What am I doing wrong? (multithreading)

Question

Here s what I'm doing in a nutshell.

In my class's cpp file I have:

std::vector<std::vector<GLdouble>> ThreadPts[4];

The thread proc looks like this:

unsigned __stdcall BezierThreadProc(void *arg)
{
    SHAPETHREADDATA *data = (SHAPETHREADDATA *) arg;

    OGLSHAPE *obj = reinterpret_cast<OGLSHAPE*>(data->objectptr);
    for(unsigned int i = data->start; i < data->end - 1; ++i)
    {


        obj->SetCubicBezier(
            obj->Contour[data->contournum].UserPoints[i],
            obj->Contour[data->contournum].UserPoints[i + 1],
            data->whichVector);

    }

    _endthreadex( 0 );
    return 0;

}

SetCubicBezier looks like this:

void OGLSHAPE::SetCubicBezier(USERFPOINT &a,USERFPOINT &b, int &currentvector )
{
    std::vector<GLdouble> temp;
    if(a.RightHandle.x == a.UserPoint.x && a.RightHandle.y == a.UserPoint.y 
        && b.LeftHandle.x == b.UserPoint.x && b.LeftHandle.y == b.UserPoint.y )
    {
        temp.clear();
        temp.push_back((GLdouble)a.UserPoint.x);
        temp.push_back((GLdouble)a.UserPoint.y);

        ThreadPts[currentvector].push_back(temp);
        temp.clear();
        temp.push_back((GLdouble)b.UserPoint.x);
        temp.push_back((GLdouble)b.UserPoint.y);


        ThreadPts[currentvector].push_back(temp);

    }
}

The code that calls the threads looks like this:

for(int i = 0; i < Contour.size(); ++i)
{
    Contour[i].DrawingPoints.clear();

 if(Contour[i].UserPoints.size() < 2)
 {
     break;
 }

HANDLE hThread[4];
SHAPETHREADDATA dat;
dat.objectptr = (void*)this;
dat.start = 0;
dat.end = floor((Contour[i].UserPoints.size() - 1) * 0.25);
dat.whichVector = 0;
dat.contournum = i;



hThread[0] = (HANDLE)_beginthreadex(NULL,0,&BezierThreadProc,&dat,0,0);

dat.start = dat.end;
dat.end = floor((Contour[i].UserPoints.size() - 1) * 0.5);
dat.whichVector = 1;

hThread[1] = (HANDLE)_beginthreadex(NULL,0,&BezierThreadProc,&dat,0,0);

dat.start = dat.end;
dat.end = floor((Contour[i].UserPoints.size() - 1) * 0.75);
dat.whichVector = 2;

hThread[2] = (HANDLE)_beginthreadex(NULL,0,&BezierThreadProc,&dat,0,0);

dat.start = dat.end;
dat.end = Contour[i].UserPoints.size();
dat.whichVector = 3;

hThread[3] = (HANDLE)_beginthreadex(NULL,0,&BezierThreadProc,&dat,0,0);

WaitForMultipleObjects(4,hThread,true,INFINITE);

}

Is there something wrong with this? I'd expect it to fill ThreadPts[4]; ... There should never be any conflicts the way I have it set up. I usually get error writing at... on the last thread where dat->whichvector = 3. If I remove:

dat.start = dat.end;
dat.end = Contour[i].UserPoints.size();
dat.whichVector = 3;

hThread[3] = (HANDLE)_beginthreadex(NULL,0,&BezierThreadProc,&dat,0,0);

Then it does not seem to crash, what could be wrong?

Thanks

Answer 1

The problem is that you're passing the same dat structure to each thread as the argument to the threadproc.

For example, When you start thread 1, there's no guarantee that it will have read the information in the dat structure before your main thread starts loading that same dat structure with the information for thread 2 (and so on). In fact, you're constantly directly using that dat structure throughout the thread's loop, so the thread won't be finished with the structure passed to it until the thread is basically done with all its work.

Also note that currentvector in SetCubicBezier() is a reference to data->whichVector, which is referring to the exact same location in a threads. So SetCubicBezier() will be performing push_back() calls on the same object in separate threads because of this.

There's a very simple fix: you should use four separate SHAPETHREADDATA instances - one to initialize each thread.