Automated merge with http://bitbucket.org/lindenlab/viewer-development

67 files changed, 5365 insertions, 3924 deletions

diff --git a/indra/llcommon/CMakeLists.txt b/indra/llcommon/CMakeLists.txt
index 66e2bc9095..28677b036d 100644
--- a/indra/llcommon/CMakeLists.txt
+++ b/indra/llcommon/CMakeLists.txt
@@ -53,7 +53,7 @@ set(llcommon_SOURCE_FILES
     lleventfilter.cpp
     llevents.cpp
     lleventtimer.cpp
-    llfasttimer_class.cpp
+    llfasttimer.cpp
     llfile.cpp
     llfindlocale.cpp
     llfixedbuffer.cpp
@@ -71,12 +71,13 @@ set(llcommon_SOURCE_FILES
     llmd5.cpp
     llmemory.cpp
     llmemorystream.cpp
-    llmemtype.cpp
     llmetrics.cpp
     llmetricperformancetester.cpp
     llmortician.cpp
+    llmutex.cpp
     lloptioninterface.cpp
     llptrto.cpp 
+    llpredicate.cpp
     llprocess.cpp
     llprocessor.cpp
     llqueuedthread.cpp
@@ -90,7 +91,6 @@ set(llcommon_SOURCE_FILES
     llsdutil.cpp
     llsecondlifeurls.cpp
     llsingleton.cpp
-    llstat.cpp
     llstacktrace.cpp
     llstreamqueue.cpp
     llstreamtools.cpp
@@ -100,6 +100,9 @@ set(llcommon_SOURCE_FILES
     llthread.cpp
     llthreadsafequeue.cpp
     lltimer.cpp
+    lltrace.cpp
+    lltracerecording.cpp
+    lltracethreadrecorder.cpp
     lluri.cpp
     lluuid.cpp
     llworkerthread.cpp
@@ -168,7 +171,6 @@ set(llcommon_HEADER_FILES
     lleventemitter.h
     llextendedstatus.h
     llfasttimer.h
-    llfasttimer_class.h
     llfile.h
     llfindlocale.h
     llfixedbuffer.h
@@ -197,13 +199,14 @@ set(llcommon_HEADER_FILES
     llmd5.h
     llmemory.h
     llmemorystream.h
-    llmemtype.h
     llmetrics.h
     llmetricperformancetester.h
     llmortician.h
+    llmutex.h
     llnametable.h
     lloptioninterface.h
     llpointer.h
+    llpredicate.h
     llpreprocessor.h
     llpriqueuemap.h
     llprocess.h
@@ -231,7 +234,6 @@ set(llcommon_HEADER_FILES
     llsortedvector.h
     llstack.h
     llstacktrace.h
-    llstat.h
     llstatenums.h
     llstl.h
     llstreamqueue.h
@@ -241,10 +243,15 @@ set(llcommon_HEADER_FILES
     llstringtable.h
     llsys.h
     llthread.h
+    llthreadlocalstorage.h
     llthreadsafequeue.h
     lltimer.h
+    lltrace.h
+    lltracerecording.h
+    lltracethreadrecorder.h
     lltreeiterators.h
     lltypeinfolookup.h
+    llunit.h
     lluri.h
     lluuid.h
     lluuidhashmap.h
@@ -337,6 +344,7 @@ if (LL_TESTS)
   LL_ADD_INTEGRATION_TEST(llstring "" "${test_libs}")
   LL_ADD_INTEGRATION_TEST(lltreeiterators "" "${test_libs}")
   LL_ADD_INTEGRATION_TEST(lluri "" "${test_libs}")
+  LL_ADD_INTEGRATION_TEST(llunits "" "${test_libs}")
   LL_ADD_INTEGRATION_TEST(reflection "" "${test_libs}")
   LL_ADD_INTEGRATION_TEST(stringize "" "${test_libs}")
   LL_ADD_INTEGRATION_TEST(lleventdispatcher "" "${test_libs}")
diff --git a/indra/llcommon/llallocator.cpp b/indra/llcommon/llallocator.cpp
index 87654b5b97..34fc28d8cc 100644
--- a/indra/llcommon/llallocator.cpp
+++ b/indra/llcommon/llallocator.cpp
@@ -35,28 +35,6 @@
 DECLARE_bool(heap_profile_use_stack_trace);
 //DECLARE_double(tcmalloc_release_rate);
 
-// static
-void LLAllocator::pushMemType(S32 type)
-{
-    if(isProfiling())
-    {
-    	PushMemType(type);
-    }
-}
-
-// static
-S32 LLAllocator::popMemType()
-{
-    if (isProfiling())
-    {
-    	return PopMemType();
-    }
-    else
-    {
-        return -1;
-    }
-}
-
 void LLAllocator::setProfilingEnabled(bool should_enable)
 {
     // NULL disables dumping to disk
@@ -94,17 +72,6 @@ std::string LLAllocator::getRawProfile()
 // stub implementations for when tcmalloc is disabled
 //
 
-// static
-void LLAllocator::pushMemType(S32 type)
-{
-}
-
-// static
-S32 LLAllocator::popMemType()
-{
-    return -1;
-}
-
 void LLAllocator::setProfilingEnabled(bool should_enable)
 {
 }
diff --git a/indra/llcommon/llallocator.h b/indra/llcommon/llallocator.h
index a91dd57d14..d26ad73c5b 100644
--- a/indra/llcommon/llallocator.h
+++ b/indra/llcommon/llallocator.h
@@ -29,16 +29,10 @@
 
 #include <string>
 
-#include "llmemtype.h"
 #include "llallocator_heap_profile.h"
 
 class LL_COMMON_API LLAllocator {
     friend class LLMemoryView;
-    friend class LLMemType;
-
-private:
-	static void pushMemType(S32 type);
-	static S32 popMemType();
 
 public:
     void setProfilingEnabled(bool should_enable);
diff --git a/indra/llcommon/llapr.cpp b/indra/llcommon/llapr.cpp
index d1c44c9403..d911f258b6 100644
--- a/indra/llcommon/llapr.cpp
+++ b/indra/llcommon/llapr.cpp
@@ -29,6 +29,7 @@
 #include "linden_common.h"
 #include "llapr.h"
 #include "apr_dso.h"
+#include "llthreadlocalstorage.h"
 
 apr_pool_t *gAPRPoolp = NULL; // Global APR memory pool
 LLVolatileAPRPool *LLAPRFile::sAPRFilePoolp = NULL ; //global volatile APR memory pool.
@@ -52,8 +53,10 @@ void ll_init_apr()
 
 	if(!LLAPRFile::sAPRFilePoolp)
 	{
-		LLAPRFile::sAPRFilePoolp = new LLVolatileAPRPool(FALSE) ;
+		LLAPRFile::sAPRFilePoolp = new LLVolatileAPRPool(FALSE);
 	}
+
+	LLThreadLocalPointerBase::initAllThreadLocalStorage();
 }
 
 
@@ -77,6 +80,9 @@ void ll_cleanup_apr()
 		apr_thread_mutex_destroy(gCallStacksLogMutexp);
 		gCallStacksLogMutexp = NULL;
 	}
+
+	LLThreadLocalPointerBase::destroyAllThreadLocalStorage();
+
 	if (gAPRPoolp)
 	{
 		apr_pool_destroy(gAPRPoolp);
@@ -84,7 +90,7 @@ void ll_cleanup_apr()
 	}
 	if (LLAPRFile::sAPRFilePoolp)
 	{
-		delete LLAPRFile::sAPRFilePoolp ;
+		delete LLAPRFile::sAPRFilePoolp ;	
 		LLAPRFile::sAPRFilePoolp = NULL ;
 	}
 	apr_terminate();
@@ -477,6 +483,77 @@ S32 LLAPRFile::seek(apr_seek_where_t where, S32 offset)
 }
 
 //
+//LLThreadLocalPointerBase
+//
+bool LLThreadLocalPointerBase::sInitialized = false;
+
+void LLThreadLocalPointerBase::set( void* value )
+{
+	llassert(sInitialized && mThreadKey);
+
+	apr_status_t result = apr_threadkey_private_set((void*)value, mThreadKey);
+	if (result != APR_SUCCESS)
+	{
+		ll_apr_warn_status(result);
+		llerrs << "Failed to set thread local data" << llendl;
+	}
+}
+
+void LLThreadLocalPointerBase::initStorage( )
+{
+	apr_status_t result = apr_threadkey_private_create(&mThreadKey, NULL, gAPRPoolp);
+	if (result != APR_SUCCESS)
+	{
+		ll_apr_warn_status(result);
+		llerrs << "Failed to allocate thread local data" << llendl;
+	}
+}
+
+void LLThreadLocalPointerBase::destroyStorage()
+{
+	if (sInitialized)
+	{
+		if (mThreadKey)
+		{
+			apr_status_t result = apr_threadkey_private_delete(mThreadKey);
+			if (result != APR_SUCCESS)
+			{
+				ll_apr_warn_status(result);
+				llerrs << "Failed to delete thread local data" << llendl;
+			}
+		}
+	}
+}
+
+void LLThreadLocalPointerBase::initAllThreadLocalStorage()
+{
+	if (!sInitialized)
+	{
+		for (LLInstanceTracker<LLThreadLocalPointerBase>::instance_iter it = beginInstances(), end_it = endInstances();
+			it != end_it;
+			++it)
+		{
+			(*it).initStorage();
+		}
+		sInitialized = true;
+	}
+}
+
+void LLThreadLocalPointerBase::destroyAllThreadLocalStorage()
+{
+	if (sInitialized)
+	{
+		//for (LLInstanceTracker<LLThreadLocalPointerBase>::instance_iter it = beginInstances(), end_it = endInstances();
+		//	it != end_it;
+		//	++it)
+		//{
+		//	(*it).destroyStorage();
+		//}
+		sInitialized = false;
+	}
+}
+
+//
 //*******************************************************************************************************************************
 //static components of LLAPRFile
 //
diff --git a/indra/llcommon/llapr.h b/indra/llcommon/llapr.h
index 034546c3f3..b3c9bfd58c 100644
--- a/indra/llcommon/llapr.h
+++ b/indra/llcommon/llapr.h
@@ -32,14 +32,6 @@
 #if LL_LINUX || LL_SOLARIS
 #include <sys/param.h>  // Need PATH_MAX in APR headers...
 #endif
-#if LL_WINDOWS
-	// Limit Windows API to small and manageable set.
-	// If you get undefined symbols, find the appropriate
-	// Windows header file and include that in your .cpp file.
-	#define WIN32_LEAN_AND_MEAN
-	#include <winsock2.h>
-	#include <windows.h>
-#endif
 
 #include <boost/noncopyable.hpp>
 
@@ -48,12 +40,26 @@
 #include "apr_getopt.h"
 #include "apr_signal.h"
 #include "apr_atomic.h"
+
 #include "llstring.h"
 
 extern LL_COMMON_API apr_thread_mutex_t* gLogMutexp;
 extern apr_thread_mutex_t* gCallStacksLogMutexp;
 
 struct apr_dso_handle_t;
+/**
+ * @brief Function which appropriately logs error or remains quiet on
+ * APR_SUCCESS.
+ * @return Returns <code>true</code> if status is an error condition.
+ */
+bool LL_COMMON_API ll_apr_warn_status(apr_status_t status);
+/// There's a whole other APR error-message function if you pass a DSO handle.
+bool LL_COMMON_API ll_apr_warn_status(apr_status_t status, apr_dso_handle_t* handle);
+
+void LL_COMMON_API ll_apr_assert_status(apr_status_t status);
+void LL_COMMON_API ll_apr_assert_status(apr_status_t status, apr_dso_handle_t* handle);
+
+extern "C" LL_COMMON_API apr_pool_t* gAPRPoolp; // Global APR memory pool
 
 /** 
  * @brief initialize the common apr constructs -- apr itself, the
@@ -163,14 +169,17 @@ public:
 	LLAtomic32<Type>(Type x) {apr_atomic_set32(&mData, apr_uint32_t(x)); };
 	~LLAtomic32<Type>() {};
 
-	operator const Type() { apr_uint32_t data = apr_atomic_read32(&mData); return Type(data); }
+	operator const Type() { return get(); }
 	Type operator =(const Type& x) { apr_atomic_set32(&mData, apr_uint32_t(x)); return Type(mData); }
 	void operator -=(Type x) { apr_atomic_sub32(&mData, apr_uint32_t(x)); }
 	void operator +=(Type x) { apr_atomic_add32(&mData, apr_uint32_t(x)); }
 	Type operator ++(int) { return apr_atomic_inc32(&mData); } // Type++
-	Type operator --(int) { return apr_atomic_dec32(&mData); } // approximately --Type (0 if final is 0, non-zero otherwise)
+	Type operator ++()	  { apr_atomic_inc32(&mData); return get();  } // ++Type
+	Type operator --(int) { const Type result(get()); apr_atomic_dec32(&mData); return result; } // Type-- 
+	Type operator --()	  { return apr_atomic_dec32(&mData); } // approximately --Type (0 if final is 0, non-zero otherwise)
 	
 private:
+	const Type get() { apr_uint32_t data = apr_atomic_read32(&mData); return Type(data); }
 	apr_uint32_t mData;
 };
 
@@ -255,18 +264,5 @@ public:
 //*******************************************************************************************************************************
 };
 
-/**
- * @brief Function which appropriately logs error or remains quiet on
- * APR_SUCCESS.
- * @return Returns <code>true</code> if status is an error condition.
- */
-bool LL_COMMON_API ll_apr_warn_status(apr_status_t status);
-/// There's a whole other APR error-message function if you pass a DSO handle.
-bool LL_COMMON_API ll_apr_warn_status(apr_status_t status, apr_dso_handle_t* handle);
-
-void LL_COMMON_API ll_apr_assert_status(apr_status_t status);
-void LL_COMMON_API ll_apr_assert_status(apr_status_t status, apr_dso_handle_t* handle);
-
-extern "C" LL_COMMON_API apr_pool_t* gAPRPoolp; // Global APR memory pool
 
 #endif // LL_LLAPR_H
diff --git a/indra/llcommon/llcommon.cpp b/indra/llcommon/llcommon.cpp
index 8be9e4f4de..c720df7555 100644
--- a/indra/llcommon/llcommon.cpp
+++ b/indra/llcommon/llcommon.cpp
@@ -29,6 +29,7 @@
 
 #include "llmemory.h"
 #include "llthread.h"
+#include "lltrace.h"
 
 //static
 BOOL LLCommon::sAprInitialized = FALSE;
@@ -44,15 +45,13 @@ void LLCommon::initClass()
 	}
 	LLTimer::initClass();
 	LLThreadSafeRefCount::initThreadSafeRefCount();
-// 	LLWorkerThread::initClass();
-// 	LLFrameCallbackManager::initClass();
+	LLTrace::init();
 }
 
 //static
 void LLCommon::cleanupClass()
 {
-// 	LLFrameCallbackManager::cleanupClass();
-// 	LLWorkerThread::cleanupClass();
+	LLTrace::cleanup();
 	LLThreadSafeRefCount::cleanupThreadSafeRefCount();
 	LLTimer::cleanupClass();
 	if (sAprInitialized)
diff --git a/indra/llcommon/lldate.cpp b/indra/llcommon/lldate.cpp
index 030ef6a3c7..2efe39e158 100644
--- a/indra/llcommon/lldate.cpp
+++ b/indra/llcommon/lldate.cpp
@@ -39,6 +39,7 @@
 
 #include "lltimer.h"
 #include "llstring.h"
+#include "llfasttimer.h"
 
 static const F64 DATE_EPOCH = 0.0;
 
@@ -48,18 +49,15 @@ static const F64 LL_APR_USEC_PER_SEC = 1000000.0;
 
 
 LLDate::LLDate() : mSecondsSinceEpoch(DATE_EPOCH)
-{
-}
+{}
 
 LLDate::LLDate(const LLDate& date) :
 	mSecondsSinceEpoch(date.mSecondsSinceEpoch)
-{
-}
+{}
 
-LLDate::LLDate(F64 seconds_since_epoch) :
-	mSecondsSinceEpoch(seconds_since_epoch)
-{
-}
+LLDate::LLDate(LLUnit<LLUnits::Seconds, F64> seconds_since_epoch) :
+	mSecondsSinceEpoch(seconds_since_epoch.value())
+{}
 
 LLDate::LLDate(const std::string& iso8601_date)
 {
diff --git a/indra/llcommon/lldate.h b/indra/llcommon/lldate.h
index 7ff8b550ad..b62a846147 100644
--- a/indra/llcommon/lldate.h
+++ b/indra/llcommon/lldate.h
@@ -33,6 +33,7 @@
 #include <string>
 
 #include "stdtypes.h"
+#include "llunit.h"
 
 /** 
  * @class LLDate
@@ -56,9 +57,9 @@ public:
 	/** 
 	 * @brief Construct a date from a seconds since epoch value.
 	 *
-	 * @pararm seconds_since_epoch The number of seconds since UTC epoch.
+	 * @param seconds_since_epoch The number of seconds since UTC epoch.
 	 */
-	LLDate(F64 seconds_since_epoch);
+	LLDate(LLUnit<LLUnits::Seconds, F64> seconds_since_epoch);
 
 	/** 
 	 * @brief Construct a date from a string representation
diff --git a/indra/llcommon/lldefs.h b/indra/llcommon/lldefs.h
index 5a4b8325f4..d57b9dccff 100644
--- a/indra/llcommon/lldefs.h
+++ b/indra/llcommon/lldefs.h
@@ -244,5 +244,8 @@ inline void llswap(LLDATATYPE& lhs, LLDATATYPE& rhs)
 	rhs = tmp;
 }
 
+#define LL_GLUE_IMPL(x, y) x##y
+#define LL_GLUE_TOKENS(x, y) LL_GLUE_IMPL(x, y)
+
 #endif // LL_LLDEFS_H
 
diff --git a/indra/llcommon/llerrorlegacy.h b/indra/llcommon/llerrorlegacy.h
index 37cee579cd..097a533b1a 100644
--- a/indra/llcommon/llerrorlegacy.h
+++ b/indra/llcommon/llerrorlegacy.h
@@ -29,6 +29,7 @@
 #define LL_LLERRORLEGACY_H
 
 #include "llpreprocessor.h"
+#include <boost/static_assert.hpp>
 
 /*
 	LEGACY -- DO NOT USE THIS STUFF ANYMORE
@@ -111,6 +112,14 @@ const int LL_ERR_PRICE_MISMATCH = -23018;
 #define llverify(func)			do {if (func) {}} while(0)
 #endif
 
+#ifdef LL_WINDOWS
+#define llstatic_assert(func, msg) static_assert(func, msg)
+#define llstatic_assert_template(type, func, msg) static_assert(func, msg)
+#else
+#define llstatic_assert(func, msg) BOOST_STATIC_ASSERT(func)
+#define llstatic_assert_template(type, func, msg) BOOST_STATIC_ASSERT(sizeof(type) != 0 && func);
+#endif
+
 // handy compile-time assert - enforce those template parameters! 
 #define cassert(expn) typedef char __C_ASSERT__[(expn)?1:-1]   /* Flawfinder: ignore */
 	//XXX: used in two places in llcommon/llskipmap.h
diff --git a/indra/llcommon/llevents.cpp b/indra/llcommon/llevents.cpp
index 0855180dcd..1c928b3db8 100644
--- a/indra/llcommon/llevents.cpp
+++ b/indra/llcommon/llevents.cpp
@@ -41,7 +41,6 @@
 #include <algorithm>
 // std headers
 #include <typeinfo>
-#include <cassert>
 #include <cmath>
 #include <cctype>
 // external library headers
diff --git a/indra/llcommon/llfasttimer.cpp b/indra/llcommon/llfasttimer.cpp
new file mode 100644
index 0000000000..e6233a094e
--- /dev/null
+++ b/indra/llcommon/llfasttimer.cpp
@@ -0,0 +1,447 @@
+/** 
+ * @file llfasttimer.cpp
+ * @brief Implementation of the fast timer.
+ *
+ * $LicenseInfo:firstyear=2004&license=viewerlgpl$
+ * Second Life Viewer Source Code
+ * Copyright (C) 2010, Linden Research, Inc.
+ * 
+ * This library is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation;
+ * version 2.1 of the License only.
+ * 
+ * This library is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ * 
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with this library; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA
+ * 
+ * Linden Research, Inc., 945 Battery Street, San Francisco, CA  94111  USA
+ * $/LicenseInfo$
+ */
+#include "linden_common.h"
+
+#include "llfasttimer.h"
+
+#include "llmemory.h"
+#include "llprocessor.h"
+#include "llsingleton.h"
+#include "lltreeiterators.h"
+#include "llsdserialize.h"
+#include "llunit.h"
+#include "llsd.h"
+#include "lltracerecording.h"
+#include "lltracethreadrecorder.h"
+
+#include <boost/bind.hpp>
+#include <queue>
+
+
+#if LL_WINDOWS
+#include "lltimer.h"
+#elif LL_LINUX || LL_SOLARIS
+#include <sys/time.h>
+#include <sched.h>
+#include "lltimer.h"
+#elif LL_DARWIN
+#include <sys/time.h>
+#include "lltimer.h"	// get_clock_count()
+#else 
+#error "architecture not supported"
+#endif
+
+namespace LLTrace
+{
+
+//////////////////////////////////////////////////////////////////////////////
+// statics
+
+bool        TimeBlock::sLog		     = false;
+std::string TimeBlock::sLogName         = "";
+bool        TimeBlock::sMetricLog       = false;
+
+#if LL_LINUX || LL_SOLARIS
+U64         TimeBlock::sClockResolution = 1000000000; // Nanosecond resolution
+#else
+U64         TimeBlock::sClockResolution = 1000000; // Microsecond resolution
+#endif
+
+static LLMutex*			sLogLock = NULL;
+static std::queue<LLSD> sLogQueue;
+
+
+// FIXME: move these declarations to the relevant modules
+
+// helper functions
+typedef LLTreeDFSPostIter<TimeBlock, TimeBlock::child_const_iter> timer_tree_bottom_up_iterator_t;
+
+static timer_tree_bottom_up_iterator_t begin_timer_tree_bottom_up(TimeBlock& id) 
+{ 
+	return timer_tree_bottom_up_iterator_t(&id, 
+							boost::bind(boost::mem_fn(&TimeBlock::beginChildren), _1), 
+							boost::bind(boost::mem_fn(&TimeBlock::endChildren), _1));
+}
+
+static timer_tree_bottom_up_iterator_t end_timer_tree_bottom_up() 
+{ 
+	return timer_tree_bottom_up_iterator_t(); 
+}
+
+typedef LLTreeDFSIter<TimeBlock, TimeBlock::child_const_iter> timer_tree_dfs_iterator_t;
+
+
+static timer_tree_dfs_iterator_t begin_timer_tree(TimeBlock& id) 
+{ 
+	return timer_tree_dfs_iterator_t(&id, 
+		boost::bind(boost::mem_fn(&TimeBlock::beginChildren), _1), 
+							boost::bind(boost::mem_fn(&TimeBlock::endChildren), _1));
+}
+
+static timer_tree_dfs_iterator_t end_timer_tree() 
+{ 
+	return timer_tree_dfs_iterator_t(); 
+}
+
+
+// sort child timers by name
+struct SortTimerByName
+{
+	bool operator()(const TimeBlock* i1, const TimeBlock* i2)
+	{
+		return i1->getName() < i2->getName();
+	}
+};
+
+TimeBlock& TimeBlock::getRootTimeBlock()
+{
+	static TimeBlock root_timer("root", true, NULL);
+	return root_timer;
+}
+
+void TimeBlock::pushLog(LLSD log)
+{
+	LLMutexLock lock(sLogLock);
+
+	sLogQueue.push(log);
+}
+
+void TimeBlock::setLogLock(LLMutex* lock)
+{
+	sLogLock = lock;
+}
+
+
+//static
+#if (LL_DARWIN || LL_LINUX || LL_SOLARIS) && !(defined(__i386__) || defined(__amd64__))
+U64 TimeBlock::countsPerSecond()
+{
+	return sClockResolution;
+}
+#else // windows or x86-mac or x86-linux or x86-solaris
+U64 TimeBlock::countsPerSecond()
+{
+#if LL_FASTTIMER_USE_RDTSC || !LL_WINDOWS
+	//getCPUFrequency returns MHz and sCPUClockFrequency wants to be in Hz
+	static LLUnit<LLUnits::Hertz, U64> sCPUClockFrequency = LLProcessorInfo().getCPUFrequency();
+
+#else
+	// If we're not using RDTSC, each fast timer tick is just a performance counter tick.
+	// Not redefining the clock frequency itself (in llprocessor.cpp/calculate_cpu_frequency())
+	// since that would change displayed MHz stats for CPUs
+	static bool firstcall = true;
+	static U64 sCPUClockFrequency;
+	if (firstcall)
+	{
+		QueryPerformanceFrequency((LARGE_INTEGER*)&sCPUClockFrequency);
+		firstcall = false;
+	}
+#endif
+	return sCPUClockFrequency.value();
+}
+#endif
+
+TimeBlock::TimeBlock(const char* name, bool open, TimeBlock* parent)
+:	TraceType<TimeBlockAccumulator>(name),
+	mCollapsed(true)
+{
+	setCollapsed(!open);
+}
+
+TimeBlockTreeNode& TimeBlock::getTreeNode() const
+{
+	TimeBlockTreeNode* nodep = LLTrace::get_thread_recorder()->getTimeBlockTreeNode(getIndex());
+	llassert(nodep);
+	return *nodep;
+}
+
+// static
+void TimeBlock::processTimes()
+{
+	get_clock_count(); // good place to calculate clock frequency
+	U64 cur_time = getCPUClockCount64();
+	BlockTimerStackRecord* stack_record = ThreadTimerStack::getInstance();
+
+	// set up initial tree
+	for (LLInstanceTracker<TimeBlock>::instance_iter it = LLInstanceTracker<TimeBlock>::beginInstances(), end_it = LLInstanceTracker<TimeBlock>::endInstances(); 
+		it != end_it; 
+		++it)
+	{
+		TimeBlock& timer = *it;
+		if (&timer == &TimeBlock::getRootTimeBlock()) continue;
+
+		// bootstrap tree construction by attaching to last timer to be on stack
+		// when this timer was called
+		if (timer.getParent() == &TimeBlock::getRootTimeBlock())
+		{
+			TimeBlockAccumulator* accumulator = timer.getPrimaryAccumulator();
+
+			if (accumulator->mLastCaller)
+			{
+				timer.setParent(accumulator->mLastCaller);
+				accumulator->mParent = accumulator->mLastCaller;
+			}
+			// no need to push up tree on first use, flag can be set spuriously
+			accumulator->mMoveUpTree = false;
+		}
+	}
+
+	// bump timers up tree if they have been flagged as being in the wrong place
+	// do this in a bottom up order to promote descendants first before promoting ancestors
+	// this preserves partial order derived from current frame's observations
+	for(timer_tree_bottom_up_iterator_t it = begin_timer_tree_bottom_up(TimeBlock::getRootTimeBlock());
+		it != end_timer_tree_bottom_up();
+		++it)
+	{
+		TimeBlock* timerp = *it;
+
+		// sort timers by time last called, so call graph makes sense
+		TimeBlockTreeNode& tree_node = timerp->getTreeNode();
+		if (tree_node.mNeedsSorting)
+		{
+			std::sort(tree_node.mChildren.begin(), tree_node.mChildren.end(), SortTimerByName());
+		}
+
+		// skip root timer
+		if (timerp != &TimeBlock::getRootTimeBlock())
+		{
+			TimeBlockAccumulator* accumulator = timerp->getPrimaryAccumulator();
+
+			if (accumulator->mMoveUpTree)
+			{
+				// since ancestors have already been visited, re-parenting won't affect tree traversal
+				//step up tree, bringing our descendants with us
+				LL_DEBUGS("FastTimers") << "Moving " << timerp->getName() << " from child of " << timerp->getParent()->getName() <<
+					" to child of " << timerp->getParent()->getParent()->getName() << LL_ENDL;
+				timerp->setParent(timerp->getParent()->getParent());
+				accumulator->mParent = timerp->getParent();
+				accumulator->mMoveUpTree = false;
+
+				// don't bubble up any ancestors until descendants are done bubbling up
+				// as ancestors may call this timer only on certain paths, so we want to resolve
+				// child-most block locations before their parents
+				it.skipAncestors();
+			}
+		}
+	}
+
+	// walk up stack of active timers and accumulate current time while leaving timing structures active
+	BlockTimer* cur_timer = stack_record->mActiveTimer;
+	TimeBlockAccumulator* accumulator = stack_record->mTimeBlock->getPrimaryAccumulator();
+	// root defined by parent pointing to self
+	while(cur_timer && cur_timer->mLastTimerData.mActiveTimer != cur_timer)
+	{
+		U64 cumulative_time_delta = cur_time - cur_timer->mStartTime;
+		U64 self_time_delta = cumulative_time_delta - stack_record->mChildTime;
+		stack_record->mChildTime = 0;
+		accumulator->mSelfTimeCounter += self_time_delta;
+		accumulator->mTotalTimeCounter += cumulative_time_delta;
+
+		cur_timer->mStartTime = cur_time;
+
+		stack_record = &cur_timer->mLastTimerData;
+		stack_record->mChildTime += cumulative_time_delta;
+		if (stack_record->mTimeBlock)
+		{
+			accumulator = stack_record->mTimeBlock->getPrimaryAccumulator();
+		}
+
+		cur_timer = cur_timer->mLastTimerData.mActiveTimer;
+	}
+
+
+	// reset for next frame
+	for (LLInstanceTracker<TimeBlock>::instance_iter it = LLInstanceTracker<TimeBlock>::beginInstances(),
+			end_it = LLInstanceTracker<TimeBlock>::endInstances();
+		it != end_it;
+		++it)
+	{
+		TimeBlock& timer = *it;
+		TimeBlockAccumulator* accumulator = timer.getPrimaryAccumulator();
+
+		accumulator->mLastCaller = NULL;
+		accumulator->mMoveUpTree = false;
+	}
+
+	// traverse tree in DFS post order, or bottom up
+	//for(timer_tree_bottom_up_iterator_t it = begin_timer_tree_bottom_up(TimeBlock::getRootTimer());
+	//	it != end_timer_tree_bottom_up();
+	//	++it)
+	//{
+	//	TimeBlock* timerp = (*it);
+	//	TimeBlockAccumulator& accumulator = timerp->getPrimaryAccumulator();
+	//	timerp->mTreeTimeCounter = accumulator.mSelfTimeCounter;
+	//	for (child_const_iter child_it = timerp->beginChildren(); child_it != timerp->endChildren(); ++child_it)
+	//	{
+	//		timerp->mTreeTimeCounter += (*child_it)->mTreeTimeCounter;
+	//	}
+	//}
+}
+
+
+std::vector<TimeBlock*>::iterator TimeBlock::beginChildren()
+{ 
+	return getTreeNode().mChildren.begin(); 
+}
+
+std::vector<TimeBlock*>::iterator TimeBlock::endChildren()
+{
+	return getTreeNode().mChildren.end();
+}
+
+std::vector<TimeBlock*>& TimeBlock::getChildren()
+{
+	return getTreeNode().mChildren;
+}
+
+//static
+void TimeBlock::logStats()
+{
+	// get ready for next frame
+	if (sLog)
+	{ //output current frame counts to performance log
+
+		static S32 call_count = 0;
+		if (call_count % 100 == 0)
+		{
+			LL_DEBUGS("FastTimers") << "countsPerSecond: " << countsPerSecond() << LL_ENDL;
+			LL_DEBUGS("FastTimers") << "LLProcessorInfo().getCPUFrequency() " << LLProcessorInfo().getCPUFrequency() << LL_ENDL;
+			LL_DEBUGS("FastTimers") << "getCPUClockCount32() " << getCPUClockCount32() << LL_ENDL;
+			LL_DEBUGS("FastTimers") << "getCPUClockCount64() " << getCPUClockCount64() << LL_ENDL;
+			LL_DEBUGS("FastTimers") << "elapsed sec " << ((F64)getCPUClockCount64()) / (LLUnit<LLUnits::Hertz, F64>(LLProcessorInfo().getCPUFrequency())) << LL_ENDL;
+		}
+		call_count++;
+
+		LLUnit<LLUnits::Seconds, F64> total_time(0);
+		LLSD sd;
+
+		{
+			for (LLInstanceTracker<TimeBlock>::instance_iter it = LLInstanceTracker<TimeBlock>::beginInstances(), 
+				end_it = LLInstanceTracker<TimeBlock>::endInstances(); 
+				it != end_it; 
+			++it)
+			{
+				TimeBlock& timer = *it;
+				LLTrace::PeriodicRecording& frame_recording = LLTrace::get_frame_recording();
+				sd[timer.getName()]["Time"] = (LLSD::Real) (frame_recording.getLastRecordingPeriod().getSum(timer).value());	
+				sd[timer.getName()]["Calls"] = (LLSD::Integer) (frame_recording.getLastRecordingPeriod().getSum(timer.callCount()));
+
+				// computing total time here because getting the root timer's getCountHistory
+				// doesn't work correctly on the first frame
+				total_time += frame_recording.getLastRecordingPeriod().getSum(timer);
+			}
+		}
+
+		sd["Total"]["Time"] = (LLSD::Real) total_time.value();
+		sd["Total"]["Calls"] = (LLSD::Integer) 1;
+
+		{		
+			LLMutexLock lock(sLogLock);
+			sLogQueue.push(sd);
+		}
+	}
+
+}
+
+//static
+void TimeBlock::dumpCurTimes()
+{
+	LLTrace::PeriodicRecording& frame_recording = LLTrace::get_frame_recording();
+	LLTrace::Recording& last_frame_recording = frame_recording.getLastRecordingPeriod();
+
+	// walk over timers in depth order and output timings
+	for(timer_tree_dfs_iterator_t it = begin_timer_tree(TimeBlock::getRootTimeBlock());
+		it != end_timer_tree();
+		++it)
+	{
+		TimeBlock* timerp = (*it);
+		LLUnit<LLUnits::Seconds, F64> total_time_ms = last_frame_recording.getSum(*timerp);
+		U32 num_calls = last_frame_recording.getSum(timerp->callCount());
+
+		// Don't bother with really brief times, keep output concise
+		if (total_time_ms < 0.1) continue;
+
+		std::ostringstream out_str;
+		TimeBlock* parent_timerp = timerp;
+		while(parent_timerp && parent_timerp != parent_timerp->getParent())
+		{
+			out_str << "\t";
+			parent_timerp = parent_timerp->getParent();
+		}
+
+		out_str << timerp->getName() << " " 
+			<< std::setprecision(3) << total_time_ms.as<LLUnits::Milliseconds, F32>().value() << " ms, "
+			<< num_calls << " calls";
+
+		llinfos << out_str.str() << llendl;
+	}
+}
+
+//static
+void TimeBlock::writeLog(std::ostream& os)
+{
+	while (!sLogQueue.empty())
+	{
+		LLSD& sd = sLogQueue.front();
+		LLSDSerialize::toXML(sd, os);
+		LLMutexLock lock(sLogLock);
+		sLogQueue.pop();
+	}
+}
+
+//////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+// TimeBlockAccumulator
+//////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+
+TimeBlockAccumulator::TimeBlockAccumulator() 
+:	mSelfTimeCounter(0),
+	mTotalTimeCounter(0),
+	mCalls(0),
+	mLastCaller(NULL),
+	mActiveCount(0),
+	mMoveUpTree(false),
+	mParent(NULL)
+{}
+
+void TimeBlockAccumulator::addSamples( const TimeBlockAccumulator& other )
+{
+	mSelfTimeCounter += other.mSelfTimeCounter;
+	mTotalTimeCounter += other.mTotalTimeCounter;
+	mCalls += other.mCalls;
+	mLastCaller = other.mLastCaller;
+	mActiveCount = other.mActiveCount;
+	mMoveUpTree = other.mMoveUpTree;
+	mParent = other.mParent;
+}
+
+void TimeBlockAccumulator::reset( const TimeBlockAccumulator* other )
+{
+	mTotalTimeCounter = 0;
+	mSelfTimeCounter = 0;
+	mCalls = 0;
+}
+
+} // namespace LLTrace
diff --git a/indra/llcommon/llfasttimer.h b/indra/llcommon/llfasttimer.h
index 2b25f2fabb..fb2868ff98 100644
--- a/indra/llcommon/llfasttimer.h
+++ b/indra/llcommon/llfasttimer.h
@@ -1,6 +1,6 @@
 /**
  * @file llfasttimer.h
- * @brief Inline implementations of fast timers.
+ * @brief Declaration of a fast timer.
  *
  * $LicenseInfo:firstyear=2004&license=viewerlgpl$
  * Second Life Viewer Source Code
@@ -27,9 +27,295 @@
 #ifndef LL_FASTTIMER_H
 #define LL_FASTTIMER_H
 
-// Implementation of getCPUClockCount32() and getCPUClockCount64 are now in llfastertimer_class.cpp.
+#include "llinstancetracker.h"
+#include "lltrace.h"
 
-// pull in the actual class definition
-#include "llfasttimer_class.h"
+#define FAST_TIMER_ON 1
+#define LL_FASTTIMER_USE_RDTSC 1
+
+class LLMutex;
+
+namespace LLTrace
+{
+
+struct BlockTimerStackRecord
+{
+	class BlockTimer*	mActiveTimer;
+	class TimeBlock*	mTimeBlock;
+	U64					mChildTime;
+};
+
+class ThreadTimerStack 
+:	public BlockTimerStackRecord, 
+	public LLThreadLocalSingleton<ThreadTimerStack>
+{
+	friend class LLThreadLocalSingleton<ThreadTimerStack>;
+	ThreadTimerStack() 
+	{}
+
+public:
+	ThreadTimerStack& operator=(const BlockTimerStackRecord& other)
+	{
+		BlockTimerStackRecord::operator=(other);
+		return *this;
+	}
+};
+
+class BlockTimer
+{
+public:
+	friend class TimeBlock;
+	typedef BlockTimer self_t;
+	typedef class TimeBlock DeclareTimer;
+
+	BlockTimer(TimeBlock& timer);
+	~BlockTimer();
+
+private:
+
+	U64				mStartTime;
+	BlockTimerStackRecord	mLastTimerData;
+};
+
+// stores a "named" timer instance to be reused via multiple BlockTimer stack instances
+class TimeBlock 
+:	public TraceType<TimeBlockAccumulator>,
+	public LLInstanceTracker<TimeBlock>
+{
+public:
+	TimeBlock(const char* name, bool open = false, TimeBlock* parent = &getRootTimeBlock());
+
+	TimeBlockTreeNode& getTreeNode() const;
+	TimeBlock* getParent() const { return getTreeNode().getParent(); }
+	void setParent(TimeBlock* parent) { getTreeNode().setParent(parent); }
+
+	typedef std::vector<TimeBlock*>::iterator child_iter;
+	typedef std::vector<TimeBlock*>::const_iterator child_const_iter;
+	child_iter beginChildren();
+	child_iter endChildren();
+	std::vector<TimeBlock*>& getChildren();
+
+	void setCollapsed(bool collapsed)	{ mCollapsed = collapsed; }
+	bool getCollapsed() const			{ return mCollapsed; }
+
+	TraceType<TimeBlockAccumulator::CallCountAspect>& callCount() 
+	{ 
+		return static_cast<TraceType<TimeBlockAccumulator::CallCountAspect>&>(*(TraceType<TimeBlockAccumulator>*)this);
+	}
+
+	TraceType<TimeBlockAccumulator::SelfTimeAspect>& selfTime() 
+	{ 
+		return static_cast<TraceType<TimeBlockAccumulator::SelfTimeAspect>&>(*(TraceType<TimeBlockAccumulator>*)this);
+	}
+
+	static TimeBlock& getRootTimeBlock();
+	static void pushLog(LLSD sd);
+	static void setLogLock(LLMutex* mutex);
+	static void writeLog(std::ostream& os);
+
+	// dumps current cumulative frame stats to log
+	// call nextFrame() to reset timers
+	static void dumpCurTimes();
+
+	//////////////////////////////////////////////////////////////////////////////
+	//
+	// Important note: These implementations must be FAST!
+	//
+
+
+#if LL_WINDOWS
+	//
+	// Windows implementation of CPU clock
+	//
+
+	//
+	// NOTE: put back in when we aren't using platform sdk anymore
+	//
+	// because MS has different signatures for these functions in winnt.h
+	// need to rename them to avoid conflicts
+	//#define _interlockedbittestandset _renamed_interlockedbittestandset
+	//#define _interlockedbittestandreset _renamed_interlockedbittestandreset
+	//#include <intrin.h>
+	//#undef _interlockedbittestandset
+	//#undef _interlockedbittestandreset
+
+	//inline U32 TimeBlock::getCPUClockCount32()
+	//{
+	//	U64 time_stamp = __rdtsc();
+	//	return (U32)(time_stamp >> 8);
+	//}
+	//
+	//// return full timer value, *not* shifted by 8 bits
+	//inline U64 TimeBlock::getCPUClockCount64()
+	//{
+	//	return __rdtsc();
+	//}
+
+	// shift off lower 8 bits for lower resolution but longer term timing
+	// on 1Ghz machine, a 32-bit word will hold ~1000 seconds of timing
+#if LL_FASTTIMER_USE_RDTSC
+	static U32 getCPUClockCount32()
+	{
+		U32 ret_val;
+		__asm
+		{
+			_emit   0x0f
+				_emit   0x31
+				shr eax,8
+				shl edx,24
+				or eax, edx
+				mov dword ptr [ret_val], eax
+		}
+		return ret_val;
+	}
+
+	// return full timer value, *not* shifted by 8 bits
+	static U64 getCPUClockCount64()
+	{
+		U64 ret_val;
+		__asm
+		{
+			_emit   0x0f
+				_emit   0x31
+				mov eax,eax
+				mov edx,edx
+				mov dword ptr [ret_val+4], edx
+				mov dword ptr [ret_val], eax
+		}
+		return ret_val;
+	}
+
+#else
+	//U64 get_clock_count(); // in lltimer.cpp
+	// These use QueryPerformanceCounter, which is arguably fine and also works on AMD architectures.
+	static U32 getCPUClockCount32()
+	{
+		return (U32)(get_clock_count()>>8);
+	}
+
+	static U64 getCPUClockCount64()
+	{
+		return get_clock_count();
+	}
+
+#endif
+
+#endif
+
+
+#if (LL_LINUX || LL_SOLARIS) && !(defined(__i386__) || defined(__amd64__))
+	//
+	// Linux and Solaris implementation of CPU clock - non-x86.
+	// This is accurate but SLOW!  Only use out of desperation.
+	//
+	// Try to use the MONOTONIC clock if available, this is a constant time counter
+	// with nanosecond resolution (but not necessarily accuracy) and attempts are
+	// made to synchronize this value between cores at kernel start. It should not
+	// be affected by CPU frequency. If not available use the REALTIME clock, but
+	// this may be affected by NTP adjustments or other user activity affecting
+	// the system time.
+	static U64 getCPUClockCount64()
+	{
+		struct timespec tp;
+
+#ifdef CLOCK_MONOTONIC // MONOTONIC supported at build-time?
+		if (-1 == clock_gettime(CLOCK_MONOTONIC,&tp)) // if MONOTONIC isn't supported at runtime then ouch, try REALTIME
+#endif
+			clock_gettime(CLOCK_REALTIME,&tp);
+
+		return (tp.tv_sec*sClockResolution)+tp.tv_nsec;        
+	}
+
+	static U32 getCPUClockCount32()
+	{
+		return (U32)(getCPUClockCount64() >> 8);
+	}
+
+#endif // (LL_LINUX || LL_SOLARIS) && !(defined(__i386__) || defined(__amd64__))
+
+
+#if (LL_LINUX || LL_SOLARIS || LL_DARWIN) && (defined(__i386__) || defined(__amd64__))
+	//
+	// Mac+Linux+Solaris FAST x86 implementation of CPU clock
+	static U32 getCPUClockCount32()
+	{
+		U64 x;
+		__asm__ volatile (".byte 0x0f, 0x31": "=A"(x));
+		return (U32)(x >> 8);
+	}
+
+	static U64 getCPUClockCount64()
+	{
+		U64 x;
+		__asm__ volatile (".byte 0x0f, 0x31": "=A"(x));
+		return x;
+	}
+
+#endif
+
+	static U64 countsPerSecond();
+
+	// updates cumulative times and hierarchy,
+	// can be called multiple times in a frame, at any point
+	static void processTimes();
+
+	// call this once a frame to periodically log timers
+	static void logStats();
+
+	bool						mCollapsed;				// don't show children
+
+	// statics
+	static std::string							sLogName;
+	static bool									sMetricLog,
+												sLog;	
+	static U64									sClockResolution;
+};
+
+LL_FORCE_INLINE BlockTimer::BlockTimer(TimeBlock& timer)
+{
+#if FAST_TIMER_ON
+	mStartTime = TimeBlock::getCPUClockCount64();
+
+	BlockTimerStackRecord* cur_timer_data = ThreadTimerStack::getIfExists();
+	TimeBlockAccumulator* accumulator = cur_timer_data->mTimeBlock->getPrimaryAccumulator();
+	accumulator->mActiveCount++;
+	// keep current parent as long as it is active when we are
+	accumulator->mMoveUpTree |= (accumulator->mParent->getPrimaryAccumulator()->mActiveCount == 0);
+
+	// store top of stack
+	mLastTimerData = *cur_timer_data;
+	// push new information
+	cur_timer_data->mActiveTimer = this;
+	cur_timer_data->mTimeBlock = &timer;
+	cur_timer_data->mChildTime = 0;
+#endif
+}
+
+LL_FORCE_INLINE BlockTimer::~BlockTimer()
+{
+#if FAST_TIMER_ON
+	U64 total_time = TimeBlock::getCPUClockCount64() - mStartTime;
+	BlockTimerStackRecord* cur_timer_data = ThreadTimerStack::getIfExists();
+	TimeBlockAccumulator* accumulator = cur_timer_data->mTimeBlock->getPrimaryAccumulator();
+
+	accumulator->mCalls++;
+	accumulator->mSelfTimeCounter += total_time - cur_timer_data->mChildTime;
+	accumulator->mTotalTimeCounter += total_time;
+	accumulator->mActiveCount--;
+
+	// store last caller to bootstrap tree creation
+	// do this in the destructor in case of recursion to get topmost caller
+	accumulator->mLastCaller = mLastTimerData.mTimeBlock;
+
+	// we are only tracking self time, so subtract our total time delta from parents
+	mLastTimerData.mChildTime += total_time;
+
+	*ThreadTimerStack::getIfExists() = mLastTimerData;
+#endif
+}
+
+}
+
+typedef LLTrace::BlockTimer LLFastTimer; 
 
 #endif // LL_LLFASTTIMER_H
diff --git a/indra/llcommon/llfasttimer_class.cpp b/indra/llcommon/llfasttimer_class.cpp
deleted file mode 100644
index 463f558c2c..0000000000
--- a/indra/llcommon/llfasttimer_class.cpp
+++ /dev/null
@@ -1,921 +0,0 @@
-/** 
- * @file llfasttimer_class.cpp
- * @brief Implementation of the fast timer.
- *
- * $LicenseInfo:firstyear=2004&license=viewerlgpl$
- * Second Life Viewer Source Code
- * Copyright (C) 2010, Linden Research, Inc.
- * 
- * This library is free software; you can redistribute it and/or
- * modify it under the terms of the GNU Lesser General Public
- * License as published by the Free Software Foundation;
- * version 2.1 of the License only.
- * 
- * This library is distributed in the hope that it will be useful,
- * but WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
- * Lesser General Public License for more details.
- * 
- * You should have received a copy of the GNU Lesser General Public
- * License along with this library; if not, write to the Free Software
- * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA
- * 
- * Linden Research, Inc., 945 Battery Street, San Francisco, CA  94111  USA
- * $/LicenseInfo$
- */
-#include "linden_common.h"
-
-#include "llfasttimer.h"
-
-#include "llmemory.h"
-#include "llprocessor.h"
-#include "llsingleton.h"
-#include "lltreeiterators.h"
-#include "llsdserialize.h"
-
-#include <boost/bind.hpp>
-
-
-#if LL_WINDOWS
-#include "lltimer.h"
-#elif LL_LINUX || LL_SOLARIS
-#include <sys/time.h>
-#include <sched.h>
-#include "lltimer.h"
-#elif LL_DARWIN
-#include <sys/time.h>
-#include "lltimer.h"	// get_clock_count()
-#else 
-#error "architecture not supported"
-#endif
-
-//////////////////////////////////////////////////////////////////////////////
-// statics
-
-S32 LLFastTimer::sCurFrameIndex = -1;
-S32 LLFastTimer::sLastFrameIndex = -1;
-U64 LLFastTimer::sLastFrameTime = LLFastTimer::getCPUClockCount64();
-bool LLFastTimer::sPauseHistory = 0;
-bool LLFastTimer::sResetHistory = 0;
-LLFastTimer::CurTimerData LLFastTimer::sCurTimerData;
-BOOL LLFastTimer::sLog = FALSE;
-std::string LLFastTimer::sLogName = "";
-BOOL LLFastTimer::sMetricLog = FALSE;
-LLMutex* LLFastTimer::sLogLock = NULL;
-std::queue<LLSD> LLFastTimer::sLogQueue;
-
-#define USE_RDTSC 0
-
-#if LL_LINUX || LL_SOLARIS
-U64 LLFastTimer::sClockResolution = 1000000000; // Nanosecond resolution
-#else
-U64 LLFastTimer::sClockResolution = 1000000; // Microsecond resolution
-#endif
-
-std::vector<LLFastTimer::FrameState>* LLFastTimer::sTimerInfos = NULL;
-U64				LLFastTimer::sTimerCycles = 0;
-U32				LLFastTimer::sTimerCalls = 0;
-
-
-// FIXME: move these declarations to the relevant modules
-
-// helper functions
-typedef LLTreeDFSPostIter<LLFastTimer::NamedTimer, LLFastTimer::NamedTimer::child_const_iter> timer_tree_bottom_up_iterator_t;
-
-static timer_tree_bottom_up_iterator_t begin_timer_tree_bottom_up(LLFastTimer::NamedTimer& id) 
-{ 
-	return timer_tree_bottom_up_iterator_t(&id, 
-							boost::bind(boost::mem_fn(&LLFastTimer::NamedTimer::beginChildren), _1), 
-							boost::bind(boost::mem_fn(&LLFastTimer::NamedTimer::endChildren), _1));
-}
-
-static timer_tree_bottom_up_iterator_t end_timer_tree_bottom_up() 
-{ 
-	return timer_tree_bottom_up_iterator_t(); 
-}
-
-typedef LLTreeDFSIter<LLFastTimer::NamedTimer, LLFastTimer::NamedTimer::child_const_iter> timer_tree_dfs_iterator_t;
-
-
-static timer_tree_dfs_iterator_t begin_timer_tree(LLFastTimer::NamedTimer& id) 
-{ 
-	return timer_tree_dfs_iterator_t(&id, 
-		boost::bind(boost::mem_fn(&LLFastTimer::NamedTimer::beginChildren), _1), 
-							boost::bind(boost::mem_fn(&LLFastTimer::NamedTimer::endChildren), _1));
-}
-
-static timer_tree_dfs_iterator_t end_timer_tree() 
-{ 
-	return timer_tree_dfs_iterator_t(); 
-}
-
-
-
-// factory class that creates NamedTimers via static DeclareTimer objects
-class NamedTimerFactory : public LLSingleton<NamedTimerFactory>
-{
-public:
-	NamedTimerFactory()
-		: mActiveTimerRoot(NULL),
-		  mTimerRoot(NULL),
-		  mAppTimer(NULL),
-		  mRootFrameState(NULL)
-	{}
-
-	/*virtual */ void initSingleton()
-	{
-		mTimerRoot = new LLFastTimer::NamedTimer("root");
-
-		mActiveTimerRoot = new LLFastTimer::NamedTimer("Frame");
-		mActiveTimerRoot->setCollapsed(false);
-
-		mRootFrameState = new LLFastTimer::FrameState(mActiveTimerRoot);
-		mRootFrameState->mParent = &mTimerRoot->getFrameState();
-		mActiveTimerRoot->setParent(mTimerRoot);
-
-		mAppTimer = new LLFastTimer(mRootFrameState);
-	}
-
-	~NamedTimerFactory()
-	{
-		std::for_each(mTimers.begin(), mTimers.end(), DeletePairedPointer());
-
-		delete mAppTimer;
-		delete mActiveTimerRoot; 
-		delete mTimerRoot;
-		delete mRootFrameState;
-	}
-
-	LLFastTimer::NamedTimer& createNamedTimer(const std::string& name)
-	{
-		timer_map_t::iterator found_it = mTimers.find(name);
-		if (found_it != mTimers.end())
-		{
-			return *found_it->second;
-		}
-
-		LLFastTimer::NamedTimer* timer = new LLFastTimer::NamedTimer(name);
-		timer->setParent(mTimerRoot);
-		mTimers.insert(std::make_pair(name, timer));
-
-		return *timer;
-	}
-
-	LLFastTimer::NamedTimer* getTimerByName(const std::string& name)
-	{
-		timer_map_t::iterator found_it = mTimers.find(name);
-		if (found_it != mTimers.end())
-		{
-			return found_it->second;
-		}
-		return NULL;
-	}
-
-	LLFastTimer::NamedTimer* getActiveRootTimer() { return mActiveTimerRoot; }
-	LLFastTimer::NamedTimer* getRootTimer() { return mTimerRoot; }
-	const LLFastTimer* getAppTimer() { return mAppTimer; }
-	LLFastTimer::FrameState& getRootFrameState() { return *mRootFrameState; }
-
-	typedef std::map<std::string, LLFastTimer::NamedTimer*> timer_map_t;
-	timer_map_t::iterator beginTimers() { return mTimers.begin(); }
-	timer_map_t::iterator endTimers() { return mTimers.end(); }
-	S32 timerCount() { return mTimers.size(); }
-
-private:
-	timer_map_t mTimers;
-
-	LLFastTimer::NamedTimer*		mActiveTimerRoot;
-	LLFastTimer::NamedTimer*		mTimerRoot;
-	LLFastTimer*						mAppTimer;
-	LLFastTimer::FrameState*		mRootFrameState;
-};
-
-void update_cached_pointers_if_changed()
-{
-	// detect when elements have moved and update cached pointers
-	static LLFastTimer::FrameState* sFirstTimerAddress = NULL;
-	if (&*(LLFastTimer::getFrameStateList().begin()) != sFirstTimerAddress)
-	{
-		LLFastTimer::DeclareTimer::updateCachedPointers();
-	}
-	sFirstTimerAddress = &*(LLFastTimer::getFrameStateList().begin());
-}
-
-LLFastTimer::DeclareTimer::DeclareTimer(const std::string& name, bool open )
-:	mTimer(NamedTimerFactory::instance().createNamedTimer(name))
-{
-	mTimer.setCollapsed(!open);
-	mFrameState = &mTimer.getFrameState();
-	update_cached_pointers_if_changed();
-}
-
-LLFastTimer::DeclareTimer::DeclareTimer(const std::string& name)
-:	mTimer(NamedTimerFactory::instance().createNamedTimer(name))
-{
-	mFrameState = &mTimer.getFrameState();
-	update_cached_pointers_if_changed();
-}
-
-// static
-void LLFastTimer::DeclareTimer::updateCachedPointers()
-{
-	// propagate frame state pointers to timer declarations
-	for (instance_iter it = beginInstances(); it != endInstances(); ++it)
-	{
-		// update cached pointer
-		it->mFrameState = &it->mTimer.getFrameState();
-	}
-
-	// also update frame states of timers on stack
-	LLFastTimer* cur_timerp = LLFastTimer::sCurTimerData.mCurTimer;
-	while(cur_timerp->mLastTimerData.mCurTimer != cur_timerp)	
-	{
-		cur_timerp->mFrameState = &cur_timerp->mFrameState->mTimer->getFrameState();
-		cur_timerp = cur_timerp->mLastTimerData.mCurTimer;
-	}
-}
-
-//static
-#if (LL_DARWIN || LL_LINUX || LL_SOLARIS) && !(defined(__i386__) || defined(__amd64__))
-U64 LLFastTimer::countsPerSecond() // counts per second for the *32-bit* timer
-{
-	return sClockResolution >> 8;
-}
-#else // windows or x86-mac or x86-linux or x86-solaris
-U64 LLFastTimer::countsPerSecond() // counts per second for the *32-bit* timer
-{
-#if USE_RDTSC || !LL_WINDOWS
-	//getCPUFrequency returns MHz and sCPUClockFrequency wants to be in Hz
-	static U64 sCPUClockFrequency = U64(LLProcessorInfo().getCPUFrequency()*1000000.0);
-
-	// we drop the low-order byte in our timers, so report a lower frequency
-#else
-	// If we're not using RDTSC, each fasttimer tick is just a performance counter tick.
-	// Not redefining the clock frequency itself (in llprocessor.cpp/calculate_cpu_frequency())
-	// since that would change displayed MHz stats for CPUs
-	static bool firstcall = true;
-	static U64 sCPUClockFrequency;
-	if (firstcall)
-	{
-		QueryPerformanceFrequency((LARGE_INTEGER*)&sCPUClockFrequency);
-		firstcall = false;
-	}
-#endif
-	return sCPUClockFrequency >> 8;
-}
-#endif
-
-LLFastTimer::FrameState::FrameState(LLFastTimer::NamedTimer* timerp)
-:	mActiveCount(0),
-	mCalls(0),
-	mSelfTimeCounter(0),
-	mParent(NULL),
-	mLastCaller(NULL),
-	mMoveUpTree(false),
-	mTimer(timerp)
-{}
-
-
-LLFastTimer::NamedTimer::NamedTimer(const std::string& name)
-:	mName(name),
-	mCollapsed(true),
-	mParent(NULL),
-	mTotalTimeCounter(0),
-	mCountAverage(0),
-	mCallAverage(0),
-	mNeedsSorting(false)
-{
-	info_list_t& frame_state_list = getFrameStateList();
-	mFrameStateIndex = frame_state_list.size();
-	getFrameStateList().push_back(FrameState(this));
-
-	mCountHistory = new U32[HISTORY_NUM];
-	memset(mCountHistory, 0, sizeof(U32) * HISTORY_NUM);
-	mCallHistory = new U32[HISTORY_NUM];
-	memset(mCallHistory, 0, sizeof(U32) * HISTORY_NUM);
-}
-
-LLFastTimer::NamedTimer::~NamedTimer()
-{
-	delete[] mCountHistory;
-	delete[] mCallHistory;
-}
-
-std::string LLFastTimer::NamedTimer::getToolTip(S32 history_idx)
-{
-	F64 ms_multiplier = 1000.0 / (F64)LLFastTimer::countsPerSecond();
-	if (history_idx < 0)
-	{
-		// by default, show average number of call
-		return llformat("%s (%d ms, %d calls)", getName().c_str(), (S32)(getCountAverage() * ms_multiplier), (S32)getCallAverage());
-	}
-	else
-	{
-		return llformat("%s (%d ms, %d calls)", getName().c_str(), (S32)(getHistoricalCount(history_idx) * ms_multiplier), (S32)getHistoricalCalls(history_idx));
-	}
-}
-
-void LLFastTimer::NamedTimer::setParent(NamedTimer* parent)
-{
-	llassert_always(parent != this);
-	llassert_always(parent != NULL);
-
-	if (mParent)
-	{
-		// subtract our accumulated from previous parent
-		for (S32 i = 0; i < HISTORY_NUM; i++)
-		{
-			mParent->mCountHistory[i] -= mCountHistory[i];
-		}
-
-		// subtract average timing from previous parent
-		mParent->mCountAverage -= mCountAverage;
-
-		std::vector<NamedTimer*>& children = mParent->getChildren();
-		std::vector<NamedTimer*>::iterator found_it = std::find(children.begin(), children.end(), this);
-		if (found_it != children.end())
-		{
-			children.erase(found_it);
-		}
-	}
-
-	mParent = parent;
-	if (parent)
-	{
-		getFrameState().mParent = &parent->getFrameState();
-		parent->getChildren().push_back(this);
-		parent->mNeedsSorting = true;
-	}
-}
-
-S32 LLFastTimer::NamedTimer::getDepth()
-{
-	S32 depth = 0;
-	NamedTimer* timerp = mParent;
-	while(timerp)
-	{
-		depth++;
-		timerp = timerp->mParent;
-	}
-	return depth;
-}
-
-// static
-void LLFastTimer::NamedTimer::processTimes()
-{
-	if (sCurFrameIndex < 0) return;
-
-	buildHierarchy();
-	accumulateTimings();
-}
-
-// sort timer info structs by depth first traversal order
-struct SortTimersDFS
-{
-	bool operator()(const LLFastTimer::FrameState& i1, const LLFastTimer::FrameState& i2)
-	{
-		return i1.mTimer->getFrameStateIndex() < i2.mTimer->getFrameStateIndex();
-	}
-};
-
-// sort child timers by name
-struct SortTimerByName
-{
-	bool operator()(const LLFastTimer::NamedTimer* i1, const LLFastTimer::NamedTimer* i2)
-	{
-		return i1->getName() < i2->getName();
-	}
-};
-
-//static
-void LLFastTimer::NamedTimer::buildHierarchy()
-{
-	if (sCurFrameIndex < 0 ) return;
-
-	// set up initial tree
-	{
-		for (instance_iter it = beginInstances(); it != endInstances(); ++it)
-		{
-			NamedTimer& timer = *it;
-			if (&timer == NamedTimerFactory::instance().getRootTimer()) continue;
-			
-			// bootstrap tree construction by attaching to last timer to be on stack
-			// when this timer was called
-			if (timer.getFrameState().mLastCaller && timer.mParent == NamedTimerFactory::instance().getRootTimer())
-			{
-				timer.setParent(timer.getFrameState().mLastCaller->mTimer);
-				// no need to push up tree on first use, flag can be set spuriously
-				timer.getFrameState().mMoveUpTree = false;
-			}
-		}
-	}
-
-	// bump timers up tree if they've been flagged as being in the wrong place
-	// do this in a bottom up order to promote descendants first before promoting ancestors
-	// this preserves partial order derived from current frame's observations
-	for(timer_tree_bottom_up_iterator_t it = begin_timer_tree_bottom_up(*NamedTimerFactory::instance().getRootTimer());
-		it != end_timer_tree_bottom_up();
-		++it)
-	{
-		NamedTimer* timerp = *it;
-		// skip root timer
-		if (timerp == NamedTimerFactory::instance().getRootTimer()) continue;
-
-		if (timerp->getFrameState().mMoveUpTree)
-		{
-			// since ancestors have already been visited, reparenting won't affect tree traversal
-			//step up tree, bringing our descendants with us
-			//llinfos << "Moving " << timerp->getName() << " from child of " << timerp->getParent()->getName() <<
-			//	" to child of " << timerp->getParent()->getParent()->getName() << llendl;
-			timerp->setParent(timerp->getParent()->getParent());
-			timerp->getFrameState().mMoveUpTree = false;
-
-			// don't bubble up any ancestors until descendants are done bubbling up
-			it.skipAncestors();
-		}
-	}
-
-	// sort timers by time last called, so call graph makes sense
-	for(timer_tree_dfs_iterator_t it = begin_timer_tree(*NamedTimerFactory::instance().getRootTimer());
-		it != end_timer_tree();
-		++it)
-	{
-		NamedTimer* timerp = (*it);
-		if (timerp->mNeedsSorting)
-		{
-			std::sort(timerp->getChildren().begin(), timerp->getChildren().end(), SortTimerByName());
-		}
-		timerp->mNeedsSorting = false;
-	}
-}
-
-//static
-void LLFastTimer::NamedTimer::accumulateTimings()
-{
-	U32 cur_time = getCPUClockCount32();
-
-	// walk up stack of active timers and accumulate current time while leaving timing structures active
-	LLFastTimer* cur_timer = sCurTimerData.mCurTimer;
-	// root defined by parent pointing to self
-	CurTimerData* cur_data = &sCurTimerData;
-	while(cur_timer->mLastTimerData.mCurTimer != cur_timer)
-	{
-		U32 cumulative_time_delta = cur_time - cur_timer->mStartTime;
-		U32 self_time_delta = cumulative_time_delta - cur_data->mChildTime;
-		cur_data->mChildTime = 0;
-		cur_timer->mFrameState->mSelfTimeCounter += self_time_delta;
-		cur_timer->mStartTime = cur_time;
-
-		cur_data = &cur_timer->mLastTimerData;
-		cur_data->mChildTime += cumulative_time_delta;
-
-		cur_timer = cur_timer->mLastTimerData.mCurTimer;
-	}
-
-	// traverse tree in DFS post order, or bottom up
-	for(timer_tree_bottom_up_iterator_t it = begin_timer_tree_bottom_up(*NamedTimerFactory::instance().getActiveRootTimer());
-		it != end_timer_tree_bottom_up();
-		++it)
-	{
-		NamedTimer* timerp = (*it);
-		timerp->mTotalTimeCounter = timerp->getFrameState().mSelfTimeCounter;
-		for (child_const_iter child_it = timerp->beginChildren(); child_it != timerp->endChildren(); ++child_it)
-		{
-			timerp->mTotalTimeCounter += (*child_it)->mTotalTimeCounter;
-		}
-
-		S32 cur_frame = sCurFrameIndex;
-		if (cur_frame >= 0)
-		{
-			// update timer history
-			int hidx = cur_frame % HISTORY_NUM;
-
-			timerp->mCountHistory[hidx] = timerp->mTotalTimeCounter;
-			timerp->mCountAverage = ((U64)timerp->mCountAverage * cur_frame + timerp->mTotalTimeCounter) / (cur_frame+1);
-			timerp->mCallHistory[hidx] = timerp->getFrameState().mCalls;
-			timerp->mCallAverage = ((U64)timerp->mCallAverage * cur_frame + timerp->getFrameState().mCalls) / (cur_frame+1);
-		}
-	}
-}
-
-// static
-void LLFastTimer::NamedTimer::resetFrame()
-{
-	if (sLog)
-	{ //output current frame counts to performance log
-
-		static S32 call_count = 0;
-		if (call_count % 100 == 0)
-		{
-			llinfos << "countsPerSecond (32 bit): " << countsPerSecond() << llendl;
-			llinfos << "get_clock_count (64 bit): " << get_clock_count() << llendl;
-			llinfos << "LLProcessorInfo().getCPUFrequency() " << LLProcessorInfo().getCPUFrequency() << llendl;
-			llinfos << "getCPUClockCount32() " << getCPUClockCount32() << llendl;
-			llinfos << "getCPUClockCount64() " << getCPUClockCount64() << llendl;
-			llinfos << "elapsed sec " << ((F64)getCPUClockCount64())/((F64)LLProcessorInfo().getCPUFrequency()*1000000.0) << llendl;
-		}
-		call_count++;
-
-		F64 iclock_freq = 1000.0 / countsPerSecond(); // good place to calculate clock frequency
-
-		F64 total_time = 0;
-		LLSD sd;
-
-		{
-			for (instance_iter it = beginInstances(); it != endInstances(); ++it)
-			{
-				NamedTimer& timer = *it;
-				FrameState& info = timer.getFrameState();
-				sd[timer.getName()]["Time"] = (LLSD::Real) (info.mSelfTimeCounter*iclock_freq);	
-				sd[timer.getName()]["Calls"] = (LLSD::Integer) info.mCalls;
-				
-				// computing total time here because getting the root timer's getCountHistory
-				// doesn't work correctly on the first frame
-				total_time = total_time + info.mSelfTimeCounter * iclock_freq;
-			}
-		}
-
-		sd["Total"]["Time"] = (LLSD::Real) total_time;
-		sd["Total"]["Calls"] = (LLSD::Integer) 1;
-
-		{		
-			LLMutexLock lock(sLogLock);
-			sLogQueue.push(sd);
-		}
-	}
-
-
-	// tag timers by position in depth first traversal of tree
-	S32 index = 0;
-	for(timer_tree_dfs_iterator_t it = begin_timer_tree(*NamedTimerFactory::instance().getRootTimer());
-		it != end_timer_tree();
-		++it)
-	{
-		NamedTimer* timerp = (*it);
-		
-		timerp->mFrameStateIndex = index;
-		index++;
-
-		llassert_always(timerp->mFrameStateIndex < (S32)getFrameStateList().size());
-	}
-
-	// sort timers by DFS traversal order to improve cache coherency
-	std::sort(getFrameStateList().begin(), getFrameStateList().end(), SortTimersDFS());
-
-	// update pointers into framestatelist now that we've sorted it
-	DeclareTimer::updateCachedPointers();
-
-	// reset for next frame
-	{
-		for (instance_iter it = beginInstances(); it != endInstances(); ++it)
-		{
-			NamedTimer& timer = *it;
-			
-			FrameState& info = timer.getFrameState();
-			info.mSelfTimeCounter = 0;
-			info.mCalls = 0;
-			info.mLastCaller = NULL;
-			info.mMoveUpTree = false;
-			// update parent pointer in timer state struct
-			if (timer.mParent)
-			{
-				info.mParent = &timer.mParent->getFrameState();
-			}
-		}
-	}
-
-	//sTimerCycles = 0;
-	//sTimerCalls = 0;
-}
-
-//static
-void LLFastTimer::NamedTimer::reset()
-{
-	resetFrame(); // reset frame data
-
-	// walk up stack of active timers and reset start times to current time
-	// effectively zeroing out any accumulated time
-	U32 cur_time = getCPUClockCount32();
-
-	// root defined by parent pointing to self
-	CurTimerData* cur_data = &sCurTimerData;
-	LLFastTimer* cur_timer = cur_data->mCurTimer;
-	while(cur_timer->mLastTimerData.mCurTimer != cur_timer)
-	{
-		cur_timer->mStartTime = cur_time;
-		cur_data->mChildTime = 0;
-
-		cur_data = &cur_timer->mLastTimerData;
-		cur_timer = cur_data->mCurTimer;
-	}
-
-	// reset all history
-	{
-		for (instance_iter it = beginInstances(); it != endInstances(); ++it)
-		{
-			NamedTimer& timer = *it;
-			if (&timer != NamedTimerFactory::instance().getRootTimer()) 
-			{
-				timer.setParent(NamedTimerFactory::instance().getRootTimer());
-			}
-			
-			timer.mCountAverage = 0;
-			timer.mCallAverage = 0;
-			memset(timer.mCountHistory, 0, sizeof(U32) * HISTORY_NUM);
-			memset(timer.mCallHistory, 0, sizeof(U32) * HISTORY_NUM);
-		}
-	}
-
-	sLastFrameIndex = 0;
-	sCurFrameIndex = 0;
-}
-
-//static 
-LLFastTimer::info_list_t& LLFastTimer::getFrameStateList() 
-{ 
-	if (!sTimerInfos) 
-	{ 
-		sTimerInfos = new info_list_t(); 
-	} 
-	return *sTimerInfos; 
-}
-
-
-U32 LLFastTimer::NamedTimer::getHistoricalCount(S32 history_index) const
-{
-	S32 history_idx = (getLastFrameIndex() + history_index) % LLFastTimer::NamedTimer::HISTORY_NUM;
-	return mCountHistory[history_idx];
-}
-
-U32 LLFastTimer::NamedTimer::getHistoricalCalls(S32 history_index ) const
-{
-	S32 history_idx = (getLastFrameIndex() + history_index) % LLFastTimer::NamedTimer::HISTORY_NUM;
-	return mCallHistory[history_idx];
-}
-
-LLFastTimer::FrameState& LLFastTimer::NamedTimer::getFrameState() const
-{
-	llassert_always(mFrameStateIndex >= 0);
-	if (this == NamedTimerFactory::instance().getActiveRootTimer()) 
-	{
-		return NamedTimerFactory::instance().getRootFrameState();
-	}
-	return getFrameStateList()[mFrameStateIndex];
-}
-
-// static
-LLFastTimer::NamedTimer& LLFastTimer::NamedTimer::getRootNamedTimer()
-{ 
-	return *NamedTimerFactory::instance().getActiveRootTimer(); 
-}
-
-std::vector<LLFastTimer::NamedTimer*>::const_iterator LLFastTimer::NamedTimer::beginChildren()
-{ 
-	return mChildren.begin(); 
-}
-
-std::vector<LLFastTimer::NamedTimer*>::const_iterator LLFastTimer::NamedTimer::endChildren()
-{
-	return mChildren.end();
-}
-
-std::vector<LLFastTimer::NamedTimer*>& LLFastTimer::NamedTimer::getChildren()
-{
-	return mChildren;
-}
-
-//static
-void LLFastTimer::nextFrame()
-{
-	countsPerSecond(); // good place to calculate clock frequency
-	U64 frame_time = getCPUClockCount64();
-	if ((frame_time - sLastFrameTime) >> 8 > 0xffffffff)
-	{
-		llinfos << "Slow frame, fast timers inaccurate" << llendl;
-	}
-
-	if (!sPauseHistory)
-	{
-		NamedTimer::processTimes();
-		sLastFrameIndex = sCurFrameIndex++;
-	}
-	
-	// get ready for next frame
-	NamedTimer::resetFrame();
-	sLastFrameTime = frame_time;
-}
-
-//static
-void LLFastTimer::dumpCurTimes()
-{
-	// accumulate timings, etc.
-	NamedTimer::processTimes();
-	
-	F64 clock_freq = (F64)countsPerSecond();
-	F64 iclock_freq = 1000.0 / clock_freq; // clock_ticks -> milliseconds
-
-	// walk over timers in depth order and output timings
-	for(timer_tree_dfs_iterator_t it = begin_timer_tree(*NamedTimerFactory::instance().getRootTimer());
-		it != end_timer_tree();
-		++it)
-	{
-		NamedTimer* timerp = (*it);
-		F64 total_time_ms = ((F64)timerp->getHistoricalCount(0) * iclock_freq);
-		// Don't bother with really brief times, keep output concise
-		if (total_time_ms < 0.1) continue;
-
-		std::ostringstream out_str;
-		for (S32 i = 0; i < timerp->getDepth(); i++)
-		{
-			out_str << "\t";
-		}
-
-
-		out_str << timerp->getName() << " " 
-			<< std::setprecision(3) << total_time_ms << " ms, "
-			<< timerp->getHistoricalCalls(0) << " calls";
-
-		llinfos << out_str.str() << llendl;
-	}
-}
-
-//static 
-void LLFastTimer::reset()
-{
-	NamedTimer::reset();
-}
-
-
-//static
-void LLFastTimer::writeLog(std::ostream& os)
-{
-	while (!sLogQueue.empty())
-	{
-		LLSD& sd = sLogQueue.front();
-		LLSDSerialize::toXML(sd, os);
-		LLMutexLock lock(sLogLock);
-		sLogQueue.pop();
-	}
-}
-
-//static
-const LLFastTimer::NamedTimer* LLFastTimer::getTimerByName(const std::string& name)
-{
-	return NamedTimerFactory::instance().getTimerByName(name);
-}
-
-LLFastTimer::LLFastTimer(LLFastTimer::FrameState* state)
-:	mFrameState(state)
-{
-	U32 start_time = getCPUClockCount32();
-	mStartTime = start_time;
-	mFrameState->mActiveCount++;
-	LLFastTimer::sCurTimerData.mCurTimer = this;
-	LLFastTimer::sCurTimerData.mFrameState = mFrameState;
-	LLFastTimer::sCurTimerData.mChildTime = 0;
-	mLastTimerData = LLFastTimer::sCurTimerData;
-}
-
-
-//////////////////////////////////////////////////////////////////////////////
-//
-// Important note: These implementations must be FAST!
-//
-
-
-#if LL_WINDOWS
-//
-// Windows implementation of CPU clock
-//
-
-//
-// NOTE: put back in when we aren't using platform sdk anymore
-//
-// because MS has different signatures for these functions in winnt.h
-// need to rename them to avoid conflicts
-//#define _interlockedbittestandset _renamed_interlockedbittestandset
-//#define _interlockedbittestandreset _renamed_interlockedbittestandreset
-//#include <intrin.h>
-//#undef _interlockedbittestandset
-//#undef _interlockedbittestandreset
-
-//inline U32 LLFastTimer::getCPUClockCount32()
-//{
-//	U64 time_stamp = __rdtsc();
-//	return (U32)(time_stamp >> 8);
-//}
-//
-//// return full timer value, *not* shifted by 8 bits
-//inline U64 LLFastTimer::getCPUClockCount64()
-//{
-//	return __rdtsc();
-//}
-
-// shift off lower 8 bits for lower resolution but longer term timing
-// on 1Ghz machine, a 32-bit word will hold ~1000 seconds of timing
-#if USE_RDTSC
-U32 LLFastTimer::getCPUClockCount32()
-{
-	U32 ret_val;
-	__asm
-	{
-        _emit   0x0f
-        _emit   0x31
-		shr eax,8
-		shl edx,24
-		or eax, edx
-		mov dword ptr [ret_val], eax
-	}
-    return ret_val;
-}
-
-// return full timer value, *not* shifted by 8 bits
-U64 LLFastTimer::getCPUClockCount64()
-{
-	U64 ret_val;
-	__asm
-	{
-        _emit   0x0f
-        _emit   0x31
-		mov eax,eax
-		mov edx,edx
-		mov dword ptr [ret_val+4], edx
-		mov dword ptr [ret_val], eax
-	}
-    return ret_val;
-}
-
-std::string LLFastTimer::sClockType = "rdtsc";
-
-#else
-//LL_COMMON_API U64 get_clock_count(); // in lltimer.cpp
-// These use QueryPerformanceCounter, which is arguably fine and also works on AMD architectures.
-U32 LLFastTimer::getCPUClockCount32()
-{
-	return (U32)(get_clock_count()>>8);
-}
-
-U64 LLFastTimer::getCPUClockCount64()
-{
-	return get_clock_count();
-}
-
-std::string LLFastTimer::sClockType = "QueryPerformanceCounter";
-#endif
-
-#endif
-
-
-#if (LL_LINUX || LL_SOLARIS) && !(defined(__i386__) || defined(__amd64__))
-//
-// Linux and Solaris implementation of CPU clock - non-x86.
-// This is accurate but SLOW!  Only use out of desperation.
-//
-// Try to use the MONOTONIC clock if available, this is a constant time counter
-// with nanosecond resolution (but not necessarily accuracy) and attempts are
-// made to synchronize this value between cores at kernel start. It should not
-// be affected by CPU frequency. If not available use the REALTIME clock, but
-// this may be affected by NTP adjustments or other user activity affecting
-// the system time.
-U64 LLFastTimer::getCPUClockCount64()
-{
-	struct timespec tp;
-	
-#ifdef CLOCK_MONOTONIC // MONOTONIC supported at build-time?
-	if (-1 == clock_gettime(CLOCK_MONOTONIC,&tp)) // if MONOTONIC isn't supported at runtime then ouch, try REALTIME
-#endif
-		clock_gettime(CLOCK_REALTIME,&tp);
-
-	return (tp.tv_sec*LLFastTimer::sClockResolution)+tp.tv_nsec;        
-}
-
-U32 LLFastTimer::getCPUClockCount32()
-{
-	return (U32)(LLFastTimer::getCPUClockCount64() >> 8);
-}
-
-std::string LLFastTimer::sClockType = "clock_gettime";
-
-#endif // (LL_LINUX || LL_SOLARIS) && !(defined(__i386__) || defined(__amd64__))
-
-
-#if (LL_LINUX || LL_SOLARIS || LL_DARWIN) && (defined(__i386__) || defined(__amd64__))
-//
-// Mac+Linux+Solaris FAST x86 implementation of CPU clock
-U32 LLFastTimer::getCPUClockCount32()
-{
-	U64 x;
-	__asm__ volatile (".byte 0x0f, 0x31": "=A"(x));
-	return (U32)(x >> 8);
-}
-
-U64 LLFastTimer::getCPUClockCount64()
-{
-	U64 x;
-	__asm__ volatile (".byte 0x0f, 0x31": "=A"(x));
-	return x;
-}
-
-std::string LLFastTimer::sClockType = "rdtsc";
-#endif
-
diff --git a/indra/llcommon/llfasttimer_class.h b/indra/llcommon/llfasttimer_class.h
deleted file mode 100644
index f481e968a6..0000000000
--- a/indra/llcommon/llfasttimer_class.h
+++ /dev/null
@@ -1,276 +0,0 @@
-/**
- * @file llfasttimer_class.h
- * @brief Declaration of a fast timer.
- *
- * $LicenseInfo:firstyear=2004&license=viewerlgpl$
- * Second Life Viewer Source Code
- * Copyright (C) 2010, Linden Research, Inc.
- * 
- * This library is free software; you can redistribute it and/or
- * modify it under the terms of the GNU Lesser General Public
- * License as published by the Free Software Foundation;
- * version 2.1 of the License only.
- * 
- * This library is distributed in the hope that it will be useful,
- * but WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
- * Lesser General Public License for more details.
- * 
- * You should have received a copy of the GNU Lesser General Public
- * License along with this library; if not, write to the Free Software
- * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA
- * 
- * Linden Research, Inc., 945 Battery Street, San Francisco, CA  94111  USA
- * $/LicenseInfo$
- */
-
-#ifndef LL_FASTTIMER_CLASS_H
-#define LL_FASTTIMER_CLASS_H
-
-#include "llinstancetracker.h"
-
-#define FAST_TIMER_ON 1
-#define TIME_FAST_TIMERS 0
-#define DEBUG_FAST_TIMER_THREADS 1
-
-class LLMutex;
-
-#include <queue>
-#include "llsd.h"
-
-LL_COMMON_API void assert_main_thread();
-
-class LL_COMMON_API LLFastTimer
-{
-public:
-	class NamedTimer;
-
-	struct LL_COMMON_API FrameState
-	{
-		FrameState(NamedTimer* timerp);
-
-		U32 				mSelfTimeCounter;
-		U32 				mCalls;
-		FrameState*			mParent;		// info for caller timer
-		FrameState*			mLastCaller;	// used to bootstrap tree construction
-		NamedTimer*			mTimer;
-		U16					mActiveCount;	// number of timers with this ID active on stack
-		bool				mMoveUpTree;	// needs to be moved up the tree of timers at the end of frame
-	};
-
-	// stores a "named" timer instance to be reused via multiple LLFastTimer stack instances
-	class LL_COMMON_API NamedTimer
-	:	public LLInstanceTracker<NamedTimer>
-	{
-		friend class DeclareTimer;
-	public:
-		~NamedTimer();
-
-		enum { HISTORY_NUM = 300 };
-
-		const std::string& getName() const { return mName; }
-		NamedTimer* getParent() const { return mParent; }
-		void setParent(NamedTimer* parent);
-		S32 getDepth();
-		std::string getToolTip(S32 history_index = -1);
-
-		typedef std::vector<NamedTimer*>::const_iterator child_const_iter;
-		child_const_iter beginChildren();
-		child_const_iter endChildren();
-		std::vector<NamedTimer*>& getChildren();
-
-		void setCollapsed(bool collapsed) { mCollapsed = collapsed; }
-		bool getCollapsed() const { return mCollapsed; }
-
-		U32 getCountAverage() const { return mCountAverage; }
-		U32 getCallAverage() const { return mCallAverage; }
-
-		U32 getHistoricalCount(S32 history_index = 0) const;
-		U32 getHistoricalCalls(S32 history_index = 0) const;
-
-		static NamedTimer& getRootNamedTimer();
-
-		S32 getFrameStateIndex() const { return mFrameStateIndex; }
-
-		FrameState& getFrameState() const;
-
-	private:
-		friend class LLFastTimer;
-		friend class NamedTimerFactory;
-
-		//
-		// methods
-		//
-		NamedTimer(const std::string& name);
-		// recursive call to gather total time from children
-		static void accumulateTimings();
-
-		// updates cumulative times and hierarchy,
-		// can be called multiple times in a frame, at any point
-		static void processTimes();
-
-		static void buildHierarchy();
-		static void resetFrame();
-		static void reset();
-
-		//
-		// members
-		//
-		S32			mFrameStateIndex;
-
-		std::string	mName;
-
-		U32 		mTotalTimeCounter;
-
-		U32 		mCountAverage;
-		U32			mCallAverage;
-
-		U32*		mCountHistory;
-		U32*		mCallHistory;
-
-		// tree structure
-		NamedTimer*					mParent;				// NamedTimer of caller(parent)
-		std::vector<NamedTimer*>	mChildren;
-		bool						mCollapsed;				// don't show children
-		bool						mNeedsSorting;			// sort children whenever child added
-	};
-
-	// used to statically declare a new named timer
-	class LL_COMMON_API DeclareTimer
-	:	public LLInstanceTracker<DeclareTimer>
-	{
-		friend class LLFastTimer;
-	public:
-		DeclareTimer(const std::string& name, bool open);
-		DeclareTimer(const std::string& name);
-
-		static void updateCachedPointers();
-
-	private:
-		NamedTimer&		mTimer;
-		FrameState*		mFrameState;
-	};
-
-public:
-	LLFastTimer(LLFastTimer::FrameState* state);
-
-	LL_FORCE_INLINE LLFastTimer(LLFastTimer::DeclareTimer& timer)
-	:	mFrameState(timer.mFrameState)
-	{
-#if TIME_FAST_TIMERS
-		U64 timer_start = getCPUClockCount64();
-#endif
-#if FAST_TIMER_ON
-		LLFastTimer::FrameState* frame_state = mFrameState;
-		mStartTime = getCPUClockCount32();
-
-		frame_state->mActiveCount++;
-		frame_state->mCalls++;
-		// keep current parent as long as it is active when we are
-		frame_state->mMoveUpTree |= (frame_state->mParent->mActiveCount == 0);
-
-		LLFastTimer::CurTimerData* cur_timer_data = &LLFastTimer::sCurTimerData;
-		mLastTimerData = *cur_timer_data;
-		cur_timer_data->mCurTimer = this;
-		cur_timer_data->mFrameState = frame_state;
-		cur_timer_data->mChildTime = 0;
-#endif
-#if TIME_FAST_TIMERS
-		U64 timer_end = getCPUClockCount64();
-		sTimerCycles += timer_end - timer_start;
-#endif
-#if DEBUG_FAST_TIMER_THREADS
-#if !LL_RELEASE
-		assert_main_thread();
-#endif
-#endif
-	}
-
-	LL_FORCE_INLINE ~LLFastTimer()
-	{
-#if TIME_FAST_TIMERS
-		U64 timer_start = getCPUClockCount64();
-#endif
-#if FAST_TIMER_ON
-		LLFastTimer::FrameState* frame_state = mFrameState;
-		U32 total_time = getCPUClockCount32() - mStartTime;
-
-		frame_state->mSelfTimeCounter += total_time - LLFastTimer::sCurTimerData.mChildTime;
-		frame_state->mActiveCount--;
-
-		// store last caller to bootstrap tree creation
-		// do this in the destructor in case of recursion to get topmost caller
-		frame_state->mLastCaller = mLastTimerData.mFrameState;
-
-		// we are only tracking self time, so subtract our total time delta from parents
-		mLastTimerData.mChildTime += total_time;
-
-		LLFastTimer::sCurTimerData = mLastTimerData;
-#endif
-#if TIME_FAST_TIMERS
-		U64 timer_end = getCPUClockCount64();
-		sTimerCycles += timer_end - timer_start;
-		sTimerCalls++;
-#endif
-	}
-
-public:
-	static LLMutex*			sLogLock;
-	static std::queue<LLSD> sLogQueue;
-	static BOOL				sLog;
-	static BOOL				sMetricLog;
-	static std::string		sLogName;
-	static bool 			sPauseHistory;
-	static bool 			sResetHistory;
-	static U64				sTimerCycles;
-	static U32				sTimerCalls;
-
-	typedef std::vector<FrameState> info_list_t;
-	static info_list_t& getFrameStateList();
-
-
-	// call this once a frame to reset timers
-	static void nextFrame();
-
-	// dumps current cumulative frame stats to log
-	// call nextFrame() to reset timers
-	static void dumpCurTimes();
-
-	// call this to reset timer hierarchy, averages, etc.
-	static void reset();
-
-	static U64 countsPerSecond();
-	static S32 getLastFrameIndex() { return sLastFrameIndex; }
-	static S32 getCurFrameIndex() { return sCurFrameIndex; }
-
-	static void writeLog(std::ostream& os);
-	static const NamedTimer* getTimerByName(const std::string& name);
-
-	struct CurTimerData
-	{
-		LLFastTimer*	mCurTimer;
-		FrameState*		mFrameState;
-		U32				mChildTime;
-	};
-	static CurTimerData		sCurTimerData;
-	static std::string sClockType;
-
-private:
-	static U32 getCPUClockCount32();
-	static U64 getCPUClockCount64();
-	static U64 sClockResolution;
-
-	static S32				sCurFrameIndex;
-	static S32				sLastFrameIndex;
-	static U64				sLastFrameTime;
-	static info_list_t*		sTimerInfos;
-
-	U32							mStartTime;
-	LLFastTimer::FrameState*	mFrameState;
-	LLFastTimer::CurTimerData	mLastTimerData;
-
-};
-
-typedef class LLFastTimer LLFastTimer;
-
-#endif // LL_LLFASTTIMER_CLASS_H
diff --git a/indra/llcommon/llfile.cpp b/indra/llcommon/llfile.cpp
index c51d042a3d..5917d7a420 100644
--- a/indra/llcommon/llfile.cpp
+++ b/indra/llcommon/llfile.cpp
@@ -28,7 +28,7 @@
  */
 
 #if LL_WINDOWS
-#include <windows.h>
+#include "llwin32headerslean.h"
 #include <stdlib.h>                 // Windows errno
 #else
 #include <errno.h>
diff --git a/indra/llcommon/llfindlocale.cpp b/indra/llcommon/llfindlocale.cpp
index cd7c0c7c09..f019bd0c64 100644
--- a/indra/llcommon/llfindlocale.cpp
+++ b/indra/llcommon/llfindlocale.cpp
@@ -33,7 +33,7 @@
 #include <ctype.h>
 
 #ifdef WIN32
-#include <windows.h>
+#include "llwin32headers.h"
 #include <winnt.h>
 #endif
 
diff --git a/indra/llcommon/llinitparam.cpp b/indra/llcommon/llinitparam.cpp
index db72aa19b9..d72e10d2fa 100644
--- a/indra/llcommon/llinitparam.cpp
+++ b/indra/llcommon/llinitparam.cpp
@@ -28,10 +28,17 @@
 #include "linden_common.h"
 
 #include "llinitparam.h"
+#include "llformat.h"
 
 
 namespace LLInitParam
 {
+
+	predicate_rule_t default_parse_rules() 
+	{ 
+		return ll_make_predicate(PROVIDED) && !ll_make_predicate(EMPTY);
+	}
+
 	//
 	// Param
 	//
@@ -164,6 +171,9 @@ namespace LLInitParam
 
 	bool BaseBlock::validateBlock(bool emit_errors) const
 	{
+		// only validate block when it hasn't already passed validation with current data
+		if (!mValidated)
+		{
 		const BlockDescriptor& block_data = mostDerivedBlockDescriptor();
 		for (BlockDescriptor::param_validation_list_t::const_iterator it = block_data.mValidationList.begin(); it != block_data.mValidationList.end(); ++it)
 		{
@@ -177,11 +187,18 @@ namespace LLInitParam
 				return false;
 			}
 		}
-		return true;
+			mValidated = true;
+		}
+		return mValidated;
 	}
 
-	void BaseBlock::serializeBlock(Parser& parser, Parser::name_stack_t& name_stack, const LLInitParam::BaseBlock* diff_block) const
+	bool BaseBlock::serializeBlock(Parser& parser, Parser::name_stack_t& name_stack, const predicate_rule_t predicate_rule, const LLInitParam::BaseBlock* diff_block) const
 	{
+		bool serialized = false;
+		if (!predicate_rule.check(ll_make_predicate(PROVIDED, isProvided())))
+		{
+			return false;
+		}
 		// named param is one like LLView::Params::follows
 		// unnamed param is like LLView::Params::rect - implicit
 		const BlockDescriptor& block_data = mostDerivedBlockDescriptor();
@@ -193,15 +210,10 @@ namespace LLInitParam
 			param_handle_t param_handle = (*it)->mParamHandle;
 			const Param* param = getParamFromHandle(param_handle);
 			ParamDescriptor::serialize_func_t serialize_func = (*it)->mSerializeFunc;
-			if (serialize_func)
+			if (serialize_func && predicate_rule.check(ll_make_predicate(PROVIDED, param->anyProvided())))
 			{
 				const Param* diff_param = diff_block ? diff_block->getParamFromHandle(param_handle) : NULL;
-				// each param descriptor remembers its serial number
-				// so we can inspect the same param under different names
-				// and see that it has the same number
-				name_stack.push_back(std::make_pair("", true));
-				serialize_func(*param, parser, name_stack, diff_param);
-				name_stack.pop_back();
+				serialized |= serialize_func(*param, parser, name_stack, predicate_rule, diff_param);
 			}
 		}
 
@@ -212,7 +224,7 @@ namespace LLInitParam
 			param_handle_t param_handle = it->second->mParamHandle;
 			const Param* param = getParamFromHandle(param_handle);
 			ParamDescriptor::serialize_func_t serialize_func = it->second->mSerializeFunc;
-			if (serialize_func && param->anyProvided())
+			if (serialize_func && predicate_rule.check(ll_make_predicate(PROVIDED, param->anyProvided())))
 			{
 				// Ensure this param has not already been serialized
 				// Prevents <rect> from being serialized as its own tag.
@@ -237,10 +249,17 @@ namespace LLInitParam
 
 				name_stack.push_back(std::make_pair(it->first, !duplicate));
 				const Param* diff_param = diff_block ? diff_block->getParamFromHandle(param_handle) : NULL;
-				serialize_func(*param, parser, name_stack, diff_param);
+				serialized |= serialize_func(*param, parser, name_stack, predicate_rule, diff_param);
 				name_stack.pop_back();
 			}
 		}
+
+		if (!serialized && predicate_rule.check(ll_make_predicate(EMPTY)))
+		{
+			serialized |= parser.writeValue(Flag(), name_stack);
+		}
+		// was anything serialized in this block?
+		return serialized;
 	}
 
 	bool BaseBlock::inspectBlock(Parser& parser, Parser::name_stack_t name_stack, S32 min_count, S32 max_count) const
@@ -359,7 +378,7 @@ namespace LLInitParam
 	}
 
 	//static 
-	void BaseBlock::addParam(BlockDescriptor& block_data, const ParamDescriptorPtr in_param, const char* char_name)
+	void BaseBlock::addParam(BlockDescriptor& block_data, ParamDescriptorPtr in_param, const char* char_name)
 	{
 		// create a copy of the param descriptor in mAllParams
 		// so other data structures can store a pointer to it
diff --git a/indra/llcommon/llinitparam.h b/indra/llcommon/llinitparam.h
index 9a6d1eff5c..502f93cbb8 100644
--- a/indra/llcommon/llinitparam.h
+++ b/indra/llcommon/llinitparam.h
@@ -29,13 +29,14 @@
 #define LL_LLPARAM_H
 
 #include <vector>
+#include <list>
 #include <boost/function.hpp>
 #include <boost/type_traits/is_convertible.hpp>
 #include <boost/unordered_map.hpp>
-#include <boost/shared_ptr.hpp>
 
 #include "llerror.h"
 #include "llstl.h"
+#include "llpredicate.h"
 
 namespace LLInitParam
 {
@@ -211,7 +212,6 @@ namespace LLInitParam
 		LOG_CLASS(Parser);
 
 	public:
-		
 		typedef std::vector<std::pair<std::string, bool> >					name_stack_t;
 		typedef std::pair<name_stack_t::iterator, name_stack_t::iterator>	name_stack_range_t;
 		typedef std::vector<std::string>									possible_values_t;
@@ -293,6 +293,19 @@ namespace LLInitParam
 
 	class Param;
 
+	enum ESerializePredicates
+	{
+		PROVIDED,
+		REQUIRED,
+		VALID,
+		HAS_DEFAULT_VALUE,
+		EMPTY
+	};
+
+	typedef LLPredicate::Rule<ESerializePredicates> predicate_rule_t;
+
+	predicate_rule_t default_parse_rules();
+
 	// various callbacks and constraints associated with an individual param
 	struct LL_COMMON_API ParamDescriptor
 	{
@@ -303,8 +316,8 @@ namespace LLInitParam
 
 		typedef bool(*merge_func_t)(Param&, const Param&, bool);
 		typedef bool(*deserialize_func_t)(Param&, Parser&, const Parser::name_stack_range_t&, bool);
-		typedef void(*serialize_func_t)(const Param&, Parser&, Parser::name_stack_t&, const Param* diff_param);
-		typedef void(*inspect_func_t)(const Param&, Parser&, Parser::name_stack_t&, S32 min_count, S32 max_count);
+		typedef bool(*serialize_func_t)(const Param&, Parser&, Parser::name_stack_t&, const predicate_rule_t, const Param*);
+		typedef void(*inspect_func_t)(const Param&, Parser&, Parser::name_stack_t&, S32, S32);
 		typedef bool(*validation_func_t)(const Param*);
 
 		ParamDescriptor(param_handle_t p, 
@@ -331,7 +344,7 @@ namespace LLInitParam
 		UserData*			mUserData;
 	};
 
-	typedef boost::shared_ptr<ParamDescriptor> ParamDescriptorPtr;
+	typedef ParamDescriptor* ParamDescriptorPtr;
 
 	// each derived Block class keeps a static data structure maintaining offsets to various params
 	class LL_COMMON_API BlockDescriptor
@@ -484,12 +497,28 @@ namespace LLInitParam
 		LOG_CLASS(BaseBlock);
 		friend class Param;
 
+		BaseBlock()
+		:	mValidated(false),
+			mParamProvided(false)
+		{}
+
 		virtual ~BaseBlock() {}
 		bool submitValue(Parser::name_stack_t& name_stack, Parser& p, bool silent=false);
 
 		param_handle_t getHandleFromParam(const Param* param) const;
 		bool validateBlock(bool emit_errors = true) const;
 
+		bool isProvided() const
+		{
+			return mParamProvided;
+		}
+
+		bool isValid() const
+		{
+			return validateBlock(false);
+		}
+
+
 		Param* getParamFromHandle(const param_handle_t param_handle)
 		{
 			if (param_handle == 0) return NULL;
@@ -507,10 +536,19 @@ namespace LLInitParam
 		void addSynonym(Param& param, const std::string& synonym);
 
 		// Blocks can override this to do custom tracking of changes
-		virtual void paramChanged(const Param& changed_param, bool user_provided) {}
+		virtual void paramChanged(const Param& changed_param, bool user_provided) 
+		{
+			if (user_provided)
+			{
+				// a child param has been explicitly changed
+				// so *some* aspect of this block is now provided
+				mValidated = false;
+				mParamProvided = true;
+			}
+		}
 
 		bool deserializeBlock(Parser& p, Parser::name_stack_range_t name_stack_range, bool new_name);
-		void serializeBlock(Parser& p, Parser::name_stack_t& name_stack, const BaseBlock* diff_block = NULL) const;
+		bool serializeBlock(Parser& p, Parser::name_stack_t& name_stack, const predicate_rule_t rule, const BaseBlock* diff_block = NULL) const;
 		bool inspectBlock(Parser& p, Parser::name_stack_t name_stack = Parser::name_stack_t(), S32 min_count = 0, S32 max_count = S32_MAX) const;
 
 		virtual const BlockDescriptor& mostDerivedBlockDescriptor() const { return selfBlockDescriptor(); }
@@ -549,6 +587,9 @@ namespace LLInitParam
 			return sBlockDescriptor;
 		}
 
+		mutable bool 	mValidated; // lazy validation flag
+		bool			mParamProvided;
+
 	private:
 		const std::string& getParamName(const BlockDescriptor& block_data, const Param* paramp) const;
 	};
@@ -688,13 +729,11 @@ namespace LLInitParam
 		typedef ParamValue<T, NAME_VALUE_LOOKUP, true>	self_t;
 
 		ParamValue() 
-		:	T(),
-			mValidated(false)
+		:	T()
 		{}
 
 		ParamValue(value_assignment_t other)
-		:	T(other),
-			mValidated(false)
+		:	T(other)
 		{}
 
 		void setValue(value_assignment_t val)
@@ -736,9 +775,6 @@ namespace LLInitParam
 
 			return *this;
 		}
-
-	protected:
-		mutable bool 	mValidated; // lazy validation flag
 	};
 
 	template<typename NAME_VALUE_LOOKUP>
@@ -836,6 +872,8 @@ namespace LLInitParam
 
 		bool isProvided() const { return Param::anyProvided(); }
 
+		bool isValid() const { return true; }
+
 		static bool deserializeParam(Param& param, Parser& parser, const Parser::name_stack_range_t& name_stack_range, bool new_name)
 		{ 
 			self_t& typed_param = static_cast<self_t&>(param);
@@ -870,10 +908,23 @@ namespace LLInitParam
 			return false;
 		}
 
-		static void serializeParam(const Param& param, Parser& parser, Parser::name_stack_t& name_stack, const Param* diff_param)
+		static bool serializeParam(const Param& param, Parser& parser, Parser::name_stack_t& name_stack, const predicate_rule_t predicate_rule, const Param* diff_param)
 		{
+			bool serialized = false;
 			const self_t& typed_param = static_cast<const self_t&>(param);
-			if (!typed_param.isProvided()) return;
+			const self_t* diff_typed_param = static_cast<const self_t*>(diff_param);
+
+			LLPredicate::Value<ESerializePredicates> predicate;
+			if (diff_typed_param && ParamCompare<T>::equals(typed_param.getValue(), diff_typed_param->getValue()))
+			{
+				predicate.set(HAS_DEFAULT_VALUE);
+			}
+
+			predicate.set(VALID, typed_param.isValid());
+			predicate.set(PROVIDED, typed_param.anyProvided());
+			predicate.set(EMPTY, false);
+
+			if (!predicate_rule.check(predicate)) return false;
 
 			if (!name_stack.empty())
 			{
@@ -886,23 +937,25 @@ namespace LLInitParam
 
 			if (!key.empty())
 			{
-				if (!diff_param || !ParamCompare<std::string>::equals(static_cast<const self_t*>(diff_param)->getValueName(), key))
+				if (!diff_typed_param || !ParamCompare<std::string>::equals(diff_typed_param->getValueName(), key))
 				{
-					parser.writeValue(key, name_stack);
+					serialized = parser.writeValue(key, name_stack);
 				}
 			}
 			// then try to serialize value directly
-			else if (!diff_param || !ParamCompare<T>::equals(typed_param.getValue(), static_cast<const self_t*>(diff_param)->getValue()))
+			else if (!diff_typed_param || ParamCompare<T>::equals(typed_param.getValue(), diff_typed_param->getValue()))
 			{
-				if (!parser.writeValue(typed_param.getValue(), name_stack)) 
+				serialized = parser.writeValue(typed_param.getValue(), name_stack);
+				if (!serialized) 
 				{
 					std::string calculated_key = typed_param.calcValueName(typed_param.getValue());
-					if (!diff_param || !ParamCompare<std::string>::equals(static_cast<const self_t*>(diff_param)->getValueName(), calculated_key))
+					if (!diff_typed_param || !ParamCompare<std::string>::equals(diff_typed_param->getValueName(), calculated_key))
 					{
-						parser.writeValue(calculated_key, name_stack);
+						serialized = parser.writeValue(calculated_key, name_stack);
 					}
 				}
 			}
+			return serialized;
 		}
 
 		static void inspectParam(const Param& param, Parser& parser, Parser::name_stack_t& name_stack, S32 min_count, S32 max_count)
@@ -953,15 +1006,15 @@ namespace LLInitParam
 	};
 
 	// parameter that is a block
-	template <typename T, typename NAME_VALUE_LOOKUP>
-	class TypedParam<T, NAME_VALUE_LOOKUP, false, true> 
+	template <typename BLOCK_T, typename NAME_VALUE_LOOKUP>
+	class TypedParam<BLOCK_T, NAME_VALUE_LOOKUP, false, true> 
 	:	public Param,
-		public ParamValue<T, NAME_VALUE_LOOKUP>
+		public ParamValue<BLOCK_T, NAME_VALUE_LOOKUP>
 	{
 	public:
-		typedef ParamValue<T, NAME_VALUE_LOOKUP>				param_value_t;
+		typedef ParamValue<BLOCK_T, NAME_VALUE_LOOKUP>				param_value_t;
 		typedef typename param_value_t::value_assignment_t		value_assignment_t;
-		typedef TypedParam<T, NAME_VALUE_LOOKUP, false, true>	self_t;
+		typedef TypedParam<BLOCK_T, NAME_VALUE_LOOKUP, false, true>	self_t;
 		typedef NAME_VALUE_LOOKUP								name_value_lookup_t;
 
 		using param_value_t::operator();
@@ -1014,10 +1067,16 @@ namespace LLInitParam
 			return false;
 		}
 
-		static void serializeParam(const Param& param, Parser& parser, Parser::name_stack_t& name_stack, const Param* diff_param)
+		static bool serializeParam(const Param& param, Parser& parser, Parser::name_stack_t& name_stack, const predicate_rule_t predicate_rule, const Param* diff_param)
 		{
 			const self_t& typed_param = static_cast<const self_t&>(param);
-			if (!typed_param.isProvided()) return;
+
+			LLPredicate::Value<ESerializePredicates> predicate;
+
+			predicate.set(VALID, typed_param.isValid());
+			predicate.set(PROVIDED, typed_param.anyProvided());
+
+			if (!predicate_rule.check(predicate)) return false;
 
 			if (!name_stack.empty())
 			{
@@ -1027,15 +1086,17 @@ namespace LLInitParam
 			std::string key = typed_param.getValueName();
 			if (!key.empty())
 			{
-				if (!parser.writeValue(key, name_stack))
+				if (parser.writeValue(key, name_stack))
 				{
-					return;
+					return true;
 				}
 			}
 			else
 			{
-				typed_param.serializeBlock(parser, name_stack, static_cast<const self_t*>(diff_param));
+				return typed_param.serializeBlock(parser, name_stack, predicate_rule, static_cast<const self_t*>(diff_param));
 			}
+			
+			return false;
 		}
 
 		static void inspectParam(const Param& param, Parser& parser, Parser::name_stack_t& name_stack, S32 min_count, S32 max_count)
@@ -1049,23 +1110,16 @@ namespace LLInitParam
 		// *and* the block as a whole validates
 		bool isProvided() const 
 		{ 
-			// only validate block when it hasn't already passed validation with current data
-			if (Param::anyProvided() && !param_value_t::mValidated)
-			{
-				// a sub-block is "provided" when it has been filled in enough to be valid
-				param_value_t::mValidated = param_value_t::validateBlock(false);
-			}
-			return Param::anyProvided() && param_value_t::mValidated;
+			return Param::anyProvided() && isValid();
 		}
 
+		using param_value_t::isValid;
+
 		// assign block contents to this param-that-is-a-block
 		void set(value_assignment_t val, bool flag_as_provided = true)
 		{
 			setValue(val);
 			param_value_t::clearValueName();
-			// force revalidation of block
-			// next call to isProvided() will update provision status based on validity
-			param_value_t::mValidated = false;
 			setProvided(flag_as_provided);
 		}
 
@@ -1080,9 +1134,6 @@ namespace LLInitParam
 			param_value_t::paramChanged(changed_param, user_provided);
 			if (user_provided)
 			{
-				// a child param has been explicitly changed
-				// so *some* aspect of this block is now provided
-				param_value_t::mValidated = false;
 				setProvided();
 				param_value_t::clearValueName();
 			}
@@ -1120,13 +1171,13 @@ namespace LLInitParam
 	};
 
 	// container of non-block parameters
-	template <typename VALUE_TYPE, typename NAME_VALUE_LOOKUP>
-	class TypedParam<VALUE_TYPE, NAME_VALUE_LOOKUP, true, false> 
+	template <typename MULTI_VALUE_T, typename NAME_VALUE_LOOKUP>
+	class TypedParam<MULTI_VALUE_T, NAME_VALUE_LOOKUP, true, false> 
 	:	public Param
 	{
 	public:
-		typedef TypedParam<VALUE_TYPE, NAME_VALUE_LOOKUP, true, false>		self_t;
-		typedef ParamValue<VALUE_TYPE, NAME_VALUE_LOOKUP>					param_value_t;
+		typedef TypedParam<MULTI_VALUE_T, NAME_VALUE_LOOKUP, true, false>		self_t;
+		typedef ParamValue<MULTI_VALUE_T, NAME_VALUE_LOOKUP>					param_value_t;
 		typedef typename std::vector<param_value_t>							container_t;
 		typedef const container_t&											value_assignment_t;
 
@@ -1134,7 +1185,9 @@ namespace LLInitParam
 		typedef NAME_VALUE_LOOKUP											name_value_lookup_t;
 		
 		TypedParam(BlockDescriptor& block_descriptor, const char* name, value_assignment_t value, ParamDescriptor::validation_func_t validate_func, S32 min_count, S32 max_count) 
-		:	Param(block_descriptor.mCurrentBlockPtr)
+		:	Param(block_descriptor.mCurrentBlockPtr),
+			mMinCount(min_count),
+			mMaxCount(max_count)
 		{
 			std::copy(value.begin(), value.end(), std::back_inserter(mValues));
 
@@ -1152,14 +1205,28 @@ namespace LLInitParam
 			}
 		} 
 
-		bool isProvided() const { return Param::anyProvided(); }
+		bool isProvided() const { return Param::anyProvided() && isValid(); }
+
+		bool isValid() const 
+		{ 
+			size_t num_elements = numValidElements();
+			return mMinCount < num_elements && num_elements < mMaxCount;
+		}
 
 		static bool deserializeParam(Param& param, Parser& parser, const Parser::name_stack_range_t& name_stack_range, bool new_name)
 		{ 
+			Parser::name_stack_range_t new_name_stack_range(name_stack_range);
 			self_t& typed_param = static_cast<self_t&>(param);
 			value_t value;
+
+			// pop first element if empty string
+			if (new_name_stack_range.first != new_name_stack_range.second && new_name_stack_range.first->first.empty())
+			{
+				++new_name_stack_range.first;
+			}
+
 			// no further names in stack, attempt to parse value now
-			if (name_stack_range.first == name_stack_range.second)
+			if (new_name_stack_range.first == new_name_stack_range.second)
 			{
 				// attempt to read value directly
 				if (parser.readValue(value))
@@ -1189,17 +1256,26 @@ namespace LLInitParam
 			return false;
 		}
 
-		static void serializeParam(const Param& param, Parser& parser, Parser::name_stack_t& name_stack, const Param* diff_param)
+		static bool serializeParam(const Param& param, Parser& parser, Parser::name_stack_t& name_stack, const predicate_rule_t predicate_rule, const Param* diff_param)
 		{
+			bool serialized = false;
 			const self_t& typed_param = static_cast<const self_t&>(param);
-			if (!typed_param.isProvided() || name_stack.empty()) return;
+			
+			LLPredicate::Value<ESerializePredicates> predicate;
+
+			predicate.set(REQUIRED, typed_param.mMinCount > 0);
+			predicate.set(VALID, typed_param.isValid());
+			predicate.set(PROVIDED, typed_param.anyProvided());
+			predicate.set(EMPTY, typed_param.mValues.empty());
+
+			if (!predicate_rule.check(predicate)) return false;
 
 			for (const_iterator it = typed_param.mValues.begin(), end_it = typed_param.mValues.end();
 				it != end_it;
 				++it)
 			{
 				std::string key = it->getValueName();
-				name_stack.back().second = true;
+				name_stack.push_back(std::make_pair(std::string(), true));
 
 				if(key.empty())
 				// not parsed via name values, write out value directly
@@ -1208,7 +1284,11 @@ namespace LLInitParam
 					if (!value_written)
 					{
 						std::string calculated_key = it->calcValueName(it->getValue());
-						if (!parser.writeValue(calculated_key, name_stack))
+						if (parser.writeValue(calculated_key, name_stack))
+						{
+							serialized = true;
+						}
+						else
 						{
 							break;
 						}
@@ -1216,17 +1296,30 @@ namespace LLInitParam
 				}
 				else 
 				{
-					if(!parser.writeValue(key, name_stack))
+					if(parser.writeValue(key, name_stack))
+					{
+						serialized = true;
+					}
+					else
 					{
 						break;
 					}
 				}
+
+				name_stack.pop_back();
+			}
+
+			if (!serialized && predicate_rule.check(ll_make_predicate(EMPTY)))
+			{
+				serialized |= parser.writeValue(Flag(), name_stack);
 			}
+
+			return serialized;
 		}
 
 		static void inspectParam(const Param& param, Parser& parser, Parser::name_stack_t& name_stack, S32 min_count, S32 max_count)
 		{
-			parser.inspectValue<VALUE_TYPE>(name_stack, min_count, max_count, NULL);
+			parser.inspectValue<MULTI_VALUE_T>(name_stack, min_count, max_count, NULL);
 			if (name_value_lookup_t::getPossibleValues())
 			{
 				parser.inspectValue<std::string>(name_stack, min_count, max_count, name_value_lookup_t::getPossibleValues());
@@ -1283,7 +1376,7 @@ namespace LLInitParam
 		bool empty() const { return mValues.empty(); }
 		size_t size() const { return mValues.size(); }
 
-		U32 numValidElements() const
+		size_t numValidElements() const
 		{
 			return mValues.size();
 		}
@@ -1313,23 +1406,27 @@ namespace LLInitParam
 		}
 
 		container_t		mValues;
+		size_t			mMinCount,
+						mMaxCount;
 	};
 
 	// container of block parameters
-	template <typename VALUE_TYPE, typename NAME_VALUE_LOOKUP>
-	class TypedParam<VALUE_TYPE, NAME_VALUE_LOOKUP, true, true> 
+	template <typename MULTI_BLOCK_T, typename NAME_VALUE_LOOKUP>
+	class TypedParam<MULTI_BLOCK_T, NAME_VALUE_LOOKUP, true, true> 
 	:	public Param
 	{
 	public:
-		typedef TypedParam<VALUE_TYPE, NAME_VALUE_LOOKUP, true, true>	self_t;
-		typedef ParamValue<VALUE_TYPE, NAME_VALUE_LOOKUP>				param_value_t;
+		typedef TypedParam<MULTI_BLOCK_T, NAME_VALUE_LOOKUP, true, true>	self_t;
+		typedef ParamValue<MULTI_BLOCK_T, NAME_VALUE_LOOKUP>				param_value_t;
 		typedef typename std::vector<param_value_t>						container_t;
 		typedef const container_t&										value_assignment_t;
 		typedef typename param_value_t::value_t							value_t;
 		typedef NAME_VALUE_LOOKUP										name_value_lookup_t;
 
 		TypedParam(BlockDescriptor& block_descriptor, const char* name, value_assignment_t value, ParamDescriptor::validation_func_t validate_func, S32 min_count, S32 max_count) 
-		:	Param(block_descriptor.mCurrentBlockPtr)
+		:	Param(block_descriptor.mCurrentBlockPtr),
+			mMinCount(min_count),
+			mMaxCount(max_count)
 		{
 			std::copy(value.begin(), value.end(), back_inserter(mValues));
 
@@ -1347,14 +1444,30 @@ namespace LLInitParam
 			}
 		} 
 
-		bool isProvided() const { return Param::anyProvided(); }
+		bool isProvided() const { return Param::anyProvided() && isValid(); }
+
+		bool isValid() const 
+		{ 
+			size_t num_elements = numValidElements();
+			return mMinCount < num_elements && num_elements < mMaxCount;
+		}
+
 
 		static bool deserializeParam(Param& param, Parser& parser, const Parser::name_stack_range_t& name_stack_range, bool new_name) 
 		{ 
+			Parser::name_stack_range_t new_name_stack_range(name_stack_range);
 			self_t& typed_param = static_cast<self_t&>(param);
 			bool new_value = false;
+			bool new_array_value = false;
+
+			// pop first element if empty string
+			if (new_name_stack_range.first != new_name_stack_range.second && new_name_stack_range.first->first.empty())
+			{
+				new_array_value = new_name_stack_range.first->second;
+				++new_name_stack_range.first;
+			}
 
-			if (new_name || typed_param.mValues.empty())
+			if (new_name || new_array_value || typed_param.mValues.empty())
 			{
 				new_value = true;
 				typed_param.mValues.push_back(value_t());
@@ -1363,9 +1476,13 @@ namespace LLInitParam
 			param_value_t& value = typed_param.mValues.back();
 
 			// attempt to parse block...
-			if(value.deserializeBlock(parser, name_stack_range, new_name))
+			if(value.deserializeBlock(parser, new_name_stack_range, new_name))
 			{
 				typed_param.setProvided();
+				if (new_array_value)
+				{
+					name_stack_range.first->second = false;
+				}
 				return true;
 			}
 			else if(name_value_lookup_t::valueNamesExist())
@@ -1379,6 +1496,10 @@ namespace LLInitParam
 					{
 						typed_param.mValues.back().setValueName(name);
 						typed_param.setProvided();
+						if (new_array_value)
+						{
+							name_stack_range.first->second = false;
+						}
 						return true;
 					}
 
@@ -1393,29 +1514,47 @@ namespace LLInitParam
 			return false;
 		}
 
-		static void serializeParam(const Param& param, Parser& parser, Parser::name_stack_t& name_stack, const Param* diff_param)
+		static bool serializeParam(const Param& param, Parser& parser, Parser::name_stack_t& name_stack, const predicate_rule_t predicate_rule, const Param* diff_param)
 		{
+			bool serialized = false;
 			const self_t& typed_param = static_cast<const self_t&>(param);
-			if (!typed_param.isProvided() || name_stack.empty()) return;
+
+			LLPredicate::Value<ESerializePredicates> predicate;
+
+			predicate.set(REQUIRED, typed_param.mMinCount > 0);
+			predicate.set(VALID, typed_param.isValid());
+			predicate.set(PROVIDED, typed_param.anyProvided());
+			predicate.set(EMPTY, typed_param.mValues.empty());
+
+			if (!predicate_rule.check(predicate)) return false;
 
 			for (const_iterator it = typed_param.mValues.begin(), end_it = typed_param.mValues.end();
 				it != end_it;
 				++it)
 			{
-				name_stack.back().second = true;
+				name_stack.push_back(std::make_pair(std::string(), true));
 
 				std::string key = it->getValueName();
 				if (!key.empty())
 				{
-					parser.writeValue(key, name_stack);
+					serialized |= parser.writeValue(key, name_stack);
 				}
 				// Not parsed via named values, write out value directly
-				// NOTE: currently we don't worry about removing default values in Multiple
+				// NOTE: currently we don't do diffing of Multiples
 				else 
 				{
-					it->serializeBlock(parser, name_stack, NULL);
+					serialized = it->serializeBlock(parser, name_stack, predicate_rule, NULL);
 				}
+
+				name_stack.pop_back();
 			}
+
+			if (!serialized && predicate_rule.check(ll_make_predicate(EMPTY)))
+			{
+				serialized |= parser.writeValue(Flag(), name_stack);
+			}
+
+			return serialized;
 		}
 
 		static void inspectParam(const Param& param, Parser& parser, Parser::name_stack_t& name_stack, S32 min_count, S32 max_count)
@@ -1471,14 +1610,14 @@ namespace LLInitParam
 		bool empty() const { return mValues.empty(); }
 		size_t size() const { return mValues.size(); }
 
-		U32 numValidElements() const
+		size_t numValidElements() const
 		{
-			U32 count = 0;
+			size_t count = 0;
 			for (const_iterator it = mValues.begin(), end_it = mValues.end();
 				it != end_it;
 				++it)
 			{
-				if(it->validateBlock(false)) count++;
+				if(it->isValid()) count++;
 			}
 			return count;
 		}
@@ -1510,6 +1649,8 @@ namespace LLInitParam
 		}
 
 		container_t			mValues;
+		size_t				mMinCount,
+							mMaxCount;
 	};
 
 	template <typename DERIVED_BLOCK, typename BASE_BLOCK = BaseBlock>
@@ -1797,7 +1938,7 @@ namespace LLInitParam
 
 			static bool validate(const Param* paramp) 
 			{
-				U32 num_valid = ((super_t*)paramp)->numValidElements();
+				size_t num_valid = ((super_t*)paramp)->numValidElements();
 				return RANGE::minCount <= num_valid && num_valid <= RANGE::maxCount;
 			}
 		};
@@ -1914,13 +2055,11 @@ namespace LLInitParam
 		typedef block_t value_t;
 
 		ParamValue()
-		:	block_t(),
-			mValidated(false)
+		:	block_t()
 		{}
 
 		ParamValue(value_assignment_t other)
-		:	block_t(other),
-			mValidated(false)
+		:	block_t(other)
 		{
 		}
 
@@ -1948,9 +2087,6 @@ namespace LLInitParam
 		{
 			return *this;
 		}
-
-	protected:
-		mutable bool 	mValidated; // lazy validation flag
 	};
 
 	template<typename T, bool IS_BLOCK>
@@ -1965,13 +2101,11 @@ namespace LLInitParam
 		typedef T value_t;
 	
 		ParamValue()
-		:	mValue(),
-			mValidated(false)
+		:	mValue()
 		{}
 
 		ParamValue(value_assignment_t other)
-		:	mValue(other),
-			mValidated(false)
+		:	mValue(other)
 		{}
 
 		void setValue(value_assignment_t val)
@@ -2004,11 +2138,11 @@ namespace LLInitParam
 			return mValue.get().deserializeBlock(p, name_stack_range, new_name);
 		}
 
-		void serializeBlock(Parser& p, Parser::name_stack_t& name_stack, const BaseBlock* diff_block = NULL) const
+		bool serializeBlock(Parser& p, Parser::name_stack_t& name_stack, const predicate_rule_t predicate_rule, const BaseBlock* diff_block = NULL) const
 		{
-			if (mValue.empty()) return;
+			if (mValue.empty()) return false;
 			
-			mValue.get().serializeBlock(p, name_stack, diff_block);
+			return mValue.get().serializeBlock(p, name_stack, predicate_rule, diff_block);
 		}
 
 		bool inspectBlock(Parser& p, Parser::name_stack_t name_stack = Parser::name_stack_t(), S32 min_count = 0, S32 max_count = S32_MAX) const
@@ -2018,9 +2152,6 @@ namespace LLInitParam
 			return mValue.get().inspectBlock(p, name_stack, min_count, max_count);
 		}
 
-	protected:
-		mutable bool 	mValidated; // lazy validation flag
-
 	private:
 		BaseBlock::Lazy<T>	mValue;
 	};
@@ -2037,12 +2168,10 @@ namespace LLInitParam
 		typedef const LLSD&	value_assignment_t;
 
 		ParamValue()
-		:	mValidated(false)
 		{}
 
 		ParamValue(value_assignment_t other)
-		:	mValue(other),
-			mValidated(false)
+		:	mValue(other)
 		{}
 
 		void setValue(value_assignment_t val) { mValue = val; }
@@ -2056,16 +2185,13 @@ namespace LLInitParam
 
 		// block param interface
 		LL_COMMON_API bool deserializeBlock(Parser& p, Parser::name_stack_range_t name_stack_range, bool new_name);
-		LL_COMMON_API void serializeBlock(Parser& p, Parser::name_stack_t& name_stack, const BaseBlock* diff_block = NULL) const;
+		LL_COMMON_API bool serializeBlock(Parser& p, Parser::name_stack_t& name_stack, const predicate_rule_t predicate_rule, const BaseBlock* diff_block = NULL) const;
 		bool inspectBlock(Parser& p, Parser::name_stack_t name_stack = Parser::name_stack_t(), S32 min_count = 0, S32 max_count = S32_MAX) const
 		{
 			//TODO: implement LLSD params as schema type Any
 			return true;
 		}
 
-	protected:
-		mutable bool 	mValidated; // lazy validation flag
-
 	private:
 		static void serializeElement(Parser& p, const LLSD& sd, Parser::name_stack_t& name_stack);
 
@@ -2094,8 +2220,7 @@ namespace LLInitParam
 
 		CustomParamValue(const T& value = T())
 		:	mValue(value),
-			mValueAge(VALUE_AUTHORITATIVE),
-			mValidated(false)
+			mValueAge(VALUE_AUTHORITATIVE)
 		{}
 
 		bool deserializeBlock(Parser& parser, Parser::name_stack_range_t name_stack_range, bool new_name)
@@ -2119,7 +2244,7 @@ namespace LLInitParam
 			return typed_param.BaseBlock::deserializeBlock(parser, name_stack_range, new_name);
 		}
 
-		void serializeBlock(Parser& parser, Parser::name_stack_t& name_stack, const BaseBlock* diff_block = NULL) const
+		bool serializeBlock(Parser& parser, Parser::name_stack_t& name_stack, const predicate_rule_t predicate_rule, const BaseBlock* diff_block = NULL) const
 		{
 			const derived_t& typed_param = static_cast<const derived_t&>(*this);
 			const derived_t* diff_param = static_cast<const derived_t*>(diff_block);
@@ -2131,14 +2256,18 @@ namespace LLInitParam
 			{
 				if (!diff_param || !ParamCompare<std::string>::equals(diff_param->getValueName(), key))
 				{
-					parser.writeValue(key, name_stack);
+					return parser.writeValue(key, name_stack);
 				}
 			}
 			// then try to serialize value directly
 			else if (!diff_param || !ParamCompare<T>::equals(typed_param.getValue(), diff_param->getValue()))
             {
 				
-				if (!parser.writeValue(typed_param.getValue(), name_stack)) 
+				if (parser.writeValue(typed_param.getValue(), name_stack)) 
+				{
+					return true;
+				}
+				else
 				{
 					//RN: *always* serialize provided components of BlockValue (don't pass diff_param on),
 					// since these tend to be viewed as the constructor arguments for the value T.  It seems
@@ -2155,14 +2284,15 @@ namespace LLInitParam
 						// and serialize those params
 						derived_t copy(typed_param);
 						copy.updateBlockFromValue(true);
-						copy.block_t::serializeBlock(parser, name_stack, NULL);
+						return copy.block_t::serializeBlock(parser, name_stack, predicate_rule, NULL);
 					}
 					else
 					{
-						block_t::serializeBlock(parser, name_stack, NULL);
+						return block_t::serializeBlock(parser, name_stack, predicate_rule, NULL);
 					}
 				}
 			}
+			return false;
 		}
 
 		bool inspectBlock(Parser& parser, Parser::name_stack_t name_stack = Parser::name_stack_t(), S32 min_count = 0, S32 max_count = S32_MAX) const
@@ -2280,8 +2410,6 @@ namespace LLInitParam
 			return block_t::mergeBlock(block_data, source, overwrite);
 		}
 
-		mutable bool 		mValidated; // lazy validation flag
-
 	private:
 		mutable T			mValue;
 		mutable EValueAge	mValueAge;
diff --git a/indra/llcommon/llinstancetracker.cpp b/indra/llcommon/llinstancetracker.cpp
index 5dc3ea5d7b..071a637cda 100644
--- a/indra/llcommon/llinstancetracker.cpp
+++ b/indra/llcommon/llinstancetracker.cpp
@@ -27,6 +27,8 @@
 #include "linden_common.h"
 // associated header
 #include "llinstancetracker.h"
+#include "llapr.h"
+
 // STL headers
 // std headers
 // external library headers
@@ -47,3 +49,19 @@ void * & LLInstanceTrackerBase::getInstances(std::type_info const & info)
 													 InstancesMap::mapped_type()))
 		.first->second;
 }
+
+void LLInstanceTrackerBase::StaticBase::incrementDepth()
+{
+	apr_atomic_inc32(&sIterationNestDepth);
+}
+
+void LLInstanceTrackerBase::StaticBase::decrementDepth()
+{
+	apr_atomic_dec32(&sIterationNestDepth);
+}
+
+U32 LLInstanceTrackerBase::StaticBase::getDepth()
+{
+	apr_uint32_t data = apr_atomic_read32(&sIterationNestDepth);
+	return data;
+}
diff --git a/indra/llcommon/llinstancetracker.h b/indra/llcommon/llinstancetracker.h
index 403df08990..9dd6d4a7ed 100644
--- a/indra/llcommon/llinstancetracker.h
+++ b/indra/llcommon/llinstancetracker.h
@@ -70,15 +70,20 @@ protected:
         StaticBase():
             sIterationNestDepth(0)
         {}
-        S32 sIterationNestDepth;
+
+		void incrementDepth();
+		void decrementDepth();
+		U32 getDepth();
+	private:
+		U32 sIterationNestDepth;
     };
 };
 
 /// This mix-in class adds support for tracking all instances of the specified class parameter T
 /// The (optional) key associates a value of type KEY with a given instance of T, for quick lookup
 /// If KEY is not provided, then instances are stored in a simple set
-/// @NOTE: see explicit specialization below for default KEY==T* case
-template<typename T, typename KEY = T*>
+/// @NOTE: see explicit specialization below for default KEY==void case
+template<typename T, typename KEY = void>
 class LLInstanceTracker : public LLInstanceTrackerBase
 {
 	typedef LLInstanceTracker<T, KEY> MyT;
@@ -99,12 +104,12 @@ public:
 		instance_iter(const typename InstanceMap::iterator& it)
 		:	mIterator(it)
 		{
-			++getStatic().sIterationNestDepth;
+			getStatic().incrementDepth();
 		}
 
 		~instance_iter()
 		{
-			--getStatic().sIterationNestDepth;
+			getStatic().decrementDepth();
 		}
 
 
@@ -133,18 +138,18 @@ public:
 		key_iter(typename InstanceMap::iterator it)
 			:	mIterator(it)
 		{
-			++getStatic().sIterationNestDepth;
+			getStatic().incrementDepth();
 		}
 
 		key_iter(const key_iter& other)
 			:	mIterator(other.mIterator)
 		{
-			++getStatic().sIterationNestDepth;
+			getStatic().incrementDepth();
 		}
 
 		~key_iter()
 		{
-			--getStatic().sIterationNestDepth;
+			getStatic().decrementDepth();
 		}
 
 
@@ -203,7 +208,7 @@ protected:
 	virtual ~LLInstanceTracker() 
 	{ 
 		// it's unsafe to delete instances of this type while all instances are being iterated over.
-		llassert_always(getStatic().sIterationNestDepth == 0);
+		llassert_always(getStatic().getDepth() == 0);
 		remove_();		
 	}
 	virtual void setKey(KEY key) { remove_(); add_(key); }
@@ -224,12 +229,12 @@ private:
 	KEY mInstanceKey;
 };
 
-/// explicit specialization for default case where KEY is T*
+/// explicit specialization for default case where KEY is void
 /// use a simple std::set<T*>
 template<typename T>
-class LLInstanceTracker<T, T*> : public LLInstanceTrackerBase
+class LLInstanceTracker<T, void> : public LLInstanceTrackerBase
 {
-	typedef LLInstanceTracker<T, T*> MyT;
+	typedef LLInstanceTracker<T, void> MyT;
 	typedef typename std::set<T*> InstanceSet;
 	struct StaticData: public StaticBase
 	{
@@ -262,18 +267,18 @@ public:
 		instance_iter(const typename InstanceSet::iterator& it)
 		:	mIterator(it)
 		{
-			++getStatic().sIterationNestDepth;
+			getStatic().incrementDepth();
 		}
 
 		instance_iter(const instance_iter& other)
 		:	mIterator(other.mIterator)
 		{
-			++getStatic().sIterationNestDepth;
+			getStatic().incrementDepth();
 		}
 
 		~instance_iter()
 		{
-			--getStatic().sIterationNestDepth;
+			getStatic().decrementDepth();
 		}
 
 	private:
@@ -306,7 +311,7 @@ protected:
 	virtual ~LLInstanceTracker()
 	{
 		// it's unsafe to delete instances of this type while all instances are being iterated over.
-		llassert_always(getStatic().sIterationNestDepth == 0);
+		llassert_always(getStatic().getDepth() == 0);
 		getSet_().erase(static_cast<T*>(this));
 	}
 
diff --git a/indra/llcommon/llleap.cpp b/indra/llcommon/llleap.cpp
index 0a57ef1c48..84d2a12f65 100644
--- a/indra/llcommon/llleap.cpp
+++ b/indra/llcommon/llleap.cpp
@@ -394,7 +394,7 @@ public:
         LLProcess::WritePipe& childin(mChild->getWritePipe(LLProcess::STDIN));
         LLEventPump& mainloop(LLEventPumps::instance().obtain("mainloop"));
         LLSD nop;
-        F64 until(LLTimer::getElapsedSeconds() + 2);
+        F64 until = (LLTimer::getElapsedSeconds() + 2).value();
         while (childin.size() && LLTimer::getElapsedSeconds() < until)
         {
             mainloop.post(nop);
diff --git a/indra/llcommon/llmd5.cpp b/indra/llcommon/llmd5.cpp
index 1409c55d1c..ed80af36d8 100644
--- a/indra/llcommon/llmd5.cpp
+++ b/indra/llcommon/llmd5.cpp
@@ -76,7 +76,6 @@ documentation and/or software.
 
 #include "llmd5.h"
 
-#include <cassert>
 #include <iostream>		// cerr
 
 // how many bytes to grab at a time when checking files
diff --git a/indra/llcommon/llmemory.cpp b/indra/llcommon/llmemory.cpp
index 70ad10ad55..c6b02df939 100644
--- a/indra/llcommon/llmemory.cpp
+++ b/indra/llcommon/llmemory.cpp
@@ -32,7 +32,6 @@
 //#endif
 
 #if defined(LL_WINDOWS)
-//# include <windows.h>
 # include <psapi.h>
 #elif defined(LL_DARWIN)
 # include <sys/types.h>
diff --git a/indra/llcommon/llmemory.h b/indra/llcommon/llmemory.h
index 10013e0f92..4ead45679f 100644
--- a/indra/llcommon/llmemory.h
+++ b/indra/llcommon/llmemory.h
@@ -26,7 +26,12 @@
 #ifndef LLMEMORY_H
 #define LLMEMORY_H
 
-#include "llmemtype.h"
+#include "linden_common.h"
+#if !LL_WINDOWS
+#include <stdint.h>
+#endif
+
+class LLMutex ;
 
 #if LL_WINDOWS && LL_DEBUG
 #define LL_CHECK_MEMORY llassert(_CrtCheckMemory());
diff --git a/indra/llcommon/llmemtype.cpp b/indra/llcommon/llmemtype.cpp
deleted file mode 100644
index 6290a7158f..0000000000
--- a/indra/llcommon/llmemtype.cpp
+++ /dev/null
@@ -1,232 +0,0 @@
-/** 
- * @file llmemtype.cpp
- * @brief Simple memory allocation/deallocation tracking stuff here
- *
- * $LicenseInfo:firstyear=2002&license=viewerlgpl$
- * Second Life Viewer Source Code
- * Copyright (C) 2010, Linden Research, Inc.
- * 
- * This library is free software; you can redistribute it and/or
- * modify it under the terms of the GNU Lesser General Public
- * License as published by the Free Software Foundation;
- * version 2.1 of the License only.
- * 
- * This library is distributed in the hope that it will be useful,
- * but WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
- * Lesser General Public License for more details.
- * 
- * You should have received a copy of the GNU Lesser General Public
- * License along with this library; if not, write to the Free Software
- * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA
- * 
- * Linden Research, Inc., 945 Battery Street, San Francisco, CA  94111  USA
- * $/LicenseInfo$
- */
-
-#include "llmemtype.h"
-#include "llallocator.h"
-
-std::vector<char const *> LLMemType::DeclareMemType::mNameList;
-
-LLMemType::DeclareMemType LLMemType::MTYPE_INIT("Init");
-LLMemType::DeclareMemType LLMemType::MTYPE_STARTUP("Startup");
-LLMemType::DeclareMemType LLMemType::MTYPE_MAIN("Main");
-LLMemType::DeclareMemType LLMemType::MTYPE_FRAME("Frame");
-
-LLMemType::DeclareMemType LLMemType::MTYPE_GATHER_INPUT("GatherInput");
-LLMemType::DeclareMemType LLMemType::MTYPE_JOY_KEY("JoyKey");
-
-LLMemType::DeclareMemType LLMemType::MTYPE_IDLE("Idle");
-LLMemType::DeclareMemType LLMemType::MTYPE_IDLE_PUMP("IdlePump");
-LLMemType::DeclareMemType LLMemType::MTYPE_IDLE_NETWORK("IdleNetwork");
-LLMemType::DeclareMemType LLMemType::MTYPE_IDLE_UPDATE_REGIONS("IdleUpdateRegions");
-LLMemType::DeclareMemType LLMemType::MTYPE_IDLE_UPDATE_VIEWER_REGION("IdleUpdateViewerRegion");
-LLMemType::DeclareMemType LLMemType::MTYPE_IDLE_UPDATE_SURFACE("IdleUpdateSurface");
-LLMemType::DeclareMemType LLMemType::MTYPE_IDLE_UPDATE_PARCEL_OVERLAY("IdleUpdateParcelOverlay");
-LLMemType::DeclareMemType LLMemType::MTYPE_IDLE_AUDIO("IdleAudio");
-
-LLMemType::DeclareMemType LLMemType::MTYPE_CACHE_PROCESS_PENDING("CacheProcessPending");
-LLMemType::DeclareMemType LLMemType::MTYPE_CACHE_PROCESS_PENDING_ASKS("CacheProcessPendingAsks");
-LLMemType::DeclareMemType LLMemType::MTYPE_CACHE_PROCESS_PENDING_REPLIES("CacheProcessPendingReplies");
-
-LLMemType::DeclareMemType LLMemType::MTYPE_MESSAGE_CHECK_ALL("MessageCheckAll");
-LLMemType::DeclareMemType LLMemType::MTYPE_MESSAGE_PROCESS_ACKS("MessageProcessAcks");
-
-LLMemType::DeclareMemType LLMemType::MTYPE_RENDER("Render");
-LLMemType::DeclareMemType LLMemType::MTYPE_SLEEP("Sleep");
-
-LLMemType::DeclareMemType LLMemType::MTYPE_NETWORK("Network");
-LLMemType::DeclareMemType LLMemType::MTYPE_PHYSICS("Physics");
-LLMemType::DeclareMemType LLMemType::MTYPE_INTERESTLIST("InterestList");
-
-LLMemType::DeclareMemType LLMemType::MTYPE_IMAGEBASE("ImageBase");
-LLMemType::DeclareMemType LLMemType::MTYPE_IMAGERAW("ImageRaw");
-LLMemType::DeclareMemType LLMemType::MTYPE_IMAGEFORMATTED("ImageFormatted");
-		
-LLMemType::DeclareMemType LLMemType::MTYPE_APPFMTIMAGE("AppFmtImage");
-LLMemType::DeclareMemType LLMemType::MTYPE_APPRAWIMAGE("AppRawImage");
-LLMemType::DeclareMemType LLMemType::MTYPE_APPAUXRAWIMAGE("AppAuxRawImage");
-		
-LLMemType::DeclareMemType LLMemType::MTYPE_DRAWABLE("Drawable");
-
-LLMemType::DeclareMemType LLMemType::MTYPE_OBJECT("Object");
-LLMemType::DeclareMemType LLMemType::MTYPE_OBJECT_PROCESS_UPDATE("ObjectProcessUpdate");
-LLMemType::DeclareMemType LLMemType::MTYPE_OBJECT_PROCESS_UPDATE_CORE("ObjectProcessUpdateCore");
-
-LLMemType::DeclareMemType LLMemType::MTYPE_DISPLAY("Display");
-LLMemType::DeclareMemType LLMemType::MTYPE_DISPLAY_UPDATE("DisplayUpdate");
-LLMemType::DeclareMemType LLMemType::MTYPE_DISPLAY_UPDATE_CAMERA("DisplayUpdateCam");
-LLMemType::DeclareMemType LLMemType::MTYPE_DISPLAY_UPDATE_GEOM("DisplayUpdateGeom");
-LLMemType::DeclareMemType LLMemType::MTYPE_DISPLAY_SWAP("DisplaySwap");
-LLMemType::DeclareMemType LLMemType::MTYPE_DISPLAY_UPDATE_HUD("DisplayUpdateHud");
-LLMemType::DeclareMemType LLMemType::MTYPE_DISPLAY_GEN_REFLECTION("DisplayGenRefl");
-LLMemType::DeclareMemType LLMemType::MTYPE_DISPLAY_IMAGE_UPDATE("DisplayImageUpdate");
-LLMemType::DeclareMemType LLMemType::MTYPE_DISPLAY_STATE_SORT("DisplayStateSort");
-LLMemType::DeclareMemType LLMemType::MTYPE_DISPLAY_SKY("DisplaySky");
-LLMemType::DeclareMemType LLMemType::MTYPE_DISPLAY_RENDER_GEOM("DisplayRenderGeom");
-LLMemType::DeclareMemType LLMemType::MTYPE_DISPLAY_RENDER_FLUSH("DisplayRenderFlush");
-LLMemType::DeclareMemType LLMemType::MTYPE_DISPLAY_RENDER_UI("DisplayRenderUI");
-LLMemType::DeclareMemType LLMemType::MTYPE_DISPLAY_RENDER_ATTACHMENTS("DisplayRenderAttach");
-
-LLMemType::DeclareMemType LLMemType::MTYPE_VERTEX_DATA("VertexData");
-LLMemType::DeclareMemType LLMemType::MTYPE_VERTEX_CONSTRUCTOR("VertexConstr");
-LLMemType::DeclareMemType LLMemType::MTYPE_VERTEX_DESTRUCTOR("VertexDestr");
-LLMemType::DeclareMemType LLMemType::MTYPE_VERTEX_CREATE_VERTICES("VertexCreateVerts");
-LLMemType::DeclareMemType LLMemType::MTYPE_VERTEX_CREATE_INDICES("VertexCreateIndices");
-LLMemType::DeclareMemType LLMemType::MTYPE_VERTEX_DESTROY_BUFFER("VertexDestroyBuff");	
-LLMemType::DeclareMemType LLMemType::MTYPE_VERTEX_DESTROY_INDICES("VertexDestroyIndices");
-LLMemType::DeclareMemType LLMemType::MTYPE_VERTEX_UPDATE_VERTS("VertexUpdateVerts");
-LLMemType::DeclareMemType LLMemType::MTYPE_VERTEX_UPDATE_INDICES("VertexUpdateIndices");
-LLMemType::DeclareMemType LLMemType::MTYPE_VERTEX_ALLOCATE_BUFFER("VertexAllocateBuffer");
-LLMemType::DeclareMemType LLMemType::MTYPE_VERTEX_RESIZE_BUFFER("VertexResizeBuffer");
-LLMemType::DeclareMemType LLMemType::MTYPE_VERTEX_MAP_BUFFER("VertexMapBuffer");
-LLMemType::DeclareMemType LLMemType::MTYPE_VERTEX_MAP_BUFFER_VERTICES("VertexMapBufferVerts");
-LLMemType::DeclareMemType LLMemType::MTYPE_VERTEX_MAP_BUFFER_INDICES("VertexMapBufferIndices");
-LLMemType::DeclareMemType LLMemType::MTYPE_VERTEX_UNMAP_BUFFER("VertexUnmapBuffer");
-LLMemType::DeclareMemType LLMemType::MTYPE_VERTEX_SET_STRIDE("VertexSetStride");
-LLMemType::DeclareMemType LLMemType::MTYPE_VERTEX_SET_BUFFER("VertexSetBuffer");
-LLMemType::DeclareMemType LLMemType::MTYPE_VERTEX_SETUP_VERTEX_BUFFER("VertexSetupVertBuff");
-LLMemType::DeclareMemType LLMemType::MTYPE_VERTEX_CLEANUP_CLASS("VertexCleanupClass");
-
-LLMemType::DeclareMemType LLMemType::MTYPE_SPACE_PARTITION("SpacePartition");
-
-LLMemType::DeclareMemType LLMemType::MTYPE_PIPELINE("Pipeline");
-LLMemType::DeclareMemType LLMemType::MTYPE_PIPELINE_INIT("PipelineInit");
-LLMemType::DeclareMemType LLMemType::MTYPE_PIPELINE_CREATE_BUFFERS("PipelineCreateBuffs");
-LLMemType::DeclareMemType LLMemType::MTYPE_PIPELINE_RESTORE_GL("PipelineRestroGL");
-LLMemType::DeclareMemType LLMemType::MTYPE_PIPELINE_UNLOAD_SHADERS("PipelineUnloadShaders");
-LLMemType::DeclareMemType LLMemType::MTYPE_PIPELINE_LIGHTING_DETAIL("PipelineLightingDetail");
-LLMemType::DeclareMemType LLMemType::MTYPE_PIPELINE_GET_POOL_TYPE("PipelineGetPoolType");
-LLMemType::DeclareMemType LLMemType::MTYPE_PIPELINE_ADD_POOL("PipelineAddPool");
-LLMemType::DeclareMemType LLMemType::MTYPE_PIPELINE_ALLOCATE_DRAWABLE("PipelineAllocDrawable");
-LLMemType::DeclareMemType LLMemType::MTYPE_PIPELINE_ADD_OBJECT("PipelineAddObj");
-LLMemType::DeclareMemType LLMemType::MTYPE_PIPELINE_CREATE_OBJECTS("PipelineCreateObjs");
-LLMemType::DeclareMemType LLMemType::MTYPE_PIPELINE_UPDATE_MOVE("PipelineUpdateMove");
-LLMemType::DeclareMemType LLMemType::MTYPE_PIPELINE_UPDATE_GEOM("PipelineUpdateGeom");
-LLMemType::DeclareMemType LLMemType::MTYPE_PIPELINE_MARK_VISIBLE("PipelineMarkVisible");
-LLMemType::DeclareMemType LLMemType::MTYPE_PIPELINE_MARK_MOVED("PipelineMarkMoved");
-LLMemType::DeclareMemType LLMemType::MTYPE_PIPELINE_MARK_SHIFT("PipelineMarkShift");
-LLMemType::DeclareMemType LLMemType::MTYPE_PIPELINE_SHIFT_OBJECTS("PipelineShiftObjs");
-LLMemType::DeclareMemType LLMemType::MTYPE_PIPELINE_MARK_TEXTURED("PipelineMarkTextured");
-LLMemType::DeclareMemType LLMemType::MTYPE_PIPELINE_MARK_REBUILD("PipelineMarkRebuild");
-LLMemType::DeclareMemType LLMemType::MTYPE_PIPELINE_UPDATE_CULL("PipelineUpdateCull");
-LLMemType::DeclareMemType LLMemType::MTYPE_PIPELINE_STATE_SORT("PipelineStateSort");
-LLMemType::DeclareMemType LLMemType::MTYPE_PIPELINE_POST_SORT("PipelinePostSort");
-		
-LLMemType::DeclareMemType LLMemType::MTYPE_PIPELINE_RENDER_HUD_ELS("PipelineHudEls");
-LLMemType::DeclareMemType LLMemType::MTYPE_PIPELINE_RENDER_HL("PipelineRenderHL");
-LLMemType::DeclareMemType LLMemType::MTYPE_PIPELINE_RENDER_GEOM("PipelineRenderGeom");
-LLMemType::DeclareMemType LLMemType::MTYPE_PIPELINE_RENDER_GEOM_DEFFERRED("PipelineRenderGeomDef");
-LLMemType::DeclareMemType LLMemType::MTYPE_PIPELINE_RENDER_GEOM_POST_DEF("PipelineRenderGeomPostDef");
-LLMemType::DeclareMemType LLMemType::MTYPE_PIPELINE_RENDER_GEOM_SHADOW("PipelineRenderGeomShadow");
-LLMemType::DeclareMemType LLMemType::MTYPE_PIPELINE_RENDER_SELECT("PipelineRenderSelect");
-LLMemType::DeclareMemType LLMemType::MTYPE_PIPELINE_REBUILD_POOLS("PipelineRebuildPools");
-LLMemType::DeclareMemType LLMemType::MTYPE_PIPELINE_QUICK_LOOKUP("PipelineQuickLookup");
-LLMemType::DeclareMemType LLMemType::MTYPE_PIPELINE_RENDER_OBJECTS("PipelineRenderObjs");
-LLMemType::DeclareMemType LLMemType::MTYPE_PIPELINE_GENERATE_IMPOSTOR("PipelineGenImpostors");
-LLMemType::DeclareMemType LLMemType::MTYPE_PIPELINE_RENDER_BLOOM("PipelineRenderBloom");
-
-LLMemType::DeclareMemType LLMemType::MTYPE_UPKEEP_POOLS("UpkeepPools");
-
-LLMemType::DeclareMemType LLMemType::MTYPE_AVATAR("Avatar");
-LLMemType::DeclareMemType LLMemType::MTYPE_AVATAR_MESH("AvatarMesh");
-LLMemType::DeclareMemType LLMemType::MTYPE_PARTICLES("Particles");
-LLMemType::DeclareMemType LLMemType::MTYPE_REGIONS("Regions");
-
-LLMemType::DeclareMemType LLMemType::MTYPE_INVENTORY("Inventory");
-LLMemType::DeclareMemType LLMemType::MTYPE_INVENTORY_DRAW("InventoryDraw");
-LLMemType::DeclareMemType LLMemType::MTYPE_INVENTORY_BUILD_NEW_VIEWS("InventoryBuildNewViews");
-LLMemType::DeclareMemType LLMemType::MTYPE_INVENTORY_DO_FOLDER("InventoryDoFolder");
-LLMemType::DeclareMemType LLMemType::MTYPE_INVENTORY_POST_BUILD("InventoryPostBuild");
-LLMemType::DeclareMemType LLMemType::MTYPE_INVENTORY_FROM_XML("InventoryFromXML");
-LLMemType::DeclareMemType LLMemType::MTYPE_INVENTORY_CREATE_NEW_ITEM("InventoryCreateNewItem");
-LLMemType::DeclareMemType LLMemType::MTYPE_INVENTORY_VIEW_INIT("InventoryViewInit");
-LLMemType::DeclareMemType LLMemType::MTYPE_INVENTORY_VIEW_SHOW("InventoryViewShow");
-LLMemType::DeclareMemType LLMemType::MTYPE_INVENTORY_VIEW_TOGGLE("InventoryViewToggle");
-
-LLMemType::DeclareMemType LLMemType::MTYPE_ANIMATION("Animation");
-LLMemType::DeclareMemType LLMemType::MTYPE_VOLUME("Volume");
-LLMemType::DeclareMemType LLMemType::MTYPE_PRIMITIVE("Primitive");
-		
-LLMemType::DeclareMemType LLMemType::MTYPE_SCRIPT("Script");
-LLMemType::DeclareMemType LLMemType::MTYPE_SCRIPT_RUN("ScriptRun");
-LLMemType::DeclareMemType LLMemType::MTYPE_SCRIPT_BYTECODE("ScriptByteCode");
-		
-LLMemType::DeclareMemType LLMemType::MTYPE_IO_PUMP("IoPump");
-LLMemType::DeclareMemType LLMemType::MTYPE_IO_TCP("IoTCP");
-LLMemType::DeclareMemType LLMemType::MTYPE_IO_BUFFER("IoBuffer");
-LLMemType::DeclareMemType LLMemType::MTYPE_IO_HTTP_SERVER("IoHttpServer");
-LLMemType::DeclareMemType LLMemType::MTYPE_IO_SD_SERVER("IoSDServer");
-LLMemType::DeclareMemType LLMemType::MTYPE_IO_SD_CLIENT("IoSDClient");
-LLMemType::DeclareMemType LLMemType::MTYPE_IO_URL_REQUEST("IOUrlRequest");
-
-LLMemType::DeclareMemType LLMemType::MTYPE_DIRECTX_INIT("DirectXInit");
-
-LLMemType::DeclareMemType LLMemType::MTYPE_TEMP1("Temp1");
-LLMemType::DeclareMemType LLMemType::MTYPE_TEMP2("Temp2");
-LLMemType::DeclareMemType LLMemType::MTYPE_TEMP3("Temp3");
-LLMemType::DeclareMemType LLMemType::MTYPE_TEMP4("Temp4");
-LLMemType::DeclareMemType LLMemType::MTYPE_TEMP5("Temp5");
-LLMemType::DeclareMemType LLMemType::MTYPE_TEMP6("Temp6");
-LLMemType::DeclareMemType LLMemType::MTYPE_TEMP7("Temp7");
-LLMemType::DeclareMemType LLMemType::MTYPE_TEMP8("Temp8");
-LLMemType::DeclareMemType LLMemType::MTYPE_TEMP9("Temp9");
-
-LLMemType::DeclareMemType LLMemType::MTYPE_OTHER("Other");
-
-
-LLMemType::DeclareMemType::DeclareMemType(char const * st)
-{
-	mID = (S32)mNameList.size();
-	mName = st;
-
-	mNameList.push_back(mName);
-}
-
-LLMemType::DeclareMemType::~DeclareMemType()
-{
-}
-
-LLMemType::LLMemType(LLMemType::DeclareMemType& dt)
-{
-	mTypeIndex = dt.mID;
-	LLAllocator::pushMemType(dt.mID);
-}
-
-LLMemType::~LLMemType()
-{
-	LLAllocator::popMemType();
-}
-
-char const * LLMemType::getNameFromID(S32 id)
-{
-	if (id < 0 || id >= (S32)DeclareMemType::mNameList.size())
-	{
-		return "INVALID";
-	}
-
-	return DeclareMemType::mNameList[id];
-}
-
-//--------------------------------------------------------------------------------------------------
diff --git a/indra/llcommon/llmemtype.h b/indra/llcommon/llmemtype.h
deleted file mode 100644
index 4945dbaf60..0000000000
--- a/indra/llcommon/llmemtype.h
+++ /dev/null
@@ -1,242 +0,0 @@
-/** 
- * @file llmemtype.h
- * @brief Runtime memory usage debugging utilities.
- *
- * $LicenseInfo:firstyear=2005&license=viewerlgpl$
- * Second Life Viewer Source Code
- * Copyright (C) 2010, Linden Research, Inc.
- * 
- * This library is free software; you can redistribute it and/or
- * modify it under the terms of the GNU Lesser General Public
- * License as published by the Free Software Foundation;
- * version 2.1 of the License only.
- * 
- * This library is distributed in the hope that it will be useful,
- * but WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
- * Lesser General Public License for more details.
- * 
- * You should have received a copy of the GNU Lesser General Public
- * License along with this library; if not, write to the Free Software
- * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA
- * 
- * Linden Research, Inc., 945 Battery Street, San Francisco, CA  94111  USA
- * $/LicenseInfo$
- */
-
-#ifndef LL_MEMTYPE_H
-#define LL_MEMTYPE_H
-
-//----------------------------------------------------------------------------
-//----------------------------------------------------------------------------
-
-//----------------------------------------------------------------------------
-
-#include "linden_common.h"
-//!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
-// WARNING: Never commit with MEM_TRACK_MEM == 1
-//!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
-#define MEM_TRACK_MEM (0 && LL_WINDOWS)
-
-#include <vector>
-
-#define MEM_TYPE_NEW(T)
-
-class LL_COMMON_API LLMemType
-{
-public:
-
-	// class we'll initialize all instances of as
-	// static members of MemType.  Then use
-	// to construct any new mem type.
-	class LL_COMMON_API DeclareMemType
-	{
-	public:
-		DeclareMemType(char const * st);
-		~DeclareMemType();
-	
-		S32 mID;
-		char const * mName;
-		
-		// array so we can map an index ID to Name
-		static std::vector<char const *> mNameList;
-	};
-
-	LLMemType(DeclareMemType& dt);
-	~LLMemType();
-
-	static char const * getNameFromID(S32 id);
-
-	static DeclareMemType MTYPE_INIT;
-	static DeclareMemType MTYPE_STARTUP;
-	static DeclareMemType MTYPE_MAIN;
-	static DeclareMemType MTYPE_FRAME;
-
-	static DeclareMemType MTYPE_GATHER_INPUT;
-	static DeclareMemType MTYPE_JOY_KEY;
-
-	static DeclareMemType MTYPE_IDLE;
-	static DeclareMemType MTYPE_IDLE_PUMP;
-	static DeclareMemType MTYPE_IDLE_NETWORK;
-	static DeclareMemType MTYPE_IDLE_UPDATE_REGIONS;
-	static DeclareMemType MTYPE_IDLE_UPDATE_VIEWER_REGION;
-	static DeclareMemType MTYPE_IDLE_UPDATE_SURFACE;
-	static DeclareMemType MTYPE_IDLE_UPDATE_PARCEL_OVERLAY;
-	static DeclareMemType MTYPE_IDLE_AUDIO;
-
-	static DeclareMemType MTYPE_CACHE_PROCESS_PENDING;
-	static DeclareMemType MTYPE_CACHE_PROCESS_PENDING_ASKS;
-	static DeclareMemType MTYPE_CACHE_PROCESS_PENDING_REPLIES;
-
-	static DeclareMemType MTYPE_MESSAGE_CHECK_ALL;
-	static DeclareMemType MTYPE_MESSAGE_PROCESS_ACKS;
-
-	static DeclareMemType MTYPE_RENDER;
-	static DeclareMemType MTYPE_SLEEP;
-
-	static DeclareMemType MTYPE_NETWORK;
-	static DeclareMemType MTYPE_PHYSICS;
-	static DeclareMemType MTYPE_INTERESTLIST;
-
-	static DeclareMemType MTYPE_IMAGEBASE;
-	static DeclareMemType MTYPE_IMAGERAW;
-	static DeclareMemType MTYPE_IMAGEFORMATTED;
-	
-	static DeclareMemType MTYPE_APPFMTIMAGE;
-	static DeclareMemType MTYPE_APPRAWIMAGE;
-	static DeclareMemType MTYPE_APPAUXRAWIMAGE;
-	
-	static DeclareMemType MTYPE_DRAWABLE;
-	
-	static DeclareMemType MTYPE_OBJECT;
-	static DeclareMemType MTYPE_OBJECT_PROCESS_UPDATE;
-	static DeclareMemType MTYPE_OBJECT_PROCESS_UPDATE_CORE;
-
-	static DeclareMemType MTYPE_DISPLAY;
-	static DeclareMemType MTYPE_DISPLAY_UPDATE;
-	static DeclareMemType MTYPE_DISPLAY_UPDATE_CAMERA;
-	static DeclareMemType MTYPE_DISPLAY_UPDATE_GEOM;
-	static DeclareMemType MTYPE_DISPLAY_SWAP;
-	static DeclareMemType MTYPE_DISPLAY_UPDATE_HUD;
-	static DeclareMemType MTYPE_DISPLAY_GEN_REFLECTION;
-	static DeclareMemType MTYPE_DISPLAY_IMAGE_UPDATE;
-	static DeclareMemType MTYPE_DISPLAY_STATE_SORT;
-	static DeclareMemType MTYPE_DISPLAY_SKY;
-	static DeclareMemType MTYPE_DISPLAY_RENDER_GEOM;
-	static DeclareMemType MTYPE_DISPLAY_RENDER_FLUSH;
-	static DeclareMemType MTYPE_DISPLAY_RENDER_UI;
-	static DeclareMemType MTYPE_DISPLAY_RENDER_ATTACHMENTS;
-
-	static DeclareMemType MTYPE_VERTEX_DATA;
-	static DeclareMemType MTYPE_VERTEX_CONSTRUCTOR;
-	static DeclareMemType MTYPE_VERTEX_DESTRUCTOR;
-	static DeclareMemType MTYPE_VERTEX_CREATE_VERTICES;
-	static DeclareMemType MTYPE_VERTEX_CREATE_INDICES;
-	static DeclareMemType MTYPE_VERTEX_DESTROY_BUFFER;	
-	static DeclareMemType MTYPE_VERTEX_DESTROY_INDICES;
-	static DeclareMemType MTYPE_VERTEX_UPDATE_VERTS;
-	static DeclareMemType MTYPE_VERTEX_UPDATE_INDICES;
-	static DeclareMemType MTYPE_VERTEX_ALLOCATE_BUFFER;
-	static DeclareMemType MTYPE_VERTEX_RESIZE_BUFFER;
-	static DeclareMemType MTYPE_VERTEX_MAP_BUFFER;
-	static DeclareMemType MTYPE_VERTEX_MAP_BUFFER_VERTICES;
-	static DeclareMemType MTYPE_VERTEX_MAP_BUFFER_INDICES;
-	static DeclareMemType MTYPE_VERTEX_UNMAP_BUFFER;
-	static DeclareMemType MTYPE_VERTEX_SET_STRIDE;
-	static DeclareMemType MTYPE_VERTEX_SET_BUFFER;
-	static DeclareMemType MTYPE_VERTEX_SETUP_VERTEX_BUFFER;
-	static DeclareMemType MTYPE_VERTEX_CLEANUP_CLASS;
-
-	static DeclareMemType MTYPE_SPACE_PARTITION;
-
-	static DeclareMemType MTYPE_PIPELINE;
-	static DeclareMemType MTYPE_PIPELINE_INIT;
-	static DeclareMemType MTYPE_PIPELINE_CREATE_BUFFERS;
-	static DeclareMemType MTYPE_PIPELINE_RESTORE_GL;
-	static DeclareMemType MTYPE_PIPELINE_UNLOAD_SHADERS;
-	static DeclareMemType MTYPE_PIPELINE_LIGHTING_DETAIL;
-	static DeclareMemType MTYPE_PIPELINE_GET_POOL_TYPE;
-	static DeclareMemType MTYPE_PIPELINE_ADD_POOL;
-	static DeclareMemType MTYPE_PIPELINE_ALLOCATE_DRAWABLE;
-	static DeclareMemType MTYPE_PIPELINE_ADD_OBJECT;
-	static DeclareMemType MTYPE_PIPELINE_CREATE_OBJECTS;
-	static DeclareMemType MTYPE_PIPELINE_UPDATE_MOVE;
-	static DeclareMemType MTYPE_PIPELINE_UPDATE_GEOM;
-	static DeclareMemType MTYPE_PIPELINE_MARK_VISIBLE;
-	static DeclareMemType MTYPE_PIPELINE_MARK_MOVED;
-	static DeclareMemType MTYPE_PIPELINE_MARK_SHIFT;
-	static DeclareMemType MTYPE_PIPELINE_SHIFT_OBJECTS;
-	static DeclareMemType MTYPE_PIPELINE_MARK_TEXTURED;
-	static DeclareMemType MTYPE_PIPELINE_MARK_REBUILD;
-	static DeclareMemType MTYPE_PIPELINE_UPDATE_CULL;
-	static DeclareMemType MTYPE_PIPELINE_STATE_SORT;
-	static DeclareMemType MTYPE_PIPELINE_POST_SORT;
-	
-	static DeclareMemType MTYPE_PIPELINE_RENDER_HUD_ELS;
-	static DeclareMemType MTYPE_PIPELINE_RENDER_HL;
-	static DeclareMemType MTYPE_PIPELINE_RENDER_GEOM;
-	static DeclareMemType MTYPE_PIPELINE_RENDER_GEOM_DEFFERRED;
-	static DeclareMemType MTYPE_PIPELINE_RENDER_GEOM_POST_DEF;
-	static DeclareMemType MTYPE_PIPELINE_RENDER_GEOM_SHADOW;
-	static DeclareMemType MTYPE_PIPELINE_RENDER_SELECT;
-	static DeclareMemType MTYPE_PIPELINE_REBUILD_POOLS;
-	static DeclareMemType MTYPE_PIPELINE_QUICK_LOOKUP;
-	static DeclareMemType MTYPE_PIPELINE_RENDER_OBJECTS;
-	static DeclareMemType MTYPE_PIPELINE_GENERATE_IMPOSTOR;
-	static DeclareMemType MTYPE_PIPELINE_RENDER_BLOOM;
-
-	static DeclareMemType MTYPE_UPKEEP_POOLS;
-
-	static DeclareMemType MTYPE_AVATAR;
-	static DeclareMemType MTYPE_AVATAR_MESH;
-	static DeclareMemType MTYPE_PARTICLES;
-	static DeclareMemType MTYPE_REGIONS;
-
-	static DeclareMemType MTYPE_INVENTORY;
-	static DeclareMemType MTYPE_INVENTORY_DRAW;
-	static DeclareMemType MTYPE_INVENTORY_BUILD_NEW_VIEWS;
-	static DeclareMemType MTYPE_INVENTORY_DO_FOLDER;
-	static DeclareMemType MTYPE_INVENTORY_POST_BUILD;
-	static DeclareMemType MTYPE_INVENTORY_FROM_XML;
-	static DeclareMemType MTYPE_INVENTORY_CREATE_NEW_ITEM;
-	static DeclareMemType MTYPE_INVENTORY_VIEW_INIT;
-	static DeclareMemType MTYPE_INVENTORY_VIEW_SHOW;
-	static DeclareMemType MTYPE_INVENTORY_VIEW_TOGGLE;
-
-	static DeclareMemType MTYPE_ANIMATION;
-	static DeclareMemType MTYPE_VOLUME;
-	static DeclareMemType MTYPE_PRIMITIVE;
-	
-	static DeclareMemType MTYPE_SCRIPT;
-	static DeclareMemType MTYPE_SCRIPT_RUN;
-	static DeclareMemType MTYPE_SCRIPT_BYTECODE;
-	
-	static DeclareMemType MTYPE_IO_PUMP;
-	static DeclareMemType MTYPE_IO_TCP;
-	static DeclareMemType MTYPE_IO_BUFFER;
-	static DeclareMemType MTYPE_IO_HTTP_SERVER;
-	static DeclareMemType MTYPE_IO_SD_SERVER;
-	static DeclareMemType MTYPE_IO_SD_CLIENT;
-	static DeclareMemType MTYPE_IO_URL_REQUEST;
-
-	static DeclareMemType MTYPE_DIRECTX_INIT;
-
-	static DeclareMemType MTYPE_TEMP1;
-	static DeclareMemType MTYPE_TEMP2;
-	static DeclareMemType MTYPE_TEMP3;
-	static DeclareMemType MTYPE_TEMP4;
-	static DeclareMemType MTYPE_TEMP5;
-	static DeclareMemType MTYPE_TEMP6;
-	static DeclareMemType MTYPE_TEMP7;
-	static DeclareMemType MTYPE_TEMP8;
-	static DeclareMemType MTYPE_TEMP9;
-
-	static DeclareMemType MTYPE_OTHER; // Special; used by display code
-
-	S32 mTypeIndex;
-};
-
-//----------------------------------------------------------------------------
-
-#endif
-
diff --git a/indra/llcommon/llmetricperformancetester.cpp b/indra/llcommon/llmetricperformancetester.cpp
index 41d3eb0bf3..aaacbfb599 100644
--- a/indra/llcommon/llmetricperformancetester.cpp
+++ b/indra/llcommon/llmetricperformancetester.cpp
@@ -29,9 +29,9 @@
 #include "indra_constants.h"
 #include "llerror.h"
 #include "llsdserialize.h"
-#include "llstat.h"
 #include "lltreeiterators.h"
 #include "llmetricperformancetester.h"
+#include "llfasttimer.h"
 
 //----------------------------------------------------------------------------------------------
 // LLMetricPerformanceTesterBasic : static methods and testers management
@@ -91,7 +91,7 @@ LLMetricPerformanceTesterBasic* LLMetricPerformanceTesterBasic::getTester(std::s
 // Return TRUE if this metric is requested or if the general default "catch all" metric is requested
 BOOL LLMetricPerformanceTesterBasic::isMetricLogRequested(std::string name)
 {
-	return (LLFastTimer::sMetricLog && ((LLFastTimer::sLogName == name) || (LLFastTimer::sLogName == DEFAULT_METRIC_NAME)));
+	return (LLTrace::TimeBlock::sMetricLog && ((LLTrace::TimeBlock::sLogName == name) || (LLTrace::TimeBlock::sLogName == DEFAULT_METRIC_NAME)));
 }
 
 /*static*/ 
@@ -194,8 +194,7 @@ void LLMetricPerformanceTesterBasic::preOutputTestResults(LLSD* sd)
 
 void LLMetricPerformanceTesterBasic::postOutputTestResults(LLSD* sd)
 {
-	LLMutexLock lock(LLFastTimer::sLogLock);
-	LLFastTimer::sLogQueue.push((*sd));
+	LLTrace::TimeBlock::pushLog(*sd);
 }
 
 void LLMetricPerformanceTesterBasic::outputTestResults() 
diff --git a/indra/llcommon/llmortician.h b/indra/llcommon/llmortician.h
index 319955ef93..9517e2db5e 100644
--- a/indra/llcommon/llmortician.h
+++ b/indra/llcommon/llmortician.h
@@ -28,6 +28,7 @@
 #define LLMORTICIAN_H
 
 #include "stdtypes.h"
+#include <list>
 
 class LL_COMMON_API LLMortician 
 {
diff --git a/indra/llcommon/llmutex.cpp b/indra/llcommon/llmutex.cpp
new file mode 100644
index 0000000000..b685bb4d60
--- /dev/null
+++ b/indra/llcommon/llmutex.cpp
@@ -0,0 +1,176 @@
+/** 
+ * @file llmutex.cpp
+ *
+ * $LicenseInfo:firstyear=2004&license=viewerlgpl$
+ * Second Life Viewer Source Code
+ * Copyright (C) 2010, Linden Research, Inc.
+ * 
+ * This library is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation;
+ * version 2.1 of the License only.
+ * 
+ * This library is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ * 
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with this library; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA
+ * 
+ * Linden Research, Inc., 945 Battery Street, San Francisco, CA  94111  USA
+ * $/LicenseInfo$
+ */
+
+#include "linden_common.h"
+#include "llapr.h"
+
+#include "apr_portable.h"
+
+#include "llmutex.h"
+#include "llthread.h"
+
+//============================================================================
+
+LLMutex::LLMutex(apr_pool_t *poolp) :
+	mAPRMutexp(NULL), mCount(0), mLockingThread(NO_THREAD)
+{
+	//if (poolp)
+	//{
+	//	mIsLocalPool = FALSE;
+	//	mAPRPoolp = poolp;
+	//}
+	//else
+	{
+		mIsLocalPool = TRUE;
+		apr_pool_create(&mAPRPoolp, NULL); // Create a subpool for this thread
+	}
+	apr_thread_mutex_create(&mAPRMutexp, APR_THREAD_MUTEX_UNNESTED, mAPRPoolp);
+}
+
+
+LLMutex::~LLMutex()
+{
+#if MUTEX_DEBUG
+	//bad assertion, the subclass LLSignal might be "locked", and that's OK
+	//llassert_always(!isLocked()); // better not be locked!
+#endif
+	apr_thread_mutex_destroy(mAPRMutexp);
+	mAPRMutexp = NULL;
+	if (mIsLocalPool)
+	{
+		apr_pool_destroy(mAPRPoolp);
+	}
+}
+
+
+void LLMutex::lock()
+{
+	if(isSelfLocked())
+	{ //redundant lock
+		mCount++;
+		return;
+	}
+	
+	apr_thread_mutex_lock(mAPRMutexp);
+	
+#if MUTEX_DEBUG
+	// Have to have the lock before we can access the debug info
+	U32 id = LLThread::currentID();
+	if (mIsLocked[id] != FALSE)
+		llerrs << "Already locked in Thread: " << id << llendl;
+	mIsLocked[id] = TRUE;
+#endif
+
+	mLockingThread = LLThread::currentID();
+}
+
+void LLMutex::unlock()
+{
+	if (mCount > 0)
+	{ //not the root unlock
+		mCount--;
+		return;
+	}
+	
+#if MUTEX_DEBUG
+	// Access the debug info while we have the lock
+	U32 id = LLThread::currentID();
+	if (mIsLocked[id] != TRUE)
+		llerrs << "Not locked in Thread: " << id << llendl;	
+	mIsLocked[id] = FALSE;
+#endif
+
+	mLockingThread = NO_THREAD;
+	apr_thread_mutex_unlock(mAPRMutexp);
+}
+
+bool LLMutex::isLocked()
+{
+	apr_status_t status = apr_thread_mutex_trylock(mAPRMutexp);
+	if (APR_STATUS_IS_EBUSY(status))
+	{
+		return true;
+	}
+	else
+	{
+		apr_thread_mutex_unlock(mAPRMutexp);
+		return false;
+	}
+}
+
+bool LLMutex::isSelfLocked()
+{
+	return mLockingThread == LLThread::currentID();
+}
+
+U32 LLMutex::lockingThread() const
+{
+	return mLockingThread;
+}
+
+//============================================================================
+
+LLCondition::LLCondition(apr_pool_t *poolp) :
+	LLMutex(poolp)
+{
+	// base class (LLMutex) has already ensured that mAPRPoolp is set up.
+
+	apr_thread_cond_create(&mAPRCondp, mAPRPoolp);
+}
+
+
+LLCondition::~LLCondition()
+{
+	apr_thread_cond_destroy(mAPRCondp);
+	mAPRCondp = NULL;
+}
+
+
+void LLCondition::wait()
+{
+	if (!isLocked())
+	{ //mAPRMutexp MUST be locked before calling apr_thread_cond_wait
+		apr_thread_mutex_lock(mAPRMutexp);
+#if MUTEX_DEBUG
+		// avoid asserts on destruction in non-release builds
+		U32 id = LLThread::currentID();
+		mIsLocked[id] = TRUE;
+#endif
+	}
+	apr_thread_cond_wait(mAPRCondp, mAPRMutexp);
+}
+
+void LLCondition::signal()
+{
+	apr_thread_cond_signal(mAPRCondp);
+}
+
+void LLCondition::broadcast()
+{
+	apr_thread_cond_broadcast(mAPRCondp);
+}
+
+
+//============================================================================
diff --git a/indra/llcommon/llmutex.h b/indra/llcommon/llmutex.h
new file mode 100644
index 0000000000..cbde4c47a9
--- /dev/null
+++ b/indra/llcommon/llmutex.h
@@ -0,0 +1,101 @@
+/** 
+ * @file llmutex.h
+ * @brief Base classes for mutex and condition handling.
+ *
+ * $LicenseInfo:firstyear=2004&license=viewerlgpl$
+ * Second Life Viewer Source Code
+ * Copyright (C) 2012, Linden Research, Inc.
+ * 
+ * This library is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation;
+ * version 2.1 of the License only.
+ * 
+ * This library is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ * 
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with this library; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA
+ * 
+ * Linden Research, Inc., 945 Battery Street, San Francisco, CA  94111  USA
+ * $/LicenseInfo$
+ */
+
+#ifndef LL_LLMUTEX_H
+#define LL_LLMUTEX_H
+
+#include "llapr.h"
+#include "apr_thread_cond.h"
+
+//============================================================================
+
+#define MUTEX_DEBUG (LL_DEBUG || LL_RELEASE_WITH_DEBUG_INFO)
+
+class LL_COMMON_API LLMutex
+{
+public:
+	typedef enum
+	{
+		NO_THREAD = 0xFFFFFFFF
+	} e_locking_thread;
+
+	LLMutex(apr_pool_t *apr_poolp = NULL); // NULL pool constructs a new pool for the mutex
+	virtual ~LLMutex();
+	
+	void lock();		// blocks
+	void unlock();
+	bool isLocked(); 	// non-blocking, but does do a lock/unlock so not free
+	bool isSelfLocked(); //return true if locked in a same thread
+	U32 lockingThread() const; //get ID of locking thread
+	
+protected:
+	apr_thread_mutex_t *mAPRMutexp;
+	mutable U32			mCount;
+	mutable U32			mLockingThread;
+	
+	apr_pool_t			*mAPRPoolp;
+	BOOL				mIsLocalPool;
+	
+#if MUTEX_DEBUG
+	std::map<U32, BOOL> mIsLocked;
+#endif
+};
+
+// Actually a condition/mutex pair (since each condition needs to be associated with a mutex).
+class LL_COMMON_API LLCondition : public LLMutex
+{
+public:
+	LLCondition(apr_pool_t *apr_poolp); // Defaults to global pool, could use the thread pool as well.
+	~LLCondition();
+	
+	void wait();		// blocks
+	void signal();
+	void broadcast();
+	
+protected:
+	apr_thread_cond_t *mAPRCondp;
+};
+
+class LLMutexLock
+{
+public:
+	LLMutexLock(LLMutex* mutex)
+	{
+		mMutex = mutex;
+		
+		if(mMutex)
+			mMutex->lock();
+	}
+	~LLMutexLock()
+	{
+		if(mMutex)
+			mMutex->unlock();
+	}
+private:
+	LLMutex* mMutex;
+};
+
+#endif // LL_LLTHREAD_H
diff --git a/indra/llcommon/llpointer.h b/indra/llcommon/llpointer.h
index 88c09c8dca..f03551045e 100644
--- a/indra/llcommon/llpointer.h
+++ b/indra/llcommon/llpointer.h
@@ -97,24 +97,13 @@ public:
 
 	LLPointer<Type>& operator =(Type* ptr)                   
 	{ 
-		if( mPointer != ptr )
-		{
-			unref(); 
-			mPointer = ptr; 
-			ref();
-		}
-
+		assign(ptr);
 		return *this; 
 	}
 
 	LLPointer<Type>& operator =(const LLPointer<Type>& ptr)  
 	{ 
-		if( mPointer != ptr.mPointer )
-		{
-			unref(); 
-			mPointer = ptr.mPointer;
-			ref();
-		}
+		assign(ptr);
 		return *this; 
 	}
 
@@ -122,12 +111,7 @@ public:
 	template<typename Subclass>
 	LLPointer<Type>& operator =(const LLPointer<Subclass>& ptr)  
 	{ 
-		if( mPointer != ptr.get() )
-		{
-			unref(); 
-			mPointer = ptr.get();
-			ref();
-		}
+		assign(ptr.get());
 		return *this; 
 	}
 	
@@ -144,6 +128,16 @@ protected:
 	void ref();                             
 	void unref();
 #else
+
+	void assign(const LLPointer<Type>& ptr)
+	{
+		if( mPointer != ptr.mPointer )
+		{
+			unref(); 
+			mPointer = ptr.mPointer;
+			ref();
+		}
+	}
 	void ref()                             
 	{ 
 		if (mPointer)
@@ -156,9 +150,9 @@ protected:
 	{
 		if (mPointer)
 		{
-			Type *tempp = mPointer;
+			Type *temp = mPointer;
 			mPointer = NULL;
-			tempp->unref();
+			temp->unref();
 			if (mPointer != NULL)
 			{
 				llwarns << "Unreference did assignment to non-NULL because of destructor" << llendl;
@@ -171,4 +165,53 @@ protected:
 	Type*	mPointer;
 };
 
+template<typename Type>
+class LLCopyOnWritePointer
+{
+public:
+	typedef LLCopyOnWritePointer<Type> self_t;
+
+	LLCopyOnWritePointer() 
+	{}
+
+	LLCopyOnWritePointer(Type* ptr) 
+	:	mPointer(ptr)
+	{}
+
+	LLCopyOnWritePointer(LLPointer<Type>& ptr)
+	:	mPointer(ptr)
+	{}
+
+	Type* write()
+	{
+		makeUnique();
+		return mPointer.get();
+	}
+
+	void makeUnique()
+	{
+		if (mPointer.notNull() && mPointer.get()->getNumRefs() > 1)
+		{
+			mPointer = new Type(*mPointer.get());
+		}
+	}
+
+	operator BOOL()  const						{ return (BOOL)mPointer; }
+	operator bool()  const						{ return (bool)mPointer; }
+	bool operator!() const						{ return !mPointer; }
+	bool isNull() const							{ return mPointer.isNull(); }
+	bool notNull() const						{ return mPointer.notNull(); }
+
+	bool operator !=(Type* ptr) const           { return (mPointer.get() != ptr); 	}
+	bool operator ==(Type* ptr) const           { return (mPointer.get() == ptr); 	}
+	bool operator ==(const LLCopyOnWritePointer<Type>& ptr) const     { return (mPointer == ptr.mPointer); 	}
+	bool operator < (const LLCopyOnWritePointer<Type>& ptr) const     { return (mPointer < ptr.mPointer); 	}
+	bool operator > (const LLCopyOnWritePointer<Type>& ptr) const     { return (mPointer > ptr.mPointer); 	}
+
+	operator const Type*()   const				{ return mPointer.get(); }
+	const Type*	operator->() const				{ return mPointer.get(); }
+protected:
+	 LLPointer<Type> mPointer;
+};
+
 #endif
diff --git a/indra/llcommon/llpredicate.cpp b/indra/llcommon/llpredicate.cpp
new file mode 100644
index 0000000000..1278948e24
--- /dev/null
+++ b/indra/llcommon/llpredicate.cpp
@@ -0,0 +1,41 @@
+/** 
+ * @file llpredicate.cpp
+ * @brief abstraction for filtering objects by predicates, with arbitrary boolean expressions
+ *
+ * $LicenseInfo:firstyear=2008&license=viewerlgpl$
+ * Second Life Viewer Source Code
+ * Copyright (C) 2010, Linden Research, Inc.
+ * 
+ * This library is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation;
+ * version 2.1 of the License only.
+ * 
+ * This library is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ * 
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with this library; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA
+ * 
+ * Linden Research, Inc., 945 Battery Street, San Francisco, CA  94111  USA
+ * $/LicenseInfo$
+ */
+#include "linden_common.h"
+
+#include "llpredicate.h"
+
+namespace LLPredicate
+{
+	const U32 cPredicateFlagsFromEnum[5] = 
+	{
+		0xAAAAaaaa, // 10101010101010101010101010101010
+		0xCCCCcccc, // 11001100110011001100110011001100
+		0xF0F0F0F0, // 11110000111100001111000011110000
+		0xFF00FF00, // 11111111000000001111111100000000
+		0xFFFF0000  // 11111111111111110000000000000000 
+	};
+}
+
diff --git a/indra/llcommon/llpredicate.h b/indra/llcommon/llpredicate.h
new file mode 100644
index 0000000000..a0e970a799
--- /dev/null
+++ b/indra/llcommon/llpredicate.h
@@ -0,0 +1,210 @@
+/** 
+ * @file llpredicate.h
+ * @brief abstraction for filtering objects by predicates, with arbitrary boolean expressions
+ *
+ * $LicenseInfo:firstyear=2008&license=viewerlgpl$
+ * Second Life Viewer Source Code
+ * Copyright (C) 2010, Linden Research, Inc.
+ * 
+ * This library is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation;
+ * version 2.1 of the License only.
+ * 
+ * This library is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ * 
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with this library; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA
+ * 
+ * Linden Research, Inc., 945 Battery Street, San Francisco, CA  94111  USA
+ * $/LicenseInfo$
+ */
+
+#ifndef LL_LLPREDICATE_H
+#define LL_LLPREDICATE_H
+
+#include "llerror.h"
+
+namespace LLPredicate
+{
+	template<typename ENUM> class Rule;
+
+	extern const U32 cPredicateFlagsFromEnum[5];
+
+	template<typename ENUM>
+	class Value
+	{
+	public:
+		typedef U32 predicate_flag_t;
+		static const S32 cMaxEnum = 5;
+
+		Value(ENUM e, bool predicate_value = true)
+		:	mPredicateFlags(predicate_value ? cPredicateFlagsFromEnum[e] : ~cPredicateFlagsFromEnum[e])
+		{
+			llassert(0 <= e && e < cMaxEnum);
+		}
+
+		Value()
+		:	mPredicateFlags(0xFFFFffff)
+		{}
+
+		Value operator!() const
+		{
+			Value new_value;
+			new_value.mPredicateFlags = ~mPredicateFlags;
+			return new_value;
+		}
+
+		Value operator &&(const Value other) const
+		{
+			Value new_value;
+			new_value.mPredicateFlags = mPredicateFlags & other.mPredicateFlags;
+			return new_value;
+		}
+
+		Value operator ||(const Value other) const
+		{
+			Value new_value;
+			new_value.mPredicateFlags = mPredicateFlags | other.mPredicateFlags;
+			return new_value;
+		}
+
+		void set(ENUM e, bool value = true)
+		{
+			llassert(0 <= e && e < cMaxEnum);
+			predicate_flag_t flags_to_modify;
+			predicate_flag_t mask = cPredicateFlagsFromEnum[e];
+			if (value)
+			{	// add predicate "e" to flags that don't contain it already
+				flags_to_modify = (mPredicateFlags & ~mask);
+				// clear flags not containing e
+				mPredicateFlags &= mask;
+				// add back flags shifted to contain e
+				mPredicateFlags |= flags_to_modify << (0x1 << e);
+			}
+			else
+			{	// remove predicate "e" from flags that contain it
+				flags_to_modify = (mPredicateFlags & mask);
+				// clear flags containing e
+				mPredicateFlags &= ~mask;
+				// add back flags shifted to not contain e
+				mPredicateFlags |= flags_to_modify >> (0x1 << e);
+			}
+		}
+
+		void forget(ENUM e)
+		{
+			set(e, true);
+			U32 flags_with_predicate = mPredicateFlags;
+			set(e, false);
+			// ambiguous value is result of adding and removing predicate at the same time!
+			mPredicateFlags |= flags_with_predicate;
+		}
+
+		bool allSet() const
+		{
+			return mPredicateFlags == ~0;
+		}
+
+		bool noneSet() const
+		{
+			return mPredicateFlags == 0;
+		}
+
+		bool someSet() const
+		{
+			return mPredicateFlags != 0;
+		}
+
+	private:
+		predicate_flag_t mPredicateFlags;
+	};
+
+	template<typename ENUM>
+	class Rule
+	{
+	public:
+		Rule(ENUM value)
+		:	mRule(value)
+		{}
+
+		Rule(const Value<ENUM> other)
+		:	mRule(other)
+		{}
+
+		Rule()
+		{}
+
+		void require(ENUM e)
+		{
+			mRule.set(e, require);
+		}
+
+		void allow(ENUM e)
+		{
+			mRule.forget(e);
+		}
+
+		bool check(const Value<ENUM> value) const
+		{
+			return (mRule && value).someSet();
+		}
+
+		bool requires(const Value<ENUM> value) const
+		{
+			return (mRule && value).someSet() && (!mRule && value).noneSet();
+		}
+
+		bool isAmbivalent(const Value<ENUM> value) const
+		{
+			return (mRule && value).someSet() && (!mRule && value).someSet();
+		}
+
+		bool acceptsAll() const
+		{
+			return mRule.allSet();
+		}
+
+		bool acceptsNone() const
+		{
+			return mRule.noneSet();
+		}
+
+		Rule operator!() const
+		{
+			Rule new_rule;
+			new_rule.mRule = !mRule;
+			return new_rule;
+		}
+
+		Rule operator &&(const Rule other) const
+		{
+			Rule new_rule;
+			new_rule.mRule = mRule && other.mRule;
+			return new_rule;
+		}
+
+		Rule operator ||(const Rule other) const
+		{
+			Rule new_rule;
+			new_rule.mRule = mRule || other.mRule;
+			return new_rule;
+		}
+
+	private:
+		Value<ENUM> mRule;
+	};
+}
+
+template<typename ENUM>
+LLPredicate::Value<ENUM> ll_make_predicate(ENUM e, bool predicate_value = true)
+{
+	 return LLPredicate::Value<ENUM>(e, predicate_value);
+}
+
+
+#endif // LL_LLPREDICATE_H
diff --git a/indra/llcommon/llprocess.h b/indra/llcommon/llprocess.h
index d711ce2f74..2fe084afcd 100644
--- a/indra/llcommon/llprocess.h
+++ b/indra/llcommon/llprocess.h
@@ -38,8 +38,7 @@
 #include <stdexcept>
 
 #if LL_WINDOWS
-#define WIN32_LEAN_AND_MEAN
-#include <windows.h>                // HANDLE (eye roll)
+#include "llwin32headerslean.h"	// for HANDLE
 #elif LL_LINUX
 #if defined(Status)
 #undef Status
diff --git a/indra/llcommon/llprocessor.cpp b/indra/llcommon/llprocessor.cpp
index fd8f603d21..5ddfa6fcef 100644
--- a/indra/llcommon/llprocessor.cpp
+++ b/indra/llcommon/llprocessor.cpp
@@ -32,9 +32,7 @@
 //#include <memory>
 
 #if LL_WINDOWS
-#	define WIN32_LEAN_AND_MEAN
-#	include <winsock2.h>
-#	include <windows.h>
+#	include "llwin32headerslean.h"
 #	define _interlockedbittestandset _renamed_interlockedbittestandset
 #	define _interlockedbittestandreset _renamed_interlockedbittestandreset
 #	include <intrin.h>
@@ -877,7 +875,7 @@ LLProcessorInfo::LLProcessorInfo() : mImpl(NULL)
 
 
 LLProcessorInfo::~LLProcessorInfo() {}
-F64 LLProcessorInfo::getCPUFrequency() const { return mImpl->getCPUFrequency(); }
+LLUnitImplicit<LLUnits::Megahertz, F64> LLProcessorInfo::getCPUFrequency() const { return mImpl->getCPUFrequency(); }
 bool LLProcessorInfo::hasSSE() const { return mImpl->hasSSE(); }
 bool LLProcessorInfo::hasSSE2() const { return mImpl->hasSSE2(); }
 bool LLProcessorInfo::hasAltivec() const { return mImpl->hasAltivec(); }
diff --git a/indra/llcommon/llprocessor.h b/indra/llcommon/llprocessor.h
index 6364d3c8bb..fbd427f484 100644
--- a/indra/llcommon/llprocessor.h
+++ b/indra/llcommon/llprocessor.h
@@ -27,6 +27,8 @@
 
 #ifndef LLPROCESSOR_H
 #define LLPROCESSOR_H
+#include "llunit.h"
+
 class LLProcessorInfoImpl;
 
 class LL_COMMON_API LLProcessorInfo
@@ -35,7 +37,7 @@ public:
 	LLProcessorInfo(); 
  	~LLProcessorInfo();
 
-	F64 getCPUFrequency() const;
+	LLUnitImplicit<LLUnits::Megahertz, F64> getCPUFrequency() const;
 	bool hasSSE() const;
 	bool hasSSE2() const;
 	bool hasAltivec() const;
diff --git a/indra/llcommon/llqueuedthread.cpp b/indra/llcommon/llqueuedthread.cpp
index abf47a0f57..956642e97a 100644
--- a/indra/llcommon/llqueuedthread.cpp
+++ b/indra/llcommon/llqueuedthread.cpp
@@ -28,6 +28,7 @@
 
 #include "llstl.h"
 #include "lltimer.h"	// ms_sleep()
+#include "lltracethreadrecorder.h"
 
 //============================================================================
 
@@ -134,8 +135,8 @@ S32 LLQueuedThread::updateQueue(F32 max_time_ms)
 		pending = getPending();
 		if(pending > 0)
 		{
-			unpause();
-		}
+		unpause();
+	}
 	}
 	else
 	{
@@ -508,6 +509,9 @@ void LLQueuedThread::run()
 		threadedUpdate();
 		
 		int res = processNextRequest();
+
+		LLTrace::get_thread_recorder()->pushToMaster();
+
 		if (res == 0)
 		{
 			mIdleThread = TRUE;
diff --git a/indra/llcommon/llsdparam.cpp b/indra/llcommon/llsdparam.cpp
index 0e29873bb0..345e30f4b4 100644
--- a/indra/llcommon/llsdparam.cpp
+++ b/indra/llcommon/llsdparam.cpp
@@ -102,13 +102,13 @@ void LLParamSDParser::readSD(const LLSD& sd, LLInitParam::BaseBlock& block, bool
 	//readSDValues(sd, block);
 }
 
-void LLParamSDParser::writeSD(LLSD& sd, const LLInitParam::BaseBlock& block)
+void LLParamSDParser::writeSDImpl(LLSD& sd, const LLInitParam::BaseBlock& block, const LLInitParam::predicate_rule_t rules, const LLInitParam::BaseBlock* diff_block)
 {
 	mNameStack.clear();
 	mWriteRootSD = &sd;
 
 	name_stack_t name_stack;
-	block.serializeBlock(*this, name_stack);
+	block.serializeBlock(*this, name_stack, rules, diff_block);
 }
 
 /*virtual*/ std::string LLParamSDParser::getCurrentElementName()
@@ -223,10 +223,14 @@ LLSD& LLParamSDParserUtilities::getSDWriteNode(LLSD& input, LLInitParam::Parser:
 	{
 		bool new_traversal = it->second;
 
-		LLSD* child_sd = it->first.empty() ? sd_to_write : &(*sd_to_write)[it->first];
-
-		if (child_sd->isArray())
+		LLSD* child_sd;
+		if (it->first.empty())
+		{
+			child_sd = sd_to_write;
+			if (child_sd->isUndefined())
 		{
+				*child_sd = LLSD::emptyArray();
+			}
 			if (new_traversal)
 			{
 				// write to new element at end
@@ -240,22 +244,7 @@ LLSD& LLParamSDParserUtilities::getSDWriteNode(LLSD& input, LLInitParam::Parser:
 		}
 		else
 		{
-			if (new_traversal 
-				&& child_sd->isDefined() 
-				&& !child_sd->isArray())
-			{
-				// copy child contents into first element of an array
-				LLSD new_array = LLSD::emptyArray();
-				new_array.append(*child_sd);
-				// assign array to slot that previously held the single value
-				*child_sd = new_array;
-				// return next element in that array
-				sd_to_write = &((*child_sd)[1]);
-			}
-			else
-			{
-				sd_to_write = child_sd;
-			}
+			sd_to_write = &(*sd_to_write)[it->first];
 		}
 		it->second = false;
 	}
@@ -283,8 +272,9 @@ void LLParamSDParserUtilities::readSDValues(read_sd_cb_t cb, const LLSD& sd, LLI
 			it != sd.endArray();
 			++it)
 		{
-			stack.back().second = true;
+			stack.push_back(make_pair(std::string(), true));
 			readSDValues(cb, *it, stack);
+			stack.pop_back();
 		}
 	}
 	else if (sd.isUndefined())
@@ -315,6 +305,12 @@ namespace LLInitParam
 	// block param interface
 	bool ParamValue<LLSD, TypeValues<LLSD>, false>::deserializeBlock(Parser& p, Parser::name_stack_range_t name_stack, bool new_name)
 	{
+		if (name_stack.first == name_stack.second
+			&& p.readValue<LLSD>(mValue))
+		{
+			return true;
+		}
+
 		LLSD& sd = LLParamSDParserUtilities::getSDWriteNode(mValue, name_stack);
 
 		LLSD::String string;
@@ -333,10 +329,14 @@ namespace LLInitParam
 		p.writeValue<LLSD::String>(sd.asString(), name_stack);
 	}
 
-	void ParamValue<LLSD, TypeValues<LLSD>, false>::serializeBlock(Parser& p, Parser::name_stack_t& name_stack, const BaseBlock* diff_block) const
+	bool ParamValue<LLSD, TypeValues<LLSD>, false>::serializeBlock(Parser& p, Parser::name_stack_t& name_stack, const predicate_rule_t predicate_rule, const BaseBlock* diff_block) const
 	{
-		// read from LLSD value and serialize out to parser (which could be LLSD, XUI, etc)
-		Parser::name_stack_t stack;
-		LLParamSDParserUtilities::readSDValues(boost::bind(&serializeElement, boost::ref(p), _1, _2), mValue, stack);
+		// attempt to write LLSD out directly
+		if (!p.writeValue<LLSD>(mValue, name_stack))
+		{
+			// otherwise read from LLSD value and serialize out to parser (which could be LLSD, XUI, etc)
+			LLParamSDParserUtilities::readSDValues(boost::bind(&serializeElement, boost::ref(p), _1, _2), mValue, name_stack);
+		}
+		return true;
 	}
 }
diff --git a/indra/llcommon/llsdparam.h b/indra/llcommon/llsdparam.h
index 6ef5debd7b..7cfc265c62 100644
--- a/indra/llcommon/llsdparam.h
+++ b/indra/llcommon/llsdparam.h
@@ -30,6 +30,7 @@
 
 #include "llinitparam.h"
 #include "boost/function.hpp"
+#include "llfasttimer.h"
 
 struct LL_COMMON_API LLParamSDParserUtilities
 {
@@ -50,11 +51,28 @@ typedef LLInitParam::Parser parser_t;
 public:
 	LLParamSDParser();
 	void readSD(const LLSD& sd, LLInitParam::BaseBlock& block, bool silent = false);
-	void writeSD(LLSD& sd, const LLInitParam::BaseBlock& block);
+	template<typename BLOCK>
+	void writeSD(LLSD& sd, 
+		const BLOCK& block, 
+		const LLInitParam::predicate_rule_t rules = LLInitParam::default_parse_rules(),
+		const LLInitParam::BaseBlock* diff_block = NULL)
+	{
+		if (!diff_block 
+			&& !rules.isAmbivalent(LLInitParam::HAS_DEFAULT_VALUE))
+		{
+			diff_block = &LLInitParam::defaultValue<BLOCK>();
+		}
+		writeSDImpl(sd, block, rules, diff_block);
+	}
 
 	/*virtual*/ std::string getCurrentElementName();
 
 private:
+	void writeSDImpl(LLSD& sd, 
+		const LLInitParam::BaseBlock& block, 
+		const LLInitParam::predicate_rule_t,
+		const LLInitParam::BaseBlock* diff_block);
+
 	void submit(LLInitParam::BaseBlock& block, const LLSD& sd, LLInitParam::Parser::name_stack_t& name_stack);
 
 	template<typename T>
diff --git a/indra/llcommon/llsingleton.h b/indra/llcommon/llsingleton.h
index 49d99f2cd0..f6b0a7194b 100644
--- a/indra/llcommon/llsingleton.h
+++ b/indra/llcommon/llsingleton.h
@@ -90,7 +90,7 @@ template <typename DERIVED_TYPE>
 class LLSingleton : private boost::noncopyable
 {
 	
-private:
+protected:
 	typedef enum e_init_state
 	{
 		UNINITIALIZED,
@@ -124,7 +124,7 @@ private:
 public:
 	virtual ~LLSingleton()
 	{
-		SingletonInstanceData& data = getData();
+		SingletonInstanceData& data = getSingletonData();
 		data.mSingletonInstance = NULL;
 		data.mInitState = DELETED;
 	}
@@ -151,29 +151,15 @@ public:
 	 */
 	static void deleteSingleton()
 	{
-		delete getData().mSingletonInstance;
-		getData().mSingletonInstance = NULL;
-		getData().mInitState = DELETED;
-	}
-
-	static SingletonInstanceData& getData()
-	{
-		// this is static to cache the lookup results
-		static void * & registry = LLSingletonRegistry::get<DERIVED_TYPE>();
-
-		// *TODO - look into making this threadsafe
-		if(NULL == registry)
-		{
-			static SingletonInstanceData data;
-			registry = &data;
-		}
-
-		return *static_cast<SingletonInstanceData *>(registry);
+		SingletonInstanceData& data = getSingletonData();
+		delete data.mSingletonInstance;
+		data.mSingletonInstance = NULL;
+		data.mInitState = DELETED;
 	}
 
 	static DERIVED_TYPE* getInstance()
 	{
-		SingletonInstanceData& data = getData();
+		SingletonInstanceData& data = getSingletonData();
 
 		if (data.mInitState == CONSTRUCTING)
 		{
@@ -197,6 +183,12 @@ public:
 		return data.mSingletonInstance;
 	}
 
+	static DERIVED_TYPE* getIfExists()
+	{
+		SingletonInstanceData& data = getSingletonData();
+		return data.mSingletonInstance;
+	}
+
 	// Reference version of getInstance()
 	// Preferred over getInstance() as it disallows checking for NULL
 	static DERIVED_TYPE& instance()
@@ -208,17 +200,31 @@ public:
 	// Use this to avoid accessing singletons before the can safely be constructed
 	static bool instanceExists()
 	{
-		return getData().mInitState == INITIALIZED;
+		return getSingletonData().mInitState == INITIALIZED;
 	}
 	
 	// Has this singleton already been deleted?
 	// Use this to avoid accessing singletons from a static object's destructor
 	static bool destroyed()
 	{
-		return getData().mInitState == DELETED;
+		return getSingletonData().mInitState == DELETED;
 	}
 
 private:
+	static SingletonInstanceData& getSingletonData()
+	{
+		// this is static to cache the lookup results
+		static void * & registry = LLSingletonRegistry::get<DERIVED_TYPE>();
+
+		// *TODO - look into making this threadsafe
+		if(NULL == registry)
+		{
+			static SingletonInstanceData data;
+			registry = &data;
+		}
+
+		return *static_cast<SingletonInstanceData *>(registry);
+	}
 	virtual void initSingleton() {}
 };
 
diff --git a/indra/llcommon/llskipmap.h b/indra/llcommon/llskipmap.h
index 49ff2928d1..ed53973baa 100644
--- a/indra/llcommon/llskipmap.h
+++ b/indra/llcommon/llskipmap.h
@@ -210,8 +210,7 @@ inline LLSkipMap<INDEX_TYPE, DATA_TYPE, BINARY_DEPTH>::LLSkipMap()
 	:	mInsertFirst(NULL),
 		mEquals(defaultEquals)
 {
-	// Skipmaps must have binary depth of at least 2
-	cassert(BINARY_DEPTH >= 2);
+	llstatic_assert(BINARY_DEPTH >= 2, "Skipmaps must have binary depth of at least 2");
 
 	S32 i;
 	for (i = 0; i < BINARY_DEPTH; i++)
@@ -229,8 +228,7 @@ inline LLSkipMap<INDEX_TYPE, DATA_TYPE, BINARY_DEPTH>::LLSkipMap(BOOL	(*insert_f
 	:	mInsertFirst(insert_first),
 		mEquals(equals)
 {
-	// Skipmaps must have binary depth of at least 2
-	cassert(BINARY_DEPTH >= 2);
+	llstatic_assert(BINARY_DEPTH >= 2, "Skipmaps must have binary depth of at least 2");
 
 	mLevel = 1;
 	S32 i;
diff --git a/indra/llcommon/llstacktrace.cpp b/indra/llcommon/llstacktrace.cpp
index ccd7ef91c2..e0e9056380 100644
--- a/indra/llcommon/llstacktrace.cpp
+++ b/indra/llcommon/llstacktrace.cpp
@@ -32,7 +32,7 @@
 #include <iostream>
 #include <sstream>
 
-#include "windows.h"
+#include "llwin32headerslean.h"
 #include "Dbghelp.h"
 
 typedef USHORT NTAPI RtlCaptureStackBackTrace_Function(
diff --git a/indra/llcommon/llstat.cpp b/indra/llcommon/llstat.cpp
deleted file mode 100644
index b82d52797e..0000000000
--- a/indra/llcommon/llstat.cpp
+++ /dev/null
@@ -1,1311 +0,0 @@
-/** 
- * @file llstat.cpp
- *
- * $LicenseInfo:firstyear=2001&license=viewerlgpl$
- * Second Life Viewer Source Code
- * Copyright (C) 2010, Linden Research, Inc.
- * 
- * This library is free software; you can redistribute it and/or
- * modify it under the terms of the GNU Lesser General Public
- * License as published by the Free Software Foundation;
- * version 2.1 of the License only.
- * 
- * This library is distributed in the hope that it will be useful,
- * but WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
- * Lesser General Public License for more details.
- * 
- * You should have received a copy of the GNU Lesser General Public
- * License along with this library; if not, write to the Free Software
- * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA
- * 
- * Linden Research, Inc., 945 Battery Street, San Francisco, CA  94111  USA
- * $/LicenseInfo$
- */
-
-#include "linden_common.h"
-
-#include "llstat.h"
-#include "lllivefile.h"
-#include "llerrorcontrol.h"
-#include "llframetimer.h"
-#include "timing.h"
-#include "llsd.h"
-#include "llsdserialize.h"
-#include "llstl.h"
-#include "u64.h"
-
-
-// statics
-S32	            LLPerfBlock::sStatsFlags = LLPerfBlock::LLSTATS_NO_OPTIONAL_STATS;       // Control what is being recorded
-LLPerfBlock::stat_map_t    LLPerfBlock::sStatMap;    // Map full path string to LLStatTime objects, tracks all active objects
-std::string        LLPerfBlock::sCurrentStatPath = "";    // Something like "/total_time/physics/physics step"
-
-//------------------------------------------------------------------------
-// Live config file to trigger stats logging
-static const char    STATS_CONFIG_FILE_NAME[]            = "/dev/shm/simperf/simperf_proc_config.llsd";
-static const F32    STATS_CONFIG_REFRESH_RATE            = 5.0;        // seconds
-
-class LLStatsConfigFile : public LLLiveFile
-{
-public:
-    LLStatsConfigFile()
-        : LLLiveFile(filename(), STATS_CONFIG_REFRESH_RATE),
-        mChanged(false), mStatsp(NULL) { }
-
-    static std::string filename();
-    
-protected:
-    /* virtual */ bool loadFile();
-
-public:
-    void init(LLPerfStats* statsp);
-    static LLStatsConfigFile& instance();
-        // return the singleton stats config file
-
-    bool mChanged;
-
-protected:
-    LLPerfStats*    mStatsp;
-};
-
-std::string LLStatsConfigFile::filename()
-{
-    return STATS_CONFIG_FILE_NAME;
-}
-
-void LLStatsConfigFile::init(LLPerfStats* statsp)
-{
-    mStatsp = statsp;
-}
-
-LLStatsConfigFile& LLStatsConfigFile::instance()
-{
-    static LLStatsConfigFile the_file;
-    return the_file;
-}
-
-
-/* virtual */
-// Load and parse the stats configuration file
-bool LLStatsConfigFile::loadFile()
-{
-    if (!mStatsp)
-    {
-        llwarns << "Tries to load performance configure file without initializing LPerfStats" << llendl;
-        return false;
-    }
-    mChanged = true;
-    
-    LLSD stats_config;
-    {
-        llifstream file(filename().c_str());
-        if (file.is_open())
-        {
-            LLSDSerialize::fromXML(stats_config, file);
-            if (stats_config.isUndefined())
-            {
-                llinfos << "Performance statistics configuration file ill-formed, not recording statistics" << llendl;
-                mStatsp->setReportPerformanceDuration( 0.f );
-                return false;
-            }
-        }
-        else 
-        {    // File went away, turn off stats if it was on
-            if ( mStatsp->frameStatsIsRunning() )
-            {
-                llinfos << "Performance statistics configuration file deleted, not recording statistics" << llendl;
-                mStatsp->setReportPerformanceDuration( 0.f );
-            }
-            return true;
-        }
-    }
-
-    F32 duration = 0.f;
-    F32 interval = 0.f;
-	S32 flags = LLPerfBlock::LLSTATS_BASIC_STATS;
-
-    const char * w = "duration";
-    if (stats_config.has(w))
-    {
-        duration = (F32)stats_config[w].asReal();
-    } 
-    w = "interval";
-    if (stats_config.has(w))
-    {
-        interval = (F32)stats_config[w].asReal();
-    } 
-    w = "flags";
-    if (stats_config.has(w))
-    {
-		flags = (S32)stats_config[w].asInteger();
-		if (flags == LLPerfBlock::LLSTATS_NO_OPTIONAL_STATS &&
-			duration > 0)
-		{   // No flags passed in, but have a duration, so reset to basic stats
-			flags = LLPerfBlock::LLSTATS_BASIC_STATS;
-		}
-    } 
-
-    mStatsp->setReportPerformanceDuration( duration, flags );
-    mStatsp->setReportPerformanceInterval( interval );
-
-    if ( duration > 0 )
-    {
-        if ( interval == 0.f )
-        {
-            llinfos << "Recording performance stats every frame for " << duration << " sec" << llendl;
-        }
-        else
-        {
-            llinfos << "Recording performance stats every " << interval << " seconds for " << duration << " seconds" << llendl;
-        }
-    }
-    else
-    {
-        llinfos << "Performance stats recording turned off" << llendl;
-    }
-	return true;
-}
-
-
-//------------------------------------------------------------------------
-
-LLPerfStats::LLPerfStats(const std::string& process_name, S32 process_pid) : 
-    mFrameStatsFileFailure(FALSE),
-    mSkipFirstFrameStats(FALSE),
-    mProcessName(process_name),
-    mProcessPID(process_pid),
-    mReportPerformanceStatInterval(1.f),
-    mReportPerformanceStatEnd(0.0) 
-{ }
-
-LLPerfStats::~LLPerfStats()
-{
-    LLPerfBlock::clearDynamicStats();
-    mFrameStatsFile.close();
-}
-
-void LLPerfStats::init()
-{
-    // Initialize the stats config file instance.
-    (void) LLStatsConfigFile::instance().init(this);
-    (void) LLStatsConfigFile::instance().checkAndReload();
-}
-
-// Open file for statistics
-void    LLPerfStats::openPerfStatsFile()
-{
-    if ( !mFrameStatsFile
-        && !mFrameStatsFileFailure )
-    {
-        std::string stats_file = llformat("/dev/shm/simperf/%s_proc.%d.llsd", mProcessName.c_str(), mProcessPID);
-        mFrameStatsFile.close();
-        mFrameStatsFile.clear();
-        mFrameStatsFile.open(stats_file, llofstream::out);
-        if ( mFrameStatsFile.fail() )
-        {
-            llinfos << "Error opening statistics log file " << stats_file << llendl;
-            mFrameStatsFileFailure = TRUE;
-        }
-        else
-        {
-            LLSD process_info = LLSD::emptyMap();
-            process_info["name"] = mProcessName;
-            process_info["pid"] = (LLSD::Integer) mProcessPID;
-            process_info["stat_rate"] = (LLSD::Integer) mReportPerformanceStatInterval;
-            // Add process-specific info.
-            addProcessHeaderInfo(process_info);
-
-            mFrameStatsFile << LLSDNotationStreamer(process_info) << std::endl; 
-        }
-    }
-}
-
-// Dump out performance metrics over some time interval
-void LLPerfStats::dumpIntervalPerformanceStats()
-{
-    // Ensure output file is OK
-    openPerfStatsFile();
-
-    if ( mFrameStatsFile )
-    {
-        LLSD stats = LLSD::emptyMap();
-
-        LLStatAccum::TimeScale scale;
-        if ( getReportPerformanceInterval() == 0.f )
-        {
-            scale = LLStatAccum::SCALE_PER_FRAME;
-        }
-        else if ( getReportPerformanceInterval() < 0.5f )
-        {
-            scale = LLStatAccum::SCALE_100MS;
-        }
-        else
-        {
-            scale = LLStatAccum::SCALE_SECOND;
-        }
-
-        // Write LLSD into log
-        stats["utc_time"] = (LLSD::String) LLError::utcTime();
-        stats["timestamp"] = U64_to_str((totalTime() / 1000) + (gUTCOffset * 1000));    // milliseconds since epoch
-        stats["frame_number"] = (LLSD::Integer) LLFrameTimer::getFrameCount();
-
-        // Add process-specific frame info.
-        addProcessFrameInfo(stats, scale);
-        LLPerfBlock::addStatsToLLSDandReset( stats, scale );
-
-        mFrameStatsFile << LLSDNotationStreamer(stats) << std::endl; 
-    }
-}
-
-// Set length of performance stat recording.  
-// If turning stats on, caller must provide flags
-void    LLPerfStats::setReportPerformanceDuration( F32 seconds, S32 flags /* = LLSTATS_NO_OPTIONAL_STATS */ )
-{ 
-	if ( seconds <= 0.f )
-	{
-		mReportPerformanceStatEnd = 0.0;
-		LLPerfBlock::setStatsFlags(LLPerfBlock::LLSTATS_NO_OPTIONAL_STATS);		// Make sure all recording is off
-		mFrameStatsFile.close();
-		LLPerfBlock::clearDynamicStats();
-	}
-	else
-	{
-		mReportPerformanceStatEnd = LLFrameTimer::getElapsedSeconds() + ((F64) seconds);
-		// Clear failure flag to try and create the log file once
-		mFrameStatsFileFailure = FALSE;
-		mSkipFirstFrameStats = TRUE;		// Skip the first report (at the end of this frame)
-		LLPerfBlock::setStatsFlags(flags);
-	}
-}
-
-void LLPerfStats::updatePerFrameStats()
-{
-    (void) LLStatsConfigFile::instance().checkAndReload();
-	static LLFrameTimer performance_stats_timer;
-	if ( frameStatsIsRunning() )
-	{
-		if ( mReportPerformanceStatInterval == 0 )
-		{	// Record info every frame
-			if ( mSkipFirstFrameStats )
-			{	// Skip the first time - was started this frame
-				mSkipFirstFrameStats = FALSE;
-			}
-			else
-			{
-				dumpIntervalPerformanceStats();
-			}
-		}
-		else
-		{
-			performance_stats_timer.setTimerExpirySec( getReportPerformanceInterval() );
-			if (performance_stats_timer.checkExpirationAndReset( mReportPerformanceStatInterval ))
-			{
-				dumpIntervalPerformanceStats();
-			}
-		}
-		
-		if ( LLFrameTimer::getElapsedSeconds() > mReportPerformanceStatEnd )
-		{	// Reached end of time, clear it to stop reporting
-			setReportPerformanceDuration(0.f);			// Don't set mReportPerformanceStatEnd directly	
-            llinfos << "Recording performance stats completed" << llendl;
-		}
-	}
-}
-
-
-//------------------------------------------------------------------------
-
-U64 LLStatAccum::sScaleTimes[NUM_SCALES] =
-{
-	USEC_PER_SEC / 10,				// 100 millisec
-	USEC_PER_SEC * 1,				// seconds
-	USEC_PER_SEC * 60,				// minutes
-#if ENABLE_LONG_TIME_STATS
-	// enable these when more time scales are desired
-	USEC_PER_SEC * 60*60,			// hours
-	USEC_PER_SEC * 24*60*60,		// days
-	USEC_PER_SEC * 7*24*60*60,		// weeks
-#endif
-};
-
-
-
-LLStatAccum::LLStatAccum(bool useFrameTimer)
-	: mUseFrameTimer(useFrameTimer),
-	  mRunning(FALSE),
-	  mLastTime(0),
-	  mLastSampleValue(0.0),
-	  mLastSampleValid(FALSE)
-{
-}
-
-LLStatAccum::~LLStatAccum()
-{
-}
-
-
-
-void LLStatAccum::reset(U64 when)
-{
-	mRunning = TRUE;
-	mLastTime = when;
-
-	for (int i = 0; i < NUM_SCALES; ++i)
-	{
-		mBuckets[i].accum = 0.0;
-		mBuckets[i].endTime = when + sScaleTimes[i];
-		mBuckets[i].lastValid = false;
-	}
-}
-
-void LLStatAccum::sum(F64 value)
-{
-	sum(value, getCurrentUsecs());
-}
-
-void LLStatAccum::sum(F64 value, U64 when)
-{
-	if (!mRunning)
-	{
-		reset(when);
-		return;
-	}
-	if (when < mLastTime)
-	{
-		// This happens a LOT on some dual core systems.
-		lldebugs << "LLStatAccum::sum clock has gone backwards from "
-			<< mLastTime << " to " << when << ", resetting" << llendl;
-
-		reset(when);
-		return;
-	}
-
-	// how long is this value for
-	U64 timeSpan = when - mLastTime;
-
-	for (int i = 0; i < NUM_SCALES; ++i)
-	{
-		Bucket& bucket = mBuckets[i];
-
-		if (when < bucket.endTime)
-		{
-			bucket.accum += value;
-		}
-		else
-		{
-			U64 timeScale = sScaleTimes[i];
-
-			U64 timeLeft = when - bucket.endTime;
-				// how much time is left after filling this bucket
-			
-			if (timeLeft < timeScale)
-			{
-				F64 valueLeft = value * timeLeft / timeSpan;
-
-				bucket.lastValid = true;
-				bucket.lastAccum = bucket.accum + (value - valueLeft);
-				bucket.accum = valueLeft;
-				bucket.endTime += timeScale;
-			}
-			else
-			{
-				U64 timeTail = timeLeft % timeScale;
-
-				bucket.lastValid = true;
-				bucket.lastAccum = value * timeScale / timeSpan;
-				bucket.accum = value * timeTail / timeSpan;
-				bucket.endTime += (timeLeft - timeTail) + timeScale;
-			}
-		}
-	}
-
-	mLastTime = when;
-}
-
-
-F32 LLStatAccum::meanValue(TimeScale scale) const
-{
-	if (!mRunning)
-	{
-		return 0.0;
-	}
-	if ( scale == SCALE_PER_FRAME )
-	{	// Per-frame not supported here
-		scale = SCALE_100MS;
-	}
-
-	if (scale < 0 || scale >= NUM_SCALES)
-	{
-		llwarns << "llStatAccum::meanValue called for unsupported scale: "
-			<< scale << llendl;
-		return 0.0;
-	}
-
-	const Bucket& bucket = mBuckets[scale];
-
-	F64 value = bucket.accum;
-	U64 timeLeft = bucket.endTime - mLastTime;
-	U64 scaleTime = sScaleTimes[scale];
-
-	if (bucket.lastValid)
-	{
-		value += bucket.lastAccum * timeLeft / scaleTime;
-	}
-	else if (timeLeft < scaleTime)
-	{
-		value *= scaleTime / (scaleTime - timeLeft);
-	}
-	else
-	{
-		value = 0.0;
-	}
-
-	return (F32)(value / scaleTime);
-}
-
-
-U64 LLStatAccum::getCurrentUsecs() const
-{
-	if (mUseFrameTimer)
-	{
-		return LLFrameTimer::getTotalTime();
-	}
-	else
-	{
-		return totalTime();
-	}
-}
-
-
-// ------------------------------------------------------------------------
-
-LLStatRate::LLStatRate(bool use_frame_timer)
-	: LLStatAccum(use_frame_timer)
-{
-}
-
-void LLStatRate::count(U32 value)
-{
-	sum((F64)value * sScaleTimes[SCALE_SECOND]);
-}
-
-
-void LLStatRate::mark()
- { 
-	// Effectively the same as count(1), but sets mLastSampleValue
-	U64 when = getCurrentUsecs();
-
-	if ( mRunning 
-		 && (when > mLastTime) )
-	{	// Set mLastSampleValue to the time from the last mark()
-		F64 duration = ((F64)(when - mLastTime)) / sScaleTimes[SCALE_SECOND];
-		if ( duration > 0.0 )
-		{
-			mLastSampleValue = 1.0 / duration;
-		}
-		else
-		{
-			mLastSampleValue = 0.0;
-		}
-	}
-
-	sum( (F64) sScaleTimes[SCALE_SECOND], when);
- }
-
-
-// ------------------------------------------------------------------------
-
-
-LLStatMeasure::LLStatMeasure(bool use_frame_timer)
-	: LLStatAccum(use_frame_timer)
-{
-}
-
-void LLStatMeasure::sample(F64 value)
-{
-	U64 when = getCurrentUsecs();
-
-	if (mLastSampleValid)
-	{
-		F64 avgValue = (value + mLastSampleValue) / 2.0;
-		F64 interval = (F64)(when - mLastTime);
-
-		sum(avgValue * interval, when);
-	}
-	else
-	{
-		reset(when);
-	}
-
-	mLastSampleValid = TRUE;
-	mLastSampleValue = value;
-}
-
-
-// ------------------------------------------------------------------------
-
-LLStatTime::LLStatTime(const std::string & key)
-	: LLStatAccum(false),
-	  mFrameNumber(LLFrameTimer::getFrameCount()),
-	  mTotalTimeInFrame(0),
-	  mKey(key)
-#if LL_DEBUG
-	  , mRunning(FALSE)
-#endif
-{
-}
-
-void LLStatTime::start()
-{
-	// Reset frame accumluation if the frame number has changed
-	U32 frame_number = LLFrameTimer::getFrameCount();
-	if ( frame_number != mFrameNumber )
-	{
-		mFrameNumber = frame_number;
-		mTotalTimeInFrame = 0;
-	}
-
-	sum(0.0);
-
-#if LL_DEBUG
-	// Shouldn't be running already
-	llassert( !mRunning );
-	mRunning = TRUE;
-#endif
-}
-
-void LLStatTime::stop()
-{
-	U64 end_time = getCurrentUsecs();
-	U64 duration = end_time - mLastTime;
-	sum(F64(duration), end_time);
-	//llinfos << "mTotalTimeInFrame incremented from  " << mTotalTimeInFrame << " to " << (mTotalTimeInFrame + duration) << llendl; 
-	mTotalTimeInFrame += duration;
-
-#if LL_DEBUG
-	mRunning = FALSE;
-#endif
-}
-
-/* virtual */ F32 LLStatTime::meanValue(TimeScale scale) const
-{
-    if ( LLStatAccum::SCALE_PER_FRAME == scale )
-    {
-        return (F32)mTotalTimeInFrame;
-    }
-    else
-    {
-        return LLStatAccum::meanValue(scale);
-    }
-}
-
-
-// ------------------------------------------------------------------------
-
-
-// Use this constructor for pre-defined LLStatTime objects
-LLPerfBlock::LLPerfBlock(LLStatTime* stat ) : mPredefinedStat(stat), mDynamicStat(NULL)
-{
-    if (mPredefinedStat)
-    {
-        // If dynamic stats are turned on, this will create a separate entry in the stat map.
-        initDynamicStat(mPredefinedStat->mKey);
-
-        // Start predefined stats.  These stats are not part of the stat map.
-        mPredefinedStat->start();
-    }
-}
-
-// Use this constructor for normal, optional LLPerfBlock time slices
-LLPerfBlock::LLPerfBlock( const char* key ) : mPredefinedStat(NULL), mDynamicStat(NULL)
-{
-    if ((sStatsFlags & LLSTATS_BASIC_STATS) == 0)
-	{	// These are off unless the base set is enabled
-		return;
-	}
-
-	initDynamicStat(key);
-}
-
-	
-// Use this constructor for dynamically created LLPerfBlock time slices
-// that are only enabled by specific control flags
-LLPerfBlock::LLPerfBlock( const char* key1, const char* key2, S32 flags ) : mPredefinedStat(NULL), mDynamicStat(NULL)
-{
-    if ((sStatsFlags & flags) == 0)
-	{
-		return;
-	}
-
-    if (NULL == key2 || strlen(key2) == 0)
-    {
-        initDynamicStat(key1);
-    }
-    else
-    {
-        std::ostringstream key;
-        key << key1 << "_" << key2;
-        initDynamicStat(key.str());
-    }
-}
-
-// Set up the result data map if dynamic stats are enabled
-void LLPerfBlock::initDynamicStat(const std::string& key)
-{
-    // Early exit if dynamic stats aren't enabled.
-    if (sStatsFlags == LLSTATS_NO_OPTIONAL_STATS) 
-		return;
-
-    mLastPath = sCurrentStatPath;		// Save and restore current path
-    sCurrentStatPath += "/" + key;		// Add key to current path
-
-    // See if the LLStatTime object already exists
-    stat_map_t::iterator iter = sStatMap.find(sCurrentStatPath);
-    if ( iter == sStatMap.end() )
-    {
-        // StatEntry object doesn't exist, so create it
-        mDynamicStat = new StatEntry( key );
-        sStatMap[ sCurrentStatPath ] = mDynamicStat;	// Set the entry for this path
-    }
-    else
-    {
-        // Found this path in the map, use the object there
-        mDynamicStat = (*iter).second;		// Get StatEntry for the current path
-    }
-
-    if (mDynamicStat)
-    {
-        mDynamicStat->mStat.start();
-        mDynamicStat->mCount++;
-    }
-    else
-    {
-        llwarns << "Initialized NULL dynamic stat at '" << sCurrentStatPath << "'" << llendl;
-       sCurrentStatPath = mLastPath;
-    }
-}
-
-
-// Destructor does the time accounting
-LLPerfBlock::~LLPerfBlock()
-{
-    if (mPredefinedStat) mPredefinedStat->stop();
-    if (mDynamicStat)
-    {
-        mDynamicStat->mStat.stop();
-        sCurrentStatPath = mLastPath;	// Restore the path in case sStatsEnabled changed during this block
-    }
-}
-
-
-// Clear the map of any dynamic stats.  Static routine
-void LLPerfBlock::clearDynamicStats()
-{
-    std::for_each(sStatMap.begin(), sStatMap.end(), DeletePairedPointer());
-    sStatMap.clear();
-}
-
-// static - Extract the stat info into LLSD
-void LLPerfBlock::addStatsToLLSDandReset( LLSD & stats,
-										  LLStatAccum::TimeScale scale )
-{
-    // If we aren't in per-frame scale, we need to go from second to microsecond.
-    U32 scale_adjustment = 1;
-    if (LLStatAccum::SCALE_PER_FRAME != scale)
-    {
-        scale_adjustment = USEC_PER_SEC;
-    }
-	stat_map_t::iterator iter = sStatMap.begin();
-	for ( ; iter != sStatMap.end(); ++iter )
-	{	// Put the entry into LLSD "/full/path/to/stat/" = microsecond total time
-		const std::string & stats_full_path = (*iter).first;
-
-		StatEntry * stat = (*iter).second;
-		if (stat)
-		{
-            if (stat->mCount > 0)
-            {
-                stats[stats_full_path] = LLSD::emptyMap();
-                stats[stats_full_path]["us"] = (LLSD::Integer) (scale_adjustment * stat->mStat.meanValue(scale));
-                if (stat->mCount > 1)
-                {
-                    stats[stats_full_path]["count"] = (LLSD::Integer) stat->mCount;
-                }
-                stat->mCount = 0;
-            }
-		}
-		else
-		{	// Shouldn't have a NULL pointer in the map.
-            llwarns << "Unexpected NULL dynamic stat at '" << stats_full_path << "'" << llendl;
-		}
-	}	
-}
-
-
-// ------------------------------------------------------------------------
-
-LLTimer LLStat::sTimer;
-LLFrameTimer LLStat::sFrameTimer;
-
-void LLStat::init()
-{
-	llassert(mNumBins > 0);
-	mNumValues = 0;
-	mLastValue = 0.f;
-	mLastTime = 0.f;
-	mCurBin = (mNumBins-1);
-	mNextBin = 0;
-	mBins      = new F32[mNumBins];
-	mBeginTime = new F64[mNumBins];
-	mTime      = new F64[mNumBins];
-	mDT        = new F32[mNumBins];
-	for (U32 i = 0; i < mNumBins; i++)
-	{
-		mBins[i]      = 0.f;
-		mBeginTime[i] = 0.0;
-		mTime[i]      = 0.0;
-		mDT[i]        = 0.f;
-	}
-
-	if (!mName.empty())
-	{
-		stat_map_t::iterator iter = getStatList().find(mName);
-		if (iter != getStatList().end())
-			llwarns << "LLStat with duplicate name: " << mName << llendl;
-		getStatList().insert(std::make_pair(mName, this));
-	}
-}
-
-LLStat::stat_map_t& LLStat::getStatList()
-{
-	static LLStat::stat_map_t stat_list;
-	return stat_list;
-}
-
-LLStat::LLStat(const U32 num_bins, const BOOL use_frame_timer)
-	: mUseFrameTimer(use_frame_timer),
-	  mNumBins(num_bins)
-{
-	init();
-}
-
-LLStat::LLStat(std::string name, U32 num_bins, BOOL use_frame_timer)
-	: mUseFrameTimer(use_frame_timer),
-	  mNumBins(num_bins),
-	  mName(name)
-{
-	init();
-}
-
-LLStat::~LLStat()
-{
-	delete[] mBins;
-	delete[] mBeginTime;
-	delete[] mTime;
-	delete[] mDT;
-
-	if (!mName.empty())
-	{
-		// handle multiple entries with the same name
-		stat_map_t::iterator iter = getStatList().find(mName);
-		while (iter != getStatList().end() && iter->second != this)
-			++iter;
-		getStatList().erase(iter);
-	}
-}
-
-void LLStat::reset()
-{
-	U32 i;
-
-	mNumValues = 0;
-	mLastValue = 0.f;
-	mCurBin = (mNumBins-1);
-	delete[] mBins;
-	delete[] mBeginTime;
-	delete[] mTime;
-	delete[] mDT;
-	mBins      = new F32[mNumBins];
-	mBeginTime = new F64[mNumBins];
-	mTime      = new F64[mNumBins];
-	mDT        = new F32[mNumBins];
-	for (i = 0; i < mNumBins; i++)
-	{
-		mBins[i]      = 0.f;
-		mBeginTime[i] = 0.0;
-		mTime[i]      = 0.0;
-		mDT[i]        = 0.f;
-	}
-}
-
-void LLStat::setBeginTime(const F64 time)
-{
-	mBeginTime[mNextBin] = time;
-}
-
-void LLStat::addValueTime(const F64 time, const F32 value)
-{
-	if (mNumValues < mNumBins)
-	{
-		mNumValues++;
-	}
-
-	// Increment the bin counters.
-	mCurBin++;
-	if ((U32)mCurBin == mNumBins)
-	{
-		mCurBin = 0;
-	}
-	mNextBin++;
-	if ((U32)mNextBin == mNumBins)
-	{
-		mNextBin = 0;
-	}
-
-	mBins[mCurBin] = value;
-	mTime[mCurBin] = time;
-	mDT[mCurBin] = (F32)(mTime[mCurBin] - mBeginTime[mCurBin]);
-	//this value is used to prime the min/max calls
-	mLastTime = mTime[mCurBin];
-	mLastValue = value;
-
-	// Set the begin time for the next stat segment.
-	mBeginTime[mNextBin] = mTime[mCurBin];
-	mTime[mNextBin] = mTime[mCurBin];
-	mDT[mNextBin] = 0.f;
-}
-
-void LLStat::start()
-{
-	if (mUseFrameTimer)
-	{
-		mBeginTime[mNextBin] = sFrameTimer.getElapsedSeconds();
-	}
-	else
-	{
-		mBeginTime[mNextBin] = sTimer.getElapsedTimeF64();
-	}
-}
-
-void LLStat::addValue(const F32 value)
-{
-	if (mNumValues < mNumBins)
-	{
-		mNumValues++;
-	}
-
-	// Increment the bin counters.
-	mCurBin++;
-	if ((U32)mCurBin == mNumBins)
-	{
-		mCurBin = 0;
-	}
-	mNextBin++;
-	if ((U32)mNextBin == mNumBins)
-	{
-		mNextBin = 0;
-	}
-
-	mBins[mCurBin] = value;
-	if (mUseFrameTimer)
-	{
-		mTime[mCurBin] = sFrameTimer.getElapsedSeconds();
-	}
-	else
-	{
-		mTime[mCurBin] = sTimer.getElapsedTimeF64();
-	}
-	mDT[mCurBin] = (F32)(mTime[mCurBin] - mBeginTime[mCurBin]);
-
-	//this value is used to prime the min/max calls
-	mLastTime = mTime[mCurBin];
-	mLastValue = value;
-
-	// Set the begin time for the next stat segment.
-	mBeginTime[mNextBin] = mTime[mCurBin];
-	mTime[mNextBin] = mTime[mCurBin];
-	mDT[mNextBin] = 0.f;
-}
-
-
-F32 LLStat::getMax() const
-{
-	U32 i;
-	F32 current_max = mLastValue;
-	if (mNumBins == 0)
-	{
-		current_max = 0.f;
-	}
-	else
-	{
-		for (i = 0; (i < mNumBins) && (i < mNumValues); i++)
-		{
-			// Skip the bin we're currently filling.
-			if (i == (U32)mNextBin)
-			{
-				continue;
-			}
-			if (mBins[i] > current_max)
-			{
-				current_max = mBins[i];
-			}
-		}
-	}
-	return current_max;
-}
-
-F32 LLStat::getMean() const
-{
-	U32 i;
-	F32 current_mean = 0.f;
-	U32 samples = 0;
-	for (i = 0; (i < mNumBins) && (i < mNumValues); i++)
-	{
-		// Skip the bin we're currently filling.
-		if (i == (U32)mNextBin)
-		{
-			continue;
-		}
-		current_mean += mBins[i];
-		samples++;
-	}
-
-	// There will be a wrap error at 2^32. :)
-	if (samples != 0)
-	{
-		current_mean /= samples;
-	}
-	else
-	{
-		current_mean = 0.f;
-	}
-	return current_mean;
-}
-
-F32 LLStat::getMin() const
-{
-	U32 i;
-	F32 current_min = mLastValue;
-
-	if (mNumBins == 0)
-	{
-		current_min = 0.f;
-	}
-	else
-	{
-		for (i = 0; (i < mNumBins) && (i < mNumValues); i++)
-		{
-			// Skip the bin we're currently filling.
-			if (i == (U32)mNextBin)
-			{
-				continue;
-			}
-			if (mBins[i] < current_min)
-			{
-				current_min = mBins[i];
-			}
-		}
-	}
-	return current_min;
-}
-
-F32 LLStat::getSum() const
-{
-	U32 i;
-	F32 sum = 0.f;
-	for (i = 0; (i < mNumBins) && (i < mNumValues); i++)
-	{
-		// Skip the bin we're currently filling.
-		if (i == (U32)mNextBin)
-		{
-			continue;
-		}
-		sum += mBins[i];
-	}
-
-	return sum;
-}
-
-F32 LLStat::getSumDuration() const
-{
-	U32 i;
-	F32 sum = 0.f;
-	for (i = 0; (i < mNumBins) && (i < mNumValues); i++)
-	{
-		// Skip the bin we're currently filling.
-		if (i == (U32)mNextBin)
-		{
-			continue;
-		}
-		sum += mDT[i];
-	}
-
-	return sum;
-}
-
-F32 LLStat::getPrev(S32 age) const
-{
-	S32 bin;
-	bin = mCurBin - age;
-
-	while (bin < 0)
-	{
-		bin += mNumBins;
-	}
-
-	if (bin == mNextBin)
-	{
-		// Bogus for bin we're currently working on.
-		return 0.f;
-	}
-	return mBins[bin];
-}
-
-F32 LLStat::getPrevPerSec(S32 age) const
-{
-	S32 bin;
-	bin = mCurBin - age;
-
-	while (bin < 0)
-	{
-		bin += mNumBins;
-	}
-
-	if (bin == mNextBin)
-	{
-		// Bogus for bin we're currently working on.
-		return 0.f;
-	}
-	return mBins[bin] / mDT[bin];
-}
-
-F64 LLStat::getPrevBeginTime(S32 age) const
-{
-	S32 bin;
-	bin = mCurBin - age;
-
-	while (bin < 0)
-	{
-		bin += mNumBins;
-	}
-
-	if (bin == mNextBin)
-	{
-		// Bogus for bin we're currently working on.
-		return 0.f;
-	}
-
-	return mBeginTime[bin];
-}
-
-F64 LLStat::getPrevTime(S32 age) const
-{
-	S32 bin;
-	bin = mCurBin - age;
-
-	while (bin < 0)
-	{
-		bin += mNumBins;
-	}
-
-	if (bin == mNextBin)
-	{
-		// Bogus for bin we're currently working on.
-		return 0.f;
-	}
-
-	return mTime[bin];
-}
-
-F32 LLStat::getBin(S32 bin) const
-{
-	return mBins[bin];
-}
-
-F32 LLStat::getBinPerSec(S32 bin) const
-{
-	return mBins[bin] / mDT[bin];
-}
-
-F64 LLStat::getBinBeginTime(S32 bin) const
-{
-	return mBeginTime[bin];
-}
-
-F64 LLStat::getBinTime(S32 bin) const
-{
-	return mTime[bin];
-}
-
-F32 LLStat::getCurrent() const
-{
-	return mBins[mCurBin];
-}
-
-F32 LLStat::getCurrentPerSec() const
-{
-	return mBins[mCurBin] / mDT[mCurBin];
-}
-
-F64 LLStat::getCurrentBeginTime() const
-{
-	return mBeginTime[mCurBin];
-}
-
-F64 LLStat::getCurrentTime() const
-{
-	return mTime[mCurBin];
-}
-
-F32 LLStat::getCurrentDuration() const
-{
-	return mDT[mCurBin];
-}
-
-F32 LLStat::getMeanPerSec() const
-{
-	U32 i;
-	F32 value = 0.f;
-	F32 dt    = 0.f;
-
-	for (i = 0; (i < mNumBins) && (i < mNumValues); i++)
-	{
-		// Skip the bin we're currently filling.
-		if (i == (U32)mNextBin)
-		{
-			continue;
-		}
-		value += mBins[i];
-		dt    += mDT[i];
-	}
-
-	if (dt > 0.f)
-	{
-		return value/dt;
-	}
-	else
-	{
-		return 0.f;
-	}
-}
-
-F32 LLStat::getMeanDuration() const
-{
-	F32 dur = 0.0f;
-	U32 count = 0;
-	for (U32 i=0; (i < mNumBins) && (i < mNumValues); i++)
-	{
-		if (i == (U32)mNextBin)
-		{
-			continue;
-		}
-		dur += mDT[i];
-		count++;
-	}
-
-	if (count > 0)
-	{
-		dur /= F32(count);
-		return dur;
-	}
-	else
-	{
-		return 0.f;
-	}
-}
-
-F32 LLStat::getMaxPerSec() const
-{
-	U32 i;
-	F32 value;
-
-	if (mNextBin != 0)
-	{
-		value = mBins[0]/mDT[0];
-	}
-	else if (mNumValues > 0)
-	{
-		value = mBins[1]/mDT[1];
-	}
-	else
-	{
-		value = 0.f;
-	}
-
-	for (i = 0; (i < mNumBins) && (i < mNumValues); i++)
-	{
-		// Skip the bin we're currently filling.
-		if (i == (U32)mNextBin)
-		{
-			continue;
-		}
-		value = llmax(value, mBins[i]/mDT[i]);
-	}
-	return value;
-}
-
-F32 LLStat::getMinPerSec() const
-{
-	U32 i;
-	F32 value;
-	
-	if (mNextBin != 0)
-	{
-		value = mBins[0]/mDT[0];
-	}
-	else if (mNumValues > 0)
-	{
-		value = mBins[1]/mDT[1];
-	}
-	else
-	{
-		value = 0.f;
-	}
-
-	for (i = 0; (i < mNumBins) && (i < mNumValues); i++)
-	{
-		// Skip the bin we're currently filling.
-		if (i == (U32)mNextBin)
-		{
-			continue;
-		}
-		value = llmin(value, mBins[i]/mDT[i]);
-	}
-	return value;
-}
-
-F32 LLStat::getMinDuration() const
-{
-	F32 dur = 0.0f;
-	for (U32 i=0; (i < mNumBins) && (i < mNumValues); i++)
-	{
-		dur = llmin(dur, mDT[i]);
-	}
-	return dur;
-}
-
-U32 LLStat::getNumValues() const
-{
-	return mNumValues;
-}
-
-S32 LLStat::getNumBins() const
-{
-	return mNumBins;
-}
-
-S32 LLStat::getCurBin() const
-{
-	return mCurBin;
-}
-
-S32 LLStat::getNextBin() const
-{
-	return mNextBin;
-}
-
-F64 LLStat::getLastTime() const
-{
-	return mLastTime;
-}
diff --git a/indra/llcommon/llstat.h b/indra/llcommon/llstat.h
deleted file mode 100644
index 1a8404cc07..0000000000
--- a/indra/llcommon/llstat.h
+++ /dev/null
@@ -1,353 +0,0 @@
-/** 
- * @file llstat.h
- * @brief Runtime statistics accumulation.
- *
- * $LicenseInfo:firstyear=2001&license=viewerlgpl$
- * Second Life Viewer Source Code
- * Copyright (C) 2010, Linden Research, Inc.
- * 
- * This library is free software; you can redistribute it and/or
- * modify it under the terms of the GNU Lesser General Public
- * License as published by the Free Software Foundation;
- * version 2.1 of the License only.
- * 
- * This library is distributed in the hope that it will be useful,
- * but WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
- * Lesser General Public License for more details.
- * 
- * You should have received a copy of the GNU Lesser General Public
- * License along with this library; if not, write to the Free Software
- * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA
- * 
- * Linden Research, Inc., 945 Battery Street, San Francisco, CA  94111  USA
- * $/LicenseInfo$
- */
-
-#ifndef LL_LLSTAT_H
-#define LL_LLSTAT_H
-
-#include <deque>
-#include <map>
-
-#include "lltimer.h"
-#include "llframetimer.h"
-#include "llfile.h"
-
-class	LLSD;
-
-// Set this if longer stats are needed
-#define ENABLE_LONG_TIME_STATS	0
-
-//
-// Accumulates statistics for an arbitrary length of time.
-// Does this by maintaining a chain of accumulators, each one
-// accumulation the results of the parent.  Can scale to arbitrary
-// amounts of time with very low memory cost.
-//
-
-class LL_COMMON_API LLStatAccum
-{
-protected:
-	LLStatAccum(bool use_frame_timer);
-	virtual ~LLStatAccum();
-
-public:
-	enum TimeScale {
-		SCALE_100MS,
-		SCALE_SECOND,
-		SCALE_MINUTE,
-#if ENABLE_LONG_TIME_STATS
-		SCALE_HOUR,
-		SCALE_DAY,
-		SCALE_WEEK,
-#endif
-		NUM_SCALES,			// Use to size storage arrays
-		SCALE_PER_FRAME		// For latest frame information - should be after NUM_SCALES since this doesn't go into the time buckets
-	};
-
-	static U64 sScaleTimes[NUM_SCALES];
-
-	virtual F32 meanValue(TimeScale scale) const;
-		// see the subclasses for the specific meaning of value
-
-	F32 meanValueOverLast100ms()  const { return meanValue(SCALE_100MS);  }
-	F32 meanValueOverLastSecond() const	{ return meanValue(SCALE_SECOND); }
-	F32 meanValueOverLastMinute() const	{ return meanValue(SCALE_MINUTE); }
-
-	void reset(U64 when);
-
-	void sum(F64 value);
-	void sum(F64 value, U64 when);
-
-	U64 getCurrentUsecs() const;
-		// Get current microseconds based on timer type
-
-	BOOL	mUseFrameTimer;
-	BOOL	mRunning;
-
-	U64		mLastTime;
-	
-	struct Bucket
-	{
-		Bucket() :
-			accum(0.0),
-			endTime(0),
-			lastValid(false),
-			lastAccum(0.0)
-		{}
-
-		F64	accum;
-		U64	endTime;
-
-		bool	lastValid;
-		F64	lastAccum;
-	};
-
-	Bucket	mBuckets[NUM_SCALES];
-
-	BOOL 	mLastSampleValid;
-	F64 	mLastSampleValue;
-};
-
-class LL_COMMON_API LLStatMeasure : public LLStatAccum
-	// gathers statistics about things that are measured
-	// ex.: tempature, time dilation
-{
-public:
-	LLStatMeasure(bool use_frame_timer = true);
-
-	void sample(F64);
-	void sample(S32 v) { sample((F64)v); }
-	void sample(U32 v) { sample((F64)v); }
-	void sample(S64 v) { sample((F64)v); }
-	void sample(U64 v) { sample((F64)v); }
-};
-
-
-class LL_COMMON_API LLStatRate : public LLStatAccum
-	// gathers statistics about things that can be counted over time
-	// ex.: LSL instructions executed, messages sent, simulator frames completed
-	// renders it in terms of rate of thing per second
-{
-public:
-	LLStatRate(bool use_frame_timer = true);
-
-	void count(U32);
-		// used to note that n items have occured
-	
-	void mark();
-		// used for counting the rate thorugh a point in the code
-};
-
-
-class LL_COMMON_API LLStatTime : public LLStatAccum
-	// gathers statistics about time spent in a block of code
-	// measure average duration per second in the block
-{
-public:
-	LLStatTime( const std::string & key = "undefined" );
-
-	U32		mFrameNumber;		// Current frame number
-	U64		mTotalTimeInFrame;	// Total time (microseconds) accumulated during the last frame
-
-	void	setKey( const std::string & key )		{ mKey = key;	};
-
-	virtual F32 meanValue(TimeScale scale) const;
-
-private:
-	void start();				// Start and stop measuring time block
-	void stop();
-
-	std::string		mKey;		// Tag representing this time block
-
-#if LL_DEBUG
-	BOOL			mRunning;	// TRUE if start() has been called
-#endif
-
-	friend class LLPerfBlock;
-};
-
-// ----------------------------------------------------------------------------
-
-
-// Use this class on the stack to record statistics about an area of code
-class LL_COMMON_API LLPerfBlock
-{
-public:
-    struct StatEntry
-    {
-            StatEntry(const std::string& key) : mStat(LLStatTime(key)), mCount(0) {}
-            LLStatTime  mStat;
-            U32         mCount;
-    };
-    typedef std::map<std::string, StatEntry*>		stat_map_t;
-
-	// Use this constructor for pre-defined LLStatTime objects
-	LLPerfBlock(LLStatTime* stat);
-
-	// Use this constructor for normal, optional LLPerfBlock time slices
-	LLPerfBlock( const char* key );
-
-	// Use this constructor for dynamically created LLPerfBlock time slices
-	// that are only enabled by specific control flags
-	LLPerfBlock( const char* key1, const char* key2, S32 flags = LLSTATS_BASIC_STATS );
-
-	~LLPerfBlock();
-
-	enum
-	{	// Stats bitfield flags
-		LLSTATS_NO_OPTIONAL_STATS	= 0x00,		// No optional stats gathering, just pre-defined LLStatTime objects
-		LLSTATS_BASIC_STATS			= 0x01,		// Gather basic optional runtime stats
-		LLSTATS_SCRIPT_FUNCTIONS	= 0x02,		// Include LSL function calls
-	};
-	static void setStatsFlags( S32 flags )	{ sStatsFlags = flags;	};
-	static S32  getStatsFlags()				{ return sStatsFlags;	};
-
-	static void clearDynamicStats();		// Reset maps to clear out dynamic objects
-	static void addStatsToLLSDandReset( LLSD & stats,		// Get current information and clear time bin
-										LLStatAccum::TimeScale scale );
-
-private:
-	// Initialize dynamically created LLStatTime objects
-    void initDynamicStat(const std::string& key);
-
-	std::string				mLastPath;				// Save sCurrentStatPath when this is called
-	LLStatTime * 			mPredefinedStat;		// LLStatTime object to get data
-	StatEntry *				mDynamicStat;   		// StatEntryobject to get data
-
-	static S32				sStatsFlags;			// Control what is being recorded
-    static stat_map_t		sStatMap;				// Map full path string to LLStatTime objects
-	static std::string		sCurrentStatPath;		// Something like "frame/physics/physics step"
-};
-
-// ----------------------------------------------------------------------------
-
-class LL_COMMON_API LLPerfStats
-{
-public:
-    LLPerfStats(const std::string& process_name = "unknown", S32 process_pid = 0);
-    virtual ~LLPerfStats();
-
-    virtual void init();    // Reset and start all stat timers
-    virtual void updatePerFrameStats();
-    // Override these function to add process-specific information to the performance log header and per-frame logging.
-    virtual void addProcessHeaderInfo(LLSD& info) { /* not implemented */ }
-    virtual void addProcessFrameInfo(LLSD& info, LLStatAccum::TimeScale scale) { /* not implemented */ }
-
-    // High-resolution frame stats
-    BOOL    frameStatsIsRunning()                                { return (mReportPerformanceStatEnd > 0.);        };
-    F32     getReportPerformanceInterval() const                { return mReportPerformanceStatInterval;        };
-    void    setReportPerformanceInterval( F32 interval )        { mReportPerformanceStatInterval = interval;    };
-    void    setReportPerformanceDuration( F32 seconds, S32 flags = LLPerfBlock::LLSTATS_NO_OPTIONAL_STATS );
-    void    setProcessName(const std::string& process_name) { mProcessName = process_name; }
-    void    setProcessPID(S32 process_pid) { mProcessPID = process_pid; }
-
-protected:
-    void    openPerfStatsFile();                    // Open file for high resolution metrics logging
-    void    dumpIntervalPerformanceStats();
-
-    llofstream      mFrameStatsFile;            // File for per-frame stats
-    BOOL            mFrameStatsFileFailure;        // Flag to prevent repeat opening attempts
-    BOOL            mSkipFirstFrameStats;        // Flag to skip one (partial) frame report
-    std::string     mProcessName;
-    S32             mProcessPID;
-
-private:
-    F32 mReportPerformanceStatInterval;    // Seconds between performance stats
-    F64 mReportPerformanceStatEnd;        // End time (seconds) for performance stats
-};
-
-// ----------------------------------------------------------------------------
-class LL_COMMON_API LLStat
-{
-private:
-	typedef std::multimap<std::string, LLStat*> stat_map_t;
-
-	void init();
-	static stat_map_t& getStatList();
-
-public:
-	LLStat(U32 num_bins = 32, BOOL use_frame_timer = FALSE);
-	LLStat(std::string name, U32 num_bins = 32, BOOL use_frame_timer = FALSE);
-	~LLStat();
-
-	void reset();
-
-	void start();	// Start the timer for the current "frame", otherwise uses the time tracked from
-					// the last addValue
-	void addValue(const F32 value = 1.f); // Adds the current value being tracked, and tracks the DT.
-	void addValue(const S32 value) { addValue((F32)value); }
-	void addValue(const U32 value) { addValue((F32)value); }
-
-	void setBeginTime(const F64 time);
-	void addValueTime(const F64 time, const F32 value = 1.f);
-	
-	S32 getCurBin() const;
-	S32 getNextBin() const;
-	
-	F32 getCurrent() const;
-	F32 getCurrentPerSec() const;
-	F64 getCurrentBeginTime() const;
-	F64 getCurrentTime() const;
-	F32 getCurrentDuration() const;
-	
-	F32 getPrev(S32 age) const;				// Age is how many "addValues" previously - zero is current
-	F32 getPrevPerSec(S32 age) const;		// Age is how many "addValues" previously - zero is current
-	F64 getPrevBeginTime(S32 age) const;
-	F64 getPrevTime(S32 age) const;
-	
-	F32 getBin(S32 bin) const;
-	F32 getBinPerSec(S32 bin) const;
-	F64 getBinBeginTime(S32 bin) const;
-	F64 getBinTime(S32 bin) const;
-
-	F32 getMax() const;
-	F32 getMaxPerSec() const;
-	
-	F32 getMean() const;
-	F32 getMeanPerSec() const;
-	F32 getMeanDuration() const;
-
-	F32 getMin() const;
-	F32 getMinPerSec() const;
-	F32 getMinDuration() const;
-
-	F32 getSum() const;
-	F32 getSumDuration() const;
-
-	U32 getNumValues() const;
-	S32 getNumBins() const;
-
-	F64 getLastTime() const;
-private:
-	BOOL mUseFrameTimer;
-	U32 mNumValues;
-	U32 mNumBins;
-	F32 mLastValue;
-	F64 mLastTime;
-	F32 *mBins;
-	F64 *mBeginTime;
-	F64 *mTime;
-	F32 *mDT;
-	S32 mCurBin;
-	S32 mNextBin;
-	
-	std::string mName;
-
-	static LLTimer sTimer;
-	static LLFrameTimer sFrameTimer;
-	
-public:
-	static LLStat* getStat(const std::string& name)
-	{
-		// return the first stat that matches 'name'
-		stat_map_t::iterator iter = getStatList().find(name);
-		if (iter != getStatList().end())
-			return iter->second;
-		else
-			return NULL;
-	}
-};
-	
-#endif // LL_STAT_
diff --git a/indra/llcommon/llstatenums.h b/indra/llcommon/llstatenums.h
index 81c4085d16..ab9b6709e8 100644
--- a/indra/llcommon/llstatenums.h
+++ b/indra/llcommon/llstatenums.h
@@ -26,7 +26,7 @@
 #ifndef LL_LLSTATENUMS_H
 #define LL_LLSTATENUMS_H
 
-enum
+enum ESimStatID
 {
 	LL_SIM_STAT_TIME_DILATION         =  0,
 	LL_SIM_STAT_FPS                   =  1,
diff --git a/indra/llcommon/llstl.h b/indra/llcommon/llstl.h
index d3941e1bc9..424138dad1 100644
--- a/indra/llcommon/llstl.h
+++ b/indra/llcommon/llstl.h
@@ -31,6 +31,7 @@
 #include <algorithm>
 #include <map>
 #include <vector>
+#include <list>
 #include <set>
 #include <deque>
 #include <typeinfo>
diff --git a/indra/llcommon/llstring.cpp b/indra/llcommon/llstring.cpp
index 0c32679744..67bbc58bf5 100644
--- a/indra/llcommon/llstring.cpp
+++ b/indra/llcommon/llstring.cpp
@@ -28,11 +28,10 @@
 
 #include "llstring.h"
 #include "llerror.h"
+#include "llfasttimer.h"
 
 #if LL_WINDOWS
-#define WIN32_LEAN_AND_MEAN
-#include <winsock2.h>
-#include <windows.h>
+#include "llwin32headerslean.h"
 #include <winnls.h> // for WideCharToMultiByte
 #endif
 
diff --git a/indra/llcommon/llstring.h b/indra/llcommon/llstring.h
index 119efc7957..9d81ac25dd 100644
--- a/indra/llcommon/llstring.h
+++ b/indra/llcommon/llstring.h
@@ -31,8 +31,9 @@
 #include <cstdio>
 #include <locale>
 #include <iomanip>
+#include <algorithm>
 #include "llsd.h"
-#include "llfasttimer.h"
+#include "llformat.h"
 
 #if LL_LINUX || LL_SOLARIS
 #include <wctype.h>
diff --git a/indra/llcommon/llsys.cpp b/indra/llcommon/llsys.cpp
index c96f2191f3..d6fe648b42 100644
--- a/indra/llcommon/llsys.cpp
+++ b/indra/llcommon/llsys.cpp
@@ -42,6 +42,7 @@
 #include "llprocessor.h"
 #include "llerrorcontrol.h"
 #include "llevents.h"
+#include "llformat.h"
 #include "lltimer.h"
 #include "llsdserialize.h"
 #include "llsdutil.h"
@@ -58,9 +59,7 @@
 using namespace llsd;
 
 #if LL_WINDOWS
-#	define WIN32_LEAN_AND_MEAN
-#	include <winsock2.h>
-#	include <windows.h>
+#	include "llwin32headerslean.h"
 #   include <psapi.h>               // GetPerformanceInfo() et al.
 #elif LL_DARWIN
 #	include <errno.h>
diff --git a/indra/llcommon/llthread.cpp b/indra/llcommon/llthread.cpp
index 1d56a52c32..6374b5398b 100644
--- a/indra/llcommon/llthread.cpp
+++ b/indra/llcommon/llthread.cpp
@@ -29,8 +29,11 @@
 #include "apr_portable.h"
 
 #include "llthread.h"
+#include "llmutex.h"
 
 #include "lltimer.h"
+#include "lltrace.h"
+#include "lltracethreadrecorder.h"
 
 #if LL_LINUX || LL_SOLARIS
 #include <sched.h>
@@ -56,12 +59,17 @@
 // 
 //----------------------------------------------------------------------------
 
-#if !LL_DARWIN
-U32 ll_thread_local sThreadID = 0;
+#if LL_DARWIN
+// statically allocated thread local storage not supported in Darwin executable formats
+#elif LL_WINDOWS
+U32 __declspec(thread) sThreadID = 0;
+#elif LL_LINUX
+U32 __thread sThreadID = 0;
 #endif 
 
 U32 LLThread::sIDIter = 0;
 
+
 LL_COMMON_API void assert_main_thread()
 {
 	static U32 s_thread_id = LLThread::currentID();
@@ -85,6 +93,8 @@ void *APR_THREAD_FUNC LLThread::staticRun(apr_thread_t *apr_threadp, void *datap
 {
 	LLThread *threadp = (LLThread *)datap;
 
+	LLTrace::ThreadRecorder* thread_recorder = new LLTrace::SlaveThreadRecorder();
+
 #if !LL_DARWIN
 	sThreadID = threadp->mID;
 #endif
@@ -97,16 +107,18 @@ void *APR_THREAD_FUNC LLThread::staticRun(apr_thread_t *apr_threadp, void *datap
 	// We're done with the run function, this thread is done executing now.
 	threadp->mStatus = STOPPED;
 
+	delete thread_recorder;
+
 	return NULL;
 }
 
-
 LLThread::LLThread(const std::string& name, apr_pool_t *poolp) :
 	mPaused(FALSE),
 	mName(name),
 	mAPRThreadp(NULL),
 	mStatus(STOPPED)
 {
+
 	mID = ++sIDIter;
 
 	// Thread creation probably CAN be paranoid about APR being initialized, if necessary
@@ -152,7 +164,7 @@ void LLThread::shutdown()
 			//llinfos << "LLThread::~LLThread() Killing thread " << mName << " Status: " << mStatus << llendl;
 			// Now wait a bit for the thread to exit
 			// It's unclear whether I should even bother doing this - this destructor
-			// should netver get called unless we're already stopped, really...
+			// should never get called unless we're already stopped, really...
 			S32 counter = 0;
 			const S32 MAX_WAIT = 600;
 			while (counter < MAX_WAIT)
@@ -282,7 +294,13 @@ void LLThread::setQuitting()
 // static
 U32 LLThread::currentID()
 {
+#if LL_DARWIN
+	// statically allocated thread local storage not supported in Darwin executable formats
 	return (U32)apr_os_thread_current();
+#else
+	return sThreadID;
+#endif
+
 }
 
 // static
@@ -315,155 +333,6 @@ void LLThread::wakeLocked()
 
 //============================================================================
 
-LLMutex::LLMutex(apr_pool_t *poolp) :
-	mAPRMutexp(NULL), mCount(0), mLockingThread(NO_THREAD)
-{
-	//if (poolp)
-	//{
-	//	mIsLocalPool = FALSE;
-	//	mAPRPoolp = poolp;
-	//}
-	//else
-	{
-		mIsLocalPool = TRUE;
-		apr_pool_create(&mAPRPoolp, NULL); // Create a subpool for this thread
-	}
-	apr_thread_mutex_create(&mAPRMutexp, APR_THREAD_MUTEX_UNNESTED, mAPRPoolp);
-}
-
-
-LLMutex::~LLMutex()
-{
-#if MUTEX_DEBUG
-	//bad assertion, the subclass LLSignal might be "locked", and that's OK
-	//llassert_always(!isLocked()); // better not be locked!
-#endif
-	apr_thread_mutex_destroy(mAPRMutexp);
-	mAPRMutexp = NULL;
-	if (mIsLocalPool)
-	{
-		apr_pool_destroy(mAPRPoolp);
-	}
-}
-
-
-void LLMutex::lock()
-{
-	if(isSelfLocked())
-	{ //redundant lock
-		mCount++;
-		return;
-	}
-	
-	apr_thread_mutex_lock(mAPRMutexp);
-	
-#if MUTEX_DEBUG
-	// Have to have the lock before we can access the debug info
-	U32 id = LLThread::currentID();
-	if (mIsLocked[id] != FALSE)
-		llerrs << "Already locked in Thread: " << id << llendl;
-	mIsLocked[id] = TRUE;
-#endif
-
-#if LL_DARWIN
-	mLockingThread = LLThread::currentID();
-#else
-	mLockingThread = sThreadID;
-#endif
-}
-
-void LLMutex::unlock()
-{
-	if (mCount > 0)
-	{ //not the root unlock
-		mCount--;
-		return;
-	}
-	
-#if MUTEX_DEBUG
-	// Access the debug info while we have the lock
-	U32 id = LLThread::currentID();
-	if (mIsLocked[id] != TRUE)
-		llerrs << "Not locked in Thread: " << id << llendl;	
-	mIsLocked[id] = FALSE;
-#endif
-
-	mLockingThread = NO_THREAD;
-	apr_thread_mutex_unlock(mAPRMutexp);
-}
-
-bool LLMutex::isLocked()
-{
-	apr_status_t status = apr_thread_mutex_trylock(mAPRMutexp);
-	if (APR_STATUS_IS_EBUSY(status))
-	{
-		return true;
-	}
-	else
-	{
-		apr_thread_mutex_unlock(mAPRMutexp);
-		return false;
-	}
-}
-
-bool LLMutex::isSelfLocked()
-{
-#if LL_DARWIN
-	return mLockingThread == LLThread::currentID();
-#else
-	return mLockingThread == sThreadID;
-#endif
-}
-
-U32 LLMutex::lockingThread() const
-{
-	return mLockingThread;
-}
-
-//============================================================================
-
-LLCondition::LLCondition(apr_pool_t *poolp) :
-	LLMutex(poolp)
-{
-	// base class (LLMutex) has already ensured that mAPRPoolp is set up.
-
-	apr_thread_cond_create(&mAPRCondp, mAPRPoolp);
-}
-
-
-LLCondition::~LLCondition()
-{
-	apr_thread_cond_destroy(mAPRCondp);
-	mAPRCondp = NULL;
-}
-
-
-void LLCondition::wait()
-{
-	if (!isLocked())
-	{ //mAPRMutexp MUST be locked before calling apr_thread_cond_wait
-		apr_thread_mutex_lock(mAPRMutexp);
-#if MUTEX_DEBUG
-		// avoid asserts on destruction in non-release builds
-		U32 id = LLThread::currentID();
-		mIsLocked[id] = TRUE;
-#endif
-	}
-	apr_thread_cond_wait(mAPRCondp, mAPRMutexp);
-}
-
-void LLCondition::signal()
-{
-	apr_thread_cond_signal(mAPRCondp);
-}
-
-void LLCondition::broadcast()
-{
-	apr_thread_cond_broadcast(mAPRCondp);
-}
-
-//============================================================================
-
 //----------------------------------------------------------------------------
 
 //static
@@ -514,7 +383,6 @@ LLThreadSafeRefCount::~LLThreadSafeRefCount()
 	}
 }
 
-
 //============================================================================
 
 LLResponder::~LLResponder()
diff --git a/indra/llcommon/llthread.h b/indra/llcommon/llthread.h
index 5c8bbca2ca..92323f5fda 100644
--- a/indra/llcommon/llthread.h
+++ b/indra/llcommon/llthread.h
@@ -30,20 +30,14 @@
 #include "llapp.h"
 #include "llapr.h"
 #include "apr_thread_cond.h"
+#include "llmutex.h"
 
-class LLThread;
-class LLMutex;
-class LLCondition;
-
-#if LL_WINDOWS
-#define ll_thread_local __declspec(thread)
-#else
-#define ll_thread_local __thread
-#endif
+LL_COMMON_API void assert_main_thread();
 
 class LL_COMMON_API LLThread
 {
 private:
+	friend class LLMutex;
 	static U32 sIDIter;
 
 public:
@@ -101,7 +95,7 @@ private:
 
 protected:
 	std::string			mName;
-	LLCondition*		mRunCondition;
+	class LLCondition*	mRunCondition;
 	LLMutex*			mDataLock;
 
 	apr_thread_t		*mAPRThreadp;
@@ -139,75 +133,6 @@ protected:
 	// mDataLock->unlock();
 };
 
-//============================================================================
-
-#define MUTEX_DEBUG (LL_DEBUG || LL_RELEASE_WITH_DEBUG_INFO)
-
-class LL_COMMON_API LLMutex
-{
-public:
-	typedef enum
-	{
-		NO_THREAD = 0xFFFFFFFF
-	} e_locking_thread;
-
-	LLMutex(apr_pool_t *apr_poolp); // NULL pool constructs a new pool for the mutex
-	virtual ~LLMutex();
-	
-	void lock();		// blocks
-	void unlock();
-	bool isLocked(); 	// non-blocking, but does do a lock/unlock so not free
-	bool isSelfLocked(); //return true if locked in a same thread
-	U32 lockingThread() const; //get ID of locking thread
-	
-protected:
-	apr_thread_mutex_t *mAPRMutexp;
-	mutable U32			mCount;
-	mutable U32			mLockingThread;
-	
-	apr_pool_t			*mAPRPoolp;
-	BOOL				mIsLocalPool;
-	
-#if MUTEX_DEBUG
-	std::map<U32, BOOL> mIsLocked;
-#endif
-};
-
-// Actually a condition/mutex pair (since each condition needs to be associated with a mutex).
-class LL_COMMON_API LLCondition : public LLMutex
-{
-public:
-	LLCondition(apr_pool_t *apr_poolp); // Defaults to global pool, could use the thread pool as well.
-	~LLCondition();
-	
-	void wait();		// blocks
-	void signal();
-	void broadcast();
-	
-protected:
-	apr_thread_cond_t *mAPRCondp;
-};
-
-class LLMutexLock
-{
-public:
-	LLMutexLock(LLMutex* mutex)
-	{
-		mMutex = mutex;
-		
-		if(mMutex)
-			mMutex->lock();
-	}
-	~LLMutexLock()
-	{
-		if(mMutex)
-			mMutex->unlock();
-	}
-private:
-	LLMutex* mMutex;
-};
-
-//============================================================================
 
 void LLThread::lockData()
 {
diff --git a/indra/llcommon/llthreadlocalstorage.h b/indra/llcommon/llthreadlocalstorage.h
new file mode 100644
index 0000000000..5a38d54eea
--- /dev/null
+++ b/indra/llcommon/llthreadlocalstorage.h
@@ -0,0 +1,275 @@
+/** 
+ * @file llthreadlocalstorage.h
+ * @author Richard
+ * @brief Class wrappers for thread local storage
+ *
+ * $LicenseInfo:firstyear=2004&license=viewerlgpl$
+ * Second Life Viewer Source Code
+ * Copyright (C) 2010, Linden Research, Inc.
+ * 
+ * This library is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation;
+ * version 2.1 of the License only.
+ * 
+ * This library is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ * 
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with this library; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA
+ * 
+ * Linden Research, Inc., 945 Battery Street, San Francisco, CA  94111  USA
+ * $/LicenseInfo$
+ */
+
+#ifndef LL_LLTHREADLOCALSTORAGE_H
+#define LL_LLTHREADLOCALSTORAGE_H
+
+#include "llinstancetracker.h"
+#include "llapr.h"
+
+class LLThreadLocalPointerBase : public LLInstanceTracker<LLThreadLocalPointerBase>
+{
+public:
+	LLThreadLocalPointerBase()
+	:	mThreadKey(NULL)
+	{
+		if (sInitialized)
+		{
+			initStorage();
+		}
+	}
+
+	LLThreadLocalPointerBase( const LLThreadLocalPointerBase& other)
+		:	mThreadKey(NULL)
+	{
+		if (sInitialized)
+		{
+			initStorage();
+		}
+	}
+
+	~LLThreadLocalPointerBase()
+	{
+		destroyStorage();
+	}
+
+	static void initAllThreadLocalStorage();
+	static void destroyAllThreadLocalStorage();
+
+protected:
+	void set(void* value);
+
+	LL_FORCE_INLINE void* get() const
+	{
+		// llassert(sInitialized);
+		void* ptr;
+		apr_status_t result =
+		apr_threadkey_private_get(&ptr, mThreadKey);
+		if (result != APR_SUCCESS)
+		{
+			ll_apr_warn_status(result);
+			llerrs << "Failed to get thread local data" << llendl;
+		}
+		return ptr;
+	}
+
+	void initStorage();
+	void destroyStorage();
+
+protected:
+	apr_threadkey_t* mThreadKey;
+	static bool		sInitialized;
+};
+
+template <typename T>
+class LLThreadLocalPointer : public LLThreadLocalPointerBase
+{
+public:
+
+	LLThreadLocalPointer()
+	{}
+
+	explicit LLThreadLocalPointer(T* value)
+	{
+		set(value);
+	}
+
+
+	LLThreadLocalPointer(const LLThreadLocalPointer<T>& other)
+	:	LLThreadLocalPointerBase(other)
+	{
+		set(other.get());		
+	}
+
+	LL_FORCE_INLINE T* get() const
+	{
+		return (T*)LLThreadLocalPointerBase::get();
+	}
+
+	T* operator -> () const
+	{
+		return (T*)get();
+	}
+
+	T& operator*() const
+	{
+		return *(T*)get();
+	}
+
+	LLThreadLocalPointer<T>& operator = (T* value)
+	{
+		set((void*)value);
+		return *this;
+	}
+
+	bool operator ==(const T* other) const
+	{
+		if (!sInitialized) return false;
+		return get() == other;
+	}
+};
+
+template<typename DERIVED_TYPE>
+class LLThreadLocalSingleton
+{
+	typedef enum e_init_state
+	{
+		UNINITIALIZED = 0,
+		CONSTRUCTING,
+		INITIALIZING,
+		INITIALIZED,
+		DELETED
+	} EInitState;
+
+public:
+	LLThreadLocalSingleton()
+	{}
+	
+	virtual ~LLThreadLocalSingleton()
+	{
+		sInstance = NULL;
+		sInitState = DELETED;
+	}
+
+	static void deleteSingleton()
+	{
+		delete sInstance;
+		sInstance = NULL;
+		sInitState = DELETED;
+	}
+
+	static DERIVED_TYPE* getInstance()
+	{
+		if (sInitState == CONSTRUCTING)
+		{
+			llerrs << "Tried to access singleton " << typeid(DERIVED_TYPE).name() << " from singleton constructor!" << llendl;
+		}
+
+		if (sInitState == DELETED)
+		{
+			llwarns << "Trying to access deleted singleton " << typeid(DERIVED_TYPE).name() << " creating new instance" << llendl;
+		}
+
+		if (!sInstance) 
+		{
+			sInitState = CONSTRUCTING;
+			sInstance = new DERIVED_TYPE(); 
+			sInitState = INITIALIZING;
+			sInstance->initSingleton(); 
+			sInitState = INITIALIZED;	
+		}
+
+		return sInstance;
+	}
+
+	static DERIVED_TYPE* getIfExists()
+	{
+		return sInstance;
+	}
+
+	// Reference version of getInstance()
+	// Preferred over getInstance() as it disallows checking for NULL
+	static DERIVED_TYPE& instance()
+	{
+		return *getInstance();
+	}
+
+	// Has this singleton been created uet?
+	// Use this to avoid accessing singletons before the can safely be constructed
+	static bool instanceExists()
+	{
+		return sInitState == INITIALIZED;
+	}
+
+	// Has this singleton already been deleted?
+	// Use this to avoid accessing singletons from a static object's destructor
+	static bool destroyed()
+	{
+		return sInitState == DELETED;
+	}
+private:
+	LLThreadLocalSingleton(const LLThreadLocalSingleton& other);
+	virtual void initSingleton() {}
+
+#ifdef LL_WINDOWS
+	static __declspec(thread) DERIVED_TYPE* sInstance;
+	static __declspec(thread) EInitState sInitState;
+#elif LL_LINUX
+	static __thread DERIVED_TYPE* sInstance;
+	static __thread EInitState sInitState;
+#endif
+};
+
+#ifdef LL_WINDOWS
+template<typename DERIVED_TYPE>
+__declspec(thread) DERIVED_TYPE* LLThreadLocalSingleton<DERIVED_TYPE>::sInstance = NULL;
+
+template<typename DERIVED_TYPE>
+__declspec(thread) typename LLThreadLocalSingleton<DERIVED_TYPE>::EInitState LLThreadLocalSingleton<DERIVED_TYPE>::sInitState = LLThreadLocalSingleton<DERIVED_TYPE>::UNINITIALIZED;
+#elif LL_LINUX
+template<typename DERIVED_TYPE>
+__thread DERIVED_TYPE* LLThreadLocalSingleton<DERIVED_TYPE>::sInstance = NULL;
+
+template<typename DERIVED_TYPE>
+__thread typename LLThreadLocalSingleton<DERIVED_TYPE>::EInitState LLThreadLocalSingleton<DERIVED_TYPE>::sInitState = LLThreadLocalSingleton<DERIVED_TYPE>::UNINITIALIZED;
+#endif
+
+template<typename DERIVED_TYPE>
+class LLThreadLocalSingletonPointer
+{
+public:
+	void operator =(DERIVED_TYPE* value)
+	{
+		sInstance = value;
+	}
+
+	LL_FORCE_INLINE static DERIVED_TYPE* getInstance()
+	{
+		return sInstance;
+	}
+
+	LL_FORCE_INLINE static void setInstance(DERIVED_TYPE* instance)
+	{
+		sInstance = instance;
+	}
+
+private:
+#ifdef LL_WINDOWS
+	static __declspec(thread) DERIVED_TYPE* sInstance;
+#elif LL_LINUX
+	static __thread DERIVED_TYPE* sInstance;
+#endif
+};
+
+template<typename DERIVED_TYPE>
+#ifdef LL_WINDOWS
+__declspec(thread) DERIVED_TYPE* LLThreadLocalSingletonPointer<DERIVED_TYPE>::sInstance = NULL;
+#elif LL_LINUX
+__thread DERIVED_TYPE* LLThreadLocalSingletonPointer<DERIVED_TYPE>::sInstance = NULL;
+#endif
+
+#endif // LL_LLTHREADLOCALSTORAGE_H
diff --git a/indra/llcommon/lltimer.cpp b/indra/llcommon/lltimer.cpp
index 9ebc6de7f4..838155d54d 100644
--- a/indra/llcommon/lltimer.cpp
+++ b/indra/llcommon/lltimer.cpp
@@ -31,11 +31,9 @@
 #include "u64.h"
 
 #if LL_WINDOWS
-#	define WIN32_LEAN_AND_MEAN
-#	include <winsock2.h>
-#	include <windows.h>
+#	include "llwin32headerslean.h"
 #elif LL_LINUX || LL_SOLARIS || LL_DARWIN
-#       include <errno.h>
+#   include <errno.h>
 #	include <sys/time.h>
 #else 
 #	error "architecture not supported"
@@ -287,14 +285,14 @@ LLTimer::~LLTimer()
 }
 
 // static
-U64 LLTimer::getTotalTime()
+LLUnitImplicit<LLUnits::Microseconds, U64> LLTimer::getTotalTime()
 {
 	// simply call into the implementation function.
 	return totalTime();
 }	
 
 // static
-F64 LLTimer::getTotalSeconds()
+LLUnitImplicit<LLUnits::Seconds, F64> LLTimer::getTotalSeconds()
 {
 	return U64_to_F64(getTotalTime()) * USEC_TO_SEC_F64;
 }
@@ -343,23 +341,23 @@ U64 getElapsedTimeAndUpdate(U64& lastClockCount)
 }
 
 
-F64 LLTimer::getElapsedTimeF64() const
+LLUnitImplicit<LLUnits::Seconds, F64> LLTimer::getElapsedTimeF64() const
 {
 	U64 last = mLastClockCount;
 	return (F64)getElapsedTimeAndUpdate(last) * gClockFrequencyInv;
 }
 
-F32 LLTimer::getElapsedTimeF32() const
+LLUnitImplicit<LLUnits::Seconds, F32> LLTimer::getElapsedTimeF32() const
 {
 	return (F32)getElapsedTimeF64();
 }
 
-F64 LLTimer::getElapsedTimeAndResetF64()
+LLUnitImplicit<LLUnits::Seconds, F64> LLTimer::getElapsedTimeAndResetF64()
 {
 	return (F64)getElapsedTimeAndUpdate(mLastClockCount) * gClockFrequencyInv;
 }
 
-F32 LLTimer::getElapsedTimeAndResetF32()
+LLUnitImplicit<LLUnits::Seconds, F32> LLTimer::getElapsedTimeAndResetF32()
 {
 	return (F32)getElapsedTimeAndResetF64();
 }
@@ -372,7 +370,7 @@ void  LLTimer::setTimerExpirySec(F32 expiration)
 		+ (U64)((F32)(expiration * gClockFrequency));
 }
 
-F32 LLTimer::getRemainingTimeF32() const
+LLUnitImplicit<LLUnits::Seconds, F32> LLTimer::getRemainingTimeF32() const
 {
 	U64 cur_ticks = get_clock_count();
 	if (cur_ticks > mExpirationTicks)
diff --git a/indra/llcommon/lltimer.h b/indra/llcommon/lltimer.h
index 513de0605d..0ba87d1e15 100644
--- a/indra/llcommon/lltimer.h
+++ b/indra/llcommon/lltimer.h
@@ -37,6 +37,7 @@
 #include <string>
 #include <list>
 // units conversions
+#include "llunit.h"
 #ifndef USEC_PER_SEC
     const U32	USEC_PER_SEC	= 1000000;
 #endif
@@ -55,7 +56,7 @@ public:
 protected:	
 	U64 mLastClockCount;
 	U64 mExpirationTicks;
-	BOOL mStarted;
+	bool mStarted;
 
 public:
 	LLTimer();
@@ -66,16 +67,16 @@ public:
 
 	// Return a high precision number of seconds since the start of
 	// this application instance.
-	static F64 getElapsedSeconds()
+	static LLUnitImplicit<LLUnits::Seconds, F64> getElapsedSeconds()
 	{
 		return sTimer->getElapsedTimeF64();
 	}
 
 	// Return a high precision usec since epoch
-	static U64 getTotalTime();
+	static LLUnitImplicit<LLUnits::Microseconds, U64> getTotalTime();
 
 	// Return a high precision seconds since epoch
-	static F64 getTotalSeconds();
+	static LLUnitImplicit<LLUnits::Seconds, F64> getTotalSeconds();
 
 
 	// MANIPULATORS
@@ -86,18 +87,18 @@ public:
 	void setTimerExpirySec(F32 expiration);
 	BOOL checkExpirationAndReset(F32 expiration);
 	BOOL hasExpired() const;
-	F32 getElapsedTimeAndResetF32();	// Returns elapsed time in seconds with reset
-	F64 getElapsedTimeAndResetF64();
+	LLUnitImplicit<LLUnits::Seconds, F32> getElapsedTimeAndResetF32();	// Returns elapsed time in seconds with reset
+	LLUnitImplicit<LLUnits::Seconds, F64> getElapsedTimeAndResetF64();
 
-	F32 getRemainingTimeF32() const;
+	LLUnitImplicit<LLUnits::Seconds, F32> getRemainingTimeF32() const;
 
 	static BOOL knownBadTimer();
 
 	// ACCESSORS
-	F32 getElapsedTimeF32() const;			// Returns elapsed time in seconds
-	F64 getElapsedTimeF64() const;			// Returns elapsed time in seconds
+	LLUnitImplicit<LLUnits::Seconds, F32> getElapsedTimeF32() const;			// Returns elapsed time in seconds
+	LLUnitImplicit<LLUnits::Seconds, F64> getElapsedTimeF64() const;			// Returns elapsed time in seconds
 
-	BOOL getStarted() const { return mStarted; }
+	bool getStarted() const { return mStarted; }
 
 
 	static U64 getCurrentClockCount();		// Returns the raw clockticks
diff --git a/indra/llcommon/lltrace.cpp b/indra/llcommon/lltrace.cpp
new file mode 100644
index 0000000000..9cadd70dd8
--- /dev/null
+++ b/indra/llcommon/lltrace.cpp
@@ -0,0 +1,110 @@
+/** 
+ * @file lltrace.cpp
+ *
+ * $LicenseInfo:firstyear=2001&license=viewerlgpl$
+ * Second Life Viewer Source Code
+ * Copyright (C) 2012, Linden Research, Inc.
+ * 
+ * This library is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation;
+ * version 2.1 of the License only.
+ * 
+ * This library is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ * 
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with this library; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA
+ * 
+ * Linden Research, Inc., 945 Battery Street, San Francisco, CA  94111  USA
+ * $/LicenseInfo$
+ */
+
+#include "linden_common.h"
+
+#include "lltrace.h"
+#include "lltracerecording.h"
+#include "lltracethreadrecorder.h"
+#include "llfasttimer.h"
+
+static bool sInitialized;
+
+namespace LLTrace
+{
+
+static MasterThreadRecorder* gMasterThreadRecorder = NULL;
+
+void init()
+{
+	gMasterThreadRecorder = new MasterThreadRecorder();
+	sInitialized = true;
+}
+
+bool isInitialized()
+{
+	return sInitialized; 
+}
+
+void cleanup()
+{
+	delete gMasterThreadRecorder;
+	gMasterThreadRecorder = NULL;
+}
+
+MasterThreadRecorder& getMasterThreadRecorder()
+{
+	llassert(gMasterThreadRecorder != NULL);
+	return *gMasterThreadRecorder;
+}
+
+LLThreadLocalPointer<ThreadRecorder>& get_thread_recorder_ptr()
+{
+	static LLThreadLocalPointer<ThreadRecorder> s_thread_recorder;
+	return s_thread_recorder;
+}
+
+const LLThreadLocalPointer<ThreadRecorder>& get_thread_recorder()
+{
+	return get_thread_recorder_ptr();
+}
+
+void set_thread_recorder(ThreadRecorder* recorder)
+{
+	get_thread_recorder_ptr() = recorder;
+}
+
+
+TimeBlockTreeNode::TimeBlockTreeNode() 
+:	mBlock(NULL),
+	mParent(NULL),
+	mNeedsSorting(false)
+{}
+
+void TimeBlockTreeNode::setParent( TimeBlock* parent )
+{
+	llassert_always(parent != mBlock);
+	llassert_always(parent != NULL);
+
+	TimeBlockTreeNode* parent_tree_node = get_thread_recorder()->getTimeBlockTreeNode(parent->getIndex());
+	if (!parent_tree_node) return;
+
+	if (mParent)
+	{
+		std::vector<TimeBlock*>& children = mParent->getChildren();
+		std::vector<TimeBlock*>::iterator found_it = std::find(children.begin(), children.end(), mBlock);
+		if (found_it != children.end())
+		{
+			children.erase(found_it);
+		}
+	}
+
+	mParent = parent;
+	mBlock->getPrimaryAccumulator()->mParent = parent;
+	parent_tree_node->mChildren.push_back(mBlock);
+	parent_tree_node->mNeedsSorting = true;
+}
+
+}
diff --git a/indra/llcommon/lltrace.h b/indra/llcommon/lltrace.h
new file mode 100644
index 0000000000..1a156e583e
--- /dev/null
+++ b/indra/llcommon/lltrace.h
@@ -0,0 +1,836 @@
+/** 
+ * @file lltrace.h
+ * @brief Runtime statistics accumulation.
+ *
+ * $LicenseInfo:firstyear=2001&license=viewerlgpl$
+ * Second Life Viewer Source Code
+ * Copyright (C) 2012, Linden Research, Inc.
+ * 
+ * This library is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation;
+ * version 2.1 of the License only.
+ * 
+ * This library is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ * 
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with this library; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA
+ * 
+ * Linden Research, Inc., 945 Battery Street, San Francisco, CA  94111  USA
+ * $/LicenseInfo$
+ */
+
+#ifndef LL_LLTRACE_H
+#define LL_LLTRACE_H
+
+#include "stdtypes.h"
+#include "llpreprocessor.h"
+
+#include "llmemory.h"
+#include "llrefcount.h"
+#include "llunit.h"
+#include "llapr.h"
+#include "llthreadlocalstorage.h"
+
+#include <list>
+
+#define LL_RECORD_BLOCK_TIME(block_timer) LLTrace::TimeBlock::Recorder LL_GLUE_TOKENS(block_time_recorder, __COUNTER__)(block_timer);
+
+namespace LLTrace
+{
+	class Recording;
+
+	typedef LLUnit<LLUnits::Bytes, F64>			Bytes;
+	typedef LLUnit<LLUnits::Kilobytes, F64>		Kilobytes;
+	typedef LLUnit<LLUnits::Megabytes, F64>		Megabytes;
+	typedef LLUnit<LLUnits::Gigabytes, F64>		Gigabytes;
+	typedef LLUnit<LLUnits::Bits, F64>			Bits;
+	typedef LLUnit<LLUnits::Kilobits, F64>		Kilobits;
+	typedef LLUnit<LLUnits::Megabits, F64>		Megabits;
+	typedef LLUnit<LLUnits::Gigabits, F64>		Gigabits;
+
+	typedef LLUnit<LLUnits::Seconds, F64>		Seconds;
+	typedef LLUnit<LLUnits::Milliseconds, F64>	Milliseconds;
+	typedef LLUnit<LLUnits::Minutes, F64>		Minutes;
+	typedef LLUnit<LLUnits::Hours, F64>			Hours;
+	typedef LLUnit<LLUnits::Milliseconds, F64>	Milliseconds;
+	typedef LLUnit<LLUnits::Microseconds, F64>	Microseconds;
+	typedef LLUnit<LLUnits::Nanoseconds, F64>	Nanoseconds;
+
+	typedef LLUnit<LLUnits::Meters, F64>		Meters;
+	typedef LLUnit<LLUnits::Kilometers, F64>	Kilometers;
+	typedef LLUnit<LLUnits::Centimeters, F64>	Centimeters;
+	typedef LLUnit<LLUnits::Millimeters, F64>	Millimeters;
+
+	void init();
+	void cleanup();
+	bool isInitialized();
+
+	const LLThreadLocalPointer<class ThreadRecorder>& get_thread_recorder();
+	void set_thread_recorder(class ThreadRecorder*);
+
+	class MasterThreadRecorder& getMasterThreadRecorder();
+
+	// one per thread per type
+	template<typename ACCUMULATOR>
+	class AccumulatorBuffer : public LLRefCount
+	{
+		typedef AccumulatorBuffer<ACCUMULATOR> self_t;
+		static const U32 DEFAULT_ACCUMULATOR_BUFFER_SIZE = 64;
+	private:
+		struct StaticAllocationMarker { };
+
+		AccumulatorBuffer(StaticAllocationMarker m)
+		:	mStorageSize(0),
+			mStorage(NULL),
+			mNextStorageSlot(0)
+		{
+		}
+
+	public:
+
+		AccumulatorBuffer(const AccumulatorBuffer& other = *getDefaultBuffer())
+		:	mStorageSize(0),
+			mStorage(NULL),
+			mNextStorageSlot(other.mNextStorageSlot)
+		{
+			resize(other.mStorageSize);
+			for (S32 i = 0; i < mNextStorageSlot; i++)
+			{
+				mStorage[i] = other.mStorage[i];
+			}
+		}
+
+		~AccumulatorBuffer()
+		{
+			if (LLThreadLocalSingletonPointer<ACCUMULATOR>::getInstance() == mStorage)
+			{
+				LLThreadLocalSingletonPointer<ACCUMULATOR>::setInstance(getDefaultBuffer()->mStorage);
+			}
+			delete[] mStorage;
+		}
+
+		LL_FORCE_INLINE ACCUMULATOR& operator[](size_t index) 
+		{ 
+			return mStorage[index]; 
+		}
+
+		LL_FORCE_INLINE const ACCUMULATOR& operator[](size_t index) const
+		{ 
+			return mStorage[index]; 
+		}
+
+		void addSamples(const AccumulatorBuffer<ACCUMULATOR>& other)
+		{
+			llassert(mNextStorageSlot == other.mNextStorageSlot);
+
+			for (size_t i = 0; i < mNextStorageSlot; i++)
+			{
+				mStorage[i].addSamples(other.mStorage[i]);
+			}
+		}
+
+		void copyFrom(const AccumulatorBuffer<ACCUMULATOR>& other)
+		{
+			for (size_t i = 0; i < mNextStorageSlot; i++)
+			{
+				mStorage[i] = other.mStorage[i];
+			}
+		}
+
+		void reset(const AccumulatorBuffer<ACCUMULATOR>* other = NULL)
+		{
+			for (size_t i = 0; i < mNextStorageSlot; i++)
+			{
+				mStorage[i].reset(other ? &other->mStorage[i] : NULL);
+			}
+		}
+
+		void makePrimary()
+		{
+			LLThreadLocalSingletonPointer<ACCUMULATOR>::setInstance(mStorage);
+		}
+
+		bool isPrimary() const
+		{
+			return LLThreadLocalSingletonPointer<ACCUMULATOR>::getInstance() == mStorage;
+		}
+
+		LL_FORCE_INLINE static ACCUMULATOR* getPrimaryStorage() 
+		{ 
+			return LLThreadLocalSingletonPointer<ACCUMULATOR>::getInstance(); 
+		}
+
+		// NOTE: this is not thread-safe.  We assume that slots are reserved in the main thread before any child threads are spawned
+		size_t reserveSlot()
+		{
+			if (LLTrace::isInitialized())
+			{
+				llerrs << "Attempting to declare trace object after program initialization.  Trace objects should be statically initialized." << llendl;
+			}
+			size_t next_slot = mNextStorageSlot++;
+			if (next_slot >= mStorageSize)
+			{
+				resize(mStorageSize + (mStorageSize >> 2));
+			}
+			llassert(mStorage && next_slot < mStorageSize);
+			return next_slot;
+		}
+
+		void resize(size_t new_size)
+		{
+			if (new_size <= mStorageSize) return;
+
+			ACCUMULATOR* old_storage = mStorage;
+			mStorage = new ACCUMULATOR[new_size];
+			if (old_storage)
+			{
+				for (S32 i = 0; i < mStorageSize; i++)
+				{
+					mStorage[i] = old_storage[i];
+				}
+			}
+			mStorageSize = new_size;
+			delete[] old_storage;
+
+			self_t* default_buffer = getDefaultBuffer();
+			if (this != default_buffer
+				&& new_size > default_buffer->size())
+			{
+				//NB: this is not thread safe, but we assume that all resizing occurs during static initialization
+				default_buffer->resize(new_size);
+			}
+		}
+
+		size_t size() const
+		{
+			return mNextStorageSlot;
+		}
+
+		static self_t* getDefaultBuffer()
+		{
+			// this buffer is allowed to leak so that trace calls from global destructors have somewhere to put their data
+			// so as not to trigger an access violation
+			static self_t* sBuffer = new AccumulatorBuffer(StaticAllocationMarker());
+			static bool sInitialized = false;
+			if (!sInitialized)
+			{
+				sBuffer->resize(DEFAULT_ACCUMULATOR_BUFFER_SIZE);
+				sInitialized = true;
+			}
+			return sBuffer;
+		}
+
+	private:
+		ACCUMULATOR*								mStorage;
+		size_t										mStorageSize;
+		size_t										mNextStorageSlot;
+	};
+
+	//TODO: replace with decltype when C++11 is enabled
+	template<typename T>
+	struct MeanValueType
+	{
+		typedef F64 type;
+	};
+
+	template<typename ACCUMULATOR>
+	class TraceType 
+	:	 public LLInstanceTracker<TraceType<ACCUMULATOR>, std::string>
+	{
+	public:
+		typedef typename MeanValueType<TraceType<ACCUMULATOR> >::type  mean_t;
+
+		TraceType(const char* name, const char* description = NULL)
+		:	LLInstanceTracker<TraceType<ACCUMULATOR>, std::string>(name),
+			mName(name),
+			mDescription(description ? description : ""),
+			mAccumulatorIndex(AccumulatorBuffer<ACCUMULATOR>::getDefaultBuffer()->reserveSlot())
+		{}
+
+		LL_FORCE_INLINE ACCUMULATOR* getPrimaryAccumulator() const
+		{
+			ACCUMULATOR* accumulator_storage = AccumulatorBuffer<ACCUMULATOR>::getPrimaryStorage();
+			return &accumulator_storage[mAccumulatorIndex];
+		}
+
+		size_t getIndex() const { return mAccumulatorIndex; }
+
+		const std::string& getName() const { return mName; }
+
+	protected:
+		const std::string	mName;
+		const std::string	mDescription;
+		const size_t		mAccumulatorIndex;
+	};
+
+	template<typename T>
+	class MeasurementAccumulator
+	{
+	public:
+		typedef T value_t;
+		typedef MeasurementAccumulator<T> self_t;
+
+		MeasurementAccumulator()
+		:	mSum(0),
+			mMin((std::numeric_limits<T>::max)()),
+			mMax((std::numeric_limits<T>::min)()),
+			mMean(0),
+			mVarianceSum(0),
+			mNumSamples(0),
+			mLastValue(0)
+		{}
+
+		LL_FORCE_INLINE void sample(T value)
+		{
+			T storage_value(value);
+			mNumSamples++;
+			mSum += storage_value;
+			if (storage_value < mMin)
+			{
+				mMin = storage_value;
+			}
+			if (storage_value > mMax)
+			{
+				mMax = storage_value;
+			}
+			F64 old_mean = mMean;
+			mMean += ((F64)storage_value - old_mean) / (F64)mNumSamples;
+			mVarianceSum += ((F64)storage_value - old_mean) * ((F64)storage_value - mMean);
+			mLastValue = storage_value;
+		}
+
+		void addSamples(const self_t& other)
+		{
+			if (other.mNumSamples)
+			{
+				mSum += other.mSum;
+				if (other.mMin < mMin)
+				{
+					mMin = other.mMin;
+				}
+				if (other.mMax > mMax)
+				{
+					mMax = other.mMax;
+				}
+				F64 weight = (F64)mNumSamples / (F64)(mNumSamples + other.mNumSamples);
+				mNumSamples += other.mNumSamples;
+				mMean = mMean * weight + other.mMean * (1.f - weight);
+
+				// combine variance (and hence standard deviation) of 2 different sized sample groups using
+				// the following formula: http://www.mrc-bsu.cam.ac.uk/cochrane/handbook/chapter_7/7_7_3_8_combining_groups.htm
+				F64 n_1 = (F64)mNumSamples,
+					n_2 = (F64)other.mNumSamples;
+				F64 m_1 = mMean,
+					m_2 = other.mMean;
+				F64 sd_1 = getStandardDeviation(),
+					sd_2 = other.getStandardDeviation();
+				if (n_1 == 0)
+				{
+					mVarianceSum = other.mVarianceSum;
+				}
+				else if (n_2 == 0)
+				{
+					// don't touch variance
+					// mVarianceSum = mVarianceSum;
+				}
+				else
+				{
+					mVarianceSum =  (F64)mNumSamples
+								* ((((n_1 - 1.f) * sd_1 * sd_1)
+									+ ((n_2 - 1.f) * sd_2 * sd_2)
+									+ (((n_1 * n_2) / (n_1 + n_2))
+										* ((m_1 * m_1) + (m_2 * m_2) - (2.f * m_1 * m_2))))
+								/ (n_1 + n_2 - 1.f));
+				}
+				mLastValue = other.mLastValue;
+			}
+		}
+
+		void reset(const self_t* other)
+		{
+			mNumSamples = 0;
+			mSum = 0;
+			mMin = 0;
+			mMax = 0;
+			mMean = 0;
+			mVarianceSum = 0;
+			mLastValue = other ? other->mLastValue : 0;
+		}
+
+		T	getSum() const { return (T)mSum; }
+		T	getMin() const { return (T)mMin; }
+		T	getMax() const { return (T)mMax; }
+		T	getLastValue() const { return (T)mLastValue; }
+		F64	getMean() const { return mMean; }
+		F64 getStandardDeviation() const { return sqrtf(mVarianceSum / mNumSamples); }
+		U32 getSampleCount() const { return mNumSamples; }
+
+	private:
+		T	mSum,
+			mMin,
+			mMax,
+			mLastValue;
+
+		F64	mMean,
+			mVarianceSum;
+
+		U32	mNumSamples;
+	};
+
+	template<typename T>
+	class CountAccumulator
+	{
+	public:
+		typedef CountAccumulator<T> self_t;
+		typedef T value_t;
+
+		CountAccumulator()
+		:	mSum(0),
+			mNumSamples(0)
+		{}
+
+		LL_FORCE_INLINE void add(T value)
+		{
+			mNumSamples++;
+			mSum += value;
+		}
+
+		void addSamples(const CountAccumulator<T>& other)
+		{
+			mSum += other.mSum;
+			mNumSamples += other.mNumSamples;
+		}
+
+		void reset(const self_t* other)
+		{
+			mNumSamples = 0;
+			mSum = 0;
+		}
+
+		T	getSum() const { return (T)mSum; }
+
+		U32 getSampleCount() const { return mNumSamples; }
+
+	private:
+		T	mSum;
+
+		U32	mNumSamples;
+	};
+
+	class TimeBlockAccumulator
+	{
+	public:
+		typedef LLUnit<LLUnits::Seconds, F64> value_t;
+		typedef TimeBlockAccumulator self_t;
+
+		// fake class that allows us to view call count aspect of timeblock accumulator
+		struct CallCountAspect 
+		{
+			typedef U32 value_t;
+		};
+
+		struct SelfTimeAspect
+		{
+			typedef LLUnit<LLUnits::Seconds, F64> value_t;
+		};
+
+		TimeBlockAccumulator();
+		void addSamples(const self_t& other);
+		void reset(const self_t* other);
+
+		//
+		// members
+		//
+		U64							mSelfTimeCounter,
+									mTotalTimeCounter;
+		U32							mCalls;
+		class TimeBlock*			mParent;		// last acknowledged parent of this time block
+		class TimeBlock*			mLastCaller;	// used to bootstrap tree construction
+		U16							mActiveCount;	// number of timers with this ID active on stack
+		bool						mMoveUpTree;	// needs to be moved up the tree of timers at the end of frame
+
+	};
+
+
+	template<>
+	struct MeanValueType<TraceType<TimeBlockAccumulator> >
+	{
+		typedef LLUnit<LLUnits::Seconds, F64> type;
+	};
+
+	template<>
+	class TraceType<TimeBlockAccumulator::CallCountAspect>
+	:	public TraceType<TimeBlockAccumulator>
+	{
+	public:
+		typedef F32 mean_t;
+
+		TraceType(const char* name, const char* description = "")
+		:	TraceType<TimeBlockAccumulator>(name, description)
+		{}
+	};
+
+	template<>
+	class TraceType<TimeBlockAccumulator::SelfTimeAspect>
+		:	public TraceType<TimeBlockAccumulator>
+	{
+	public:
+		typedef F32 mean_t;
+
+		TraceType(const char* name, const char* description = "")
+			:	TraceType<TimeBlockAccumulator>(name, description)
+		{}
+	};
+
+	class TimeBlock;
+	class TimeBlockTreeNode
+	{
+	public:
+		TimeBlockTreeNode();
+
+		void setParent(TimeBlock* parent);
+		TimeBlock* getParent() { return mParent; }
+
+		TimeBlock*					mBlock;
+		TimeBlock*					mParent;	
+		std::vector<TimeBlock*>		mChildren;
+		bool						mNeedsSorting;
+	};
+
+
+	template <typename T = F64>
+	class Measurement
+	:	public TraceType<MeasurementAccumulator<typename LLUnits::HighestPrecisionType<T>::type_t> >
+	{
+	public:
+		typedef typename LLUnits::HighestPrecisionType<T>::type_t storage_t;
+		typedef TraceType<MeasurementAccumulator<typename LLUnits::HighestPrecisionType<T>::type_t> > trace_t;
+
+		Measurement(const char* name, const char* description = NULL) 
+		:	trace_t(name, description)
+		{}
+
+		template<typename UNIT_T>
+		void sample(UNIT_T value)
+		{
+			T converted_value(value);
+			trace_t::getPrimaryAccumulator()->sample(LLUnits::rawValue(converted_value));
+		}
+	};
+
+	template <typename T = F64>
+	class Count 
+	:	public TraceType<CountAccumulator<typename LLUnits::HighestPrecisionType<T>::type_t> >
+	{
+	public:
+		typedef typename LLUnits::HighestPrecisionType<T>::type_t storage_t;
+		typedef TraceType<CountAccumulator<typename LLUnits::HighestPrecisionType<T>::type_t> > trace_t;
+
+		Count(const char* name, const char* description = NULL) 
+		:	trace_t(name)
+		{}
+
+		template<typename UNIT_T>
+		void add(UNIT_T value)
+		{
+			T converted_value(value);
+			trace_t::getPrimaryAccumulator()->add(LLUnits::rawValue(converted_value));
+		}
+	};
+
+struct MemStatAccumulator
+{
+	MemStatAccumulator()
+	:	mSize(0),
+		mChildSize(0),
+		mAllocatedCount(0),
+		mDeallocatedCount(0)
+	{}
+
+	void addSamples(const MemStatAccumulator& other)
+	{
+		mSize += other.mSize;
+		mChildSize += other.mChildSize;
+		mAllocatedCount += other.mAllocatedCount;
+		mDeallocatedCount += other.mDeallocatedCount;
+	}
+
+	void reset(const MemStatAccumulator* other)
+	{
+		mSize = 0;
+		mChildSize = 0;
+		mAllocatedCount = 0;
+		mDeallocatedCount = 0;
+	}
+
+	size_t		mSize,
+				mChildSize;
+	int			mAllocatedCount,
+				mDeallocatedCount;
+};
+
+class MemStat : public TraceType<MemStatAccumulator>
+{
+public:
+	typedef TraceType<MemStatAccumulator> trace_t;
+	MemStat(const char* name)
+	:	trace_t(name)
+	{}
+};
+
+// measures effective memory footprint of specified type
+// specialize to cover different types
+
+template<typename T>
+struct MemFootprint
+{
+	static size_t measure(const T& value)
+	{
+		return sizeof(T);
+	}
+
+	static size_t measure()
+	{
+		return sizeof(T);
+	}
+};
+
+template<typename T>
+struct MemFootprint<T*>
+{
+	static size_t measure(const T* value)
+	{
+		if (!value)
+		{
+			return 0;
+		}
+		return MemFootprint<T>::measure(*value);
+	}
+
+	static size_t measure()
+	{
+		return MemFootprint<T>::measure();
+	}
+};
+
+template<typename T>
+struct MemFootprint<std::basic_string<T> >
+{
+	static size_t measure(const std::basic_string<T>& value)
+	{
+		return value.capacity() * sizeof(T);
+	}
+
+	static size_t measure()
+	{
+		return sizeof(std::basic_string<T>);
+	}
+};
+
+template<typename T>
+struct MemFootprint<std::vector<T> >
+{
+	static size_t measure(const std::vector<T>& value)
+	{
+		return value.capacity() * MemFootprint<T>::measure();
+	}
+
+	static size_t measure()
+	{
+		return sizeof(std::vector<T>);
+	}
+};
+
+template<typename T>
+struct MemFootprint<std::list<T> >
+{
+	static size_t measure(const std::list<T>& value)
+	{
+		return value.size() * (MemFootprint<T>::measure() + sizeof(void*) * 2);
+	}
+
+	static size_t measure()
+	{
+		return sizeof(std::list<T>);
+	}
+};
+
+template<typename DERIVED>
+class MemTrackable
+{
+	template<typename TRACKED, typename TRACKED_IS_TRACKER>
+	struct TrackMemImpl;
+
+	typedef MemTrackable<DERIVED> mem_trackable_t;
+
+public:
+	typedef void mem_trackable_tag_t;
+
+	~MemTrackable()
+	{
+		memDisclaim(mMemFootprint);
+	}
+
+	void* operator new(size_t allocation_size) 
+	{
+		// reserve 8 bytes for allocation size (and preserving 8 byte alignment of structs)
+		void* allocation = ::operator new(allocation_size + 8);
+		*(size_t*)allocation = allocation_size;
+		MemStatAccumulator* accumulator = DERIVED::sMemStat.getPrimaryAccumulator();
+		if (accumulator)
+		{
+			accumulator->mSize += allocation_size;
+			accumulator->mAllocatedCount++;
+		}
+		return (void*)((char*)allocation + 8);
+	}
+
+	void operator delete(void* ptr)
+	{
+		size_t* allocation_size = (size_t*)((char*)ptr - 8);
+		MemStatAccumulator* accumulator = DERIVED::sMemStat.getPrimaryAccumulator();
+		if (accumulator)
+		{
+			accumulator->mSize -= *allocation_size;
+			accumulator->mAllocatedCount--;
+			accumulator->mDeallocatedCount++;
+		}
+		::delete((char*)ptr - 8);
+	}
+
+	void *operator new [](size_t size)
+	{
+		size_t* result = (size_t*)malloc(size + 8);
+		*result = size;
+		MemStatAccumulator* accumulator = DERIVED::sMemStat.getPrimaryAccumulator();
+		if (accumulator)
+		{
+			accumulator->mSize += size;
+			accumulator->mAllocatedCount++;
+		}
+		return (void*)((char*)result + 8);
+	}
+
+	void operator delete[](void* ptr)
+	{
+		size_t* allocation_size = (size_t*)((char*)ptr - 8);
+		MemStatAccumulator* accumulator = DERIVED::sMemStat.getPrimaryAccumulator();
+		if (accumulator)
+		{
+			accumulator->mSize -= *allocation_size;
+			accumulator->mAllocatedCount--;
+			accumulator->mDeallocatedCount++;
+		}
+		::delete[]((char*)ptr - 8);
+	}
+
+	// claim memory associated with other objects/data as our own, adding to our calculated footprint
+	template<typename CLAIM_T>
+	CLAIM_T& memClaim(CLAIM_T& value)
+	{
+		TrackMemImpl<CLAIM_T>::claim(*this, value);
+		return value;
+	}
+
+	template<typename CLAIM_T>
+	const CLAIM_T& memClaim(const CLAIM_T& value)
+	{
+		TrackMemImpl<CLAIM_T>::claim(*this, value);
+		return value;
+	}
+
+
+	void memClaim(size_t size)
+	{
+		MemStatAccumulator* accumulator = DERIVED::sMemStat.getPrimaryAccumulator();
+		mMemFootprint += size;
+		if (accumulator)
+		{
+			accumulator->mSize += size;
+		}
+	}
+
+	// remove memory we had claimed from our calculated footprint
+	template<typename CLAIM_T>
+	CLAIM_T& memDisclaim(CLAIM_T& value)
+	{
+		TrackMemImpl<CLAIM_T>::disclaim(*this, value);
+		return value;
+	}
+
+	template<typename CLAIM_T>
+	const CLAIM_T& memDisclaim(const CLAIM_T& value)
+	{
+		TrackMemImpl<CLAIM_T>::disclaim(*this, value);
+		return value;
+	}
+
+	void memDisclaim(size_t size)
+	{
+		MemStatAccumulator* accumulator = DERIVED::sMemStat.getPrimaryAccumulator();
+		if (accumulator)
+		{
+			accumulator->mSize -= size;
+		}
+		mMemFootprint -= size;
+	}
+
+private:
+	size_t mMemFootprint;
+
+	template<typename TRACKED, typename TRACKED_IS_TRACKER = void>
+	struct TrackMemImpl
+	{
+		static void claim(mem_trackable_t& tracker, const TRACKED& tracked)
+		{
+			MemStatAccumulator* accumulator = DERIVED::sMemStat.getPrimaryAccumulator();
+			if (accumulator)
+			{
+				size_t footprint = MemFootprint<TRACKED>::measure(tracked);
+				accumulator->mSize += footprint;
+				tracker.mMemFootprint += footprint;
+			}
+		}
+
+		static void disclaim(mem_trackable_t& tracker, const TRACKED& tracked)
+		{
+			MemStatAccumulator* accumulator = DERIVED::sMemStat.getPrimaryAccumulator();
+			if (accumulator)
+			{
+				size_t footprint = MemFootprint<TRACKED>::measure(tracked);
+				accumulator->mSize -= footprint;
+				tracker.mMemFootprint -= footprint;
+			}
+		}
+	};
+
+	template<typename TRACKED>
+	struct TrackMemImpl<TRACKED, typename TRACKED::mem_trackable_tag_t>
+	{
+		static void claim(mem_trackable_t& tracker, TRACKED& tracked)
+		{
+			MemStatAccumulator* accumulator = DERIVED::sMemStat.getPrimaryAccumulator();
+			if (accumulator)
+			{
+				accumulator->mChildSize += MemFootprint<TRACKED>::measure(tracked);
+			}
+		}
+
+		static void disclaim(mem_trackable_t& tracker, TRACKED& tracked)
+		{
+			MemStatAccumulator* accumulator = DERIVED::sMemStat.getPrimaryAccumulator();
+			if (accumulator)
+			{
+				accumulator->mChildSize -= MemFootprint<TRACKED>::measure(tracked);
+			}
+		}
+	};
+};
+
+}
+#endif // LL_LLTRACE_H
diff --git a/indra/llcommon/lltracerecording.cpp b/indra/llcommon/lltracerecording.cpp
new file mode 100644
index 0000000000..c110e64380
--- /dev/null
+++ b/indra/llcommon/lltracerecording.cpp
@@ -0,0 +1,636 @@
+/** 
+ * @file lltracesampler.cpp
+ *
+ * $LicenseInfo:firstyear=2001&license=viewerlgpl$
+ * Second Life Viewer Source Code
+ * Copyright (C) 2012, Linden Research, Inc.
+ * 
+ * This library is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation;
+ * version 2.1 of the License only.
+ * 
+ * This library is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ * 
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with this library; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA
+ * 
+ * Linden Research, Inc., 945 Battery Street, San Francisco, CA  94111  USA
+ * $/LicenseInfo$
+ */
+
+#include "linden_common.h"
+
+#include "lltrace.h"
+#include "llfasttimer.h"
+#include "lltracerecording.h"
+#include "lltracethreadrecorder.h"
+#include "llthread.h"
+
+namespace LLTrace
+{
+
+///////////////////////////////////////////////////////////////////////
+// Recording
+///////////////////////////////////////////////////////////////////////
+
+Recording::Recording() 
+:	mElapsedSeconds(0),
+	mCountsFloat(new AccumulatorBuffer<CountAccumulator<F64> >()),
+	mMeasurementsFloat(new AccumulatorBuffer<MeasurementAccumulator<F64> >()),
+	mCounts(new AccumulatorBuffer<CountAccumulator<S64> >()),
+	mMeasurements(new AccumulatorBuffer<MeasurementAccumulator<S64> >()),
+	mStackTimers(new AccumulatorBuffer<TimeBlockAccumulator>()),
+	mMemStats(new AccumulatorBuffer<MemStatAccumulator>())
+{}
+
+Recording::Recording( const Recording& other )
+{
+	llassert(other.mCountsFloat.notNull());
+	mSamplingTimer     = other.mSamplingTimer;
+	mElapsedSeconds    = other.mElapsedSeconds;
+	mCountsFloat       = other.mCountsFloat;
+	mMeasurementsFloat = other.mMeasurementsFloat;
+	mCounts            = other.mCounts;
+	mMeasurements      = other.mMeasurements;
+	mStackTimers       = other.mStackTimers;
+	mMemStats		   = other.mMemStats;
+
+	LLStopWatchControlsMixin<Recording>::initTo(other.getPlayState());
+}
+
+
+Recording::~Recording()
+{
+	stop();
+	llassert(isStopped());
+}
+
+void Recording::update()
+{
+	if (isStarted())
+	{
+		LLTrace::get_thread_recorder()->update(this);
+		mSamplingTimer.reset();
+	}
+}
+
+void Recording::handleReset()
+{
+	mCountsFloat.write()->reset();
+	mMeasurementsFloat.write()->reset();
+	mCounts.write()->reset();
+	mMeasurements.write()->reset();
+	mStackTimers.write()->reset();
+	mMemStats.write()->reset();
+
+	mElapsedSeconds = 0.0;
+	mSamplingTimer.reset();
+}
+
+void Recording::handleStart()
+{
+	mSamplingTimer.reset();
+	LLTrace::get_thread_recorder()->activate(this);
+}
+
+void Recording::handleStop()
+{
+	mElapsedSeconds += mSamplingTimer.getElapsedTimeF64();
+	LLTrace::TimeBlock::processTimes();
+	LLTrace::get_thread_recorder()->deactivate(this);
+}
+
+void Recording::handleSplitTo(Recording& other)
+{
+	stop();
+	other.restart();
+	other.mMeasurementsFloat.write()->reset(mMeasurementsFloat);
+	other.mMeasurements.write()->reset(mMeasurements);
+	//TODO: figure out how to get seamless handoff of timing stats
+}
+
+void Recording::makePrimary()
+{
+	mCountsFloat.write()->makePrimary();
+	mMeasurementsFloat.write()->makePrimary();
+	mCounts.write()->makePrimary();
+	mMeasurements.write()->makePrimary();
+	mStackTimers.write()->makePrimary();
+	mMemStats.write()->makePrimary();
+
+	ThreadRecorder* thread_recorder = get_thread_recorder().get();
+	AccumulatorBuffer<TimeBlockAccumulator>& timer_accumulator_buffer = *mStackTimers.write();
+	// update stacktimer parent pointers
+	for (S32 i = 0, end_i = mStackTimers->size(); i < end_i; i++)
+	{
+		TimeBlockTreeNode* tree_node = thread_recorder->getTimeBlockTreeNode(i);
+		if (tree_node)
+		{
+			timer_accumulator_buffer[i].mParent = tree_node->mParent;
+		}
+	}
+}
+
+bool Recording::isPrimary() const
+{
+	return mCounts->isPrimary();
+}
+
+void Recording::makeUnique()
+{
+	mCountsFloat.makeUnique();
+	mMeasurementsFloat.makeUnique();
+	mCounts.makeUnique();
+	mMeasurements.makeUnique();
+	mStackTimers.makeUnique();
+	mMemStats.makeUnique();
+}
+
+void Recording::appendRecording( const Recording& other )
+{
+	mCountsFloat.write()->addSamples(*other.mCountsFloat);
+	mMeasurementsFloat.write()->addSamples(*other.mMeasurementsFloat);
+	mCounts.write()->addSamples(*other.mCounts);
+	mMeasurements.write()->addSamples(*other.mMeasurements);
+	mMemStats.write()->addSamples(*other.mMemStats);
+
+	mStackTimers.write()->addSamples(*other.mStackTimers);
+	mElapsedSeconds += other.mElapsedSeconds;
+}
+
+void Recording::mergeRecording( const Recording& other)
+{
+	mCountsFloat.write()->addSamples(*other.mCountsFloat);
+	mMeasurementsFloat.write()->addSamples(*other.mMeasurementsFloat);
+	mCounts.write()->addSamples(*other.mCounts);
+	mMeasurements.write()->addSamples(*other.mMeasurements);
+	mMemStats.write()->addSamples(*other.mMemStats);
+}
+
+LLUnit<LLUnits::Seconds, F64> Recording::getSum(const TraceType<TimeBlockAccumulator>& stat) const
+{
+	return (F64)(*mStackTimers)[stat.getIndex()].mTotalTimeCounter / (F64)LLTrace::TimeBlock::countsPerSecond();
+}
+
+LLUnit<LLUnits::Seconds, F64> Recording::getSum(const TraceType<TimeBlockAccumulator::SelfTimeAspect>& stat) const
+{
+	return (F64)(*mStackTimers)[stat.getIndex()].mSelfTimeCounter / (F64)LLTrace::TimeBlock::countsPerSecond();
+}
+
+
+U32 Recording::getSum(const TraceType<TimeBlockAccumulator::CallCountAspect>& stat) const
+{
+	return (*mStackTimers)[stat.getIndex()].mCalls;
+}
+
+LLUnit<LLUnits::Seconds, F64> Recording::getPerSec(const TraceType<TimeBlockAccumulator>& stat) const
+{
+	return (F64)(*mStackTimers)[stat.getIndex()].mTotalTimeCounter / ((F64)LLTrace::TimeBlock::countsPerSecond() * mElapsedSeconds);
+}
+
+LLUnit<LLUnits::Seconds, F64> Recording::getPerSec(const TraceType<TimeBlockAccumulator::SelfTimeAspect>& stat) const
+{
+	return (F64)(*mStackTimers)[stat.getIndex()].mSelfTimeCounter / ((F64)LLTrace::TimeBlock::countsPerSecond() * mElapsedSeconds);
+}
+
+F32 Recording::getPerSec(const TraceType<TimeBlockAccumulator::CallCountAspect>& stat) const
+{
+	return (F32)(*mStackTimers)[stat.getIndex()].mCalls / mElapsedSeconds;
+}
+
+LLUnit<LLUnits::Bytes, U32> Recording::getSum(const TraceType<MemStatAccumulator>& stat) const
+{
+	return (*mMemStats)[stat.getIndex()].mAllocatedCount;
+}
+
+LLUnit<LLUnits::Bytes, F32> Recording::getPerSec(const TraceType<MemStatAccumulator>& stat) const
+{
+	return (F32)(*mMemStats)[stat.getIndex()].mAllocatedCount / mElapsedSeconds;
+}
+
+
+F64 Recording::getSum( const TraceType<CountAccumulator<F64> >& stat ) const
+{
+	return (*mCountsFloat)[stat.getIndex()].getSum();
+}
+
+S64 Recording::getSum( const TraceType<CountAccumulator<S64> >& stat ) const
+{
+	return (*mCounts)[stat.getIndex()].getSum();
+}
+
+F64 Recording::getSum( const TraceType<MeasurementAccumulator<F64> >& stat ) const
+{
+	return (F64)(*mMeasurementsFloat)[stat.getIndex()].getSum();
+}
+
+S64 Recording::getSum( const TraceType<MeasurementAccumulator<S64> >& stat ) const
+{
+	return (S64)(*mMeasurements)[stat.getIndex()].getSum();
+}
+
+
+
+F64 Recording::getPerSec( const TraceType<CountAccumulator<F64> >& stat ) const
+{
+	F64 sum = (*mCountsFloat)[stat.getIndex()].getSum();
+	return  (sum != 0.0) 
+		? (sum / mElapsedSeconds)
+		: 0.0;
+}
+
+F64 Recording::getPerSec( const TraceType<CountAccumulator<S64> >& stat ) const
+{
+	S64 sum = (*mCounts)[stat.getIndex()].getSum();
+	return (sum != 0) 
+		? ((F64)sum / mElapsedSeconds)
+		: 0.0;
+}
+
+U32 Recording::getSampleCount( const TraceType<CountAccumulator<F64> >& stat ) const
+{
+	return (*mCountsFloat)[stat.getIndex()].getSampleCount();
+}
+
+U32 Recording::getSampleCount( const TraceType<CountAccumulator<S64> >& stat ) const
+{
+	return (*mMeasurementsFloat)[stat.getIndex()].getSampleCount();
+}
+
+
+F64 Recording::getPerSec( const TraceType<MeasurementAccumulator<F64> >& stat ) const
+{
+	F64 sum = (*mMeasurementsFloat)[stat.getIndex()].getSum();
+	return  (sum != 0.0) 
+		? (sum / mElapsedSeconds)
+		: 0.0;
+}
+
+F64 Recording::getPerSec( const TraceType<MeasurementAccumulator<S64> >& stat ) const
+{
+	S64 sum = (*mMeasurements)[stat.getIndex()].getSum();
+	return (sum != 0) 
+		? ((F64)sum / mElapsedSeconds)
+		: 0.0;
+}
+
+F64 Recording::getMin( const TraceType<MeasurementAccumulator<F64> >& stat ) const
+{
+	return (*mMeasurementsFloat)[stat.getIndex()].getMin();
+}
+
+S64 Recording::getMin( const TraceType<MeasurementAccumulator<S64> >& stat ) const
+{
+	return (*mMeasurements)[stat.getIndex()].getMin();
+}
+
+F64 Recording::getMax( const TraceType<MeasurementAccumulator<F64> >& stat ) const
+{
+	return (*mMeasurementsFloat)[stat.getIndex()].getMax();
+}
+
+S64 Recording::getMax( const TraceType<MeasurementAccumulator<S64> >& stat ) const
+{
+	return (*mMeasurements)[stat.getIndex()].getMax();
+}
+
+F64 Recording::getMean( const TraceType<MeasurementAccumulator<F64> >& stat ) const
+{
+	return (*mMeasurementsFloat)[stat.getIndex()].getMean();
+}
+
+F64 Recording::getMean( const TraceType<MeasurementAccumulator<S64> >& stat ) const
+{
+	return (*mMeasurements)[stat.getIndex()].getMean();
+}
+
+F64 Recording::getStandardDeviation( const TraceType<MeasurementAccumulator<F64> >& stat ) const
+{
+	return (*mMeasurementsFloat)[stat.getIndex()].getStandardDeviation();
+}
+
+F64 Recording::getStandardDeviation( const TraceType<MeasurementAccumulator<S64> >& stat ) const
+{
+	return (*mMeasurements)[stat.getIndex()].getStandardDeviation();
+}
+
+F64 Recording::getLastValue( const TraceType<MeasurementAccumulator<F64> >& stat ) const
+{
+	return (*mMeasurementsFloat)[stat.getIndex()].getLastValue();
+}
+
+S64 Recording::getLastValue( const TraceType<MeasurementAccumulator<S64> >& stat ) const
+{
+	return (*mMeasurements)[stat.getIndex()].getLastValue();
+}
+
+U32 Recording::getSampleCount( const TraceType<MeasurementAccumulator<F64> >& stat ) const
+{
+	return (*mMeasurementsFloat)[stat.getIndex()].getSampleCount();
+}
+
+U32 Recording::getSampleCount( const TraceType<MeasurementAccumulator<S64> >& stat ) const
+{
+	return (*mMeasurements)[stat.getIndex()].getSampleCount();
+}
+
+
+
+///////////////////////////////////////////////////////////////////////
+// PeriodicRecording
+///////////////////////////////////////////////////////////////////////
+
+PeriodicRecording::PeriodicRecording( S32 num_periods, EStopWatchState state) 
+:	mNumPeriods(num_periods),
+	mCurPeriod(0),
+	mTotalValid(false),
+	mRecordingPeriods( new Recording[num_periods])
+{
+	llassert(mNumPeriods > 0);
+	initTo(state);
+}
+
+PeriodicRecording::PeriodicRecording(PeriodicRecording& other)
+:	mNumPeriods(other.mNumPeriods),
+	mCurPeriod(other.mCurPeriod),
+	mTotalValid(other.mTotalValid),
+	mTotalRecording(other.mTotalRecording)
+{
+	mRecordingPeriods = new Recording[mNumPeriods];
+	for (S32 i = 0; i < mNumPeriods; i++)
+	{
+		mRecordingPeriods[i] = other.mRecordingPeriods[i];
+	}
+}
+
+
+PeriodicRecording::~PeriodicRecording()
+{
+	delete[] mRecordingPeriods;
+}
+
+
+void PeriodicRecording::nextPeriod()
+{
+	EStopWatchState play_state = getPlayState();
+	Recording& old_recording = getCurRecordingPeriod();
+	mCurPeriod = (mCurPeriod + 1) % mNumPeriods;
+	old_recording.splitTo(getCurRecordingPeriod());
+
+	switch(play_state)
+	{
+	case STOPPED:
+		getCurRecordingPeriod().stop();
+		break;
+	case PAUSED:
+		getCurRecordingPeriod().pause();
+		break;
+	case STARTED:
+		break;
+	}
+	// new period, need to recalculate total
+	mTotalValid = false;
+}
+
+Recording& PeriodicRecording::getTotalRecording()
+{
+	if (!mTotalValid)
+	{
+		mTotalRecording.reset();
+		for (S32 i = mCurPeriod + 1; i < mCurPeriod + mNumPeriods; i++)
+		{
+			mTotalRecording.appendRecording(mRecordingPeriods[i % mNumPeriods]);
+		}
+	}
+	mTotalValid = true;
+	return mTotalRecording;
+}
+
+void PeriodicRecording::start()
+{
+	getCurRecordingPeriod().start();
+}
+
+void PeriodicRecording::stop()
+{
+	getCurRecordingPeriod().stop();
+}
+
+void PeriodicRecording::pause()
+{
+	getCurRecordingPeriod().pause();
+}
+
+void PeriodicRecording::resume()
+{
+	getCurRecordingPeriod().resume();
+}
+
+void PeriodicRecording::restart()
+{
+	getCurRecordingPeriod().restart();
+}
+
+void PeriodicRecording::reset()
+{
+	getCurRecordingPeriod().reset();
+}
+
+void PeriodicRecording::splitTo(PeriodicRecording& other)
+{
+	getCurRecordingPeriod().splitTo(other.getCurRecordingPeriod());
+}
+
+void PeriodicRecording::splitFrom(PeriodicRecording& other)
+{
+	getCurRecordingPeriod().splitFrom(other.getCurRecordingPeriod());
+}
+
+
+///////////////////////////////////////////////////////////////////////
+// ExtendableRecording
+///////////////////////////////////////////////////////////////////////
+
+void ExtendableRecording::extend()
+{
+	mAcceptedRecording.appendRecording(mPotentialRecording);
+	mPotentialRecording.reset();
+}
+
+void ExtendableRecording::start()
+{
+	mPotentialRecording.start();
+}
+
+void ExtendableRecording::stop()
+{
+	mPotentialRecording.stop();
+}
+
+void ExtendableRecording::pause()
+{
+	mPotentialRecording.pause();
+}
+
+void ExtendableRecording::resume()
+{
+	mPotentialRecording.resume();
+}
+
+void ExtendableRecording::restart()
+{
+	mAcceptedRecording.reset();
+	mPotentialRecording.restart();
+}
+
+void ExtendableRecording::reset()
+{
+	mAcceptedRecording.reset();
+	mPotentialRecording.reset();
+}
+
+void ExtendableRecording::splitTo(ExtendableRecording& other)
+{
+	mPotentialRecording.splitTo(other.mPotentialRecording);
+}
+
+void ExtendableRecording::splitFrom(ExtendableRecording& other)
+{
+	mPotentialRecording.splitFrom(other.mPotentialRecording);
+}
+
+PeriodicRecording& get_frame_recording()
+{
+	static LLThreadLocalPointer<PeriodicRecording> sRecording(new PeriodicRecording(64, PeriodicRecording::STARTED));
+	return *sRecording;
+}
+
+}
+
+void LLStopWatchControlsMixinCommon::start()
+{
+	switch (mState)
+	{
+	case STOPPED:
+		handleReset();
+		handleStart();
+		break;
+	case PAUSED:
+		handleStart();
+		break;
+	case STARTED:
+		handleReset();
+		break;
+	default:
+		llassert(false);
+		break;
+	}
+	mState = STARTED;
+}
+
+void LLStopWatchControlsMixinCommon::stop()
+{
+	switch (mState)
+	{
+	case STOPPED:
+		break;
+	case PAUSED:
+		handleStop();
+		break;
+	case STARTED:
+		handleStop();
+		break;
+	default:
+		llassert(false);
+		break;
+	}
+	mState = STOPPED;
+}
+
+void LLStopWatchControlsMixinCommon::pause()
+{
+	switch (mState)
+	{
+	case STOPPED:
+		break;
+	case PAUSED:
+		break;
+	case STARTED:
+		handleStop();
+		break;
+	default:
+		llassert(false);
+		break;
+	}
+	mState = PAUSED;
+}
+
+void LLStopWatchControlsMixinCommon::resume()
+{
+	switch (mState)
+	{
+	case STOPPED:
+		handleStart();
+		break;
+	case PAUSED:
+		handleStart();
+		break;
+	case STARTED:
+		break;
+	default:
+		llassert(false);
+		break;
+	}
+	mState = STARTED;
+}
+
+void LLStopWatchControlsMixinCommon::restart()
+{
+	switch (mState)
+	{
+	case STOPPED:
+		handleReset();
+		handleStart();
+		break;
+	case PAUSED:
+		handleReset();
+		handleStart();
+		break;
+	case STARTED:
+		handleReset();
+		break;
+	default:
+		llassert(false);
+		break;
+	}
+	mState = STARTED;
+}
+
+void LLStopWatchControlsMixinCommon::reset()
+{
+	handleReset();
+}
+
+void LLStopWatchControlsMixinCommon::initTo( EStopWatchState state )
+{
+	switch(state)
+	{
+	case STOPPED:
+		break;
+	case PAUSED:
+		break;
+	case STARTED:
+		handleStart();
+		break;
+	default:
+		llassert(false);
+		break;
+	}
+
+	mState = state;
+}
diff --git a/indra/llcommon/lltracerecording.h b/indra/llcommon/lltracerecording.h
new file mode 100644
index 0000000000..aa3200e5ad
--- /dev/null
+++ b/indra/llcommon/lltracerecording.h
@@ -0,0 +1,410 @@
+/** 
+ * @file lltracerecording.h
+ * @brief Sampling object for collecting runtime statistics originating from lltrace.
+ *
+ * $LicenseInfo:firstyear=2001&license=viewerlgpl$
+ * Second Life Viewer Source Code
+ * Copyright (C) 2012, Linden Research, Inc.
+ * 
+ * This library is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation;
+ * version 2.1 of the License only.
+ * 
+ * This library is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ * 
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with this library; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA
+ * 
+ * Linden Research, Inc., 945 Battery Street, San Francisco, CA  94111  USA
+ * $/LicenseInfo$
+ */
+
+#ifndef LL_LLTRACERECORDING_H
+#define LL_LLTRACERECORDING_H
+
+#include "stdtypes.h"
+#include "llpreprocessor.h"
+
+#include "llpointer.h"
+#include "lltimer.h"
+#include "lltrace.h"
+
+class LLStopWatchControlsMixinCommon
+{
+public:
+	virtual ~LLStopWatchControlsMixinCommon() {}
+
+	enum EStopWatchState
+	{
+		STOPPED,
+		PAUSED,
+		STARTED
+	};
+
+	virtual void start();
+	virtual void stop();
+	virtual void pause();
+	virtual void resume();
+	virtual void restart();
+	virtual void reset();
+
+	bool isStarted() const { return mState == STARTED; }
+	bool isPaused() const  { return mState == PAUSED; }
+	bool isStopped() const { return mState == STOPPED; }
+	EStopWatchState getPlayState() const { return mState; }
+
+protected:
+	LLStopWatchControlsMixinCommon()
+	:	mState(STOPPED)
+	{}
+
+	// derived classes can call this from their copy constructor in order
+	// to duplicate play state of source
+	void initTo(EStopWatchState state);
+private:
+	// trigger active behavior (without reset)
+	virtual void handleStart(){};
+	// stop active behavior
+	virtual void handleStop(){};
+	// clear accumulated state, can be called while started
+	virtual void handleReset(){};
+
+	EStopWatchState mState;
+};
+
+template<typename DERIVED>
+class LLStopWatchControlsMixin
+:	public LLStopWatchControlsMixinCommon
+{
+public:
+	typedef LLStopWatchControlsMixin<DERIVED> self_t;
+	virtual void splitTo(DERIVED& other)
+	{
+		handleSplitTo(other);
+	}
+
+	virtual void splitFrom(DERIVED& other)
+	{
+		static_cast<self_t&>(other).handleSplitTo(*static_cast<DERIVED*>(this));
+	}
+private:
+	// atomically stop this object while starting the other
+	// no data can be missed in between stop and start
+	virtual void handleSplitTo(DERIVED& other) {};
+
+};
+
+namespace LLTrace
+{
+	class Recording : public LLStopWatchControlsMixin<Recording>
+	{
+	public:
+		Recording();
+
+		Recording(const Recording& other);
+		~Recording();
+
+		void makePrimary();
+		bool isPrimary() const;
+
+		void makeUnique();
+
+		// accumulate data from subsequent, non-overlapping recording
+		void appendRecording(const Recording& other);
+
+		// gather data from recording, ignoring time relationship (for example, pulling data from slave threads)
+		void mergeRecording(const Recording& other);
+
+		void update();
+
+		// Timer accessors
+		LLUnit<LLUnits::Seconds, F64> getSum(const TraceType<TimeBlockAccumulator>& stat) const;
+		LLUnit<LLUnits::Seconds, F64> getSum(const TraceType<TimeBlockAccumulator::SelfTimeAspect>& stat) const;
+		U32 getSum(const TraceType<TimeBlockAccumulator::CallCountAspect>& stat) const;
+
+		LLUnit<LLUnits::Seconds, F64> getPerSec(const TraceType<TimeBlockAccumulator>& stat) const;
+		LLUnit<LLUnits::Seconds, F64> getPerSec(const TraceType<TimeBlockAccumulator::SelfTimeAspect>& stat) const;
+		F32 getPerSec(const TraceType<TimeBlockAccumulator::CallCountAspect>& stat) const;
+
+		// Memory accessors
+		LLUnit<LLUnits::Bytes, U32> getSum(const TraceType<MemStatAccumulator>& stat) const;
+		LLUnit<LLUnits::Bytes, F32> getPerSec(const TraceType<MemStatAccumulator>& stat) const;
+
+		// Count accessors
+		F64 getSum(const TraceType<CountAccumulator<F64> >& stat) const;
+		S64 getSum(const TraceType<CountAccumulator<S64> >& stat) const;
+		template <typename T>
+		T getSum(const Count<T>& stat) const
+		{
+			return (T)getSum(static_cast<const TraceType<CountAccumulator<typename LLUnits::HighestPrecisionType<T>::type_t> >&> (stat));
+		}
+
+		F64 getPerSec(const TraceType<CountAccumulator<F64> >& stat) const;
+		F64 getPerSec(const TraceType<CountAccumulator<S64> >& stat) const;
+		template <typename T>
+		T getPerSec(const Count<T>& stat) const
+		{
+			return (T)getPerSec(static_cast<const TraceType<CountAccumulator<typename LLUnits::HighestPrecisionType<T>::type_t> >&> (stat));
+		}
+
+		U32 getSampleCount(const TraceType<CountAccumulator<F64> >& stat) const;
+		U32 getSampleCount(const TraceType<CountAccumulator<S64> >& stat) const;
+
+
+		// Measurement accessors
+		F64 getSum(const TraceType<MeasurementAccumulator<F64> >& stat) const;
+		S64 getSum(const TraceType<MeasurementAccumulator<S64> >& stat) const;
+		template <typename T>
+		T getSum(const Measurement<T>& stat) const
+		{
+			return (T)getSum(static_cast<const TraceType<MeasurementAccumulator<typename LLUnits::HighestPrecisionType<T>::type_t> >&> (stat));
+		}
+
+		F64 getPerSec(const TraceType<MeasurementAccumulator<F64> >& stat) const;
+		F64 getPerSec(const TraceType<MeasurementAccumulator<S64> >& stat) const;
+		template <typename T>
+		T getPerSec(const Measurement<T>& stat) const
+		{
+			return (T)getPerSec(static_cast<const TraceType<MeasurementAccumulator<typename LLUnits::HighestPrecisionType<T>::type_t> >&> (stat));
+		}
+
+		F64 getMin(const TraceType<MeasurementAccumulator<F64> >& stat) const;
+		S64 getMin(const TraceType<MeasurementAccumulator<S64> >& stat) const;
+		template <typename T>
+		T getMin(const Measurement<T>& stat) const
+		{
+			return (T)getMin(static_cast<const TraceType<MeasurementAccumulator<typename LLUnits::HighestPrecisionType<T>::type_t> >&> (stat));
+		}
+
+		F64 getMax(const TraceType<MeasurementAccumulator<F64> >& stat) const;
+		S64 getMax(const TraceType<MeasurementAccumulator<S64> >& stat) const;
+		template <typename T>
+		T getMax(const Measurement<T>& stat) const
+		{
+			return (T)getMax(static_cast<const TraceType<MeasurementAccumulator<typename LLUnits::HighestPrecisionType<T>::type_t> >&> (stat));
+		}
+
+		F64 getMean(const TraceType<MeasurementAccumulator<F64> >& stat) const;
+		F64 getMean(const TraceType<MeasurementAccumulator<S64> >& stat) const;
+		template <typename T>
+		T getMean(Measurement<T>& stat) const
+		{
+			return (T)getMean(static_cast<const TraceType<MeasurementAccumulator<typename LLUnits::HighestPrecisionType<T>::type_t> >&> (stat));
+		}
+
+		F64 getStandardDeviation(const TraceType<MeasurementAccumulator<F64> >& stat) const;
+		F64 getStandardDeviation(const TraceType<MeasurementAccumulator<S64> >& stat) const;
+		template <typename T>
+		T getStandardDeviation(const Measurement<T>& stat) const
+		{
+			return (T)getMean(static_cast<const TraceType<MeasurementAccumulator<typename LLUnits::HighestPrecisionType<T>::type_t> >&> (stat));
+		}
+
+		F64 getLastValue(const TraceType<MeasurementAccumulator<F64> >& stat) const;
+		S64 getLastValue(const TraceType<MeasurementAccumulator<S64> >& stat) const;
+		template <typename T>
+		T getLastValue(const Measurement<T>& stat) const
+		{
+			return (T)getLastValue(static_cast<const TraceType<MeasurementAccumulator<typename LLUnits::HighestPrecisionType<T>::type_t> >&> (stat));
+		}
+
+		U32 getSampleCount(const TraceType<MeasurementAccumulator<F64> >& stat) const;
+		U32 getSampleCount(const TraceType<MeasurementAccumulator<S64> >& stat) const;
+
+		LLUnit<LLUnits::Seconds, F64> getDuration() const { return LLUnit<LLUnits::Seconds, F64>(mElapsedSeconds); }
+
+	private:
+		friend class ThreadRecorder;
+
+		// implementation for LLStopWatchControlsMixin
+		/*virtual*/ void handleStart();
+		/*virtual*/ void handleStop();
+		/*virtual*/ void handleReset();
+		/*virtual*/ void handleSplitTo(Recording& other);
+
+		// returns data for current thread
+		class ThreadRecorder* getThreadRecorder(); 
+
+		LLCopyOnWritePointer<AccumulatorBuffer<CountAccumulator<F64> > >		mCountsFloat;
+		LLCopyOnWritePointer<AccumulatorBuffer<MeasurementAccumulator<F64> > >	mMeasurementsFloat;
+		LLCopyOnWritePointer<AccumulatorBuffer<CountAccumulator<S64> > >		mCounts;
+		LLCopyOnWritePointer<AccumulatorBuffer<MeasurementAccumulator<S64> > >	mMeasurements;
+		LLCopyOnWritePointer<AccumulatorBuffer<TimeBlockAccumulator> >			mStackTimers;
+		LLCopyOnWritePointer<AccumulatorBuffer<MemStatAccumulator> >			mMemStats;
+
+		LLTimer			mSamplingTimer;
+		F64				mElapsedSeconds;
+	};
+
+	class LL_COMMON_API PeriodicRecording
+	:	public LLStopWatchControlsMixin<PeriodicRecording>
+	{
+	public:
+		PeriodicRecording(S32 num_periods, EStopWatchState state = STOPPED);
+		PeriodicRecording(PeriodicRecording& recording);
+		~PeriodicRecording();
+
+		void nextPeriod();
+		S32 getNumPeriods() { return mNumPeriods; }
+
+		Recording& getLastRecordingPeriod()
+		{
+			return mRecordingPeriods[(mCurPeriod + mNumPeriods - 1) % mNumPeriods];
+		}
+
+		const Recording& getLastRecordingPeriod() const
+		{
+			return getPrevRecordingPeriod(1);
+		}
+
+		Recording& getCurRecordingPeriod()
+		{
+			return mRecordingPeriods[mCurPeriod];
+		}
+
+		const Recording& getCurRecordingPeriod() const
+		{
+			return mRecordingPeriods[mCurPeriod];
+		}
+
+		Recording& getPrevRecordingPeriod(S32 offset)
+		{
+			offset = llclamp(offset, 0, mNumPeriods - 1);
+			return mRecordingPeriods[(mCurPeriod + mNumPeriods - offset) % mNumPeriods];
+		}
+
+		const Recording& getPrevRecordingPeriod(S32 offset) const
+		{
+			offset = llclamp(offset, 0, mNumPeriods - 1);
+			return mRecordingPeriods[(mCurPeriod + mNumPeriods - offset) % mNumPeriods];
+		}
+
+		Recording snapshotCurRecordingPeriod() const
+		{
+			Recording recording_copy(getCurRecordingPeriod());
+			recording_copy.stop();
+			return recording_copy;
+		}
+
+		Recording& getTotalRecording();
+
+		template <typename T>
+		typename T::value_t getPeriodMin(const TraceType<T>& stat) const
+		{
+			typename T::value_t min_val = (std::numeric_limits<typename T::value_t>::max)();
+			for (S32 i = 0; i < mNumPeriods; i++)
+			{
+				min_val = llmin(min_val, mRecordingPeriods[i].getSum(stat));
+			}
+			return min_val;
+		}
+
+		template <typename T>
+		F64 getPeriodMinPerSec(const TraceType<T>& stat) const
+		{
+			F64 min_val = (std::numeric_limits<F64>::max)();
+			for (S32 i = 0; i < mNumPeriods; i++)
+			{
+				min_val = llmin(min_val, mRecordingPeriods[i].getPerSec(stat));
+			}
+			return min_val;
+		}
+
+		template <typename T>
+		typename T::value_t getPeriodMax(const TraceType<T>& stat) const
+		{
+			typename T::value_t max_val = (std::numeric_limits<typename T::value_t>::min)();
+			for (S32 i = 0; i < mNumPeriods; i++)
+			{
+				max_val = llmax(max_val, mRecordingPeriods[i].getSum(stat));
+			}
+			return max_val;
+		}
+
+		template <typename T>
+		F64 getPeriodMaxPerSec(const TraceType<T>& stat) const
+		{
+			F64 max_val = (std::numeric_limits<F64>::min)();
+			for (S32 i = 0; i < mNumPeriods; i++)
+			{
+				max_val = llmax(max_val, mRecordingPeriods[i].getPerSec(stat));
+			}
+			return max_val;
+		}
+
+		template <typename T>
+		typename TraceType<T>::mean_t getPeriodMean(const TraceType<T>& stat) const
+		{
+			typename TraceType<T>::mean_t mean = 0.0;
+			for (S32 i = 0; i < mNumPeriods; i++)
+			{
+				if (mRecordingPeriods[i].getDuration() > 0.f)
+				{
+					mean += mRecordingPeriods[i].getSum(stat);
+				}
+			}
+			mean /= mNumPeriods;
+			return mean;
+		}
+
+		template <typename T>
+		typename TraceType<T>::mean_t getPeriodMeanPerSec(const TraceType<T>& stat) const
+		{
+			typename TraceType<T>::mean_t mean = 0.0;
+			for (S32 i = 0; i < mNumPeriods; i++)
+			{
+				if (mRecordingPeriods[i].getDuration() > 0.f)
+				{
+					mean += mRecordingPeriods[i].getPerSec(stat);
+				}
+			}
+			mean /= mNumPeriods;
+			return mean;
+		}
+
+		// implementation for LLStopWatchControlsMixin
+		/*virtual*/ void start();
+		/*virtual*/ void stop();
+		/*virtual*/ void pause();
+		/*virtual*/ void resume();
+		/*virtual*/ void restart();
+		/*virtual*/ void reset();
+		/*virtual*/ void splitTo(PeriodicRecording& other);
+		/*virtual*/ void splitFrom(PeriodicRecording& other);
+
+	private:
+		Recording*	mRecordingPeriods;
+		Recording	mTotalRecording;
+		bool		mTotalValid;
+		S32			mNumPeriods,
+					mCurPeriod;
+	};
+
+	PeriodicRecording& get_frame_recording();
+
+	class ExtendableRecording
+	:	public LLStopWatchControlsMixin<ExtendableRecording>
+	{
+		void extend();
+
+		// implementation for LLStopWatchControlsMixin
+		/*virtual*/ void start();
+		/*virtual*/ void stop();
+		/*virtual*/ void pause();
+		/*virtual*/ void resume();
+		/*virtual*/ void restart();
+		/*virtual*/ void reset();
+		/*virtual*/ void splitTo(ExtendableRecording& other);
+		/*virtual*/ void splitFrom(ExtendableRecording& other);
+	private:
+		Recording mAcceptedRecording;
+		Recording mPotentialRecording;
+	};
+}
+
+#endif // LL_LLTRACERECORDING_H
diff --git a/indra/llcommon/lltracethreadrecorder.cpp b/indra/llcommon/lltracethreadrecorder.cpp
new file mode 100644
index 0000000000..9fb789c62d
--- /dev/null
+++ b/indra/llcommon/lltracethreadrecorder.cpp
@@ -0,0 +1,287 @@
+/** 
+ * @file lltracethreadrecorder.cpp
+ *
+ * $LicenseInfo:firstyear=2001&license=viewerlgpl$
+ * Second Life Viewer Source Code
+ * Copyright (C) 2012, Linden Research, Inc.
+ * 
+ * This library is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation;
+ * version 2.1 of the License only.
+ * 
+ * This library is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ * 
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with this library; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA
+ * 
+ * Linden Research, Inc., 945 Battery Street, San Francisco, CA  94111  USA
+ * $/LicenseInfo$
+ */
+
+#include "linden_common.h"
+
+#include "lltracethreadrecorder.h"
+#include "llfasttimer.h"
+
+namespace LLTrace
+{
+
+
+///////////////////////////////////////////////////////////////////////
+// ThreadRecorder
+///////////////////////////////////////////////////////////////////////
+
+ThreadRecorder::ThreadRecorder()
+{
+	//NB: the ordering of initialization in this function is very fragile due to a large number of implicit dependencies
+	set_thread_recorder(this);
+	TimeBlock& root_time_block = TimeBlock::getRootTimeBlock();
+
+	ThreadTimerStack* timer_stack = ThreadTimerStack::getInstance();
+	timer_stack->mTimeBlock = &root_time_block;
+	timer_stack->mActiveTimer = NULL;
+
+	mNumTimeBlockTreeNodes = AccumulatorBuffer<TimeBlockAccumulator>::getDefaultBuffer()->size();
+	mTimeBlockTreeNodes = new TimeBlockTreeNode[mNumTimeBlockTreeNodes];
+
+	mThreadRecording.start();
+
+	// initialize time block parent pointers
+	for (LLInstanceTracker<TimeBlock>::instance_iter it = LLInstanceTracker<TimeBlock>::beginInstances(), end_it = LLInstanceTracker<TimeBlock>::endInstances(); 
+		it != end_it; 
+		++it)
+	{
+		TimeBlock& time_block = *it;
+		TimeBlockTreeNode& tree_node = mTimeBlockTreeNodes[it->getIndex()];
+		tree_node.mBlock = &time_block;
+		tree_node.mParent = &root_time_block;
+
+		it->getPrimaryAccumulator()->mParent = &root_time_block;
+	}
+
+	mRootTimer = new BlockTimer(root_time_block);
+	timer_stack->mActiveTimer = mRootTimer;
+
+	TimeBlock::getRootTimeBlock().getPrimaryAccumulator()->mActiveCount = 1;
+}
+
+ThreadRecorder::~ThreadRecorder()
+{
+	delete mRootTimer;
+
+	while(mActiveRecordings.size())
+	{
+		mActiveRecordings.front().mTargetRecording->stop();
+	}
+	set_thread_recorder(NULL);
+	delete[] mTimeBlockTreeNodes;
+}
+
+TimeBlockTreeNode* ThreadRecorder::getTimeBlockTreeNode(S32 index)
+{
+	if (0 <= index && index < mNumTimeBlockTreeNodes)
+	{
+		return &mTimeBlockTreeNodes[index];
+	}
+	return NULL;
+}
+
+
+void ThreadRecorder::activate( Recording* recording )
+{
+	mActiveRecordings.push_front(ActiveRecording(recording));
+	mActiveRecordings.front().mBaseline.makePrimary();
+}
+
+std::list<ThreadRecorder::ActiveRecording>::iterator ThreadRecorder::update( Recording* recording )
+{
+	std::list<ActiveRecording>::iterator it, end_it;
+	for (it = mActiveRecordings.begin(), end_it = mActiveRecordings.end();
+		it != end_it;
+		++it)
+	{
+		std::list<ActiveRecording>::iterator next_it = it;
+		++next_it;
+
+		// if we have another recording further down in the stack...
+		if (next_it != mActiveRecordings.end())
+		{
+			// ...push our gathered data down to it
+			next_it->mBaseline.appendRecording(it->mBaseline);
+		}
+
+		// copy accumulated measurements into result buffer and clear accumulator (mBaseline)
+		it->moveBaselineToTarget();
+
+		if (it->mTargetRecording == recording)
+		{
+			// found the recording, so return it
+			break;
+		}
+	}
+
+	if (it == end_it)
+	{
+		llerrs << "Recording not active on this thread" << llendl;
+	}
+
+	return it;
+}
+
+AccumulatorBuffer<CountAccumulator<F64> > 		gCountsFloat;
+AccumulatorBuffer<MeasurementAccumulator<F64> >	gMeasurementsFloat;
+AccumulatorBuffer<CountAccumulator<S64> >		gCounts;
+AccumulatorBuffer<MeasurementAccumulator<S64> >	gMeasurements;
+AccumulatorBuffer<TimeBlockAccumulator>			gStackTimers;
+AccumulatorBuffer<MemStatAccumulator>			gMemStats;
+
+void ThreadRecorder::deactivate( Recording* recording )
+{
+	std::list<ActiveRecording>::iterator it = update(recording);
+	if (it != mActiveRecordings.end())
+	{
+		// and if we've found the recording we wanted to update
+		std::list<ActiveRecording>::iterator next_it = it;
+		++next_it;
+		if (next_it != mActiveRecordings.end())
+		{
+			next_it->mTargetRecording->makePrimary();
+		}
+
+		mActiveRecordings.erase(it);
+	}
+}
+
+ThreadRecorder::ActiveRecording::ActiveRecording( Recording* target ) 
+:	mTargetRecording(target)
+{
+}
+
+void ThreadRecorder::ActiveRecording::moveBaselineToTarget()
+{
+	mTargetRecording->mMeasurementsFloat.write()->addSamples(*mBaseline.mMeasurementsFloat);
+	mTargetRecording->mCountsFloat.write()->addSamples(*mBaseline.mCountsFloat);
+	mTargetRecording->mMeasurements.write()->addSamples(*mBaseline.mMeasurements);
+	mTargetRecording->mCounts.write()->addSamples(*mBaseline.mCounts);
+	mTargetRecording->mStackTimers.write()->addSamples(*mBaseline.mStackTimers);
+	mBaseline.mMeasurementsFloat.write()->reset();
+	mBaseline.mCountsFloat.write()->reset();
+	mBaseline.mMeasurements.write()->reset();
+	mBaseline.mCounts.write()->reset();
+	mBaseline.mStackTimers.write()->reset();
+}
+
+
+///////////////////////////////////////////////////////////////////////
+// SlaveThreadRecorder
+///////////////////////////////////////////////////////////////////////
+
+SlaveThreadRecorder::SlaveThreadRecorder()
+{
+	getMasterThreadRecorder().addSlaveThread(this);
+}
+
+SlaveThreadRecorder::~SlaveThreadRecorder()
+{
+	getMasterThreadRecorder().removeSlaveThread(this);
+}
+
+void SlaveThreadRecorder::pushToMaster()
+{
+	mThreadRecording.stop();
+	{
+		LLMutexLock(getMasterThreadRecorder().getSlaveListMutex());
+		mSharedData.appendFrom(mThreadRecording);
+	}
+	mThreadRecording.start();
+}
+
+void SlaveThreadRecorder::SharedData::appendFrom( const Recording& source )
+{
+	LLMutexLock lock(&mRecordingMutex);
+	mRecording.appendRecording(source);
+}
+
+void SlaveThreadRecorder::SharedData::appendTo( Recording& sink )
+{
+	LLMutexLock lock(&mRecordingMutex);
+	sink.appendRecording(mRecording);
+}
+
+void SlaveThreadRecorder::SharedData::mergeFrom( const Recording& source )
+{
+	LLMutexLock lock(&mRecordingMutex);
+	mRecording.mergeRecording(source);
+}
+
+void SlaveThreadRecorder::SharedData::mergeTo( Recording& sink )
+{
+	LLMutexLock lock(&mRecordingMutex);
+	sink.mergeRecording(mRecording);
+}
+
+void SlaveThreadRecorder::SharedData::reset()
+{
+	LLMutexLock lock(&mRecordingMutex);
+	mRecording.reset();
+}
+
+
+///////////////////////////////////////////////////////////////////////
+// MasterThreadRecorder
+///////////////////////////////////////////////////////////////////////
+
+LLFastTimer::DeclareTimer FTM_PULL_TRACE_DATA_FROM_SLAVES("Pull slave trace data");
+void MasterThreadRecorder::pullFromSlaveThreads()
+{
+	LLFastTimer _(FTM_PULL_TRACE_DATA_FROM_SLAVES);
+	if (mActiveRecordings.empty()) return;
+
+	LLMutexLock lock(&mSlaveListMutex);
+
+	Recording& target_recording = mActiveRecordings.front().mBaseline;
+	for (slave_thread_recorder_list_t::iterator it = mSlaveThreadRecorders.begin(), end_it = mSlaveThreadRecorders.end();
+		it != end_it;
+		++it)
+	{
+		// ignore block timing info for now
+		(*it)->mSharedData.mergeTo(target_recording);
+		(*it)->mSharedData.reset();
+	}
+}
+
+void MasterThreadRecorder::addSlaveThread( class SlaveThreadRecorder* child )
+{
+	LLMutexLock lock(&mSlaveListMutex);
+
+	mSlaveThreadRecorders.push_back(child);
+}
+
+void MasterThreadRecorder::removeSlaveThread( class SlaveThreadRecorder* child )
+{
+	LLMutexLock lock(&mSlaveListMutex);
+
+	for (slave_thread_recorder_list_t::iterator it = mSlaveThreadRecorders.begin(), end_it = mSlaveThreadRecorders.end();
+		it != end_it;
+		++it)
+	{
+		if ((*it) == child)
+		{
+			mSlaveThreadRecorders.erase(it);
+			break;
+		}
+	}
+}
+
+void MasterThreadRecorder::pushToMaster()
+{}
+
+MasterThreadRecorder::MasterThreadRecorder()
+{}
+
+}
diff --git a/indra/llcommon/lltracethreadrecorder.h b/indra/llcommon/lltracethreadrecorder.h
new file mode 100644
index 0000000000..337035974c
--- /dev/null
+++ b/indra/llcommon/lltracethreadrecorder.h
@@ -0,0 +1,123 @@
+/** 
+ * @file lltrace.h
+ * @brief Runtime statistics accumulation.
+ *
+ * $LicenseInfo:firstyear=2001&license=viewerlgpl$
+ * Second Life Viewer Source Code
+ * Copyright (C) 2012, Linden Research, Inc.
+ * 
+ * This library is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation;
+ * version 2.1 of the License only.
+ * 
+ * This library is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ * 
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with this library; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA
+ * 
+ * Linden Research, Inc., 945 Battery Street, San Francisco, CA  94111  USA
+ * $/LicenseInfo$
+ */
+
+#ifndef LL_LLTRACETHREADRECORDER_H
+#define LL_LLTRACETHREADRECORDER_H
+
+#include "stdtypes.h"
+#include "llpreprocessor.h"
+
+#include "llmutex.h"
+#include "lltracerecording.h"
+
+namespace LLTrace
+{
+	class LL_COMMON_API ThreadRecorder
+	{
+	protected:
+		struct ActiveRecording;
+	public:
+		ThreadRecorder();
+
+		virtual ~ThreadRecorder();
+
+		void activate(Recording* recording);
+		std::list<struct ActiveRecording>::iterator update(Recording* recording);
+		void deactivate(Recording* recording);
+
+		virtual void pushToMaster() = 0;
+
+		TimeBlockTreeNode* getTimeBlockTreeNode(S32 index);
+
+	protected:
+		struct ActiveRecording
+		{
+			ActiveRecording(Recording* target);
+
+			Recording*	mTargetRecording;
+			Recording	mBaseline;
+
+			void moveBaselineToTarget();
+		};
+		Recording					mThreadRecording;
+		std::list<ActiveRecording>	mActiveRecordings;
+
+		class BlockTimer*				mRootTimer;
+		TimeBlockTreeNode*				mTimeBlockTreeNodes;
+		size_t							mNumTimeBlockTreeNodes;
+	};
+
+	class LL_COMMON_API MasterThreadRecorder : public ThreadRecorder
+	{
+	public:
+		MasterThreadRecorder();
+
+		void addSlaveThread(class SlaveThreadRecorder* child);
+		void removeSlaveThread(class SlaveThreadRecorder* child);
+
+		/*virtual */ void pushToMaster();
+
+		// call this periodically to gather stats data from slave threads
+		void pullFromSlaveThreads();
+
+		LLMutex* getSlaveListMutex() { return &mSlaveListMutex; }
+
+	private:
+
+		typedef std::list<class SlaveThreadRecorder*> slave_thread_recorder_list_t;
+
+		slave_thread_recorder_list_t	mSlaveThreadRecorders;
+		LLMutex							mSlaveListMutex;
+	};
+
+	class LL_COMMON_API SlaveThreadRecorder : public ThreadRecorder
+	{
+	public:
+		SlaveThreadRecorder();
+		~SlaveThreadRecorder();
+
+		// call this periodically to gather stats data for master thread to consume
+		/*virtual*/ void pushToMaster();
+
+		MasterThreadRecorder* 	mMaster;
+
+		class SharedData
+		{
+		public:
+			void appendFrom(const Recording& source);
+			void appendTo(Recording& sink);
+			void mergeFrom(const Recording& source);
+			void mergeTo(Recording& sink);
+			void reset();
+		private:
+			LLMutex		mRecordingMutex;
+			Recording	mRecording;
+		};
+		SharedData		mSharedData;
+	};
+}
+
+#endif // LL_LLTRACETHREADRECORDER_H
diff --git a/indra/llcommon/llunit.h b/indra/llcommon/llunit.h
new file mode 100644
index 0000000000..c600883607
--- /dev/null
+++ b/indra/llcommon/llunit.h
@@ -0,0 +1,475 @@
+/** 
+ * @file llunit.h
+ * @brief Unit conversion classes
+ *
+ * $LicenseInfo:firstyear=2001&license=viewerlgpl$
+ * Second Life Viewer Source Code
+ * Copyright (C) 2012, Linden Research, Inc.
+ * 
+ * This library is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation;
+ * version 2.1 of the License only.
+ * 
+ * This library is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ * 
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with this library; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA
+ * 
+ * Linden Research, Inc., 945 Battery Street, San Francisco, CA  94111  USA
+ * $/LicenseInfo$
+ */
+
+#ifndef LL_LLUNIT_H
+#define LL_LLUNIT_H
+
+#include "stdtypes.h"
+#include "llpreprocessor.h"
+#include "llerrorlegacy.h"
+
+namespace LLUnits
+{
+
+template<typename T>
+struct HighestPrecisionType
+{
+	typedef T type_t;
+};
+
+template<> struct HighestPrecisionType<F32> { typedef F64 type_t; };
+template<> struct HighestPrecisionType<S32> { typedef S64 type_t; };
+template<> struct HighestPrecisionType<U32> { typedef S64 type_t; };
+template<> struct HighestPrecisionType<S16> { typedef S64 type_t; };
+template<> struct HighestPrecisionType<U16> { typedef S64 type_t; };
+template<> struct HighestPrecisionType<S8> { typedef S64 type_t; };
+template<> struct HighestPrecisionType<U8> { typedef S64 type_t; };
+
+template<typename DERIVED_UNITS_TAG, typename BASE_UNITS_TAG, typename VALUE_TYPE>
+struct ConversionFactor
+{
+	static typename HighestPrecisionType<VALUE_TYPE>::type_t get()
+	{
+		// spurious use of dependent type to stop gcc from triggering the static assertion before instantiating the template
+		llstatic_assert_template(DERIVED_UNITS_TAG, false,  "Cannot convert between types.");
+	}
+};
+
+template<typename BASE_UNITS_TAG, typename VALUE_TYPE>
+struct ConversionFactor<BASE_UNITS_TAG, BASE_UNITS_TAG, VALUE_TYPE>
+{
+	static typename HighestPrecisionType<VALUE_TYPE>::type_t get() 
+	{ 
+		return 1; 
+	}
+};
+
+}
+
+template<typename UNIT_TYPE, typename STORAGE_TYPE>
+struct LLUnit
+{
+	typedef LLUnit<UNIT_TYPE, STORAGE_TYPE> self_t;
+	typedef STORAGE_TYPE storage_t;
+
+	// value initialization
+	LLUnit(storage_t value = storage_t())
+	:	mValue(value)
+	{}
+
+	// unit initialization and conversion
+	template<typename OTHER_UNIT, typename OTHER_STORAGE>
+	LLUnit(LLUnit<OTHER_UNIT, OTHER_STORAGE> other)
+	:	mValue(convert(other))
+	{}
+	
+	// value assignment
+	self_t& operator = (storage_t value)
+	{
+		mValue = value;
+		return *this;
+	}
+
+	// unit assignment
+	template<typename OTHER_UNIT, typename OTHER_STORAGE>
+	self_t& operator = (LLUnit<OTHER_UNIT, OTHER_STORAGE> other)
+	{
+		mValue = convert(other);
+		return *this;
+	}
+
+	storage_t value() const
+	{
+		return mValue;
+	}
+
+	template<typename NEW_UNIT_TYPE, typename NEW_STORAGE_TYPE> LLUnit<NEW_UNIT_TYPE, NEW_STORAGE_TYPE> as()
+	{
+		return LLUnit<NEW_UNIT_TYPE, NEW_STORAGE_TYPE>(*this);
+	}
+
+	void operator += (storage_t value)
+	{
+		mValue += value;
+	}
+
+	template<typename OTHER_UNIT, typename OTHER_STORAGE>
+	void operator += (LLUnit<OTHER_UNIT, OTHER_STORAGE> other)
+	{
+		mValue += convert(other);
+	}
+
+	void operator -= (storage_t value)
+	{
+		mValue -= value;
+	}
+
+	template<typename OTHER_UNIT, typename OTHER_STORAGE>
+	void operator -= (LLUnit<OTHER_UNIT, OTHER_STORAGE> other)
+	{
+		mValue -= convert(other);
+	}
+
+	void operator *= (storage_t multiplicand)
+	{
+		mValue *= multiplicand;
+	}
+
+	template<typename OTHER_UNIT, typename OTHER_STORAGE>
+	void operator *= (LLUnit<OTHER_UNIT, OTHER_STORAGE> multiplicand)
+	{
+		// spurious use of dependent type to stop gcc from triggering the static assertion before instantiating the template
+		llstatic_assert_template(OTHER_UNIT, 0, "Multiplication of unit types not supported.");
+	}
+
+	void operator /= (storage_t divisor)
+	{
+		mValue /= divisor;
+	}
+
+	template<typename OTHER_UNIT, typename OTHER_STORAGE>
+	void operator /= (LLUnit<OTHER_UNIT, OTHER_STORAGE> divisor)
+	{
+		// spurious use of dependent type to stop gcc from triggering the static assertion before instantiating the template
+		llstatic_assert_template(OTHER_UNIT, 0, "Illegal in-place division of unit types.");
+	}
+
+	template<typename SOURCE_UNITS, typename SOURCE_STORAGE>
+	static storage_t convert(LLUnit<SOURCE_UNITS, SOURCE_STORAGE> v) 
+	{ 
+		return (storage_t)(v.value() 
+			* LLUnits::ConversionFactor<SOURCE_UNITS, typename UNIT_TYPE::base_unit_t, SOURCE_STORAGE>::get() 
+			* LLUnits::ConversionFactor<typename UNIT_TYPE::base_unit_t, UNIT_TYPE, STORAGE_TYPE>::get()); 
+	}
+
+protected:
+	storage_t mValue;
+};
+
+template<typename UNIT_TYPE, typename STORAGE_TYPE>
+struct LLUnitImplicit : public LLUnit<UNIT_TYPE, STORAGE_TYPE>
+{
+	typedef LLUnitImplicit<UNIT_TYPE, STORAGE_TYPE> self_t;
+	typedef typename LLUnit<UNIT_TYPE, STORAGE_TYPE>::storage_t storage_t;
+	typedef LLUnit<UNIT_TYPE, STORAGE_TYPE> base_t;
+
+	LLUnitImplicit(storage_t value = storage_t())
+	:	base_t(value)
+	{}
+
+	template<typename OTHER_UNIT, typename OTHER_STORAGE>
+	LLUnitImplicit(LLUnit<OTHER_UNIT, OTHER_STORAGE> other)
+	:	base_t(convert(other))
+	{}
+
+	// unlike LLUnit, LLUnitImplicit is *implicitly* convertable to a POD scalar (F32, S32, etc)
+	// this allows for interoperability with legacy code
+	operator storage_t() const
+	{
+		return base_t::value();
+	}
+};
+
+//
+// operator +
+//
+template<typename UNIT_TYPE1, typename STORAGE_TYPE1, typename UNIT_TYPE2, typename STORAGE_TYPE2>
+LLUnit<UNIT_TYPE1, STORAGE_TYPE1> operator + (LLUnit<UNIT_TYPE1, STORAGE_TYPE1> first, LLUnit<UNIT_TYPE2, STORAGE_TYPE2> second)
+{
+	LLUnit<UNIT_TYPE1, STORAGE_TYPE1> result(first);
+	result += second;
+	return result;
+}
+
+template<typename UNIT_TYPE, typename STORAGE_TYPE, typename SCALAR_TYPE>
+LLUnit<UNIT_TYPE, STORAGE_TYPE> operator + (LLUnit<UNIT_TYPE, STORAGE_TYPE> first, SCALAR_TYPE second)
+{
+	LLUnit<UNIT_TYPE, STORAGE_TYPE> result(first);
+	result += second;
+	return result;
+}
+
+template<typename UNIT_TYPE, typename STORAGE_TYPE, typename SCALAR_TYPE>
+LLUnit<UNIT_TYPE, STORAGE_TYPE> operator + (SCALAR_TYPE first, LLUnit<UNIT_TYPE, STORAGE_TYPE> second)
+{
+	LLUnit<UNIT_TYPE, STORAGE_TYPE> result(first);
+	result += second;
+	return result;
+}
+
+template<typename UNIT_TYPE1, typename STORAGE_TYPE1, typename UNIT_TYPE2, typename STORAGE_TYPE2>
+LLUnitImplicit<UNIT_TYPE1, STORAGE_TYPE1> operator + (LLUnitImplicit<UNIT_TYPE1, STORAGE_TYPE1> first, LLUnit<UNIT_TYPE2, STORAGE_TYPE2> second)
+{
+	LLUnitImplicit<UNIT_TYPE1, STORAGE_TYPE1> result(first);
+	result += second;
+	return result;
+}
+
+template<typename UNIT_TYPE, typename STORAGE_TYPE, typename SCALAR_TYPE>
+LLUnitImplicit<UNIT_TYPE, STORAGE_TYPE> operator + (LLUnitImplicit<UNIT_TYPE, STORAGE_TYPE> first, SCALAR_TYPE second)
+{
+	LLUnitImplicit<UNIT_TYPE, STORAGE_TYPE> result(first);
+	result += second;
+	return result;
+}
+
+template<typename UNIT_TYPE1, typename STORAGE_TYPE1, typename UNIT_TYPE2, typename STORAGE_TYPE2>
+LLUnitImplicit<UNIT_TYPE1, STORAGE_TYPE1> operator + (LLUnitImplicit<UNIT_TYPE1, STORAGE_TYPE1> first, LLUnitImplicit<UNIT_TYPE2, STORAGE_TYPE2> second)
+{
+	LLUnitImplicit<UNIT_TYPE1, STORAGE_TYPE1> result(first);
+	result += second;
+	return result;
+}
+
+//
+// operator -
+//
+template<typename UNIT_TYPE1, typename STORAGE_TYPE1, typename UNIT_TYPE2, typename STORAGE_TYPE2>
+LLUnit<UNIT_TYPE1, STORAGE_TYPE1> operator - (LLUnit<UNIT_TYPE1, STORAGE_TYPE1> first, LLUnit<UNIT_TYPE2, STORAGE_TYPE2> second)
+{
+	LLUnit<UNIT_TYPE1, STORAGE_TYPE1> result(first);
+	result -= second;
+	return result;
+}
+
+template<typename UNIT_TYPE, typename STORAGE_TYPE, typename SCALAR_TYPE>
+LLUnit<UNIT_TYPE, STORAGE_TYPE> operator - (LLUnit<UNIT_TYPE, STORAGE_TYPE> first, SCALAR_TYPE second)
+{
+	LLUnit<UNIT_TYPE, STORAGE_TYPE> result(first);
+	result -= second;
+	return result;
+}
+
+template<typename UNIT_TYPE, typename STORAGE_TYPE, typename SCALAR_TYPE>
+LLUnit<UNIT_TYPE, STORAGE_TYPE> operator - (SCALAR_TYPE first, LLUnit<UNIT_TYPE, STORAGE_TYPE> second)
+{
+	LLUnit<UNIT_TYPE, STORAGE_TYPE> result(first);
+	result -= second;
+	return result;
+}
+
+template<typename UNIT_TYPE1, typename STORAGE_TYPE1, typename UNIT_TYPE2, typename STORAGE_TYPE2>
+LLUnitImplicit<UNIT_TYPE1, STORAGE_TYPE1> operator - (LLUnitImplicit<UNIT_TYPE1, STORAGE_TYPE1> first, LLUnitImplicit<UNIT_TYPE2, STORAGE_TYPE2> second)
+{
+	LLUnitImplicit<UNIT_TYPE1, STORAGE_TYPE1> result(first);
+	result -= second;
+	return result;
+}
+
+template<typename UNIT_TYPE, typename STORAGE_TYPE, typename SCALAR_TYPE>
+LLUnitImplicit<UNIT_TYPE, STORAGE_TYPE> operator - (LLUnitImplicit<UNIT_TYPE, STORAGE_TYPE> first, SCALAR_TYPE second)
+{
+	LLUnitImplicit<UNIT_TYPE, STORAGE_TYPE> result(first);
+	result -= second;
+	return result;
+}
+
+template<typename UNIT_TYPE, typename STORAGE_TYPE, typename SCALAR_TYPE>
+LLUnitImplicit<UNIT_TYPE, STORAGE_TYPE> operator - (SCALAR_TYPE first, LLUnitImplicit<UNIT_TYPE, STORAGE_TYPE> second)
+{
+	LLUnitImplicit<UNIT_TYPE, STORAGE_TYPE> result(first);
+	result -= second;
+	return result;
+}
+
+//
+// operator *
+//
+template<typename UNIT_TYPE, typename STORAGE_TYPE, typename SCALAR_TYPE>
+LLUnit<UNIT_TYPE, STORAGE_TYPE> operator * (SCALAR_TYPE first, LLUnit<UNIT_TYPE, STORAGE_TYPE> second)
+{
+	return LLUnit<UNIT_TYPE, STORAGE_TYPE>(first * second.value());
+}
+
+template<typename UNIT_TYPE, typename STORAGE_TYPE, typename SCALAR_TYPE>
+LLUnit<UNIT_TYPE, STORAGE_TYPE> operator * (LLUnit<UNIT_TYPE, STORAGE_TYPE> first, SCALAR_TYPE second)
+{
+	return LLUnit<UNIT_TYPE, STORAGE_TYPE>(first.value() * second);
+}
+
+template<typename UNIT_TYPE1, typename STORAGE_TYPE1, typename UNIT_TYPE2, typename STORAGE_TYPE2>
+LLUnit<UNIT_TYPE1, STORAGE_TYPE1> operator * (LLUnit<UNIT_TYPE1, STORAGE_TYPE1>, LLUnit<UNIT_TYPE2, STORAGE_TYPE2>)
+{
+	// spurious use of dependent type to stop gcc from triggering the static assertion before instantiating the template
+	llstatic_assert_template(STORAGE_TYPE1, 0, "Multiplication of unit types results in new unit type - not supported.");
+	return LLUnit<UNIT_TYPE1, STORAGE_TYPE1>();
+}
+
+template<typename UNIT_TYPE, typename STORAGE_TYPE, typename SCALAR_TYPE>
+LLUnitImplicit<UNIT_TYPE, STORAGE_TYPE> operator * (SCALAR_TYPE first, LLUnitImplicit<UNIT_TYPE, STORAGE_TYPE> second)
+{
+	return LLUnitImplicit<UNIT_TYPE, STORAGE_TYPE>(first * second.value());
+}
+
+template<typename UNIT_TYPE, typename STORAGE_TYPE, typename SCALAR_TYPE>
+LLUnitImplicit<UNIT_TYPE, STORAGE_TYPE> operator * (LLUnitImplicit<UNIT_TYPE, STORAGE_TYPE> first, SCALAR_TYPE second)
+{
+	return LLUnitImplicit<UNIT_TYPE, STORAGE_TYPE>(first.value() * second);
+}
+
+template<typename UNIT_TYPE1, typename STORAGE_TYPE1, typename UNIT_TYPE2, typename STORAGE_TYPE2>
+LLUnitImplicit<UNIT_TYPE1, STORAGE_TYPE1> operator * (LLUnitImplicit<UNIT_TYPE1, STORAGE_TYPE1>, LLUnitImplicit<UNIT_TYPE2, STORAGE_TYPE2>)
+{
+	// spurious use of dependent type to stop gcc from triggering the static assertion before instantiating the template
+	llstatic_assert_template(STORAGE_TYPE1, 0, "Multiplication of unit types results in new unit type - not supported.");
+	return LLUnitImplicit<UNIT_TYPE1, STORAGE_TYPE1>();
+}
+
+//
+// operator /
+//
+template<typename UNIT_TYPE, typename STORAGE_TYPE, typename SCALAR_TYPE>
+SCALAR_TYPE operator / (SCALAR_TYPE first, LLUnit<UNIT_TYPE, STORAGE_TYPE> second)
+{
+	return SCALAR_TYPE(first / second.value());
+}
+
+template<typename UNIT_TYPE, typename STORAGE_TYPE, typename SCALAR_TYPE>
+LLUnit<UNIT_TYPE, STORAGE_TYPE> operator / (LLUnit<UNIT_TYPE, STORAGE_TYPE> first, SCALAR_TYPE second)
+{
+	return LLUnit<UNIT_TYPE, STORAGE_TYPE>(first.value() / second);
+}
+
+template<typename UNIT_TYPE1, typename STORAGE_TYPE1, typename UNIT_TYPE2, typename STORAGE_TYPE2>
+STORAGE_TYPE1 operator / (LLUnit<UNIT_TYPE1, STORAGE_TYPE1> first, LLUnit<UNIT_TYPE2, STORAGE_TYPE2> second)
+{
+	// spurious use of dependent type to stop gcc from triggering the static assertion before instantiating the template
+	return STORAGE_TYPE1(first.value() / second.value());
+}
+
+template<typename UNIT_TYPE, typename STORAGE_TYPE, typename SCALAR_TYPE>
+LLUnitImplicit<UNIT_TYPE, STORAGE_TYPE> operator / (LLUnitImplicit<UNIT_TYPE, STORAGE_TYPE> first, SCALAR_TYPE second)
+{
+	return LLUnitImplicit<UNIT_TYPE, STORAGE_TYPE>(first.value() / second);
+}
+
+template<typename UNIT_TYPE1, typename STORAGE_TYPE1, typename UNIT_TYPE2, typename STORAGE_TYPE2>
+STORAGE_TYPE1 operator / (LLUnitImplicit<UNIT_TYPE1, STORAGE_TYPE1> first, LLUnitImplicit<UNIT_TYPE2, STORAGE_TYPE2> second)
+{
+	// spurious use of dependent type to stop gcc from triggering the static assertion before instantiating the template
+	return STORAGE_TYPE1(first.value() / second.value());
+}
+
+#define COMPARISON_OPERATORS(op)                                                                                     \
+template<typename UNIT_TYPE, typename STORAGE_TYPE, typename SCALAR_TYPE>                                            \
+bool operator op (SCALAR_TYPE first, LLUnit<UNIT_TYPE, STORAGE_TYPE> second)                                         \
+{                                                                                                                    \
+	return first op second.value();                                                                                  \
+}                                                                                                                    \
+	                                                                                                                 \
+template<typename UNIT_TYPE, typename STORAGE_TYPE, typename SCALAR_TYPE>                                            \
+bool operator op (LLUnit<UNIT_TYPE, STORAGE_TYPE> first, SCALAR_TYPE second)                                         \
+{                                                                                                                    \
+	return first.value() op second;                                                                                  \
+}                                                                                                                    \
+	                                                                                                                 \
+template<typename UNIT_TYPE1, typename STORAGE_TYPE1, typename UNIT_TYPE2, typename STORAGE_TYPE2>                   \
+bool operator op (LLUnitImplicit<UNIT_TYPE1, STORAGE_TYPE1> first, LLUnitImplicit<UNIT_TYPE2, STORAGE_TYPE2> second) \
+{                                                                                                                    \
+	return first.value() op first.convert(second);                                                                   \
+}                                                                                                                    \
+	                                                                                                                 \
+template<typename UNIT_TYPE1, typename STORAGE_TYPE1, typename UNIT_TYPE2, typename STORAGE_TYPE2>                   \
+	bool operator op (LLUnit<UNIT_TYPE1, STORAGE_TYPE1> first, LLUnit<UNIT_TYPE2, STORAGE_TYPE2> second)             \
+{                                                                                                                    \
+	return first.value() op first.convert(second);                                                                   \
+}
+
+COMPARISON_OPERATORS(<)
+COMPARISON_OPERATORS(<=)
+COMPARISON_OPERATORS(>)
+COMPARISON_OPERATORS(>=)
+COMPARISON_OPERATORS(==)
+COMPARISON_OPERATORS(!=)
+
+namespace LLUnits
+{
+template<typename T>
+T rawValue(T val) { return val; }
+
+template<typename UNIT_TYPE, typename STORAGE_TYPE> 
+STORAGE_TYPE rawValue(LLUnit<UNIT_TYPE, STORAGE_TYPE> val) { return val.value(); }
+
+template<typename UNIT_TYPE, typename STORAGE_TYPE> 
+STORAGE_TYPE rawValue(LLUnitImplicit<UNIT_TYPE, STORAGE_TYPE> val) { return val.value(); }
+
+template<typename UNIT_TYPE, typename STORAGE_TYPE> 
+struct HighestPrecisionType<LLUnit<UNIT_TYPE, STORAGE_TYPE> >
+{
+	typedef typename HighestPrecisionType<STORAGE_TYPE>::type_t type_t;
+};
+
+#define LL_DECLARE_DERIVED_UNIT(conversion_factor, base_unit_name, unit_name)                   \
+struct unit_name                                                                                \
+{                                                                                               \
+	typedef base_unit_name base_unit_t;                                                         \
+};                                                                                              \
+template<typename STORAGE_TYPE>                                                                 \
+struct ConversionFactor<unit_name, base_unit_name, STORAGE_TYPE>                                \
+{                                                                                               \
+	static typename HighestPrecisionType<STORAGE_TYPE>::type_t get()                            \
+	{                                                                                           \
+		return typename HighestPrecisionType<STORAGE_TYPE>::type_t(conversion_factor);          \
+	}                                                                                           \
+};                                                                                              \
+	                                                                                            \
+template<typename STORAGE_TYPE>                                                                 \
+struct ConversionFactor<base_unit_name, unit_name, STORAGE_TYPE>						        \
+{                                                                                               \
+	static typename HighestPrecisionType<STORAGE_TYPE>::type_t get()                            \
+	{                                                                                           \
+		return typename HighestPrecisionType<STORAGE_TYPE>::type_t(1.0 / (conversion_factor));  \
+	}                                                                                           \
+}
+
+struct Bytes { typedef Bytes base_unit_t; };
+LL_DECLARE_DERIVED_UNIT(1024, Bytes, Kilobytes);
+LL_DECLARE_DERIVED_UNIT(1024 * 1024, Bytes, Megabytes);
+LL_DECLARE_DERIVED_UNIT(1024 * 1024 * 1024, Bytes, Gigabytes);
+LL_DECLARE_DERIVED_UNIT(1.0 / 8.0, Bytes, Bits);
+LL_DECLARE_DERIVED_UNIT(1024 / 8, Bytes, Kilobits);
+LL_DECLARE_DERIVED_UNIT(1024 / 8, Bytes, Megabits);
+LL_DECLARE_DERIVED_UNIT(1024 * 1024 * 1024 / 8, Bytes, Gigabits);
+
+struct Seconds { typedef Seconds base_unit_t; };
+LL_DECLARE_DERIVED_UNIT(60, Seconds, Minutes);
+LL_DECLARE_DERIVED_UNIT(60 * 60, Seconds, Hours);
+LL_DECLARE_DERIVED_UNIT(1.0 / 1000.0, Seconds, Milliseconds);
+LL_DECLARE_DERIVED_UNIT(1.0 / 1000000.0, Seconds, Microseconds);
+LL_DECLARE_DERIVED_UNIT(1.0 / 1000000000.0, Seconds, Nanoseconds);
+
+struct Meters { typedef Meters base_unit_t; };
+LL_DECLARE_DERIVED_UNIT(1000, Meters, Kilometers);
+LL_DECLARE_DERIVED_UNIT(1.0 / 100.0, Meters, Centimeters);
+LL_DECLARE_DERIVED_UNIT(1.0 / 1000.0, Meters, Millimeters);
+
+struct Hertz { typedef Hertz base_unit_t; };
+LL_DECLARE_DERIVED_UNIT(1000, Hertz, Kilohertz);
+LL_DECLARE_DERIVED_UNIT(1000 * 1000, Hertz, Megahertz);
+LL_DECLARE_DERIVED_UNIT(1000 * 1000 * 1000, Hertz, Gigahertz);
+} // namespace LLUnits
+
+#endif // LL_LLUNIT_H
diff --git a/indra/llcommon/lluuid.cpp b/indra/llcommon/lluuid.cpp
index 0aaa50d231..ba4b670b9a 100644
--- a/indra/llcommon/lluuid.cpp
+++ b/indra/llcommon/lluuid.cpp
@@ -27,9 +27,7 @@
 
 // We can't use WIN32_LEAN_AND_MEAN here, needs lots of includes.
 #if LL_WINDOWS
-#undef WIN32_LEAN_AND_MEAN
-#include <winsock2.h>
-#include <windows.h>
+#include "llwin32headers.h"
 // ugh, this is ugly.  We need to straighten out our linking for this library
 #pragma comment(lib, "IPHLPAPI.lib")
 #include <iphlpapi.h>
diff --git a/indra/llcommon/llwin32headers.h b/indra/llcommon/llwin32headers.h
new file mode 100644
index 0000000000..9c89b6b280
--- /dev/null
+++ b/indra/llcommon/llwin32headers.h
@@ -0,0 +1,36 @@
+/** 
+ * @file llwindows.h
+ * @brief sanitized include of windows header files
+ *
+ * $LicenseInfo:firstyear=2001&license=viewerlgpl$
+ * Second Life Viewer Source Code
+ * Copyright (C) 2010, Linden Research, Inc.
+ * 
+ * This library is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation;
+ * version 2.1 of the License only.
+ * 
+ * This library is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ * 
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with this library; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA
+ * 
+ * Linden Research, Inc., 945 Battery Street, San Francisco, CA  94111  USA
+ * $/LicenseInfo$
+ */
+
+#ifndef LL_LLWINDOWS_H
+#define LL_LLWINDOWS_H
+
+#ifdef LL_WINDOWS
+#undef WIN32_LEAN_AND_MEAN
+#include <winsock2.h>
+#include <windows.h>
+#endif
+
+#endif
diff --git a/indra/llcommon/llwin32headerslean.h b/indra/llcommon/llwin32headerslean.h
new file mode 100644
index 0000000000..d3fb90d4b1
--- /dev/null
+++ b/indra/llcommon/llwin32headerslean.h
@@ -0,0 +1,36 @@
+/** 
+ * @file llwindows.h
+ * @brief sanitized include of windows header files
+ *
+ * $LicenseInfo:firstyear=2001&license=viewerlgpl$
+ * Second Life Viewer Source Code
+ * Copyright (C) 2010, Linden Research, Inc.
+ * 
+ * This library is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation;
+ * version 2.1 of the License only.
+ * 
+ * This library is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ * 
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with this library; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA
+ * 
+ * Linden Research, Inc., 945 Battery Street, San Francisco, CA  94111  USA
+ * $/LicenseInfo$
+ */
+
+#ifndef LL_LLWINDOWS_H
+#define LL_LLWINDOWS_H
+
+#ifdef LL_WINDOWS
+#define WIN32_LEAN_AND_MEAN
+#include <winsock2.h>
+#include <windows.h>
+#endif
+
+#endif
diff --git a/indra/llcommon/llworkerthread.h b/indra/llcommon/llworkerthread.h
index be46394d6e..09776816a8 100644
--- a/indra/llcommon/llworkerthread.h
+++ b/indra/llcommon/llworkerthread.h
@@ -26,10 +26,11 @@
 #ifndef LL_LLWORKERTHREAD_H
 #define LL_LLWORKERTHREAD_H
 
-#include <queue>
-#include <string>
+#include <list>
 #include <map>
+#include <queue>
 #include <set>
+#include <string>
 
 #include "llqueuedthread.h"
 #include "llapr.h"
diff --git a/indra/llcommon/tests/llunits_test.cpp b/indra/llcommon/tests/llunits_test.cpp
new file mode 100644
index 0000000000..2a941e8229
--- /dev/null
+++ b/indra/llcommon/tests/llunits_test.cpp
@@ -0,0 +1,208 @@
+/** 
+ * @file llsingleton_test.cpp
+ * @date 2011-08-11
+ * @brief Unit test for the LLSingleton class
+ *
+ * $LicenseInfo:firstyear=2011&license=viewerlgpl$
+ * Second Life Viewer Source Code
+ * Copyright (C) 2011, Linden Research, Inc.
+ *
+ * This library is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation;
+ * version 2.1 of the License only.
+ *
+ * This library is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with this library; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA
+ *
+ * Linden Research, Inc., 945 Battery Street, San Francisco, CA  94111  USA
+ * $/LicenseInfo$
+ */
+
+#include "linden_common.h"
+
+#include "llunit.h"
+#include "../test/lltut.h"
+
+namespace LLUnits
+{
+	// using powers of 2 to allow strict floating point equality
+	struct Quatloos { typedef Quatloos base_unit_t; };
+	LL_DECLARE_DERIVED_UNIT(4, Quatloos, Latinum);
+	LL_DECLARE_DERIVED_UNIT((1.0 / 4.0), Quatloos, Solari);
+}
+
+namespace tut
+{
+	using namespace LLUnits;
+	struct units
+	{
+	};
+
+	typedef test_group<units> units_t;
+	typedef units_t::object units_object_t;
+	tut::units_t tut_singleton("LLUnit");
+
+	// storage type conversions
+	template<> template<>
+	void units_object_t::test<1>()
+	{
+		LLUnit<Quatloos, F32> float_quatloos;
+		ensure(float_quatloos.value() == 0.f);
+
+		LLUnit<Quatloos, S32> int_quatloos;
+		ensure(int_quatloos.value() == 0);
+
+		int_quatloos = 42;
+		ensure(int_quatloos.value() == 42);
+		float_quatloos = int_quatloos;
+		ensure(float_quatloos.value() == 42.f);
+
+		int_quatloos = float_quatloos;
+		ensure(int_quatloos.value() == 42);
+
+		float_quatloos = 42.1f;
+		ensure(float_quatloos.value() == 42.1f);
+		int_quatloos = float_quatloos;
+		ensure(int_quatloos.value() == 42);
+		LLUnit<Quatloos, U32> unsigned_int_quatloos(float_quatloos);
+		ensure(unsigned_int_quatloos.value() == 42);
+	}
+
+	// conversions to/from base unit
+	template<> template<>
+	void units_object_t::test<2>()
+	{
+		LLUnit<Quatloos, F32> quatloos(1.f);
+		ensure(quatloos.value() == 1.f);
+		LLUnit<Latinum, F32> latinum_bars(quatloos);
+		ensure(latinum_bars.value() == 1.f / 4.f);
+
+		latinum_bars = 256;
+		quatloos = latinum_bars;
+		ensure(quatloos.value() == 1024);
+
+		LLUnit<Solari, F32> solari(quatloos);
+		ensure(solari.value() == 4096);
+	}
+
+	// conversions across non-base units
+	template<> template<>
+	void units_object_t::test<3>()
+	{
+		LLUnit<Solari, F32> solari = 4.f;
+		LLUnit<Latinum, F32> latinum_bars = solari;
+		ensure(latinum_bars.value() == 0.25f);
+	}
+
+	// math operations
+	template<> template<>
+	void units_object_t::test<4>()
+	{
+		LLUnit<Quatloos, F32> quatloos = 1.f;
+		quatloos *= 4.f;
+		ensure(quatloos.value() == 4);
+		quatloos = quatloos * 2;
+		ensure(quatloos.value() == 8);
+		quatloos = 2.f * quatloos;
+		ensure(quatloos.value() == 16);
+
+		quatloos += 4.f;
+		ensure(quatloos.value() == 20);
+		quatloos += 4;
+		ensure(quatloos.value() == 24);
+		quatloos = quatloos + 4;
+		ensure(quatloos.value() == 28);
+		quatloos = 4 + quatloos;
+		ensure(quatloos.value() == 32);
+		quatloos += quatloos * 3;
+		ensure(quatloos.value() == 128);
+
+		quatloos -= quatloos / 4 * 3;
+		ensure(quatloos.value() == 32);
+		quatloos = quatloos - 8;
+		ensure(quatloos.value() == 24);
+		quatloos -= 4;
+		ensure(quatloos.value() == 20);
+		quatloos -= 4.f;
+		ensure(quatloos.value() == 16);
+
+		quatloos *= 2.f;
+		ensure(quatloos.value() == 32);
+		quatloos = quatloos * 2.f;
+		ensure(quatloos.value() == 64);
+		quatloos = 0.5f * quatloos;
+		ensure(quatloos.value() == 32);
+
+		quatloos /= 2.f;
+		ensure(quatloos.value() == 16);
+		quatloos = quatloos / 4;
+		ensure(quatloos.value() == 4);
+
+		F32 ratio = quatloos / LLUnit<Quatloos, F32>(4.f);
+		ensure(ratio == 1);
+
+		quatloos += LLUnit<Solari, F32>(4.f);
+		ensure(quatloos.value() == 5);
+		quatloos -= LLUnit<Latinum, F32>(1.f);
+		ensure(quatloos.value() == 1);
+	}
+
+	// implicit units
+	template<> template<>
+	void units_object_t::test<5>()
+	{
+		// 0-initialized
+		LLUnit<Quatloos, F32> quatloos(0);
+		// initialize implicit unit from explicit
+		LLUnitImplicit<Quatloos, F32> quatloos_implicit = quatloos + 1;
+		ensure(quatloos_implicit.value() == 1);
+
+		// assign implicit to explicit, or perform math operations
+		quatloos = quatloos_implicit;
+		ensure(quatloos.value() == 1);
+		quatloos += quatloos_implicit;
+		ensure(quatloos.value() == 2);
+
+		// math operations on implicits
+		quatloos_implicit = 1;
+		ensure(quatloos_implicit == 1);
+
+		quatloos_implicit += 2;
+		ensure(quatloos_implicit == 3);
+
+		quatloos_implicit *= 2;
+		ensure(quatloos_implicit == 6);
+
+		quatloos_implicit -= 1;
+		ensure(quatloos_implicit == 5);
+
+		quatloos_implicit /= 5;
+		ensure(quatloos_implicit == 1);
+
+		quatloos_implicit = quatloos_implicit + 3 + quatloos_implicit;
+		ensure(quatloos_implicit == 5);
+
+		quatloos_implicit = 10 - quatloos_implicit - 1;
+		ensure(quatloos_implicit == 4);
+
+		quatloos_implicit = 2 * quatloos_implicit * 2;
+		ensure(quatloos_implicit == 16);
+
+		F32 one_half = quatloos_implicit / (quatloos_implicit * 2);
+		ensure(one_half == 0.5f);
+
+		// implicit conversion to POD
+		F32 float_val = quatloos_implicit;
+		ensure(float_val == 16);
+
+		S32 int_val = quatloos_implicit;
+		ensure(int_val == 16);
+	}
+}