parallel_for_each casue memory leak!

Question

 I just wrote a few code about PPL, as following:

        vector<wstring> testPath, retPath;
        testPath.push_back(L"niha");
        testPath.push_back(L"nnnn");
        parallel_for_each(testPath.begin(),testPath.end(),
            [/*retPath*/](wstring path)
        {
            // TODO:
        });

After dubugging those code , memory leak happened. I tried google for the root cause, but found nothing of help?

Does anyone ever get the same question?

Answer 1

 I just wrote a few code about PPL, as following:

        vector<wstring> testPath, retPath;
        testPath.push_back(L"niha");
        testPath.push_back(L"nnnn");
        parallel_for_each(testPath.begin(),testPath.end(),
            [/*retPath*/](wstring path)
        {
            // TODO:
        });

After dubugging those code , memory leak happened. I tried google for the root cause, but found nothing of help?

Does anyone ever get the same question?

I got this problem when I develop a MFC application.

Answer 2

I get memory leaks too using parallel_for in a very simplified test console application in VS2010:

 

void test(int t)
{
   for (int i = 0; i < t; i++)
      i = i;
}

int _tmain(int argc, _TCHAR* argv[])
{
   _CrtSetDbgFlag(_CrtSetDbgFlag(_CRTDBG_REPORT_FLAG) | _CRTDBG_LEAK_CHECK_DF);
   Concurrency::parallel_for(0, 20000, test);
   return 0;
}

 

Using _crtBreakAlloc, it seems that the ThreadScheduler created in the following code (in ThreadScheduler.cpp) is never released:

 

 ThreadScheduler* ThreadScheduler::Create(__in const ::Concurrency::SchedulerPolicy& policy)
    {
        return new ThreadScheduler(policy);
    }

 

Am I missing something or is there a bug in the PPL?

Answer 3

AceLencho,

Apologies for the delayed response.

In VS 2010 some memory allocations were falsely reported as leaks by the CRT error reporting mechanism. This issues was fixed in the SP1.

---

Artur Laksberg - MSFT