skl_mb_c.cpp

mpeg4编解码器

extern SKL_MB_QP_PASS_SIGNATURE(Skl_V_Pass_Avrg_Up_8_Add_Rnd0_MMX);

SKL_MB_FUNCS Skl_MB_Funcs_Copy_Rnd1_MMX = {
  { Skl_Copy_16x8_FF_MMX, Skl_Copy_16x8_FH_Rnd1_MMX, Skl_Copy_16x8_HF_Rnd1_MMX, Skl_Copy_16x8_HH_Rnd1_MMX }
, { Skl_Copy_8x8_FF_MMX,  Skl_Copy_8x8_FH_Rnd1_MMX,  Skl_Copy_8x8_HF_Rnd1_MMX,  Skl_Copy_8x8_HH_Rnd1_MMX }
, { Skl_Copy_8x4_FF_MMX,  Skl_Copy_8x4_FH_Rnd1_MMX,  Skl_Copy_8x4_HF_Rnd1_MMX,  Skl_Copy_8x4_HH_Rnd1_MMX  }
,   Skl_H_Pass_16_Copy_Rnd1_MMX, Skl_H_Pass_Avrg_16_Copy_Rnd1_MMX, Skl_H_Pass_Avrg_Up_16_Copy_Rnd1_MMX
,   Skl_H_Pass_16_Copy_Rnd1_MMX, Skl_H_Pass_Avrg_16_Copy_Rnd1_MMX, Skl_H_Pass_Avrg_Up_16_Copy_Rnd1_MMX
,   Skl_V_Pass_16_Copy_Rnd1_MMX, Skl_V_Pass_Avrg_16_Copy_Rnd1_MMX, Skl_V_Pass_Avrg_Up_16_Copy_Rnd1_MMX
,   Skl_H_Pass_8_Copy_Rnd1_MMX,  Skl_H_Pass_Avrg_8_Copy_Rnd1_MMX,  Skl_H_Pass_Avrg_Up_8_Copy_Rnd1_MMX
,   Skl_H_Pass_8_Copy_Rnd1_MMX,  Skl_H_Pass_Avrg_8_Copy_Rnd1_MMX,  Skl_H_Pass_Avrg_Up_8_Copy_Rnd1_MMX
,   Skl_V_Pass_8_Copy_Rnd1_MMX,  Skl_V_Pass_Avrg_8_Copy_Rnd1_MMX,  Skl_V_Pass_Avrg_Up_8_Copy_Rnd1_MMX
,   Skl_SAD_HP_16x16_Rnd1_C, Skl_SAD_HP_8x8_Rnd1_C
};
SKL_MB_FUNCS Skl_MB_Funcs_Copy_Rnd0_MMX = {
  { Skl_Copy_16x8_FF_MMX, Skl_Copy_16x8_FH_Rnd0_MMX, Skl_Copy_16x8_HF_Rnd0_MMX, Skl_Copy_16x8_HH_Rnd0_MMX }
, { Skl_Copy_8x8_FF_MMX,  Skl_Copy_8x8_FH_Rnd0_MMX,  Skl_Copy_8x8_HF_Rnd0_MMX,  Skl_Copy_8x8_HH_Rnd0_MMX }
, { Skl_Copy_8x4_FF_MMX,  Skl_Copy_8x4_FH_Rnd0_MMX,  Skl_Copy_8x4_HF_Rnd0_MMX,  Skl_Copy_8x4_HH_Rnd0_MMX  }
,   Skl_H_Pass_16_Copy_Rnd0_MMX, Skl_H_Pass_Avrg_16_Copy_Rnd0_MMX, Skl_H_Pass_Avrg_Up_16_Copy_Rnd0_MMX
,   Skl_H_Pass_16_Copy_Rnd0_MMX, Skl_H_Pass_Avrg_16_Copy_Rnd0_MMX, Skl_H_Pass_Avrg_Up_16_Copy_Rnd0_MMX
,   Skl_V_Pass_16_Copy_Rnd0_MMX, Skl_V_Pass_Avrg_16_Copy_Rnd0_MMX, Skl_V_Pass_Avrg_Up_16_Copy_Rnd0_MMX
,   Skl_H_Pass_8_Copy_Rnd0_MMX,  Skl_H_Pass_Avrg_8_Copy_Rnd0_MMX,  Skl_H_Pass_Avrg_Up_8_Copy_Rnd0_MMX
,   Skl_H_Pass_8_Copy_Rnd0_MMX,  Skl_H_Pass_Avrg_8_Copy_Rnd0_MMX,  Skl_H_Pass_Avrg_Up_8_Copy_Rnd0_MMX
,   Skl_V_Pass_8_Copy_Rnd0_MMX,  Skl_V_Pass_Avrg_8_Copy_Rnd0_MMX,  Skl_V_Pass_Avrg_Up_8_Copy_Rnd0_MMX
,   Skl_SAD_HP_16x16_Rnd0_C, Skl_SAD_HP_8x8_Rnd0_C
};
SKL_MB_FUNCS Skl_MB_Funcs_Add_Rnd0_MMX = {
  { Skl_Add_16x8_FF_MMX, Skl_Add_16x8_FH_Rnd0_MMX, Skl_Add_16x8_HF_Rnd0_MMX, Skl_Add_16x8_HH_Rnd0_MMX }
, { Skl_Add_8x8_FF_MMX,  Skl_Add_8x8_FH_Rnd0_MMX,  Skl_Add_8x8_HF_Rnd0_MMX,  Skl_Add_8x8_HH_Rnd0_MMX }
, { Skl_Add_8x4_FF_MMX,  Skl_Add_8x4_FH_Rnd0_MMX,  Skl_Add_8x4_HF_Rnd0_MMX,  Skl_Add_8x4_HH_Rnd0_MMX  }
,   Skl_H_Pass_16_Add_Rnd0_MMX,  Skl_H_Pass_Avrg_16_Add_Rnd0_MMX,  Skl_H_Pass_Avrg_Up_16_Add_Rnd0_MMX
,   Skl_H_Pass_16_Copy_Rnd0_MMX, Skl_H_Pass_Avrg_16_Copy_Rnd0_MMX, Skl_H_Pass_Avrg_Up_16_Copy_Rnd0_MMX
,   Skl_V_Pass_16_Add_Rnd0_MMX,  Skl_V_Pass_Avrg_16_Add_Rnd0_MMX,  Skl_V_Pass_Avrg_Up_16_Add_Rnd0_MMX
,   Skl_H_Pass_8_Add_Rnd0_MMX,   Skl_H_Pass_Avrg_8_Add_Rnd0_MMX,   Skl_H_Pass_Avrg_Up_8_Add_Rnd0_MMX
,   Skl_H_Pass_8_Copy_Rnd0_MMX,  Skl_H_Pass_Avrg_8_Copy_Rnd0_MMX,  Skl_H_Pass_Avrg_Up_8_Copy_Rnd0_MMX
,   Skl_V_Pass_8_Add_Rnd0_MMX,   Skl_V_Pass_Avrg_8_Add_Rnd0_MMX,   Skl_V_Pass_Avrg_Up_8_Add_Rnd0_MMX
,   Skl_SAD_HP_16x16_Rnd0_C, Skl_SAD_HP_8x8_Rnd0_C
};
extern MB_FUNC(Skl_H_Pass_2Taps_MMX);
extern MB_FUNC(Skl_V_Pass_2Taps_MMX);
extern MB_FUNC(Skl_HV_Pass_2Taps_MMX);

const SKL_HV_FILTER Skl_Filter_2_MMX = {
  Skl_H_Pass_2Taps_MMX, Skl_V_Pass_2Taps_MMX, Skl_HV_Pass_2Taps_MMX
};


//////////////////////////////////////////////////////////
// tables sent to ASM code

extern SKL_INT16 Skl_MMX_Expand[256][4];

  // 17 tables, 2K each => 34K
  // Mirroring can be acheived composing 11 basic tables
  // (for instance: (23,19,-6,3)=(20,20,-6,3)+(3,-1,0,0)
  // Using Symmetries (and bswap) could reduce further
  // the memory to 7 tables (->14K).

extern SKL_INT16 Skl_FIR_1_0_0_0[256][4];
extern SKL_INT16 Skl_FIR_3_1_0_0[256][4];
extern SKL_INT16 Skl_FIR_6_3_1_0[256][4];
extern SKL_INT16 Skl_FIR_14_3_2_1[256][4];
extern SKL_INT16 Skl_FIR_20_6_3_1[256][4];
extern SKL_INT16 Skl_FIR_20_20_6_3[256][4];
extern SKL_INT16 Skl_FIR_23_19_6_3[256][4];
extern SKL_INT16 Skl_FIR_7_20_20_6[256][4];
extern SKL_INT16 Skl_FIR_6_20_20_6[256][4];
extern SKL_INT16 Skl_FIR_6_20_20_7[256][4];
extern SKL_INT16 Skl_FIR_3_6_20_20[256][4];
extern SKL_INT16 Skl_FIR_3_6_19_23[256][4];
extern SKL_INT16 Skl_FIR_1_3_6_20[256][4];
extern SKL_INT16 Skl_FIR_1_2_3_14[256][4];
extern SKL_INT16 Skl_FIR_0_1_3_6[256][4];
extern SKL_INT16 Skl_FIR_0_0_1_3[256][4];
extern SKL_INT16 Skl_FIR_0_0_0_1[256][4];

extern void Skl_Init_QP_MMX();

//////////////////////////////////////////////////////////

SKL_INT16 Skl_FIR_1_0_0_0[256][4];
SKL_INT16 Skl_FIR_3_1_0_0[256][4];
SKL_INT16 Skl_FIR_6_3_1_0[256][4];
SKL_INT16 Skl_FIR_14_3_2_1[256][4];
SKL_INT16 Skl_FIR_20_6_3_1[256][4];
SKL_INT16 Skl_FIR_20_20_6_3[256][4];
SKL_INT16 Skl_FIR_23_19_6_3[256][4];
SKL_INT16 Skl_FIR_7_20_20_6[256][4];
SKL_INT16 Skl_FIR_6_20_20_6[256][4];
SKL_INT16 Skl_FIR_6_20_20_7[256][4];
SKL_INT16 Skl_FIR_3_6_20_20[256][4];
SKL_INT16 Skl_FIR_3_6_19_23[256][4];
SKL_INT16 Skl_FIR_1_3_6_20[256][4];
SKL_INT16 Skl_FIR_1_2_3_14[256][4];
SKL_INT16 Skl_FIR_0_1_3_6[256][4];
SKL_INT16 Skl_FIR_0_0_1_3[256][4];
SKL_INT16 Skl_FIR_0_0_0_1[256][4];

SKL_INT16 Skl_MMX_Expand[256][4];
static void Init_FIR_Table(SKL_INT16 Tab[][4],
                           int A, int B, int C, int D)
{
  for(int i=0; i<256; ++i) {
    Tab[i][0] = i*A;
    Tab[i][1] = i*B;
    Tab[i][2] = i*C;
    Tab[i][3] = i*D;
  }
}

void Skl_Init_QP_MMX()
{
  for(int i=0; i<256; ++i) {
    Skl_MMX_Expand[i][0] = i;
    Skl_MMX_Expand[i][1] = i;
    Skl_MMX_Expand[i][2] = i;
    Skl_MMX_Expand[i][3] = i;
  }
  Init_FIR_Table(Skl_FIR_1_0_0_0,   -1,  0,  0,  0);
  Init_FIR_Table(Skl_FIR_3_1_0_0,    3, -1,  0,  0);
  Init_FIR_Table(Skl_FIR_6_3_1_0,   -6,  3, -1,  0);
  Init_FIR_Table(Skl_FIR_14_3_2_1,  14, -3,  2, -1);
  Init_FIR_Table(Skl_FIR_20_6_3_1,  20, -6,  3, -1);
  Init_FIR_Table(Skl_FIR_20_20_6_3, 20, 20, -6,  3);
  Init_FIR_Table(Skl_FIR_23_19_6_3, 23, 19, -6,  3);
  Init_FIR_Table(Skl_FIR_7_20_20_6, -7, 20, 20, -6);
  Init_FIR_Table(Skl_FIR_6_20_20_6, -6, 20, 20, -6);
  Init_FIR_Table(Skl_FIR_6_20_20_7, -6, 20, 20, -7);
  Init_FIR_Table(Skl_FIR_3_6_20_20,  3, -6, 20, 20);
  Init_FIR_Table(Skl_FIR_3_6_19_23,  3, -6, 19, 23);
  Init_FIR_Table(Skl_FIR_1_3_6_20,  -1,  3, -6, 20);
  Init_FIR_Table(Skl_FIR_1_2_3_14,  -1,  2, -3, 14);
  Init_FIR_Table(Skl_FIR_0_1_3_6,    0, -1,  3, -6);
  Init_FIR_Table(Skl_FIR_0_0_1_3,    0,  0, -1,  3);
  Init_FIR_Table(Skl_FIR_0_0_0_1,    0,  0,  0, -1);
}


//////////////////////////////////////////////////////////
// SSE version

extern SKL_MB_FUNCS Skl_MB_Funcs_Add_SSE;
extern SKL_MB_FUNCS Skl_MB_Funcs_Copy_SSE;

extern MB_FUNC(Skl_Add_8x4_FF_SSE);
extern MB_FUNC(Skl_Add_8x4_FH_Rnd0_SSE);
extern MB_FUNC(Skl_Add_8x4_HF_Rnd0_SSE);
extern MB_FUNC(Skl_Add_8x4_HH_Rnd0_SSE);
extern MB_FUNC(Skl_Add_8x8_FF_SSE);
extern MB_FUNC(Skl_Add_8x8_FH_Rnd0_SSE);
extern MB_FUNC(Skl_Add_8x8_HF_Rnd0_SSE);
extern MB_FUNC(Skl_Add_8x8_HH_Rnd0_SSE);
extern MB_FUNC(Skl_Add_16x8_FF_SSE);
extern MB_FUNC(Skl_Add_16x8_FH_Rnd0_SSE);
extern MB_FUNC(Skl_Add_16x8_HF_Rnd0_SSE);
extern MB_FUNC(Skl_Add_16x8_HH_Rnd0_SSE);
extern MB_FUNC(Skl_Copy_8x4_FH_Rnd1_SSE);
extern MB_FUNC(Skl_Copy_8x4_HF_Rnd1_SSE);
extern MB_FUNC(Skl_Copy_8x4_HH_Rnd1_SSE);
extern MB_FUNC(Skl_Copy_8x4_FH_Rnd0_SSE);
extern MB_FUNC(Skl_Copy_8x4_HF_Rnd0_SSE);
extern MB_FUNC(Skl_Copy_8x4_HH_Rnd0_SSE);
extern MB_FUNC(Skl_Copy_8x8_FH_Rnd1_SSE);
extern MB_FUNC(Skl_Copy_8x8_HF_Rnd1_SSE);
extern MB_FUNC(Skl_Copy_8x8_HH_Rnd1_SSE);
extern MB_FUNC(Skl_Copy_8x8_FH_Rnd0_SSE);
extern MB_FUNC(Skl_Copy_8x8_HF_Rnd0_SSE);
extern MB_FUNC(Skl_Copy_8x8_HH_Rnd0_SSE);
extern MB_FUNC(Skl_Copy_16x8_FH_Rnd1_SSE);
extern MB_FUNC(Skl_Copy_16x8_HF_Rnd1_SSE);
extern MB_FUNC(Skl_Copy_16x8_HH_Rnd1_SSE);
extern MB_FUNC(Skl_Copy_16x8_FH_Rnd0_SSE);
extern MB_FUNC(Skl_Copy_16x8_HF_Rnd0_SSE);
extern MB_FUNC(Skl_Copy_16x8_HH_Rnd0_SSE);


SKL_MB_FUNCS Skl_MB_Funcs_Copy_Rnd1_SSE = {
  { Skl_Copy_16x8_FF_MMX, Skl_Copy_16x8_FH_Rnd1_SSE, Skl_Copy_16x8_HF_Rnd1_SSE, Skl_Copy_16x8_HH_Rnd1_SSE }
, { Skl_Copy_8x8_FF_MMX,  Skl_Copy_8x8_FH_Rnd1_SSE,  Skl_Copy_8x8_HF_Rnd1_SSE,  Skl_Copy_8x8_HH_Rnd1_SSE }
, { Skl_Copy_8x4_FF_MMX,  Skl_Copy_8x4_FH_Rnd1_SSE,  Skl_Copy_8x4_HF_Rnd1_SSE,  Skl_Copy_8x4_HH_Rnd1_SSE  }
,   Skl_H_Pass_16_Copy_Rnd1_MMX, Skl_H_Pass_Avrg_16_Copy_Rnd1_MMX, Skl_H_Pass_Avrg_Up_16_Copy_Rnd1_MMX
,   Skl_H_Pass_16_Copy_Rnd1_MMX, Skl_H_Pass_Avrg_16_Copy_Rnd1_MMX, Skl_H_Pass_Avrg_Up_16_Copy_Rnd1_MMX
,   Skl_V_Pass_16_Copy_Rnd1_MMX, Skl_V_Pass_Avrg_16_Copy_Rnd1_MMX, Skl_V_Pass_Avrg_Up_16_Copy_Rnd1_MMX
,   Skl_H_Pass_8_Copy_Rnd1_MMX,  Skl_H_Pass_Avrg_8_Copy_Rnd1_MMX,  Skl_H_Pass_Avrg_Up_8_Copy_Rnd1_MMX
,   Skl_H_Pass_8_Copy_Rnd1_MMX,  Skl_H_Pass_Avrg_8_Copy_Rnd1_MMX,  Skl_H_Pass_Avrg_Up_8_Copy_Rnd1_MMX 
,   Skl_V_Pass_8_Copy_Rnd1_MMX,  Skl_V_Pass_Avrg_8_Copy_Rnd1_MMX,  Skl_V_Pass_Avrg_Up_8_Copy_Rnd1_MMX
,   Skl_SAD_HP_16x16_Rnd1_C, Skl_SAD_HP_8x8_Rnd1_C
};
SKL_MB_FUNCS Skl_MB_Funcs_Copy_Rnd0_SSE = {
  { Skl_Copy_16x8_FF_MMX, Skl_Copy_16x8_FH_Rnd0_SSE, Skl_Copy_16x8_HF_Rnd0_SSE, Skl_Copy_16x8_HH_Rnd0_SSE }
, { Skl_Copy_8x8_FF_MMX,  Skl_Copy_8x8_FH_Rnd0_SSE,  Skl_Copy_8x8_HF_Rnd0_SSE,  Skl_Copy_8x8_HH_Rnd0_SSE }
, { Skl_Copy_8x4_FF_MMX,  Skl_Copy_8x4_FH_Rnd0_SSE,  Skl_Copy_8x4_HF_Rnd0_SSE,  Skl_Copy_8x4_HH_Rnd0_SSE  }
,   Skl_H_Pass_16_Copy_Rnd0_MMX, Skl_H_Pass_Avrg_16_Copy_Rnd0_MMX, Skl_H_Pass_Avrg_Up_16_Copy_Rnd0_MMX
,   Skl_H_Pass_16_Copy_Rnd0_MMX, Skl_H_Pass_Avrg_16_Copy_Rnd0_MMX, Skl_H_Pass_Avrg_Up_16_Copy_Rnd0_MMX
,   Skl_V_Pass_16_Copy_Rnd0_MMX, Skl_V_Pass_Avrg_16_Copy_Rnd0_MMX, Skl_V_Pass_Avrg_Up_16_Copy_Rnd0_MMX
,   Skl_H_Pass_8_Copy_Rnd0_MMX,  Skl_H_Pass_Avrg_8_Copy_Rnd0_MMX,  Skl_H_Pass_Avrg_Up_8_Copy_Rnd0_MMX
,   Skl_H_Pass_8_Copy_Rnd0_MMX,  Skl_H_Pass_Avrg_8_Copy_Rnd0_MMX,  Skl_H_Pass_Avrg_Up_8_Copy_Rnd0_MMX 
,   Skl_V_Pass_8_Copy_Rnd0_MMX,  Skl_V_Pass_Avrg_8_Copy_Rnd0_MMX,  Skl_V_Pass_Avrg_Up_8_Copy_Rnd0_MMX
,   Skl_SAD_HP_16x16_Rnd0_C, Skl_SAD_HP_8x8_Rnd0_C
};
SKL_MB_FUNCS Skl_MB_Funcs_Add_Rnd0_SSE = {
  { Skl_Add_16x8_FF_SSE, Skl_Add_16x8_FH_Rnd0_SSE, Skl_Add_16x8_HF_Rnd0_SSE, Skl_Add_16x8_HH_Rnd0_SSE }
, { Skl_Add_8x8_FF_SSE,  Skl_Add_8x8_FH_Rnd0_SSE,  Skl_Add_8x8_HF_Rnd0_SSE,  Skl_Add_8x8_HH_Rnd0_SSE }
, { Skl_Add_8x4_FF_SSE,  Skl_Add_8x4_FH_Rnd0_SSE,  Skl_Add_8x4_HF_Rnd0_SSE,  Skl_Add_8x4_HH_Rnd0_SSE  }
,   Skl_H_Pass_16_Add_Rnd0_MMX,  Skl_H_Pass_Avrg_16_Add_Rnd0_MMX,  Skl_H_Pass_Avrg_Up_16_Add_Rnd0_MMX
,   Skl_H_Pass_16_Copy_Rnd0_MMX, Skl_H_Pass_Avrg_16_Copy_Rnd0_MMX, Skl_H_Pass_Avrg_Up_16_Copy_Rnd0_MMX
,   Skl_V_Pass_16_Add_Rnd0_MMX,  Skl_V_Pass_Avrg_16_Add_Rnd0_MMX,  Skl_V_Pass_Avrg_Up_16_Add_Rnd0_MMX
,   Skl_H_Pass_8_Add_Rnd0_MMX,   Skl_H_Pass_Avrg_8_Add_Rnd0_MMX,   Skl_H_Pass_Avrg_Up_8_Add_Rnd0_MMX
,   Skl_H_Pass_8_Copy_Rnd0_MMX,  Skl_H_Pass_Avrg_8_Copy_Rnd0_MMX,  Skl_H_Pass_Avrg_Up_8_Copy_Rnd0_MMX 
,   Skl_V_Pass_8_Add_Rnd0_MMX,   Skl_V_Pass_Avrg_8_Add_Rnd0_MMX,   Skl_V_Pass_Avrg_Up_8_Add_Rnd0_MMX
,   Skl_SAD_HP_16x16_Rnd0_C, Skl_SAD_HP_8x8_Rnd0_C
};

//////////////////////////////////////////////////////////

extern MB_FUNC(Skl_H_Pass_2Taps_SSE);
extern MB_FUNC(Skl_V_Pass_2Taps_SSE);
extern MB_FUNC(Skl_HV_Pass_2Taps_SSE);

const SKL_HV_FILTER Skl_Filter_2_SSE = {
  Skl_H_Pass_2Taps_SSE, Skl_V_Pass_2Taps_SSE, Skl_HV_Pass_2Taps_SSE
};

//////////////////////////////////////////////////////////
// SSE2 version

extern SKL_MB_FUNCS Skl_MB_Funcs_Add_SSE2;
extern SKL_MB_FUNCS Skl_MB_Funcs_Copy_SSE2;

extern MB_FUNC(Skl_Add_16x8_FF_SSE2);
extern MB_FUNC(Skl_Add_16x8_FH_Rnd0_SSE2);
extern MB_FUNC(Skl_Add_16x8_HF_Rnd0_SSE2);
extern MB_FUNC(Skl_Add_16x8_HH_Rnd0_SSE2);
extern MB_FUNC(Skl_Copy_16x8_FF_SSE2);
extern MB_FUNC(Skl_Copy_16x8_FH_Rnd0_SSE2);
extern MB_FUNC(Skl_Copy_16x8_HF_Rnd0_SSE2);
extern MB_FUNC(Skl_Copy_16x8_HH_Rnd0_SSE2);
extern MB_FUNC(Skl_Copy_16x8_FH_Rnd1_SSE2);
extern MB_FUNC(Skl_Copy_16x8_HF_Rnd1_SSE2);
extern MB_FUNC(Skl_Copy_16x8_HH_Rnd1_SSE2);

SKL_MB_FUNCS Skl_MB_Funcs_Copy_Rnd1_SSE2 = {
  { Skl_Copy_16x8_FF_SSE2, Skl_Copy_16x8_FH_Rnd1_SSE2, Skl_Copy_16x8_HF_Rnd1_SSE2, Skl_Copy_16x8_HH_Rnd1_SSE2 }
, { Skl_Copy_8x8_FF_MMX,   Skl_Copy_8x8_FH_Rnd1_SSE,   Skl_Copy_8x8_HF_Rnd1_SSE,   Skl_Copy_8x8_HH_Rnd1_SSE }
, { Skl_Copy_8x4_FF_MMX,   Skl_Copy_8x4_FH_Rnd1_SSE,   Skl_Copy_8x4_HF_Rnd1_SSE,   Skl_Copy_8x4_HH_Rnd1_SSE }
,   Skl_H_Pass_16_Copy_Rnd1_MMX, Skl_H_Pass_Avrg_16_Copy_Rnd1_MMX, Skl_H_Pass_Avrg_Up_16_Copy_Rnd1_MMX
,   Skl_H_Pass_16_Copy_Rnd1_MMX, Skl_H_Pass_Avrg_16_Copy_Rnd1_MMX, Skl_H_Pass_Avrg_Up_16_Copy_Rnd1_MMX
,   Skl_V_Pass_16_Copy_Rnd1_MMX, Skl_V_Pass_Avrg_16_Copy_Rnd1_MMX, Skl_V_Pass_Avrg_Up_16_Copy_Rnd1_MMX
,   Skl_H_Pass_8_Copy_Rnd1_MMX,  Skl_H_Pass_Avrg_8_Copy_Rnd1_MMX,  Skl_H_Pass_Avrg_Up_8_Copy_Rnd1_MMX
,   Skl_H_Pass_8_Copy_Rnd1_MMX,  Skl_H_Pass_Avrg_8_Copy_Rnd1_MMX,  Skl_H_Pass_Avrg_Up_8_Copy_Rnd1_MMX 
,   Skl_V_Pass_8_Copy_Rnd1_MMX,  Skl_V_Pass_Avrg_8_Copy_Rnd1_MMX,  Skl_V_Pass_Avrg_Up_8_Copy_Rnd1_MMX
,   Skl_SAD_HP_16x16_Rnd1_C, Skl_SAD_HP_8x8_Rnd1_C
};
SKL_MB_FUNCS Skl_MB_Funcs_Copy_Rnd0_SSE2 = {
  { Skl_Copy_16x8_FF_SSE2, Skl_Copy_16x8_FH_Rnd0_SSE2, Skl_Copy_16x8_HF_Rnd0_SSE2, Skl_Copy_16x8_HH_Rnd0_SSE2 }
, { Skl_Copy_8x8_FF_MMX,   Skl_Copy_8x8_FH_Rnd0_SSE,   Skl_Copy_8x8_HF_Rnd0_SSE,   Skl_Copy_8x8_HH_Rnd0_SSE }
, { Skl_Copy_8x4_FF_MMX,   Skl_Copy_8x4_FH_Rnd0_SSE,   Skl_Copy_8x4_HF_Rnd0_SSE,   Skl_Copy_8x4_HH_Rnd0_SSE }
,   Skl_H_Pass_16_Copy_Rnd0_MMX, Skl_H_Pass_Avrg_16_Copy_Rnd0_MMX, Skl_H_Pass_Avrg_Up_16_Copy_Rnd0_MMX
,   Skl_H_Pass_16_Copy_Rnd0_MMX, Skl_H_Pass_Avrg_16_Copy_Rnd0_MMX, Skl_H_Pass_Avrg_Up_16_Copy_Rnd0_MMX
,   Skl_V_Pass_16_Copy_Rnd0_MMX, Skl_V_Pass_Avrg_16_Copy_Rnd0_MMX, Skl_V_Pass_Avrg_Up_16_Copy_Rnd0_MMX
,   Skl_H_Pass_8_Copy_Rnd0_MMX,  Skl_H_Pass_Avrg_8_Copy_Rnd0_MMX,  Skl_H_Pass_Avrg_Up_8_Copy_Rnd0_MMX
,   Skl_H_Pass_8_Copy_Rnd0_MMX,  Skl_H_Pass_Avrg_8_Copy_Rnd0_MMX,  Skl_H_Pass_Avrg_Up_8_Copy_Rnd0_MMX
,   Skl_V_Pass_8_Copy_Rnd0_MMX,  Skl_V_Pass_Avrg_8_Copy_Rnd0_MMX,  Skl_V_Pass_Avrg_Up_8_Copy_Rnd0_MMX
,   Skl_SAD_HP_16x16_Rnd0_C, Skl_SAD_HP_8x8_Rnd0_C
};
SKL_MB_FUNCS Skl_MB_Funcs_Add_Rnd0_SSE2 = {
  { Skl_Add_16x8_FF_SSE2, Skl_Add_16x8_FH_Rnd0_SSE2, Skl_Add_16x8_HF_Rnd0_SSE2, Skl_Add_16x8_HH_Rnd0_SSE2 }
, { Skl_Add_8x8_FF_MMX,   Skl_Add_8x8_FH_Rnd0_SSE,   Skl_Add_8x8_HF_Rnd0_SSE,   Skl_Add_8x8_HH_Rnd0_SSE }
, { Skl_Add_8x4_FF_MMX,   Skl_Add_8x4_FH_Rnd0_SSE,   Skl_Add_8x4_HF_Rnd0_SSE,   Skl_Add_8x4_HH_Rnd0_SSE }
,   Skl_H_Pass_16_Add_Rnd0_MMX,  Skl_H_Pass_Avrg_16_Add_Rnd0_MMX,  Skl_H_Pass_Avrg_Up_16_Add_Rnd0_MMX
,   Skl_H_Pass_16_Copy_Rnd0_MMX, Skl_H_Pass_Avrg_16_Copy_Rnd0_MMX, Skl_H_Pass_Avrg_Up_16_Copy_Rnd0_MMX
,   Skl_V_Pass_16_Add_Rnd0_MMX,  Skl_V_Pass_Avrg_16_Add_Rnd0_MMX,  Skl_V_Pass_Avrg_Up_16_Add_Rnd0_MMX
,   Skl_H_Pass_8_Add_Rnd0_MMX,   Skl_H_Pass_Avrg_8_Add_Rnd0_MMX,   Skl_H_Pass_Avrg_Up_8_Add_Rnd0_MMX
,   Skl_H_Pass_8_Copy_Rnd0_MMX,  Skl_H_Pass_Avrg_8_Copy_Rnd0_MMX,  Skl_H_Pass_Avrg_Up_8_Copy_Rnd0_MMX
,   Skl_V_Pass_8_Add_Rnd0_MMX,   Skl_V_Pass_Avrg_8_Add_Rnd0_MMX,   Skl_V_Pass_Avrg_Up_8_Add_Rnd0_MMX
,   Skl_SAD_HP_16x16_Rnd0_C, Skl_SAD_HP_8x8_Rnd0_C
};

#endif    /* SKL_USE_ASM */

//////////////////////////////////////////////////////////
// Upsampling (3,1) filter

static inline void Filter_31(SKL_BYTE *Dst1, SKL_BYTE *Dst2,
                             const SKL_INT16 *Src1, const SKL_INT16 *Src2)
{
    /* Src[] is assumed to be >=0. So we can use ">>2" instead of "/2" */
  SKL_INT16 a = (3*Src1[0]+  Src2[0]+2) >> 2;
  SKL_INT16 b = (  Src1[0]+3*Src2[0]+2) >> 2;
  COPY( Dst1[0], a );
  COPY( Dst2[0], b );

}

static inline void Filter_9331(SKL_BYTE *Dst1, SKL_BYTE *Dst2,
                               const SKL_INT16 *Src1, const SKL_INT16 *Src2)
{
    /* Src[] is assumed to be >=0. So we can use ">>4" instead of "/16" */
  SKL_INT16 a = (9*Src1[0]+  3*Src1[1]+ 3*Src2[0] + 1*Src2[1] + 8) >> 4;
  SKL_INT16 b = (3*Src1[0]+  9*Src1[1]+ 1*Src2[0] + 3*Src2[1] + 8) >> 4;
  SKL_INT16 c = (3*Src1[0]+  1*Src1[1]+ 9*Src2[0] + 3*Src2[1] + 8) >> 4;
  SKL_INT16 d = (1*Src1[0]+  3*Src1[1]+ 3*Src2[0] + 9*Src2[1] + 8) >> 4;
  COPY( Dst1[0], a );
  COPY( Dst1[1], b );
  COPY( Dst2[0], c );
  COPY( Dst2[1], d );
}

void Skl_Copy_Upsampled_8x8_16To8_C(SKL_BYTE *Dst, const SKL_INT16 *Src, const int BpS)
{
  int x, y;

  COPY( Dst[0], Src[0] );
  for(x=0; x<7; ++x) Filter_31(Dst+2*x+1, Dst+2*x+2, Src+x, Src+x+1);
  COPY( Dst[15], Src[7] );
  Dst += BpS;
  for(y=0; y<7; ++y) {
    SKL_BYTE *const Dst2 = Dst + BpS;
    Filter_31(Dst, Dst2, Src, Src+8);
    for(x=0; x<7; ++x)
      Filter_9331(Dst+2*x+1, Dst2+2*x+1, Src+x, Src+x+8);
    Filter_31(Dst+15, Dst2+15, Src+7, Src+7+8);
    Src += 8; 
    Dst += 2*BpS;
  }
  COPY( Dst[0], Src[0] );
  for(x=0; x<7; ++x) Filter_31(Dst+2*x+1, Dst+2*x+2, Src+x, Src+x+1);
  COPY( Dst[15], Src[7] );
}

static inline void Filter_Add_31(SKL_BYTE *Dst1, SKL_BYTE *Dst2,
                             const SKL_INT16 *Src1, const SKL_INT16 *Src2)
{
    /* Here, we must use "/4", since Src[] is in [-256, 255] */
  SKL_INT16 a = (3*Src1[0]+  Src2[0] + 2) / 4;
  SKL_INT16 b = (  Src1[0]+3*Src2[0] + 2) / 4;
  ADD(Dst1[0], a);
  ADD(Dst2[0], b);
}

static inline void Filter_Add_9331(SKL_BYTE *Dst1, SKL_BYTE *Dst2,
                                   const SKL_INT16 *Src1, const SKL_INT16 *Src2)
{
  SKL_INT16 a = (9*Src1[0]+  3*Src1[1]+ 3*Src2[0] + 1*Src2[1] + 8) / 16;
  SKL_INT16 b = (3*Src1[0]+  9*Src1[1]+ 1*Src2[0] + 3*Src2[1] + 8) / 16;
  SKL_INT16 c = (3*Src1[0]+  1*Src1[1]+ 9*Src2[0] + 3*Src2[1] + 8) / 16;
  SKL_INT16 d = (1*Src1[0]+  3*Src1[1]+ 3*Src2[0] + 9*Src2[1] + 8) / 16;
  ADD(Dst1[0], a);
  ADD(Dst1[1], b);
  ADD(Dst2[0], c);
  ADD(Dst2[1], d);
}

void Skl_Add_Upsampled_8x8_16To8_C(SKL_BYTE *Dst, const SKL_INT16 *Src, const int BpS)
{
  int x, y;