansi_error = createANSI("31"); // red
             static std::string s_ansi_warn  = createANSI("34"); // blue
             static std::string s_ansi_debug = createANSI("35"); // magenta
@@ -220,7 +223,8 @@ namespace {
 			}
             else
             {
-                 fprintf(stderr, "%s\n", message.c_str());
+                LL_PROFILE_ZONE_NAMED("fprintf");
+                fprintf(stderr, "%s\n", message.c_str());
             }
 		}
 	
@@ -229,6 +233,7 @@ namespace {
 
         LL_FORCE_INLINE void writeANSI(const std::string& ansi_code, const std::string& message)
 		{
+            LL_PROFILE_ZONE_SCOPED
             static std::string s_ansi_bold  = createANSI("1");  // bold
             static std::string s_ansi_reset = createANSI("0");  // reset
 			// ANSI color code escape sequence, message, and reset in one fprintf call
@@ -265,6 +270,7 @@ namespace {
 		virtual void recordMessage(LLError::ELevel level,
 								   const std::string& message) override
 		{
+            LL_PROFILE_ZONE_SCOPED
 			mBuffer->addLine(message);
 		}
 	
@@ -291,6 +297,7 @@ namespace {
 		virtual void recordMessage(LLError::ELevel level,
 								   const std::string& message) override
 		{
+            LL_PROFILE_ZONE_SCOPED
 			debugger_print(message);
 		}
 	};
@@ -1178,6 +1185,7 @@ namespace
 
 	void writeToRecorders(const LLError::CallSite& site, const std::string& message)
 	{
+        LL_PROFILE_ZONE_SCOPED
 		LLError::ELevel level = site.mLevel;
 		LLError::SettingsConfigPtr s = LLError::Settings::getInstance()->getSettingsConfig();
 
@@ -1311,6 +1319,7 @@ namespace LLError
 
 	bool Log::shouldLog(CallSite& site)
 	{
+        LL_PROFILE_ZONE_SCOPED
 		LLMutexTrylock lock(getMutex<LOG_MUTEX>(), 5);
 		if (!lock.isLocked())
 		{
@@ -1354,6 +1363,7 @@ namespace LLError
 
 	void Log::flush(const std::ostringstream& out, const CallSite& site)
 	{
+        LL_PROFILE_ZONE_SCOPED
 		LLMutexTrylock lock(getMutex<LOG_MUTEX>(),5);
 		if (!lock.isLocked())
 		{
diff --git a/indra/llcommon/llerror.h b/indra/llcommon/llerror.h
index d439136ca8..d06c0e2132 100644
--- a/indra/llcommon/llerror.h
+++ b/indra/llcommon/llerror.h
@@ -35,7 +35,9 @@
 
 #include "stdtypes.h"
 
+#include "llprofiler.h"
 #include "llpreprocessor.h"
+
 #include <boost/static_assert.hpp>
 
 const int LL_ERR_NOERR = 0;
@@ -348,7 +350,8 @@ typedef LLError::NoClassInfo _LL_CLASS_TO_LOG;
 // if (condition) LL_INFOS() << "True" << LL_ENDL; else LL_INFOS()() << "False" << LL_ENDL;
 
 #define lllog(level, once, ...)                                         \
-	do {                                                                \
+    do {                                                                \
+        LL_PROFILE_ZONE_NAMED("lllog");                                 \
 		const char* tags[] = {"", ##__VA_ARGS__};                       \
 		static LLError::CallSite _site(lllog_site_args_(level, once, tags)); \
 		lllog_test_()
diff --git a/indra/llcommon/llerrorcontrol.h b/indra/llcommon/llerrorcontrol.h
index e87bb7bf35..57f10b7895 100644
--- a/indra/llcommon/llerrorcontrol.h
+++ b/indra/llcommon/llerrorcontrol.h
@@ -190,6 +190,7 @@ namespace LLError
         {}
         void recordMessage(LLError::ELevel level, const std::string& message) override
         {
+            LL_PROFILE_ZONE_SCOPED
             mCallable(level, message);
         }
     private:
diff --git a/indra/llcommon/lleventfilter.h b/indra/llcommon/lleventfilter.h
index 48c2570732..7613850fb2 100644
--- a/indra/llcommon/lleventfilter.h
+++ b/indra/llcommon/lleventfilter.h
@@ -429,6 +429,8 @@ public:
     // path, then stores it to mTarget.
     virtual bool post(const LLSD& event)
     {
+        LL_PROFILE_ZONE_SCOPED
+
         // Extract the element specified by 'mPath' from 'event'. To perform a
         // generic type-appropriate store through mTarget, construct an
         // LLSDParam<T> and store that, thus engaging LLSDParam's custom
diff --git a/indra/llcommon/llmemory.cpp b/indra/llcommon/llmemory.cpp
index ea84e4c1ea..849867586a 100644
--- a/indra/llcommon/llmemory.cpp
+++ b/indra/llcommon/llmemory.cpp
@@ -82,6 +82,7 @@ void LLMemory::initMaxHeapSizeGB(F32Gigabytes max_heap_size)
 //static 
 void LLMemory::updateMemoryInfo() 
 {
+	LL_PROFILE_ZONE_SCOPED
 #if LL_WINDOWS
 	PROCESS_MEMORY_COUNTERS counters;
 
@@ -145,6 +146,7 @@ void* LLMemory::tryToAlloc(void* address, U32 size)
 //static 
 void LLMemory::logMemoryInfo(BOOL update)
 {
+	LL_PROFILE_ZONE_SCOPED
 	if(update)
 	{
 		updateMemoryInfo() ;
diff --git a/indra/llcommon/llmutex.cpp b/indra/llcommon/llmutex.cpp
index 4d73c04d07..a49002b5dc 100644
--- a/indra/llcommon/llmutex.cpp
+++ b/indra/llcommon/llmutex.cpp
@@ -44,6 +44,7 @@ LLMutex::~LLMutex()
 
 void LLMutex::lock()
 {
+    LL_PROFILE_ZONE_SCOPED
 	if(isSelfLocked())
 	{ //redundant lock
 		mCount++;
@@ -65,6 +66,7 @@ void LLMutex::lock()
 
 void LLMutex::unlock()
 {
+    LL_PROFILE_ZONE_SCOPED
 	if (mCount > 0)
 	{ //not the root unlock
 		mCount--;
@@ -85,6 +87,7 @@ void LLMutex::unlock()
 
 bool LLMutex::isLocked()
 {
+    LL_PROFILE_ZONE_SCOPED
 	if (!mMutex.try_lock())
 	{
 		return true;
@@ -108,6 +111,7 @@ LLThread::id_t LLMutex::lockingThread() const
 
 bool LLMutex::trylock()
 {
+    LL_PROFILE_ZONE_SCOPED
 	if(isSelfLocked())
 	{ //redundant lock
 		mCount++;
@@ -146,17 +150,20 @@ LLCondition::~LLCondition()
 
 void LLCondition::wait()
 {
+    LL_PROFILE_ZONE_SCOPED
 	std::unique_lock< std::mutex > lock(mMutex);
 	mCond.wait(lock);
 }
 
 void LLCondition::signal()
 {
+    LL_PROFILE_ZONE_SCOPED
 	mCond.notify_one();
 }
 
 void LLCondition::broadcast()
 {
+    LL_PROFILE_ZONE_SCOPED
 	mCond.notify_all();
 }
 
@@ -166,6 +173,7 @@ LLMutexTrylock::LLMutexTrylock(LLMutex* mutex)
     : mMutex(mutex),
     mLocked(false)
 {
+    LL_PROFILE_ZONE_SCOPED
     if (mMutex)
         mLocked = mMutex->trylock();
 }
@@ -174,6 +182,7 @@ LLMutexTrylock::LLMutexTrylock(LLMutex* mutex, U32 aTries, U32 delay_ms)
     : mMutex(mutex),
     mLocked(false)
 {
+    LL_PROFILE_ZONE_SCOPED
     if (!mMutex)
         return;
 
@@ -188,6 +197,7 @@ LLMutexTrylock::LLMutexTrylock(LLMutex* mutex, U32 aTries, U32 delay_ms)
 
 LLMutexTrylock::~LLMutexTrylock()
 {
+    LL_PROFILE_ZONE_SCOPED
     if (mMutex && mLocked)
         mMutex->unlock();
 }
@@ -199,6 +209,7 @@ LLMutexTrylock::~LLMutexTrylock()
 //
 LLScopedLock::LLScopedLock(std::mutex* mutex) : mMutex(mutex)
 {
+    LL_PROFILE_ZONE_SCOPED
 	if(mutex)
 	{
 		mutex->lock();
@@ -217,6 +228,7 @@ LLScopedLock::~LLScopedLock()
 
 void LLScopedLock::unlock()
 {
+    LL_PROFILE_ZONE_SCOPED
 	if(mLocked)
 	{
 		mMutex->unlock();
diff --git a/indra/llcommon/llprofiler.h b/indra/llcommon/llprofiler.h
index 062c9360dd..49510df913 100644
--- a/indra/llcommon/llprofiler.h
+++ b/indra/llcommon/llprofiler.h
@@ -45,22 +45,55 @@
 //      #define TRACY_ONLY_LOCALHOST 1
         #define TRACY_ONLY_IPV4      1
         #include "Tracy.hpp"
+
+        // Mutually exclusive with detailed memory tracing
+        #define LL_PROFILER_ENABLE_TRACY_OPENGL 0
     #endif
 
     #if LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_TRACY
-        #define LL_PROFILER_FRAME_END               FrameMark
-        #define LL_PROFILER_SET_THREAD_NAME( name ) tracy::SetThreadName( name )
-        #define LL_RECORD_BLOCK_TIME(name)          ZoneNamedN( ___tracy_scoped_zone, #name, true );
+        #define LL_PROFILER_FRAME_END                   FrameMark
+        #define LL_PROFILER_SET_THREAD_NAME( name )     tracy::SetThreadName( name )
+        #define LL_RECORD_BLOCK_TIME(name)              ZoneScoped // Want descriptive names; was: ZoneNamedN( ___tracy_scoped_zone, #name, true );
+        #define LL_PROFILE_ZONE_NAMED(name)             ZoneNamedN( ___tracy_scoped_zone, name, true );
+        #define LL_PROFILE_ZONE_NAMED_COLOR(name,color) ZoneNamedNC( ___tracy_scopped_zone, name, color, true ) // RGB
+        #define LL_PROFILE_ZONE_SCOPED                  ZoneScoped
+
+        #define LL_PROFILE_ZONE_NUM( val )              ZoneValue( val )
+        #define LL_PROFILE_ZONE_TEXT( text, size )      ZoneText( text, size )
+
+        #define LL_PROFILE_ZONE_ERR(name)               LL_PROFILE_ZONE_NAMED_COLOR( name, 0XFF0000  )  // RGB yellow
+        #define LL_PROFILE_ZONE_INFO(name)              LL_PROFILE_ZONE_NAMED_COLOR( name, 0X00FFFF  )  // RGB cyan
+        #define LL_PROFILE_ZONE_WARN(name)              LL_PROFILE_ZONE_NAMED_COLOR( name, 0x0FFFF00 )  // RGB red
     #endif
     #if LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_FAST_TIMER
         #define LL_PROFILER_FRAME_END
-        #define LL_PROFILER_SET_THREAD_NAME( name ) (void)(name)
+        #define LL_PROFILER_SET_THREAD_NAME( name )      (void)(name)
         #define LL_RECORD_BLOCK_TIME(name)                                                                  const LLTrace::BlockTimer& LL_GLUE_TOKENS(block_time_recorder, __LINE__)(LLTrace::timeThisBlock(name)); (void)LL_GLUE_TOKENS(block_time_recorder, __LINE__);
+        #define LL_PROFILE_ZONE_NAMED(name)             // LL_PROFILE_ZONE_NAMED is a no-op when Tracy is disabled
+        #define LL_PROFILE_ZONE_SCOPED                  // LL_PROFILE_ZONE_SCOPED is a no-op when Tracy is disabled
+        #define LL_PROFILE_ZONE_COLOR(name,color)       // LL_RECORD_BLOCK_TIME(name)
+
+        #define LL_PROFILE_ZONE_NUM( val )              (void)( val );                // Not supported
+        #define LL_PROFILE_ZONE_TEXT( text, size )      (void)( text ); void( size ); // Not supported
+
+        #define LL_PROFILE_ZONE_ERR(name)               (void)(name); // Not supported
+        #define LL_PROFILE_ZONE_INFO(name)              (void)(name); // Not supported
+        #define LL_PROFILE_ZONE_WARN(name)              (void)(name); // Not supported
     #endif
     #if LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_TRACY_FAST_TIMER
-        #define LL_PROFILER_FRAME_END               FrameMark
-        #define LL_PROFILER_SET_THREAD_NAME( name ) tracy::SetThreadName( name )
-        #define LL_RECORD_BLOCK_TIME(name)          ZoneNamedN( ___tracy_scoped_zone, #timer_stat, true )   const LLTrace::BlockTimer& LL_GLUE_TOKENS(block_time_recorder, __LINE__)(LLTrace::timeThisBlock(name)); (void)LL_GLUE_TOKENS(block_time_recorder, __LINE__);
+        #define LL_PROFILER_FRAME_END                   FrameMark
+        #define LL_PROFILER_SET_THREAD_NAME( name )     tracy::SetThreadName( name )
+        #define LL_RECORD_BLOCK_TIME(name)              ZoneScoped                                          const LLTrace::BlockTimer& LL_GLUE_TOKENS(block_time_recorder, __LINE__)(LLTrace::timeThisBlock(name)); (void)LL_GLUE_TOKENS(block_time_recorder, __LINE__);
+        #define LL_PROFILE_ZONE_NAMED(name)             ZoneNamedN( ___tracy_scoped_zone, #name, true );
+        #define LL_PROFILE_ZONE_NAMED_COLOR(name,color) ZoneNamedNC( ___tracy_scopped_zone, name, color, true ) // RGB
+        #define LL_PROFILE_ZONE_SCOPED                  ZoneScoped
+
+        #define LL_PROFILE_ZONE_NUM( val )              ZoneValue( val )
+        #define LL_PROFILE_ZONE_TEXT( text, size )      ZoneText( text, size )
+
+        #define LL_PROFILE_ZONE_ERR(name)               LL_PROFILE_ZONE_NAMED_COLOR( name, 0XFF0000  )  // RGB yellow
+        #define LL_PROFILE_ZONE_INFO(name)              LL_PROFILE_ZONE_NAMED_COLOR( name, 0X00FFFF  )  // RGB cyan
+        #define LL_PROFILE_ZONE_WARN(name)              LL_PROFILE_ZONE_NAMED_COLOR( name, 0x0FFFF00 )  // RGB red
     #endif
 #else
     #define LL_PROFILER_FRAME_END
diff --git a/indra/llcommon/llsdutil.cpp b/indra/llcommon/llsdutil.cpp
index eb3a96b133..c2fe15e9b7 100644
--- a/indra/llcommon/llsdutil.cpp
+++ b/indra/llcommon/llsdutil.cpp
@@ -214,6 +214,8 @@ BOOL compare_llsd_with_template(
 	const LLSD& template_llsd,
 	LLSD& resultant_llsd)
 {
+    LL_PROFILE_ZONE_SCOPED
+
 	if (
 		llsd_to_test.isUndefined() &&
 		template_llsd.isDefined() )
@@ -335,6 +337,8 @@ bool filter_llsd_with_template(
 	const LLSD & template_llsd,
 	LLSD & resultant_llsd)
 {
+    LL_PROFILE_ZONE_SCOPED
+
 	if (llsd_to_test.isUndefined() && template_llsd.isDefined())
 	{
 		resultant_llsd = template_llsd;
@@ -529,6 +533,8 @@ class TypeLookup
 public:
     TypeLookup()
     {
+        LL_PROFILE_ZONE_SCOPED
+
         for (const Data *di(boost::begin(typedata)), *dend(boost::end(typedata)); di != dend; ++di)
         {
             mMap[di->type] = di->name;
@@ -537,6 +543,8 @@ public:
 
     std::string lookup(LLSD::Type type) const
     {
+        LL_PROFILE_ZONE_SCOPED
+
         MapType::const_iterator found = mMap.find(type);
         if (found != mMap.end())
         {
@@ -587,6 +595,8 @@ static std::string match_types(LLSD::Type expect, // prototype.type()
                                LLSD::Type actual,        // type we're checking
                                const std::string& pfx)   // as for llsd_matches
 {
+    LL_PROFILE_ZONE_SCOPED
+
     // Trivial case: if the actual type is exactly what we expect, we're good.
     if (actual == expect)
         return "";
@@ -624,6 +634,8 @@ static std::string match_types(LLSD::Type expect, // prototype.type()
 // see docstring in .h file
 std::string llsd_matches(const LLSD& prototype, const LLSD& data, const std::string& pfx)
 {
+    LL_PROFILE_ZONE_SCOPED
+
     // An undefined prototype means that any data is valid.
     // An undefined slot in an array or map prototype means that any data
     // may fill that slot.
@@ -756,6 +768,8 @@ std::string llsd_matches(const LLSD& prototype, const LLSD& data, const std::str
 
 bool llsd_equals(const LLSD& lhs, const LLSD& rhs, int bits)
 {
+    LL_PROFILE_ZONE_SCOPED
+
     // We're comparing strict equality of LLSD representation rather than
     // performing any conversions. So if the types aren't equal, the LLSD
     // values aren't equal.
@@ -864,6 +878,8 @@ namespace llsd
 
 LLSD& drill(LLSD& blob, const LLSD& rawPath)
 {
+    LL_PROFILE_ZONE_SCOPED
+
     // Treat rawPath uniformly as an array. If it's not already an array,
     // store it as the only entry in one. (But let's say Undefined means an
     // empty array.)
@@ -889,6 +905,8 @@ LLSD& drill(LLSD& blob, const LLSD& rawPath)
     // path entry that's bad.
     for (LLSD::Integer i = 0; i < path.size(); ++i)
     {
+        LL_PROFILE_ZONE_NUM( i )
+
         const LLSD& key{path[i]};
         if (key.isString())
         {
@@ -917,6 +935,8 @@ LLSD& drill(LLSD& blob, const LLSD& rawPath)
 
 LLSD drill(const LLSD& blob, const LLSD& path)
 {
+    LL_PROFILE_ZONE_SCOPED
+
     // non-const drill() does exactly what we want. Temporarily cast away
     // const-ness and use that.
     return drill(const_cast<LLSD&>(blob), path);
@@ -929,6 +949,8 @@ LLSD drill(const LLSD& blob, const LLSD& path)
 // filter may be include to exclude/include keys in a map. 
 LLSD llsd_clone(LLSD value, LLSD filter)
 {
+    LL_PROFILE_ZONE_SCOPED
+
     LLSD clone;
     bool has_filter(filter.isMap());
 
diff --git a/indra/llcommon/llsys.cpp b/indra/llcommon/llsys.cpp
index 4e61fb8a58..6d5d043e8d 100644
--- a/indra/llcommon/llsys.cpp
+++ b/indra/llcommon/llsys.cpp
@@ -861,6 +861,7 @@ LLSD LLMemoryInfo::getStatsMap() const
 
 LLMemoryInfo& LLMemoryInfo::refresh()
 {
+	LL_PROFILE_ZONE_SCOPED
 	mStatsMap = loadStatsMap();
 
 	LL_DEBUGS("LLMemoryInfo") << "Populated mStatsMap:\n";
diff --git a/indra/llcommon/llthread.cpp b/indra/llcommon/llthread.cpp
index a8cc750437..11f5a015f1 100644
--- a/indra/llcommon/llthread.cpp
+++ b/indra/llcommon/llthread.cpp
@@ -333,6 +333,7 @@ bool LLThread::runCondition(void)
 // Stop thread execution if requested until unpaused.
 void LLThread::checkPause()
 {
+    LL_PROFILE_ZONE_SCOPED
     mDataLock->lock();
 
     // This is in a while loop because the pthread API allows for spurious wakeups.
@@ -364,17 +365,20 @@ void LLThread::setQuitting()
 // static
 LLThread::id_t LLThread::currentID()
 {
+    LL_PROFILE_ZONE_SCOPED
     return std::this_thread::get_id();
 }
 
 // static
 void LLThread::yield()
 {
+    LL_PROFILE_ZONE_SCOPED
     std::this_thread::yield();
 }
 
 void LLThread::wake()
 {
+    LL_PROFILE_ZONE_SCOPED
     mDataLock->lock();
     if(!shouldSleep())
     {
@@ -385,6 +389,7 @@ void LLThread::wake()
 
 void LLThread::wakeLocked()
 {
+    LL_PROFILE_ZONE_SCOPED
     if(!shouldSleep())
     {
         mRunCondition->signal();
@@ -393,11 +398,13 @@ void LLThread::wakeLocked()
 
 void LLThread::lockData()
 {
+    LL_PROFILE_ZONE_SCOPED
     mDataLock->lock();
 }
 
 void LLThread::unlockData()
 {
+    LL_PROFILE_ZONE_SCOPED
     mDataLock->unlock();
 }
 
diff --git a/indra/llcommon/lluuid.h b/indra/llcommon/lluuid.h
index fe7482ba29..86a396ab06 100644
--- a/indra/llcommon/lluuid.h
+++ b/indra/llcommon/lluuid.h
@@ -184,6 +184,17 @@ struct boost::hash<LLUUID>
     }
 };
 
+// Adapt boost hash to std hash
+namespace std
+{
+    template<> struct hash<LLUUID>
+    {
+        std::size_t operator()(LLUUID const& s) const noexcept
+        {
+            return boost::hash<LLUUID>()(s);
+        }
+    };
+}
 #endif
 
 
diff --git a/indra/llmath/llmatrix4a.h b/indra/llmath/llmatrix4a.h
index 7ba347062f..5291a05607 100644
--- a/indra/llmath/llmatrix4a.h
+++ b/indra/llmath/llmatrix4a.h
@@ -36,6 +36,26 @@ class LLMatrix4a
 public:
 	LL_ALIGN_16(LLVector4a mMatrix[4]);
 
+    LLMatrix4a()
+    {
+
+    }
+
+    explicit LLMatrix4a(const LLMatrix4& val)
+    {
+        loadu(val);
+    }
+
+    inline F32* getF32ptr()
+    {
+        return (F32*) &mMatrix;
+    }
+
+    inline const F32* getF32ptr() const
+    {
+        return (F32*)&mMatrix;
+    }
+
 	inline void clear()
 	{
 		mMatrix[0].clear();
@@ -44,6 +64,14 @@ public:
 		mMatrix[3].clear();
 	}
 
+    inline void setIdentity()
+    {
+        mMatrix[0].set(1.f, 0.f, 0.f, 0.f);
+        mMatrix[1].set(0.f, 1.f, 0.f, 0.f);
+        mMatrix[2].set(0.f, 0.f, 1.f, 0.f);
+        mMatrix[3].set(0.f, 0.f, 0.f, 1.f);
+    }
+
 	inline void loadu(const LLMatrix4& src)
 	{
 		mMatrix[0] = _mm_loadu_ps(src.mMatrix[0]);
@@ -105,7 +133,7 @@ public:
 		mMatrix[3].setAdd(a.mMatrix[3],d3);
 	}
 
-	inline void rotate(const LLVector4a& v, LLVector4a& res)
+	inline void rotate(const LLVector4a& v, LLVector4a& res) const
 	{
 		LLVector4a y,z;
 
@@ -151,6 +179,8 @@ public:
     {
         affineTransformSSE(v,res);
     }
+
+    const LLVector4a& getTranslation() const { return mMatrix[3]; }
 };
 
 inline LLVector4a rowMul(const LLVector4a &row, const LLMatrix4a &mat)
@@ -176,6 +206,15 @@ inline void matMul(const LLMatrix4a &a, const LLMatrix4a &b, LLMatrix4a &res)
     res.mMatrix[3] = row3;
 }
 
+//Faster version of matMul wehere res must not be a or b
+inline void matMulUnsafe(const LLMatrix4a &a, const LLMatrix4a &b, LLMatrix4a &res)
+{
+    res.mMatrix[0] = rowMul(a.mMatrix[0], b);
+    res.mMatrix[1] = rowMul(a.mMatrix[1], b);
+    res.mMatrix[2] = rowMul(a.mMatrix[2], b);
+    res.mMatrix[3] = rowMul(a.mMatrix[3], b);
+}
+
 inline std::ostream& operator<<(std::ostream& s, const LLMatrix4a& m)
 {
     s << "[" << m.mMatrix[0] << ", " << m.mMatrix[1] << ", " << m.mMatrix[2] << ", " << m.mMatrix[3] << "]";
diff --git a/indra/llmath/llvector4a.h b/indra/llmath/llvector4a.h
index 27abf39537..5a02928374 100644
--- a/indra/llmath/llvector4a.h
+++ b/indra/llmath/llvector4a.h
@@ -46,10 +46,9 @@ class LLRotation;
 // of this writing, July 08, 2010) about getting it implemented before you resort to
 // LLVector3/LLVector4. 
 /////////////////////////////////
-struct LLVector4a;
 
 LL_ALIGN_PREFIX(16)
-struct LLVector4a
+class LLVector4a
 {
 public:
 
@@ -138,10 +137,10 @@ public:
 	// BASIC GET/SET 
 	////////////////////////////////////
 	
-	// Return a "this" as an F32 pointer. Do not use unless you have a very good reason.  (Not sure? Ask Falcon)
+	// Return a "this" as an F32 pointer.
 	inline F32* getF32ptr();
 	
-	// Return a "this" as a const F32 pointer. Do not use unless you have a very good reason.  (Not sure? Ask Falcon)
+	// Return a "this" as a const F32 pointer.
 	inline const F32* const getF32ptr() const;
 	
 	// Read-only access a single float in this vector. Do not use in proximity to any function call that manipulates
diff --git a/indra/llmath/llvector4a.inl b/indra/llmath/llvector4a.inl
index 69d3d01efe..8be1c1b114 100644
--- a/indra/llmath/llvector4a.inl
+++ b/indra/llmath/llvector4a.inl
@@ -58,13 +58,13 @@ inline void LLVector4a::store4a(F32* dst) const
 // BASIC GET/SET 
 ////////////////////////////////////
 
-// Return a "this" as an F32 pointer. Do not use unless you have a very good reason.  (Not sure? Ask Falcon)
+// Return a "this" as an F32 pointer.
 F32* LLVector4a::getF32ptr()
 {
 	return (F32*) &mQ;
 }
 
-// Return a "this" as a const F32 pointer. Do not use unless you have a very good reason.  (Not sure? Ask Falcon)
+// Return a "this" as a const F32 pointer.
 const F32* const LLVector4a::getF32ptr() const
 {
 	return (const F32* const) &mQ;
diff --git a/indra/llmath/llvolume.cpp b/indra/llmath/llvolume.cpp
index e085fa6ada..130f30bedc 100644
--- a/indra/llmath/llvolume.cpp
+++ b/indra/llmath/llvolume.cpp
@@ -383,6 +383,7 @@ public:
 	virtual void visit(const LLOctreeNode<LLVolumeTriangle>* branch)
 	{ //this is a depth first traversal, so it's safe to assum all children have complete
 		//bounding data
+	LL_PROFILE_ZONE_SCOPED
 
 		LLVolumeOctreeListener* node = (LLVolumeOctreeListener*) branch->getListener(0);
 
@@ -822,6 +823,8 @@ S32 LLProfile::getNumPoints(const LLProfileParams& params, BOOL path_open,F32 de
 BOOL LLProfile::generate(const LLProfileParams& params, BOOL path_open,F32 detail, S32 split,
 						 BOOL is_sculpted, S32 sculpt_size)
 {
+	LL_PROFILE_ZONE_SCOPED
+
 	if ((!mDirty) && (!is_sculpted))
 	{
 		return FALSE;
@@ -1302,6 +1305,8 @@ S32 LLPath::getNumNGonPoints(const LLPathParams& params, S32 sides, F32 startOff
 
 void LLPath::genNGon(const LLPathParams& params, S32 sides, F32 startOff, F32 end_scale, F32 twist_scale)
 {
+	LL_PROFILE_ZONE_SCOPED
+
 	// Generates a circular path, starting at (1, 0, 0), counterclockwise along the xz plane.
 	static const F32 tableScale[] = { 1, 1, 1, 0.5f, 0.707107f, 0.53f, 0.525f, 0.5f };
 
@@ -1536,6 +1541,8 @@ S32 LLPath::getNumPoints(const LLPathParams& params, F32 detail)
 BOOL LLPath::generate(const LLPathParams& params, F32 detail, S32 split,
 					  BOOL is_sculpted, S32 sculpt_size)
 {
+	LL_PROFILE_ZONE_SCOPED
+
 	if ((!mDirty) && (!is_sculpted))
 	{
 		return FALSE;
@@ -2112,6 +2119,8 @@ LLVolume::~LLVolume()
 
 BOOL LLVolume::generate()
 {
+	LL_PROFILE_ZONE_SCOPED
+
 	LL_CHECK_MEMORY
 	llassert_always(mProfilep);
 	
@@ -2370,6 +2379,8 @@ bool LLVolumeFace::VertexData::compareNormal(const LLVolumeFace::VertexData& rhs
 
 bool LLVolume::unpackVolumeFaces(std::istream& is, S32 size)
 {
+	LL_PROFILE_ZONE_SCOPED
+
 	//input stream is now pointing at a zlib compressed block of LLSD
 	//decompress block
 	LLSD mdl;
@@ -2755,6 +2766,8 @@ S32	LLVolume::getNumFaces() const
 
 void LLVolume::createVolumeFaces()
 {
+	LL_PROFILE_ZONE_SCOPED
+
 	if (mGenerateSingleFace)
 	{
 		// do nothing
@@ -3720,6 +3733,8 @@ void LLVolume::generateSilhouetteVertices(std::vector<LLVector3> &vertices,
 										  const LLMatrix3& norm_mat_in,
 										  S32 face_mask)
 {
+	LL_PROFILE_ZONE_SCOPED
+
 	LLMatrix4a mat;
 	mat.loadu(mat_in);
 
@@ -4846,6 +4861,8 @@ void LLVolumeFace::freeData()
 
 BOOL LLVolumeFace::create(LLVolume* volume, BOOL partial_build)
 {
+	LL_PROFILE_ZONE_SCOPED
+
 	//tree for this face is no longer valid
 	delete mOctree;
 	mOctree = NULL;
@@ -5514,6 +5531,8 @@ bool LLVolumeFace::cacheOptimize()
 
 void LLVolumeFace::createOctree(F32 scaler, const LLVector4a& center, const LLVector4a& size)
 {
+	LL_PROFILE_ZONE_SCOPED
+
 	if (mOctree)
 	{
 		return;
@@ -6287,6 +6306,8 @@ void CalculateTangentArray(U32 vertexCount, const LLVector4a *vertex, const LLVe
 
 void LLVolumeFace::createTangents()
 {
+	LL_PROFILE_ZONE_SCOPED
+
 	if (!mTangents)
 	{
 		allocateTangents(mNumVertices);
@@ -6482,6 +6503,8 @@ void LLVolumeFace::fillFromLegacyData(std::vector<LLVolumeFace::VertexData>& v,
 
 BOOL LLVolumeFace::createSide(LLVolume* volume, BOOL partial_build)
 {
+	LL_PROFILE_ZONE_SCOPED
+
 	LL_CHECK_MEMORY
 	BOOL flat = mTypeMask & FLAT_MASK;
 
@@ -6974,6 +6997,8 @@ BOOL LLVolumeFace::createSide(LLVolume* volume, BOOL partial_build)
 void CalculateTangentArray(U32 vertexCount, const LLVector4a *vertex, const LLVector4a *normal,
         const LLVector2 *texcoord, U32 triangleCount, const U16* index_array, LLVector4a *tangent)
 {
+	LL_PROFILE_ZONE_SCOPED
+
     //LLVector4a *tan1 = new LLVector4a[vertexCount * 2];
 	LLVector4a* tan1 = (LLVector4a*) ll_aligned_malloc_16(vertexCount*2*sizeof(LLVector4a));
 	// new(tan1) LLVector4a;
diff --git a/indra/llmath/m4math.cpp b/indra/llmath/m4math.cpp
index 3baf1bad18..6e40dae30b 100644
--- a/indra/llmath/m4math.cpp
+++ b/indra/llmath/m4math.cpp
@@ -32,8 +32,7 @@
 #include "m4math.h"
 #include "m3math.h"
 #include "llquaternion.h"
-
-
+#include "llmatrix4a.h"
 
 
 // LLMatrix4
@@ -115,6 +114,12 @@ LLMatrix4::LLMatrix4(const LLQuaternion &q)
 	*this = initRotation(q);
 }
 
+LLMatrix4::LLMatrix4(const LLMatrix4a& mat)
+    : LLMatrix4(mat.getF32ptr())
+{
+    
+}
+
 LLMatrix4::LLMatrix4(const LLQuaternion &q, const LLVector4 &pos)
 {
 	*this = initRotTrans(q, pos);
diff --git a/indra/llmath/m4math.h b/indra/llmath/m4math.h
index bf60adb9b6..b9da970cde 100644
--- a/indra/llmath/m4math.h
+++ b/indra/llmath/m4math.h
@@ -32,6 +32,7 @@
 class LLVector4;
 class LLMatrix3;
 class LLQuaternion;
+class LLMatrix4a;
 
 // NOTA BENE: Currently assuming a right-handed, x-forward, y-left, z-up universe
 
@@ -104,6 +105,7 @@ public:
 	explicit LLMatrix4(const F32 *mat);								// Initializes Matrix to values in mat
 	explicit LLMatrix4(const LLMatrix3 &mat);						// Initializes Matrix to values in mat and sets position to (0,0,0)
 	explicit LLMatrix4(const LLQuaternion &q);						// Initializes Matrix with rotation q and sets position to (0,0,0)
+    explicit LLMatrix4(const LLMatrix4a& mat);
 
 	LLMatrix4(const LLMatrix3 &mat, const LLVector4 &pos);	// Initializes Matrix to values in mat and pos
 
diff --git a/indra/llmath/v3math.cpp b/indra/llmath/v3math.cpp
index b04c67d926..93010d2250 100644
--- a/indra/llmath/v3math.cpp
+++ b/indra/llmath/v3math.cpp
@@ -316,6 +316,12 @@ LLVector3::LLVector3(const LLVector4 &vec)
 	mV[VZ] = (F32)vec.mV[VZ];
 }
 
+LLVector3::LLVector3(const LLVector4a& vec)
+    : LLVector3(vec.getF32ptr())
+{
+
+}
+
 LLVector3::LLVector3(const LLSD& sd)
 {
 	setValue(sd);
diff --git a/indra/llmath/v3math.h b/indra/llmath/v3math.h
index 6f857d7061..068f489020 100644
--- a/indra/llmath/v3math.h
+++ b/indra/llmath/v3math.h
@@ -33,6 +33,7 @@
 #include "llsd.h"
 class LLVector2;
 class LLVector4;
+class LLVector4a;
 class LLMatrix3;
 class LLMatrix4;
 class LLVector3d;
@@ -62,7 +63,9 @@ class LLVector3
 		explicit LLVector3(const LLVector2 &vec);				// Initializes LLVector3 to (vec[0]. vec[1], 0)
 		explicit LLVector3(const LLVector3d &vec);				// Initializes LLVector3 to (vec[0]. vec[1], vec[2])
 		explicit LLVector3(const LLVector4 &vec);				// Initializes LLVector4 to (vec[0]. vec[1], vec[2])
-		explicit LLVector3(const LLSD& sd);
+        explicit LLVector3(const LLVector4a& vec);              // Initializes LLVector4 to (vec[0]. vec[1], vec[2])
+        explicit LLVector3(const LLSD& sd);
+        
 
 		LLSD getValue() const;
 
diff --git a/indra/llprimitive/lldaeloader.cpp b/indra/llprimitive/lldaeloader.cpp
index dfa29fb539..8343de0cbc 100644
--- a/indra/llprimitive/lldaeloader.cpp
+++ b/indra/llprimitive/lldaeloader.cpp
@@ -1173,17 +1173,19 @@ void LLDAELoader::processDomModel(LLModel* model, DAE* dae, daeElement* root, do
 
 			LLMeshSkinInfo& skin_info = model->mSkinInfo;
 
+            LLMatrix4 mat;
 			for (int i = 0; i < 4; i++)
 			{
 				for(int j = 0; j < 4; j++)
 				{
-					skin_info.mBindShapeMatrix.mMatrix[i][j] = dom_value[i + j*4];
+                    mat.mMatrix[i][j] = dom_value[i + j*4];
 				}
 			}
 
-			LLMatrix4 trans = normalized_transformation;
-			trans *= skin_info.mBindShapeMatrix;
-			skin_info.mBindShapeMatrix = trans;							
+            skin_info.mBindShapeMatrix.loadu(mat);
+
+			LLMatrix4a trans(normalized_transformation);
+            matMul(trans, skin_info.mBindShapeMatrix, skin_info.mBindShapeMatrix);
 		}
 
 
@@ -1401,7 +1403,7 @@ void LLDAELoader::processDomModel(LLModel* model, DAE* dae, daeElement* root, do
 									mat.mMatrix[i][j] = transform[k*16 + i + j*4];
 								}
 							}
-							model->mSkinInfo.mInvBindMatrix.push_back(mat);
+							model->mSkinInfo.mInvBindMatrix.push_back(LLMatrix4a(mat));
 						}
 					}
 				}
@@ -1475,9 +1477,9 @@ void LLDAELoader::processDomModel(LLModel* model, DAE* dae, daeElement* root, do
 			if (mJointMap.find(lookingForJoint) != mJointMap.end()
 				&& model->mSkinInfo.mInvBindMatrix.size() > i)
 			{
-				LLMatrix4 newInverse = model->mSkinInfo.mInvBindMatrix[i];
+				LLMatrix4 newInverse = LLMatrix4(model->mSkinInfo.mInvBindMatrix[i].getF32ptr());
 				newInverse.setTranslation( mJointList[lookingForJoint].getTranslation() );
-				model->mSkinInfo.mAlternateBindMatrix.push_back( newInverse );
+				model->mSkinInfo.mAlternateBindMatrix.push_back( LLMatrix4a(newInverse) );
             }
 			else
 			{
diff --git a/indra/llprimitive/llmodel.cpp b/indra/llprimitive/llmodel.cpp
index 702a1b5238..a23b991f1d 100644
--- a/indra/llprimitive/llmodel.cpp
+++ b/indra/llprimitive/llmodel.cpp
@@ -1396,7 +1396,7 @@ void LLMeshSkinInfo::fromLLSD(LLSD& skin)
 				}
 			}
 
-			mInvBindMatrix.push_back(mat);
+			mInvBindMatrix.push_back(LLMatrix4a(mat));
 		}
 
         if (mJointNames.size() != mInvBindMatrix.size())
@@ -1410,13 +1410,15 @@ void LLMeshSkinInfo::fromLLSD(LLSD& skin)
 
 	if (skin.has("bind_shape_matrix"))
 	{
+        LLMatrix4 mat;
 		for (U32 j = 0; j < 4; j++)
 		{
 			for (U32 k = 0; k < 4; k++)
 			{
-				mBindShapeMatrix.mMatrix[j][k] = skin["bind_shape_matrix"][j*4+k].asReal();
+				mat.mMatrix[j][k] = skin["bind_shape_matrix"][j*4+k].asReal();
 			}
 		}
+        mBindShapeMatrix.loadu(mat);
 	}
 
 	if (skin.has("alt_inverse_bind_matrix"))
@@ -1432,7 +1434,7 @@ void LLMeshSkinInfo::fromLLSD(LLSD& skin)
 				}
 			}
 			
-			mAlternateBindMatrix.push_back(mat);
+			mAlternateBindMatrix.push_back(LLMatrix4a(mat));
 		}
 	}
 
diff --git a/indra/llprimitive/llmodel.h b/indra/llprimitive/llmodel.h
index 51fa2f8079..96d4582b4f 100644
--- a/indra/llprimitive/llmodel.h
+++ b/indra/llprimitive/llmodel.h
@@ -33,6 +33,8 @@
 #include "m4math.h"
 #include <queue>
 
+#include <boost/align/aligned_allocator.hpp>
+
 class daeElement;
 class domMesh;
 
@@ -49,10 +51,11 @@ public:
 	LLUUID mMeshID;
 	std::vector<std::string> mJointNames;
     mutable std::vector<S32> mJointNums;
-	std::vector<LLMatrix4> mInvBindMatrix;
-	std::vector<LLMatrix4> mAlternateBindMatrix;
+    typedef std::vector<LLMatrix4a, boost::alignment::aligned_allocator<LLMatrix4a, 16>> matrix_list_t;
+	matrix_list_t mInvBindMatrix;
+	matrix_list_t mAlternateBindMatrix;
 
-	LLMatrix4 mBindShapeMatrix;
+	LLMatrix4a mBindShapeMatrix;
 	float mPelvisOffset;
     bool mLockScaleIfJointPosition;
     bool mInvalidJointsScrubbed;
diff --git a/indra/llrender/llglheaders.h b/indra/llrender/llglheaders.h
index 6bca3623e0..3d93cc0762 100644
--- a/indra/llrender/llglheaders.h
+++ b/indra/llrender/llglheaders.h
@@ -812,4 +812,23 @@ extern void glGetBufferPointervARB (GLenum, GLenum, GLvoid* *);
 #define GL_RENDERBUFFER_FREE_MEMORY_ATI            0x87FD
 #endif
 
+#if defined(TRACY_ENABLE) && LL_PROFILER_ENABLE_TRACY_OPENGL
+    // Tracy uses the following:
+    //    glGenQueries
+    //    glGetQueryiv
+    //    glGetQueryObjectiv
+    #define glGenQueries        glGenQueriesARB
+    #define glGetQueryiv        glGetQueryivARB
+    #define glGetQueryObjectiv  glGetQueryObjectivARB
+    #include <tracy/TracyOpenGL.hpp>
+
+    #define LL_PROFILER_GPU_ZONEC(name,color) TracyGpuZoneC(name,color);
+    #define LL_PROFILER_GPU_COLLECT           TracyGpuCollect
+    #define LL_PROFILER_GPU_CONTEXT           TracyGpuContext
+#else
+    #define LL_PROFILER_GPU_ZONEC(name,color) (void)name;(void)color;
+    #define LL_PROFILER_GPU_COLLECT
+    #define LL_PROFILER_GPU_CONTEXT
+#endif
+
 #endif // LL_LLGLHEADERS_H
diff --git a/indra/llrender/llglslshader.cpp b/indra/llrender/llglslshader.cpp
index 4351f6e2c8..9ab38d25a9 100644
--- a/indra/llrender/llglslshader.cpp
+++ b/indra/llrender/llglslshader.cpp
@@ -126,6 +126,7 @@ struct LLGLSLShaderCompareTimeElapsed
 //static
 void LLGLSLShader::finishProfile(bool emit_report)
 {
+    LL_PROFILE_ZONE_SCOPED
     sProfileEnabled = false;
 
     if (emit_report)
@@ -1346,6 +1347,7 @@ void LLGLSLShader::uniformMatrix3fv(U32 index, U32 count, GLboolean transpose, c
 
 void LLGLSLShader::uniformMatrix3x4fv(U32 index, U32 count, GLboolean transpose, const GLfloat *v)
 {
+    LL_PROFILE_ZONE_SCOPED;
 	if (mProgramObject)
 	{	
 		if (mUniform.size() <= index)
diff --git a/indra/llrender/llimagegl.cpp b/indra/llrender/llimagegl.cpp
index 0151d20128..276fa55e15 100644
--- a/indra/llrender/llimagegl.cpp
+++ b/indra/llrender/llimagegl.cpp
@@ -1304,7 +1304,10 @@ void LLImageGL::setManualImage(U32 target, S32 miplevel, S32 intformat, S32 widt
 	}
 
 	stop_glerror();
-	glTexImage2D(target, miplevel, intformat, width, height, 0, pixformat, pixtype, use_scratch ? scratch : pixels);
+	{
+		LL_PROFILE_ZONE_NAMED("glTexImage2D");
+		glTexImage2D(target, miplevel, intformat, width, height, 0, pixformat, pixtype, use_scratch ? scratch : pixels);
+	}
 	stop_glerror();
 
 	if (use_scratch)
diff --git a/indra/llrender/llrender.cpp b/indra/llrender/llrender.cpp
index 03b6aac20c..43b4441ea8 100644
--- a/indra/llrender/llrender.cpp
+++ b/indra/llrender/llrender.cpp
@@ -1848,6 +1848,7 @@ LLLightState* LLRender::getLight(U32 index)
 
 void LLRender::setAmbientLightColor(const LLColor4& color)
 {
+	LL_PROFILE_ZONE_SCOPED
 	if (color != mAmbientLightColor)
 	{
 		++mLightHash;
diff --git a/indra/llrender/llrendertarget.cpp b/indra/llrender/llrendertarget.cpp
index e3c0255290..401085a00b 100644
--- a/indra/llrender/llrendertarget.cpp
+++ b/indra/llrender/llrendertarget.cpp
@@ -437,11 +437,13 @@ void LLRenderTarget::bindTarget()
 									GL_COLOR_ATTACHMENT1,
 									GL_COLOR_ATTACHMENT2,
 									GL_COLOR_ATTACHMENT3};
+			LL_PROFILER_GPU_ZONEC( "gl.DrawBuffersARB", 0x4000FF )
 			glDrawBuffersARB(mTex.size(), drawbuffers);
 		}
 			
 		if (mTex.empty())
 		{ //no color buffer to draw to
+			LL_PROFILER_GPU_ZONEC( "gl.DrawBuffer", 0x0000FF )
 			glDrawBuffer(GL_NONE);
 			glReadBuffer(GL_NONE);
 		}
diff --git a/indra/llrender/llvertexbuffer.cpp b/indra/llrender/llvertexbuffer.cpp
index 7d2b09ca4a..0449ac392c 100644
--- a/indra/llrender/llvertexbuffer.cpp
+++ b/indra/llrender/llvertexbuffer.cpp
@@ -91,6 +91,8 @@ LLVBOPool LLVertexBuffer::sDynamicIBOPool(GL_DYNAMIC_DRAW_ARB, GL_ELEMENT_ARRAY_
 
 U32 LLVBOPool::sBytesPooled = 0;
 U32 LLVBOPool::sIndexBytesPooled = 0;
+U32 LLVBOPool::sNameIdx = 0;
+U32 LLVBOPool::sNamePool[1024];
 
 std::list<U32> LLVertexBuffer::sAvailableVAOName;
 U32 LLVertexBuffer::sCurVAOName = 1;
@@ -121,15 +123,20 @@ bool LLVertexBuffer::sPreferStreamDraw = false;
 
 U32 LLVBOPool::genBuffer()
 {
-	U32 ret = 0;
+	LL_PROFILE_ZONE_SCOPED
 
-	glGenBuffersARB(1, &ret);
-	
-	return ret;
+	if (sNameIdx == 0)
+	{
+		glGenBuffersARB(1024, sNamePool);
+		sNameIdx = 1024;
+	}
+
+	return sNamePool[--sNameIdx];
 }
 
 void LLVBOPool::deleteBuffer(U32 name)
 {
+	LL_PROFILE_ZONE_SCOPED
 	if (gGLManager.mInited)
 	{
 		LLVertexBuffer::unbind();
@@ -152,6 +159,7 @@ LLVBOPool::LLVBOPool(U32 vboUsage, U32 vboType)
 
 volatile U8* LLVBOPool::allocate(U32& name, U32 size, bool for_seed)
 {
+	LL_PROFILE_ZONE_SCOPED
 	llassert(vbo_block_size(size) == size);
 	
 	volatile U8* ret = NULL;
@@ -267,10 +275,12 @@ void LLVBOPool::release(U32 name, volatile U8* buffer, U32 size)
 
 void LLVBOPool::seedPool()
 {
+	LL_PROFILE_ZONE_SCOPED
 	U32 dummy_name = 0;
 
 	if (mFreeList.size() < LL_VBO_POOL_SEED_COUNT)
 	{
+		LL_PROFILE_ZONE_NAMED("VBOPool Resize");
 		mFreeList.resize(LL_VBO_POOL_SEED_COUNT);
 	}
 
@@ -411,6 +421,7 @@ void LLVertexBuffer::releaseVAOName(U32 name)
 //static
 void LLVertexBuffer::seedPools()
 {
+	LL_PROFILE_ZONE_SCOPED
 	sStreamVBOPool.seedPool();
 	sDynamicVBOPool.seedPool();
 	sDynamicCopyVBOPool.seedPool();
@@ -615,6 +626,7 @@ void LLVertexBuffer::drawArrays(U32 mode, const std::vector<LLVector3>& pos, con
 		glNormalPointer(GL_FLOAT, 0, norm[0].mV);
 	}
 	LLGLSLShader::startProfile();
+	LL_PROFILER_GPU_ZONEC( "gl.DrawArrays", 0xFF0000 )
 	glDrawArrays(sGLMode[mode], 0, count);
 	LLGLSLShader::stopProfile(count, mode);
 }
@@ -654,6 +666,7 @@ void LLVertexBuffer::drawElements(U32 mode, const LLVector4a* pos, const LLVecto
 	}
 
 	LLGLSLShader::startProfile();
+    LL_PROFILER_GPU_ZONEC( "gl.DrawElements", 0x80FF80 )
 	glDrawElements(sGLMode[mode], num_indices, GL_UNSIGNED_SHORT, indicesp);
 	LLGLSLShader::stopProfile(num_indices, mode);
 }
@@ -763,6 +776,7 @@ void LLVertexBuffer::drawRange(U32 mode, U32 start, U32 end, U32 count, U32 indi
 
 	stop_glerror();
 	LLGLSLShader::startProfile();
+    LL_PROFILER_GPU_ZONEC( "gl.DrawRangeElements", 0xFFFF00 )
 	glDrawRangeElements(sGLMode[mode], start, end, count, GL_UNSIGNED_SHORT, 
 		idx);
 	LLGLSLShader::stopProfile(count, mode);
@@ -814,6 +828,7 @@ void LLVertexBuffer::draw(U32 mode, U32 count, U32 indices_offset) const
 
 	stop_glerror();
 	LLGLSLShader::startProfile();
+    LL_PROFILER_GPU_ZONEC( "gl.DrawElements", 0xA0FFA0 )
 	glDrawElements(sGLMode[mode], count, GL_UNSIGNED_SHORT,
 		((U16*) getIndicesPointer()) + indices_offset);
 	LLGLSLShader::stopProfile(count, mode);
@@ -861,6 +876,7 @@ void LLVertexBuffer::drawArrays(U32 mode, U32 first, U32 count) const
 		stop_glerror();
 		LLGLSLShader::startProfile();
 		stop_glerror();
+        LL_PROFILER_GPU_ZONEC( "gl.DrawArrays", 0xFF4040 )
 		glDrawArrays(sGLMode[mode], first, count);
 		stop_glerror();
 		LLGLSLShader::stopProfile(count, mode);
diff --git a/indra/llrender/llvertexbuffer.h b/indra/llrender/llvertexbuffer.h
index dbe1a3687f..1d60970df4 100644
--- a/indra/llrender/llvertexbuffer.h
+++ b/indra/llrender/llvertexbuffer.h
@@ -89,6 +89,9 @@ public:
 	std::vector<record_list_t> mFreeList;
 	std::vector<U32> mMissCount;
 
+	//used to avoid calling glGenBuffers for every VBO creation
+	static U32 sNamePool[1024];
+	static U32 sNameIdx;
 };
 
 
@@ -127,7 +130,7 @@ public:
 	static LLVBOPool sDynamicCopyVBOPool;
 	static LLVBOPool sStreamIBOPool;
 	static LLVBOPool sDynamicIBOPool;
-	
+
 	static std::list<U32> sAvailableVAOName;
 	static U32 sCurVAOName;
 
diff --git a/indra/llwindow/llwindowwin32.cpp b/indra/llwindow/llwindowwin32.cpp
index b2b123f0da..714e07da31 100644
--- a/indra/llwindow/llwindowwin32.cpp
+++ b/indra/llwindow/llwindowwin32.cpp
@@ -784,7 +784,7 @@ LLWindowWin32::LLWindowWin32(LLWindowCallbacks* callbacks,
 	LLCoordScreen windowPos(x,y);
 	LLCoordScreen windowSize(window_rect.right - window_rect.left,
 							 window_rect.bottom - window_rect.top);
-	if (!switchContext(mFullscreen, windowSize, TRUE, &windowPos))
+	if (!switchContext(mFullscreen, windowSize, disable_vsync, &windowPos))
 	{
 		return;
 	}
@@ -1748,6 +1748,8 @@ const	S32   max_format  = (S32)num_formats - 1;
 		return FALSE;
 	}
 
+	LL_PROFILER_GPU_CONTEXT
+
 	if (!gGLManager.initGL())
 	{
 		close();
@@ -1764,6 +1766,7 @@ const	S32   max_format  = (S32)num_formats - 1;
 	else
 	{
 		LL_DEBUGS("Window") << "Keeping vertical sync" << LL_ENDL;
+        wglSwapIntervalEXT(1);
 	}
 
 	SetWindowLongPtr(mWindowHandle, GWLP_USERDATA, (LONG_PTR)this);
@@ -3474,6 +3477,8 @@ BOOL LLWindowWin32::resetDisplayResolution()
 void LLWindowWin32::swapBuffers()
 {
 	SwapBuffers(mhDC);
+
+    LL_PROFILER_GPU_COLLECT
 }
 
 
diff --git a/indra/newview/app_settings/settings.xml b/indra/newview/app_settings/settings.xml
index b1120c18b2..05c3fc3bfe 100644
--- a/indra/newview/app_settings/settings.xml
+++ b/indra/newview/app_settings/settings.xml
@@ -812,17 +812,6 @@
       <key>Value</key>
       <integer>0</integer>
     </map>
-    <key>FramePerSecondLimit</key>
-    <map>
-      <key>Comment</key>
-      <string>Controls upper limit of frames per second</string>
-      <key>Persist</key>
-      <integer>1</integer>
-      <key>Type</key>
-      <string>U32</string>
-      <key>Value</key>
-      <integer>120</integer>
-    </map>
     <key>BackgroundYieldTime</key>
     <map>
       <key>Comment</key>
@@ -3366,13 +3355,13 @@
     <key>DisableVerticalSync</key>
     <map>
       <key>Comment</key>
-      <string>Update frames as fast as possible (FALSE = update frames between display scans)</string>
+      <string>Update frames as fast as possible (FALSE = update frames between display scans).  Requires restart.</string>
       <key>Persist</key>
       <integer>1</integer>
       <key>Type</key>
       <string>Boolean</string>
       <key>Value</key>
-      <integer>1</integer>
+      <integer>0</integer>
     </map>
     <key>EnableGroupChatPopups</key>
     <map>
diff --git a/indra/newview/llappviewer.cpp b/indra/newview/llappviewer.cpp
index 94f0b31ecd..3d49a9eb78 100644
--- a/indra/newview/llappviewer.cpp
+++ b/indra/newview/llappviewer.cpp
@@ -694,8 +694,7 @@ LLAppViewer::LLAppViewer()
 	mPeriodicSlowFrame(LLCachedControl<bool>(gSavedSettings,"Periodic Slow Frame", FALSE)),
 	mFastTimerLogThread(NULL),
 	mSettingsLocationList(NULL),
-	mIsFirstRun(false),
-	mMinMicroSecPerFrame(0.f)
+	mIsFirstRun(false)
 {
 	if(NULL != sInstance)
 	{
@@ -1326,10 +1325,6 @@ bool LLAppViewer::init()
 	joystick = LLViewerJoystick::getInstance();
 	joystick->setNeedsReset(true);
 	/*----------------------------------------------------------------------*/
-
-	gSavedSettings.getControl("FramePerSecondLimit")->getSignal()->connect(boost::bind(&LLAppViewer::onChangeFrameLimit, this, _2));
-	onChangeFrameLimit(gSavedSettings.getLLSD("FramePerSecondLimit"));
-
 	// Load User's bindings
 	loadKeyBindings();
 
@@ -1358,7 +1353,8 @@ void LLAppViewer::initMaxHeapSize()
 }
 
 static LLTrace::BlockTimerStatHandle FTM_MESSAGES("System Messages");
-static LLTrace::BlockTimerStatHandle FTM_SLEEP("Sleep");
+static LLTrace::BlockTimerStatHandle FTM_SLEEP1("Sleep1");
+static LLTrace::BlockTimerStatHandle FTM_SLEEP2("Sleep2");
 static LLTrace::BlockTimerStatHandle FTM_YIELD("Yield");
 
 static LLTrace::BlockTimerStatHandle FTM_TEXTURE_CACHE("Texture Cache");
@@ -1420,13 +1416,17 @@ bool LLAppViewer::frame()
 
 bool LLAppViewer::doFrame()
 {
+	LL_RECORD_BLOCK_TIME(FTM_FRAME);
+
 	LLEventPump& mainloop(LLEventPumps::instance().obtain("mainloop"));
 	LLSD newFrame;
 
-	LL_RECORD_BLOCK_TIME(FTM_FRAME);
-	LLTrace::BlockTimer::processTimes();
-	LLTrace::get_frame_recording().nextPeriod();
-	LLTrace::BlockTimer::logStats();
+	{
+		LL_PROFILE_ZONE_NAMED( "df blocktimer" )
+		LLTrace::BlockTimer::processTimes();
+		LLTrace::get_frame_recording().nextPeriod();
+		LLTrace::BlockTimer::logStats();
+	}
 
 	LLTrace::get_thread_recorder()->pullFromChildren();
 
@@ -1434,6 +1434,7 @@ bool LLAppViewer::doFrame()
 	LL_CLEAR_CALLSTACKS();
 
 	{
+		LL_PROFILE_ZONE_NAMED( "df processMiscNativeEvents" )
 		pingMainloopTimeout("Main:MiscNativeWindowEvents");
 
 		if (gViewerWindow)
@@ -1442,7 +1443,10 @@ bool LLAppViewer::doFrame()
 			gViewerWindow->getWindow()->processMiscNativeEvents();
 		}
 
-		pingMainloopTimeout("Main:GatherInput");
+		{
+			LL_PROFILE_ZONE_NAMED( "df gatherInput" )
+			pingMainloopTimeout("Main:GatherInput");
+		}
 
 		if (gViewerWindow)
 		{
@@ -1466,13 +1470,21 @@ bool LLAppViewer::doFrame()
 			}
 		}
 
-		// canonical per-frame event
-		mainloop.post(newFrame);
-		// give listeners a chance to run
-		llcoro::suspend();
+		{
+			LL_PROFILE_ZONE_NAMED( "df mainloop" )
+			// canonical per-frame event
+			mainloop.post(newFrame);
+		}
+
+		{
+			LL_PROFILE_ZONE_NAMED( "df suspend" )
+			// give listeners a chance to run
+			llcoro::suspend();
+		}
 
 		if (!LLApp::isExiting())
 		{
+			LL_PROFILE_ZONE_NAMED( "df JoystickKeyboard" )
 			pingMainloopTimeout("Main:JoystickKeyboard");
 
 			// Scan keyboard for movement keys.  Command keys and typing
@@ -1493,12 +1505,18 @@ bool LLAppViewer::doFrame()
 
 			// Update state based on messages, user input, object idle.
 			{
-				pauseMainloopTimeout(); // *TODO: Remove. Messages shouldn't be stalling for 20+ seconds!
+				{
+					LL_PROFILE_ZONE_NAMED( "df pauseMainloopTimeout" )
+					pauseMainloopTimeout(); // *TODO: Remove. Messages shouldn't be stalling for 20+ seconds!
+				}
 
 				LL_RECORD_BLOCK_TIME(FTM_IDLE);
 				idle();
 
-				resumeMainloopTimeout();
+				{
+					LL_PROFILE_ZONE_NAMED( "df resumeMainloopTimeout" )
+					resumeMainloopTimeout();
+				}
 			}
 
 			if (gDoDisconnect && (LLStartUp::getStartupState() == STATE_STARTED))
@@ -1513,46 +1531,40 @@ bool LLAppViewer::doFrame()
 			// *TODO: Should we run display() even during gHeadlessClient?  DK 2011-02-18
 			if (!LLApp::isExiting() && !gHeadlessClient && gViewerWindow)
 			{
+				LL_PROFILE_ZONE_NAMED( "df Display" )
 				pingMainloopTimeout("Main:Display");
 				gGLActive = TRUE;
 
 				display();
 
-				static U64 last_call = 0;
-				if (!gTeleportDisplay)
 				{
-					// Frame/draw throttling, controlled by FramePerSecondLimit
-					U64 elapsed_time = LLTimer::getTotalTime() - last_call;
-					if (elapsed_time < mMinMicroSecPerFrame)
-					{
-						LL_RECORD_BLOCK_TIME(FTM_SLEEP);
-						// llclamp for when time function gets funky
-						U64 sleep_time = llclamp(mMinMicroSecPerFrame - elapsed_time, (U64)1, (U64)1e6);
-						micro_sleep(sleep_time, 0);
-					}
-				}
-				last_call = LLTimer::getTotalTime();
-
-				pingMainloopTimeout("Main:Snapshot");
-				LLFloaterSnapshot::update(); // take snapshots
+					LL_PROFILE_ZONE_NAMED( "df Snapshot" )
+					pingMainloopTimeout("Main:Snapshot");
+					LLFloaterSnapshot::update(); // take snapshots
 					LLFloaterOutfitSnapshot::update();
-				gGLActive = FALSE;
+					gGLActive = FALSE;
+				}
 			}
 		}
 
-		pingMainloopTimeout("Main:Sleep");
+		{
+			LL_PROFILE_ZONE_NAMED( "df pauseMainloopTimeout" )
+			pingMainloopTimeout("Main:Sleep");
 
-		pauseMainloopTimeout();
+			pauseMainloopTimeout();
+		}
 
 		// Sleep and run background threads
 		{
-			LL_RECORD_BLOCK_TIME(FTM_SLEEP);
+			//LL_RECORD_BLOCK_TIME(SLEEP2);
+			LL_PROFILE_ZONE_WARN( "Sleep2" )
 
 			// yield some time to the os based on command line option
 			static LLCachedControl<S32> yield_time(gSavedSettings, "YieldTime", -1);
 			if(yield_time >= 0)
 			{
 				LL_RECORD_BLOCK_TIME(FTM_YIELD);
+				LL_PROFILE_ZONE_NUM( yield_time )
 				ms_sleep(yield_time);
 			}
 
@@ -1615,16 +1627,22 @@ bool LLAppViewer::doFrame()
 				total_io_pending += io_pending ;
 
 			}
-			gMeshRepo.update() ;
+
+			{
+				LL_PROFILE_ZONE_NAMED( "df gMeshRepo" )
+				gMeshRepo.update() ;
+			}
 
 			if(!total_work_pending) //pause texture fetching threads if nothing to process.
 			{
+				LL_PROFILE_ZONE_NAMED( "df getTextureCache" )
 				LLAppViewer::getTextureCache()->pause();
 				LLAppViewer::getImageDecodeThread()->pause();
 				LLAppViewer::getTextureFetch()->pause();
 			}
 			if(!total_io_pending) //pause file threads if nothing to process.
 			{
+				LL_PROFILE_ZONE_NAMED( "df LLVFSThread" )
 				LLVFSThread::sLocal->pause();
 				LLLFSThread::sLocal->pause();
 			}
@@ -1632,6 +1650,7 @@ bool LLAppViewer::doFrame()
 			//texture fetching debugger
 			if(LLTextureFetchDebugger::isEnabled())
 			{
+				LL_PROFILE_ZONE_NAMED( "df tex_fetch_debugger_instance" )
 				LLFloaterTextureFetchDebugger* tex_fetch_debugger_instance =
 					LLFloaterReg::findTypedInstance<LLFloaterTextureFetchDebugger>("tex_fetch_debugger");
 				if(tex_fetch_debugger_instance)
@@ -1640,8 +1659,10 @@ bool LLAppViewer::doFrame()
 				}
 			}
 
-			resumeMainloopTimeout();
-
+			{
+				LL_PROFILE_ZONE_NAMED( "df resumeMainloopTimeout" )
+				resumeMainloopTimeout();
+			}
 			pingMainloopTimeout("Main:End");
 		}
 	}
@@ -2480,12 +2501,7 @@ bool LLAppViewer::initConfiguration()
 #ifndef	LL_RELEASE_FOR_DOWNLOAD
 	// provide developer build only overrides for these control variables that are not
 	// persisted to settings.xml
-	LLControlVariable* c = gSavedSettings.getControl("ShowConsoleWindow");
-	if (c)
-	{
-		c->setValue(true, false);
-	}
-	c = gSavedSettings.getControl("AllowMultipleViewers");
+	LLControlVariable* c = gSavedSettings.getControl("AllowMultipleViewers");
 	if (c)
 	{
 		c->setValue(true, false);
@@ -5613,19 +5629,6 @@ void LLAppViewer::disconnectViewer()
 	LLUrlEntryParcel::setDisconnected(gDisconnected);
 }
 
-bool LLAppViewer::onChangeFrameLimit(LLSD const & evt)
-{
-	if (evt.asInteger() > 0)
-	{
-		mMinMicroSecPerFrame = (U64)(1000000.0f / F32(evt.asInteger()));
-	}
-	else
-	{
-		mMinMicroSecPerFrame = 0;
-	}
-	return false;
-}
-
 void LLAppViewer::forceErrorLLError()
 {
    	LL_ERRS() << "This is a deliberate llerror" << LL_ENDL;
diff --git a/indra/newview/llappviewer.h b/indra/newview/llappviewer.h
index 0afb70958c..64e7caa36b 100644
--- a/indra/newview/llappviewer.h
+++ b/indra/newview/llappviewer.h
@@ -262,8 +262,6 @@ private:
     void sendLogoutRequest();
     void disconnectViewer();
 
-	bool onChangeFrameLimit(LLSD const & evt);
-
 	// *FIX: the app viewer class should be some sort of singleton, no?
 	// Perhaps its child class is the singleton and this should be an abstract base.
 	static LLAppViewer* sInstance; 
@@ -319,10 +317,7 @@ private:
 	// llcorehttp library init/shutdown helper
 	LLAppCoreHttp mAppCoreHttp;
 
-        bool mIsFirstRun;
-	U64 mMinMicroSecPerFrame; // frame throttling
-
-
+    bool mIsFirstRun;
 };
 
 // consts from viewer.h
diff --git a/indra/newview/llappviewerwin32.cpp b/indra/newview/llappviewerwin32.cpp
index 9daea515e5..fb53a7648d 100644
--- a/indra/newview/llappviewerwin32.cpp
+++ b/indra/newview/llappviewerwin32.cpp
@@ -323,6 +323,10 @@ int APIENTRY WINMAIN(HINSTANCE hInstance,
                      PWSTR     pCmdLine,
                      int       nCmdShow)
 {
+    // Call Tracy first thing to have it allocate memory
+    // https://github.com/wolfpld/tracy/issues/196
+    LL_PROFILER_FRAME_END
+
 	const S32 MAX_HEAPS = 255;
 	DWORD heap_enable_lfh_error[MAX_HEAPS];
 	S32 num_heaps = 0;
diff --git a/indra/newview/llcontrolavatar.cpp b/indra/newview/llcontrolavatar.cpp
index fab249f988..606e670805 100644
--- a/indra/newview/llcontrolavatar.cpp
+++ b/indra/newview/llcontrolavatar.cpp
@@ -241,7 +241,7 @@ void LLControlAvatar::matchVolumeTransform()
 			if (skin_info)
 			{
                 LL_DEBUGS("BindShape") << getFullname() << " bind shape " << skin_info->mBindShapeMatrix << LL_ENDL;
-                bind_rot = LLSkinningUtil::getUnscaledQuaternion(skin_info->mBindShapeMatrix);
+                bind_rot = LLSkinningUtil::getUnscaledQuaternion(LLMatrix4(skin_info->mBindShapeMatrix));
 			}
 #endif
 			setRotation(bind_rot*obj_rot);
diff --git a/indra/newview/lldrawable.cpp b/indra/newview/lldrawable.cpp
index 507af56cb0..30c4a21e1c 100644
--- a/indra/newview/lldrawable.cpp
+++ b/indra/newview/lldrawable.cpp
@@ -101,6 +101,8 @@ LLDrawable::LLDrawable(LLViewerObject *vobj, bool new_entry)
 
 void LLDrawable::init(bool new_entry)
 {
+	LL_PROFILE_ZONE_SCOPED
+
 	// mXform
 	mParent = NULL;
 	mRenderType = 0;
@@ -232,6 +234,8 @@ void LLDrawable::markDead()
 
 LLVOVolume* LLDrawable::getVOVolume() const
 {
+	LL_PROFILE_ZONE_SCOPED
+
 	LLViewerObject* objectp = mVObjp;
 	if ( !isDead() && objectp && (objectp->getPCode() == LL_PCODE_VOLUME))
 	{
@@ -335,6 +339,7 @@ static LLTrace::BlockTimerStatHandle FTM_ALLOCATE_FACE("Allocate Face");
 
 LLFace*	LLDrawable::addFace(LLFacePool *poolp, LLViewerTexture *texturep)
 {
+	LL_PROFILE_ZONE_SCOPED
 	
 	LLFace *face;
 	{
@@ -363,6 +368,8 @@ LLFace*	LLDrawable::addFace(LLFacePool *poolp, LLViewerTexture *texturep)
 
 LLFace*	LLDrawable::addFace(const LLTextureEntry *te, LLViewerTexture *texturep)
 {
+	LL_PROFILE_ZONE_SCOPED
+
 	LLFace *face;
 
 	{
@@ -387,6 +394,8 @@ LLFace*	LLDrawable::addFace(const LLTextureEntry *te, LLViewerTexture *texturep)
 
 LLFace*	LLDrawable::addFace(const LLTextureEntry *te, LLViewerTexture *texturep, LLViewerTexture *normalp)
 {
+	LL_PROFILE_ZONE_SCOPED
+
 	LLFace *face;
 	face = new LLFace(this, mVObjp);
 	
@@ -408,6 +417,8 @@ LLFace*	LLDrawable::addFace(const LLTextureEntry *te, LLViewerTexture *texturep,
 
 LLFace*	LLDrawable::addFace(const LLTextureEntry *te, LLViewerTexture *texturep, LLViewerTexture *normalp, LLViewerTexture *specularp)
 {
+	LL_PROFILE_ZONE_SCOPED
+
 	LLFace *face;
 	face = new LLFace(this, mVObjp);
 	
@@ -430,6 +441,8 @@ LLFace*	LLDrawable::addFace(const LLTextureEntry *te, LLViewerTexture *texturep,
 
 void LLDrawable::setNumFaces(const S32 newFaces, LLFacePool *poolp, LLViewerTexture *texturep)
 {
+	LL_PROFILE_ZONE_SCOPED
+
 	if (newFaces == (S32)mFaces.size())
 	{
 		return;
@@ -453,6 +466,8 @@ void LLDrawable::setNumFaces(const S32 newFaces, LLFacePool *poolp, LLViewerText
 
 void LLDrawable::setNumFacesFast(const S32 newFaces, LLFacePool *poolp, LLViewerTexture *texturep)
 {
+	LL_PROFILE_ZONE_SCOPED
+
 	if (newFaces <= (S32)mFaces.size() && newFaces >= (S32)mFaces.size()/2)
 	{
 		return;
@@ -476,6 +491,8 @@ void LLDrawable::setNumFacesFast(const S32 newFaces, LLFacePool *poolp, LLViewer
 
 void LLDrawable::mergeFaces(LLDrawable* src)
 {
+	LL_PROFILE_ZONE_SCOPED
+
 	U32 face_count = mFaces.size() + src->mFaces.size();
 
 	mFaces.reserve(face_count);
@@ -509,6 +526,8 @@ void LLDrawable::updateMaterial()
 
 void LLDrawable::makeActive()
 {		
+	LL_PROFILE_ZONE_SCOPED
+
 #if !LL_RELEASE_FOR_DOWNLOAD
 	if (mVObjp.notNull())
 	{
@@ -572,6 +591,8 @@ void LLDrawable::makeActive()
 
 void LLDrawable::makeStatic(BOOL warning_enabled)
 {
+	LL_PROFILE_ZONE_SCOPED
+
 	if (isState(ACTIVE) && 
 		!isState(ACTIVE_CHILD) && 
 		!mVObjp->isAttachment() && 
@@ -618,6 +639,8 @@ void LLDrawable::makeStatic(BOOL warning_enabled)
 // Returns "distance" between target destination and resulting xfrom
 F32 LLDrawable::updateXform(BOOL undamped)
 {
+	LL_PROFILE_ZONE_SCOPED
+
 	BOOL damped = !undamped;
 
 	// Position
@@ -769,6 +792,8 @@ void LLDrawable::moveUpdatePipeline(BOOL moved)
 
 void LLDrawable::movePartition()
 {
+	LL_PROFILE_ZONE_SCOPED
+
 	LLSpatialPartition* part = getSpatialPartition();
 	if (part)
 	{
@@ -813,6 +838,8 @@ BOOL LLDrawable::updateMoveUndamped()
 
 void LLDrawable::updatePartition()
 {
+	LL_PROFILE_ZONE_SCOPED
+
 	if (!getVOVolume())
 	{
 		movePartition();
@@ -830,6 +857,8 @@ void LLDrawable::updatePartition()
 
 BOOL LLDrawable::updateMoveDamped()
 {
+	LL_PROFILE_ZONE_SCOPED
+
 	F32 dist_squared = updateXform(FALSE);
 
 	mGeneration++;
@@ -853,6 +882,8 @@ BOOL LLDrawable::updateMoveDamped()
 
 void LLDrawable::updateDistance(LLCamera& camera, bool force_update)
 {
+	LL_PROFILE_ZONE_SCOPED
+
 	if (LLViewerCamera::sCurCameraID != LLViewerCamera::CAMERA_WORLD)
 	{
 		LL_WARNS() << "Attempted to update distance for non-world camera." << LL_ENDL;
@@ -957,6 +988,8 @@ void LLDrawable::updateTexture()
 
 BOOL LLDrawable::updateGeometry(BOOL priority)
 {
+	LL_PROFILE_ZONE_SCOPED
+
 	llassert(mVObjp.notNull());
 	BOOL res = mVObjp->updateGeometry(this);
 	return res;
@@ -1034,6 +1067,8 @@ const LLVector3& LLDrawable::getBounds(LLVector3& min, LLVector3& max) const
 
 void LLDrawable::updateSpatialExtents()
 {
+	LL_PROFILE_ZONE_SCOPED
+
 	if (mVObjp)
 	{
 		const LLVector4a* exts = getSpatialExtents();
@@ -1164,6 +1199,8 @@ void LLDrawable::setGroup(LLViewerOctreeGroup *groupp)
 
 LLSpatialPartition* LLDrawable::getSpatialPartition()
 { 
+	LL_PROFILE_ZONE_SCOPED
+
 	LLSpatialPartition* retval = NULL;
 	
 	if (!mVObjp || 
@@ -1247,6 +1284,8 @@ LLSpatialBridge::LLSpatialBridge(LLDrawable* root, BOOL render_by_group, U32 dat
 	LLDrawable(root->getVObj(), true),
 	LLSpatialPartition(data_mask, render_by_group, GL_STREAM_DRAW_ARB, regionp)
 {
+	LL_PROFILE_ZONE_SCOPED
+
 	mBridge = this;
 	mDrawable = root;
 	root->setSpatialBridge(this);
@@ -1292,6 +1331,8 @@ void LLSpatialBridge::destroyTree()
 
 void LLSpatialBridge::updateSpatialExtents()
 {
+	LL_PROFILE_ZONE_SCOPED
+
 	LLSpatialGroup* root = (LLSpatialGroup*) mOctree->getListener(0);
 	
 	{
@@ -1455,6 +1496,8 @@ public:
 
 void LLSpatialBridge::setVisible(LLCamera& camera_in, std::vector<LLDrawable*>* results, BOOL for_select)
 {
+	LL_PROFILE_ZONE_SCOPED
+
 	if (!gPipeline.hasRenderType(mDrawableType))
 	{
 		return;
@@ -1552,6 +1595,8 @@ void LLSpatialBridge::setVisible(LLCamera& camera_in, std::vector<LLDrawable*>*
 
 void LLSpatialBridge::updateDistance(LLCamera& camera_in, bool force_update)
 {
+	LL_PROFILE_ZONE_SCOPED
+
 	if (mDrawable == NULL)
 	{
 		markDead();
diff --git a/indra/newview/lldrawpoolavatar.cpp b/indra/newview/lldrawpoolavatar.cpp
index 687b13d2c8..02ab316256 100644
--- a/indra/newview/lldrawpoolavatar.cpp
+++ b/indra/newview/lldrawpoolavatar.cpp
@@ -121,6 +121,8 @@ LLDrawPoolAvatar::~LLDrawPoolAvatar()
 // virtual
 BOOL LLDrawPoolAvatar::isDead()
 {
+    LL_PROFILE_ZONE_SCOPED
+
     if (!LLFacePool::isDead())
     {
         return FALSE;
@@ -138,11 +140,15 @@ BOOL LLDrawPoolAvatar::isDead()
 
 S32 LLDrawPoolAvatar::getShaderLevel() const
 {
+    LL_PROFILE_ZONE_SCOPED
+
 	return (S32) LLViewerShaderMgr::instance()->getShaderLevel(LLViewerShaderMgr::SHADER_AVATAR);
 }
 
 void LLDrawPoolAvatar::prerender()
 {
+    LL_PROFILE_ZONE_SCOPED
+
 	mShaderLevel = LLViewerShaderMgr::instance()->getShaderLevel(LLViewerShaderMgr::SHADER_AVATAR);
 	
 	sShaderLevel = mShaderLevel;
@@ -163,12 +169,15 @@ void LLDrawPoolAvatar::prerender()
 		{
 			LLVOAvatar* avatarp = (LLVOAvatar *)facep->getDrawable()->getVObj().get();
 			updateRiggedVertexBuffers(avatarp);
+            updateSkinInfoMatrixPalettes(avatarp);
 		}
 	}
 }
 
 LLMatrix4& LLDrawPoolAvatar::getModelView()
 {
+    LL_PROFILE_ZONE_SCOPED
+
 	static LLMatrix4 ret;
 
 	ret.initRows(LLVector4(gGLModelView+0),
@@ -257,6 +266,8 @@ void LLDrawPoolAvatar::endDeferredPass(S32 pass)
 
 void LLDrawPoolAvatar::renderDeferred(S32 pass)
 {
+    LL_PROFILE_ZONE_SCOPED
+
 	render(pass);
 }
 
@@ -267,6 +278,8 @@ S32 LLDrawPoolAvatar::getNumPostDeferredPasses()
 
 void LLDrawPoolAvatar::beginPostDeferredPass(S32 pass)
 {
+    LL_PROFILE_ZONE_SCOPED
+
 	switch (pass)
 	{
 	case 0:
@@ -295,6 +308,8 @@ void LLDrawPoolAvatar::beginPostDeferredPass(S32 pass)
 
 void LLDrawPoolAvatar::beginPostDeferredAlpha()
 {
+    LL_PROFILE_ZONE_SCOPED
+
 	sSkipOpaque = TRUE;
 	sShaderLevel = mShaderLevel;
 	sVertexProgram = &gDeferredAvatarAlphaProgram;
@@ -309,6 +324,8 @@ void LLDrawPoolAvatar::beginPostDeferredAlpha()
 
 void LLDrawPoolAvatar::beginDeferredRiggedAlpha()
 {
+    LL_PROFILE_ZONE_SCOPED
+
 	sVertexProgram = &gDeferredSkinnedAlphaProgram;
 	gPipeline.bindDeferredShader(*sVertexProgram);
 	sDiffuseChannel = sVertexProgram->enableTexture(LLViewerShaderMgr::DIFFUSE_MAP);
@@ -317,6 +334,8 @@ void LLDrawPoolAvatar::beginDeferredRiggedAlpha()
 
 void LLDrawPoolAvatar::beginDeferredRiggedMaterialAlpha(S32 pass)
 {
+    LL_PROFILE_ZONE_SCOPED
+
 	switch (pass)
 	{
 	case 0: pass = 1; break;
@@ -343,6 +362,8 @@ void LLDrawPoolAvatar::beginDeferredRiggedMaterialAlpha(S32 pass)
 
 void LLDrawPoolAvatar::endDeferredRiggedAlpha()
 {
+    LL_PROFILE_ZONE_SCOPED
+
 	LLVertexBuffer::unbind();
 	gPipeline.unbindDeferredShader(*sVertexProgram);
 	sDiffuseChannel = 0;
@@ -353,6 +374,8 @@ void LLDrawPoolAvatar::endDeferredRiggedAlpha()
 
 void LLDrawPoolAvatar::endPostDeferredPass(S32 pass)
 {
+    LL_PROFILE_ZONE_SCOPED
+
 	switch (pass)
 	{
 	case 0:
@@ -381,6 +404,8 @@ void LLDrawPoolAvatar::endPostDeferredPass(S32 pass)
 
 void LLDrawPoolAvatar::endPostDeferredAlpha()
 {
+    LL_PROFILE_ZONE_SCOPED
+
 	// if we're in software-blending, remember to set the fence _after_ we draw so we wait till this rendering is done
 	sRenderingSkinned = FALSE;
 	sSkipOpaque = FALSE;
@@ -392,6 +417,8 @@ void LLDrawPoolAvatar::endPostDeferredAlpha()
 
 void LLDrawPoolAvatar::renderPostDeferred(S32 pass)
 {
+    LL_PROFILE_ZONE_SCOPED
+
 	static const S32 actual_pass[] =
 	{ //map post deferred pass numbers to what render() expects
 		2, //skinned
@@ -428,225 +455,235 @@ S32 LLDrawPoolAvatar::getNumShadowPasses()
 void LLDrawPoolAvatar::beginShadowPass(S32 pass)
 {
 	LL_RECORD_BLOCK_TIME(FTM_SHADOW_AVATAR);
+    {
+        LL_PROFILE_ZONE_SCOPED;
 
-	if (pass == SHADOW_PASS_AVATAR_OPAQUE)
-	{
-		sVertexProgram = &gDeferredAvatarShadowProgram;
-		
-		if ((sShaderLevel > 0))  // for hardware blending
-		{
-			sRenderingSkinned = TRUE;
-			sVertexProgram->bind();
-		}
+        if (pass == SHADOW_PASS_AVATAR_OPAQUE)
+        {
+            sVertexProgram = &gDeferredAvatarShadowProgram;
 
-		gGL.diffuseColor4f(1,1,1,1);
-	}
-    else if (pass == SHADOW_PASS_AVATAR_ALPHA_BLEND)
-	{
-		sVertexProgram = &gDeferredAvatarAlphaShadowProgram;
+            if ((sShaderLevel > 0))  // for hardware blending
+            {
+                sRenderingSkinned = TRUE;
+                sVertexProgram->bind();
+            }
 
-        // bind diffuse tex so we can reference the alpha channel...
-        S32 loc = sVertexProgram->getUniformLocation(LLViewerShaderMgr::DIFFUSE_MAP);
-        sDiffuseChannel = 0;
-        if (loc != -1)
+            gGL.diffuseColor4f(1, 1, 1, 1);
+        }
+        else if (pass == SHADOW_PASS_AVATAR_ALPHA_BLEND)
         {
-            sDiffuseChannel = sVertexProgram->enableTexture(LLViewerShaderMgr::DIFFUSE_MAP);
-		}
+            sVertexProgram = &gDeferredAvatarAlphaShadowProgram;
 
-		if ((sShaderLevel > 0))  // for hardware blending
-		{
-			sRenderingSkinned = TRUE;
-			sVertexProgram->bind();
-		}
+            // bind diffuse tex so we can reference the alpha channel...
+            S32 loc = sVertexProgram->getUniformLocation(LLViewerShaderMgr::DIFFUSE_MAP);
+            sDiffuseChannel = 0;
+            if (loc != -1)
+            {
+                sDiffuseChannel = sVertexProgram->enableTexture(LLViewerShaderMgr::DIFFUSE_MAP);
+            }
 
-		gGL.diffuseColor4f(1,1,1,1);
-	}
-    else if (pass == SHADOW_PASS_AVATAR_ALPHA_MASK)
-	{
-		sVertexProgram = &gDeferredAvatarAlphaMaskShadowProgram;
+            if ((sShaderLevel > 0))  // for hardware blending
+            {
+                sRenderingSkinned = TRUE;
+                sVertexProgram->bind();
+            }
 
-        // bind diffuse tex so we can reference the alpha channel...
-        S32 loc = sVertexProgram->getUniformLocation(LLViewerShaderMgr::DIFFUSE_MAP);
-        sDiffuseChannel = 0;
-        if (loc != -1)
+            gGL.diffuseColor4f(1, 1, 1, 1);
+        }
+        else if (pass == SHADOW_PASS_AVATAR_ALPHA_MASK)
         {
-            sDiffuseChannel = sVertexProgram->enableTexture(LLViewerShaderMgr::DIFFUSE_MAP);
-		}
+            sVertexProgram = &gDeferredAvatarAlphaMaskShadowProgram;
 
-		if ((sShaderLevel > 0))  // for hardware blending
-		{
-			sRenderingSkinned = TRUE;
-			sVertexProgram->bind();
-		}
+            // bind diffuse tex so we can reference the alpha channel...
+            S32 loc = sVertexProgram->getUniformLocation(LLViewerShaderMgr::DIFFUSE_MAP);
+            sDiffuseChannel = 0;
+            if (loc != -1)
+            {
+                sDiffuseChannel = sVertexProgram->enableTexture(LLViewerShaderMgr::DIFFUSE_MAP);
+            }
 
-		gGL.diffuseColor4f(1,1,1,1);
-	}
-    else if (pass == SHADOW_PASS_ATTACHMENT_ALPHA_BLEND)
-	{
-		sVertexProgram = &gDeferredAttachmentAlphaShadowProgram;
+            if ((sShaderLevel > 0))  // for hardware blending
+            {
+                sRenderingSkinned = TRUE;
+                sVertexProgram->bind();
+            }
 
-        // bind diffuse tex so we can reference the alpha channel...
-        S32 loc = sVertexProgram->getUniformLocation(LLViewerShaderMgr::DIFFUSE_MAP);
-        sDiffuseChannel = 0;
-        if (loc != -1)
+            gGL.diffuseColor4f(1, 1, 1, 1);
+        }
+        else if (pass == SHADOW_PASS_ATTACHMENT_ALPHA_BLEND)
         {
-            sDiffuseChannel = sVertexProgram->enableTexture(LLViewerShaderMgr::DIFFUSE_MAP);
-		}
-		
-		if ((sShaderLevel > 0))  // for hardware blending
-		{
-			sRenderingSkinned = TRUE;
-			sVertexProgram->bind();
-		}
+            sVertexProgram = &gDeferredAttachmentAlphaShadowProgram;
 
-		gGL.diffuseColor4f(1,1,1,1);
-	}
-    else if (pass == SHADOW_PASS_ATTACHMENT_ALPHA_MASK)
-	{
-		sVertexProgram = &gDeferredAttachmentAlphaMaskShadowProgram;
+            // bind diffuse tex so we can reference the alpha channel...
+            S32 loc = sVertexProgram->getUniformLocation(LLViewerShaderMgr::DIFFUSE_MAP);
+            sDiffuseChannel = 0;
+            if (loc != -1)
+            {
+                sDiffuseChannel = sVertexProgram->enableTexture(LLViewerShaderMgr::DIFFUSE_MAP);
+            }
+
+            if ((sShaderLevel > 0))  // for hardware blending
+            {
+                sRenderingSkinned = TRUE;
+                sVertexProgram->bind();
+            }
 
-        // bind diffuse tex so we can reference the alpha channel...
-		S32 loc = sVertexProgram->getUniformLocation(LLViewerShaderMgr::DIFFUSE_MAP);
-        sDiffuseChannel = 0;
-        if (loc != -1)
+            gGL.diffuseColor4f(1, 1, 1, 1);
+        }
+        else if (pass == SHADOW_PASS_ATTACHMENT_ALPHA_MASK)
         {
-            sDiffuseChannel = sVertexProgram->enableTexture(LLViewerShaderMgr::DIFFUSE_MAP);
-		}
+            sVertexProgram = &gDeferredAttachmentAlphaMaskShadowProgram;
 
-		if ((sShaderLevel > 0))  // for hardware blending
-		{
-			sRenderingSkinned = TRUE;
-			sVertexProgram->bind();
-		}
+            // bind diffuse tex so we can reference the alpha channel...
+            S32 loc = sVertexProgram->getUniformLocation(LLViewerShaderMgr::DIFFUSE_MAP);
+            sDiffuseChannel = 0;
+            if (loc != -1)
+            {
+                sDiffuseChannel = sVertexProgram->enableTexture(LLViewerShaderMgr::DIFFUSE_MAP);
+            }
 
-		gGL.diffuseColor4f(1,1,1,1);
-	}
-	else // SHADOW_PASS_ATTACHMENT_OPAQUE
-	{
-		sVertexProgram = &gDeferredAttachmentShadowProgram;
-		S32 loc = sVertexProgram->getUniformLocation(LLViewerShaderMgr::DIFFUSE_MAP);
-        sDiffuseChannel = 0;
-        if (loc != -1)
+            if ((sShaderLevel > 0))  // for hardware blending
+            {
+                sRenderingSkinned = TRUE;
+                sVertexProgram->bind();
+            }
+
+            gGL.diffuseColor4f(1, 1, 1, 1);
+        }
+        else // SHADOW_PASS_ATTACHMENT_OPAQUE
         {
-            sDiffuseChannel = sVertexProgram->enableTexture(LLViewerShaderMgr::DIFFUSE_MAP);
-		}
-		sVertexProgram->bind();
-	}
+            sVertexProgram = &gDeferredAttachmentShadowProgram;
+            S32 loc = sVertexProgram->getUniformLocation(LLViewerShaderMgr::DIFFUSE_MAP);
+            sDiffuseChannel = 0;
+            if (loc != -1)
+            {
+                sDiffuseChannel = sVertexProgram->enableTexture(LLViewerShaderMgr::DIFFUSE_MAP);
+            }
+            sVertexProgram->bind();
+        }
+    }
 }
 
 void LLDrawPoolAvatar::endShadowPass(S32 pass)
 {
 	LL_RECORD_BLOCK_TIME(FTM_SHADOW_AVATAR);
+    {
+        LL_PROFILE_ZONE_SCOPED;
 
-	if (pass == SHADOW_PASS_ATTACHMENT_OPAQUE)
-	{
-		LLVertexBuffer::unbind();
-	}
+        if (pass == SHADOW_PASS_ATTACHMENT_OPAQUE)
+        {
+            LLVertexBuffer::unbind();
+        }
 
-    if (sShaderLevel > 0)
-	{			
-		sVertexProgram->unbind();
-	}
-    sVertexProgram = NULL;
-    sRenderingSkinned = FALSE;
-    LLDrawPoolAvatar::sShadowPass = -1;
+        if (sShaderLevel > 0)
+        {
+            sVertexProgram->unbind();
+        }
+        sVertexProgram = NULL;
+        sRenderingSkinned = FALSE;
+        LLDrawPoolAvatar::sShadowPass = -1;
+    }
 }
 
 void LLDrawPoolAvatar::renderShadow(S32 pass)
 {
-	LL_RECORD_BLOCK_TIME(FTM_SHADOW_AVATAR);
+    LL_RECORD_BLOCK_TIME(FTM_SHADOW_AVATAR);
+    {
+        LL_PROFILE_ZONE_SCOPED;
 
-	if (mDrawFace.empty())
-	{
-		return;
-	}
+        if (mDrawFace.empty())
+        {
+            return;
+        }
 
-	const LLFace *facep = mDrawFace[0];
-	if (!facep->getDrawable())
-	{
-		return;
-	}
-	LLVOAvatar *avatarp = (LLVOAvatar *)facep->getDrawable()->getVObj().get();
+        const LLFace *facep = mDrawFace[0];
+        if (!facep->getDrawable())
+        {
+            return;
+        }
+        LLVOAvatar *avatarp = (LLVOAvatar *)facep->getDrawable()->getVObj().get();
 
-	if (avatarp->isDead() || avatarp->isUIAvatar() || avatarp->mDrawable.isNull())
-	{
-		return;
-	}
-	LLVOAvatar::AvatarOverallAppearance oa = avatarp->getOverallAppearance();
-	BOOL impostor = !LLPipeline::sImpostorRender && avatarp->isImpostor();
-	if (oa == LLVOAvatar::AOA_INVISIBLE ||
-		(impostor && oa == LLVOAvatar::AOA_JELLYDOLL))
-	{
-		// No shadows for jellydolled or invisible avs.
-		return;
-	}
-	
-    LLDrawPoolAvatar::sShadowPass = pass;
+        if (avatarp->isDead() || avatarp->isUIAvatar() || avatarp->mDrawable.isNull())
+        {
+            return;
+        }
+        LLVOAvatar::AvatarOverallAppearance oa = avatarp->getOverallAppearance();
+        BOOL impostor = !LLPipeline::sImpostorRender && avatarp->isImpostor();
+        if (impostor || (oa == LLVOAvatar::AOA_INVISIBLE))
+        {
+            // No shadows for impostored (including jellydolled) or invisible avs.
+            return;
+        }
 
-	if (pass == SHADOW_PASS_AVATAR_OPAQUE)
-	{
-        LLDrawPoolAvatar::sSkipTransparent = true;
-		avatarp->renderSkinned();
-        LLDrawPoolAvatar::sSkipTransparent = false;
-	}
-    else if (pass == SHADOW_PASS_AVATAR_ALPHA_BLEND)
-	{
-        LLDrawPoolAvatar::sSkipOpaque = true;
-		avatarp->renderSkinned();
-        LLDrawPoolAvatar::sSkipOpaque = false;
-	}
-    else if (pass == SHADOW_PASS_AVATAR_ALPHA_MASK)
-	{
-        LLDrawPoolAvatar::sSkipOpaque = true;
-		avatarp->renderSkinned();
-        LLDrawPoolAvatar::sSkipOpaque = false;
-	}
-    else if (pass == SHADOW_PASS_ATTACHMENT_ALPHA_BLEND) // rigged alpha
-	{
-        LLDrawPoolAvatar::sSkipOpaque = true;
-        renderRigged(avatarp, RIGGED_MATERIAL_ALPHA);
-        renderRigged(avatarp, RIGGED_MATERIAL_ALPHA_EMISSIVE);
-        renderRigged(avatarp, RIGGED_ALPHA);
-        renderRigged(avatarp, RIGGED_FULLBRIGHT_ALPHA);
-        renderRigged(avatarp, RIGGED_GLOW);
-        renderRigged(avatarp, RIGGED_SPECMAP_BLEND);
-        renderRigged(avatarp, RIGGED_NORMMAP_BLEND);
-        renderRigged(avatarp, RIGGED_NORMSPEC_BLEND);
-        LLDrawPoolAvatar::sSkipOpaque = false;
-	}
-    else if (pass == SHADOW_PASS_ATTACHMENT_ALPHA_MASK) // rigged alpha mask
-	{
-        LLDrawPoolAvatar::sSkipOpaque = true;
-        renderRigged(avatarp, RIGGED_MATERIAL_ALPHA_MASK);
-        renderRigged(avatarp, RIGGED_NORMMAP_MASK);
-        renderRigged(avatarp, RIGGED_SPECMAP_MASK);
-		renderRigged(avatarp, RIGGED_NORMSPEC_MASK);    
-        renderRigged(avatarp, RIGGED_GLOW);
-        LLDrawPoolAvatar::sSkipOpaque = false;
-	}
-	else // rigged opaque (SHADOW_PASS_ATTACHMENT_OPAQUE
-	{
-        LLDrawPoolAvatar::sSkipTransparent = true;
-		renderRigged(avatarp, RIGGED_MATERIAL);
-        renderRigged(avatarp, RIGGED_SPECMAP);
-		renderRigged(avatarp, RIGGED_SPECMAP_EMISSIVE);
-		renderRigged(avatarp, RIGGED_NORMMAP);		
-		renderRigged(avatarp, RIGGED_NORMMAP_EMISSIVE);
-		renderRigged(avatarp, RIGGED_NORMSPEC);
-		renderRigged(avatarp, RIGGED_NORMSPEC_EMISSIVE);
-		renderRigged(avatarp, RIGGED_SIMPLE);
-		renderRigged(avatarp, RIGGED_FULLBRIGHT);
-		renderRigged(avatarp, RIGGED_SHINY);
-		renderRigged(avatarp, RIGGED_FULLBRIGHT_SHINY);
-		renderRigged(avatarp, RIGGED_GLOW);
-		renderRigged(avatarp, RIGGED_DEFERRED_BUMP);
-		renderRigged(avatarp, RIGGED_DEFERRED_SIMPLE);
-        LLDrawPoolAvatar::sSkipTransparent = false;
-	}
+        LLDrawPoolAvatar::sShadowPass = pass;
+
+        if (pass == SHADOW_PASS_AVATAR_OPAQUE)
+        {
+            LLDrawPoolAvatar::sSkipTransparent = true;
+            avatarp->renderSkinned();
+            LLDrawPoolAvatar::sSkipTransparent = false;
+        }
+        else if (pass == SHADOW_PASS_AVATAR_ALPHA_BLEND)
+        {
+            LLDrawPoolAvatar::sSkipOpaque = true;
+            avatarp->renderSkinned();
+            LLDrawPoolAvatar::sSkipOpaque = false;
+        }
+        else if (pass == SHADOW_PASS_AVATAR_ALPHA_MASK)
+        {
+            LLDrawPoolAvatar::sSkipOpaque = true;
+            avatarp->renderSkinned();
+            LLDrawPoolAvatar::sSkipOpaque = false;
+        }
+        else if (pass == SHADOW_PASS_ATTACHMENT_ALPHA_BLEND) // rigged alpha
+        {
+            LLDrawPoolAvatar::sSkipOpaque = true;
+            renderRigged(avatarp, RIGGED_MATERIAL_ALPHA);
+            renderRigged(avatarp, RIGGED_MATERIAL_ALPHA_EMISSIVE);
+            renderRigged(avatarp, RIGGED_ALPHA);
+            renderRigged(avatarp, RIGGED_FULLBRIGHT_ALPHA);
+            renderRigged(avatarp, RIGGED_GLOW);
+            renderRigged(avatarp, RIGGED_SPECMAP_BLEND);
+            renderRigged(avatarp, RIGGED_NORMMAP_BLEND);
+            renderRigged(avatarp, RIGGED_NORMSPEC_BLEND);
+            LLDrawPoolAvatar::sSkipOpaque = false;
+        }
+        else if (pass == SHADOW_PASS_ATTACHMENT_ALPHA_MASK) // rigged alpha mask
+        {
+            LLDrawPoolAvatar::sSkipOpaque = true;
+            renderRigged(avatarp, RIGGED_MATERIAL_ALPHA_MASK);
+            renderRigged(avatarp, RIGGED_NORMMAP_MASK);
+            renderRigged(avatarp, RIGGED_SPECMAP_MASK);
+            renderRigged(avatarp, RIGGED_NORMSPEC_MASK);
+            renderRigged(avatarp, RIGGED_GLOW);
+            LLDrawPoolAvatar::sSkipOpaque = false;
+        }
+        else // rigged opaque (SHADOW_PASS_ATTACHMENT_OPAQUE
+        {
+            LLDrawPoolAvatar::sSkipTransparent = true;
+            renderRigged(avatarp, RIGGED_MATERIAL);
+            renderRigged(avatarp, RIGGED_SPECMAP);
+            renderRigged(avatarp, RIGGED_SPECMAP_EMISSIVE);
+            renderRigged(avatarp, RIGGED_NORMMAP);
+            renderRigged(avatarp, RIGGED_NORMMAP_EMISSIVE);
+            renderRigged(avatarp, RIGGED_NORMSPEC);
+            renderRigged(avatarp, RIGGED_NORMSPEC_EMISSIVE);
+            renderRigged(avatarp, RIGGED_SIMPLE);
+            renderRigged(avatarp, RIGGED_FULLBRIGHT);
+            renderRigged(avatarp, RIGGED_SHINY);
+            renderRigged(avatarp, RIGGED_FULLBRIGHT_SHINY);
+            renderRigged(avatarp, RIGGED_GLOW);
+            renderRigged(avatarp, RIGGED_DEFERRED_BUMP);
+            renderRigged(avatarp, RIGGED_DEFERRED_SIMPLE);
+            LLDrawPoolAvatar::sSkipTransparent = false;
+        }
+    }
 }
 
 S32 LLDrawPoolAvatar::getNumPasses()
 {
+    LL_PROFILE_ZONE_SCOPED
+
 	if (LLPipeline::sImpostorRender)
 	{
 		return 8;
@@ -660,6 +697,8 @@ S32 LLDrawPoolAvatar::getNumPasses()
 
 S32 LLDrawPoolAvatar::getNumDeferredPasses()
 {
+    LL_PROFILE_ZONE_SCOPED
+
 	if (LLPipeline::sImpostorRender)
 	{
 		return 19;
@@ -780,6 +819,8 @@ void LLDrawPoolAvatar::endRenderPass(S32 pass)
 
 void LLDrawPoolAvatar::beginImpostor()
 {
+    LL_PROFILE_ZONE_SCOPED
+
 	if (!LLPipeline::sReflectionRender)
 	{
 		LLVOAvatar::sRenderDistance = llclamp(LLVOAvatar::sRenderDistance, 16.f, 256.f);
@@ -798,6 +839,8 @@ void LLDrawPoolAvatar::beginImpostor()
 
 void LLDrawPoolAvatar::endImpostor()
 {
+    LL_PROFILE_ZONE_SCOPED
+
 	if (LLGLSLShader::sNoFixedFunction)
 	{
 		gImpostorProgram.unbind();
@@ -807,6 +850,8 @@ void LLDrawPoolAvatar::endImpostor()
 
 void LLDrawPoolAvatar::beginRigid()
 {
+    LL_PROFILE_ZONE_SCOPED
+
 	if (gPipeline.canUseVertexShaders())
 	{
 		if (LLPipeline::sUnderWaterRender)
@@ -840,6 +885,8 @@ void LLDrawPoolAvatar::beginRigid()
 
 void LLDrawPoolAvatar::endRigid()
 {
+    LL_PROFILE_ZONE_SCOPED
+
 	sShaderLevel = mShaderLevel;
 	if (sVertexProgram != NULL)
 	{
@@ -849,6 +896,8 @@ void LLDrawPoolAvatar::endRigid()
 
 void LLDrawPoolAvatar::beginDeferredImpostor()
 {
+    LL_PROFILE_ZONE_SCOPED
+
 	if (!LLPipeline::sReflectionRender)
 	{
 		LLVOAvatar::sRenderDistance = llclamp(LLVOAvatar::sRenderDistance, 16.f, 256.f);
@@ -865,6 +914,8 @@ void LLDrawPoolAvatar::beginDeferredImpostor()
 
 void LLDrawPoolAvatar::endDeferredImpostor()
 {
+    LL_PROFILE_ZONE_SCOPED
+
 	sShaderLevel = mShaderLevel;
 	sVertexProgram->disableTexture(LLViewerShaderMgr::DEFERRED_NORMAL);
 	sVertexProgram->disableTexture(LLViewerShaderMgr::SPECULAR_MAP);
@@ -876,6 +927,8 @@ void LLDrawPoolAvatar::endDeferredImpostor()
 
 void LLDrawPoolAvatar::beginDeferredRigid()
 {
+    LL_PROFILE_ZONE_SCOPED
+
 	sVertexProgram = &gDeferredNonIndexedDiffuseAlphaMaskNoColorProgram;
 	sDiffuseChannel = sVertexProgram->enableTexture(LLViewerShaderMgr::DIFFUSE_MAP);
 	sVertexProgram->bind();
@@ -892,6 +945,8 @@ void LLDrawPoolAvatar::beginDeferredRigid()
 
 void LLDrawPoolAvatar::endDeferredRigid()
 {
+    LL_PROFILE_ZONE_SCOPED
+
 	sShaderLevel = mShaderLevel;
 	sVertexProgram->disableTexture(LLViewerShaderMgr::DIFFUSE_MAP);
 	sVertexProgram->unbind();
@@ -901,6 +956,8 @@ void LLDrawPoolAvatar::endDeferredRigid()
 
 void LLDrawPoolAvatar::beginSkinned()
 {
+    LL_PROFILE_ZONE_SCOPED
+
 	if (sShaderLevel > 0)
 	{
 		if (LLPipeline::sUnderWaterRender)
@@ -967,6 +1024,8 @@ void LLDrawPoolAvatar::beginSkinned()
 
 void LLDrawPoolAvatar::endSkinned()
 {
+    LL_PROFILE_ZONE_SCOPED
+
 	// if we're in software-blending, remember to set the fence _after_ we draw so we wait till this rendering is done
 	if (sShaderLevel > 0)
 	{
@@ -991,6 +1050,8 @@ void LLDrawPoolAvatar::endSkinned()
 
 void LLDrawPoolAvatar::beginRiggedSimple()
 {
+    LL_PROFILE_ZONE_SCOPED
+
 	if (sShaderLevel > 0)
 	{
 		if (LLPipeline::sUnderWaterRender)
@@ -1031,6 +1092,8 @@ void LLDrawPoolAvatar::beginRiggedSimple()
 
 void LLDrawPoolAvatar::endRiggedSimple()
 {
+    LL_PROFILE_ZONE_SCOPED
+
 	LLVertexBuffer::unbind();
 	if (sShaderLevel > 0 || gPipeline.canUseVertexShaders())
 	{
@@ -1041,27 +1104,37 @@ void LLDrawPoolAvatar::endRiggedSimple()
 
 void LLDrawPoolAvatar::beginRiggedAlpha()
 {
+    LL_PROFILE_ZONE_SCOPED
+
 	beginRiggedSimple();
 }
 
 void LLDrawPoolAvatar::endRiggedAlpha()
 {
+    LL_PROFILE_ZONE_SCOPED
+
 	endRiggedSimple();
 }
 
 
 void LLDrawPoolAvatar::beginRiggedFullbrightAlpha()
 {
+    LL_PROFILE_ZONE_SCOPED
+
 	beginRiggedFullbright();
 }
 
 void LLDrawPoolAvatar::endRiggedFullbrightAlpha()
 {
+    LL_PROFILE_ZONE_SCOPED
+
 	endRiggedFullbright();
 }
 
 void LLDrawPoolAvatar::beginRiggedGlow()
 {
+    LL_PROFILE_ZONE_SCOPED
+
 	if (sShaderLevel > 0)
 	{
 		if (LLPipeline::sUnderWaterRender)
@@ -1108,11 +1181,15 @@ void LLDrawPoolAvatar::beginRiggedGlow()
 
 void LLDrawPoolAvatar::endRiggedGlow()
 {
+    LL_PROFILE_ZONE_SCOPED
+
 	endRiggedFullbright();
 }
 
 void LLDrawPoolAvatar::beginRiggedFullbright()
 {
+    LL_PROFILE_ZONE_SCOPED
+
 	if (sShaderLevel > 0)
 	{
 		if (LLPipeline::sUnderWaterRender)
@@ -1170,6 +1247,8 @@ void LLDrawPoolAvatar::beginRiggedFullbright()
 
 void LLDrawPoolAvatar::endRiggedFullbright()
 {
+    LL_PROFILE_ZONE_SCOPED
+
 	LLVertexBuffer::unbind();
 	if (sShaderLevel > 0 || gPipeline.canUseVertexShaders())
 	{
@@ -1180,6 +1259,8 @@ void LLDrawPoolAvatar::endRiggedFullbright()
 
 void LLDrawPoolAvatar::beginRiggedShinySimple()
 {
+    LL_PROFILE_ZONE_SCOPED
+
 	if (sShaderLevel > 0)
 	{
 		if (LLPipeline::sUnderWaterRender)
@@ -1220,6 +1301,8 @@ void LLDrawPoolAvatar::beginRiggedShinySimple()
 
 void LLDrawPoolAvatar::endRiggedShinySimple()
 {
+    LL_PROFILE_ZONE_SCOPED
+
 	LLVertexBuffer::unbind();
 	if (sShaderLevel > 0 || gPipeline.canUseVertexShaders())
 	{
@@ -1231,6 +1314,8 @@ void LLDrawPoolAvatar::endRiggedShinySimple()
 
 void LLDrawPoolAvatar::beginRiggedFullbrightShiny()
 {
+    LL_PROFILE_ZONE_SCOPED
+
 	if (sShaderLevel > 0)
 	{
 		if (LLPipeline::sUnderWaterRender)
@@ -1296,6 +1381,8 @@ void LLDrawPoolAvatar::beginRiggedFullbrightShiny()
 
 void LLDrawPoolAvatar::endRiggedFullbrightShiny()
 {
+    LL_PROFILE_ZONE_SCOPED
+
 	LLVertexBuffer::unbind();
 	if (sShaderLevel > 0 || gPipeline.canUseVertexShaders())
 	{
@@ -1308,6 +1395,8 @@ void LLDrawPoolAvatar::endRiggedFullbrightShiny()
 
 void LLDrawPoolAvatar::beginDeferredRiggedSimple()
 {
+    LL_PROFILE_ZONE_SCOPED
+
 	sVertexProgram = &gDeferredSkinnedDiffuseProgram;
 	sDiffuseChannel = 0;
 	sVertexProgram->bind();
@@ -1323,6 +1412,8 @@ void LLDrawPoolAvatar::beginDeferredRiggedSimple()
 
 void LLDrawPoolAvatar::endDeferredRiggedSimple()
 {
+    LL_PROFILE_ZONE_SCOPED
+
 	LLVertexBuffer::unbind();
 	sVertexProgram->unbind();
 	sVertexProgram = NULL;
@@ -1330,6 +1421,8 @@ void LLDrawPoolAvatar::endDeferredRiggedSimple()
 
 void LLDrawPoolAvatar::beginDeferredRiggedBump()
 {
+    LL_PROFILE_ZONE_SCOPED
+
 	sVertexProgram = &gDeferredSkinnedBumpProgram;
 	sVertexProgram->bind();
     if (LLPipeline::sRenderingHUDs)
@@ -1346,6 +1439,8 @@ void LLDrawPoolAvatar::beginDeferredRiggedBump()
 
 void LLDrawPoolAvatar::endDeferredRiggedBump()
 {
+    LL_PROFILE_ZONE_SCOPED
+
 	LLVertexBuffer::unbind();
 	sVertexProgram->disableTexture(LLViewerShaderMgr::BUMP_MAP);
 	sVertexProgram->disableTexture(LLViewerShaderMgr::DIFFUSE_MAP);
@@ -1357,6 +1452,8 @@ void LLDrawPoolAvatar::endDeferredRiggedBump()
 
 void LLDrawPoolAvatar::beginDeferredRiggedMaterial(S32 pass)
 {
+    LL_PROFILE_ZONE_SCOPED
+
 	if (pass == 1 ||
 		pass == 5 ||
 		pass == 9 ||
@@ -1387,6 +1484,8 @@ void LLDrawPoolAvatar::beginDeferredRiggedMaterial(S32 pass)
 
 void LLDrawPoolAvatar::endDeferredRiggedMaterial(S32 pass)
 {
+    LL_PROFILE_ZONE_SCOPED
+
 	if (pass == 1 ||
 		pass == 5 ||
 		pass == 9 ||
@@ -1407,6 +1506,8 @@ void LLDrawPoolAvatar::endDeferredRiggedMaterial(S32 pass)
 
 void LLDrawPoolAvatar::beginDeferredSkinned()
 {
+    LL_PROFILE_ZONE_SCOPED
+
 	sShaderLevel = mShaderLevel;
 	sVertexProgram = &gDeferredAvatarProgram;
 	sRenderingSkinned = TRUE;
@@ -1428,6 +1529,8 @@ void LLDrawPoolAvatar::beginDeferredSkinned()
 
 void LLDrawPoolAvatar::endDeferredSkinned()
 {
+    LL_PROFILE_ZONE_SCOPED
+
 	// if we're in software-blending, remember to set the fence _after_ we draw so we wait till this rendering is done
 	sRenderingSkinned = FALSE;
 	sVertexProgram->unbind();
@@ -1740,6 +1843,8 @@ void LLDrawPoolAvatar::getRiggedGeometry(
     LLVolume* volume,
     const LLVolumeFace& vol_face)
 {
+    LL_PROFILE_ZONE_SCOPED
+
     face->setGeomIndex(0);
     face->setIndicesIndex(0);
 
@@ -1794,7 +1899,7 @@ void LLDrawPoolAvatar::getRiggedGeometry(
 
 	U16 offset = 0;
 		
-	LLMatrix4 mat_vert = skin->mBindShapeMatrix;
+	LLMatrix4 mat_vert = LLMatrix4(skin->mBindShapeMatrix);
 	glh::matrix4f m((F32*) mat_vert.mMatrix);
 	m = m.inverse().transpose();
 		
@@ -1836,24 +1941,26 @@ void LLDrawPoolAvatar::getRiggedGeometry(
 void LLDrawPoolAvatar::updateRiggedFaceVertexBuffer(
     LLVOAvatar* avatar,
     LLFace* face,
-    const LLMeshSkinInfo* skin,
+    const LLVOVolume* vobj,
     LLVolume* volume,
     LLVolumeFace& vol_face)
 {
+    LL_PROFILE_ZONE_SCOPED;
+
 	LLVector4a* weights = vol_face.mWeights;
 	if (!weights)
 	{
 		return;
 	}
 
+    if (!vobj || vobj->isNoLOD())
+    {
+        return;
+    }
+
 	LLPointer<LLVertexBuffer> buffer = face->getVertexBuffer();
 	LLDrawable* drawable = face->getDrawable();
 
-	if (drawable->getVOVolume() && drawable->getVOVolume()->isNoLOD())
-	{
-		return;
-	}
-
     const U32 max_joints = LLSkinningUtil::getMaxJointCount();
 
 #if USE_SEPARATE_JOINT_INDICES_AND_WEIGHTS
@@ -1893,23 +2000,26 @@ void LLDrawPoolAvatar::updateRiggedFaceVertexBuffer(
     }
 #endif
 
-    // FIXME ugly const cast
-    LLSkinningUtil::scrubInvalidJoints(avatar, const_cast<LLMeshSkinInfo*>(skin));
-
-	U32 data_mask = face->getRiggedVertexBufferDataMask();
+    U32 data_mask = face->getRiggedVertexBufferDataMask();
+    const LLMeshSkinInfo* skin = nullptr;
 
-    if (!vol_face.mWeightsScrubbed)
-    {
-        LLSkinningUtil::scrubSkinWeights(weights, vol_face.mNumVertices, skin);
-        vol_face.mWeightsScrubbed = TRUE;
-    }
-	
 	if (buffer.isNull() || 
 		buffer->getTypeMask() != data_mask ||
 		buffer->getNumVerts() != vol_face.mNumVertices ||
 		buffer->getNumIndices() != vol_face.mNumIndices ||
 		(drawable && drawable->isState(LLDrawable::REBUILD_ALL)))
 	{
+        LL_PROFILE_ZONE_NAMED("Rigged VBO Rebuild");
+        skin = vobj->getSkinInfo();
+        // FIXME ugly const cast
+        LLSkinningUtil::scrubInvalidJoints(avatar, const_cast<LLMeshSkinInfo*>(skin));
+
+        if (!vol_face.mWeightsScrubbed)
+        {
+            LLSkinningUtil::scrubSkinWeights(weights, vol_face.mNumVertices, skin);
+            vol_face.mWeightsScrubbed = TRUE;
+        }
+
 		if (drawable && drawable->isState(LLDrawable::REBUILD_ALL))
 		{
             //rebuild EVERY face in the drawable, not just this one, to avoid missing drawable wide rebuild issues
@@ -1935,18 +2045,13 @@ void LLDrawPoolAvatar::updateRiggedFaceVertexBuffer(
 		}
 	}
 
-	if (buffer.isNull() ||
-		buffer->getNumVerts() != vol_face.mNumVertices ||
-		buffer->getNumIndices() != vol_face.mNumIndices)
-	{
-		// Allocation failed
-		return;
-	}
-
-	if (!buffer.isNull() && 
-		sShaderLevel <= 0 && 
-		face->mLastSkinTime < avatar->getLastSkinTime())
+	if (sShaderLevel <= 0 && 
+        face->mLastSkinTime < avatar->getLastSkinTime() &&
+        !buffer.isNull() &&
+        buffer->getNumVerts() == vol_face.mNumVertices &&
+        buffer->getNumIndices() == vol_face.mNumIndices)
 	{
+        LL_PROFILE_ZONE_NAMED("Software Skinning");
 		//perform software vertex skinning for this face
 		LLStrider<LLVector3> position;
 		LLStrider<LLVector3> normal;
@@ -1963,14 +2068,16 @@ void LLDrawPoolAvatar::updateRiggedFaceVertexBuffer(
 
 		LLVector4a* norm = has_normal ? (LLVector4a*) normal.get() : NULL;
 		
-		//build matrix palette
-		LLMatrix4a mat[LL_MAX_JOINTS_PER_MESH_OBJECT];
-        U32 count = LLSkinningUtil::getMeshJointCount(skin);
-        LLSkinningUtil::initSkinningMatrixPalette((LLMatrix4*)mat, count, skin, avatar);
-        LLSkinningUtil::checkSkinWeights(weights, buffer->getNumVerts(), skin);
+        if (skin == nullptr)
+        {
+            skin = vobj->getSkinInfo();
+        }
 
-		LLMatrix4a bind_shape_matrix;
-		bind_shape_matrix.loadu(skin->mBindShapeMatrix);
+        const MatrixPaletteCache& mpc = updateSkinInfoMatrixPalette(avatar, skin);
+        const LLMatrix4a* mat = &(mpc.mMatrixPalette[0]);
+
+        LLSkinningUtil::checkSkinWeights(weights, buffer->getNumVerts(), skin);
+		const LLMatrix4a& bind_shape_matrix = skin->mBindShapeMatrix;
 
 #if USE_SEPARATE_JOINT_INDICES_AND_WEIGHTS
         U8* joint_indices_cursor = vol_face.mJointIndices;
@@ -2038,6 +2145,8 @@ void LLDrawPoolAvatar::updateRiggedFaceVertexBuffer(
 
 void LLDrawPoolAvatar::renderRigged(LLVOAvatar* avatar, U32 type, bool glow)
 {
+    LL_PROFILE_ZONE_SCOPED
+
 	if (!avatar->shouldRenderRigged())
 	{
 		return;
@@ -2045,15 +2154,18 @@ void LLDrawPoolAvatar::renderRigged(LLVOAvatar* avatar, U32 type, bool glow)
 
 	stop_glerror();
 
+    const LLMeshSkinInfo* lastSkin = nullptr;
+
 	for (U32 i = 0; i < mRiggedFace[type].size(); ++i)
 	{
+        LL_PROFILE_ZONE_NAMED("Render Rigged Face");
 		LLFace* face = mRiggedFace[type][i];
 
         S32 offset = face->getIndicesStart();
 		U32 count = face->getIndicesCount();
 
         U16 start = face->getGeomStart();
-		U16 end = start + face->getGeomCount()-1;			
+		U16 end = start + face->getGeomCount()-1;
 
 		LLDrawable* drawable = face->getDrawable();
 		if (!drawable)
@@ -2175,52 +2287,32 @@ void LLDrawPoolAvatar::renderRigged(LLVOAvatar* avatar, U32 type, bool glow)
         }
 
 		if (buff)
-		{        
+		{
 			if (sShaderLevel > 0)
 			{
-                // upload matrix palette to shader
-				LLMatrix4a mat[LL_MAX_JOINTS_PER_MESH_OBJECT];
-				U32 count = LLSkinningUtil::getMeshJointCount(skin);
-                LLSkinningUtil::initSkinningMatrixPalette((LLMatrix4*)mat, count, skin, avatar);
-
-				stop_glerror();
-
-				F32 mp[LL_MAX_JOINTS_PER_MESH_OBJECT*12];
-
-				for (U32 i = 0; i < count; ++i)
-				{
-					F32* m = (F32*) mat[i].mMatrix[0].getF32ptr();
-
-					U32 idx = i*12;
-
-					mp[idx+0] = m[0];
-					mp[idx+1] = m[1];
-					mp[idx+2] = m[2];
-					mp[idx+3] = m[12];
+                if (lastSkin != skin) // <== only upload matrix palette to GL if the skininfo changed
+                {
+                    // upload matrix palette to shader
+                    const MatrixPaletteCache& mpc = updateSkinInfoMatrixPalette(avatar, skin);
+                    U32 count = mpc.mMatrixPalette.size();
 
-					mp[idx+4] = m[4];
-					mp[idx+5] = m[5];
-					mp[idx+6] = m[6];
-					mp[idx+7] = m[13];
+                    stop_glerror();
 
-					mp[idx+8] = m[8];
-					mp[idx+9] = m[9];
-					mp[idx+10] = m[10];
-					mp[idx+11] = m[14];
-				}
+                    LLDrawPoolAvatar::sVertexProgram->uniformMatrix3x4fv(LLViewerShaderMgr::AVATAR_MATRIX,
+                        count,
+                        FALSE,
+                        (GLfloat*) &(mpc.mGLMp[0]));
 
-				LLDrawPoolAvatar::sVertexProgram->uniformMatrix3x4fv(LLViewerShaderMgr::AVATAR_MATRIX, 
-					count,
-					FALSE,
-					(GLfloat*) mp);
-
-				stop_glerror();
+                    stop_glerror();
+                }
 			}
 			else
 			{
 				data_mask &= ~LLVertexBuffer::MAP_WEIGHT4;
 			}
 
+            lastSkin = skin;
+
 			/*if (glow)
 			{
 				gGL.diffuseColor4f(0,0,0,face->getTextureEntry()->getGlow());
@@ -2330,16 +2422,22 @@ void LLDrawPoolAvatar::renderRigged(LLVOAvatar* avatar, U32 type, bool glow)
 
 void LLDrawPoolAvatar::renderDeferredRiggedSimple(LLVOAvatar* avatar)
 {
+    LL_PROFILE_ZONE_SCOPED
+
 	renderRigged(avatar, RIGGED_DEFERRED_SIMPLE);
 }
 
 void LLDrawPoolAvatar::renderDeferredRiggedBump(LLVOAvatar* avatar)
 {
+    LL_PROFILE_ZONE_SCOPED
+
 	renderRigged(avatar, RIGGED_DEFERRED_BUMP);
 }
 
 void LLDrawPoolAvatar::renderDeferredRiggedMaterial(LLVOAvatar* avatar, S32 pass)
 {
+    LL_PROFILE_ZONE_SCOPED
+
 	renderRigged(avatar, pass);
 }
 
@@ -2352,8 +2450,10 @@ void LLDrawPoolAvatar::updateRiggedVertexBuffers(LLVOAvatar* avatar)
 	//update rigged vertex buffers
 	for (U32 type = 0; type < NUM_RIGGED_PASSES; ++type)
 	{
+        LL_PROFILE_ZONE_NAMED("Pass");
 		for (U32 i = 0; i < mRiggedFace[type].size(); ++i)
 		{
+            LL_PROFILE_ZONE_NAMED("Face");
 			LLFace* face = mRiggedFace[type][i];
 			LLDrawable* drawable = face->getDrawable();
 			if (!drawable)
@@ -2376,43 +2476,111 @@ void LLDrawPoolAvatar::updateRiggedVertexBuffers(LLVOAvatar* avatar)
 				continue;
 			}
 
-			const LLMeshSkinInfo* skin = vobj->getSkinInfo();
-			if (!skin)
-			{
-				continue;
-			}
-
 			stop_glerror();
 
 			LLVolumeFace& vol_face = volume->getVolumeFace(te);
-			updateRiggedFaceVertexBuffer(avatar, face, skin, volume, vol_face);
+			updateRiggedFaceVertexBuffer(avatar, face, vobj, volume, vol_face);
 		}
 	}
 }
 
+void LLDrawPoolAvatar::updateSkinInfoMatrixPalettes(LLVOAvatar* avatarp)
+{
+    LL_PROFILE_ZONE_SCOPED;
+    //evict matrix palettes from the cache that haven't been updated in 10 frames
+    for (matrix_palette_cache_t::iterator iter = mMatrixPaletteCache.begin(); iter != mMatrixPaletteCache.end(); )
+    {
+        if (gFrameCount - iter->second.mFrame > 10)
+        {
+            iter = mMatrixPaletteCache.erase(iter);
+        }
+        else
+        {
+            ++iter;
+        }
+    }
+}
+
+const LLDrawPoolAvatar::MatrixPaletteCache& LLDrawPoolAvatar::updateSkinInfoMatrixPalette(LLVOAvatar * avatarp, const LLMeshSkinInfo* skin)
+{
+    MatrixPaletteCache& entry = mMatrixPaletteCache[skin];
+
+    if (entry.mFrame != gFrameCount)
+    {
+        LL_PROFILE_ZONE_SCOPED;
+        entry.mFrame = gFrameCount;
+        //build matrix palette
+        U32 count = LLSkinningUtil::getMeshJointCount(skin);
+        entry.mMatrixPalette.resize(count);
+        LLSkinningUtil::initSkinningMatrixPalette(&(entry.mMatrixPalette[0]), count, skin, avatarp);
+
+        const LLMatrix4a* mat = &(entry.mMatrixPalette[0]);
+
+        stop_glerror();
+        
+        entry.mGLMp.resize(count * 12);
+
+        F32* mp = &(entry.mGLMp[0]);
+        
+        for (U32 i = 0; i < count; ++i)
+        {
+            F32* m = (F32*)mat[i].mMatrix[0].getF32ptr();
+
+            U32 idx = i * 12;
+
+            mp[idx + 0] = m[0];
+            mp[idx + 1] = m[1];
+            mp[idx + 2] = m[2];
+            mp[idx + 3] = m[12];
+
+            mp[idx + 4] = m[4];
+            mp[idx + 5] = m[5];
+            mp[idx + 6] = m[6];
+            mp[idx + 7] = m[13];
+
+            mp[idx + 8] = m[8];
+            mp[idx + 9] = m[9];
+            mp[idx + 10] = m[10];
+            mp[idx + 11] = m[14];
+        }
+    }
+
+    return entry;
+}
+
 void LLDrawPoolAvatar::renderRiggedSimple(LLVOAvatar* avatar)
 {
+    LL_PROFILE_ZONE_SCOPED
+
 	renderRigged(avatar, RIGGED_SIMPLE);
 }
 
 void LLDrawPoolAvatar::renderRiggedFullbright(LLVOAvatar* avatar)
 {
+    LL_PROFILE_ZONE_SCOPED
+
 	renderRigged(avatar, RIGGED_FULLBRIGHT);
 }
 
 	
 void LLDrawPoolAvatar::renderRiggedShinySimple(LLVOAvatar* avatar)
 {
+    LL_PROFILE_ZONE_SCOPED
+
 	renderRigged(avatar, RIGGED_SHINY);
 }
 
 void LLDrawPoolAvatar::renderRiggedFullbrightShiny(LLVOAvatar* avatar)
 {
+    LL_PROFILE_ZONE_SCOPED
+
 	renderRigged(avatar, RIGGED_FULLBRIGHT_SHINY);
 }
 
 void LLDrawPoolAvatar::renderRiggedAlpha(LLVOAvatar* avatar)
 {
+    LL_PROFILE_ZONE_SCOPED
+
 	if (!mRiggedFace[RIGGED_ALPHA].empty())
 	{
 		LLGLEnable blend(GL_BLEND);
@@ -2430,6 +2598,8 @@ void LLDrawPoolAvatar::renderRiggedAlpha(LLVOAvatar* avatar)
 
 void LLDrawPoolAvatar::renderRiggedFullbrightAlpha(LLVOAvatar* avatar)
 {
+    LL_PROFILE_ZONE_SCOPED
+
 	if (!mRiggedFace[RIGGED_FULLBRIGHT_ALPHA].empty())
 	{
 		LLGLEnable blend(GL_BLEND);
@@ -2447,6 +2617,8 @@ void LLDrawPoolAvatar::renderRiggedFullbrightAlpha(LLVOAvatar* avatar)
 
 void LLDrawPoolAvatar::renderRiggedGlow(LLVOAvatar* avatar)
 {
+    LL_PROFILE_ZONE_SCOPED
+
 	if (!mRiggedFace[RIGGED_GLOW].empty())
 	{
 		LLGLEnable blend(GL_BLEND);
@@ -2474,6 +2646,8 @@ void LLDrawPoolAvatar::renderRiggedGlow(LLVOAvatar* avatar)
 //-----------------------------------------------------------------------------
 LLViewerTexture *LLDrawPoolAvatar::getDebugTexture()
 {
+    LL_PROFILE_ZONE_SCOPED
+
 	if (mReferences.empty())
 	{
 		return NULL;
@@ -2497,6 +2671,8 @@ LLColor3 LLDrawPoolAvatar::getDebugColor() const
 
 void LLDrawPoolAvatar::addRiggedFace(LLFace* facep, U32 type)
 {
+    LL_PROFILE_ZONE_SCOPED
+
     llassert (facep->isState(LLFace::RIGGED));
     llassert(getType() == LLDrawPool::POOL_AVATAR || getType() == LLDrawPool::POOL_CONTROL_AV);
     if (facep->getPool() && facep->getPool() != this)
@@ -2519,6 +2695,8 @@ void LLDrawPoolAvatar::addRiggedFace(LLFace* facep, U32 type)
 
 void LLDrawPoolAvatar::removeRiggedFace(LLFace* facep)
 {
+    LL_PROFILE_ZONE_SCOPED
+
     llassert (facep->isState(LLFace::RIGGED));
     llassert(getType() == LLDrawPool::POOL_AVATAR || getType() == LLDrawPool::POOL_CONTROL_AV);
     if (facep->getPool() != this)
@@ -2556,7 +2734,7 @@ LLVertexBufferAvatar::LLVertexBufferAvatar()
 : LLVertexBuffer(sDataMask, 
 	GL_STREAM_DRAW_ARB) //avatars are always stream draw due to morph targets
 {
-
+    LL_PROFILE_ZONE_SCOPED
 }
 
 
diff --git a/indra/newview/lldrawpoolavatar.h b/indra/newview/lldrawpoolavatar.h
index 92a8538958..0c1ee2cced 100644
--- a/indra/newview/lldrawpoolavatar.h
+++ b/indra/newview/lldrawpoolavatar.h
@@ -28,14 +28,18 @@
 #define LL_LLDRAWPOOLAVATAR_H
 
 #include "lldrawpool.h"
+#include "llmodel.h"
+
+#include <unordered_map>
 
 class LLVOAvatar;
+class LLVOVolume;
 class LLGLSLShader;
 class LLFace;
-class LLMeshSkinInfo;
 class LLVolume;
 class LLVolumeFace;
 
+extern U32 gFrameCount;
 
 class LLDrawPoolAvatar : public LLFacePool
 {
@@ -253,11 +257,13 @@ typedef enum
 	void getRiggedGeometry(LLFace* face, LLPointer<LLVertexBuffer>& buffer, U32 data_mask, const LLMeshSkinInfo* skin, LLVolume* volume, const LLVolumeFace& vol_face);
 	void updateRiggedFaceVertexBuffer(LLVOAvatar* avatar,
 									  LLFace* facep, 
-									  const LLMeshSkinInfo* skin, 
+									  const LLVOVolume* vobj,
 									  LLVolume* volume,
 									  LLVolumeFace& vol_face);
 	void updateRiggedVertexBuffers(LLVOAvatar* avatar);
 
+    void updateSkinInfoMatrixPalettes(LLVOAvatar* avatarp);
+
 	void renderRigged(LLVOAvatar* avatar, U32 type, bool glow = false);
 	void renderRiggedSimple(LLVOAvatar* avatar);
 	void renderRiggedAlpha(LLVOAvatar* avatar);
@@ -277,6 +283,26 @@ typedef enum
 
 	std::vector<LLFace*> mRiggedFace[NUM_RIGGED_PASSES];
 
+    class MatrixPaletteCache
+    {
+    public:
+        U32 mFrame;
+        LLMeshSkinInfo::matrix_list_t mMatrixPalette;
+        
+        // Float array ready to be sent to GL
+        std::vector<F32> mGLMp;
+
+        MatrixPaletteCache() :
+            mFrame(gFrameCount-1)
+        {
+        }
+    };
+    
+    const MatrixPaletteCache& updateSkinInfoMatrixPalette(LLVOAvatar* avatarp, const LLMeshSkinInfo* skin);
+
+    typedef std::unordered_map<const LLMeshSkinInfo*, MatrixPaletteCache> matrix_palette_cache_t;
+    matrix_palette_cache_t mMatrixPaletteCache;
+
 	/*virtual*/ LLViewerTexture *getDebugTexture();
 	/*virtual*/ LLColor3 getDebugColor() const; // For AGP debug display
 
diff --git a/indra/newview/llexperiencelog.cpp b/indra/newview/llexperiencelog.cpp
index ee5d561927..c441fbc09f 100644
--- a/indra/newview/llexperiencelog.cpp
+++ b/indra/newview/llexperiencelog.cpp
@@ -149,10 +149,6 @@ std::string LLExperienceLog::getPermissionString( const LLSD& message, const std
 		{
 			buf.str(entry);
 		}
-		else
-		{
-			buf.str();
-		}
 	}
 
 	if(buf.str().empty())
diff --git a/indra/newview/llface.cpp b/indra/newview/llface.cpp
index 4a802ad9aa..34448a780d 100644
--- a/indra/newview/llface.cpp
+++ b/indra/newview/llface.cpp
@@ -241,6 +241,8 @@ void LLFace::setPool(LLFacePool* pool)
 
 void LLFace::setPool(LLFacePool* new_pool, LLViewerTexture *texturep)
 {
+	LL_PROFILE_ZONE_SCOPED
+
 	if (!new_pool)
 	{
 		LL_ERRS() << "Setting pool to null!" << LL_ENDL;
@@ -320,6 +322,8 @@ void LLFace::setSpecularMap(LLViewerTexture* tex)
 
 void LLFace::dirtyTexture()
 {
+	LL_PROFILE_ZONE_SCOPED
+
 	LLDrawable* drawablep = getDrawable();
 
 	if (mVObjp.notNull() && mVObjp->getVolume())
@@ -535,6 +539,8 @@ void LLFace::updateCenterAgent()
 
 void LLFace::renderSelected(LLViewerTexture *imagep, const LLColor4& color)
 {
+	LL_PROFILE_ZONE_SCOPED
+
 	if (mDrawablep == NULL || mDrawablep->getSpatialGroup() == NULL)
 	{
 		return;
@@ -585,6 +591,7 @@ void LLFace::renderSelected(LLViewerTexture *imagep, const LLColor4& color)
 						glTexCoordPointer(2, GL_FLOAT, 8, vol_face.mTexCoords);
 					}
 					gGL.syncMatrices();
+					LL_PROFILER_GPU_ZONEC( "gl.DrawElements", 0x00FF00 );
 					glDrawElements(GL_TRIANGLES, vol_face.mNumIndices, GL_UNSIGNED_SHORT, vol_face.mIndices);
 					glDisableClientState(GL_TEXTURE_COORD_ARRAY);
 				}
@@ -605,6 +612,8 @@ void LLFace::renderSelected(LLViewerTexture *imagep, const LLColor4& color)
 
 void renderFace(LLDrawable* drawable, LLFace *face)
 {
+	LL_PROFILE_ZONE_SCOPED
+
     LLVOVolume* vobj = drawable->getVOVolume();
     if (vobj)
     {
@@ -891,6 +900,8 @@ bool less_than_max_mag(const LLVector4a& vec)
 BOOL LLFace::genVolumeBBoxes(const LLVolume &volume, S32 f,
                              const LLMatrix4& mat_vert_in, BOOL global_volume)
 {
+	LL_PROFILE_ZONE_SCOPED
+
 	//get bounding box
 	if (mDrawablep->isState(LLDrawable::REBUILD_VOLUME | LLDrawable::REBUILD_POSITION | LLDrawable::REBUILD_RIGGED))
 	{
@@ -2375,6 +2386,8 @@ F32 LLFace::getTextureVirtualSize()
 
 BOOL LLFace::calcPixelArea(F32& cos_angle_to_view_dir, F32& radius)
 {
+	LL_PROFILE_ZONE_SCOPED
+
 	//VECTORIZE THIS
 	//get area of circle around face
 	LLVector4a center;
@@ -2654,6 +2667,8 @@ const LLMatrix4& LLFace::getRenderMatrix() const
 
 S32 LLFace::renderElements(const U16 *index_array) const
 {
+	LL_PROFILE_ZONE_SCOPED
+
 	S32 ret = 0;
 	
 	if (isState(GLOBAL))
@@ -2673,6 +2688,8 @@ S32 LLFace::renderElements(const U16 *index_array) const
 
 S32 LLFace::renderIndexed()
 {
+	LL_PROFILE_ZONE_SCOPED
+
 	if(mDrawablep == NULL || mDrawPoolp == NULL)
 	{
 		return 0;
@@ -2683,6 +2700,8 @@ S32 LLFace::renderIndexed()
 
 S32 LLFace::renderIndexed(U32 mask)
 {
+	LL_PROFILE_ZONE_SCOPED
+
 	if (mVertexBuffer.isNull())
 	{
 		return 0;
diff --git a/indra/newview/llfloatermodelpreview.cpp b/indra/newview/llfloatermodelpreview.cpp
index d9edd4dc30..0e54b66ea9 100644
--- a/indra/newview/llfloatermodelpreview.cpp
+++ b/indra/newview/llfloatermodelpreview.cpp
@@ -1357,31 +1357,31 @@ void LLFloaterModelPreview::clearAvatarTab()
 			}
 
 void LLFloaterModelPreview::updateAvatarTab(bool highlight_overrides)
-			{
+{
     S32 display_lod = mModelPreview->mPreviewLOD;
     if (mModelPreview->mModel[display_lod].empty())
-				{
+    {
         mSelectedJointName.clear();
         return;
-					}
+    }
 
     // Joints will be listed as long as they are listed in mAlternateBindMatrix
     // even if they are for some reason identical to defaults.
     // Todo: Are overrides always identical for all lods? They normally are, but there might be situations where they aren't.
     if (mJointOverrides[display_lod].empty())
-					{
+    {
         // populate map
         for (LLModelLoader::scene::iterator iter = mModelPreview->mScene[display_lod].begin(); iter != mModelPreview->mScene[display_lod].end(); ++iter)
-					{
+        {
             for (LLModelLoader::model_instance_list::iterator model_iter = iter->second.begin(); model_iter != iter->second.end(); ++model_iter)
-					{
+            {
                 LLModelInstance& instance = *model_iter;
                 LLModel* model = instance.mModel;
                 const LLMeshSkinInfo *skin = &model->mSkinInfo;
                 U32 joint_count = LLSkinningUtil::getMeshJointCount(skin);
                 U32 bind_count = highlight_overrides ? skin->mAlternateBindMatrix.size() : 0; // simply do not include overrides if data is not needed
                 if (bind_count > 0 && bind_count != joint_count)
-						{
+                {
                     std::ostringstream out;
                     out << "Invalid joint overrides for model " << model->getName();
                     out << ". Amount of joints " << joint_count;
@@ -1390,68 +1390,68 @@ void LLFloaterModelPreview::updateAvatarTab(bool highlight_overrides)
                     addStringToLog(out.str(), true);
                     // Disable overrides for this model
                     bind_count = 0;
-						}
+                }
                 if (bind_count > 0)
-						{
+                {
                     for (U32 j = 0; j < joint_count; ++j)
-							{
-                        const LLVector3& joint_pos = skin->mAlternateBindMatrix[j].getTranslation();
+                    {
+                        const LLVector3& joint_pos = LLVector3(skin->mAlternateBindMatrix[j].getTranslation());
                         LLJointOverrideData &data = mJointOverrides[display_lod][skin->mJointNames[j]];
 
                         LLJoint* pJoint = LLModelPreview::lookupJointByName(skin->mJointNames[j], mModelPreview);
                         if (pJoint)
-							{
+                        {
                             // see how voavatar uses aboveJointPosThreshold
                             if (pJoint->aboveJointPosThreshold(joint_pos))
-				{
+                            {
                                 // valid override
                                 if (data.mPosOverrides.size() > 0
                                     && (data.mPosOverrides.begin()->second - joint_pos).lengthSquared() > (LL_JOINT_TRESHOLD_POS_OFFSET * LL_JOINT_TRESHOLD_POS_OFFSET))
-					{
+                                {
                                     // File contains multiple meshes with conflicting joint offsets
                                     // preview may be incorrect, upload result might wary (depends onto
                                     // mesh_id that hasn't been generated yet).
                                     data.mHasConflicts = true;
-							}
+                                }
                                 data.mPosOverrides[model->getName()] = joint_pos;
-						}
-						else
-						{
+                            }
+                            else
+                            {
                                 // default value, it won't be accounted for by avatar
                                 data.mModelsNoOverrides.insert(model->getName());
-					}
-					}
-				}
-			}
-			else
-			{
+                            }
+                        }
+                    }
+                }
+                else
+                {
                     for (U32 j = 0; j < joint_count; ++j)
-				{				
+                    {
                         LLJointOverrideData &data = mJointOverrides[display_lod][skin->mJointNames[j]];
                         data.mModelsNoOverrides.insert(model->getName());
                     }
                 }
-			}
-		}
-	}
+            }
+        }
+    }
 
     LLPanel *panel = mTabContainer->getPanelByName("rigging_panel");
     LLScrollListCtrl *joints_list = panel->getChild<LLScrollListCtrl>("joints_list");
 
     if (joints_list->isEmpty())
-	{
+    {
         // Populate table
 
-    std::map<std::string, std::string> joint_alias_map;
+        std::map<std::string, std::string> joint_alias_map;
         mModelPreview->getJointAliases(joint_alias_map);
-    
+
         S32 conflicts = 0;
         joint_override_data_map_t::iterator joint_iter = mJointOverrides[display_lod].begin();
         joint_override_data_map_t::iterator joint_end = mJointOverrides[display_lod].end();
         while (joint_iter != joint_end)
-	{
+        {
             const std::string& listName = joint_iter->first;
-        
+
             LLScrollListItem::Params item_params;
             item_params.value(listName);
 
@@ -1459,38 +1459,38 @@ void LLFloaterModelPreview::updateAvatarTab(bool highlight_overrides)
             cell_params.font = LLFontGL::getFontSansSerif();
             cell_params.value = listName;
             if (joint_alias_map.find(listName) == joint_alias_map.end())
-	{
+            {
                 // Missing names
                 cell_params.color = LLColor4::red;
-	}
+            }
             if (joint_iter->second.mHasConflicts)
-	{
+            {
                 // Conflicts
                 cell_params.color = LLColor4::orange;
                 conflicts++;
-	}
+            }
             if (highlight_overrides && joint_iter->second.mPosOverrides.size() > 0)
-	{
+            {
                 cell_params.font.style = "BOLD";
-	}
+            }
 
             item_params.columns.add(cell_params);
 
             joints_list->addRow(item_params, ADD_BOTTOM);
             joint_iter++;
-	}
+        }
         joints_list->selectFirstItem();
         LLScrollListItem *selected = joints_list->getFirstSelected();
         if (selected)
-{
+        {
             mSelectedJointName = selected->getValue().asString();
-	}
+        }
 
         LLTextBox *joint_conf_descr = panel->getChild<LLTextBox>("conflicts_description");
         joint_conf_descr->setTextArg("[CONFLICTS]", llformat("%d", conflicts));
         joint_conf_descr->setTextArg("[JOINTS_COUNT]", llformat("%d", mJointOverrides[display_lod].size()));
-		}
-	}
+    }
+}
 
 //-----------------------------------------------------------------------------
 // addStringToLogTab()
diff --git a/indra/newview/llhudnametag.cpp b/indra/newview/llhudnametag.cpp
index 9d49c30a49..55a4b5a457 100644
--- a/indra/newview/llhudnametag.cpp
+++ b/indra/newview/llhudnametag.cpp
@@ -224,6 +224,7 @@ BOOL LLHUDNameTag::lineSegmentIntersect(const LLVector4a& start, const LLVector4
 
 void LLHUDNameTag::render()
 {
+    LL_PROFILE_ZONE_SCOPED;
 	if (sDisplayText)
 	{
 		LLGLDepthTest gls_depth(GL_TRUE, GL_FALSE);
@@ -731,6 +732,7 @@ void LLHUDNameTag::updateSize()
 
 void LLHUDNameTag::updateAll()
 {
+    LL_PROFILE_ZONE_SCOPED;
 	// iterate over all text objects, calculate their restoration forces,
 	// and add them to the visible set if they are on screen and close enough
 	sVisibleTextObjects.clear();
diff --git a/indra/newview/llmeshrepository.cpp b/indra/newview/llmeshrepository.cpp
index 2c1c1191da..8ac64dbd15 100644
--- a/indra/newview/llmeshrepository.cpp
+++ b/indra/newview/llmeshrepository.cpp
@@ -4046,28 +4046,26 @@ S32 LLMeshRepository::getActualMeshLOD(const LLVolumeParams& mesh_params, S32 lo
 const LLMeshSkinInfo* LLMeshRepository::getSkinInfo(const LLUUID& mesh_id, const LLVOVolume* requesting_obj)
 {
 	LL_RECORD_BLOCK_TIME(FTM_MESH_FETCH);
+    if (mesh_id.notNull())
+    {
+        skin_map::iterator iter = mSkinMap.find(mesh_id);
+        if (iter != mSkinMap.end())
+        {
+            return &(iter->second);
+        }
 
-	if (mesh_id.notNull())
-	{
-		skin_map::iterator iter = mSkinMap.find(mesh_id);
-		if (iter != mSkinMap.end())
-		{
-			return &(iter->second);
-		}
-		
-		//no skin info known about given mesh, try to fetch it
-		{
-			LLMutexLock lock(mMeshMutex);
-			//add volume to list of loading meshes
-			skin_load_map::iterator iter = mLoadingSkins.find(mesh_id);
-			if (iter == mLoadingSkins.end())
-			{ //no request pending for this skin info
-				mPendingSkinRequests.push(mesh_id);
-			}
-			mLoadingSkins[mesh_id].insert(requesting_obj->getID());
-		}
-	}
-
+        //no skin info known about given mesh, try to fetch it
+        {
+            LLMutexLock lock(mMeshMutex);
+            //add volume to list of loading meshes
+            skin_load_map::iterator iter = mLoadingSkins.find(mesh_id);
+            if (iter == mLoadingSkins.end())
+            { //no request pending for this skin info
+                mPendingSkinRequests.push(mesh_id);
+            }
+            mLoadingSkins[mesh_id].insert(requesting_obj->getID());
+        }
+    }
 	return NULL;
 }
 
diff --git a/indra/newview/llmeshrepository.h b/indra/newview/llmeshrepository.h
index 81e49cb1d8..c1698194cb 100644
--- a/indra/newview/llmeshrepository.h
+++ b/indra/newview/llmeshrepository.h
@@ -27,6 +27,7 @@
 #ifndef LL_MESH_REPOSITORY_H
 #define LL_MESH_REPOSITORY_H
 
+#include <unordered_map>
 #include "llassettype.h"
 #include "llmodel.h"
 #include "lluuid.h"
@@ -613,7 +614,7 @@ public:
 	typedef std::map<LLVolumeParams, std::set<LLUUID> > mesh_load_map;
 	mesh_load_map mLoadingMeshes[4];
 	
-	typedef std::map<LLUUID, LLMeshSkinInfo> skin_map;
+	typedef std::unordered_map<LLUUID, LLMeshSkinInfo> skin_map;
 	skin_map mSkinMap;
 
 	typedef std::map<LLUUID, LLModel::Decomposition*> decomposition_map;
diff --git a/indra/newview/llmodelpreview.cpp b/indra/newview/llmodelpreview.cpp
index a9e80ab5da..01bddd781d 100644
--- a/indra/newview/llmodelpreview.cpp
+++ b/indra/newview/llmodelpreview.cpp
@@ -591,7 +591,7 @@ void LLModelPreview::rebuildUploadData()
                 bool upload_skinweights = fmp && fmp->childGetValue("upload_skin").asBoolean();
                 if (upload_skinweights && high_lod_model->mSkinInfo.mJointNames.size() > 0)
                 {
-                    LLQuaternion bind_rot = LLSkinningUtil::getUnscaledQuaternion(high_lod_model->mSkinInfo.mBindShapeMatrix);
+                    LLQuaternion bind_rot = LLSkinningUtil::getUnscaledQuaternion(LLMatrix4(high_lod_model->mSkinInfo.mBindShapeMatrix));
                     LLQuaternion identity;
                     if (!bind_rot.isEqualEps(identity, 0.01))
                     {
@@ -3298,7 +3298,7 @@ BOOL LLModelPreview::render()
                                 LLJoint *joint = getPreviewAvatar()->getJoint(skin->mJointNums[j]);
                                 if (joint)
                                 {
-                                    const LLVector3& jointPos = skin->mAlternateBindMatrix[j].getTranslation();
+                                    const LLVector3& jointPos = LLVector3(skin->mAlternateBindMatrix[j].getTranslation());
                                     if (joint->aboveJointPosThreshold(jointPos))
                                     {
                                         bool override_changed;
@@ -3340,11 +3340,10 @@ BOOL LLModelPreview::render()
                             //build matrix palette
 
                             LLMatrix4a mat[LL_MAX_JOINTS_PER_MESH_OBJECT];
-                            LLSkinningUtil::initSkinningMatrixPalette((LLMatrix4*)mat, joint_count,
+                            LLSkinningUtil::initSkinningMatrixPalette(mat, joint_count,
                                 skin, getPreviewAvatar());
 
-                            LLMatrix4a bind_shape_matrix;
-                            bind_shape_matrix.loadu(skin->mBindShapeMatrix);
+                            const LLMatrix4a& bind_shape_matrix = skin->mBindShapeMatrix;
                             U32 max_joints = LLSkinningUtil::getMaxJointCount();
                             for (U32 j = 0; j < buffer->getNumVerts(); ++j)
                             {
diff --git a/indra/newview/llskinningutil.cpp b/indra/newview/llskinningutil.cpp
index e02b21f036..dc12de29fb 100644
--- a/indra/newview/llskinningutil.cpp
+++ b/indra/newview/llskinningutil.cpp
@@ -35,7 +35,6 @@
 #include "llrigginginfo.h"
 
 #define DEBUG_SKINNING  LL_DEBUG
-#define MAT_USE_SSE     1
 
 void dump_avatar_and_skin_state(const std::string& reason, LLVOAvatar *avatar, const LLMeshSkinInfo *skin)
 {
@@ -120,36 +119,26 @@ void LLSkinningUtil::scrubInvalidJoints(LLVOAvatar *avatar, LLMeshSkinInfo* skin
     skin->mInvalidJointsScrubbed = true;
 }
 
-#define MAT_USE_SSE 1
-
 void LLSkinningUtil::initSkinningMatrixPalette(
-    LLMatrix4* mat,
+    LLMatrix4a* mat,
     S32 count, 
     const LLMeshSkinInfo* skin,
     LLVOAvatar *avatar)
 {
+    LL_PROFILE_ZONE_SCOPED;
+
     initJointNums(const_cast<LLMeshSkinInfo*>(skin), avatar);
+
+    LLMatrix4a world[LL_CHARACTER_MAX_ANIMATED_JOINTS];
+
     for (U32 j = 0; j < count; ++j)
     {
         S32 joint_num = skin->mJointNums[j];
-        LLJoint *joint = NULL;
-        if (joint_num >= 0 && joint_num < LL_CHARACTER_MAX_ANIMATED_JOINTS)
-        {
-            joint = avatar->getJoint(joint_num);
-        }
-        llassert(joint);
+        LLJoint *joint = avatar->getJoint(joint_num);
+
         if (joint)
         {
-#ifdef MAT_USE_SSE
-            LLMatrix4a bind, world, res;
-            bind.loadu(skin->mInvBindMatrix[j]);
-            world.loadu(joint->getWorldMatrix());
-            matMul(bind,world,res);
-            memcpy(mat[j].mMatrix,res.mMatrix,16*sizeof(float));
-#else
-            mat[j] = skin->mInvBindMatrix[j];
-            mat[j] *= joint->getWorldMatrix();
-#endif
+            world[j] = joint->getWorldMatrix4a();
         }
         else
         {
@@ -159,16 +148,27 @@ void LLSkinningUtil::initSkinningMatrixPalette(
             // rendering  should  be disabled  unless  all joints  are
             // valid.  In other  cases of  skinned  rendering, invalid
             // joints should already have  been removed during scrubInvalidJoints().
-            LL_WARNS_ONCE("Avatar") << avatar->getFullname() 
-                                    << " rigged to invalid joint name " << skin->mJointNames[j] 
-                                    << " num " << skin->mJointNums[j] << LL_ENDL;
-            LL_WARNS_ONCE("Avatar") << avatar->getFullname() 
-                                    << " avatar build state: isBuilt() " << avatar->isBuilt() 
-                                    << " mInitFlags " << avatar->mInitFlags << LL_ENDL;
+            LL_WARNS_ONCE("Avatar") << avatar->getFullname()
+                << " rigged to invalid joint name " << skin->mJointNames[j]
+                << " num " << skin->mJointNums[j] << LL_ENDL;
+            LL_WARNS_ONCE("Avatar") << avatar->getFullname()
+                << " avatar build state: isBuilt() " << avatar->isBuilt()
+                << " mInitFlags " << avatar->mInitFlags << LL_ENDL;
 #endif
             dump_avatar_and_skin_state("initSkinningMatrixPalette joint not found", avatar, skin);
         }
     }
+
+    //NOTE: pointer striders used here as a micro-optimization over vector/array lookups
+    const LLMatrix4a* invBind = &(skin->mInvBindMatrix[0]);
+    const LLMatrix4a* w = world;
+    LLMatrix4a* m = mat;
+    LLMatrix4a* end = m + count;
+
+    while (m < end)
+    {
+        matMulUnsafe(*(invBind++), *(w++), *(m++));
+    }
 }
 
 void LLSkinningUtil::checkSkinWeights(LLVector4a* weights, U32 num_vertices, const LLMeshSkinInfo* skin)
@@ -212,7 +212,7 @@ void LLSkinningUtil::scrubSkinWeights(LLVector4a* weights, U32 num_vertices, con
 
 void LLSkinningUtil::getPerVertexSkinMatrix(
     F32* weights,
-    LLMatrix4a* mat,
+    const LLMatrix4a* mat,
     bool handle_bad_scale,
     LLMatrix4a& final_mat,
     U32 max_joints)
@@ -270,6 +270,7 @@ void LLSkinningUtil::initJointNums(LLMeshSkinInfo* skin, LLVOAvatar *avatar)
 {
     if (!skin->mJointNumsInitialized)
     {
+        LL_PROFILE_ZONE_SCOPED;
         for (U32 j = 0; j < skin->mJointNames.size(); ++j)
         {
     #if DEBUG_SKINNING     
@@ -357,13 +358,11 @@ void LLSkinningUtil::updateRiggingInfo(const LLMeshSkinInfo* skin, LLVOAvatar *a
                                 rig_info_tab[joint_num].setIsRiggedTo(true);
 
                                 // FIXME could precompute these matMuls.
-                                LLMatrix4a bind_shape;
-                                LLMatrix4a inv_bind;
+                                const LLMatrix4a& bind_shape = skin->mBindShapeMatrix;
+                                const LLMatrix4a& inv_bind = skin->mInvBindMatrix[joint_index];
                                 LLMatrix4a mat;
                                 LLVector4a pos_joint_space;
 
-                                bind_shape.loadu(skin->mBindShapeMatrix);
-                                inv_bind.loadu(skin->mInvBindMatrix[joint_index]);
                                 matMul(bind_shape, inv_bind, mat);
 
                                 mat.affineTransform(pos, pos_joint_space);
@@ -426,3 +425,4 @@ LLQuaternion LLSkinningUtil::getUnscaledQuaternion(const LLMatrix4& mat4)
     bind_rot.normalize();
     return bind_rot;
 }
+
diff --git a/indra/newview/llskinningutil.h b/indra/newview/llskinningutil.h
index efe7c85997..807418f983 100644
--- a/indra/newview/llskinningutil.h
+++ b/indra/newview/llskinningutil.h
@@ -42,10 +42,10 @@ namespace LLSkinningUtil
     S32 getMaxJointCount();
     U32 getMeshJointCount(const LLMeshSkinInfo *skin);
     void scrubInvalidJoints(LLVOAvatar *avatar, LLMeshSkinInfo* skin);
-    void initSkinningMatrixPalette(LLMatrix4* mat, S32 count, const LLMeshSkinInfo* skin, LLVOAvatar *avatar);
+    void initSkinningMatrixPalette(LLMatrix4a* mat, S32 count, const LLMeshSkinInfo* skin, LLVOAvatar *avatar);
     void checkSkinWeights(LLVector4a* weights, U32 num_vertices, const LLMeshSkinInfo* skin);
     void scrubSkinWeights(LLVector4a* weights, U32 num_vertices, const LLMeshSkinInfo* skin);
-    void getPerVertexSkinMatrix(F32* weights, LLMatrix4a* mat, bool handle_bad_scale, LLMatrix4a& final_mat, U32 max_joints);
+    void getPerVertexSkinMatrix(F32* weights, const LLMatrix4a* mat, bool handle_bad_scale, LLMatrix4a& final_mat, U32 max_joints);
 
     LL_FORCE_INLINE void getPerVertexSkinMatrixWithIndices(
         F32*        weights,
diff --git a/indra/newview/llspatialpartition.cpp b/indra/newview/llspatialpartition.cpp
index efa4a7fd66..253b6b9953 100644
--- a/indra/newview/llspatialpartition.cpp
+++ b/indra/newview/llspatialpartition.cpp
@@ -503,7 +503,9 @@ LLSpatialGroup* LLSpatialGroup::getParent()
 	}
 
 BOOL LLSpatialGroup::removeObject(LLDrawable *drawablep, BOOL from_octree)
-	{
+{
+	LL_PROFILE_ZONE_SCOPED
+
 	if(!drawablep)
 	{
 		return FALSE;
@@ -591,6 +593,8 @@ public:
 
 void LLSpatialGroup::setState(U32 state, S32 mode) 
 {
+	LL_PROFILE_ZONE_SCOPED
+
 	llassert(state <= LLSpatialGroup::STATE_MASK);
 	
 	if (mode > STATE_MODE_SINGLE)
@@ -638,6 +642,8 @@ public:
 
 void LLSpatialGroup::clearState(U32 state, S32 mode)
 {
+	LL_PROFILE_ZONE_SCOPED
+
 	llassert(state <= LLSpatialGroup::STATE_MASK);
 
 	if (mode > STATE_MODE_SINGLE)
@@ -724,6 +730,8 @@ void LLSpatialGroup::updateDistance(LLCamera &camera)
 
 F32 LLSpatialPartition::calcDistance(LLSpatialGroup* group, LLCamera& camera)
 {
+	LL_PROFILE_ZONE_SCOPED
+
 	LLVector4a eye;
 	LLVector4a origin;
 	origin.load3(camera.getOrigin().mV);
@@ -815,6 +823,8 @@ F32 LLSpatialGroup::getUpdateUrgency() const
 
 BOOL LLSpatialGroup::changeLOD()
 {
+	LL_PROFILE_ZONE_SCOPED
+
 	if (hasState(ALPHA_DIRTY | OBJECT_DIRTY))
 	{
 		//a rebuild is going to happen, update distance and LoD
@@ -907,6 +917,8 @@ void LLSpatialGroup::handleDestruction(const TreeNode* node)
 
 void LLSpatialGroup::handleChildAddition(const OctreeNode* parent, OctreeNode* child) 
 {
+	LL_PROFILE_ZONE_SCOPED
+
 	if (child->getListenerCount() == 0)
 	{
 		new LLSpatialGroup(child, getSpatialPartition());
@@ -2700,11 +2712,17 @@ void renderPhysicsShape(LLDrawable* drawable, LLVOVolume* volume)
 			glVertexPointer(3, GL_FLOAT, 16, phys_volume->mHullPoints);
 			gGL.diffuseColor4fv(line_color.mV);
 			gGL.syncMatrices();
-			glDrawElements(GL_TRIANGLES, phys_volume->mNumHullIndices, GL_UNSIGNED_SHORT, phys_volume->mHullIndices);
+			{
+				LL_PROFILER_GPU_ZONEC( "gl.DrawElements", 0x20FF20 )
+				glDrawElements(GL_TRIANGLES, phys_volume->mNumHullIndices, GL_UNSIGNED_SHORT, phys_volume->mHullIndices);
+			}
 			
 			gGL.diffuseColor4fv(color.mV);
 			glPolygonMode(GL_FRONT_AND_BACK, GL_FILL);
-			glDrawElements(GL_TRIANGLES, phys_volume->mNumHullIndices, GL_UNSIGNED_SHORT, phys_volume->mHullIndices);			
+			{
+				LL_PROFILER_GPU_ZONEC( "gl.DrawElements", 0x40FF40 )
+				glDrawElements(GL_TRIANGLES, phys_volume->mNumHullIndices, GL_UNSIGNED_SHORT, phys_volume->mHullIndices);
+			}
 		}
 		else
 		{
@@ -3222,6 +3240,7 @@ void renderRaycast(LLDrawable* drawablep)
 						gGL.diffuseColor4f(0,1,1,0.5f);
 						glVertexPointer(3, GL_FLOAT, sizeof(LLVector4a), face.mPositions);
 						gGL.syncMatrices();
+						LL_PROFILER_GPU_ZONEC( "gl.DrawElements", 0x60FF60 );
 						glDrawElements(GL_TRIANGLES, face.mNumIndices, GL_UNSIGNED_SHORT, face.mIndices);
 					}
 					
diff --git a/indra/newview/llsurface.cpp b/indra/newview/llsurface.cpp
index cb356726e6..ea36e1d7be 100644
--- a/indra/newview/llsurface.cpp
+++ b/indra/newview/llsurface.cpp
@@ -1214,6 +1214,7 @@ F32 LLSurface::getWaterHeight() const
 BOOL LLSurface::generateWaterTexture(const F32 x, const F32 y,
 									 const F32 width, const F32 height)
 {
+	LL_PROFILE_ZONE_SCOPED
 	if (!getWaterTexture())
 	{
 		return FALSE;
diff --git a/indra/newview/llsurfacepatch.cpp b/indra/newview/llsurfacepatch.cpp
index 5e056944e9..aeefcd6fb8 100644
--- a/indra/newview/llsurfacepatch.cpp
+++ b/indra/newview/llsurfacepatch.cpp
@@ -728,6 +728,7 @@ BOOL LLSurfacePatch::updateTexture()
 
 void LLSurfacePatch::updateGL()
 {
+	LL_PROFILE_ZONE_SCOPED
 	F32 meters_per_grid = getSurface()->getMetersPerGrid();
 	F32 grids_per_patch_edge = (F32)getSurface()->getGridsPerPatchEdge();
 
diff --git a/indra/newview/llviewerdisplay.cpp b/indra/newview/llviewerdisplay.cpp
index a590237440..33842497d1 100644
--- a/indra/newview/llviewerdisplay.cpp
+++ b/indra/newview/llviewerdisplay.cpp
@@ -208,9 +208,11 @@ void display_update_camera()
 // Write some stats to LL_INFOS()
 void display_stats()
 {
+	LL_PROFILE_ZONE_SCOPED
 	F32 fps_log_freq = gSavedSettings.getF32("FPSLogFrequency");
 	if (fps_log_freq > 0.f && gRecentFPSTime.getElapsedTimeF32() >= fps_log_freq)
 	{
+		LL_PROFILE_ZONE_NAMED("DS - FPS");
 		F32 fps = gRecentFrameCount / fps_log_freq;
 		LL_INFOS() << llformat("FPS: %.02f", fps) << LL_ENDL;
 		gRecentFrameCount = 0;
@@ -219,6 +221,7 @@ void display_stats()
 	F32 mem_log_freq = gSavedSettings.getF32("MemoryLogFrequency");
 	if (mem_log_freq > 0.f && gRecentMemoryTime.getElapsedTimeF32() >= mem_log_freq)
 	{
+		LL_PROFILE_ZONE_NAMED("DS - Memory");
 		gMemoryAllocated = U64Bytes(LLMemory::getCurrentRSS());
 		U32Megabytes memory = gMemoryAllocated;
 		LL_INFOS() << "MEMORY: " << memory << LL_ENDL;
@@ -228,6 +231,7 @@ void display_stats()
     F32 asset_storage_log_freq = gSavedSettings.getF32("AssetStorageLogFrequency");
     if (asset_storage_log_freq > 0.f && gAssetStorageLogTime.getElapsedTimeF32() >= asset_storage_log_freq)
     {
+		LL_PROFILE_ZONE_NAMED("DS - Asset Storage");
         gAssetStorageLogTime.reset();
         gAssetStorage->logAssetStorageInfo();
     }
@@ -630,6 +634,7 @@ void display(BOOL rebuild, F32 zoom_factor, int subfield, BOOL for_snapshot)
 	
 	if (!gDisconnected)
 	{
+		LL_PROFILE_ZONE_NAMED("display - 1");
 		LLAppViewer::instance()->pingMainloopTimeout("Display:Update");
 		if (gPipeline.hasRenderType(LLPipeline::RENDER_TYPE_HUD))
 		{ //don't draw hud objects in this frame
@@ -722,6 +727,7 @@ void display(BOOL rebuild, F32 zoom_factor, int subfield, BOOL for_snapshot)
 		LLAppViewer::instance()->pingMainloopTimeout("Display:Swap");
 		
 		{ 
+			LL_PROFILE_ZONE_NAMED("display - 2")
 			if (gResizeScreenTexture)
 			{
 				gResizeScreenTexture = FALSE;
@@ -777,6 +783,7 @@ void display(BOOL rebuild, F32 zoom_factor, int subfield, BOOL for_snapshot)
 
 		//if (!for_snapshot)
 		{
+			LL_PROFILE_ZONE_NAMED("display - 3")
 			LLAppViewer::instance()->pingMainloopTimeout("Display:Imagery");
 			gPipeline.generateWaterReflection(*LLViewerCamera::getInstance());
 			gPipeline.generateHighlight(*LLViewerCamera::getInstance());
@@ -825,7 +832,7 @@ void display(BOOL rebuild, F32 zoom_factor, int subfield, BOOL for_snapshot)
 				LLImageGL::deleteDeadTextures();
 				stop_glerror();
 			}*/
-			}
+		}
 
 		LLGLState::checkStates();
 		LLGLState::checkClientArrays();
@@ -840,6 +847,7 @@ void display(BOOL rebuild, F32 zoom_factor, int subfield, BOOL for_snapshot)
 		//
 		LLAppViewer::instance()->pingMainloopTimeout("Display:StateSort");
 		{
+			LL_PROFILE_ZONE_NAMED("display - 3")
 			LLViewerCamera::sCurCameraID = LLViewerCamera::CAMERA_WORLD;
 			gPipeline.stateSort(*LLViewerCamera::getInstance(), result);
 			stop_glerror();
@@ -948,6 +956,7 @@ void display(BOOL rebuild, F32 zoom_factor, int subfield, BOOL for_snapshot)
 		if (!(LLAppViewer::instance()->logoutRequestSent() && LLAppViewer::instance()->hasSavedFinalSnapshot())
 				&& !gRestoreGL)
 		{
+			LL_PROFILE_ZONE_NAMED("display - 4")
 			LLViewerCamera::sCurCameraID = LLViewerCamera::CAMERA_WORLD;
 
 			if (gSavedSettings.getBOOL("RenderDepthPrePass") && LLGLSLShader::sNoFixedFunction)
diff --git a/indra/newview/llviewerobjectlist.cpp b/indra/newview/llviewerobjectlist.cpp
index 63e48d1dd0..0b20556104 100644
--- a/indra/newview/llviewerobjectlist.cpp
+++ b/indra/newview/llviewerobjectlist.cpp
@@ -168,6 +168,8 @@ U64 LLViewerObjectList::getIndex(const U32 local_id,
 
 BOOL LLViewerObjectList::removeFromLocalIDTable(const LLViewerObject* objectp)
 {
+	LL_PROFILE_ZONE_SCOPED
+
 	if(objectp && objectp->getRegion())
 	{
 		U32 local_id = objectp->mLocalID;		
@@ -303,6 +305,8 @@ static LLTrace::BlockTimerStatHandle FTM_PROCESS_OBJECTS("Process Objects");
 
 LLViewerObject* LLViewerObjectList::processObjectUpdateFromCache(LLVOCacheEntry* entry, LLViewerRegion* regionp)
 {
+	LL_PROFILE_ZONE_SCOPED
+
 	LLDataPacker *cached_dpp = entry->getDP();
 
 	if (!cached_dpp)
@@ -848,6 +852,8 @@ static LLTrace::BlockTimerStatHandle FTM_IDLE_COPY("Idle Copy");
 
 void LLViewerObjectList::update(LLAgent &agent)
 {
+	LL_PROFILE_ZONE_SCOPED
+
 	// Update globals
 	LLViewerObject::setVelocityInterpolate( gSavedSettings.getBOOL("VelocityInterpolate") );
 	LLViewerObject::setPingInterpolate( gSavedSettings.getBOOL("PingInterpolate") );
@@ -1293,6 +1299,8 @@ void LLViewerObjectList::clearDebugText()
 
 void LLViewerObjectList::cleanupReferences(LLViewerObject *objectp)
 {
+	LL_PROFILE_ZONE_SCOPED
+
 	bool new_dead_object = true;
 	if (mDeadObjects.find(objectp->mID) != mDeadObjects.end())
 	{
@@ -1523,6 +1531,8 @@ void LLViewerObjectList::removeFromActiveList(LLViewerObject* objectp)
 
 void LLViewerObjectList::updateActive(LLViewerObject *objectp)
 {
+	LL_PROFILE_ZONE_SCOPED
+
 	if (objectp->isDead())
 	{
 		return; // We don't update dead objects!
@@ -1843,6 +1853,8 @@ void LLViewerObjectList::renderObjectBounds(const LLVector3 &center)
 
 void LLViewerObjectList::generatePickList(LLCamera &camera)
 {
+	LL_PROFILE_ZONE_SCOPED
+
 		LLViewerObject *objectp;
 		S32 i;
 		// Reset all of the GL names to zero.
@@ -2104,6 +2116,8 @@ LLViewerObject *LLViewerObjectList::replaceObject(const LLUUID &id, const LLPCod
 
 S32 LLViewerObjectList::findReferences(LLDrawable *drawablep) const
 {
+	LL_PROFILE_ZONE_SCOPED
+
 	LLViewerObject *objectp;
 	S32 num_refs = 0;
 	
@@ -2167,6 +2181,8 @@ void LLViewerObjectList::orphanize(LLViewerObject *childp, U32 parent_id, U32 ip
 
 void LLViewerObjectList::findOrphans(LLViewerObject* objectp, U32 ip, U32 port)
 {
+	LL_PROFILE_ZONE_SCOPED
+
 	if (objectp->isDead())
 	{
 		LL_WARNS() << "Trying to find orphans for dead obj " << objectp->mID 
diff --git a/indra/newview/llviewerparceloverlay.cpp b/indra/newview/llviewerparceloverlay.cpp
index 7c3dd00e1a..02f7bbeed8 100644
--- a/indra/newview/llviewerparceloverlay.cpp
+++ b/indra/newview/llviewerparceloverlay.cpp
@@ -847,6 +847,7 @@ void LLViewerParcelOverlay::setDirty()
 
 void LLViewerParcelOverlay::updateGL()
 {
+	LL_PROFILE_ZONE_SCOPED
 	updateOverlayTexture();
 }
 
diff --git a/indra/newview/llviewertexture.cpp b/indra/newview/llviewertexture.cpp
index ca01bb46aa..af55f68cd2 100644
--- a/indra/newview/llviewertexture.cpp
+++ b/indra/newview/llviewertexture.cpp
@@ -208,6 +208,7 @@ void  LLViewerTextureManager::findTextures(const LLUUID& id, std::vector<LLViewe
 
 LLViewerFetchedTexture* LLViewerTextureManager::findFetchedTexture(const LLUUID& id, S32 tex_type)
 {
+    LL_PROFILE_ZONE_SCOPED;
     return gTextureList.findImage(id, (ETexListType)tex_type);
 }
 
diff --git a/indra/newview/llviewertexturelist.cpp b/indra/newview/llviewertexturelist.cpp
index 561319ca5d..12495078e9 100644
--- a/indra/newview/llviewertexturelist.cpp
+++ b/indra/newview/llviewertexturelist.cpp
@@ -620,6 +620,7 @@ void LLViewerTextureList::findTexturesByID(const LLUUID &image_id, std::vector<L
 
 LLViewerFetchedTexture *LLViewerTextureList::findImage(const LLTextureKey &search_key)
 {
+    LL_PROFILE_ZONE_SCOPED;
     uuid_map_t::iterator iter = mUUIDMap.find(search_key);
     if (iter == mUUIDMap.end())
         return NULL;
diff --git a/indra/newview/llviewerwindow.cpp b/indra/newview/llviewerwindow.cpp
index 1d13a306ef..2157585364 100644
--- a/indra/newview/llviewerwindow.cpp
+++ b/indra/newview/llviewerwindow.cpp
@@ -5280,6 +5280,7 @@ void LLViewerWindow::setup3DRender()
 
 void LLViewerWindow::setup3DViewport(S32 x_offset, S32 y_offset)
 {
+	LL_PROFILE_ZONE_SCOPED
 	gGLViewport[0] = mWorldViewRectRaw.mLeft + x_offset;
 	gGLViewport[1] = mWorldViewRectRaw.mBottom + y_offset;
 	gGLViewport[2] = mWorldViewRectRaw.getWidth();
diff --git a/indra/newview/llvlcomposition.cpp b/indra/newview/llvlcomposition.cpp
index c63c5f6b23..46beac8255 100644
--- a/indra/newview/llvlcomposition.cpp
+++ b/indra/newview/llvlcomposition.cpp
@@ -254,6 +254,7 @@ BOOL LLVLComposition::generateComposition()
 BOOL LLVLComposition::generateTexture(const F32 x, const F32 y,
 									  const F32 width, const F32 height)
 {
+	LL_PROFILE_ZONE_SCOPED
 	llassert(mSurfacep);
 	llassert(x >= 0.f);
 	llassert(y >= 0.f);
diff --git a/indra/newview/llvoavatar.cpp b/indra/newview/llvoavatar.cpp
index e085a945a8..a588d05ff7 100644
--- a/indra/newview/llvoavatar.cpp
+++ b/indra/newview/llvoavatar.cpp
@@ -3136,6 +3136,8 @@ void LLVOAvatar::idleUpdateWindEffect()
 
 void LLVOAvatar::idleUpdateNameTag(const LLVector3& root_pos_last)
 {
+    LL_PROFILE_ZONE_SCOPED;
+
 	// update chat bubble
 	//--------------------------------------------------------------------
 	// draw text label over character's head
@@ -4884,6 +4886,8 @@ bool LLVOAvatar::shouldAlphaMask()
 //-----------------------------------------------------------------------------
 U32 LLVOAvatar::renderSkinned()
 {
+    LL_PROFILE_ZONE_SCOPED;
+
 	U32 num_indices = 0;
 
 	if (!mIsBuilt)
@@ -6158,27 +6162,29 @@ LLJoint *LLVOAvatar::getJoint( const std::string &name )
 LLJoint *LLVOAvatar::getJoint( S32 joint_num )
 {
     LLJoint *pJoint = NULL;
-    S32 collision_start = mNumBones;
-    S32 attachment_start = mNumBones + mNumCollisionVolumes;
-    if (joint_num>=attachment_start)
+    if (joint_num >= 0)
     {
-        // Attachment IDs start at 1
-        S32 attachment_id = joint_num - attachment_start + 1;
-        attachment_map_t::iterator iter = mAttachmentPoints.find(attachment_id);
-        if (iter != mAttachmentPoints.end())
+        if (joint_num < mNumBones)
         {
-            pJoint = iter->second;
+            pJoint = mSkeleton[joint_num];
+        }
+        else if (joint_num < mNumBones + mNumCollisionVolumes)
+        {
+            S32 collision_id = joint_num - mNumBones;
+            pJoint = &mCollisionVolumes[collision_id];
+        }
+        else
+        {
+            // Attachment IDs start at 1
+            S32 attachment_id = joint_num - (mNumBones + mNumCollisionVolumes) + 1;
+            attachment_map_t::iterator iter = mAttachmentPoints.find(attachment_id);
+            if (iter != mAttachmentPoints.end())
+            {
+                pJoint = iter->second;
+            }
         }
     }
-    else if (joint_num>=collision_start)
-    {
-        S32 collision_id = joint_num-collision_start;
-        pJoint = &mCollisionVolumes[collision_id];
-    }
-    else if (joint_num>=0)
-    {
-        pJoint = mSkeleton[joint_num];
-    }
+    
 	llassert(!pJoint || pJoint->getJointNum() == joint_num);
     return pJoint;
 }
@@ -6513,7 +6519,7 @@ void LLVOAvatar::addAttachmentOverridesForObject(LLViewerObject *vo, std::set<LL
 					LLJoint* pJoint = getJoint( lookingForJoint );
 					if (pJoint)
 					{   									
-						const LLVector3& jointPos = pSkinData->mAlternateBindMatrix[i].getTranslation();									
+						const LLVector3& jointPos = LLVector3(pSkinData->mAlternateBindMatrix[i].getTranslation());
                         if (pJoint->aboveJointPosThreshold(jointPos))
                         {
                             bool override_changed;
@@ -7117,6 +7123,7 @@ void LLVOAvatar::updateGL()
 {
 	if (mMeshTexturesDirty)
 	{
+		LL_PROFILE_ZONE_SCOPED
 		updateMeshTextures();
 		mMeshTexturesDirty = FALSE;
 	}
@@ -7862,6 +7869,8 @@ void LLVOAvatar::onGlobalColorChanged(const LLTexGlobalColor* global_color)
 // Do rigged mesh attachments display with this av?
 bool LLVOAvatar::shouldRenderRigged() const
 {
+    LL_PROFILE_ZONE_SCOPED;
+
 	if (getOverallAppearance() == AOA_NORMAL)
 	{
 		return true;
@@ -8353,6 +8362,7 @@ void LLVOAvatar::updateMeshVisibility()
 // virtual
 void LLVOAvatar::updateMeshTextures()
 {
+	LL_PROFILE_ZONE_SCOPED
 	static S32 update_counter = 0;
 	mBakedTextureDebugText.clear();
 	
@@ -10949,6 +10959,7 @@ void LLVOAvatar::updateOverallAppearanceAnimations()
 // Based on isVisuallyMuted(), but has 3 possible results.
 LLVOAvatar::AvatarOverallAppearance LLVOAvatar::getOverallAppearance() const
 {
+    LL_PROFILE_ZONE_SCOPED;
 	AvatarOverallAppearance result = AOA_NORMAL;
 
 	// Priority order (highest priority first)
diff --git a/indra/newview/llvoavatar.h b/indra/newview/llvoavatar.h
index 74ef589ca4..39adaab8ca 100644
--- a/indra/newview/llvoavatar.h
+++ b/indra/newview/llvoavatar.h
@@ -209,6 +209,11 @@ public:
 	virtual LLJoint*		getJoint(const std::string &name);
 	LLJoint*		        getJoint(S32 num);
 
+    //if you KNOW joint_num is a valid animated joint index, use getSkeletonJoint for efficiency
+    inline LLJoint* getSkeletonJoint(S32 joint_num) { return mSkeleton[joint_num]; }
+    inline size_t getSkeletonJointCount() const { return mSkeleton.size(); }
+
+
 	void 					addAttachmentOverridesForObject(LLViewerObject *vo, std::set<LLUUID>* meshes_seen = NULL, bool recursive = true);
 	void					removeAttachmentOverridesForObject(const LLUUID& mesh_id);
 	void					removeAttachmentOverridesForObject(LLViewerObject *vo);
diff --git a/indra/newview/llvosurfacepatch.cpp b/indra/newview/llvosurfacepatch.cpp
index 897bace4e1..b5560d3d79 100644
--- a/indra/newview/llvosurfacepatch.cpp
+++ b/indra/newview/llvosurfacepatch.cpp
@@ -218,6 +218,7 @@ void LLVOSurfacePatch::updateGL()
 {
 	if (mPatchp)
 	{
+		LL_PROFILE_ZONE_SCOPED
 		mPatchp->updateGL();
 	}
 }
diff --git a/indra/newview/llvovolume.cpp b/indra/newview/llvovolume.cpp
index f9b2285989..e5a4b0f374 100644
--- a/indra/newview/llvovolume.cpp
+++ b/indra/newview/llvovolume.cpp
@@ -4807,7 +4807,7 @@ void LLRiggedVolume::update(const LLMeshSkinInfo* skin, LLVOAvatar* avatar, cons
 
 	LLMatrix4a mat[kMaxJoints];
 	U32 maxJoints = LLSkinningUtil::getMeshJointCount(skin);
-    LLSkinningUtil::initSkinningMatrixPalette((LLMatrix4*)mat, maxJoints, skin, avatar);
+    LLSkinningUtil::initSkinningMatrixPalette(mat, maxJoints, skin, avatar);
 
     S32 rigged_vert_count = 0;
     S32 rigged_face_count = 0;
@@ -4823,8 +4823,7 @@ void LLRiggedVolume::update(const LLMeshSkinInfo* skin, LLVOAvatar* avatar, cons
 		if ( weight )
 		{
             LLSkinningUtil::checkSkinWeights(weight, dst_face.mNumVertices, skin);
-			LLMatrix4a bind_shape_matrix;
-			bind_shape_matrix.loadu(skin->mBindShapeMatrix);
+			const LLMatrix4a& bind_shape_matrix = skin->mBindShapeMatrix;
 
 			LLVector4a* pos = dst_face.mPositions;
 
diff --git a/indra/newview/llworld.cpp b/indra/newview/llworld.cpp
index a1a1db35d6..c7b0a2bfb4 100644
--- a/indra/newview/llworld.cpp
+++ b/indra/newview/llworld.cpp
@@ -1284,6 +1284,7 @@ void send_agent_pause()
 
 void send_agent_resume()
 {
+	LL_PROFILE_ZONE_SCOPED
 	// Note: used to check for LLWorld initialization before it became a singleton.
 	// Rather than just remove this check I'm changing it to assure that the message 
 	// system has been initialized. -MG
diff --git a/indra/newview/pipeline.cpp b/indra/newview/pipeline.cpp
index e7c2d4db39..d7acf2ec0e 100644
--- a/indra/newview/pipeline.cpp
+++ b/indra/newview/pipeline.cpp
@@ -1897,6 +1897,7 @@ void LLPipeline::createObject(LLViewerObject* vobj)
 
 void LLPipeline::resetFrameStats()
 {
+	LL_PROFILE_ZONE_SCOPED
 	assertInitialized();
 
 	sCompiles        = 0;
@@ -2125,6 +2126,7 @@ void LLPipeline::grabReferences(LLCullResult& result)
 
 void LLPipeline::clearReferences()
 {
+	LL_PROFILE_ZONE_SCOPED
 	sCull = NULL;
 	mGroupSaveQ1.clear();
 }
@@ -11146,6 +11148,7 @@ void LLPipeline::generateImpostor(LLVOAvatar* avatar)
 		if (LLPipeline::sRenderDeferred)
 		{
 			GLuint buff = GL_COLOR_ATTACHMENT0;
+			LL_PROFILER_GPU_ZONEC( "gl.DrawBuffersARB", 0x8000FF );
 			glDrawBuffersARB(1, &buff);
 		}
 
diff --git a/indra/test/test.cpp b/indra/test/test.cpp
index 87c4a8d8a3..bb48216b2b 100644
--- a/indra/test/test.cpp
+++ b/indra/test/test.cpp
@@ -112,6 +112,7 @@ public:
 
 	virtual void recordMessage(LLError::ELevel level, const std::string& message)
 	{
+        LL_PROFILE_ZONE_SCOPED
 		mFile << message << std::endl;
 	}