Error in code

[pHySiQuE]

Around line 59 of dgThread.cpp, change this:

Code: Select all

#ifdef DG_THREAD_EMULATION

dgThread::dgSemaphore::dgSemaphore ()
   :m_sem (0)
{
}

To this:

Code: Select all

#ifdef DG_THREAD_EMULATION

dgThread::dgSemaphore::dgSemaphore ()
   :m_sem ()
{
}

Android NDK doesn't like using the 0 for a pointer.

[Julio Jerez]

but the secudn mode will leave un inlitlizes

can to try m_sem (NULL)

[pHySiQuE]

Using NULL gives me this error:

/svn/Leadwerks3D/Engine/Projects/Android2.2b/jni/../../../Source/../../../NewtonDynamics/coreLibrary_300/source/core/dgThread.cpp: In constructor 'dgThread::dgSemaphore::dgSemaphore()':
/svn/Leadwerks3D/Engine/Projects/Android2.2b/jni/../../../Source/../../../NewtonDynamics/coreLibrary_300/source/core/dgThread.cpp:62: error: no matching function for call to 'sem_t::sem_t(NULL)'
/Applications/AndroidSDK/android-ndk-r7/platforms/android-8/arch-arm/usr/include/semaphore.h:37: note: candidates are: sem_t::sem_t(const sem_t&)
/Applications/AndroidSDK/android-ndk-r7/platforms/android-8/arch-arm/usr/include/semaphore.h:37: note: sem_t::sem_t()

[Julio Jerez]

Ok I change dot this

Code: Select all

dgThread::dgSemaphore::dgSemaphore ()
{
   m_sem = NULL;
}

sinc and comple again.
If there any reason why you are compiling with thread in emulations. does not the android sdk supports POSIT threads?

[pHySiQuE]

You're right, threads are working fine on Android.

[Julio Jerez]

how many core does android has?

but even with a single core, I found that using twice as many threads per core yield better result. I can not explain why but that seems to be the case in window and linux.

[pHySiQuE]

Okay, threading works on Mac and Android. On iOS, both the simulator and the actual device, the program stops with the error "Program received signal 'SIGABRT" on TickCallback in this function:

Code: Select all

void dgMutexThread::Execute (dgInt32 threadID)
{
   // suspend this tread until the call thread decide to
   _ASSERTE (threadID == m_id);
   while (!m_terminate) {
      // wait for the main thread to signal an update
      dgInterlockedExchange(&m_isBusy, 0);
      SuspendExecution(m_myMutex);
      dgInterlockedExchange(&m_isBusy, 1);
      if (!m_terminate) {
         TickCallback(threadID); // Program received signal '"SIGABRT"
         // let main thread resume execution
         m_callerMutex.Release();
      }
   }
   dgInterlockedExchange(&m_isBusy, 0);
}

Code: Select all

pure virtual method called[Switching to process 9219 thread 0x2403]
[Switching to process 9219 thread 0x2403]
Single stepping until exit from function __pthread_kill,
which has no line number information.
warning: Remote failure reply: E37
Current language:  auto; currently c++
Single stepping until exit from function abort,
which has no line number information.
pure virtual method called[Switching to process 9731 thread 0x2603]
[Switching to process 9731 thread 0x2603]

I can live without threading on iOS, but it is strange.

[carli2]

how many core does android has?

Depends on which ARM processor you use.
There are up to 4 cores in the latest Tegra chips.

but even with a single core, I found that using twice as many threads per core yield better result. I can not explain why but that seems to be the case in window and linux.

Does your CPU have hyper threading? (All intel except atom have)

[Julio Jerez]

yes My linux machine in a 2.4 pentium 4 hypertreading. But I had never seem multtreaded code or sse code actualy running faster than simple x87 until wrote the new dgTread class for core 300.
as far as I remember I was so disappointed with hypertreading than I dissabled at the bios level.

I am very saitified with the new dgthread class, it is sleep free, in core 200 it was not and after 2 maybe 4 threads It got dimishing returns.
now core 300 the ghread count can be unlimited. I put a limit to 16 but it is an arbirtrary number.
in core 300 in my sytem Intel7 each time I double the thread count the performance almost double until it reaches 8 threads, after that point it still gain performance but it is marginal. it never slower witch was the case before.
for this reason now Newton is thread even in a single core. unless it is compiled with thread emulation on

[Julio Jerez]

pHySiQuE does than happens on Mac OSX or Iphone?

[pHySiQuE]

Hmmm, I guess it happens on all of these:
-OSX
-iOS simulator
-iOS device

It's complaining about a "pure virtual method called" but I don't see why. In the debugger, it says the threadID is 0.

[Julio Jerez]

Ok, Iit is probably teh atimic not working right in OSX, I am not using OSX specifc atomics, inestead GCC specific onces.

can you build the mac without thread for now, until I have time to debug that.
I am writing a native mac framework fo the Mac but I put in hold for teh momnet until O move the ething I am workng at the momnet a litle further alone.

[pHySiQuE]

Sure, no prob.

[pHySiQuE]

In the latest version, I have DG_USE_THREAD_EMULATION defined. (I am attempting to build without threads.)

I get a crash in Xcode that says "Thread 13: signal SIGABRT" on line 62 of dgMutexThread:

Screen Shot 2013-01-23 at 6.55.48 PM.jpg (155.81 KiB) Viewed 5612 times

[Julio Jerez]

if you defined DG_USE_THREAD_EMULATION you should had a compiler error, in thgsio function

Code: Select all

void dgThreadHive::QueueJob (dgWorkerThreadTaskCallback callback, void* const context0, void* const context1)
{
   if (!m_beesCount) {
      callback (context0, context1, 0);
   } else {
      #ifdef DG_USE_THREAD_EMULATION
         callback (context0, context1, 0);
      #else
         dgThreadJob job (context0, context1, callback);
         m_jobsPool.Push(job);
         if (m_jobsPool.IsFull()) {
            SynchronizationBarrier ();
         }
      #endif
   }
}

I my last check int I modified the thread code to pass two parameters, primary and secundady, this is to simply the high level interface.
only one paratemt is hard to mass stiff like call pointer and data context without having to do lot of extra dat copied.

I forget to compile the non threaded version. It is fixed now.