contrib/patch-x264-aq.patch

   1 Index: common/common.c
   2 ===================================================================
   3 --- common/common.c     (revision 669)
   4 +++ common/common.c     (working copy)
   5 @@ -123,6 +123,9 @@
   6      param->analyse.i_chroma_qp_offset = 0;
   7      param->analyse.b_fast_pskip = 1;
   8      param->analyse.b_dct_decimate = 1;
   9 +    param->analyse.b_aq = 0;
  10 +    param->analyse.f_aq_strength = 0.0;
  11 +    param->analyse.f_aq_sensitivity = 15;
  12      param->analyse.i_luma_deadzone[0] = 21;
  13      param->analyse.i_luma_deadzone[1] = 11;
  14      param->analyse.b_psnr = 1;
  15 @@ -455,6 +458,13 @@
  16          p->analyse.b_fast_pskip = atobool(value);
  17      OPT("dct-decimate")
  18          p->analyse.b_dct_decimate = atobool(value);
  19 +    OPT("aq-strength")
  20 +    {
  21 +        p->analyse.f_aq_strength = atof(value);
  22 +        p->analyse.b_aq = (p->analyse.f_aq_strength > 0.0);
  23 +    }
  24 +    OPT("aq-sensitivity")
  25 +        p->analyse.f_aq_sensitivity = atof(value);
  26      OPT("deadzone-inter")
  27          p->analyse.i_luma_deadzone[0] = atoi(value);
  28      OPT("deadzone-intra")
  29 @@ -939,6 +949,9 @@
  30              s += sprintf( s, " zones" );
  31      }
  32
  33 +    if( p->analyse.b_aq )
  34 +        s += sprintf( s, " aq=1:%.1f:%.1f", p->analyse.f_aq_strength, p->analyse.f_aq_sensitivity );
  35 +
  36      return buf;
  37  }
  38
  39 Index: common/pixel.c
  40 ===================================================================
  41 --- common/pixel.c      (revision 669)
  42 +++ common/pixel.c      (working copy)
  43 @@ -213,6 +213,14 @@
  44  PIXEL_SATD_C( x264_pixel_satd_4x8,   4, 8 )
  45  PIXEL_SATD_C( x264_pixel_satd_4x4,   4, 4 )
  46
  47 +static int x264_pixel_count_8x8( uint8_t *pix, int i_pix, uint32_t threshold )
  48 +{
  49 +    int x, y, sum = 0;
  50 +    for( y=0; y<8; y++, pix += i_pix )
  51 +        for( x=0; x<8; x++ )
  52 +            sum += pix[x] > (uint8_t)threshold;
  53 +    return sum;
  54 +}
  55
  56  /****************************************************************************
  57   * pixel_sa8d_WxH: sum of 8x8 Hadamard transformed differences
  58 @@ -470,6 +478,8 @@
  59      pixf->ads[PIXEL_16x8] = pixel_ads2;
  60      pixf->ads[PIXEL_8x8] = pixel_ads1;
  61
  62 +    pixf->count_8x8 = x264_pixel_count_8x8;
  63 +
  64  #ifdef HAVE_MMX
  65      if( cpu&X264_CPU_MMX )
  66      {
  67 Index: common/pixel.h
  68 ===================================================================
  69 --- common/pixel.h      (revision 669)
  70 +++ common/pixel.h      (working copy)
  71 @@ -90,6 +90,8 @@
  72      void (*ads[7])( int enc_dc[4], uint16_t *sums, int delta,
  73                      uint16_t *res, int width );
  74
  75 +    int (*count_8x8)( uint8_t *pix, int i_pix, uint32_t threshold );
  76 +
  77      /* calculate satd of V, H, and DC modes.
  78       * may be NULL, in which case just use pred+satd instead. */
  79      void (*intra_satd_x3_16x16)( uint8_t *fenc, uint8_t *fdec, int res[3] );
  80 Index: encoder/analyse.c
  81 ===================================================================
  82 --- encoder/analyse.c   (revision 669)
  83 +++ encoder/analyse.c   (working copy)
  84 @@ -29,6 +29,7 @@
  85  #endif
  86
  87  #include "common/common.h"
  88 +#include "common/cpu.h"
  89  #include "macroblock.h"
  90  #include "me.h"
  91  #include "ratecontrol.h"
  92 @@ -2029,8 +2030,68 @@
  93      }
  94  }
  95
  96 +static int x264_sum_dctq( int16_t dct[8][8] )
  97 +{
  98 +    int i, t = 0;
  99 +    int16_t *p = &dct[0][0];
 100 +    for( i=1; i<64; i++ )
 101 +        t += abs(p[i]) * x264_dct8_weight_tab[i];
 102 +    return t;
 103 +}
 104
 105  /*****************************************************************************
 106 + * x264_adaptive_quant:
 107 + * check if mb is "flat", i.e. has most energy in low frequency components, and
 108 + * adjust qp down if it is
 109 + *****************************************************************************/
 110 +void x264_adaptive_quant( x264_t *h, x264_mb_analysis_t *a )
 111 +{
 112 +    DECLARE_ALIGNED( static uint8_t, zero[FDEC_STRIDE*8], 16 );
 113 +    DECLARE_ALIGNED( int16_t, dct[8][8], 16 );
 114 +    float fc;
 115 +    int total = 0;
 116 +    int qp = h->mb.i_qp, qp_adj;
 117 +    int i;
 118 +
 119 +    if( qp <= 10 ) /* AQ is probably not needed at such low QP */
 120 +        return;
 121 +
 122 +    if( h->pixf.sad[PIXEL_16x16](h->mb.pic.p_fenc[0], FENC_STRIDE, zero, 16) > 64*16*16 )
 123 +    {   /* light places */
 124 +        if( h->pixf.count_8x8(h->mb.pic.p_fenc[1], FENC_STRIDE, 0x81818181) < 40 )
 125 +            /* not enough "blue" pixels */
 126 +            return;
 127 +
 128 +        if( h->pixf.count_8x8(h->mb.pic.p_fenc[2], FENC_STRIDE, 0x87878787) > 24 )
 129 +            /* too many "red" pixels */
 130 +            return;
 131 +    }
 132 +
 133 +    for( i=0; i<4; i++ )
 134 +    {
 135 +        h->dctf.sub8x8_dct8( dct, h->mb.pic.p_fenc[0] + (i&1)*8 + (i>>1)*FENC_STRIDE, zero );
 136 +        total += x264_sum_dctq( dct );
 137 +    }
 138 +
 139 +    if( total == 0 ) /* no AC coefficients, nothing to do */
 140 +        return;
 141 +
 142 +    x264_cpu_restore( h->param.cpu );
 143 +
 144 +    fc = expf(-5e-13 * total * total);
 145 +
 146 +    /* the function is chosen such that it stays close to 0 in almost all
 147 +      * range of 0..1, and rapidly goes up to 1 near 1.0 */
 148 +    qp_adj = (int)(qp * h->param.analyse.f_aq_strength / pow(2 - fc, h->param.analyse.f_aq_sensitivity));
 149 +
 150 +    /* don't adjust by more than this amount */
 151 +    qp_adj = X264_MIN(qp_adj, qp/2);
 152 +
 153 +    h->mb.i_qp = a->i_qp = qp - qp_adj;
 154 +    h->mb.i_chroma_qp = i_chroma_qp_table[x264_clip3( h->mb.i_qp + h->pps->i_chroma_qp_index_offset, 0, 51 )];
 155 +}
 156 +
 157 +/*****************************************************************************
 158   * x264_macroblock_analyse:
 159   *****************************************************************************/
 160  void x264_macroblock_analyse( x264_t *h )
 161 @@ -2038,9 +2099,14 @@
 162      x264_mb_analysis_t analysis;
 163      int i_cost = COST_MAX;
 164      int i;
 165 +
 166 +    h->mb.i_qp = x264_ratecontrol_qp( h );
 167
 168 +    if( h->param.analyse.b_aq )
 169 +        x264_adaptive_quant( h, &analysis );
 170 +
 171      /* init analysis */
 172 -    x264_mb_analyse_init( h, &analysis, x264_ratecontrol_qp( h ) );
 173 +    x264_mb_analyse_init( h, &analysis, h->mb.i_qp );
 174
 175      /*--------------------------- Do the analysis ---------------------------*/
 176      if( h->sh.i_type == SLICE_TYPE_I )
 177 Index: encoder/encoder.c
 178 ===================================================================
 179 --- encoder/encoder.c   (revision 669)
 180 +++ encoder/encoder.c   (working copy)
 181 @@ -477,6 +477,8 @@
 182      if( !h->param.b_cabac )
 183          h->param.analyse.i_trellis = 0;
 184      h->param.analyse.i_trellis = x264_clip3( h->param.analyse.i_trellis, 0, 2 );
 185 +    if( h->param.analyse.b_aq && h->param.analyse.f_aq_strength <= 0 )
 186 +        h->param.analyse.b_aq = 0;
 187      h->param.analyse.i_noise_reduction = x264_clip3( h->param.analyse.i_noise_reduction, 0, 1<<16 );
 188
 189      {
 190 Index: x264.c
 191 ===================================================================
 192 --- x264.c      (revision 669)
 193 +++ x264.c      (working copy)
 194 @@ -243,6 +243,12 @@
 195          "                                  - 2: enabled on all mode decisions\n", defaults->analyse.i_trellis );
 196      H0( "      --no-fast-pskip         Disables early SKIP detection on P-frames\n" );
 197      H0( "      --no-dct-decimate       Disables coefficient thresholding on P-frames\n" );
 198 +    H0( "      --aq-strength <float>   Amount to adjust QP per MB [%.1f]\n"
 199 +        "                                  0.0: no AQ\n"
 200 +        "                                  1.1: strong AQ\n", defaults->analyse.f_aq_strength );
 201 +    H0( "      --aq-sensitivity <float> \"Flatness\" threshold to trigger AQ [%.1f]\n"
 202 +        "                                    5: applies to almost all blocks\n"
 203 +        "                                   22: only flat blocks\n", defaults->analyse.f_aq_sensitivity );
 204      H0( "      --nr <integer>          Noise reduction [%d]\n", defaults->analyse.i_noise_reduction );
 205      H1( "\n" );
 206      H1( "      --deadzone-inter <int>  Set the size of the inter luma quantization deadzone [%d]\n", defaults->analyse.i_luma_deadzone[0] );
 207 @@ -406,6 +412,8 @@
 208              { "trellis", required_argument, NULL, 't' },
 209              { "no-fast-pskip", no_argument, NULL, 0 },
 210              { "no-dct-decimate", no_argument, NULL, 0 },
 211 +            { "aq-strength", required_argument, NULL, 0 },
 212 +            { "aq-sensitivity", required_argument, NULL, 0 },
 213              { "deadzone-inter", required_argument, NULL, '0' },
 214              { "deadzone-intra", required_argument, NULL, '0' },
 215              { "level",   required_argument, NULL, 0 },
 216 Index: x264.h
 217 ===================================================================
 218 --- x264.h      (revision 669)
 219 +++ x264.h      (working copy)
 220 @@ -227,6 +227,9 @@
 221          int          i_trellis;  /* trellis RD quantization */
 222          int          b_fast_pskip; /* early SKIP detection on P-frames */
 223          int          b_dct_decimate; /* transform coefficient thresholding on P-frames */
 224 +        int          b_aq; /* psy adaptive QP */
 225 +        float        f_aq_strength;
 226 +        float        f_aq_sensitivity;
 227          int          i_noise_reduction; /* adaptive pseudo-deadzone */
 228
 229          /* the deadzone size that will be used in luma quantization */