Cuda ffmpeg上的视频解码器

时间:2013-04-25 19:37:25

标签: cuda ffmpeg nvidia decoder libav

我开始实现使用cuda HW解码器生成YUV帧以便对其进行编码的custum视频解码器。

如何填写“CUVIDPICPARAMS”struc ??? 有可能吗?

我的算法是:

获取视频流数据包我使用的是ffmpeg-dev libs avcodec,avformat ...

我的步骤:

1)打开输入文件:

avformat_open_input(&ff_formatContext,in_filename,nullptr,nullptr);

2)获取视频流属性:

avformat_find_stream_info(ff_formatContext,nullptr);

3)获取视频流:

ff_video_stream=ff_formatContext->streams[i];

4)获取CUDA设备并启动它:

cuDeviceGet(&cu_device,0);
CUcontext cu_vid_ctx;

5)初始化视频CUDA解码器并设置创建参数:

CUVIDDECODECREATEINFO *cu_decoder_info=new CUVIDDECODECREATEINFO;
memset(cu_decoder_info,0,sizeof(CUVIDDECODECREATEINFO));
...
cuvidCreateDecoder(cu_video_decoder,cu_decoder_info);

6)将帧数据读取到AVpacket

av_read_frame(ff_formatContext,ff_packet);

现在我需要在CUDA视频解码器上解码帧数据包,理论上是:

cuvidDecodePicture(pDecoder,&picParams);

但在我需要填写CUVIDPICPARAMS之前

CUVIDPICPARAMS picParams; // =新CUVIDPICPARAMS; memset(& picParams,0,sizeof(CUVIDPICPARAMS));

HOW CAN I FILL "CUVIDPICPARAMS" struc ???

typedef struct _CUVIDPICPARAMS
{
    int PicWidthInMbs;      // Coded Frame Size
    int FrameHeightInMbs;   // Coded Frame Height
    int CurrPicIdx;         // Output index of the current picture
    int field_pic_flag;     // 0=frame picture, 1=field picture
    int bottom_field_flag;  // 0=top field, 1=bottom field (ignored if field_pic_flag=0)
    int second_field;       // Second field of a complementary field pair
    // Bitstream data
    unsigned int nBitstreamDataLen;        // Number of bytes in bitstream data buffer
    const unsigned char *pBitstreamData;   // Ptr to bitstream data for this picture (slice-layer)
    unsigned int nNumSlices;               // Number of slices in this picture
    const unsigned int *pSliceDataOffsets; // nNumSlices entries, contains offset of each slice within the bitstream data buffer
    int ref_pic_flag;       // This picture is a reference picture
    int intra_pic_flag;     // This picture is entirely intra coded
    unsigned int Reserved[30];             // Reserved for future use
    // Codec-specific data
    union {
        CUVIDMPEG2PICPARAMS mpeg2;          // Also used for MPEG-1
        CUVIDH264PICPARAMS h264;
        CUVIDVC1PICPARAMS vc1;
        CUVIDMPEG4PICPARAMS mpeg4;
        CUVIDJPEGPICPARAMS jpeg;
        unsigned int CodecReserved[1024];
    } CodecSpecific;
} CUVIDPICPARAMS;

typedef struct _CUVIDH264PICPARAMS
{
    // SPS
    int log2_max_frame_num_minus4;
    int pic_order_cnt_type;
    int log2_max_pic_order_cnt_lsb_minus4;
    int delta_pic_order_always_zero_flag;
    int frame_mbs_only_flag;
    int direct_8x8_inference_flag;
    int num_ref_frames;             // NOTE: shall meet level 4.1 restrictions
    unsigned char residual_colour_transform_flag;
    unsigned char bit_depth_luma_minus8;    // Must be 0 (only 8-bit supported)
    unsigned char bit_depth_chroma_minus8;  // Must be 0 (only 8-bit supported)
    unsigned char qpprime_y_zero_transform_bypass_flag;
    // PPS
    int entropy_coding_mode_flag;
    int pic_order_present_flag;
    int num_ref_idx_l0_active_minus1;
    int num_ref_idx_l1_active_minus1;
    int weighted_pred_flag;
    int weighted_bipred_idc;
    int pic_init_qp_minus26;
    int deblocking_filter_control_present_flag;
    int redundant_pic_cnt_present_flag;
    int transform_8x8_mode_flag;
    int MbaffFrameFlag;
    int constrained_intra_pred_flag;
    int chroma_qp_index_offset;
    int second_chroma_qp_index_offset;
    int ref_pic_flag;
    int frame_num;
    int CurrFieldOrderCnt[2];
    // DPB
    CUVIDH264DPBENTRY dpb[16];          // List of reference frames within the DPB
    // Quantization Matrices (raster-order)
    unsigned char WeightScale4x4[6][16];
    unsigned char WeightScale8x8[2][64];
    // FMO/ASO
    unsigned char fmo_aso_enable;
    unsigned char num_slice_groups_minus1;
    unsigned char slice_group_map_type;
    signed char pic_init_qs_minus26;
    unsigned int slice_group_change_rate_minus1;
    union
    {
        unsigned long long slice_group_map_addr;
        const unsigned char *pMb2SliceGroupMap;
    } fmo;
    unsigned int  Reserved[12];
    // SVC/MVC
    union
    {
        CUVIDH264MVCEXT mvcext;
        CUVIDH264SVCEXT svcext;
    };
} CUVIDH264PICPARAMS;

1 个答案:

答案 0 :(得分:3)

这是CUvideoparser对象的目的。您通过cuvidParseVideoData逐帧向数据流提供数据流,当CUVIDPICPARAMS检测到已准备好完整的帧时,它会通过{{1}}回拨给您,并准备传递给解码器。

D3D9解码样本中提供了所有这些以及更多内容,可用here。我建议详细研究它,因为除此之外没有太多关于此API的文档。