IPP 7 H264VideoDecoder high CPU patch

[Steve_Browne]

Since my last report of high CPU when using multiple threads with the H264VideoDecoder got dismissed as being an OpenMP issue. I figured I would start a new thread to make it easier for people to find the real solution. Hopefully this will get fixed in a newer IPP update as I would consider this a pretty major problem.

See the attached patch file for how to fix it.

[Chao_Y_Intel]

Hello,

Thanks for sharing it. We will have further check.

Regards,
Chao

[Arvin_Chan]

I encountered the same issue as well and fixed it in the following way:

// umc_h264_task_broker.cpp line 1751

bool TaskBrokerTwoThread::GetNextTaskInternal(H264Task *pTask)

{

...

if (!pTask->m_iThreadNumber)

{

SwitchCurrentAU();

m_completedQueue.clear();

return false;

}

// Only signal threads if there are tasks

Ipp32s cnt_tasks = GetNumberOfTasks(true);

if (cnt_tasks)

AwakeThreads();

...

}

Note, this does lead to a hang when closing the H264 decoder which I then fixed as follows

// umc_h264_task_supplier.cpp line 1027

void TaskSupplier::Close()

{

m_threadGroup.Release();

if (m_pTaskBroker)

{

m_pTaskBroker->Release();

}

// from reset

//m_threadGroup.Release();

...

}

// umc_h264_thread.cpp line 121

void H264Thread::Release()

{

m_bQuit = true;

Reset();

if (IsValid())

{

//m_bQuit = true;

Awake();

Wait();

}

Close();

} // void Release(void)

[j_miles]

Hi Chao,

Can the Intel team verify the findings and the solution outlined by Steve (or Arvin)? Has a fix been included in IPP 7.1 beta?

And is this issue really related to (or equal to) the one outlined in thread #81357 ("H.264 performence issues in IPP 7.0.2"), which seems very similar and is announced as having been fixed in 7.1 beta?

Thanks.

Best regards,

- Jay