MAINT-8190 Avatar Physics safety checks and better iteration spreading

indra/newview/llphysicsmotion.cpp

@@ -44,7 +44,9 @@ typedef std::map<std::string, std::string> controller_map_t;
 typedef std::map<std::string, F32> default_controller_map_t;
 
 #define MIN_REQUIRED_PIXEL_AREA_AVATAR_PHYSICS_MOTION 0.f
-#define TIME_ITERATION_STEP 0.05f
+// we use TIME_ITERATION_STEP_MAX in division operation, make sure this is a simple
+// value and devision result won't end with repeated/recurring tail like 1.333(3)
+#define TIME_ITERATION_STEP_MAX 0.05f // minimal step size will end up as 0.025
 
 inline F64 llsgn(const F64 a)
 {
@@ -480,7 +482,7 @@ BOOL LLPhysicsMotion::onUpdate(F32 time)
         if (!mParamDriver)
                 return FALSE;
 
-        if (!mLastTime)
+        if (!mLastTime || mLastTime >= time)
         {
                 mLastTime = time;
                 return FALSE;
@@ -561,14 +563,10 @@ BOOL LLPhysicsMotion::onUpdate(F32 time)
 	// bounce at right (relatively) position.
 	// Note: this doesn't look to be optimal, since it provides only "roughly same" behavior, but
 	// irregularity at higher fps looks to be insignificant so it works good enough for low fps.
-	for (F32 time_iteration = 0; time_iteration <= time_delta; time_iteration += TIME_ITERATION_STEP)
+	U32 steps = (U32)(time_delta / TIME_ITERATION_STEP_MAX) + 1;
+	F32 time_iteration_step = time_delta / (F32)steps; //minimal step size ends up as 0.025
+	for (U32 i = 0; i < steps; i++)
 	{
-		F32 time_iteration_step = TIME_ITERATION_STEP;
-		if (time_iteration + TIME_ITERATION_STEP > time_delta)
-		{
-			time_iteration_step = time_delta-time_iteration;
-		}
-		
 		// mPositon_local should be in normalized 0,1 range already.  Just making sure...
 		const F32 position_current_local = llclamp(mPosition_local,
 							   0.0f,