H.264 stream decoding issue

[guerzoni__filippo]

Hi,

I modified the sample_multi transcode example to decode H.264 image sequences (streams) to raw NV12.

Streams are sent via RTP-UDP to the decoder which reconstructs bitstreams heades and FU NALS and dumps incoming data to file.

I noticed that after a while (400 frames) the decoding always stops working with va_res=VA_STATUS_ERROR_DECODING_ERROR(0x17) that gets remapped to mfx_res=MFX_ERR_UNKNOWN(-1):

va_res = m_libva->vaSyncSurface(m_dpy, *(vaapi_mid->m_surface));                     (LockFrame() - vaapi_allocator.cpp) 
    ...
    sts = Surface2BS(&VppExtSurface, &m_BSPool.back()->Bitstream,m_mfxVppParams.vpp.Out.FourCC);     (Transcode() - pipeline-transcode.cpp)
        ...
        while (MFX_ERR_NONE == pContext->transcodingSts)                    (ThranscodeRoutine() - pipeline-transcode.cpp)

and the subsequent call to vaQuerySurfaceError returns:
{status = -670922364, start_mb = 32767, end_mb = 0, decode_error_type = VADecodeSliceMissing, num_mb = 0}

I'm confident about the received compressed data because calling unmodified sample_multi_transcode on the dumped file does not returns any errors.

What sounds strange to me is start_mb = 32767, end_mb = 0 on the vaQuerySurfaceError returned value. It seems like an overflow on macroblocks.

I would like to solve the issue but even a workaround to recover the decoder without loosing data would be fine.

I'm on a Ubuntu 16.4 machine running Intel® Media Server Studio – Community Edition    Version 2017 (2017_R2).

Thanks in advance

[guerzoni__filippo · Accepted solution]

Solved, the issue is related to application specific code.

The buffer I'm sending to decoder sometimes gets corrupted and decoder complains about that.