Automated merge with bundle:d:\code\viewer-interesting+c:\users\richard\appdata\local\temp\thg.7jrj8n\https__bitbucket.org_lindenlab_viewer-interesting_jch9cd.hg

16 files changed, 597 insertions, 503 deletions

diff --git a/indra/llcommon/llcriticaldamp.cpp b/indra/llcommon/llcriticaldamp.cpp
index 2f013fe255..575fc4149e 100755
--- a/indra/llcommon/llcriticaldamp.cpp
+++ b/indra/llcommon/llcriticaldamp.cpp
@@ -81,7 +81,7 @@ void LLSmoothInterpolation::updateInterpolants()
 //-----------------------------------------------------------------------------
 // getInterpolant()
 //-----------------------------------------------------------------------------
-F32 LLSmoothInterpolation::getInterpolant(LLUnit<LLUnits::Seconds, F32> time_constant, bool use_cache)
+F32 LLSmoothInterpolation::getInterpolant(LLUnit<F32, LLUnits::Seconds> time_constant, bool use_cache)
 {
 	if (time_constant == 0.f)
 	{
diff --git a/indra/llcommon/llcriticaldamp.h b/indra/llcommon/llcriticaldamp.h
index ab5d4ba6e2..e174643cd0 100755
--- a/indra/llcommon/llcriticaldamp.h
+++ b/indra/llcommon/llcriticaldamp.h
@@ -42,10 +42,10 @@ public:
 	static void updateInterpolants();
 
 	// ACCESSORS
-	static F32 getInterpolant(LLUnit<LLUnits::Seconds, F32> time_constant, bool use_cache = true);
+	static F32 getInterpolant(LLUnit<F32, LLUnits::Seconds> time_constant, bool use_cache = true);
 
 	template<typename T> 
-	static T lerp(T a, T b, LLUnit<LLUnits::Seconds, F32> time_constant, bool use_cache = true)
+	static T lerp(T a, T b, LLUnit<F32, LLUnits::Seconds> time_constant, bool use_cache = true)
 	{
 		F32 interpolant = getInterpolant(time_constant, use_cache);
 		return ((a * (1.f - interpolant)) 
diff --git a/indra/llcommon/lldate.cpp b/indra/llcommon/lldate.cpp
index 2efe39e158..7892269e35 100755
--- a/indra/llcommon/lldate.cpp
+++ b/indra/llcommon/lldate.cpp
@@ -55,7 +55,7 @@ LLDate::LLDate(const LLDate& date) :
 	mSecondsSinceEpoch(date.mSecondsSinceEpoch)
 {}
 
-LLDate::LLDate(LLUnit<LLUnits::Seconds, F64> seconds_since_epoch) :
+LLDate::LLDate(LLUnit<F64, LLUnits::Seconds> seconds_since_epoch) :
 	mSecondsSinceEpoch(seconds_since_epoch.value())
 {}
 
diff --git a/indra/llcommon/lldate.h b/indra/llcommon/lldate.h
index b62a846147..1067ac5280 100755
--- a/indra/llcommon/lldate.h
+++ b/indra/llcommon/lldate.h
@@ -59,7 +59,7 @@ public:
 	 *
 	 * @param seconds_since_epoch The number of seconds since UTC epoch.
 	 */
-	LLDate(LLUnit<LLUnits::Seconds, F64> seconds_since_epoch);
+	LLDate(LLUnit<F64, LLUnits::Seconds> seconds_since_epoch);
 
 	/** 
 	 * @brief Construct a date from a string representation
diff --git a/indra/llcommon/llfasttimer.cpp b/indra/llcommon/llfasttimer.cpp
index dfc72bd2ce..4da9c3fd6c 100755
--- a/indra/llcommon/llfasttimer.cpp
+++ b/indra/llcommon/llfasttimer.cpp
@@ -146,8 +146,8 @@ 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();
-
+	static LLUnit<U64, LLUnits::Hertz> sCPUClockFrequency = LLProcessorInfo().getCPUFrequency();
+	return sCPUClockFrequency.value();
 #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())
@@ -159,8 +159,8 @@ U64 TimeBlock::countsPerSecond()
 		QueryPerformanceFrequency((LARGE_INTEGER*)&sCPUClockFrequency);
 		firstcall = false;
 	}
-#endif
 	return sCPUClockFrequency.value();
+#endif
 }
 #endif
 
@@ -178,43 +178,38 @@ TimeBlockTreeNode& TimeBlock::getTreeNode() const
 	return *nodep;
 }
 
-static LLFastTimer::DeclareTimer FTM_PROCESS_TIMES("Process FastTimer Times");
 
-// not thread safe, so only call on main thread
-//static
-void TimeBlock::processTimes()
+void TimeBlock::bootstrapTimerTree()
 {
-	LLFastTimer _(FTM_PROCESS_TIMES);
-	get_clock_count(); // good place to calculate clock frequency
-	U64 cur_time = getCPUClockCount64();
-
-	// set up initial tree
 	for (LLInstanceTracker<TimeBlock>::instance_iter begin_it = LLInstanceTracker<TimeBlock>::beginInstances(), end_it = LLInstanceTracker<TimeBlock>::endInstances(), it = begin_it; 
 		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
+			// 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
+// 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
+void TimeBlock::incrementalUpdateTimerTree()
+{
 	for(timer_tree_bottom_up_iterator_t it = begin_timer_tree_bottom_up(TimeBlock::getRootTimeBlock());
 		it != end_timer_tree_bottom_up();
 		++it)
@@ -240,27 +235,35 @@ void TimeBlock::processTimes()
 				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;
+				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
+				// as ancestors may call this timer only on certain paths, so we want to resolve
+				// child-most block locations before their parents
 				it.skipAncestors();
 			}
 		}
 	}
+}
+
+
+void TimeBlock::updateTimes()
+{
+	U64 cur_time = getCPUClockCount64();
 
 	// walk up stack of active timers and accumulate current time while leaving timing structures active
-	BlockTimerStackRecord* stack_record			= ThreadTimerStack::getInstance();
-	BlockTimer* cur_timer						= stack_record->mActiveTimer;
-	TimeBlockAccumulator* accumulator = stack_record->mTimeBlock->getPrimaryAccumulator();
+	BlockTimerStackRecord* stack_record	= ThreadTimerStack::getInstance();
+	BlockTimer* cur_timer				= stack_record->mActiveTimer;
+	TimeBlockAccumulator* accumulator	= stack_record->mTimeBlock->getPrimaryAccumulator();
 
 	while(cur_timer 
 		&& cur_timer->mParentTimerData.mActiveTimer != cur_timer) // root defined by parent pointing to self
 	{
 		U64 cumulative_time_delta = cur_time - cur_timer->mStartTime;
-		accumulator->mTotalTimeCounter += cumulative_time_delta - (accumulator->mTotalTimeCounter - cur_timer->mBlockStartTotalTimeCounter);
+		accumulator->mTotalTimeCounter += cumulative_time_delta 
+			- (accumulator->mTotalTimeCounter 
+			- cur_timer->mBlockStartTotalTimeCounter);
 		accumulator->mSelfTimeCounter += cumulative_time_delta - stack_record->mChildTime;
 		stack_record->mChildTime = 0;
 
@@ -268,11 +271,28 @@ void TimeBlock::processTimes()
 		cur_timer->mBlockStartTotalTimeCounter = accumulator->mTotalTimeCounter;
 
 		stack_record = &cur_timer->mParentTimerData;
-		accumulator = stack_record->mTimeBlock->getPrimaryAccumulator();
-		cur_timer = stack_record->mActiveTimer;
+		accumulator  = stack_record->mTimeBlock->getPrimaryAccumulator();
+		cur_timer    = stack_record->mActiveTimer;
 
 		stack_record->mChildTime += cumulative_time_delta;
 	}
+}
+
+static LLFastTimer::DeclareTimer FTM_PROCESS_TIMES("Process FastTimer Times");
+
+// not thread safe, so only call on main thread
+//static
+void TimeBlock::processTimes()
+{
+	LLFastTimer _(FTM_PROCESS_TIMES);
+	get_clock_count(); // good place to calculate clock frequency
+
+	// set up initial tree
+	bootstrapTimerTree();
+
+	incrementalUpdateTimerTree();
+
+	updateTimes();
 
 	// reset for next frame
 	for (LLInstanceTracker<TimeBlock>::instance_iter it = LLInstanceTracker<TimeBlock>::beginInstances(),
@@ -288,14 +308,13 @@ void TimeBlock::processTimes()
 	}
 }
 
-
 std::vector<TimeBlock*>::iterator TimeBlock::beginChildren()
-		{
+{
 	return getTreeNode().mChildren.begin(); 
-		}
+}
 
 std::vector<TimeBlock*>::iterator TimeBlock::endChildren()
-		{
+{
 	return getTreeNode().mChildren.end();
 }
 
@@ -318,11 +337,11 @@ void TimeBlock::logStats()
 			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;
+			LL_DEBUGS("FastTimers") << "elapsed sec " << ((F64)getCPUClockCount64()) / (LLUnit<F64, LLUnits::Hertz>(LLProcessorInfo().getCPUFrequency())) << LL_ENDL;
 		}
 		call_count++;
 
-		LLUnit<LLUnits::Seconds, F64> total_time(0);
+		LLUnit<F64, LLUnits::Seconds> total_time(0);
 		LLSD sd;
 
 		{
@@ -365,11 +384,11 @@ void TimeBlock::dumpCurTimes()
 		++it)
 	{
 		TimeBlock* timerp = (*it);
-		LLUnit<LLUnits::Seconds, F64> total_time_ms = last_frame_recording.getSum(*timerp);
+		LLUnit<F64, LLUnits::Seconds> total_time = 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;
+		if (total_time < LLUnit<F32, LLUnits::Milliseconds>(0.1)) continue;
 
 		std::ostringstream out_str;
 		TimeBlock* parent_timerp = timerp;
@@ -380,7 +399,7 @@ void TimeBlock::dumpCurTimes()
 		}
 
 		out_str << timerp->getName() << " " 
-			<< std::setprecision(3) << total_time_ms.as<LLUnits::Milliseconds>().value() << " ms, "
+			<< std::setprecision(3) << total_time.getAs<LLUnits::Milliseconds>() << " ms, "
 			<< num_calls << " calls";
 
 		llinfos << out_str.str() << llendl;
@@ -449,7 +468,7 @@ void TimeBlockAccumulator::reset( const TimeBlockAccumulator* other )
 	}
 }
 
-LLUnit<LLUnits::Seconds, F64> BlockTimer::getElapsedTime()
+LLUnit<F64, LLUnits::Seconds> BlockTimer::getElapsedTime()
 {
 	U64 total_time = TimeBlock::getCPUClockCount64() - mStartTime;
 
diff --git a/indra/llcommon/llfasttimer.h b/indra/llcommon/llfasttimer.h
index 20514d1638..e800befd9f 100755
--- a/indra/llcommon/llfasttimer.h
+++ b/indra/llcommon/llfasttimer.h
@@ -71,7 +71,7 @@ public:
 	BlockTimer(TimeBlock& timer);
 	~BlockTimer();
 
-	LLUnit<LLUnits::Seconds, F64> getElapsedTime();
+	LLUnit<F64, LLUnits::Seconds> getElapsedTime();
 
 private:
 
@@ -115,6 +115,7 @@ public:
 	static void pushLog(LLSD sd);
 	static void setLogLock(LLMutex* mutex);
 	static void writeLog(std::ostream& os);
+	static void updateTimes();
 
 	// dumps current cumulative frame stats to log
 	// call nextFrame() to reset timers
@@ -262,6 +263,9 @@ public:
 	// can be called multiple times in a frame, at any point
 	static void processTimes();
 
+	static void bootstrapTimerTree();
+	static void incrementalUpdateTimerTree();
+
 	// call this once a frame to periodically log timers
 	static void logStats();
 
diff --git a/indra/llcommon/llprocessor.cpp b/indra/llcommon/llprocessor.cpp
index 5ddfa6fcef..b80e813d84 100755
--- a/indra/llcommon/llprocessor.cpp
+++ b/indra/llcommon/llprocessor.cpp
@@ -875,7 +875,7 @@ LLProcessorInfo::LLProcessorInfo() : mImpl(NULL)
 
 
 LLProcessorInfo::~LLProcessorInfo() {}
-LLUnitImplicit<LLUnits::Megahertz, F64> LLProcessorInfo::getCPUFrequency() const { return mImpl->getCPUFrequency(); }
+LLUnitImplicit<F64, LLUnits::Megahertz> 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 fbd427f484..7f220467b0 100755
--- a/indra/llcommon/llprocessor.h
+++ b/indra/llcommon/llprocessor.h
@@ -37,7 +37,7 @@ public:
 	LLProcessorInfo(); 
  	~LLProcessorInfo();
 
-	LLUnitImplicit<LLUnits::Megahertz, F64> getCPUFrequency() const;
+	LLUnitImplicit<F64, LLUnits::Megahertz> getCPUFrequency() const;
 	bool hasSSE() const;
 	bool hasSSE2() const;
 	bool hasAltivec() const;
diff --git a/indra/llcommon/lltimer.cpp b/indra/llcommon/lltimer.cpp
index 838155d54d..693809b622 100755
--- a/indra/llcommon/lltimer.cpp
+++ b/indra/llcommon/lltimer.cpp
@@ -285,14 +285,14 @@ LLTimer::~LLTimer()
 }
 
 // static
-LLUnitImplicit<LLUnits::Microseconds, U64> LLTimer::getTotalTime()
+LLUnitImplicit<U64, LLUnits::Microseconds> LLTimer::getTotalTime()
 {
 	// simply call into the implementation function.
 	return totalTime();
 }	
 
 // static
-LLUnitImplicit<LLUnits::Seconds, F64> LLTimer::getTotalSeconds()
+LLUnitImplicit<F64, LLUnits::Seconds> LLTimer::getTotalSeconds()
 {
 	return U64_to_F64(getTotalTime()) * USEC_TO_SEC_F64;
 }
@@ -341,23 +341,23 @@ U64 getElapsedTimeAndUpdate(U64& lastClockCount)
 }
 
 
-LLUnitImplicit<LLUnits::Seconds, F64> LLTimer::getElapsedTimeF64() const
+LLUnitImplicit<F64, LLUnits::Seconds> LLTimer::getElapsedTimeF64() const
 {
 	U64 last = mLastClockCount;
 	return (F64)getElapsedTimeAndUpdate(last) * gClockFrequencyInv;
 }
 
-LLUnitImplicit<LLUnits::Seconds, F32> LLTimer::getElapsedTimeF32() const
+LLUnitImplicit<F32, LLUnits::Seconds> LLTimer::getElapsedTimeF32() const
 {
 	return (F32)getElapsedTimeF64();
 }
 
-LLUnitImplicit<LLUnits::Seconds, F64> LLTimer::getElapsedTimeAndResetF64()
+LLUnitImplicit<F64, LLUnits::Seconds> LLTimer::getElapsedTimeAndResetF64()
 {
 	return (F64)getElapsedTimeAndUpdate(mLastClockCount) * gClockFrequencyInv;
 }
 
-LLUnitImplicit<LLUnits::Seconds, F32> LLTimer::getElapsedTimeAndResetF32()
+LLUnitImplicit<F32, LLUnits::Seconds> LLTimer::getElapsedTimeAndResetF32()
 {
 	return (F32)getElapsedTimeAndResetF64();
 }
@@ -370,7 +370,7 @@ void  LLTimer::setTimerExpirySec(F32 expiration)
 		+ (U64)((F32)(expiration * gClockFrequency));
 }
 
-LLUnitImplicit<LLUnits::Seconds, F32> LLTimer::getRemainingTimeF32() const
+LLUnitImplicit<F32, LLUnits::Seconds> 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 0ba87d1e15..9e464c4b1a 100755
--- a/indra/llcommon/lltimer.h
+++ b/indra/llcommon/lltimer.h
@@ -67,16 +67,16 @@ public:
 
 	// Return a high precision number of seconds since the start of
 	// this application instance.
-	static LLUnitImplicit<LLUnits::Seconds, F64> getElapsedSeconds()
+	static LLUnitImplicit<F64, LLUnits::Seconds> getElapsedSeconds()
 	{
 		return sTimer->getElapsedTimeF64();
 	}
 
 	// Return a high precision usec since epoch
-	static LLUnitImplicit<LLUnits::Microseconds, U64> getTotalTime();
+	static LLUnitImplicit<U64, LLUnits::Microseconds> getTotalTime();
 
 	// Return a high precision seconds since epoch
-	static LLUnitImplicit<LLUnits::Seconds, F64> getTotalSeconds();
+	static LLUnitImplicit<F64, LLUnits::Seconds> getTotalSeconds();
 
 
 	// MANIPULATORS
@@ -87,16 +87,16 @@ public:
 	void setTimerExpirySec(F32 expiration);
 	BOOL checkExpirationAndReset(F32 expiration);
 	BOOL hasExpired() const;
-	LLUnitImplicit<LLUnits::Seconds, F32> getElapsedTimeAndResetF32();	// Returns elapsed time in seconds with reset
-	LLUnitImplicit<LLUnits::Seconds, F64> getElapsedTimeAndResetF64();
+	LLUnitImplicit<F32, LLUnits::Seconds> getElapsedTimeAndResetF32();	// Returns elapsed time in seconds with reset
+	LLUnitImplicit<F64, LLUnits::Seconds> getElapsedTimeAndResetF64();
 
-	LLUnitImplicit<LLUnits::Seconds, F32> getRemainingTimeF32() const;
+	LLUnitImplicit<F32, LLUnits::Seconds> getRemainingTimeF32() const;
 
 	static BOOL knownBadTimer();
 
 	// ACCESSORS
-	LLUnitImplicit<LLUnits::Seconds, F32> getElapsedTimeF32() const;			// Returns elapsed time in seconds
-	LLUnitImplicit<LLUnits::Seconds, F64> getElapsedTimeF64() const;			// Returns elapsed time in seconds
+	LLUnitImplicit<F32, LLUnits::Seconds> getElapsedTimeF32() const;			// Returns elapsed time in seconds
+	LLUnitImplicit<F64, LLUnits::Seconds> getElapsedTimeF64() const;			// Returns elapsed time in seconds
 
 	bool getStarted() const { return mStarted; }
 
diff --git a/indra/llcommon/lltrace.h b/indra/llcommon/lltrace.h
index cfe1273b4b..6292534a03 100644
--- a/indra/llcommon/lltrace.h
+++ b/indra/llcommon/lltrace.h
@@ -44,27 +44,37 @@ namespace LLTrace
 {
 class Recording;
 
-typedef LLUnit<LLUnits::Bytes, F64>			Bytes;
-typedef LLUnit<LLUnits::Kibibytes, F64>		Kibibytes;
-typedef LLUnit<LLUnits::Mibibytes, F64>		Mibibytes;
-typedef LLUnit<LLUnits::Gibibytes, F64>		Gibibytes;
-typedef LLUnit<LLUnits::Bits, F64>			Bits;
-typedef LLUnit<LLUnits::Kibibits, F64>		Kibibits;
-typedef LLUnit<LLUnits::Mibibits, F64>		Mibibits;
-typedef LLUnit<LLUnits::Gibibits, F64>		Gibibits;
-
-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;
+typedef LLUnit<F64, LLUnits::Bytes>			Bytes;
+typedef LLUnit<F64, LLUnits::Kibibytes>		Kibibytes;
+typedef LLUnit<F64, LLUnits::Mibibytes>		Mibibytes;
+typedef LLUnit<F64, LLUnits::Gibibytes>		Gibibytes;
+typedef LLUnit<F64, LLUnits::Bits>			Bits;
+typedef LLUnit<F64, LLUnits::Kibibits>		Kibibits;
+typedef LLUnit<F64, LLUnits::Mibibits>		Mibibits;
+typedef LLUnit<F64, LLUnits::Gibibits>		Gibibits;
+
+typedef LLUnit<F64, LLUnits::Seconds>		Seconds;
+typedef LLUnit<F64, LLUnits::Milliseconds>	Milliseconds;
+typedef LLUnit<F64, LLUnits::Minutes>		Minutes;
+typedef LLUnit<F64, LLUnits::Hours>			Hours;
+typedef LLUnit<F64, LLUnits::Milliseconds>	Milliseconds;
+typedef LLUnit<F64, LLUnits::Microseconds>	Microseconds;
+typedef LLUnit<F64, LLUnits::Nanoseconds>	Nanoseconds;
+
+typedef LLUnit<F64, LLUnits::Meters>		Meters;
+typedef LLUnit<F64, LLUnits::Kilometers>	Kilometers;
+typedef LLUnit<F64, LLUnits::Centimeters>	Centimeters;
+typedef LLUnit<F64, LLUnits::Millimeters>	Millimeters;
+
+
+template<typename T>
+T storage_value(T val) { return val; }
+
+template<typename UNIT_TYPE, typename STORAGE_TYPE> 
+STORAGE_TYPE storage_value(LLUnit<STORAGE_TYPE, UNIT_TYPE> val) { return val.value(); }
+
+template<typename UNIT_TYPE, typename STORAGE_TYPE> 
+STORAGE_TYPE storage_value(LLUnitImplicit<STORAGE_TYPE, UNIT_TYPE> val) { return val.value(); }
 
 void init();
 void cleanup();
@@ -147,12 +157,12 @@ public:
 		}
 	}
 
-	void flush()
+	void flush(LLUnitImplicit<F64, LLUnits::Seconds> time_stamp)
 	{
 		llassert(mStorageSize >= sNextStorageSlot);
 		for (size_t i = 0; i < sNextStorageSlot; i++)
 		{
-			mStorage[i].flush();
+			mStorage[i].flush(time_stamp);
 		}
 	}
 
@@ -217,6 +227,11 @@ public:
 
 	size_t size() const
 	{
+		return getNumIndices();
+	}
+
+	static size_t getNumIndices() 
+	{
 		return sNextStorageSlot;
 	}
 
@@ -263,6 +278,7 @@ public:
 	}
 
 	size_t getIndex() const { return mAccumulatorIndex; }
+	static size_t getNumIndices() { return AccumulatorBuffer<ACCUMULATOR>::getNumIndices(); }
 
 	virtual const char* getUnitLabel() { return ""; }
 
@@ -364,7 +380,7 @@ public:
 		mLastValue = other ? other->mLastValue : 0;
 	}
 
-	void flush() {}
+	void flush(LLUnitImplicit<F64, LLUnits::Seconds>) {}
 
 	F64	getSum() const { return mSum; }
 	F64	getMin() const { return mMin; }
@@ -408,8 +424,8 @@ public:
 
 	void sample(F64 value)
 	{
-		LLUnitImplicit<LLUnits::Seconds, F64> time_stamp = LLTimer::getTotalSeconds();
-		LLUnitImplicit<LLUnits::Seconds, F64> delta_time = time_stamp - mLastSampleTimeStamp;
+		LLUnitImplicit<F64, LLUnits::Seconds> time_stamp = LLTimer::getTotalSeconds();
+		LLUnitImplicit<F64, LLUnits::Seconds> delta_time = time_stamp - mLastSampleTimeStamp;
 		mLastSampleTimeStamp = time_stamp;
 
 		if (mHasValue)
@@ -496,10 +512,9 @@ public:
 		mHasValue = other ? other->mHasValue : false;
 	}
 
-	void flush()
+	void flush(LLUnitImplicit<F64, LLUnits::Seconds> time_stamp)
 	{
-		LLUnitImplicit<LLUnits::Seconds, F64> time_stamp = LLTimer::getTotalSeconds();
-		LLUnitImplicit<LLUnits::Seconds, F64> delta_time = time_stamp - mLastSampleTimeStamp;
+		LLUnitImplicit<F64, LLUnits::Seconds> delta_time = time_stamp - mLastSampleTimeStamp;
 
 		if (mHasValue)
 		{
@@ -528,7 +543,7 @@ private:
 	F64	mMean,
 		mVarianceSum;
 
-	LLUnitImplicit<LLUnits::Seconds, F64>	mLastSampleTimeStamp,
+	LLUnitImplicit<F64, LLUnits::Seconds>	mLastSampleTimeStamp,
 											mTotalSamplingTime;
 
 	U32	mNumSamples;
@@ -563,7 +578,7 @@ public:
 		mSum = 0;
 	}
 
-	void flush() {}
+	void flush(LLUnitImplicit<F64, LLUnits::Seconds>) {}
 
 	F64	getSum() const { return mSum; }
 
@@ -578,8 +593,8 @@ private:
 class TimeBlockAccumulator
 {
 public:
-	typedef LLUnit<LLUnits::Seconds, F64> value_t;
-	typedef LLUnit<LLUnits::Seconds, F64> mean_t;
+	typedef LLUnit<F64, LLUnits::Seconds> value_t;
+	typedef LLUnit<F64, LLUnits::Seconds> mean_t;
 	typedef TimeBlockAccumulator self_t;
 
 	// fake classes that allows us to view different facets of underlying statistic
@@ -591,14 +606,14 @@ public:
 
 	struct SelfTimeFacet
 	{
-		typedef LLUnit<LLUnits::Seconds, F64> value_t;
-		typedef LLUnit<LLUnits::Seconds, F64> mean_t;
+		typedef LLUnit<F64, LLUnits::Seconds> value_t;
+		typedef LLUnit<F64, LLUnits::Seconds> mean_t;
 	};
 
 	TimeBlockAccumulator();
 	void addSamples(const self_t& other, bool /*append*/);
 	void reset(const self_t* other);
-	void flush() {}
+	void flush(LLUnitImplicit<F64, LLUnits::Seconds>) {}
 
 	//
 	// members
@@ -672,7 +687,7 @@ template<typename T, typename VALUE_T>
 void record(EventStatHandle<T>& measurement, VALUE_T value)
 {
 	T converted_value(value);
-	measurement.getPrimaryAccumulator()->record(LLUnits::rawValue(converted_value));
+	measurement.getPrimaryAccumulator()->record(storage_value(converted_value));
 }
 
 template <typename T = F64>
@@ -694,7 +709,7 @@ template<typename T, typename VALUE_T>
 void sample(SampleStatHandle<T>& measurement, VALUE_T value)
 {
 	T converted_value(value);
-	measurement.getPrimaryAccumulator()->sample(LLUnits::rawValue(converted_value));
+	measurement.getPrimaryAccumulator()->sample(storage_value(converted_value));
 }
 
 template <typename T = F64>
@@ -716,7 +731,7 @@ template<typename T, typename VALUE_T>
 void add(CountStatHandle<T>& count, VALUE_T value)
 {
 	T converted_value(value);
-	count.getPrimaryAccumulator()->add(LLUnits::rawValue(converted_value));
+	count.getPrimaryAccumulator()->add(storage_value(converted_value));
 }
 
 
@@ -739,8 +754,8 @@ struct MemStatAccumulator
 
 	struct ChildMemFacet
 	{
-		typedef LLUnit<LLUnits::Bytes, F64> value_t;
-		typedef LLUnit<LLUnits::Bytes, F64> mean_t;
+		typedef LLUnit<F64, LLUnits::Bytes> value_t;
+		typedef LLUnit<F64, LLUnits::Bytes> mean_t;
 	};
 
 	MemStatAccumulator()
@@ -764,10 +779,10 @@ struct MemStatAccumulator
 		mDeallocatedCount = 0;
 	}
 
-	void flush() 
+	void flush(LLUnitImplicit<F64, LLUnits::Seconds> time_stamp) 
 	{
-		mSize.flush();
-		mChildSize.flush();
+		mSize.flush(time_stamp);
+		mChildSize.flush(time_stamp);
 	}
 
 	SampleAccumulator	mSize,
diff --git a/indra/llcommon/lltracerecording.cpp b/indra/llcommon/lltracerecording.cpp
index d32504b014..33002929ea 100644
--- a/indra/llcommon/lltracerecording.cpp
+++ b/indra/llcommon/lltracerecording.cpp
@@ -107,7 +107,9 @@ void RecordingBuffers::reset(RecordingBuffers* other)
 
 void RecordingBuffers::flush()
 {
-	mSamples.flush();
+	LLUnitImplicit<F64, LLUnits::Seconds> time_stamp = LLTimer::getTotalSeconds();
+
+	mSamples.flush(time_stamp);
 }
 
 ///////////////////////////////////////////////////////////////////////
@@ -122,21 +124,26 @@ Recording::Recording()
 
 Recording::Recording( const Recording& other )
 {
+	*this = other;
+}
+
+Recording& Recording::operator = (const Recording& other)
+{
 	// this will allow us to seamlessly start without affecting any data we've acquired from other
 	setPlayState(PAUSED);
 
 	Recording& mutable_other = const_cast<Recording&>(other);
+	mutable_other.update();
 	EPlayState other_play_state = other.getPlayState();
-	mutable_other.pause();
 
-	mBuffers = other.mBuffers;
+	mBuffers = mutable_other.mBuffers;
 
 	LLStopWatchControlsMixin<Recording>::setPlayState(other_play_state);
-	mutable_other.setPlayState(other_play_state);
 
 	// above call will clear mElapsedSeconds as a side effect, so copy it here
 	mElapsedSeconds = other.mElapsedSeconds;
 	mSamplingTimer = other.mSamplingTimer;
+	return *this;
 }
 
 
@@ -186,240 +193,198 @@ void Recording::handleSplitTo(Recording& other)
 
 void Recording::appendRecording( const Recording& other )
 {
-	EPlayState play_state = getPlayState();
-	{
-		pause();
-		mBuffers.write()->append(*other.mBuffers);
-		mElapsedSeconds += other.mElapsedSeconds;
-	}
-	setPlayState(play_state);
+	update();
+	mBuffers.write()->append(*other.mBuffers);
+	mElapsedSeconds += other.mElapsedSeconds;
 }
 
 void Recording::mergeRecording( const Recording& other)
 {
-	EPlayState play_state = getPlayState();
-	{
-		pause();
-		mBuffers.write()->merge(*other.mBuffers);
-	}
-	setPlayState(play_state);
+	update();
+	mBuffers.write()->merge(*other.mBuffers);
 }
 
-LLUnit<LLUnits::Seconds, F64> Recording::getSum(const TraceType<TimeBlockAccumulator>& stat)
+LLUnit<F64, LLUnits::Seconds> Recording::getSum(const TraceType<TimeBlockAccumulator>& stat)
 {
 	const TimeBlockAccumulator& accumulator = mBuffers->mStackTimers[stat.getIndex()];
-	update();
 	return (F64)(accumulator.mTotalTimeCounter - accumulator.mStartTotalTimeCounter) 
 				/ (F64)LLTrace::TimeBlock::countsPerSecond();
 }
 
-LLUnit<LLUnits::Seconds, F64> Recording::getSum(const TraceType<TimeBlockAccumulator::SelfTimeFacet>& stat)
+LLUnit<F64, LLUnits::Seconds> Recording::getSum(const TraceType<TimeBlockAccumulator::SelfTimeFacet>& stat)
 {
 	const TimeBlockAccumulator& accumulator = mBuffers->mStackTimers[stat.getIndex()];
-	update();
 	return (F64)(accumulator.mSelfTimeCounter) / (F64)LLTrace::TimeBlock::countsPerSecond();
 }
 
 
 U32 Recording::getSum(const TraceType<TimeBlockAccumulator::CallCountFacet>& stat)
 {
-	update();
 	return mBuffers->mStackTimers[stat.getIndex()].mCalls;
 }
 
-LLUnit<LLUnits::Seconds, F64> Recording::getPerSec(const TraceType<TimeBlockAccumulator>& stat)
+LLUnit<F64, LLUnits::Seconds> Recording::getPerSec(const TraceType<TimeBlockAccumulator>& stat)
 {
 	const TimeBlockAccumulator& accumulator = mBuffers->mStackTimers[stat.getIndex()];
 
-	update();
 	return (F64)(accumulator.mTotalTimeCounter - accumulator.mStartTotalTimeCounter) 
-				/ ((F64)LLTrace::TimeBlock::countsPerSecond() * mElapsedSeconds);
+				/ ((F64)LLTrace::TimeBlock::countsPerSecond() * mElapsedSeconds.value());
 }
 
-LLUnit<LLUnits::Seconds, F64> Recording::getPerSec(const TraceType<TimeBlockAccumulator::SelfTimeFacet>& stat)
+LLUnit<F64, LLUnits::Seconds> Recording::getPerSec(const TraceType<TimeBlockAccumulator::SelfTimeFacet>& stat)
 {
 	const TimeBlockAccumulator& accumulator = mBuffers->mStackTimers[stat.getIndex()];
 
-	update();
 	return (F64)(accumulator.mSelfTimeCounter) 
-			/ ((F64)LLTrace::TimeBlock::countsPerSecond() * mElapsedSeconds);
+			/ ((F64)LLTrace::TimeBlock::countsPerSecond() * mElapsedSeconds.value());
 }
 
 F32 Recording::getPerSec(const TraceType<TimeBlockAccumulator::CallCountFacet>& stat)
 {
-	update();
-	return (F32)mBuffers->mStackTimers[stat.getIndex()].mCalls / mElapsedSeconds;
+	return (F32)mBuffers->mStackTimers[stat.getIndex()].mCalls / mElapsedSeconds.value();
 }
 
-LLUnit<LLUnits::Bytes, F64> Recording::getMin(const TraceType<MemStatAccumulator>& stat)
+LLUnit<F64, LLUnits::Bytes> Recording::getMin(const TraceType<MemStatAccumulator>& stat)
 {
-	update();
 	return mBuffers->mMemStats[stat.getIndex()].mSize.getMin();
 }
 
-LLUnit<LLUnits::Bytes, F64> Recording::getMean(const TraceType<MemStatAccumulator>& stat)
+LLUnit<F64, LLUnits::Bytes> Recording::getMean(const TraceType<MemStatAccumulator>& stat)
 {
-	update();
 	return mBuffers->mMemStats[stat.getIndex()].mSize.getMean();
 }
 
-LLUnit<LLUnits::Bytes, F64> Recording::getMax(const TraceType<MemStatAccumulator>& stat)
+LLUnit<F64, LLUnits::Bytes> Recording::getMax(const TraceType<MemStatAccumulator>& stat)
 {
-	update();
 	return mBuffers->mMemStats[stat.getIndex()].mSize.getMax();
 }
 
-LLUnit<LLUnits::Bytes, F64> Recording::getStandardDeviation(const TraceType<MemStatAccumulator>& stat)
+LLUnit<F64, LLUnits::Bytes> Recording::getStandardDeviation(const TraceType<MemStatAccumulator>& stat)
 {
-	update();
 	return mBuffers->mMemStats[stat.getIndex()].mSize.getStandardDeviation();
 }
 
-LLUnit<LLUnits::Bytes, F64> Recording::getLastValue(const TraceType<MemStatAccumulator>& stat)
+LLUnit<F64, LLUnits::Bytes> Recording::getLastValue(const TraceType<MemStatAccumulator>& stat)
 {
-	update();
 	return mBuffers->mMemStats[stat.getIndex()].mSize.getLastValue();
 }
 
-LLUnit<LLUnits::Bytes, F64> Recording::getMin(const TraceType<MemStatAccumulator::ChildMemFacet>& stat)
+LLUnit<F64, LLUnits::Bytes> Recording::getMin(const TraceType<MemStatAccumulator::ChildMemFacet>& stat)
 {
-	update();
 	return mBuffers->mMemStats[stat.getIndex()].mChildSize.getMin();
 }
 
-LLUnit<LLUnits::Bytes, F64> Recording::getMean(const TraceType<MemStatAccumulator::ChildMemFacet>& stat)
+LLUnit<F64, LLUnits::Bytes> Recording::getMean(const TraceType<MemStatAccumulator::ChildMemFacet>& stat)
 {
-	update();
 	return mBuffers->mMemStats[stat.getIndex()].mChildSize.getMean();
 }
 
-LLUnit<LLUnits::Bytes, F64> Recording::getMax(const TraceType<MemStatAccumulator::ChildMemFacet>& stat)
+LLUnit<F64, LLUnits::Bytes> Recording::getMax(const TraceType<MemStatAccumulator::ChildMemFacet>& stat)
 {
-	update();
 	return mBuffers->mMemStats[stat.getIndex()].mChildSize.getMax();
 }
 
-LLUnit<LLUnits::Bytes, F64> Recording::getStandardDeviation(const TraceType<MemStatAccumulator::ChildMemFacet>& stat)
+LLUnit<F64, LLUnits::Bytes> Recording::getStandardDeviation(const TraceType<MemStatAccumulator::ChildMemFacet>& stat)
 {
-	update();
 	return mBuffers->mMemStats[stat.getIndex()].mChildSize.getStandardDeviation();
 }
 
-LLUnit<LLUnits::Bytes, F64> Recording::getLastValue(const TraceType<MemStatAccumulator::ChildMemFacet>& stat)
+LLUnit<F64, LLUnits::Bytes> Recording::getLastValue(const TraceType<MemStatAccumulator::ChildMemFacet>& stat)
 {
-	update();
 	return mBuffers->mMemStats[stat.getIndex()].mChildSize.getLastValue();
 }
 
 U32 Recording::getSum(const TraceType<MemStatAccumulator::AllocationCountFacet>& stat)
 {
-	update();
 	return mBuffers->mMemStats[stat.getIndex()].mAllocatedCount;
 }
 
 U32 Recording::getSum(const TraceType<MemStatAccumulator::DeallocationCountFacet>& stat)
 {
-	update();
 	return mBuffers->mMemStats[stat.getIndex()].mAllocatedCount;
 }
 
 
 F64 Recording::getSum( const TraceType<CountAccumulator>& stat )
 {
-	update();
 	return mBuffers->mCounts[stat.getIndex()].getSum();
 }
 
 F64 Recording::getSum( const TraceType<EventAccumulator>& stat )
 {
-	update();
 	return (F64)mBuffers->mEvents[stat.getIndex()].getSum();
 }
 
 F64 Recording::getPerSec( const TraceType<CountAccumulator>& stat )
 {
-	update();
 	F64 sum = mBuffers->mCounts[stat.getIndex()].getSum();
 	return  (sum != 0.0) 
-		? (sum / mElapsedSeconds)
+		? (sum / mElapsedSeconds.value())
 		: 0.0;
 }
 
 U32 Recording::getSampleCount( const TraceType<CountAccumulator>& stat )
 {
-	update();
 	return mBuffers->mCounts[stat.getIndex()].getSampleCount();
 }
 
 F64 Recording::getMin( const TraceType<SampleAccumulator>& stat )
 {
-	update();
 	return mBuffers->mSamples[stat.getIndex()].getMin();
 }
 
 F64 Recording::getMax( const TraceType<SampleAccumulator>& stat )
 {
-	update();
 	return mBuffers->mSamples[stat.getIndex()].getMax();
 }
 
 F64 Recording::getMean( const TraceType<SampleAccumulator>& stat )
 {
-	update();
 	return mBuffers->mSamples[stat.getIndex()].getMean();
 }
 
 F64 Recording::getStandardDeviation( const TraceType<SampleAccumulator>& stat )
 {
-	update();
 	return mBuffers->mSamples[stat.getIndex()].getStandardDeviation();
 }
 
 F64 Recording::getLastValue( const TraceType<SampleAccumulator>& stat )
 {
-	update();
 	return mBuffers->mSamples[stat.getIndex()].getLastValue();
 }
 
 U32 Recording::getSampleCount( const TraceType<SampleAccumulator>& stat )
 {
-	update();
 	return mBuffers->mSamples[stat.getIndex()].getSampleCount();
 }
 
 F64 Recording::getMin( const TraceType<EventAccumulator>& stat )
 {
-	update();
 	return mBuffers->mEvents[stat.getIndex()].getMin();
 }
 
 F64 Recording::getMax( const TraceType<EventAccumulator>& stat )
 {
-	update();
 	return mBuffers->mEvents[stat.getIndex()].getMax();
 }
 
 F64 Recording::getMean( const TraceType<EventAccumulator>& stat )
 {
-	update();
 	return mBuffers->mEvents[stat.getIndex()].getMean();
 }
 
 F64 Recording::getStandardDeviation( const TraceType<EventAccumulator>& stat )
 {
-	update();
 	return mBuffers->mEvents[stat.getIndex()].getStandardDeviation();
 }
 
 F64 Recording::getLastValue( const TraceType<EventAccumulator>& stat )
 {
-	update();
 	return mBuffers->mEvents[stat.getIndex()].getLastValue();
 }
 
 U32 Recording::getSampleCount( const TraceType<EventAccumulator>& stat )
 {
-	update();
 	return mBuffers->mEvents[stat.getIndex()].getSampleCount();
 }
 
@@ -430,6 +395,7 @@ U32 Recording::getSampleCount( const TraceType<EventAccumulator>& stat )
 PeriodicRecording::PeriodicRecording( U32 num_periods, EPlayState state) 
 :	mAutoResize(num_periods == 0),
 	mCurPeriod(0),
+	mNumPeriods(0),
 	mRecordingPeriods(num_periods ? num_periods : 1)
 {
 	setPlayState(state);
@@ -443,9 +409,16 @@ void PeriodicRecording::nextPeriod()
 	}
 
 	Recording& old_recording = getCurRecording();
-
 	mCurPeriod = (mCurPeriod + 1) % mRecordingPeriods.size();
 	old_recording.splitTo(getCurRecording());
+
+	mNumPeriods = llmin(mRecordingPeriods.size(), mNumPeriods + 1);
+}
+
+void PeriodicRecording::appendRecording(Recording& recording)
+{
+	getCurRecording().appendRecording(recording);
+	nextPeriod();
 }
 
 
@@ -453,77 +426,71 @@ void PeriodicRecording::appendPeriodicRecording( PeriodicRecording& other )
 {
 	if (other.mRecordingPeriods.empty()) return;
 
-	EPlayState play_state = getPlayState();
-	pause();
-
-	EPlayState other_play_state = other.getPlayState();
-	other.pause();
-
-	U32 other_recording_count = other.mRecordingPeriods.size();
-
-	Recording& other_oldest_recording = other.mRecordingPeriods[(other.mCurPeriod + 1) % other.mRecordingPeriods.size()];
+	getCurRecording().update();
+	other.getCurRecording().update();
 
-	// if I have a recording of any length, then close it off and start a fresh one
-	if (getCurRecording().getDuration().value())
+	if (mAutoResize)
 	{
-		nextPeriod();
-	}
-	getCurRecording().appendRecording(other_oldest_recording);
+		S32 other_index = (other.mCurPeriod + 1) % other.mRecordingPeriods.size();
+		S32 end_index = (other.mCurPeriod) % other.mRecordingPeriods.size(); 
 
-	if (other_recording_count > 1)
-	{
-		if (mAutoResize)
+		do
 		{
-			for (S32 other_index = (other.mCurPeriod + 2) % other_recording_count,
-				end_index = (other.mCurPeriod + 1) % other_recording_count; 
-				other_index != end_index; 
-				other_index = (other_index + 1) % other_recording_count)
+			if (other.mRecordingPeriods[other_index].getDuration().value())
 			{
-				llassert(other.mRecordingPeriods[other_index].getDuration() != 0.f 
-							&& (mRecordingPeriods.empty() 
-								|| other.mRecordingPeriods[other_index].getDuration() != mRecordingPeriods.back().getDuration()));
 				mRecordingPeriods.push_back(other.mRecordingPeriods[other_index]);
 			}
-
-			mCurPeriod = mRecordingPeriods.size() - 1;
+			other_index = (other_index + 1) % other.mRecordingPeriods.size();
 		}
-		else
+		while(other_index != end_index);
+
+		mCurPeriod = mRecordingPeriods.size() - 1;
+		mNumPeriods = mRecordingPeriods.size();
+	}
+	else
+	{
+		//FIXME: get proper number of recordings from other...might not have used all its slots
+		size_t num_to_copy = llmin(	mRecordingPeriods.size(), other.getNumRecordedPeriods());
+		std::vector<Recording>::iterator src_it = other.mRecordingPeriods.begin() 
+													+ (	(other.mCurPeriod + 1									// oldest period
+															+ (other.mRecordingPeriods.size() - num_to_copy))	// minus room for copy
+														% other.mRecordingPeriods.size());
+		std::vector<Recording>::iterator dest_it = mRecordingPeriods.begin() + mCurPeriod;
+
+		for(size_t i = 0; i < num_to_copy; i++)
 		{
-			size_t num_to_copy = llmin(	mRecordingPeriods.size(), other.mRecordingPeriods.size() - 1);
-			std::vector<Recording>::iterator src_it = other.mRecordingPeriods.begin() 
-														+ (	(other.mCurPeriod + 1									// oldest period
-																+ (other.mRecordingPeriods.size() - num_to_copy))	// minus room for copy
-															% other.mRecordingPeriods.size());
-			std::vector<Recording>::iterator dest_it = mRecordingPeriods.begin() + ((mCurPeriod + 1) % mRecordingPeriods.size());
-
-			for(S32 i = 0; i < num_to_copy; i++)
-			{
-				*dest_it = *src_it;
+			*dest_it = *src_it;
 
-				if (++src_it == other.mRecordingPeriods.end())
-				{
-					src_it = other.mRecordingPeriods.begin();
-				}
+			if (++src_it == other.mRecordingPeriods.end())
+			{
+				src_it = other.mRecordingPeriods.begin();
+			}
 
-				if (++dest_it == mRecordingPeriods.end())
-				{
-					dest_it = mRecordingPeriods.begin();
-				}
+			if (++dest_it == mRecordingPeriods.end())
+			{
+				dest_it = mRecordingPeriods.begin();
 			}
-		
-			mCurPeriod = (mCurPeriod + num_to_copy) % mRecordingPeriods.size();
 		}
+		
+		// want argument to % to be positive, otherwise result could be negative and thus out of bounds
+		llassert(num_to_copy >= 1);
+		// advance to last recording period copied, so we can check if the last period had actually carried any data, in which case we'll advance below
+		// using nextPeriod() which retains continuity (mLastValue, etc)
+		mCurPeriod = (mCurPeriod + num_to_copy - 1) % mRecordingPeriods.size();
+		mNumPeriods = llmin(mRecordingPeriods.size(), mNumPeriods + num_to_copy);
 	}
 
-	nextPeriod();
-
-	setPlayState(play_state);
-	other.setPlayState(other_play_state);
+	if (getCurRecording().getDuration().value())
+	{
+		//call this to chain last period copied to new active period
+		nextPeriod();
+	}
+	getCurRecording().setPlayState(getPlayState());
 }
 
-LLUnit<LLUnits::Seconds, F64> PeriodicRecording::getDuration() const
+LLUnit<F64, LLUnits::Seconds> PeriodicRecording::getDuration() const
 {
-	LLUnit<LLUnits::Seconds, F64> duration;
+	LLUnit<F64, LLUnits::Seconds> duration;
 	size_t num_periods = mRecordingPeriods.size();
 	for (size_t i = 1; i <= num_periods; i++)
 	{
@@ -615,7 +582,7 @@ void PeriodicRecording::handleSplitTo(PeriodicRecording& other)
 F64 PeriodicRecording::getPeriodMean( const TraceType<EventAccumulator>& stat, size_t num_periods /*= U32_MAX*/ )
 {
 	size_t total_periods = mRecordingPeriods.size();
-	num_periods = llmin(num_periods, total_periods);
+	num_periods = llmin(num_periods, isStarted() ? total_periods - 1 : total_periods);
 
 	F64 mean = 0;
 	if (num_periods <= 0) { return mean; }
@@ -643,7 +610,7 @@ F64 PeriodicRecording::getPeriodMean( const TraceType<EventAccumulator>& stat, s
 F64 PeriodicRecording::getPeriodMin( const TraceType<EventAccumulator>& stat, size_t num_periods /*= U32_MAX*/ )
 {
 	size_t total_periods = mRecordingPeriods.size();
-	num_periods = llmin(num_periods, total_periods);
+	num_periods = llmin(num_periods, isStarted() ? total_periods - 1 : total_periods);
 
 	F64 min_val = std::numeric_limits<F64>::max();
 	for (S32 i = 1; i <= num_periods; i++)
@@ -657,7 +624,7 @@ F64 PeriodicRecording::getPeriodMin( const TraceType<EventAccumulator>& stat, si
 F64 PeriodicRecording::getPeriodMax( const TraceType<EventAccumulator>& stat, size_t num_periods /*= U32_MAX*/ )
 {
 	size_t total_periods = mRecordingPeriods.size();
-	num_periods = llmin(num_periods, total_periods);
+	num_periods = llmin(num_periods, isStarted() ? total_periods - 1 : total_periods);
 
 	F64 max_val = std::numeric_limits<F64>::min();
 	for (S32 i = 1; i <= num_periods; i++)
@@ -671,7 +638,7 @@ F64 PeriodicRecording::getPeriodMax( const TraceType<EventAccumulator>& stat, si
 F64 PeriodicRecording::getPeriodMin( const TraceType<SampleAccumulator>& stat, size_t num_periods /*= U32_MAX*/ )
 {
 	size_t total_periods = mRecordingPeriods.size();
-	num_periods = llmin(num_periods, total_periods);
+	num_periods = llmin(num_periods, isStarted() ? total_periods - 1 : total_periods);
 
 	F64 min_val = std::numeric_limits<F64>::max();
 	for (S32 i = 1; i <= num_periods; i++)
@@ -685,7 +652,7 @@ F64 PeriodicRecording::getPeriodMin( const TraceType<SampleAccumulator>& stat, s
 F64 PeriodicRecording::getPeriodMax(const TraceType<SampleAccumulator>& stat, size_t num_periods /*= U32_MAX*/)
 {
 	size_t total_periods = mRecordingPeriods.size();
-	num_periods = llmin(num_periods, total_periods);
+	num_periods = llmin(num_periods, isStarted() ? total_periods - 1 : total_periods);
 
 	F64 max_val = std::numeric_limits<F64>::min();
 	for (S32 i = 1; i <= num_periods; i++)
@@ -700,9 +667,9 @@ F64 PeriodicRecording::getPeriodMax(const TraceType<SampleAccumulator>& stat, si
 F64 PeriodicRecording::getPeriodMean( const TraceType<SampleAccumulator>& stat, size_t num_periods /*= U32_MAX*/ )
 {
 	size_t total_periods = mRecordingPeriods.size();
-	num_periods = llmin(num_periods, total_periods);
+	num_periods = llmin(num_periods, isStarted() ? total_periods - 1 : total_periods);
 
-	LLUnit<LLUnits::Seconds, F64> total_duration = 0.f;
+	LLUnit<F64, LLUnits::Seconds> total_duration = 0.f;
 
 	F64 mean = 0;
 	if (num_periods <= 0) { return mean; }
@@ -712,7 +679,7 @@ F64 PeriodicRecording::getPeriodMean( const TraceType<SampleAccumulator>& stat,
 		S32 index = (mCurPeriod + total_periods - i) % total_periods;
 		if (mRecordingPeriods[index].getDuration() > 0.f)
 		{
-			LLUnit<LLUnits::Seconds, F64> recording_duration = mRecordingPeriods[index].getDuration();
+			LLUnit<F64, LLUnits::Seconds> recording_duration = mRecordingPeriods[index].getDuration();
 			mean += mRecordingPeriods[index].getMean(stat) * recording_duration.value();
 			total_duration += recording_duration;
 		}
@@ -734,13 +701,11 @@ F64 PeriodicRecording::getPeriodMean( const TraceType<SampleAccumulator>& stat,
 void ExtendableRecording::extend()
 {
 	// stop recording to get latest data
-	mPotentialRecording.stop();
+	mPotentialRecording.update();
 	// push the data back to accepted recording
 	mAcceptedRecording.appendRecording(mPotentialRecording);
 	// flush data, so we can start from scratch
 	mPotentialRecording.reset();
-	// go back to play state we were in initially
-	mPotentialRecording.setPlayState(getPlayState());
 }
 
 void ExtendableRecording::handleStart()
@@ -777,15 +742,10 @@ ExtendablePeriodicRecording::ExtendablePeriodicRecording()
 
 void ExtendablePeriodicRecording::extend()
 {
-	llassert(mPotentialRecording.getPlayState() == getPlayState());
-	// stop recording to get latest data
-	mPotentialRecording.pause();
 	// push the data back to accepted recording
 	mAcceptedRecording.appendPeriodicRecording(mPotentialRecording);
 	// flush data, so we can start from scratch
 	mPotentialRecording.reset();
-	// go back to play state we were in initially
-	mPotentialRecording.setPlayState(getPlayState());
 }
 
 
diff --git a/indra/llcommon/lltracerecording.h b/indra/llcommon/lltracerecording.h
index 4651bfcb61..b839e85de0 100644
--- a/indra/llcommon/lltracerecording.h
+++ b/indra/llcommon/lltracerecording.h
@@ -135,6 +135,8 @@ namespace LLTrace
 		Recording(const Recording& other);
 		~Recording();
 
+		Recording& operator = (const Recording& other);
+
 		// accumulate data from subsequent, non-overlapping recording
 		void appendRecording(const Recording& other);
 
@@ -148,26 +150,26 @@ namespace LLTrace
 		void makeUnique() { mBuffers.makeUnique(); }
 
 		// Timer accessors
-		LLUnit<LLUnits::Seconds, F64> getSum(const TraceType<TimeBlockAccumulator>& stat);
-		LLUnit<LLUnits::Seconds, F64> getSum(const TraceType<TimeBlockAccumulator::SelfTimeFacet>& stat);
+		LLUnit<F64, LLUnits::Seconds> getSum(const TraceType<TimeBlockAccumulator>& stat);
+		LLUnit<F64, LLUnits::Seconds> getSum(const TraceType<TimeBlockAccumulator::SelfTimeFacet>& stat);
 		U32 getSum(const TraceType<TimeBlockAccumulator::CallCountFacet>& stat);
 
-		LLUnit<LLUnits::Seconds, F64> getPerSec(const TraceType<TimeBlockAccumulator>& stat);
-		LLUnit<LLUnits::Seconds, F64> getPerSec(const TraceType<TimeBlockAccumulator::SelfTimeFacet>& stat);
+		LLUnit<F64, LLUnits::Seconds> getPerSec(const TraceType<TimeBlockAccumulator>& stat);
+		LLUnit<F64, LLUnits::Seconds> getPerSec(const TraceType<TimeBlockAccumulator::SelfTimeFacet>& stat);
 		F32 getPerSec(const TraceType<TimeBlockAccumulator::CallCountFacet>& stat);
 
 		// Memory accessors
-		LLUnit<LLUnits::Bytes, F64> getMin(const TraceType<MemStatAccumulator>& stat);
-		LLUnit<LLUnits::Bytes, F64> getMean(const TraceType<MemStatAccumulator>& stat);
-		LLUnit<LLUnits::Bytes, F64> getMax(const TraceType<MemStatAccumulator>& stat);
-		LLUnit<LLUnits::Bytes, F64> getStandardDeviation(const TraceType<MemStatAccumulator>& stat);
-		LLUnit<LLUnits::Bytes, F64> getLastValue(const TraceType<MemStatAccumulator>& stat);
-
-		LLUnit<LLUnits::Bytes, F64> getMin(const TraceType<MemStatAccumulator::ChildMemFacet>& stat);
-		LLUnit<LLUnits::Bytes, F64> getMean(const TraceType<MemStatAccumulator::ChildMemFacet>& stat);
-		LLUnit<LLUnits::Bytes, F64> getMax(const TraceType<MemStatAccumulator::ChildMemFacet>& stat);
-		LLUnit<LLUnits::Bytes, F64> getStandardDeviation(const TraceType<MemStatAccumulator::ChildMemFacet>& stat);
-		LLUnit<LLUnits::Bytes, F64> getLastValue(const TraceType<MemStatAccumulator::ChildMemFacet>& stat);
+		LLUnit<F64, LLUnits::Bytes> getMin(const TraceType<MemStatAccumulator>& stat);
+		LLUnit<F64, LLUnits::Bytes> getMean(const TraceType<MemStatAccumulator>& stat);
+		LLUnit<F64, LLUnits::Bytes> getMax(const TraceType<MemStatAccumulator>& stat);
+		LLUnit<F64, LLUnits::Bytes> getStandardDeviation(const TraceType<MemStatAccumulator>& stat);
+		LLUnit<F64, LLUnits::Bytes> getLastValue(const TraceType<MemStatAccumulator>& stat);
+
+		LLUnit<F64, LLUnits::Bytes> getMin(const TraceType<MemStatAccumulator::ChildMemFacet>& stat);
+		LLUnit<F64, LLUnits::Bytes> getMean(const TraceType<MemStatAccumulator::ChildMemFacet>& stat);
+		LLUnit<F64, LLUnits::Bytes> getMax(const TraceType<MemStatAccumulator::ChildMemFacet>& stat);
+		LLUnit<F64, LLUnits::Bytes> getStandardDeviation(const TraceType<MemStatAccumulator::ChildMemFacet>& stat);
+		LLUnit<F64, LLUnits::Bytes> getLastValue(const TraceType<MemStatAccumulator::ChildMemFacet>& stat);
 
 		U32 getSum(const TraceType<MemStatAccumulator::AllocationCountFacet>& stat);
 		U32 getSum(const TraceType<MemStatAccumulator::DeallocationCountFacet>& stat);
@@ -273,7 +275,7 @@ namespace LLTrace
 
 		U32 getSampleCount(const TraceType<EventAccumulator>& stat);
 
-		LLUnit<LLUnits::Seconds, F64> getDuration() const { return LLUnit<LLUnits::Seconds, F64>(mElapsedSeconds); }
+		LLUnit<F64, LLUnits::Seconds> getDuration() const { return mElapsedSeconds; }
 
 	protected:
 		friend class ThreadRecorder;
@@ -288,7 +290,7 @@ namespace LLTrace
 		class ThreadRecorder* getThreadRecorder(); 
 
 		LLTimer				mSamplingTimer;
-		F64					mElapsedSeconds;
+		LLUnit<F64, LLUnits::Seconds>			mElapsedSeconds;
 		LLCopyOnWritePointer<RecordingBuffers>	mBuffers;
 	};
 
@@ -299,11 +301,12 @@ namespace LLTrace
 		PeriodicRecording(U32 num_periods, EPlayState state = STOPPED);
 
 		void nextPeriod();
-		U32 getNumPeriods() { return mRecordingPeriods.size(); }
+		size_t getNumRecordedPeriods() { return mNumPeriods; }
 
-		LLUnit<LLUnits::Seconds, F64> getDuration() const;
+		LLUnit<F64, LLUnits::Seconds> getDuration() const;
 
 		void appendPeriodicRecording(PeriodicRecording& other);
+		void appendRecording(Recording& recording);
 		Recording& getLastRecording();
 		const Recording& getLastRecording() const;
 		Recording& getCurRecording();
@@ -317,7 +320,7 @@ namespace LLTrace
 		typename T::value_t getPeriodMin(const TraceType<T>& stat, size_t num_periods = U32_MAX)
 		{
 			size_t total_periods = mRecordingPeriods.size();
-			num_periods = llmin(num_periods, total_periods);
+			num_periods = llmin(num_periods, isStarted() ? total_periods - 1 : total_periods);
 
 			typename T::value_t min_val = std::numeric_limits<typename T::value_t>::max();
 			for (S32 i = 1; i <= num_periods; i++)
@@ -346,7 +349,7 @@ namespace LLTrace
 		F64 getPeriodMinPerSec(const TraceType<T>& stat, size_t num_periods = U32_MAX)
 		{
 			size_t total_periods = mRecordingPeriods.size();
-			num_periods = llmin(num_periods, total_periods);
+			num_periods = llmin(num_periods, isStarted() ? total_periods - 1 : total_periods);
 
 			F64 min_val = std::numeric_limits<F64>::max();
 			for (S32 i = 1; i <= num_periods; i++)
@@ -362,7 +365,7 @@ namespace LLTrace
 		typename T::value_t getPeriodMax(const TraceType<T>& stat, size_t num_periods = U32_MAX)
 		{
 			size_t total_periods = mRecordingPeriods.size();
-			num_periods = llmin(num_periods, total_periods);
+			num_periods = llmin(num_periods, isStarted() ? total_periods - 1 : total_periods);
 
 			typename T::value_t max_val = std::numeric_limits<typename T::value_t>::min();
 			for (S32 i = 1; i <= num_periods; i++)
@@ -391,7 +394,7 @@ namespace LLTrace
 		F64 getPeriodMaxPerSec(const TraceType<T>& stat, size_t num_periods = U32_MAX)
 		{
 			size_t total_periods = mRecordingPeriods.size();
-			num_periods = llmin(num_periods, total_periods);
+			num_periods = llmin(num_periods, isStarted() ? total_periods - 1 : total_periods);
 
 			F64 max_val = std::numeric_limits<F64>::min();
 			for (S32 i = 1; i <= num_periods; i++)
@@ -407,7 +410,7 @@ namespace LLTrace
 		typename T::mean_t getPeriodMean(const TraceType<T >& stat, size_t num_periods = U32_MAX)
 		{
 			size_t total_periods = mRecordingPeriods.size();
-			num_periods = llmin(num_periods, total_periods);
+			num_periods = llmin(num_periods, isStarted() ? total_periods - 1 : total_periods);
 
 			typename T::mean_t mean = 0;
 			if (num_periods <= 0) { return mean; }
@@ -442,7 +445,7 @@ namespace LLTrace
 		typename T::mean_t getPeriodMeanPerSec(const TraceType<T>& stat, size_t num_periods = U32_MAX)
 		{
 			size_t total_periods = mRecordingPeriods.size();
-			num_periods = llmin(num_periods, total_periods);
+			num_periods = llmin(num_periods, isStarted() ? total_periods - 1 : total_periods);
 
 			typename T::mean_t mean = 0;
 			if (num_periods <= 0) { return mean; }
@@ -468,8 +471,9 @@ namespace LLTrace
 
 	private:
 		std::vector<Recording>	mRecordingPeriods;
-		const bool	mAutoResize;
-		S32			mCurPeriod;
+		const bool				mAutoResize;
+		size_t					mCurPeriod;
+		size_t					mNumPeriods;
 	};
 
 	PeriodicRecording& get_frame_recording();
diff --git a/indra/llcommon/lltracethreadrecorder.cpp b/indra/llcommon/lltracethreadrecorder.cpp
index c1a0700eff..54006f4e5b 100644
--- a/indra/llcommon/lltracethreadrecorder.cpp
+++ b/indra/llcommon/lltracethreadrecorder.cpp
@@ -111,6 +111,7 @@ ThreadRecorder::active_recording_list_t::reverse_iterator ThreadRecorder::bringU
 	if (mActiveRecordings.empty()) return mActiveRecordings.rend();
 
 	mActiveRecordings.back()->mPartialRecording.flush();
+	TimeBlock::updateTimes();
 
 	active_recording_list_t::reverse_iterator it, end_it;
 	for (it = mActiveRecordings.rbegin(), end_it = mActiveRecordings.rend();
diff --git a/indra/llcommon/llunit.h b/indra/llcommon/llunit.h
index f48cbe0e11..5229fe69d7 100644
--- a/indra/llcommon/llunit.h
+++ b/indra/llcommon/llunit.h
@@ -30,36 +30,12 @@
 #include "stdtypes.h"
 #include "llpreprocessor.h"
 #include "llerrorlegacy.h"
+#include <boost/type_traits/is_same.hpp>
 
-namespace LLUnits
-{
-
-template<typename DERIVED_UNITS_TAG, typename BASE_UNITS_TAG, typename VALUE_TYPE>
-struct ConversionFactor
-{
-	static F64 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.");
-        return 0;
-	}
-};
-
-template<typename BASE_UNITS_TAG, typename VALUE_TYPE>
-struct ConversionFactor<BASE_UNITS_TAG, BASE_UNITS_TAG, VALUE_TYPE>
-{
-	static F64 get() 
-	{ 
-		return 1; 
-	}
-};
-
-}
-
-template<typename UNIT_TYPE, typename STORAGE_TYPE>
+template<typename STORAGE_TYPE, typename UNIT_TYPE>
 struct LLUnit
 {
-	typedef LLUnit<UNIT_TYPE, STORAGE_TYPE> self_t;
+	typedef LLUnit<STORAGE_TYPE, UNIT_TYPE> self_t;
 	typedef STORAGE_TYPE storage_t;
 
 	// value initialization
@@ -68,11 +44,16 @@ struct LLUnit
 	{}
 
 	// unit initialization and conversion
-	template<typename OTHER_UNIT, typename OTHER_STORAGE>
-	LLUnit(LLUnit<OTHER_UNIT, OTHER_STORAGE> other)
-	:	mValue(convert(other))
+	template<typename OTHER_STORAGE, typename OTHER_UNIT>
+	LLUnit(LLUnit<OTHER_STORAGE, OTHER_UNIT> other)
+	:	mValue(convert(other).mValue)
 	{}
 	
+	bool operator == (const self_t& other)
+	{
+		return mValue = other.mValue;
+	}
+
 	// value assignment
 	self_t& operator = (storage_t value)
 	{
@@ -81,10 +62,10 @@ struct LLUnit
 	}
 
 	// unit assignment
-	template<typename OTHER_UNIT, typename OTHER_STORAGE>
-	self_t& operator = (LLUnit<OTHER_UNIT, OTHER_STORAGE> other)
+	template<typename OTHER_STORAGE, typename OTHER_UNIT>
+	self_t& operator = (LLUnit<OTHER_STORAGE, OTHER_UNIT> other)
 	{
-		mValue = convert(other);
+		mValue = convert(other).mValue;
 		return *this;
 	}
 
@@ -93,21 +74,27 @@ struct LLUnit
 		return mValue;
 	}
 
-	template<typename NEW_UNIT_TYPE> LLUnit<NEW_UNIT_TYPE, STORAGE_TYPE> as()
+	template<typename NEW_UNIT_TYPE> 
+	STORAGE_TYPE getAs()
 	{
-		return LLUnit<NEW_UNIT_TYPE, STORAGE_TYPE>(*this);
+		return LLUnit<STORAGE_TYPE, NEW_UNIT_TYPE>(*this).value();
 	}
 
+	template<typename NEW_UNIT_TYPE> 
+	STORAGE_TYPE setAs(STORAGE_TYPE val)
+	{
+		*this = LLUnit<STORAGE_TYPE, NEW_UNIT_TYPE>(val);
+	}
 
 	void operator += (storage_t value)
 	{
 		mValue += value;
 	}
 
-	template<typename OTHER_UNIT, typename OTHER_STORAGE>
-	void operator += (LLUnit<OTHER_UNIT, OTHER_STORAGE> other)
+	template<typename OTHER_STORAGE, typename OTHER_UNIT>
+	void operator += (LLUnit<OTHER_STORAGE, OTHER_UNIT> other)
 	{
-		mValue += convert(other);
+		mValue += convert(other).mValue;
 	}
 
 	void operator -= (storage_t value)
@@ -115,10 +102,10 @@ struct LLUnit
 		mValue -= value;
 	}
 
-	template<typename OTHER_UNIT, typename OTHER_STORAGE>
-	void operator -= (LLUnit<OTHER_UNIT, OTHER_STORAGE> other)
+	template<typename OTHER_STORAGE, typename OTHER_UNIT>
+	void operator -= (LLUnit<OTHER_STORAGE, OTHER_UNIT> other)
 	{
-		mValue -= convert(other);
+		mValue -= convert(other).mValue;
 	}
 
 	void operator *= (storage_t multiplicand)
@@ -127,7 +114,7 @@ struct LLUnit
 	}
 
 	template<typename OTHER_UNIT, typename OTHER_STORAGE>
-	void operator *= (LLUnit<OTHER_UNIT, OTHER_STORAGE> multiplicand)
+	void operator *= (LLUnit<OTHER_STORAGE, OTHER_UNIT> multiplicand)
 	{
 		// spurious use of dependent type to stop gcc from triggering the static assertion before instantiating the template
 		llstatic_assert_template(OTHER_UNIT, false, "Multiplication of unit types not supported.");
@@ -139,37 +126,37 @@ struct LLUnit
 	}
 
 	template<typename OTHER_UNIT, typename OTHER_STORAGE>
-	void operator /= (LLUnit<OTHER_UNIT, OTHER_STORAGE> divisor)
+	void operator /= (LLUnit<OTHER_STORAGE, OTHER_UNIT> divisor)
 	{
 		// spurious use of dependent type to stop gcc from triggering the static assertion before instantiating the template
 		llstatic_assert_template(OTHER_UNIT, false, "Illegal in-place division of unit types.");
 	}
 
-	template<typename SOURCE_UNITS, typename SOURCE_STORAGE>
-	static storage_t convert(LLUnit<SOURCE_UNITS, SOURCE_STORAGE> v) 
+	template<typename SOURCE_STORAGE, typename SOURCE_UNITS>
+	static self_t convert(LLUnit<SOURCE_STORAGE, SOURCE_UNITS> 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()); 
+		self_t result;
+		ll_convert_units(v, result);
+		return result;
 	}
 
 protected:
 	storage_t mValue;
 };
 
-template<typename UNIT_TYPE, typename STORAGE_TYPE>
-struct LLUnitImplicit : public LLUnit<UNIT_TYPE, STORAGE_TYPE>
+template<typename STORAGE_TYPE, typename UNIT_TYPE>
+struct LLUnitImplicit : public LLUnit<STORAGE_TYPE, UNIT_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;
+	typedef LLUnitImplicit<STORAGE_TYPE, UNIT_TYPE> self_t;
+	typedef typename LLUnit<STORAGE_TYPE, UNIT_TYPE>::storage_t storage_t;
+	typedef LLUnit<STORAGE_TYPE, UNIT_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)
+	template<typename OTHER_STORAGE, typename OTHER_UNIT>
+	LLUnitImplicit(LLUnit<OTHER_STORAGE, OTHER_UNIT> other)
 	:	base_t(convert(other))
 	{}
 
@@ -181,53 +168,86 @@ struct LLUnitImplicit : public LLUnit<UNIT_TYPE, STORAGE_TYPE>
 	}
 };
 
+
+template<typename S1, typename T1, typename S2, typename T2>
+LL_FORCE_INLINE void ll_convert_units(LLUnit<S1, T1> in, LLUnit<S2, T2>& out, ...)
+{
+	static_assert(boost::is_same<T1, T2>::value 
+					|| !boost::is_same<T1, typename T1::base_unit_t>::value 
+					|| !boost::is_same<T2, typename T2::base_unit_t>::value, 
+				"invalid conversion");
+
+	if (boost::is_same<T1, typename T1::base_unit_t>::value)
+	{
+		if (boost::is_same<T2, typename T2::base_unit_t>::value)
+		{
+			// T1 and T2 fully reduced and equal...just copy
+			out = (S2)in.value();
+		}
+		else
+		{
+			// reduce T2
+			LLUnit<S2, typename T2::base_unit_t> new_out;
+			ll_convert_units(in, new_out);
+			ll_convert_units(new_out, out);
+		}
+	}
+	else
+	{
+		// reduce T1
+		LLUnit<S1, typename T1::base_unit_t> new_in;
+		ll_convert_units(in, new_in);
+		ll_convert_units(new_in, out);
+	}
+}
+
 //
 // 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)
+template<typename STORAGE_TYPE1, typename UNIT_TYPE1, typename STORAGE_TYPE2, typename UNIT_TYPE2>
+LLUnit<STORAGE_TYPE1, UNIT_TYPE1> operator + (LLUnit<STORAGE_TYPE1, UNIT_TYPE1> first, LLUnit<STORAGE_TYPE2, UNIT_TYPE2> second)
 {
-	LLUnit<UNIT_TYPE1, STORAGE_TYPE1> result(first);
+	LLUnit<STORAGE_TYPE1, UNIT_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)
+template<typename STORAGE_TYPE, typename UNIT_TYPE, typename SCALAR_TYPE>
+LLUnit<STORAGE_TYPE, UNIT_TYPE> operator + (LLUnit<STORAGE_TYPE, UNIT_TYPE> first, SCALAR_TYPE second)
 {
-	LLUnit<UNIT_TYPE, STORAGE_TYPE> result(first);
+	LLUnit<STORAGE_TYPE, UNIT_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)
+template<typename STORAGE_TYPE, typename UNIT_TYPE, typename SCALAR_TYPE>
+LLUnit<STORAGE_TYPE, UNIT_TYPE> operator + (SCALAR_TYPE first, LLUnit<STORAGE_TYPE, UNIT_TYPE> second)
 {
-	LLUnit<UNIT_TYPE, STORAGE_TYPE> result(first);
+	LLUnit<STORAGE_TYPE, UNIT_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)
+template<typename STORAGE_TYPE1, typename UNIT_TYPE1, typename STORAGE_TYPE2, typename UNIT_TYPE2>
+LLUnitImplicit<STORAGE_TYPE1, UNIT_TYPE1> operator + (LLUnitImplicit<STORAGE_TYPE1, UNIT_TYPE1> first, LLUnit<STORAGE_TYPE2, UNIT_TYPE2> second)
 {
-	LLUnitImplicit<UNIT_TYPE1, STORAGE_TYPE1> result(first);
+	LLUnitImplicit<STORAGE_TYPE1, UNIT_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)
+template<typename STORAGE_TYPE, typename UNIT_TYPE, typename SCALAR_TYPE>
+LLUnitImplicit<STORAGE_TYPE, UNIT_TYPE> operator + (LLUnitImplicit<STORAGE_TYPE, UNIT_TYPE> first, SCALAR_TYPE second)
 {
-	LLUnitImplicit<UNIT_TYPE, STORAGE_TYPE> result(first);
+	LLUnitImplicit<STORAGE_TYPE, UNIT_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)
+template<typename STORAGE_TYPE1, typename UNIT_TYPE1, typename STORAGE_TYPE2, typename UNIT_TYPE2>
+LLUnitImplicit<STORAGE_TYPE1, UNIT_TYPE1> operator + (LLUnitImplicit<STORAGE_TYPE1, UNIT_TYPE1> first, LLUnitImplicit<STORAGE_TYPE2, UNIT_TYPE2> second)
 {
-	LLUnitImplicit<UNIT_TYPE1, STORAGE_TYPE1> result(first);
+	LLUnitImplicit<STORAGE_TYPE1, UNIT_TYPE1> result(first);
 	result += second;
 	return result;
 }
@@ -235,50 +255,50 @@ LLUnitImplicit<UNIT_TYPE1, STORAGE_TYPE1> operator + (LLUnitImplicit<UNIT_TYPE1,
 //
 // 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)
+template<typename STORAGE_TYPE1, typename UNIT_TYPE1, typename STORAGE_TYPE2, typename UNIT_TYPE2>
+LLUnit<STORAGE_TYPE1, UNIT_TYPE1> operator - (LLUnit<STORAGE_TYPE1, UNIT_TYPE1> first, LLUnit<STORAGE_TYPE2, UNIT_TYPE2> second)
 {
-	LLUnit<UNIT_TYPE1, STORAGE_TYPE1> result(first);
+	LLUnit<STORAGE_TYPE1, UNIT_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)
+template<typename STORAGE_TYPE, typename UNIT_TYPE, typename SCALAR_TYPE>
+LLUnit<STORAGE_TYPE, UNIT_TYPE> operator - (LLUnit<STORAGE_TYPE, UNIT_TYPE> first, SCALAR_TYPE second)
 {
-	LLUnit<UNIT_TYPE, STORAGE_TYPE> result(first);
+	LLUnit<STORAGE_TYPE, UNIT_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)
+template<typename STORAGE_TYPE, typename UNIT_TYPE, typename SCALAR_TYPE>
+LLUnit<STORAGE_TYPE, UNIT_TYPE> operator - (SCALAR_TYPE first, LLUnit<STORAGE_TYPE, UNIT_TYPE> second)
 {
-	LLUnit<UNIT_TYPE, STORAGE_TYPE> result(first);
+	LLUnit<STORAGE_TYPE, UNIT_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)
+template<typename STORAGE_TYPE1, typename UNIT_TYPE1, typename STORAGE_TYPE2, typename UNIT_TYPE2>
+LLUnitImplicit<STORAGE_TYPE1, UNIT_TYPE1> operator - (LLUnitImplicit<STORAGE_TYPE1, UNIT_TYPE1> first, LLUnitImplicit<STORAGE_TYPE2, UNIT_TYPE2> second)
 {
-	LLUnitImplicit<UNIT_TYPE1, STORAGE_TYPE1> result(first);
+	LLUnitImplicit<STORAGE_TYPE1, UNIT_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)
+template<typename STORAGE_TYPE, typename UNIT_TYPE, typename SCALAR_TYPE>
+LLUnitImplicit<STORAGE_TYPE, UNIT_TYPE> operator - (LLUnitImplicit<STORAGE_TYPE, UNIT_TYPE> first, SCALAR_TYPE second)
 {
-	LLUnitImplicit<UNIT_TYPE, STORAGE_TYPE> result(first);
+	LLUnitImplicit<STORAGE_TYPE, UNIT_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)
+template<typename STORAGE_TYPE, typename UNIT_TYPE, typename SCALAR_TYPE>
+LLUnitImplicit<STORAGE_TYPE, UNIT_TYPE> operator - (SCALAR_TYPE first, LLUnitImplicit<STORAGE_TYPE, UNIT_TYPE> second)
 {
-	LLUnitImplicit<UNIT_TYPE, STORAGE_TYPE> result(first);
+	LLUnitImplicit<STORAGE_TYPE, UNIT_TYPE> result(first);
 	result -= second;
 	return result;
 }
@@ -286,102 +306,100 @@ LLUnitImplicit<UNIT_TYPE, STORAGE_TYPE> operator - (SCALAR_TYPE first, LLUnitImp
 //
 // 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)
+template<typename STORAGE_TYPE, typename UNIT_TYPE, typename SCALAR_TYPE>
+LLUnit<STORAGE_TYPE, UNIT_TYPE> operator * (SCALAR_TYPE first, LLUnit<STORAGE_TYPE, UNIT_TYPE> second)
 {
-	return LLUnit<UNIT_TYPE, STORAGE_TYPE>((STORAGE_TYPE)(first * second.value()));
+	return LLUnit<STORAGE_TYPE, 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)
+template<typename STORAGE_TYPE, typename UNIT_TYPE, typename SCALAR_TYPE>
+LLUnit<STORAGE_TYPE, UNIT_TYPE> operator * (LLUnit<STORAGE_TYPE, UNIT_TYPE> first, SCALAR_TYPE second)
 {
-	return LLUnit<UNIT_TYPE, STORAGE_TYPE>((STORAGE_TYPE)(first.value() * second));
+	return LLUnit<STORAGE_TYPE, 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>)
+template<typename STORAGE_TYPE1, typename UNIT_TYPE1, typename STORAGE_TYPE2, typename UNIT_TYPE2>
+LLUnit<STORAGE_TYPE1, UNIT_TYPE1> operator * (LLUnit<STORAGE_TYPE1, UNIT_TYPE1>, LLUnit<STORAGE_TYPE2, UNIT_TYPE2>)
 {
 	// spurious use of dependent type to stop gcc from triggering the static assertion before instantiating the template
 	llstatic_assert_template(STORAGE_TYPE1, false, "Multiplication of unit types results in new unit type - not supported.");
-	return LLUnit<UNIT_TYPE1, STORAGE_TYPE1>();
+	return LLUnit<STORAGE_TYPE1, UNIT_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)
+template<typename STORAGE_TYPE, typename UNIT_TYPE, typename SCALAR_TYPE>
+LLUnitImplicit<STORAGE_TYPE, UNIT_TYPE> operator * (SCALAR_TYPE first, LLUnitImplicit<STORAGE_TYPE, UNIT_TYPE> second)
 {
-	return LLUnitImplicit<UNIT_TYPE, STORAGE_TYPE>(first * second.value());
+	return LLUnitImplicit<STORAGE_TYPE, UNIT_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)
+template<typename STORAGE_TYPE, typename UNIT_TYPE, typename SCALAR_TYPE>
+LLUnitImplicit<STORAGE_TYPE, UNIT_TYPE> operator * (LLUnitImplicit<STORAGE_TYPE, UNIT_TYPE> first, SCALAR_TYPE second)
 {
-	return LLUnitImplicit<UNIT_TYPE, STORAGE_TYPE>(first.value() * second);
+	return LLUnitImplicit<STORAGE_TYPE, UNIT_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>)
+template<typename STORAGE_TYPE1, typename UNIT_TYPE1, typename STORAGE_TYPE2, typename UNIT_TYPE2>
+LLUnitImplicit<STORAGE_TYPE1, UNIT_TYPE1> operator * (LLUnitImplicit<STORAGE_TYPE1, UNIT_TYPE1>, LLUnitImplicit<STORAGE_TYPE2, UNIT_TYPE2>)
 {
 	// spurious use of dependent type to stop gcc from triggering the static assertion before instantiating the template
 	llstatic_assert_template(STORAGE_TYPE1, false, "Multiplication of unit types results in new unit type - not supported.");
-	return LLUnitImplicit<UNIT_TYPE1, STORAGE_TYPE1>();
+	return LLUnitImplicit<STORAGE_TYPE1, UNIT_TYPE1>();
 }
 
 //
 // operator /
 //
-template<typename UNIT_TYPE, typename STORAGE_TYPE, typename SCALAR_TYPE>
-SCALAR_TYPE operator / (SCALAR_TYPE first, LLUnit<UNIT_TYPE, STORAGE_TYPE> second)
+template<typename STORAGE_TYPE, typename UNIT_TYPE, typename SCALAR_TYPE>
+SCALAR_TYPE operator / (SCALAR_TYPE first, LLUnit<STORAGE_TYPE, UNIT_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)
+template<typename STORAGE_TYPE, typename UNIT_TYPE, typename SCALAR_TYPE>
+LLUnit<STORAGE_TYPE, UNIT_TYPE> operator / (LLUnit<STORAGE_TYPE, UNIT_TYPE> first, SCALAR_TYPE second)
 {
-	return LLUnit<UNIT_TYPE, STORAGE_TYPE>((STORAGE_TYPE)(first.value() / second));
+	return LLUnit<STORAGE_TYPE, 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)
+template<typename STORAGE_TYPE1, typename UNIT_TYPE1, typename STORAGE_TYPE2, typename UNIT_TYPE2>
+STORAGE_TYPE1 operator / (LLUnit<STORAGE_TYPE1, UNIT_TYPE1> first, LLUnit<STORAGE_TYPE2, UNIT_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());
+	return STORAGE_TYPE1(first.value() / first.convert(second));
 }
 
-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)
+template<typename STORAGE_TYPE, typename UNIT_TYPE, typename SCALAR_TYPE>
+LLUnitImplicit<STORAGE_TYPE, UNIT_TYPE> operator / (LLUnitImplicit<STORAGE_TYPE, UNIT_TYPE> first, SCALAR_TYPE second)
 {
-	return LLUnitImplicit<UNIT_TYPE, STORAGE_TYPE>((STORAGE_TYPE)(first.value() / second));
+	return LLUnitImplicit<STORAGE_TYPE, 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)
+template<typename STORAGE_TYPE1, typename UNIT_TYPE1, typename STORAGE_TYPE2, typename UNIT_TYPE2>
+STORAGE_TYPE1 operator / (LLUnitImplicit<STORAGE_TYPE1, UNIT_TYPE1> first, LLUnitImplicit<STORAGE_TYPE2, UNIT_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());
+	return STORAGE_TYPE1(first.value() / first.convert(second));
 }
 
 #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)                                         \
+template<typename STORAGE_TYPE, typename UNIT_TYPE, typename SCALAR_TYPE>                                            \
+bool operator op (SCALAR_TYPE first, LLUnit<STORAGE_TYPE, UNIT_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)                                         \
+template<typename STORAGE_TYPE, typename UNIT_TYPE, typename SCALAR_TYPE>                                            \
+bool operator op (LLUnit<STORAGE_TYPE, UNIT_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) \
+template<typename STORAGE_TYPE1, typename UNIT_TYPE1, typename STORAGE_TYPE2, typename UNIT_TYPE2>                   \
+bool operator op (LLUnitImplicit<STORAGE_TYPE1, UNIT_TYPE1> first, LLUnitImplicit<STORAGE_TYPE2, UNIT_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)             \
+template<typename STORAGE_TYPE1, typename UNIT_TYPE1, typename STORAGE_TYPE2, typename UNIT_TYPE2>                   \
+	bool operator op (LLUnit<STORAGE_TYPE1, UNIT_TYPE1> first, LLUnit<STORAGE_TYPE2, UNIT_TYPE2> second)             \
 {                                                                                                                    \
 	return first.value() op first.convert(second);                                                                   \
 }
@@ -401,80 +419,142 @@ struct LLGetUnitLabel
 };
 
 template<typename T, typename STORAGE_T>
-struct LLGetUnitLabel<LLUnit<T, STORAGE_T> >
+struct LLGetUnitLabel<LLUnit<STORAGE_T, T> >
 {
 	static const char* getUnitLabel() { return T::getUnitLabel(); }
 };
 
-//
-// Unit declarations
-//
-namespace LLUnits
+template<typename VALUE_TYPE>
+struct LLUnitLinearOps
 {
-template<typename T>
-T rawValue(T val) { return val; }
+	typedef LLUnitLinearOps<VALUE_TYPE> self_t;
+	LLUnitLinearOps(VALUE_TYPE val) : mValue (val) {}
 
-template<typename UNIT_TYPE, typename STORAGE_TYPE> 
-STORAGE_TYPE rawValue(LLUnit<UNIT_TYPE, STORAGE_TYPE> val) { return val.value(); }
+	operator VALUE_TYPE() const { return mValue; }
+	VALUE_TYPE mValue;
 
-template<typename UNIT_TYPE, typename STORAGE_TYPE> 
-STORAGE_TYPE rawValue(LLUnitImplicit<UNIT_TYPE, STORAGE_TYPE> val) { return val.value(); }
+	template<typename T>
+	self_t operator * (T other)
+	{
+		return mValue * other;
+	}
+
+	template<typename T>
+	self_t operator / (T other)
+	{
+		return mValue / other;
+	}
+
+	template<typename T>
+	self_t operator + (T other)
+	{
+		return mValue + other;
+	}
+
+	template<typename T>
+	self_t operator - (T other)
+	{
+		return mValue - other;
+	}
+};
 
-#define LL_DECLARE_DERIVED_UNIT(conversion_factor, base_unit_name, unit_name, unit_label)		\
+template<typename VALUE_TYPE>
+struct LLUnitInverseLinearOps
+{
+	typedef LLUnitInverseLinearOps<VALUE_TYPE> self_t;
+
+	LLUnitInverseLinearOps(VALUE_TYPE val) : mValue (val) {}
+	operator VALUE_TYPE() const { return mValue; }
+	VALUE_TYPE mValue;
+
+	template<typename T>
+	self_t operator * (T other)
+	{
+		return mValue / other;
+	}
+
+	template<typename T>
+	self_t operator / (T other)
+	{
+		return mValue * other;
+	}
+
+	template<typename T>
+	self_t operator + (T other)
+	{
+		return mValue - other;
+	}
+
+	template<typename T>
+	self_t operator - (T other)
+	{
+		return mValue + other;
+	}
+};
+
+#define LL_DECLARE_BASE_UNIT(base_unit_name, unit_label) \
+struct base_unit_name { typedef base_unit_name base_unit_t; static const char* getUnitLabel() { return unit_label; }}
+
+#define LL_DECLARE_DERIVED_UNIT(unit_name, unit_label, base_unit_name, conversion_operation)	\
 struct unit_name                                                                                \
 {                                                                                               \
 	typedef base_unit_name base_unit_t;                                                         \
 	static const char* getUnitLabel() { return unit_label; }									\
 };                                                                                              \
-template<typename STORAGE_TYPE>                                                                 \
-struct ConversionFactor<unit_name, base_unit_name, STORAGE_TYPE>                                \
-{                                                                                               \
-	static F64 get()                                                                            \
-	{                                                                                           \
-		return (F64)conversion_factor;                                                          \
-	}                                                                                           \
-};                                                                                              \
 	                                                                                            \
-template<typename STORAGE_TYPE>                                                                 \
-struct ConversionFactor<base_unit_name, unit_name, STORAGE_TYPE>						        \
+template<typename S1, typename S2>                                                              \
+void ll_convert_units(LLUnit<S1, unit_name> in, LLUnit<S2, base_unit_name>& out)                \
 {                                                                                               \
-	static F64 get()                                                                            \
-	{                                                                                           \
-		return (F64)(1.0 / (conversion_factor));                                                \
-	}                                                                                           \
-}
+	out = (S2)(LLUnitLinearOps<S1>(in.value()) conversion_operation).mValue;                    \
+}                                                                                               \
+                                                                                                \
+template<typename S1, typename S2>                                                              \
+void ll_convert_units(LLUnit<S1, base_unit_name> in, LLUnit<S2, unit_name>& out)                \
+{                                                                                               \
+	out = (S2)(LLUnitInverseLinearOps<S1>(in.value()) conversion_operation).mValue;             \
+}                                                                                               
 
-#define LL_DECLARE_BASE_UNIT(base_unit_name, unit_label) \
-struct base_unit_name { typedef base_unit_name base_unit_t; static const char* getUnitLabel() { return unit_label; }}
+//
+// Unit declarations
+//
 
+namespace LLUnits
+{
 LL_DECLARE_BASE_UNIT(Bytes, "B");
-LL_DECLARE_DERIVED_UNIT(1024, Bytes, Kibibytes, "KiB");
-LL_DECLARE_DERIVED_UNIT(1024 * 1024, Bytes, Mibibytes, "MiB");
-LL_DECLARE_DERIVED_UNIT(1024 * 1024 * 1024, Bytes, Gibibytes, "GiB");
-LL_DECLARE_DERIVED_UNIT(1.0 / 8.0, Bytes, Bits, "b");
-LL_DECLARE_DERIVED_UNIT(1024 / 8, Bytes, Kibibits, "Kib");
-LL_DECLARE_DERIVED_UNIT(1024 / 8, Bytes, Mibibits, "Mib");
-LL_DECLARE_DERIVED_UNIT(1024 * 1024 * 1024 / 8, Bytes, Gibibits, "Gib");
+LL_DECLARE_DERIVED_UNIT(Kilobytes, "KB", Bytes, * 1000);
+LL_DECLARE_DERIVED_UNIT(Megabytes, "MB", Kilobytes, * 1000);
+LL_DECLARE_DERIVED_UNIT(Gigabytes, "GB", Megabytes, * 1000);
+LL_DECLARE_DERIVED_UNIT(Kibibytes, "KiB", Bytes, * 1024);
+LL_DECLARE_DERIVED_UNIT(Mibibytes, "MiB", Kibibytes, * 1024);
+LL_DECLARE_DERIVED_UNIT(Gibibytes, "GiB", Mibibytes, * 1024);
+
+LL_DECLARE_DERIVED_UNIT(Bits, "b", Bytes, / 8);
+LL_DECLARE_DERIVED_UNIT(Kilobits, "Kb", Bytes, * 1000 / 8);
+LL_DECLARE_DERIVED_UNIT(Megabits, "Mb", Kilobits, * 1000 / 8);
+LL_DECLARE_DERIVED_UNIT(Gigabits, "Gb", Megabits, * 1000 / 8);
+LL_DECLARE_DERIVED_UNIT(Kibibits, "Kib", Bytes, * 1024 / 8);
+LL_DECLARE_DERIVED_UNIT(Mibibits, "Mib", Kibibits, * 1024 / 8);
+LL_DECLARE_DERIVED_UNIT(Gibibits, "Gib", Mibibits, * 1024 / 8);
 
 LL_DECLARE_BASE_UNIT(Seconds, "s");
-LL_DECLARE_DERIVED_UNIT(60, Seconds, Minutes, "min");
-LL_DECLARE_DERIVED_UNIT(60 * 60, Seconds, Hours, "h");
-LL_DECLARE_DERIVED_UNIT(1.0 / 1000.0, Seconds, Milliseconds, "ms");
-LL_DECLARE_DERIVED_UNIT(1.0 / 1000000.0, Seconds, Microseconds, "\x09\x3cs");
-LL_DECLARE_DERIVED_UNIT(1.0 / 1000000000.0, Seconds, Nanoseconds, "ns");
+LL_DECLARE_DERIVED_UNIT(Minutes, "min", Seconds, * 60);
+LL_DECLARE_DERIVED_UNIT(Hours, "h", Seconds, * 60 * 60);
+LL_DECLARE_DERIVED_UNIT(Milliseconds, "ms", Seconds, / 1000);
+LL_DECLARE_DERIVED_UNIT(Microseconds, "\x09\x3cs", Milliseconds, / 1000);
+LL_DECLARE_DERIVED_UNIT(Nanoseconds, "ns", Microseconds, / 1000);
 
 LL_DECLARE_BASE_UNIT(Meters, "m");
-LL_DECLARE_DERIVED_UNIT(1000, Meters, Kilometers, "km");
-LL_DECLARE_DERIVED_UNIT(1.0 / 100.0, Meters, Centimeters, "cm");
-LL_DECLARE_DERIVED_UNIT(1.0 / 1000.0, Meters, Millimeters, "mm");
+LL_DECLARE_DERIVED_UNIT(Kilometers, "km", Meters, * 1000);
+LL_DECLARE_DERIVED_UNIT(Centimeters, "cm", Meters, / 100);
+LL_DECLARE_DERIVED_UNIT(Millimeters, "mm", Meters, / 1000);
 
 LL_DECLARE_BASE_UNIT(Hertz, "Hz");
-LL_DECLARE_DERIVED_UNIT(1000, Hertz, Kilohertz, "KHz");
-LL_DECLARE_DERIVED_UNIT(1000 * 1000, Hertz, Megahertz, "MHz");
-LL_DECLARE_DERIVED_UNIT(1000 * 1000 * 1000, Hertz, Gigahertz, "GHz");
+LL_DECLARE_DERIVED_UNIT(Kilohertz, "KHz", Hertz, * 1000);
+LL_DECLARE_DERIVED_UNIT(Megahertz, "MHz", Kilohertz, * 1000);
+LL_DECLARE_DERIVED_UNIT(Gigahertz, "GHz", Megahertz, * 1000);
 
 LL_DECLARE_BASE_UNIT(Radians, "rad");
-LL_DECLARE_DERIVED_UNIT(0.01745329251994, Radians, Degrees, "deg");
+LL_DECLARE_DERIVED_UNIT(Degrees, "deg", Radians, * 0.01745329251994);
 
 
 } // namespace LLUnits
diff --git a/indra/llcommon/tests/llunits_test.cpp b/indra/llcommon/tests/llunits_test.cpp
index 33e30f9688..a5df51f6de 100644
--- a/indra/llcommon/tests/llunits_test.cpp
+++ b/indra/llcommon/tests/llunits_test.cpp
@@ -34,8 +34,8 @@ namespace LLUnits
 {
 	// using powers of 2 to allow strict floating point equality
 	LL_DECLARE_BASE_UNIT(Quatloos, "Quat");
-	LL_DECLARE_DERIVED_UNIT(4, Quatloos, Latinum, "Lat");
-	LL_DECLARE_DERIVED_UNIT((1.0 / 4.0), Quatloos, Solari, "Sol");
+	LL_DECLARE_DERIVED_UNIT(Latinum, "Lat", Quatloos, * 4);
+	LL_DECLARE_DERIVED_UNIT(Solari, "Sol", Latinum, / 16);
 }
 
 namespace tut
@@ -53,105 +53,110 @@ namespace tut
 	template<> template<>
 	void units_object_t::test<1>()
 	{
-		LLUnit<Quatloos, F32> float_quatloos;
-		ensure(float_quatloos.value() == 0.f);
+		LLUnit<F32, Quatloos> float_quatloos;
+		ensure(float_quatloos == 0.f);
 
-		LLUnit<Quatloos, S32> int_quatloos;
-		ensure(int_quatloos.value() == 0);
+		LLUnit<F32, Quatloos> float_initialize_quatloos(1);
+		ensure(float_initialize_quatloos == 1.f);
+
+		LLUnit<S32, Quatloos> int_quatloos;
+		ensure(int_quatloos == 0);
 
 		int_quatloos = 42;
-		ensure(int_quatloos.value() == 42);
+		ensure(int_quatloos == 42);
 		float_quatloos = int_quatloos;
-		ensure(float_quatloos.value() == 42.f);
+		ensure(float_quatloos == 42.f);
 
 		int_quatloos = float_quatloos;
-		ensure(int_quatloos.value() == 42);
+		ensure(int_quatloos == 42);
 
 		float_quatloos = 42.1f;
-		ensure(float_quatloos.value() == 42.1f);
+		ensure(float_quatloos == 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);
+		ensure(int_quatloos == 42);
+		LLUnit<U32, Quatloos> unsigned_int_quatloos(float_quatloos);
+		ensure(unsigned_int_quatloos == 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);
+		LLUnit<F32, Quatloos> quatloos(1.f);
+		ensure(quatloos == 1.f);
+		LLUnit<F32, Latinum> latinum_bars(quatloos);
+		ensure(latinum_bars == 1.f / 4.f);
 
 		latinum_bars = 256;
 		quatloos = latinum_bars;
-		ensure(quatloos.value() == 1024);
+		ensure(quatloos == 1024);
 
-		LLUnit<Solari, F32> solari(quatloos);
-		ensure(solari.value() == 4096);
+		LLUnit<F32, Solari> solari(quatloos);
+		ensure(solari == 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);
+		LLUnit<F32, Solari> solari = 4.f;
+		LLUnit<F32, Latinum> latinum_bars = solari;
+		ensure(latinum_bars == 0.25f);
 	}
 
 	// math operations
 	template<> template<>
 	void units_object_t::test<4>()
 	{
-		LLUnit<Quatloos, F32> quatloos = 1.f;
+		LLUnit<F32, Quatloos> quatloos = 1.f;
 		quatloos *= 4.f;
-		ensure(quatloos.value() == 4);
+		ensure(quatloos == 4);
 		quatloos = quatloos * 2;
-		ensure(quatloos.value() == 8);
+		ensure(quatloos == 8);
 		quatloos = 2.f * quatloos;
-		ensure(quatloos.value() == 16);
+		ensure(quatloos == 16);
 
 		quatloos += 4.f;
-		ensure(quatloos.value() == 20);
+		ensure(quatloos == 20);
 		quatloos += 4;
-		ensure(quatloos.value() == 24);
+		ensure(quatloos == 24);
 		quatloos = quatloos + 4;
-		ensure(quatloos.value() == 28);
+		ensure(quatloos == 28);
 		quatloos = 4 + quatloos;
-		ensure(quatloos.value() == 32);
+		ensure(quatloos == 32);
 		quatloos += quatloos * 3;
-		ensure(quatloos.value() == 128);
+		ensure(quatloos == 128);
 
 		quatloos -= quatloos / 4 * 3;
-		ensure(quatloos.value() == 32);
+		ensure(quatloos == 32);
 		quatloos = quatloos - 8;
-		ensure(quatloos.value() == 24);
+		ensure(quatloos == 24);
 		quatloos -= 4;
-		ensure(quatloos.value() == 20);
+		ensure(quatloos == 20);
 		quatloos -= 4.f;
-		ensure(quatloos.value() == 16);
+		ensure(quatloos == 16);
 
 		quatloos *= 2.f;
-		ensure(quatloos.value() == 32);
+		ensure(quatloos == 32);
 		quatloos = quatloos * 2.f;
-		ensure(quatloos.value() == 64);
+		ensure(quatloos == 64);
 		quatloos = 0.5f * quatloos;
-		ensure(quatloos.value() == 32);
+		ensure(quatloos == 32);
 
 		quatloos /= 2.f;
-		ensure(quatloos.value() == 16);
+		ensure(quatloos == 16);
 		quatloos = quatloos / 4;
-		ensure(quatloos.value() == 4);
+		ensure(quatloos == 4);
 
-		F32 ratio = quatloos / LLUnit<Quatloos, F32>(4.f);
+		F32 ratio = quatloos / LLUnit<F32, Quatloos>(4.f);
+		ensure(ratio == 1);
+		ratio = quatloos / LLUnit<F32, Solari>(16.f);
 		ensure(ratio == 1);
 
-		quatloos += LLUnit<Solari, F32>(4.f);
-		ensure(quatloos.value() == 5);
-		quatloos -= LLUnit<Latinum, F32>(1.f);
-		ensure(quatloos.value() == 1);
+		quatloos += LLUnit<F32, Solari>(4.f);
+		ensure(quatloos == 5);
+		quatloos -= LLUnit<F32, Latinum>(1.f);
+		ensure(quatloos == 1);
 	}
 
 	// implicit units
@@ -159,16 +164,16 @@ namespace tut
 	void units_object_t::test<5>()
 	{
 		// 0-initialized
-		LLUnit<Quatloos, F32> quatloos(0);
+		LLUnit<F32, Quatloos> quatloos(0);
 		// initialize implicit unit from explicit
-		LLUnitImplicit<Quatloos, F32> quatloos_implicit = quatloos + 1;
-		ensure(quatloos_implicit.value() == 1);
+		LLUnitImplicit<F32, Quatloos> quatloos_implicit = quatloos + 1;
+		ensure(quatloos_implicit == 1);
 
 		// assign implicit to explicit, or perform math operations
 		quatloos = quatloos_implicit;
-		ensure(quatloos.value() == 1);
+		ensure(quatloos == 1);
 		quatloos += quatloos_implicit;
-		ensure(quatloos.value() == 2);
+		ensure(quatloos == 2);
 
 		// math operations on implicits
 		quatloos_implicit = 1;
@@ -204,5 +209,11 @@ namespace tut
 
 		S32 int_val = quatloos_implicit;
 		ensure(int_val == 16);
+
+		// conversion of implicits
+		LLUnitImplicit<F32, Latinum> latinum_implicit(2);
+		ensure(latinum_implicit == 2);
+
+		ensure(latinum_implicit * 2 == quatloos_implicit);
 	}
 }