macroblock.c 32.1 KB
Newer Older
Laurent Aimar's avatar
Laurent Aimar committed
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26
/*****************************************************************************
 * macroblock.c: h264 encoder library
 *****************************************************************************
 * Copyright (C) 2003 Laurent Aimar
 * $Id: macroblock.c,v 1.1 2004/06/03 19:27:08 fenrir Exp $
 *
 * Authors: Laurent Aimar <fenrir@via.ecp.fr>
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation; either version 2 of the License, or
 * (at your option) any later version.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with this program; if not, write to the Free Software
 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA  02111, USA.
 *****************************************************************************/

#include <stdio.h>
#include <string.h>

27
#include "common/common.h"
Laurent Aimar's avatar
Laurent Aimar committed
28 29 30
#include "macroblock.h"


Loren Merritt's avatar
Loren Merritt committed
31 32 33 34
/* def_quant4_mf only for probe_skip; actual encoding uses matrices from set.c */
/* FIXME this seems to make better decisions with cqm=jvt, but could screw up
 * with general custom matrices. */
static const int def_quant4_mf[6][4][4] =
Laurent Aimar's avatar
Laurent Aimar committed
35
{
36 37 38 39 40 41 42 43 44 45 46 47 48 49
    { { 13107, 8066, 13107, 8066 }, { 8066, 5243, 8066, 5243 },
      { 13107, 8066, 13107, 8066 }, { 8066, 5243, 8066, 5243 } },
    { { 11916, 7490, 11916, 7490 }, { 7490, 4660, 7490, 4660 },
      { 11916, 7490, 11916, 7490 }, { 7490, 4660, 7490, 4660 } },
    { { 10082, 6554, 10082, 6554 }, { 6554, 4194, 6554, 4194 },
      { 10082, 6554, 10082, 6554 }, { 6554, 4194, 6554, 4194 } },
    { {  9362, 5825,  9362, 5825 }, { 5825, 3647, 5825, 3647 },
      {  9362, 5825,  9362, 5825 }, { 5825, 3647, 5825, 3647 } },
    { {  8192, 5243,  8192, 5243 }, { 5243, 3355, 5243, 3355 },
      {  8192, 5243,  8192, 5243 }, { 5243, 3355, 5243, 3355 } },
    { {  7282, 4559,  7282, 4559 }, { 4559, 2893, 4559, 2893 },
      {  7282, 4559,  7282, 4559 }, { 4559, 2893, 4559, 2893 } }
};

Laurent Aimar's avatar
Laurent Aimar committed
50 51 52 53
/****************************************************************************
 * Scan and Quant functions
 ****************************************************************************/

54
#define ZIG(i,y,x) level[i] = dct[x][y];
55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73
static inline void scan_zigzag_8x8full( int level[64], int16_t dct[8][8] )
{
    ZIG( 0,0,0) ZIG( 1,0,1) ZIG( 2,1,0) ZIG( 3,2,0)
    ZIG( 4,1,1) ZIG( 5,0,2) ZIG( 6,0,3) ZIG( 7,1,2)
    ZIG( 8,2,1) ZIG( 9,3,0) ZIG(10,4,0) ZIG(11,3,1)
    ZIG(12,2,2) ZIG(13,1,3) ZIG(14,0,4) ZIG(15,0,5)
    ZIG(16,1,4) ZIG(17,2,3) ZIG(18,3,2) ZIG(19,4,1)
    ZIG(20,5,0) ZIG(21,6,0) ZIG(22,5,1) ZIG(23,4,2)
    ZIG(24,3,3) ZIG(25,2,4) ZIG(26,1,5) ZIG(27,0,6)
    ZIG(28,0,7) ZIG(29,1,6) ZIG(30,2,5) ZIG(31,3,4)
    ZIG(32,4,3) ZIG(33,5,2) ZIG(34,6,1) ZIG(35,7,0)
    ZIG(36,7,1) ZIG(37,6,2) ZIG(38,5,3) ZIG(39,4,4)
    ZIG(40,3,5) ZIG(41,2,6) ZIG(42,1,7) ZIG(43,2,7)
    ZIG(44,3,6) ZIG(45,4,5) ZIG(46,5,4) ZIG(47,6,3)
    ZIG(48,7,2) ZIG(49,7,3) ZIG(50,6,4) ZIG(51,5,5)
    ZIG(52,4,6) ZIG(53,3,7) ZIG(54,4,7) ZIG(55,5,6)
    ZIG(56,6,5) ZIG(57,7,4) ZIG(58,7,5) ZIG(59,6,6)
    ZIG(60,5,7) ZIG(61,6,7) ZIG(62,7,6) ZIG(63,7,7)
}
Laurent Aimar's avatar
Laurent Aimar committed
74 75
static inline void scan_zigzag_4x4full( int level[16], int16_t dct[4][4] )
{
76 77 78 79
    ZIG( 0,0,0) ZIG( 1,0,1) ZIG( 2,1,0) ZIG( 3,2,0)
    ZIG( 4,1,1) ZIG( 5,0,2) ZIG( 6,0,3) ZIG( 7,1,2)
    ZIG( 8,2,1) ZIG( 9,3,0) ZIG(10,3,1) ZIG(11,2,2)
    ZIG(12,1,3) ZIG(13,2,3) ZIG(14,3,2) ZIG(15,3,3)
Laurent Aimar's avatar
Laurent Aimar committed
80 81 82
}
static inline void scan_zigzag_4x4( int level[15], int16_t dct[4][4] )
{
83 84 85 86
                ZIG( 0,0,1) ZIG( 1,1,0) ZIG( 2,2,0)
    ZIG( 3,1,1) ZIG( 4,0,2) ZIG( 5,0,3) ZIG( 6,1,2)
    ZIG( 7,2,1) ZIG( 8,3,0) ZIG( 9,3,1) ZIG(10,2,2)
    ZIG(11,1,3) ZIG(12,2,3) ZIG(13,3,2) ZIG(14,3,3)
Laurent Aimar's avatar
Laurent Aimar committed
87 88 89
}
static inline void scan_zigzag_2x2_dc( int level[4], int16_t dct[2][2] )
{
90 91 92 93
    ZIG(0,0,0)
    ZIG(1,0,1)
    ZIG(2,1,0)
    ZIG(3,1,1)
Laurent Aimar's avatar
Laurent Aimar committed
94
}
95
#undef ZIG
Laurent Aimar's avatar
Laurent Aimar committed
96

Loren Merritt's avatar
Loren Merritt committed
97
#define ZIG(i,y,x) {\
98 99 100 101
    int oe = x+y*FENC_STRIDE;\
    int od = x+y*FDEC_STRIDE;\
    level[i] = p_src[oe] - p_dst[od];\
    p_dst[od] = p_src[oe];\
Loren Merritt's avatar
Loren Merritt committed
102
}
103
static inline void sub_zigzag_4x4full( int level[16], const uint8_t *p_src, uint8_t *p_dst )
Loren Merritt's avatar
Loren Merritt committed
104 105 106 107 108 109
{
    ZIG( 0,0,0) ZIG( 1,0,1) ZIG( 2,1,0) ZIG( 3,2,0)
    ZIG( 4,1,1) ZIG( 5,0,2) ZIG( 6,0,3) ZIG( 7,1,2)
    ZIG( 8,2,1) ZIG( 9,3,0) ZIG(10,3,1) ZIG(11,2,2)
    ZIG(12,1,3) ZIG(13,2,3) ZIG(14,3,2) ZIG(15,3,3)
}
110
static inline void sub_zigzag_4x4( int level[15], const uint8_t *p_src, uint8_t *p_dst )
Loren Merritt's avatar
Loren Merritt committed
111 112 113 114 115 116 117 118
{
                ZIG( 0,0,1) ZIG( 1,1,0) ZIG( 2,2,0)
    ZIG( 3,1,1) ZIG( 4,0,2) ZIG( 5,0,3) ZIG( 6,1,2)
    ZIG( 7,2,1) ZIG( 8,3,0) ZIG( 9,3,1) ZIG(10,2,2)
    ZIG(11,1,3) ZIG(12,2,3) ZIG(13,3,2) ZIG(14,3,3)
}
#undef ZIG

119
static void quant_8x8( x264_t *h, int16_t dct[8][8], int quant_mf[6][8][8], int i_qscale, int b_intra )
120 121 122
{
    const int i_qbits = 16 + i_qscale / 6;
    const int i_mf = i_qscale % 6;
Loren Merritt's avatar
Loren Merritt committed
123
    const int f = ( 1 << (i_qbits + b_intra) ) / 6;
124
    h->quantf.quant_8x8_core( dct, quant_mf[i_mf], i_qbits, f );
125
}
126
static void quant_4x4( x264_t *h, int16_t dct[4][4], int quant_mf[6][4][4], int i_qscale, int b_intra )
Laurent Aimar's avatar
Laurent Aimar committed
127 128 129
{
    const int i_qbits = 15 + i_qscale / 6;
    const int i_mf = i_qscale % 6;
Loren Merritt's avatar
Loren Merritt committed
130
    const int f = ( 1 << (i_qbits + b_intra) ) / 6;
131
    h->quantf.quant_4x4_core( dct, quant_mf[i_mf], i_qbits, f );
Laurent Aimar's avatar
Laurent Aimar committed
132
}
133
static void quant_4x4_dc( x264_t *h, int16_t dct[4][4], int quant_mf[6][4][4], int i_qscale )
Laurent Aimar's avatar
Laurent Aimar committed
134
{
135 136 137 138
    const int i_qbits = 16 + i_qscale / 6;
    const int i_mf = i_qscale % 6;
    const int f = ( 1 << i_qbits ) / 3;
    h->quantf.quant_4x4_dc_core( dct, quant_mf[i_mf][0][0], i_qbits, f );
Laurent Aimar's avatar
Laurent Aimar committed
139
}
140
static void quant_2x2_dc( x264_t *h, int16_t dct[2][2], int quant_mf[6][4][4], int i_qscale, int b_intra )
Laurent Aimar's avatar
Laurent Aimar committed
141
{
142 143
    const int i_qbits = 16 + i_qscale / 6;
    const int i_mf = i_qscale % 6;
Loren Merritt's avatar
Loren Merritt committed
144
    const int f = ( 1 << (i_qbits + b_intra) ) / 6;
145
    h->quantf.quant_2x2_dc_core( dct, quant_mf[i_mf][0][0], i_qbits, f );
Laurent Aimar's avatar
Laurent Aimar committed
146
}
147

Laurent Aimar's avatar
Laurent Aimar committed
148 149 150 151 152 153 154 155 156 157
/* (ref: JVT-B118)
 * x264_mb_decimate_score: given dct coeffs it returns a score to see if we could empty this dct coeffs
 * to 0 (low score means set it to null)
 * Used in inter macroblock (luma and chroma)
 *  luma: for a 8x8 block: if score < 4 -> null
 *        for the complete mb: if score < 6 -> null
 *  chroma: for the complete mb: if score < 7 -> null
 */
static int x264_mb_decimate_score( int *dct, int i_max )
{
158 159 160 161 162 163 164 165 166
    static const int i_ds_table4[16] = {
        3,2,2,1,1,1,0,0,0,0,0,0,0,0,0,0 };
    static const int i_ds_table8[64] = {
        3,3,3,3,2,2,2,2,2,2,2,2,1,1,1,1,
        1,1,1,1,1,1,1,1,0,0,0,0,0,0,0,0,
        0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
        0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0 };

    const int *ds_table = (i_max == 64) ? i_ds_table8 : i_ds_table4;
Laurent Aimar's avatar
Laurent Aimar committed
167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185
    int i_score = 0;
    int idx = i_max - 1;

    while( idx >= 0 && dct[idx] == 0 )
        idx--;

    while( idx >= 0 )
    {
        int i_run;

        if( abs( dct[idx--] ) > 1 )
            return 9;

        i_run = 0;
        while( idx >= 0 && dct[idx] == 0 )
        {
            idx--;
            i_run++;
        }
186
        i_score += ds_table[i_run];
Laurent Aimar's avatar
Laurent Aimar committed
187 188 189 190 191 192 193
    }

    return i_score;
}

void x264_mb_encode_i4x4( x264_t *h, int idx, int i_qscale )
{
194 195 196 197
    int x = 4 * block_idx_x[idx];
    int y = 4 * block_idx_y[idx];
    uint8_t *p_src = &h->mb.pic.p_fenc[0][x+y*FENC_STRIDE];
    uint8_t *p_dst = &h->mb.pic.p_fdec[0][x+y*FDEC_STRIDE];
Laurent Aimar's avatar
Laurent Aimar committed
198 199
    int16_t dct4x4[4][4];

Loren Merritt's avatar
Loren Merritt committed
200 201
    if( h->mb.b_lossless )
    {
202
        sub_zigzag_4x4full( h->dct.block[idx].luma4x4, p_src, p_dst );
Loren Merritt's avatar
Loren Merritt committed
203 204 205
        return;
    }

206
    h->dctf.sub4x4_dct( dct4x4, p_src, p_dst );
Loren Merritt's avatar
Loren Merritt committed
207 208 209 210 211 212

    if( h->mb.b_trellis )
        x264_quant_4x4_trellis( h, dct4x4, CQM_4IY, i_qscale, DCT_LUMA_4x4, 1 );
    else
        quant_4x4( h, dct4x4, h->quant4_mf[CQM_4IY], i_qscale, 1 );

Laurent Aimar's avatar
Laurent Aimar committed
213
    scan_zigzag_4x4full( h->dct.block[idx].luma4x4, dct4x4 );
214
    h->quantf.dequant_4x4( dct4x4, h->dequant4_mf[CQM_4IY], i_qscale );
Laurent Aimar's avatar
Laurent Aimar committed
215 216

    /* output samples to fdec */
217
    h->dctf.add4x4_idct( p_dst, dct4x4 );
Laurent Aimar's avatar
Laurent Aimar committed
218 219
}

220 221
void x264_mb_encode_i8x8( x264_t *h, int idx, int i_qscale )
{
222 223 224 225
    int x = 8 * (idx&1);
    int y = 8 * (idx>>1);
    uint8_t *p_src = &h->mb.pic.p_fenc[0][x+y*FENC_STRIDE];
    uint8_t *p_dst = &h->mb.pic.p_fdec[0][x+y*FDEC_STRIDE];
226 227
    int16_t dct8x8[8][8];

228
    h->dctf.sub8x8_dct8( dct8x8, p_src, p_dst );
Loren Merritt's avatar
Loren Merritt committed
229 230 231 232 233 234

    if( h->mb.b_trellis )
        x264_quant_8x8_trellis( h, dct8x8, CQM_8IY, i_qscale, 1 );
    else 
        quant_8x8( h, dct8x8, h->quant8_mf[CQM_8IY], i_qscale, 1 );

235
    scan_zigzag_8x8full( h->dct.luma8x8[idx], dct8x8 );
236
    h->quantf.dequant_8x8( dct8x8, h->dequant8_mf[CQM_8IY], i_qscale );
237
    h->dctf.add8x8_idct8( p_dst, dct8x8 );
238 239
}

Laurent Aimar's avatar
Laurent Aimar committed
240 241 242 243 244 245 246 247 248
static void x264_mb_encode_i16x16( x264_t *h, int i_qscale )
{
    uint8_t  *p_src = h->mb.pic.p_fenc[0];
    uint8_t  *p_dst = h->mb.pic.p_fdec[0];

    int16_t dct4x4[16+1][4][4];

    int i;

Loren Merritt's avatar
Loren Merritt committed
249 250 251 252
    if( h->mb.b_lossless )
    {
        for( i = 0; i < 16; i++ )
        {
253 254 255
            int oe = block_idx_x[i]*4 + block_idx_y[i]*4*FENC_STRIDE;
            int od = block_idx_x[i]*4 + block_idx_y[i]*4*FDEC_STRIDE;
            sub_zigzag_4x4( h->dct.block[i].residual_ac, p_src+oe, p_dst+od );
256
            dct4x4[0][block_idx_x[i]][block_idx_y[i]] = p_src[oe] - p_dst[od];
257
            p_dst[od] = p_src[oe];
Loren Merritt's avatar
Loren Merritt committed
258 259 260 261 262
        }
        scan_zigzag_4x4full( h->dct.luma16x16_dc, dct4x4[0] );
        return;
    }

263
    h->dctf.sub16x16_dct( &dct4x4[1], p_src, p_dst );
Laurent Aimar's avatar
Laurent Aimar committed
264 265 266 267 268 269
    for( i = 0; i < 16; i++ )
    {
        /* copy dc coeff */
        dct4x4[0][block_idx_y[i]][block_idx_x[i]] = dct4x4[1+i][0][0];

        /* quant/scan/dequant */
Loren Merritt's avatar
Loren Merritt committed
270 271 272 273 274
        if( h->mb.b_trellis )
            x264_quant_4x4_trellis( h, dct4x4[1+i], CQM_4IY, i_qscale, DCT_LUMA_AC, 1 );
        else
            quant_4x4( h, dct4x4[1+i], h->quant4_mf[CQM_4IY], i_qscale, 1 );

Laurent Aimar's avatar
Laurent Aimar committed
275
        scan_zigzag_4x4( h->dct.block[i].residual_ac, dct4x4[1+i] );
276
        h->quantf.dequant_4x4( dct4x4[1+i], h->dequant4_mf[CQM_4IY], i_qscale );
Laurent Aimar's avatar
Laurent Aimar committed
277 278 279
    }

    h->dctf.dct4x4dc( dct4x4[0] );
280
    quant_4x4_dc( h, dct4x4[0], h->quant4_mf[CQM_4IY], i_qscale );
Laurent Aimar's avatar
Laurent Aimar committed
281 282 283 284
    scan_zigzag_4x4full( h->dct.luma16x16_dc, dct4x4[0] );

    /* output samples to fdec */
    h->dctf.idct4x4dc( dct4x4[0] );
Loren Merritt's avatar
Loren Merritt committed
285
    x264_mb_dequant_4x4_dc( dct4x4[0], h->dequant4_mf[CQM_4IY], i_qscale );  /* XXX not inversed */
Laurent Aimar's avatar
Laurent Aimar committed
286 287 288 289 290 291 292 293

    /* calculate dct coeffs */
    for( i = 0; i < 16; i++ )
    {
        /* copy dc coeff */
        dct4x4[1+i][0][0] = dct4x4[0][block_idx_y[i]][block_idx_x[i]];
    }
    /* put pixels to fdec */
294
    h->dctf.add16x16_idct( p_dst, &dct4x4[1] );
Laurent Aimar's avatar
Laurent Aimar committed
295 296
}

297
static void x264_mb_encode_8x8_chroma( x264_t *h, int b_inter, int i_qscale )
Laurent Aimar's avatar
Laurent Aimar committed
298 299
{
    int i, ch;
300
    int b_decimate = b_inter && (h->sh.i_type == SLICE_TYPE_B || h->param.analyse.b_dct_decimate);
Laurent Aimar's avatar
Laurent Aimar committed
301 302 303 304 305 306 307 308 309 310

    for( ch = 0; ch < 2; ch++ )
    {
        uint8_t  *p_src = h->mb.pic.p_fenc[1+ch];
        uint8_t  *p_dst = h->mb.pic.p_fdec[1+ch];
        int i_decimate_score = 0;

        int16_t dct2x2[2][2];
        int16_t dct4x4[4][4][4];

Loren Merritt's avatar
Loren Merritt committed
311 312 313 314
        if( h->mb.b_lossless )
        {
            for( i = 0; i < 4; i++ )
            {
315 316 317 318 319
                int oe = block_idx_x[i]*4 + block_idx_y[i]*4*FENC_STRIDE;
                int od = block_idx_x[i]*4 + block_idx_y[i]*4*FDEC_STRIDE;
                sub_zigzag_4x4( h->dct.block[16+i+ch*4].residual_ac, p_src+oe, p_dst+od );
                h->dct.chroma_dc[ch][i] = p_src[oe] - p_dst[od];
                p_dst[od] = p_src[oe];
Loren Merritt's avatar
Loren Merritt committed
320 321 322 323
            }
            continue;
        }
            
324
        h->dctf.sub8x8_dct( dct4x4, p_src, p_dst );
Laurent Aimar's avatar
Laurent Aimar committed
325 326 327 328 329 330
        /* calculate dct coeffs */
        for( i = 0; i < 4; i++ )
        {
            /* copy dc coeff */
            dct2x2[block_idx_y[i]][block_idx_x[i]] = dct4x4[i][0][0];

Loren Merritt's avatar
Loren Merritt committed
331
            /* no trellis; it doesn't seem to help chroma noticeably */
332
            quant_4x4( h, dct4x4[i], h->quant4_mf[CQM_4IC + b_inter], i_qscale, !b_inter );
Laurent Aimar's avatar
Laurent Aimar committed
333 334
            scan_zigzag_4x4( h->dct.block[16+i+ch*4].residual_ac, dct4x4[i] );

335
            if( b_decimate )
Laurent Aimar's avatar
Laurent Aimar committed
336 337 338 339 340 341
            {
                i_decimate_score += x264_mb_decimate_score( h->dct.block[16+i+ch*4].residual_ac, 15 );
            }
        }

        h->dctf.dct2x2dc( dct2x2 );
342
        quant_2x2_dc( h, dct2x2, h->quant4_mf[CQM_4IC + b_inter], i_qscale, !b_inter );
Laurent Aimar's avatar
Laurent Aimar committed
343 344 345 346
        scan_zigzag_2x2_dc( h->dct.chroma_dc[ch], dct2x2 );

        /* output samples to fdec */
        h->dctf.idct2x2dc( dct2x2 );
Loren Merritt's avatar
Loren Merritt committed
347
        x264_mb_dequant_2x2_dc( dct2x2, h->dequant4_mf[CQM_4IC + b_inter], i_qscale );  /* XXX not inversed */
Laurent Aimar's avatar
Laurent Aimar committed
348

349
        if( b_decimate && i_decimate_score < 7 )
Laurent Aimar's avatar
Laurent Aimar committed
350 351
        {
            /* Near null chroma 8x8 block so make it null (bits saving) */
Loren Merritt's avatar
Loren Merritt committed
352
            memset( &h->dct.block[16+ch*4], 0, 4 * sizeof( *h->dct.block ) );
353 354 355
            if( !array_non_zero( (int*)dct2x2, sizeof(dct2x2)/sizeof(int) ) )
                continue;
            memset( dct4x4, 0, sizeof( dct4x4 ) );
Laurent Aimar's avatar
Laurent Aimar committed
356
        }
357 358 359 360 361
        else
        {
            for( i = 0; i < 4; i++ )
                h->quantf.dequant_4x4( dct4x4[i], h->dequant4_mf[CQM_4IC + b_inter], i_qscale );
        }
Laurent Aimar's avatar
Laurent Aimar committed
362 363

        for( i = 0; i < 4; i++ )
364
            dct4x4[i][0][0] = dct2x2[0][i];
365
        h->dctf.add8x8_idct( p_dst, dct4x4 );
Laurent Aimar's avatar
Laurent Aimar committed
366 367 368
    }
}

369 370 371 372 373 374 375 376 377 378 379 380 381 382 383
static void x264_macroblock_encode_skip( x264_t *h )
{
    int i;
    h->mb.i_cbp_luma = 0x00;
    h->mb.i_cbp_chroma = 0x00;

    for( i = 0; i < 16+8; i++ )
    {
        h->mb.cache.non_zero_count[x264_scan8[i]] = 0;
    }

    /* store cbp */
    h->mb.cbp[h->mb.i_mb_xy] = 0;
}

Laurent Aimar's avatar
Laurent Aimar committed
384 385 386 387 388 389
/*****************************************************************************
 * x264_macroblock_encode_pskip:
 *  Encode an already marked skip block
 *****************************************************************************/
void x264_macroblock_encode_pskip( x264_t *h )
{
390 391 392 393
    const int mvx = x264_clip3( h->mb.cache.mv[0][x264_scan8[0]][0],
                                h->mb.mv_min[0], h->mb.mv_max[0] );
    const int mvy = x264_clip3( h->mb.cache.mv[0][x264_scan8[0]][1],
                                h->mb.mv_min[1], h->mb.mv_max[1] );
Laurent Aimar's avatar
Laurent Aimar committed
394 395

    /* Motion compensation XXX probably unneeded */
396
    h->mc.mc_luma( h->mb.pic.p_fref[0][0], h->mb.pic.i_stride[0],
397 398
                   h->mb.pic.p_fdec[0],    FDEC_STRIDE,
                   mvx, mvy, 16, 16 );
Laurent Aimar's avatar
Laurent Aimar committed
399 400

    /* Chroma MC */
401
    h->mc.mc_chroma( h->mb.pic.p_fref[0][0][4], h->mb.pic.i_stride[1],
402 403
                     h->mb.pic.p_fdec[1],       FDEC_STRIDE,
                     mvx, mvy, 8, 8 );
Laurent Aimar's avatar
Laurent Aimar committed
404

405
    h->mc.mc_chroma( h->mb.pic.p_fref[0][0][5], h->mb.pic.i_stride[2],
406 407
                     h->mb.pic.p_fdec[2],       FDEC_STRIDE,
                     mvx, mvy, 8, 8 );
Laurent Aimar's avatar
Laurent Aimar committed
408

409
    x264_macroblock_encode_skip( h );
Laurent Aimar's avatar
Laurent Aimar committed
410 411 412 413 414 415 416 417
}

/*****************************************************************************
 * x264_macroblock_encode:
 *****************************************************************************/
void x264_macroblock_encode( x264_t *h )
{
    int i_cbp_dc = 0;
418
    int i_qp = h->mb.i_qp;
419
    int b_decimate = h->sh.i_type == SLICE_TYPE_B || h->param.analyse.b_dct_decimate;
Laurent Aimar's avatar
Laurent Aimar committed
420 421 422 423 424 425 426 427
    int i;

    if( h->mb.i_type == P_SKIP )
    {
        /* A bit special */
        x264_macroblock_encode_pskip( h );
        return;
    }
428 429 430 431 432 433 434
    if( h->mb.i_type == B_SKIP )
    {
        /* XXX motion compensation is probably unneeded */
        x264_mb_mc( h );
        x264_macroblock_encode_skip( h );
        return;
    }
Laurent Aimar's avatar
Laurent Aimar committed
435 436 437 438

    if( h->mb.i_type == I_16x16 )
    {
        const int i_mode = h->mb.i_intra16x16_pred_mode;
439
        h->mb.b_transform_8x8 = 0;
Laurent Aimar's avatar
Laurent Aimar committed
440
        /* do the right prediction */
441
        h->predict_16x16[i_mode]( h->mb.pic.p_fdec[0] );
Laurent Aimar's avatar
Laurent Aimar committed
442 443

        /* encode the 16x16 macroblock */
444
        x264_mb_encode_i16x16( h, i_qp );
Laurent Aimar's avatar
Laurent Aimar committed
445
    }
446 447
    else if( h->mb.i_type == I_8x8 )
    {
448
        DECLARE_ALIGNED( uint8_t, edge[33], 8 );
449
        h->mb.b_transform_8x8 = 1;
450 451
        for( i = 0; i < 4; i++ )
        {
452
            uint8_t  *p_dst = &h->mb.pic.p_fdec[0][8 * (i&1) + 8 * (i>>1) * FDEC_STRIDE];
453 454
            int      i_mode = h->mb.cache.intra4x4_pred_mode[x264_scan8[4*i]];

455 456
            x264_predict_8x8_filter( p_dst, edge, h->mb.i_neighbour8[i], x264_pred_i4x4_neighbors[i_mode] );
            h->predict_8x8[i_mode]( p_dst, edge );
457
            x264_mb_encode_i8x8( h, i, i_qp );
458 459
        }
    }
Laurent Aimar's avatar
Laurent Aimar committed
460 461
    else if( h->mb.i_type == I_4x4 )
    {
462
        h->mb.b_transform_8x8 = 0;
Laurent Aimar's avatar
Laurent Aimar committed
463 464
        for( i = 0; i < 16; i++ )
        {
465
            uint8_t  *p_dst = &h->mb.pic.p_fdec[0][4 * block_idx_x[i] + 4 * block_idx_y[i] * FDEC_STRIDE];
Laurent Aimar's avatar
Laurent Aimar committed
466 467
            int      i_mode = h->mb.cache.intra4x4_pred_mode[x264_scan8[i]];

468 469
            if( (h->mb.i_neighbour4[i] & (MB_TOPRIGHT|MB_TOP)) == MB_TOP )
                /* emulate missing topright samples */
470
                *(uint32_t*) &p_dst[4-FDEC_STRIDE] = p_dst[3-FDEC_STRIDE] * 0x01010101U;
471

472
            h->predict_4x4[i_mode]( p_dst );
473
            x264_mb_encode_i4x4( h, i, i_qp );
Laurent Aimar's avatar
Laurent Aimar committed
474 475 476 477 478 479 480 481 482 483
        }
    }
    else    /* Inter MB */
    {
        int i8x8, i4x4, idx;
        int i_decimate_mb = 0;

        /* Motion compensation */
        x264_mb_mc( h );

Loren Merritt's avatar
Loren Merritt committed
484 485 486 487
        if( h->mb.b_lossless )
        {
            for( i4x4 = 0; i4x4 < 16; i4x4++ )
            {
488 489 490 491 492
                int x = 4*block_idx_x[i4x4];
                int y = 4*block_idx_y[i4x4];
                sub_zigzag_4x4full( h->dct.block[i4x4].luma4x4,
                                    h->mb.pic.p_fenc[0]+x+y*FENC_STRIDE,
                                    h->mb.pic.p_fdec[0]+x+y*FDEC_STRIDE );
Loren Merritt's avatar
Loren Merritt committed
493 494 495
            }
        }
        else if( h->mb.b_transform_8x8 )
Laurent Aimar's avatar
Laurent Aimar committed
496
        {
497
            int16_t dct8x8[4][8][8];
498
            int nnz8x8[4] = {1,1,1,1};
499
            b_decimate &= !h->mb.b_trellis; // 8x8 trellis is inherently optimal decimation
500
            h->dctf.sub16x16_dct8( dct8x8, h->mb.pic.p_fenc[0], h->mb.pic.p_fdec[0] );
Laurent Aimar's avatar
Laurent Aimar committed
501

502
            for( idx = 0; idx < 4; idx++ )
Laurent Aimar's avatar
Laurent Aimar committed
503
            {
504 505
                if( h->mb.b_noise_reduction )
                    x264_denoise_dct( h, (int16_t*)dct8x8[idx] );
Loren Merritt's avatar
Loren Merritt committed
506 507 508 509
                if( h->mb.b_trellis )
                    x264_quant_8x8_trellis( h, dct8x8[idx], CQM_8PY, i_qp, 0 );
                else
                    quant_8x8( h, dct8x8[idx], h->quant8_mf[CQM_8PY], i_qp, 0 );
Laurent Aimar's avatar
Laurent Aimar committed
510

511
                scan_zigzag_8x8full( h->dct.luma8x8[idx], dct8x8[idx] );
Laurent Aimar's avatar
Laurent Aimar committed
512

513
                if( b_decimate )
514
                {
Loren Merritt's avatar
Loren Merritt committed
515 516 517 518 519 520
                    int i_decimate_8x8 = x264_mb_decimate_score( h->dct.luma8x8[idx], 64 );
                    i_decimate_mb += i_decimate_8x8;
                    if( i_decimate_8x8 < 4 )
                    {
                        memset( h->dct.luma8x8[idx], 0, sizeof( h->dct.luma8x8[idx] ) );
                        memset( dct8x8[idx], 0, sizeof( dct8x8[idx] ) );
521
                        nnz8x8[idx] = 0;
Loren Merritt's avatar
Loren Merritt committed
522
                    }
523
                }
524 525
                else
                    nnz8x8[idx] = array_non_zero( (int*)dct8x8[idx], sizeof(*dct8x8)/sizeof(int) );
Laurent Aimar's avatar
Laurent Aimar committed
526 527
            }

528
            if( i_decimate_mb < 6 && b_decimate )
529 530
                memset( h->dct.luma8x8, 0, sizeof( h->dct.luma8x8 ) );
            else
531 532 533 534 535
            {
                for( idx = 0; idx < 4; idx++ )
                    if( nnz8x8[idx] )
                    {
                        h->quantf.dequant_8x8( dct8x8[idx], h->dequant8_mf[CQM_8PY], i_qp );
536
                        h->dctf.add8x8_idct8( &h->mb.pic.p_fdec[0][(idx&1)*8 + (idx>>1)*8*FDEC_STRIDE], dct8x8[idx] );
537 538
                    }
            }
539 540 541 542
        }
        else
        {
            int16_t dct4x4[16][4][4];
543
            int nnz8x8[4] = {1,1,1,1};
544
            h->dctf.sub16x16_dct( dct4x4, h->mb.pic.p_fenc[0], h->mb.pic.p_fdec[0] );
545 546

            for( i8x8 = 0; i8x8 < 4; i8x8++ )
Laurent Aimar's avatar
Laurent Aimar committed
547
            {
548 549 550 551
                int i_decimate_8x8;

                /* encode one 4x4 block */
                i_decimate_8x8 = 0;
Laurent Aimar's avatar
Laurent Aimar committed
552 553 554
                for( i4x4 = 0; i4x4 < 4; i4x4++ )
                {
                    idx = i8x8 * 4 + i4x4;
555

556 557
                    if( h->mb.b_noise_reduction )
                        x264_denoise_dct( h, (int16_t*)dct4x4[idx] );
Loren Merritt's avatar
Loren Merritt committed
558 559 560 561 562
                    if( h->mb.b_trellis )
                        x264_quant_4x4_trellis( h, dct4x4[idx], CQM_4PY, i_qp, DCT_LUMA_4x4, 0 );
                    else
                        quant_4x4( h, dct4x4[idx], h->quant4_mf[CQM_4PY], i_qp, 0 );

563
                    scan_zigzag_4x4full( h->dct.block[idx].luma4x4, dct4x4[idx] );
564 565 566
                    
                    if( b_decimate )
                        i_decimate_8x8 += x264_mb_decimate_score( h->dct.block[idx].luma4x4, 16 );
Laurent Aimar's avatar
Laurent Aimar committed
567 568
                }

569 570
                /* decimate this 8x8 block */
                i_decimate_mb += i_decimate_8x8;
571
                if( i_decimate_8x8 < 4 && b_decimate )
Laurent Aimar's avatar
Laurent Aimar committed
572
                {
Loren Merritt's avatar
Loren Merritt committed
573 574
                    memset( &dct4x4[i8x8*4], 0, 4 * sizeof( *dct4x4 ) );
                    memset( &h->dct.block[i8x8*4], 0, 4 * sizeof( *h->dct.block ) );
575
                    nnz8x8[i8x8] = 0;
Laurent Aimar's avatar
Laurent Aimar committed
576 577
                }
            }
578

579
            if( i_decimate_mb < 6 && b_decimate )
Loren Merritt's avatar
Loren Merritt committed
580
                memset( h->dct.block, 0, 16 * sizeof( *h->dct.block ) );
581
            else
582 583 584 585 586 587
            {
                for( i8x8 = 0; i8x8 < 4; i8x8++ )
                    if( nnz8x8[i8x8] )
                    {
                        for( i = 0; i < 4; i++ )
                            h->quantf.dequant_4x4( dct4x4[i8x8*4+i], h->dequant4_mf[CQM_4PY], i_qp );
588
                        h->dctf.add8x8_idct( &h->mb.pic.p_fdec[0][(i8x8&1)*8 + (i8x8>>1)*8*FDEC_STRIDE], &dct4x4[i8x8*4] );
589 590
                    }
            }
Laurent Aimar's avatar
Laurent Aimar committed
591 592 593 594
        }
    }

    /* encode chroma */
595
    i_qp = i_chroma_qp_table[x264_clip3( i_qp + h->pps->i_chroma_qp_index_offset, 0, 51 )];
Laurent Aimar's avatar
Laurent Aimar committed
596 597 598
    if( IS_INTRA( h->mb.i_type ) )
    {
        const int i_mode = h->mb.i_chroma_pred_mode;
599 600
        h->predict_8x8c[i_mode]( h->mb.pic.p_fdec[1] );
        h->predict_8x8c[i_mode]( h->mb.pic.p_fdec[2] );
Laurent Aimar's avatar
Laurent Aimar committed
601 602 603
    }

    /* encode the 8x8 blocks */
604
    x264_mb_encode_8x8_chroma( h, !IS_INTRA( h->mb.i_type ), i_qp );
Laurent Aimar's avatar
Laurent Aimar committed
605

Loren Merritt's avatar
Loren Merritt committed
606
    /* Calculate the Luma/Chroma pattern and non_zero_count */
607
    h->mb.i_cbp_luma = 0x00;
Laurent Aimar's avatar
Laurent Aimar committed
608 609 610 611 612 613 614 615
    if( h->mb.i_type == I_16x16 )
    {
        for( i = 0; i < 16; i++ )
        {
            const int nz = array_non_zero_count( h->dct.block[i].residual_ac, 15 );
            h->mb.cache.non_zero_count[x264_scan8[i]] = nz;
            if( nz > 0 )
                h->mb.i_cbp_luma = 0x0f;
616 617 618 619
        }
    }
    else if( h->mb.b_transform_8x8 )
    {
Loren Merritt's avatar
Loren Merritt committed
620 621
        /* coded_block_flag is enough for CABAC.
         * the full non_zero_count is done only in CAVLC. */
622 623 624 625 626 627 628 629
        for( i = 0; i < 4; i++ )
        {
            const int nz = array_non_zero( h->dct.luma8x8[i], 64 );
            int j;
            for( j = 0; j < 4; j++ )
                h->mb.cache.non_zero_count[x264_scan8[4*i+j]] = nz;
            if( nz > 0 )
                h->mb.i_cbp_luma |= 1 << i;
Laurent Aimar's avatar
Laurent Aimar committed
630 631 632 633 634 635 636 637 638 639 640 641 642
        }
    }
    else
    {
        for( i = 0; i < 16; i++ )
        {
            const int nz = array_non_zero_count( h->dct.block[i].luma4x4, 16 );
            h->mb.cache.non_zero_count[x264_scan8[i]] = nz;
            if( nz > 0 )
                h->mb.i_cbp_luma |= 1 << (i/4);
        }
    }

Loren Merritt's avatar
Loren Merritt committed
643
    /* Calculate the chroma pattern */
Laurent Aimar's avatar
Laurent Aimar committed
644 645 646 647 648 649 650 651 652 653
    h->mb.i_cbp_chroma = 0x00;
    for( i = 0; i < 8; i++ )
    {
        const int nz = array_non_zero_count( h->dct.block[16+i].residual_ac, 15 );
        h->mb.cache.non_zero_count[x264_scan8[16+i]] = nz;
        if( nz > 0 )
        {
            h->mb.i_cbp_chroma = 0x02;    /* dc+ac (we can't do only ac) */
        }
    }
654
    if( h->mb.i_cbp_chroma == 0x00 && array_non_zero( h->dct.chroma_dc[0], 8 ) )
Laurent Aimar's avatar
Laurent Aimar committed
655 656 657 658 659 660
    {
        h->mb.i_cbp_chroma = 0x01;    /* dc only */
    }

    if( h->param.b_cabac )
    {
661 662 663
        i_cbp_dc = ( h->mb.i_type == I_16x16 && array_non_zero( h->dct.luma16x16_dc, 16 ) )
                 | array_non_zero( h->dct.chroma_dc[0], 4 ) << 1
                 | array_non_zero( h->dct.chroma_dc[1], 4 ) << 2;
Laurent Aimar's avatar
Laurent Aimar committed
664 665 666 667 668 669 670 671 672
    }

    /* store cbp */
    h->mb.cbp[h->mb.i_mb_xy] = (i_cbp_dc << 8) | (h->mb.i_cbp_chroma << 4) | h->mb.i_cbp_luma;

    /* Check for P_SKIP
     * XXX: in the me perhaps we should take x264_mb_predict_mv_pskip into account
     *      (if multiple mv give same result)*/
    if( h->mb.i_type == P_L0 && h->mb.i_partition == D_16x16 &&
673 674
        h->mb.i_cbp_luma == 0x00 && h->mb.i_cbp_chroma== 0x00 &&
        h->mb.cache.ref[0][x264_scan8[0]] == 0 )
Laurent Aimar's avatar
Laurent Aimar committed
675
    {
676
        int mvp[2];
Laurent Aimar's avatar
Laurent Aimar committed
677

678 679 680 681 682
        x264_mb_predict_mv_pskip( h, mvp );
        if( h->mb.cache.mv[0][x264_scan8[0]][0] == mvp[0] &&
            h->mb.cache.mv[0][x264_scan8[0]][1] == mvp[1] )
        {
            h->mb.i_type = P_SKIP;
Laurent Aimar's avatar
Laurent Aimar committed
683 684
        }
    }
685 686 687 688 689

    /* Check for B_SKIP */
    if( h->mb.i_type == B_DIRECT &&
        h->mb.i_cbp_luma == 0x00 && h->mb.i_cbp_chroma== 0x00 )
    {
690
        h->mb.i_type = B_SKIP;
691
    }
Laurent Aimar's avatar
Laurent Aimar committed
692 693 694
}

/*****************************************************************************
695 696
 * x264_macroblock_probe_skip:
 *  Check if the current MB could be encoded as a [PB]_SKIP (it supposes you use
Laurent Aimar's avatar
Laurent Aimar committed
697 698
 *  the previous QP
 *****************************************************************************/
699
int x264_macroblock_probe_skip( x264_t *h, int b_bidir )
Laurent Aimar's avatar
Laurent Aimar committed
700 701 702 703 704
{
    DECLARE_ALIGNED( int16_t, dct4x4[16][4][4], 16 );
    DECLARE_ALIGNED( int16_t, dct2x2[2][2], 16 );
    DECLARE_ALIGNED( int,     dctscan[16], 16 );

705
    int i_qp = h->mb.i_qp;
Laurent Aimar's avatar
Laurent Aimar committed
706 707 708 709 710 711
    int mvp[2];
    int ch;

    int i8x8, i4x4;
    int i_decimate_mb;

712 713 714 715
    if( !b_bidir )
    {
        /* Get the MV */
        x264_mb_predict_mv_pskip( h, mvp );
716 717
        mvp[0] = x264_clip3( mvp[0], h->mb.mv_min[0], h->mb.mv_max[0] );
        mvp[1] = x264_clip3( mvp[1], h->mb.mv_min[1], h->mb.mv_max[1] );
Laurent Aimar's avatar
Laurent Aimar committed
718

719
        /* Motion compensation */
720
        h->mc.mc_luma( h->mb.pic.p_fref[0][0], h->mb.pic.i_stride[0],
721 722
                       h->mb.pic.p_fdec[0],    FDEC_STRIDE,
                       mvp[0], mvp[1], 16, 16 );
723
    }
Laurent Aimar's avatar
Laurent Aimar committed
724 725

    /* get luma diff */
726 727
    h->dctf.sub16x16_dct( dct4x4, h->mb.pic.p_fenc[0],
                                  h->mb.pic.p_fdec[0] );
Laurent Aimar's avatar
Laurent Aimar committed
728 729 730 731 732 733 734 735

    for( i8x8 = 0, i_decimate_mb = 0; i8x8 < 4; i8x8++ )
    {
        /* encode one 4x4 block */
        for( i4x4 = 0; i4x4 < 4; i4x4++ )
        {
            const int idx = i8x8 * 4 + i4x4;

736
            quant_4x4( h, dct4x4[idx], (int(*)[4][4])def_quant4_mf, i_qp, 0 );
Laurent Aimar's avatar
Laurent Aimar committed
737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756
            scan_zigzag_4x4full( dctscan, dct4x4[idx] );

            i_decimate_mb += x264_mb_decimate_score( dctscan, 16 );

            if( i_decimate_mb >= 6 )
            {
                /* not as P_SKIP */
                return 0;
            }
        }
    }

    /* encode chroma */
    i_qp = i_chroma_qp_table[x264_clip3( i_qp + h->pps->i_chroma_qp_index_offset, 0, 51 )];

    for( ch = 0; ch < 2; ch++ )
    {
        uint8_t  *p_src = h->mb.pic.p_fenc[1+ch];
        uint8_t  *p_dst = h->mb.pic.p_fdec[1+ch];

757 758
        if( !b_bidir )
        {
759 760 761
            h->mc.mc_chroma( h->mb.pic.p_fref[0][0][4+ch], h->mb.pic.i_stride[1+ch],
                             h->mb.pic.p_fdec[1+ch],       FDEC_STRIDE,
                             mvp[0], mvp[1], 8, 8 );
762
        }
Laurent Aimar's avatar
Laurent Aimar committed
763

764
        h->dctf.sub8x8_dct( dct4x4, p_src, p_dst );
Laurent Aimar's avatar
Laurent Aimar committed
765 766 767 768 769 770 771

        /* calculate dct DC */
        dct2x2[0][0] = dct4x4[0][0][0];
        dct2x2[0][1] = dct4x4[1][0][0];
        dct2x2[1][0] = dct4x4[2][0][0];
        dct2x2[1][1] = dct4x4[3][0][0];
        h->dctf.dct2x2dc( dct2x2 );
772
        quant_2x2_dc( h, dct2x2, (int(*)[4][4])def_quant4_mf, i_qp, 0 );
Laurent Aimar's avatar
Laurent Aimar committed
773 774 775 776 777 778 779 780 781
        if( dct2x2[0][0] || dct2x2[0][1] || dct2x2[1][0] || dct2x2[1][1]  )
        {
            /* can't be */
            return 0;
        }

        /* calculate dct coeffs */
        for( i4x4 = 0, i_decimate_mb = 0; i4x4 < 4; i4x4++ )
        {
782
            quant_4x4( h, dct4x4[i4x4], (int(*)[4][4])def_quant4_mf, i_qp, 0 );
Laurent Aimar's avatar
Laurent Aimar committed
783 784 785 786 787 788 789 790 791 792 793 794
            scan_zigzag_4x4( dctscan, dct4x4[i4x4] );

            i_decimate_mb += x264_mb_decimate_score( dctscan, 15 );
            if( i_decimate_mb >= 7 )
            {
                return 0;
            }
        }
    }

    return 1;
}
795 796 797 798 799 800 801 802 803

/****************************************************************************
 * DCT-domain noise reduction / adaptive deadzone
 * from libavcodec
 ****************************************************************************/

void x264_noise_reduction_update( x264_t *h )
{
    int cat, i;
804
    for( cat = 0; cat < 2; cat++ )
805
    {
806 807
        int size = cat ? 64 : 16;
        const int *weight = cat ? x264_dct8_weight2_tab : x264_dct4_weight2_tab;
808

809
        if( h->nr_count[cat] > (cat ? (1<<16) : (1<<18)) )
810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825
        {
            for( i = 0; i < size; i++ )
                h->nr_residual_sum[cat][i] >>= 1;
            h->nr_count[cat] >>= 1;
        }

        for( i = 0; i < size; i++ )
            h->nr_offset[cat][i] =
                ((uint64_t)h->param.analyse.i_noise_reduction * h->nr_count[cat]
                 + h->nr_residual_sum[cat][i]/2)
              / ((uint64_t)h->nr_residual_sum[cat][i] * weight[i]/256 + 1);
    }
}

void x264_denoise_dct( x264_t *h, int16_t *dct )
{
826
    const int cat = h->mb.b_transform_8x8;
827 828 829 830
    int i;

    h->nr_count[cat]++;

831
    for( i = (cat ? 63 : 15); i >= 1; i-- )
832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853
    {
        int level = dct[i];
        if( level )
        {
            if( level > 0 )
            {
                h->nr_residual_sum[cat][i] += level;
                level -= h->nr_offset[cat][i];
                if( level < 0 )
                    level = 0;
            }
            else
            {
                h->nr_residual_sum[cat][i] -= level;
                level += h->nr_offset[cat][i];
                if( level > 0 )
                    level = 0;
            }
            dct[i] = level;
        }
    }
}
854 855 856 857 858 859 860 861 862 863

/*****************************************************************************
 * RD only; 4 calls to this do not make up for one macroblock_encode.
 * doesn't transform chroma dc.
 *****************************************************************************/
void x264_macroblock_encode_p8x8( x264_t *h, int i8 )
{
    int i_qp = h->mb.i_qp;
    uint8_t *p_fenc = h->mb.pic.p_fenc[0] + (i8&1)*8 + (i8>>1)*8*FENC_STRIDE;
    uint8_t *p_fdec = h->mb.pic.p_fdec[0] + (i8&1)*8 + (i8>>1)*8*FDEC_STRIDE;
864 865
    int b_decimate = h->sh.i_type == SLICE_TYPE_B || h->param.analyse.b_dct_decimate;
    int nnz8x8;
866 867 868 869 870 871 872
    int ch;

    x264_mb_mc_8x8( h, i8 );

    if( h->mb.b_transform_8x8 )
    {
        int16_t dct8x8[8][8];
873
        h->dctf.sub8x8_dct8( dct8x8, p_fenc, p_fdec );
874 875 876
        quant_8x8( h, dct8x8, h->quant8_mf[CQM_8PY], i_qp, 0 );
        scan_zigzag_8x8full( h->dct.luma8x8[i8], dct8x8 );

877 878 879 880 881
        if( b_decimate )
            nnz8x8 = 4 <= x264_mb_decimate_score( h->dct.luma8x8[i8], 64 );
        else
            nnz8x8 = array_non_zero( (int*)dct8x8, sizeof(dct8x8)/sizeof(int) );

882 883 884
        if( nnz8x8 )
        {
            h->quantf.dequant_8x8( dct8x8, h->dequant8_mf[CQM_8PY], i_qp );
885
            h->dctf.add8x8_idct8( p_fdec, dct8x8 );
886 887 888 889
        }
    }
    else
    {
890
        int i4;
891
        int16_t dct4x4[4][4][4];
892
        h->dctf.sub8x8_dct( dct4x4, p_fenc, p_fdec );
893 894 895 896
        quant_4x4( h, dct4x4[0], h->quant4_mf[CQM_4PY], i_qp, 0 );
        quant_4x4( h, dct4x4[1], h->quant4_mf[CQM_4PY], i_qp, 0 );
        quant_4x4( h, dct4x4[2], h->quant4_mf[CQM_4PY], i_qp, 0 );
        quant_4x4( h, dct4x4[3], h->quant4_mf[CQM_4PY], i_qp, 0 );
897
        for( i4 = 0; i4 < 4; i4++ )
898
            scan_zigzag_4x4full( h->dct.block[i8*4+i4].luma4x4, dct4x4[i4] );
899

900
        if( b_decimate )
901
        {
902 903 904 905
            int i_decimate_8x8 = 0;
            for( i4 = 0; i4 < 4 && i_decimate_8x8 < 4; i4++ )
                i_decimate_8x8 += x264_mb_decimate_score( h->dct.block[i8*4+i4].luma4x4, 16 );
            nnz8x8 = 4 <= i_decimate_8x8;
906
        }
907 908 909
        else
            nnz8x8 = array_non_zero( (int*)dct4x4, sizeof(dct4x4)/sizeof(int) );

910 911 912
        if( nnz8x8 )
        {
            for( i4 = 0; i4 < 4; i4++ )
913
                h->quantf.dequant_4x4( dct4x4[i4], h->dequant4_mf[CQM_4PY], i_qp );
914
            h->dctf.add8x8_idct( p_fdec, dct4x4 );
915 916 917 918 919 920 921 922 923 924 925
        }
    }

    i_qp = i_chroma_qp_table[x264_clip3( i_qp + h->pps->i_chroma_qp_index_offset, 0, 51 )];

    for( ch = 0; ch < 2; ch++ )
    {
        int16_t dct4x4[4][4];
        p_fenc = h->mb.pic.p_fenc[1+ch] + (i8&1)*4 + (i8>>1)*4*FENC_STRIDE;
        p_fdec = h->mb.pic.p_fdec[1+ch] + (i8&1)*4 + (i8>>1)*4*FDEC_STRIDE;

926
        h->dctf.sub4x4_dct( dct4x4, p_fenc, p_fdec );
927 928
        quant_4x4( h, dct4x4, h->quant4_mf[CQM_4PC], i_qp, 0 );
        scan_zigzag_4x4( h->dct.block[16+i8+ch*4].residual_ac, dct4x4 );
929 930 931 932 933
        if( array_non_zero( (int*)dct4x4, sizeof(dct4x4)/sizeof(int) ) )
        {
            h->quantf.dequant_4x4( dct4x4, h->dequant4_mf[CQM_4PC], i_qp );
            h->dctf.add4x4_idct( p_fdec, dct4x4 );
        }
934 935 936 937 938 939 940 941
    }

    if( nnz8x8 )
        h->mb.i_cbp_luma |= (1 << i8);
    else
        h->mb.i_cbp_luma &= ~(1 << i8);
    h->mb.i_cbp_chroma = 0x02;
}