QVideoFrame bits() access too slow

[HAWK0044]

Hello, I'am using QMediaPlayer + derived video surface with present function override. Frame mapping takes 0ms to process, but if I want access bits(), copy them into GPU memory, or just copy to another byte-bufer (created and allocated in code) it takes about 20ms on my hardware (laptop, core i5-4210M, GeForce 940M/Intel HD Graphics 4600). So 20ms is not a big time, it's ok, also tried on another PCs and laptops it's just fine. But if I'am trying to run this app on specific PC (Radeon HD7970, Core i7-4770) it takes about 2000ms, two seconds to copy bufer! If I'am trying to copy another byte-bufer (not from QVideoFrame) it's ok and takes about 2-4ms. So the problem only with mapped QVideoFrame (even if I wrapping it with QImage).

there is an example (I've already tried to copy from videoFrame.bits(), and unmap after coping data, have no effect)

bool QMyVideoSurface::present(const QVideoFrame &frame){
...
QVideoFrame videoFrame(frame);

    if(videoFrame.map(QAbstractVideoBuffer::ReadWrite))
    {
        if(frame.pixelFormat() != QVideoFrame::Format_ARGB32){
            return true;
        }

        image = QImage(videoFrame.bits(),
                           videoFrame.width(),
                           videoFrame.height(),
                           videoFrame.bytesPerLine(), QImage::Format_ARGB32_Premultiplied);

        videoFrame.unmap();

        // bufer allocation skipped here.

        QElapsedTimer t;
        t.start();

        buf = (uchar*)image.bits();

        for(int i=0;i<image.byteCount();i++){
            imgBuf[i] = buf[i];
        }

        qDebug() << "Taken" << t.elapsed(); //shows 20ms on my laptop, and 2000ms on powerful PC
    }
...
}

Any suggestions are welcome. Please, help me, deadline is near.

UDP:
If I trying to create static bufer in .map(){} scope, e.g:

if(videoFrame.map(QAbstractVideoBuffer::ReadWrite))
{
...
   uchar abuf[4096*4096*4];
   memcpy(aib, videoFrame.bits(), videoFrame.mappedBytes());
...
}

then this taking about 1-2ms. But if I moving abuf declaration out of this scope (for example placing it in class definition, or in the top of file: then memcpy will take about 2000ms (actually depends on video resolution)

[SGaist]

Hi,

Do you really need to map your video frame as ReadWrite ?

Where do you need these data to go to ?

[HAWK0044]

@SGaist Tried ReadOnly too. The solve is change compiler. Looks like it's MSVC2015 x64 Qt 5.6 issue, with Qt 5.2.0 MinGW x86 it works perfectly.

[SGaist]

Then it sounds like a regression… Can you check against the latest 5.7 to see if it still happens ?