分析HM代码,首先必须把class TComDataCU完全搞明白!

分析HM代码,首先必须把class TComDataCU完全搞明白!

为了找到合适的模式或者合适的分割, 基本上都是采用两个TComDataCU来操作, TComDataCU*& rpcBestCU, TComDataCU*& rpcTempCU.

相当于两个指针对应两个乒乓buffer,  m_ppcBestCU[uiDepth], m_ppcTempCU[uiDepth], 当然每一个uiDepth都有一个对应的BestCU和TempCU的buffer.

 

 

class TComDataCU
{
private:
  
  // -------------------------------------------------------
  // class pointers
  // -------------------------------------------------------
  TComPic*      m_pcPic;              ///< picture class pointer
  TComSlice*    m_pcSlice;            ///< slice header pointer
  TComPattern*  m_pcPattern;          ///< neighbour access class pointer
  // -------------------------------------------------------
  // CU description
  // -------------------------------------------------------
  UInt          m_uiCUAddr;           ///< CU address in a slice
  UInt          m_uiAbsIdxInLCU;      ///< absolute address in a CU. It's Z scan order
  UInt          m_uiCUPelX;           ///< CU position in a pixel (X)
  UInt          m_uiCUPelY;           ///< CU position in a pixel (Y)
  UInt          m_uiNumPartition;     ///< total number of minimum partitions in a CU
  UChar*        m_puhWidth;           ///< array of widths
  UChar*        m_puhHeight;          ///< array of heights
  UChar*        m_puhDepth;           ///< array of depths
  Int           m_unitSize;           ///< size of a "minimum partition"
  // -------------------------------------------------------
  // CU data
  // -------------------------------------------------------
  Bool*         m_skipFlag;           ///< array of skip flags
  Char*         m_pePartSize;         ///< array of partition sizes
  Char*         m_pePredMode;         ///< array of prediction modes
  Bool*         m_CUTransquantBypass;   ///< array of cu_transquant_bypass flags
  Char*         m_phQP;               ///< array of QP values
  UChar*        m_puhTrIdx;           ///< array of transform indices
  UChar*        m_puhTransformSkip[3];///< array of transform skipping flags
  UChar*        m_puhCbf[3];          ///< array of coded block flags (CBF)
  TComCUMvField m_acCUMvField[2];     ///< array of motion vectors
  TCoeff*       m_pcTrCoeffY;         ///< transformed coefficient buffer (Y)
  TCoeff*       m_pcTrCoeffCb;        ///< transformed coefficient buffer (Cb)
  TCoeff*       m_pcTrCoeffCr;        ///< transformed coefficient buffer (Cr)
#if ADAPTIVE_QP_SELECTION
  Int*          m_pcArlCoeffY;        ///< ARL coefficient buffer (Y)
  Int*          m_pcArlCoeffCb;       ///< ARL coefficient buffer (Cb)
  Int*          m_pcArlCoeffCr;       ///< ARL coefficient buffer (Cr)
  bool          m_ArlCoeffIsAliasedAllocation; ///< ARL coefficient buffer is an alias of the global buffer and must not be free()'d
  static Int*   m_pcGlbArlCoeffY;     ///< ARL coefficient buffer (Y)
  static Int*   m_pcGlbArlCoeffCb;    ///< ARL coefficient buffer (Cb)
  static Int*   m_pcGlbArlCoeffCr;    ///< ARL coefficient buffer (Cr)
#endif
  Pel*          m_pcIPCMSampleY;      ///< PCM sample buffer (Y)
  Pel*          m_pcIPCMSampleCb;     ///< PCM sample buffer (Cb)
  Pel*          m_pcIPCMSampleCr;     ///< PCM sample buffer (Cr)
  Int*          m_piSliceSUMap;       ///< pointer of slice ID map
  std::vector<NDBFBlockInfo> m_vNDFBlock;
 .................. 
 .............
}

 

上面是HM9.0的class TComDataCU的部分代码, 那些地方有疑问大家可以都贴出来一起讨论啊!