VideoParser.cpp from the cudaDecodeGL example. Here I directly feed the a parsed frame into the decoder and when I the display callback is called I try to map the frame ... which always fails except the first 2 calls.

VideoParser.cpp

int
CUDAAPI
VideoParser::HandlePictureDecode(void *pUserData, CUVIDPICPARAMS *pPicParams)
{
    VideoParserData *pParserData = reinterpret_cast<VideoParserData *>(pUserData);

    printf("CUVIDPICPARAMS.frame_num: %d\n", pPicParams->CodecSpecific.h264.frame_num);

#if 0    
    bool bFrameAvailable = pParserData->pFrameQueue->waitUntilFrameAvailable(pPicParams->CurrPicIdx);

    if (!bFrameAvailable)
        return false;

    if (pParserData->pVideoDecoder->decodePicture(pPicParams, pParserData->pContext) != CUDA_SUCCESS)
    {
        return false;
    }
#else
    if (pParserData->pVideoDecoder->decodePicture(pPicParams, pParserData->pContext) != CUDA_SUCCESS)
    {
      printf("ERROR: failed to decode the picture. \n");
      return false;
    }
#endif
    return true;
}

int
CUDAAPI
VideoParser::HandlePictureDisplay(void *pUserData, CUVIDPARSERDISPINFO *pPicParams)
{
    // std::cout << *pPicParams << std::endl;

    VideoParserData *pParserData = reinterpret_cast<VideoParserData *>(pUserData);
    
#if 0  
    pParserData->pFrameQueue->enqueue(pPicParams);
#else
    /* @todo testing by directly mapping the image. */
    CUVIDPROCPARAMS vpp;
    CUdeviceptr device_ptr = 0;
    unsigned int pitch = 0;
    CUresult r = CUDA_SUCCESS;

    memset(&vpp, 0, sizeof(CUVIDPROCPARAMS));
    vpp.progressive_frame = pPicParams->progressive_frame;
    vpp.top_field_first = pPicParams->top_field_first;
    vpp.unpaired_field = (pPicParams->repeat_first_field < 0);
    
    r = pParserData->pVideoDecoder->mapFrame(pPicParams->picture_index, &device_ptr, &pitch,&vpp);
    if (CUDA_SUCCESS == r) {
      printf("---------------------------------------------------------\n");
    }
    else {
      printf(":-(\n");
    }
    
#endif

    return 1;
}