iOS platform and Android don't have __thread local storage, so avoid multi-threaded profiler on those (only on _WIN32, __linux__ and __APPLE__ and not TARGET_OS_IPHONE

Add a Sleep(0) for Windows to yield threads (and not Sleep(1))

examples/ExampleBrowser/GwenGUISupport/GwenProfileWindow.cpp

@@ -293,8 +293,10 @@ void	processProfileData( MyProfileWindow* profWindow, bool idle)
 {
 	if (profWindow)
 	{
-		
+		if (profWindow->profIter)
-		profWindow->UpdateText(profWindow->profIter, idle);
+		{
+			profWindow->UpdateText(profWindow->profIter, idle);
+		}
 	}	
 }
 

examples/ExampleBrowser/InProcessExampleBrowser.cpp

@@ -255,6 +255,7 @@ void	ExampleBrowserThreadFunc(void* userPtr,void* lsMemory)
 
 		do
 		{
+			B3_PROFILE("ExampleBrowserThreadFunc");
 			float deltaTimeInSeconds = clock.getTimeMicroseconds()/1000000.f;
 			{
 				if (deltaTimeInSeconds > 0.1)
@@ -263,9 +264,11 @@ void	ExampleBrowserThreadFunc(void* userPtr,void* lsMemory)
 				}
 				if (deltaTimeInSeconds < (gMinUpdateTimeMicroSecs/1e6))
 				{
+					B3_PROFILE("clock.usleep");
 					clock.usleep(gMinUpdateTimeMicroSecs/10.);
 				} else
 				{
+					B3_PROFILE("exampleBrowser->update");
 					clock.reset();
 					exampleBrowser->update(deltaTimeInSeconds);
 				}

examples/ExampleBrowser/OpenGLExampleBrowser.cpp

@@ -116,7 +116,6 @@ bool gAllowRetina = true;
 bool gDisableDemoSelection = false;
 static class ExampleEntries* gAllExamples=0;
 bool sUseOpenGL2 = false;
-bool drawGUI=true;
 #ifndef USE_OPENGL3
 extern bool useShadowMap;
 #endif
@@ -298,17 +297,20 @@ struct btTimings
 	int m_activeBuffer;
 	btAlignedObjectArray<btTiming> m_timings[1];
 };
-
+#ifndef BT_NO_PROFILE
-btTimings gTimings[BT_MAX_THREAD_COUNT];
+btTimings gTimings[BT_QUICKPROF_MAX_THREAD_COUNT];
+#define MAX_NESTING 1024
+int gStackDepths[BT_QUICKPROF_MAX_THREAD_COUNT] = {0};
+const char* gFuncNames[BT_QUICKPROF_MAX_THREAD_COUNT][MAX_NESTING];
+unsigned long long int gStartTimes[BT_QUICKPROF_MAX_THREAD_COUNT][MAX_NESTING];
+#endif
 
 btClock clk;
 
-#define MAX_NESTING 1024
+
 
 bool gProfileDisabled = true;
-int gStackDepths[BT_MAX_THREAD_COUNT] = {0};
+
-const char* gFuncNames[BT_MAX_THREAD_COUNT][MAX_NESTING];
-unsigned long long int gStartTimes[BT_MAX_THREAD_COUNT][MAX_NESTING];
 
 void MyDummyEnterProfileZoneFunc(const char* msg)
 {
@@ -322,8 +324,11 @@ void MyEnterProfileZoneFunc(const char* msg)
 {
 	if (gProfileDisabled)
 		return;
-	int threadId = btGetCurrentThreadIndex();
+#ifndef BT_NO_PROFILE
-
+	int threadId = btQuickprofGetCurrentThreadIndex2();
+	if (threadId<0)
+		return;
+	
 	if (gStackDepths[threadId]>=MAX_NESTING)
 	{
 		btAssert(0);
@@ -336,14 +341,18 @@ void MyEnterProfileZoneFunc(const char* msg)
 		gStartTimes[threadId][gStackDepths[threadId]]=1+gStartTimes[threadId][gStackDepths[threadId]-1];
 	}
 	gStackDepths[threadId]++;
+#endif
+
 }
 void MyLeaveProfileZoneFunc()
 {
 	if (gProfileDisabled)
 		return;
-
+#ifndef BT_NO_PROFILE
-	int threadId = btGetCurrentThreadIndex();
+	int threadId = btQuickprofGetCurrentThreadIndex2();
-
+	if (threadId<0)
+		return;
+	
 	if (gStackDepths[threadId]<=0)
 	{
 		return;
@@ -356,6 +365,7 @@ void MyLeaveProfileZoneFunc()
 	
 	unsigned long long int endTime = clk.getTimeNanoseconds();
 	gTimings[threadId].addTiming(name,threadId,startTime,endTime);
+#endif //BT_NO_PROFILE
 }
 
 
@@ -455,6 +465,7 @@ void MyKeyboardCallback(int key, int state)
 
 	if (key=='p')
 	{
+#ifndef BT_NO_PROFILE
 		if (state)
 		{
 			m_firstTiming = true;
@@ -479,7 +490,7 @@ void MyKeyboardCallback(int key, int state)
 			gTimingFile = fopen(fileName,"w");
 			fprintf(gTimingFile,"{\"traceEvents\":[\n");
 			//dump the content to file
-			for (int i=0;i<BT_MAX_THREAD_COUNT;i++)
+			for (int i=0;i<BT_QUICKPROF_MAX_THREAD_COUNT;i++)
 			{
 				if (gTimings[i].m_numTimings)
 				{
@@ -492,6 +503,7 @@ void MyKeyboardCallback(int key, int state)
 			gTimingFile = 0;
 
 		}
+#endif //BT_NO_PROFILE
 	}
 
 #ifndef NO_OPENGL3

examples/SharedMemory/PhysicsServerExample.cpp

@@ -272,10 +272,10 @@ struct MotionThreadLocalStorage
 	int threadId;
 };
 
-int skip = 0;
+
-int skip1 = 0;
 float clampedDeltaTime  = 0.2;
 
+
 void	MotionThreadFunc(void* userPtr,void* lsMemory)
 {
 	printf("MotionThreadFunc thread started\n");
@@ -299,24 +299,19 @@ void	MotionThreadFunc(void* userPtr,void* lsMemory)
 		do
 		{
 			BT_PROFILE("loop");
-			deltaTimeInSeconds+= double(clock.getTimeMicroseconds())/1000000.;
+			
+			{
+				BT_PROFILE("Sleep(0)");
+				b3Clock::usleep(0);
+			}
+			double dt = double(clock.getTimeMicroseconds())/1000000.;
+
+			deltaTimeInSeconds+= dt;
+		
 			clock.reset();
 
-			if (deltaTimeInSeconds<(1./5000.))
+			
 			{
-
-				skip++;
-				skip1++;
-				//if (skip1>105)
-				if (skip1>5)
-				{
-					BT_PROFILE("b3Clock::usleep(250)");
-					b3Clock::usleep(250);
-					skip1 = 0;
-				}
-			} else
-			{
-				skip1=0;
 				
 				//process special controller commands, such as
 				//VR controller button press/release and controller motion
@@ -450,9 +445,6 @@ void	MotionThreadFunc(void* userPtr,void* lsMemory)
 	}
 
 
-	printf("finished, #skip = %d, skip1 = %d\n",skip,skip1);
-	skip=0;
-	skip1=0;
 	//do nothing
 
 }

examples/Utils/b3Clock.cpp

@@ -187,12 +187,16 @@ double b3Clock::getTimeInSeconds()
 void b3Clock::usleep(int microSeconds)
 {
 #ifdef _WIN32
-	int millis = microSeconds/1000;
+	if (microSeconds==0)
-	if (millis < 1)
 	{
-		millis = 1;
+		Sleep(0);
+	} else
+	{
+		int millis = microSeconds/1000;
+		if (millis<1)
+			millis=1;
+		Sleep(millis);
 	}
-	Sleep(millis);
 #else
 
     ::usleep(microSeconds); 

src/LinearMath/btQuickprof.cpp

@@ -466,11 +466,10 @@ void	CProfileIterator::Enter_Parent( void )
 **
 ***************************************************************************************************/
 
-#include "btThreads.h"
 
 
 
-CProfileNode	gRoots[BT_MAX_THREAD_COUNT]={
+CProfileNode	gRoots[BT_QUICKPROF_MAX_THREAD_COUNT]={
 	CProfileNode("Root",NULL),CProfileNode("Root",NULL),CProfileNode("Root",NULL),CProfileNode("Root",NULL),
 	CProfileNode("Root",NULL),CProfileNode("Root",NULL),CProfileNode("Root",NULL),CProfileNode("Root",NULL),
 	CProfileNode("Root",NULL),CProfileNode("Root",NULL),CProfileNode("Root",NULL),CProfileNode("Root",NULL),
@@ -490,7 +489,7 @@ CProfileNode	gRoots[BT_MAX_THREAD_COUNT]={
 };
 
 
-CProfileNode* gCurrentNodes[BT_MAX_THREAD_COUNT]=
+CProfileNode* gCurrentNodes[BT_QUICKPROF_MAX_THREAD_COUNT]=
 {
 	&gRoots[ 0],	&gRoots[ 1],	&gRoots[ 2],	&gRoots[ 3],
 	&gRoots[ 4],	&gRoots[ 5],	&gRoots[ 6],	&gRoots[ 7],
@@ -517,13 +516,16 @@ unsigned long int			CProfileManager::ResetTime = 0;
 CProfileIterator *	CProfileManager::Get_Iterator( void )
 { 
 
-		int threadIndex = btGetCurrentThreadIndex();
+		int threadIndex = btQuickprofGetCurrentThreadIndex2();
+		if (threadIndex<0)
+			return 0;
+
 		return new CProfileIterator( &gRoots[threadIndex]); 
 }
 
 void						CProfileManager::CleanupMemory(void)
 {
-	for (int i=0;i<BT_MAX_THREAD_COUNT;i++)
+	for (int i=0;i<BT_QUICKPROF_MAX_THREAD_COUNT;i++)
 	{
 		gRoots[i].CleanupMemory();
 	}
@@ -545,7 +547,10 @@ void						CProfileManager::CleanupMemory(void)
  *=============================================================================================*/
 void	CProfileManager::Start_Profile( const char * name )
 {
-	int threadIndex = btGetCurrentThreadIndex();
+	int threadIndex = btQuickprofGetCurrentThreadIndex2();
+	if (threadIndex<0)
+		return;
+
 	if (name != gCurrentNodes[threadIndex]->Get_Name()) {
 		gCurrentNodes[threadIndex] = gCurrentNodes[threadIndex]->Get_Sub_Node( name );
 	}
@@ -559,7 +564,10 @@ void	CProfileManager::Start_Profile( const char * name )
  *=============================================================================================*/
 void	CProfileManager::Stop_Profile( void )
 {
-	int threadIndex = btGetCurrentThreadIndex();
+	int threadIndex = btQuickprofGetCurrentThreadIndex2();
+	if (threadIndex<0)
+		return;
+
 	// Return will indicate whether we should back up to our parent (we may
 	// be profiling a recursive function)
 	if (gCurrentNodes[threadIndex]->Return()) {
@@ -568,14 +576,7 @@ void	CProfileManager::Stop_Profile( void )
 }
 
 
-void	btEnterProfileZoneDefault(const char* name)
+
-{
-	CProfileManager::Start_Profile( name ); 
-}
-void	btLeaveProfileZoneDefault()
-{
-	CProfileManager::Stop_Profile(); 
-}
 
 
 
@@ -587,8 +588,11 @@ void	btLeaveProfileZoneDefault()
 void	CProfileManager::Reset( void )
 {
 	gProfileClock.reset();
-	gRoots[btGetCurrentThreadIndex()].Reset();
+	int threadIndex = btQuickprofGetCurrentThreadIndex2();
-	gRoots[btGetCurrentThreadIndex()].Call();
+	if (threadIndex<0)
+		return;
+	gRoots[threadIndex].Reset();
+	gRoots[threadIndex].Call();
 	FrameCounter = 0;
 	Profile_Get_Ticks(&ResetTime);
 }
@@ -676,18 +680,63 @@ void	CProfileManager::dumpAll()
 }
 
 
-#else
 
+
+unsigned int btQuickprofGetCurrentThreadIndex2()
+{
+	const unsigned int kNullIndex = ~0U;
+#ifdef _WIN32
+	__declspec( thread ) static unsigned int sThreadIndex = kNullIndex;
+#else
+#ifdef __APPLE__
+	#if TARGET_OS_IPHONE
+		unsigned int sThreadIndex = 0;
+		return -1;
+	#else
+		static __thread unsigned int sThreadIndex = kNullIndex;
+	#endif
+#else//__APPLE__
+#if __linux__
+	static __thread unsigned int sThreadIndex = kNullIndex;
+#else
+	unsigned int sThreadIndex = 0;
+	return -1;
+#endif
+#endif//__APPLE__
+	
+#endif
+	static int gThreadCounter=0;
+
+	if ( sThreadIndex == kNullIndex )
+	{
+		sThreadIndex = gThreadCounter++;
+	}
+	return sThreadIndex;
+}
+
+void	btEnterProfileZoneDefault(const char* name)
+{
+	CProfileManager::Start_Profile( name ); 
+}
+void	btLeaveProfileZoneDefault()
+{
+	CProfileManager::Stop_Profile(); 
+}
+
+
+#else
 void	btEnterProfileZoneDefault(const char* name)
 {
 }
 void	btLeaveProfileZoneDefault()
 {
 }
-
 #endif //BT_NO_PROFILE
 
 
+
+
+
 static btEnterProfileZoneFunc* bts_enterFunc = btEnterProfileZoneDefault;
 static btLeaveProfileZoneFunc* bts_leaveFunc = btLeaveProfileZoneDefault;
 

src/LinearMath/btQuickprof.h

@@ -60,6 +60,8 @@ typedef void (btLeaveProfileZoneFunc)();
 btEnterProfileZoneFunc* btGetCurrentEnterProfileZoneFunc();
 btLeaveProfileZoneFunc* btGetCurrentLeaveProfileZoneFunc();
 
+
+
 void btSetCustomEnterProfileZoneFunc(btEnterProfileZoneFunc* enterFunc);
 void btSetCustomLeaveProfileZoneFunc(btLeaveProfileZoneFunc* leaveFunc);
 
@@ -67,6 +69,11 @@ void btSetCustomLeaveProfileZoneFunc(btLeaveProfileZoneFunc* leaveFunc);
 //To disable built-in profiling, please comment out next line
 //#define BT_NO_PROFILE 1
 #ifndef BT_NO_PROFILE
+//btQuickprofGetCurrentThreadIndex will return -1 if thread index cannot be determined, 
+//otherwise returns thread index in range [0..maxThreads]
+unsigned int btQuickprofGetCurrentThreadIndex2();
+const unsigned int BT_QUICKPROF_MAX_THREAD_COUNT = 64;
+
 #include <stdio.h>//@todo remove this, backwards compatibility
 
 #include "btAlignedAllocator.h"

src/LinearMath/btThreads.cpp

@@ -226,23 +226,6 @@ bool btSpinMutex::tryLock()
     return true;
 }
 
-unsigned int btGetCurrentThreadIndex()
-{
-	const unsigned int kNullIndex = ~0U;
-#ifdef _WIN32
-	__declspec( thread ) static unsigned int sThreadIndex = kNullIndex;
-
-#else
-	static __thread unsigned int sThreadIndex = kNullIndex;
-#endif
-	static int gThreadCounter=0;
-
-	if ( sThreadIndex == kNullIndex )
-	{
-		sThreadIndex = gThreadCounter++;
-	}
-	return sThreadIndex;
-}
 
 #endif // #if BT_THREADSAFE
 

src/LinearMath/btThreads.h

@@ -69,8 +69,6 @@ const unsigned int BT_MAX_THREAD_COUNT = 64;
 SIMD_FORCE_INLINE void btMutexLock( btSpinMutex* ) {}
 SIMD_FORCE_INLINE void btMutexUnlock( btSpinMutex* ) {}
 SIMD_FORCE_INLINE bool btMutexTryLock( btSpinMutex* ) {return true;}
-unsigned int btGetCurrentThreadIndex();
-const unsigned int BT_MAX_THREAD_COUNT = 64;
 #endif