macroblock.c 29.3 KB
Newer Older
Laurent Aimar's avatar
Laurent Aimar committed
1 2 3
/*****************************************************************************
 * macroblock.c: h264 encoder library
 *****************************************************************************
4
 * Copyright (C) 2003-2008 x264 project
Laurent Aimar's avatar
Laurent Aimar committed
5 6
 *
 * Authors: Laurent Aimar <fenrir@via.ecp.fr>
7 8
 *          Loren Merritt <lorenm@u.washington.edu>
 *          Fiona Glaser <fiona@x264.com>
Laurent Aimar's avatar
Laurent Aimar committed
9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
 *
 * 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.
 *****************************************************************************/

25
#include "common/common.h"
Laurent Aimar's avatar
Laurent Aimar committed
26 27 28
#include "macroblock.h"


29
#define ZIG(i,y,x) level[i] = dct[x][y];
30
static inline void zigzag_scan_2x2_dc( int16_t level[4], int16_t dct[2][2] )
Laurent Aimar's avatar
Laurent Aimar committed
31
{
32 33 34 35
    ZIG(0,0,0)
    ZIG(1,0,1)
    ZIG(2,1,0)
    ZIG(3,1,1)
Laurent Aimar's avatar
Laurent Aimar committed
36
}
37
#undef ZIG
Laurent Aimar's avatar
Laurent Aimar committed
38 39 40 41 42 43 44 45 46

/* (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
 */
47
static int x264_mb_decimate_score( int16_t *dct, int i_max )
Laurent Aimar's avatar
Laurent Aimar committed
48
{
49 50 51 52 53 54 55 56 57
    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
58 59 60 61 62 63 64 65 66 67
    int i_score = 0;
    int idx = i_max - 1;

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

    while( idx >= 0 )
    {
        int i_run;

68
        if( (unsigned)(dct[idx--] + 1) > 2 )
Laurent Aimar's avatar
Laurent Aimar committed
69 70 71 72 73 74 75 76
            return 9;

        i_run = 0;
        while( idx >= 0 && dct[idx] == 0 )
        {
            idx--;
            i_run++;
        }
77
        i_score += ds_table[i_run];
Laurent Aimar's avatar
Laurent Aimar committed
78 79 80 81 82 83 84
    }

    return i_score;
}

void x264_mb_encode_i4x4( x264_t *h, int idx, int i_qscale )
{
85 86 87 88
    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];
Loren Merritt's avatar
Loren Merritt committed
89
    DECLARE_ALIGNED_16( int16_t dct4x4[4][4] );
Laurent Aimar's avatar
Laurent Aimar committed
90

Loren Merritt's avatar
Loren Merritt committed
91 92
    if( h->mb.b_lossless )
    {
93
        h->zigzagf.sub_4x4( h->dct.luma4x4[idx], p_src, p_dst );
Loren Merritt's avatar
Loren Merritt committed
94 95 96
        return;
    }

97
    h->dctf.sub4x4_dct( dct4x4, p_src, p_dst );
Loren Merritt's avatar
Loren Merritt committed
98 99 100 101

    if( h->mb.b_trellis )
        x264_quant_4x4_trellis( h, dct4x4, CQM_4IY, i_qscale, DCT_LUMA_4x4, 1 );
    else
Loren Merritt's avatar
Loren Merritt committed
102
        h->quantf.quant_4x4( dct4x4, h->quant4_mf[CQM_4IY][i_qscale], h->quant4_bias[CQM_4IY][i_qscale] );
Loren Merritt's avatar
Loren Merritt committed
103

104 105 106 107
    if( array_non_zero( dct4x4 ) )
    {
        h->zigzagf.scan_4x4( h->dct.luma4x4[idx], dct4x4 );
        h->quantf.dequant_4x4( dct4x4, h->dequant4_mf[CQM_4IY], i_qscale );
Laurent Aimar's avatar
Laurent Aimar committed
108

109 110 111 112 113
        /* output samples to fdec */
        h->dctf.add4x4_idct( p_dst, dct4x4 );
    }
    else
        memset( h->dct.luma4x4[idx], 0, sizeof(h->dct.luma4x4[idx]));
Laurent Aimar's avatar
Laurent Aimar committed
114 115
}

116 117
void x264_mb_encode_i8x8( x264_t *h, int idx, int i_qscale )
{
118 119 120 121
    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];
Loren Merritt's avatar
Loren Merritt committed
122
    DECLARE_ALIGNED_16( int16_t dct8x8[8][8] );
123

124
    h->dctf.sub8x8_dct8( dct8x8, p_src, p_dst );
Loren Merritt's avatar
Loren Merritt committed
125 126 127 128

    if( h->mb.b_trellis )
        x264_quant_8x8_trellis( h, dct8x8, CQM_8IY, i_qscale, 1 );
    else 
Loren Merritt's avatar
Loren Merritt committed
129
        h->quantf.quant_8x8( dct8x8, h->quant8_mf[CQM_8IY][i_qscale], h->quant8_bias[CQM_8IY][i_qscale] );
Loren Merritt's avatar
Loren Merritt committed
130

131
    h->zigzagf.scan_8x8( h->dct.luma8x8[idx], dct8x8 );
132
    h->quantf.dequant_8x8( dct8x8, h->dequant8_mf[CQM_8IY], i_qscale );
133
    h->dctf.add8x8_idct8( p_dst, dct8x8 );
134 135
}

Laurent Aimar's avatar
Laurent Aimar committed
136 137 138 139 140
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];

141 142
    DECLARE_ALIGNED_16( int16_t dct4x4[16][4][4] );
    DECLARE_ALIGNED_16( int16_t dct_dc4x4[4][4] );
Laurent Aimar's avatar
Laurent Aimar committed
143 144 145

    int i;

Loren Merritt's avatar
Loren Merritt committed
146 147 148 149
    if( h->mb.b_lossless )
    {
        for( i = 0; i < 16; i++ )
        {
150 151
            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;
152
            h->zigzagf.sub_4x4( h->dct.luma4x4[i], p_src+oe, p_dst+od );
153
            dct_dc4x4[block_idx_x[i]][block_idx_y[i]] = h->dct.luma4x4[i][0];
Fiona Glaser's avatar
Fiona Glaser committed
154
            h->dct.luma4x4[i][0] = 0;
Loren Merritt's avatar
Loren Merritt committed
155
        }
156
        h->zigzagf.scan_4x4( h->dct.luma16x16_dc, dct_dc4x4 );
Loren Merritt's avatar
Loren Merritt committed
157 158 159
        return;
    }

160
    h->dctf.sub16x16_dct( dct4x4, p_src, p_dst );
Laurent Aimar's avatar
Laurent Aimar committed
161 162 163
    for( i = 0; i < 16; i++ )
    {
        /* copy dc coeff */
164 165
        dct_dc4x4[block_idx_y[i]][block_idx_x[i]] = dct4x4[i][0][0];
        dct4x4[i][0][0] = 0;
Laurent Aimar's avatar
Laurent Aimar committed
166 167

        /* quant/scan/dequant */
Loren Merritt's avatar
Loren Merritt committed
168
        if( h->mb.b_trellis )
169
            x264_quant_4x4_trellis( h, dct4x4[i], CQM_4IY, i_qscale, DCT_LUMA_AC, 1 );
Loren Merritt's avatar
Loren Merritt committed
170
        else
171
            h->quantf.quant_4x4( dct4x4[i], h->quant4_mf[CQM_4IY][i_qscale], h->quant4_bias[CQM_4IY][i_qscale] );
Loren Merritt's avatar
Loren Merritt committed
172

173 174
        h->zigzagf.scan_4x4( h->dct.luma4x4[i], dct4x4[i] );
        h->quantf.dequant_4x4( dct4x4[i], h->dequant4_mf[CQM_4IY], i_qscale );
Laurent Aimar's avatar
Laurent Aimar committed
175 176
    }

177 178 179
    h->dctf.dct4x4dc( dct_dc4x4 );
    h->quantf.quant_4x4_dc( dct_dc4x4, h->quant4_mf[CQM_4IY][i_qscale][0]>>1, h->quant4_bias[CQM_4IY][i_qscale][0]<<1 );
    h->zigzagf.scan_4x4( h->dct.luma16x16_dc, dct_dc4x4 );
Laurent Aimar's avatar
Laurent Aimar committed
180 181

    /* output samples to fdec */
182 183
    h->dctf.idct4x4dc( dct_dc4x4 );
    x264_mb_dequant_4x4_dc( dct_dc4x4, h->dequant4_mf[CQM_4IY], i_qscale );  /* XXX not inversed */
Laurent Aimar's avatar
Laurent Aimar committed
184 185 186 187 188

    /* calculate dct coeffs */
    for( i = 0; i < 16; i++ )
    {
        /* copy dc coeff */
189
        dct4x4[i][0][0] = dct_dc4x4[block_idx_y[i]][block_idx_x[i]];
Laurent Aimar's avatar
Laurent Aimar committed
190 191
    }
    /* put pixels to fdec */
192
    h->dctf.add16x16_idct( p_dst, dct4x4 );
Laurent Aimar's avatar
Laurent Aimar committed
193 194
}

195
void x264_mb_encode_8x8_chroma( x264_t *h, int b_inter, int i_qscale )
Laurent Aimar's avatar
Laurent Aimar committed
196 197
{
    int i, ch;
198
    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
199 200 201 202 203 204 205

    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;

Loren Merritt's avatar
Loren Merritt committed
206 207
        DECLARE_ALIGNED_16( int16_t dct2x2[2][2]  );
        DECLARE_ALIGNED_16( int16_t dct4x4[4][4][4] );
Laurent Aimar's avatar
Laurent Aimar committed
208

Loren Merritt's avatar
Loren Merritt committed
209 210 211 212
        if( h->mb.b_lossless )
        {
            for( i = 0; i < 4; i++ )
            {
213 214
                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;
215
                h->zigzagf.sub_4x4( h->dct.luma4x4[16+i+ch*4], p_src+oe, p_dst+od );
Loren Merritt's avatar
Loren Merritt committed
216
                h->dct.chroma_dc[ch][i] = h->dct.luma4x4[16+i+ch*4][0];
Fiona Glaser's avatar
Fiona Glaser committed
217
                h->dct.luma4x4[16+i+ch*4][0] = 0;
Loren Merritt's avatar
Loren Merritt committed
218 219 220 221
            }
            continue;
        }
            
222
        h->dctf.sub8x8_dct( dct4x4, p_src, p_dst );
Laurent Aimar's avatar
Laurent Aimar committed
223 224 225 226 227
        /* 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];
228
            dct4x4[i][0][0] = 0;
Laurent Aimar's avatar
Laurent Aimar committed
229

Loren Merritt's avatar
Loren Merritt committed
230
            /* no trellis; it doesn't seem to help chroma noticeably */
Loren Merritt's avatar
Loren Merritt committed
231
            h->quantf.quant_4x4( dct4x4[i], h->quant4_mf[CQM_4IC+b_inter][i_qscale], h->quant4_bias[CQM_4IC+b_inter][i_qscale] );
232
            h->zigzagf.scan_4x4( h->dct.luma4x4[16+i+ch*4], dct4x4[i] );
Laurent Aimar's avatar
Laurent Aimar committed
233

234
            if( b_decimate )
Laurent Aimar's avatar
Laurent Aimar committed
235
            {
236
                i_decimate_score += x264_mb_decimate_score( h->dct.luma4x4[16+i+ch*4]+1, 15 );
Laurent Aimar's avatar
Laurent Aimar committed
237 238 239 240
            }
        }

        h->dctf.dct2x2dc( dct2x2 );
Loren Merritt's avatar
Loren Merritt committed
241
        h->quantf.quant_2x2_dc( dct2x2, h->quant4_mf[CQM_4IC+b_inter][i_qscale][0]>>1, h->quant4_bias[CQM_4IC+b_inter][i_qscale][0]<<1 );
242
        zigzag_scan_2x2_dc( h->dct.chroma_dc[ch], dct2x2 );
Laurent Aimar's avatar
Laurent Aimar committed
243 244 245

        /* output samples to fdec */
        h->dctf.idct2x2dc( dct2x2 );
Loren Merritt's avatar
Loren Merritt committed
246
        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
247

248
        if( b_decimate && i_decimate_score < 7 )
Laurent Aimar's avatar
Laurent Aimar committed
249 250
        {
            /* Near null chroma 8x8 block so make it null (bits saving) */
251
            memset( &h->dct.luma4x4[16+ch*4], 0, 4 * sizeof( *h->dct.luma4x4 ) );
252
            if( !array_non_zero( dct2x2 ) )
253 254
                continue;
            memset( dct4x4, 0, sizeof( dct4x4 ) );
Laurent Aimar's avatar
Laurent Aimar committed
255
        }
256 257 258 259 260
        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
261 262

        for( i = 0; i < 4; i++ )
263
            dct4x4[i][0][0] = dct2x2[0][i];
264
        h->dctf.add8x8_idct( p_dst, dct4x4 );
Laurent Aimar's avatar
Laurent Aimar committed
265
    }
266 267 268 269 270

    /* coded block pattern */
    h->mb.i_cbp_chroma = 0;
    for( i = 0; i < 8; i++ )
    {
271
        int nz = array_non_zero( h->dct.luma4x4[16+i] );
272 273 274 275 276
        h->mb.cache.non_zero_count[x264_scan8[16+i]] = nz;
        h->mb.i_cbp_chroma |= nz;
    }
    if( h->mb.i_cbp_chroma )
        h->mb.i_cbp_chroma = 2;    /* dc+ac (we can't do only ac) */
277
    else if( array_non_zero( h->dct.chroma_dc ) )
278
        h->mb.i_cbp_chroma = 1;    /* dc only */
Laurent Aimar's avatar
Laurent Aimar committed
279 280
}

281 282 283 284
static void x264_macroblock_encode_skip( x264_t *h )
{
    h->mb.i_cbp_luma = 0x00;
    h->mb.i_cbp_chroma = 0x00;
285
    memset( h->mb.cache.non_zero_count, 0, X264_SCAN8_SIZE );
286 287 288 289
    /* store cbp */
    h->mb.cbp[h->mb.i_mb_xy] = 0;
}

Laurent Aimar's avatar
Laurent Aimar committed
290 291 292 293 294 295
/*****************************************************************************
 * x264_macroblock_encode_pskip:
 *  Encode an already marked skip block
 *****************************************************************************/
void x264_macroblock_encode_pskip( x264_t *h )
{
296 297 298 299
    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
300

301
    /* don't do pskip motion compensation if it was already done in macroblock_analyse */
302
    if( !h->mb.b_skip_mc )
303 304 305 306
    {
        h->mc.mc_luma( h->mb.pic.p_fdec[0],    FDEC_STRIDE,
                       h->mb.pic.p_fref[0][0], h->mb.pic.i_stride[0],
                       mvx, mvy, 16, 16 );
Laurent Aimar's avatar
Laurent Aimar committed
307

308 309 310
        h->mc.mc_chroma( h->mb.pic.p_fdec[1],       FDEC_STRIDE,
                         h->mb.pic.p_fref[0][0][4], h->mb.pic.i_stride[1],
                         mvx, mvy, 8, 8 );
Laurent Aimar's avatar
Laurent Aimar committed
311

312 313 314 315
        h->mc.mc_chroma( h->mb.pic.p_fdec[2],       FDEC_STRIDE,
                         h->mb.pic.p_fref[0][0][5], h->mb.pic.i_stride[2],
                         mvx, mvy, 8, 8 );
    }
Laurent Aimar's avatar
Laurent Aimar committed
316

317
    x264_macroblock_encode_skip( h );
Laurent Aimar's avatar
Laurent Aimar committed
318 319 320 321 322 323 324 325
}

/*****************************************************************************
 * x264_macroblock_encode:
 *****************************************************************************/
void x264_macroblock_encode( x264_t *h )
{
    int i_cbp_dc = 0;
326
    int i_qp = h->mb.i_qp;
327
    int b_decimate = h->sh.i_type == SLICE_TYPE_B || h->param.analyse.b_dct_decimate;
328
    int b_force_no_skip = 0;
329 330
    int i,j,idx;
    uint8_t nnz8x8[4] = {1,1,1,1};
Laurent Aimar's avatar
Laurent Aimar committed
331

332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348
    if( h->sh.b_mbaff
        && h->mb.i_mb_xy == h->sh.i_first_mb + h->mb.i_mb_stride
        && IS_SKIP(h->mb.type[h->sh.i_first_mb]) )
    {
        /* The first skip is predicted to be a frame mb pair.
         * We don't yet support the aff part of mbaff, so force it to non-skip
         * so that we can pick the aff flag. */
        b_force_no_skip = 1;
        if( IS_SKIP(h->mb.i_type) )
        {
            if( h->mb.i_type == P_SKIP )
                h->mb.i_type = P_L0;
            else if( h->mb.i_type == B_SKIP )
                h->mb.i_type = B_DIRECT;
        }
    }

Laurent Aimar's avatar
Laurent Aimar committed
349 350 351 352 353 354
    if( h->mb.i_type == P_SKIP )
    {
        /* A bit special */
        x264_macroblock_encode_pskip( h );
        return;
    }
355 356
    if( h->mb.i_type == B_SKIP )
    {
357
        /* don't do bskip motion compensation if it was already done in macroblock_analyse */
358
        if( !h->mb.b_skip_mc )
359
            x264_mb_mc( h );
360 361 362
        x264_macroblock_encode_skip( h );
        return;
    }
Laurent Aimar's avatar
Laurent Aimar committed
363 364 365 366

    if( h->mb.i_type == I_16x16 )
    {
        const int i_mode = h->mb.i_intra16x16_pred_mode;
367
        h->mb.b_transform_8x8 = 0;
Laurent Aimar's avatar
Laurent Aimar committed
368
        /* do the right prediction */
369
        h->predict_16x16[i_mode]( h->mb.pic.p_fdec[0] );
Laurent Aimar's avatar
Laurent Aimar committed
370 371

        /* encode the 16x16 macroblock */
372
        x264_mb_encode_i16x16( h, i_qp );
Laurent Aimar's avatar
Laurent Aimar committed
373
    }
374 375
    else if( h->mb.i_type == I_8x8 )
    {
Loren Merritt's avatar
Loren Merritt committed
376
        DECLARE_ALIGNED_16( uint8_t edge[33] );
377
        h->mb.b_transform_8x8 = 1;
378 379 380 381 382 383 384 385 386
        /* If we already encoded 3 of the 4 i8x8 blocks, we don't have to do them again. */
        if( h->mb.i_skip_intra )
        {
            h->mc.copy[PIXEL_16x16]( h->mb.pic.p_fdec[0], FDEC_STRIDE, h->mb.pic.i8x8_fdec_buf, 16, 16 );
            /* In RD mode, restore the now-overwritten DCT data. */
            if( h->mb.i_skip_intra == 2 )
                h->mc.memcpy_aligned( h->dct.luma8x8, h->mb.pic.i8x8_dct_buf, sizeof(h->mb.pic.i8x8_dct_buf) );
        }
        for( i = h->mb.i_skip_intra ? 3 : 0 ; i < 4; i++ )
387
        {
388
            uint8_t  *p_dst = &h->mb.pic.p_fdec[0][8 * (i&1) + 8 * (i>>1) * FDEC_STRIDE];
389 390
            int      i_mode = h->mb.cache.intra4x4_pred_mode[x264_scan8[4*i]];

391 392
            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 );
393
            x264_mb_encode_i8x8( h, i, i_qp );
394
        }
395 396
        for( i = 0; i < 4; i++ )
            nnz8x8[i] = array_non_zero( h->dct.luma8x8[i] );
397
    }
Laurent Aimar's avatar
Laurent Aimar committed
398 399
    else if( h->mb.i_type == I_4x4 )
    {
400
        h->mb.b_transform_8x8 = 0;
401 402 403 404 405 406
        /* If we already encoded 15 of the 16 i4x4 blocks, we don't have to do them again. */
        if( h->mb.i_skip_intra )
        {
            h->mc.copy[PIXEL_16x16]( h->mb.pic.p_fdec[0], FDEC_STRIDE, h->mb.pic.i4x4_fdec_buf, 16, 16 );
            /* In RD mode, restore the now-overwritten DCT data. */
            if( h->mb.i_skip_intra == 2 )
407
                h->mc.memcpy_aligned( h->dct.luma4x4, h->mb.pic.i4x4_dct_buf, sizeof(h->mb.pic.i4x4_dct_buf) );
408 409
        }
        for( i = h->mb.i_skip_intra ? 15 : 0 ; i < 16; i++ )
Laurent Aimar's avatar
Laurent Aimar committed
410
        {
411
            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
412 413
            int      i_mode = h->mb.cache.intra4x4_pred_mode[x264_scan8[i]];

414 415
            if( (h->mb.i_neighbour4[i] & (MB_TOPRIGHT|MB_TOP)) == MB_TOP )
                /* emulate missing topright samples */
416
                *(uint32_t*) &p_dst[4-FDEC_STRIDE] = p_dst[3-FDEC_STRIDE] * 0x01010101U;
417

418
            h->predict_4x4[i_mode]( p_dst );
419
            x264_mb_encode_i4x4( h, i, i_qp );
Laurent Aimar's avatar
Laurent Aimar committed
420 421 422 423
        }
    }
    else    /* Inter MB */
    {
424
        int i8x8, i4x4;
Laurent Aimar's avatar
Laurent Aimar committed
425 426
        int i_decimate_mb = 0;

427 428 429
        /* Don't repeat motion compensation if it was already done in non-RD transform analysis */
        if( !h->mb.b_skip_mc )
            x264_mb_mc( h );
Laurent Aimar's avatar
Laurent Aimar committed
430

Loren Merritt's avatar
Loren Merritt committed
431 432 433 434
        if( h->mb.b_lossless )
        {
            for( i4x4 = 0; i4x4 < 16; i4x4++ )
            {
435 436
                int x = 4*block_idx_x[i4x4];
                int y = 4*block_idx_y[i4x4];
437
                h->zigzagf.sub_4x4( h->dct.luma4x4[i4x4],
438
                                    h->mb.pic.p_fenc[0]+x+y*FENC_STRIDE,
439
                                    h->mb.pic.p_fdec[0]+x+y*FDEC_STRIDE );
Loren Merritt's avatar
Loren Merritt committed
440 441 442
            }
        }
        else if( h->mb.b_transform_8x8 )
Laurent Aimar's avatar
Laurent Aimar committed
443
        {
Loren Merritt's avatar
Loren Merritt committed
444
            DECLARE_ALIGNED_16( int16_t dct8x8[4][8][8] );
445
            b_decimate &= !h->mb.b_trellis; // 8x8 trellis is inherently optimal decimation
446
            h->dctf.sub16x16_dct8( dct8x8, h->mb.pic.p_fenc[0], h->mb.pic.p_fdec[0] );
Fiona Glaser's avatar
Fiona Glaser committed
447
            h->nr_count[1] += h->mb.b_noise_reduction * 4;
Laurent Aimar's avatar
Laurent Aimar committed
448

449
            for( idx = 0; idx < 4; idx++ )
Laurent Aimar's avatar
Laurent Aimar committed
450
            {
451
                if( h->mb.b_noise_reduction )
Fiona Glaser's avatar
Fiona Glaser committed
452
                    h->quantf.denoise_dct_core( *dct8x8[idx], h->nr_residual_sum[1], h->nr_offset[1], 64 );
Loren Merritt's avatar
Loren Merritt committed
453 454 455
                if( h->mb.b_trellis )
                    x264_quant_8x8_trellis( h, dct8x8[idx], CQM_8PY, i_qp, 0 );
                else
Loren Merritt's avatar
Loren Merritt committed
456
                    h->quantf.quant_8x8( dct8x8[idx], h->quant8_mf[CQM_8PY][i_qp], h->quant8_bias[CQM_8PY][i_qp] );
Laurent Aimar's avatar
Laurent Aimar committed
457

458
                h->zigzagf.scan_8x8( h->dct.luma8x8[idx], dct8x8[idx] );
Laurent Aimar's avatar
Laurent Aimar committed
459

460
                if( b_decimate )
461
                {
Loren Merritt's avatar
Loren Merritt committed
462 463 464
                    int i_decimate_8x8 = x264_mb_decimate_score( h->dct.luma8x8[idx], 64 );
                    i_decimate_mb += i_decimate_8x8;
                    if( i_decimate_8x8 < 4 )
465
                        nnz8x8[idx] = 0;
466
                }
467
                else
468
                    nnz8x8[idx] = array_non_zero( dct8x8[idx] );
Laurent Aimar's avatar
Laurent Aimar committed
469 470
            }

471
            if( i_decimate_mb < 6 && b_decimate )
472
                *(uint32_t*)nnz8x8 = 0;
473
            else
474 475 476 477 478
            {
                for( idx = 0; idx < 4; idx++ )
                    if( nnz8x8[idx] )
                    {
                        h->quantf.dequant_8x8( dct8x8[idx], h->dequant8_mf[CQM_8PY], i_qp );
479
                        h->dctf.add8x8_idct8( &h->mb.pic.p_fdec[0][(idx&1)*8 + (idx>>1)*8*FDEC_STRIDE], dct8x8[idx] );
480 481
                    }
            }
482 483 484
        }
        else
        {
Loren Merritt's avatar
Loren Merritt committed
485
            DECLARE_ALIGNED_16( int16_t dct4x4[16][4][4] );
486
            h->dctf.sub16x16_dct( dct4x4, h->mb.pic.p_fenc[0], h->mb.pic.p_fdec[0] );
Fiona Glaser's avatar
Fiona Glaser committed
487
            h->nr_count[0] += h->mb.b_noise_reduction * 16;
488 489

            for( i8x8 = 0; i8x8 < 4; i8x8++ )
Laurent Aimar's avatar
Laurent Aimar committed
490
            {
491 492 493 494
                int i_decimate_8x8;

                /* encode one 4x4 block */
                i_decimate_8x8 = 0;
Laurent Aimar's avatar
Laurent Aimar committed
495 496 497
                for( i4x4 = 0; i4x4 < 4; i4x4++ )
                {
                    idx = i8x8 * 4 + i4x4;
498

499
                    if( h->mb.b_noise_reduction )
Fiona Glaser's avatar
Fiona Glaser committed
500
                        h->quantf.denoise_dct_core( *dct4x4[idx], h->nr_residual_sum[0], h->nr_offset[0], 16 );
Loren Merritt's avatar
Loren Merritt committed
501 502 503
                    if( h->mb.b_trellis )
                        x264_quant_4x4_trellis( h, dct4x4[idx], CQM_4PY, i_qp, DCT_LUMA_4x4, 0 );
                    else
Loren Merritt's avatar
Loren Merritt committed
504
                        h->quantf.quant_4x4( dct4x4[idx], h->quant4_mf[CQM_4PY][i_qp], h->quant4_bias[CQM_4PY][i_qp] );
Loren Merritt's avatar
Loren Merritt committed
505

506
                    h->zigzagf.scan_4x4( h->dct.luma4x4[idx], dct4x4[idx] );
507 508

                    if( b_decimate && i_decimate_8x8 <= 6 )
509
                        i_decimate_8x8 += x264_mb_decimate_score( h->dct.luma4x4[idx], 16 );
Laurent Aimar's avatar
Laurent Aimar committed
510 511
                }

512 513
                /* decimate this 8x8 block */
                i_decimate_mb += i_decimate_8x8;
514
                if( i_decimate_8x8 < 4 && b_decimate )
515
                    nnz8x8[i8x8] = 0;
Laurent Aimar's avatar
Laurent Aimar committed
516
            }
517

518
            if( i_decimate_mb < 6 && b_decimate )
519
                *(uint32_t*)nnz8x8 = 0;
520
            else
521 522 523 524 525 526
            {
                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 );
527
                        h->dctf.add8x8_idct( &h->mb.pic.p_fdec[0][(i8x8&1)*8 + (i8x8>>1)*8*FDEC_STRIDE], &dct4x4[i8x8*4] );
528 529
                    }
            }
Laurent Aimar's avatar
Laurent Aimar committed
530 531 532 533 534 535 536
        }
    }

    /* encode chroma */
    if( IS_INTRA( h->mb.i_type ) )
    {
        const int i_mode = h->mb.i_chroma_pred_mode;
537 538
        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
539 540 541
    }

    /* encode the 8x8 blocks */
542
    x264_mb_encode_8x8_chroma( h, !IS_INTRA( h->mb.i_type ), h->mb.i_chroma_qp );
Laurent Aimar's avatar
Laurent Aimar committed
543

544
    /* coded block pattern and non_zero_count */
545
    h->mb.i_cbp_luma = 0x00;
Laurent Aimar's avatar
Laurent Aimar committed
546 547 548 549
    if( h->mb.i_type == I_16x16 )
    {
        for( i = 0; i < 16; i++ )
        {
550
            int nz = array_non_zero( h->dct.luma4x4[i] );
Laurent Aimar's avatar
Laurent Aimar committed
551
            h->mb.cache.non_zero_count[x264_scan8[i]] = nz;
552
            h->mb.i_cbp_luma |= nz;
Laurent Aimar's avatar
Laurent Aimar committed
553
        }
554
        h->mb.i_cbp_luma *= 0xf;
Laurent Aimar's avatar
Laurent Aimar committed
555 556 557
    }
    else
    {
558
        for( i = 0; i < 4; i++)
Laurent Aimar's avatar
Laurent Aimar committed
559
        {
560
            if(!nnz8x8[i])
561 562 563 564
            {
                *(uint16_t*)&h->mb.cache.non_zero_count[x264_scan8[0+i*4]] = 0;
                *(uint16_t*)&h->mb.cache.non_zero_count[x264_scan8[2+i*4]] = 0;
            }
565 566
            else if( h->mb.b_transform_8x8 )
            {
567 568 569
                *(uint16_t*)&h->mb.cache.non_zero_count[x264_scan8[0+4*i]] = nnz8x8[i] * 0x0101;
                *(uint16_t*)&h->mb.cache.non_zero_count[x264_scan8[2+4*i]] = nnz8x8[i] * 0x0101;
                h->mb.i_cbp_luma |= nnz8x8[i] << i;
570 571 572
            }
            else
            {
573
                int nz, cbp = 0;
574 575
                for( j = 0; j < 4; j++ )
                {
576 577 578
                    nz = array_non_zero( h->dct.luma4x4[j+4*i] );
                    h->mb.cache.non_zero_count[x264_scan8[j+4*i]] = nz;
                    cbp |= nz;
579
                }
580
                h->mb.i_cbp_luma |= cbp << i;
581
            }
Laurent Aimar's avatar
Laurent Aimar committed
582 583 584 585 586
        }
    }

    if( h->param.b_cabac )
    {
587 588 589
        i_cbp_dc = ( h->mb.i_type == I_16x16 && array_non_zero( h->dct.luma16x16_dc ) )
                 | array_non_zero( h->dct.chroma_dc[0] ) << 1
                 | array_non_zero( h->dct.chroma_dc[1] ) << 2;
Laurent Aimar's avatar
Laurent Aimar committed
590 591 592 593 594 595 596 597
    }

    /* 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)*/
598
    if( !b_force_no_skip )
Laurent Aimar's avatar
Laurent Aimar committed
599
    {
600
        if( h->mb.i_type == P_L0 && h->mb.i_partition == D_16x16 &&
601 602 603
            !(h->mb.i_cbp_luma | h->mb.i_cbp_chroma) && 
            *(uint32_t*)h->mb.cache.mv[0][x264_scan8[0]] == *(uint32_t*)h->mb.cache.pskip_mv
            && h->mb.cache.ref[0][x264_scan8[0]] == 0 )
604
        {
605
            h->mb.i_type = P_SKIP;
Laurent Aimar's avatar
Laurent Aimar committed
606
        }
607

608
        /* Check for B_SKIP */
609
        if( h->mb.i_type == B_DIRECT && !(h->mb.i_cbp_luma | h->mb.i_cbp_chroma) )
610 611 612
        {
            h->mb.i_type = B_SKIP;
        }
613
    }
Laurent Aimar's avatar
Laurent Aimar committed
614 615 616
}

/*****************************************************************************
617 618
 * 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
619 620
 *  the previous QP
 *****************************************************************************/
621
int x264_macroblock_probe_skip( x264_t *h, const int b_bidir )
Laurent Aimar's avatar
Laurent Aimar committed
622
{
Fiona Glaser's avatar
Fiona Glaser committed
623
    DECLARE_ALIGNED_16( int16_t dct4x4[4][4][4] );
Loren Merritt's avatar
Loren Merritt committed
624 625
    DECLARE_ALIGNED_16( int16_t dct2x2[2][2] );
    DECLARE_ALIGNED_16( int16_t dctscan[16] );
Laurent Aimar's avatar
Laurent Aimar committed
626

627
    int i_qp = h->mb.i_qp;
Laurent Aimar's avatar
Laurent Aimar committed
628
    int mvp[2];
629
    int ch, thresh;
Laurent Aimar's avatar
Laurent Aimar committed
630 631 632 633

    int i8x8, i4x4;
    int i_decimate_mb;

634 635 636
    if( !b_bidir )
    {
        /* Get the MV */
637 638
        mvp[0] = x264_clip3( h->mb.cache.pskip_mv[0], h->mb.mv_min[0], h->mb.mv_max[0] );
        mvp[1] = x264_clip3( h->mb.cache.pskip_mv[1], h->mb.mv_min[1], h->mb.mv_max[1] );
Laurent Aimar's avatar
Laurent Aimar committed
639

640
        /* Motion compensation */
641 642
        h->mc.mc_luma( h->mb.pic.p_fdec[0],    FDEC_STRIDE,
                       h->mb.pic.p_fref[0][0], h->mb.pic.i_stride[0],
643
                       mvp[0], mvp[1], 16, 16 );
644
    }
Laurent Aimar's avatar
Laurent Aimar committed
645 646 647

    for( i8x8 = 0, i_decimate_mb = 0; i8x8 < 4; i8x8++ )
    {
Fiona Glaser's avatar
Fiona Glaser committed
648 649 650 651 652
        int fenc_offset = (i8x8&1) * 8 + (i8x8>>1) * FENC_STRIDE * 8;
        int fdec_offset = (i8x8&1) * 8 + (i8x8>>1) * FDEC_STRIDE * 8;
        /* get luma diff */
        h->dctf.sub8x8_dct( dct4x4, h->mb.pic.p_fenc[0] + fenc_offset,
                                    h->mb.pic.p_fdec[0] + fdec_offset );
Laurent Aimar's avatar
Laurent Aimar committed
653 654 655
        /* encode one 4x4 block */
        for( i4x4 = 0; i4x4 < 4; i4x4++ )
        {
Fiona Glaser's avatar
Fiona Glaser committed
656 657 658 659
            h->quantf.quant_4x4( dct4x4[i4x4], h->quant4_mf[CQM_4PY][i_qp], h->quant4_bias[CQM_4PY][i_qp] );
            if( !array_non_zero(dct4x4[i4x4]) )
                continue;
            h->zigzagf.scan_4x4( dctscan, dct4x4[i4x4] );
Laurent Aimar's avatar
Laurent Aimar committed
660 661 662 663 664 665 666
            i_decimate_mb += x264_mb_decimate_score( dctscan, 16 );
            if( i_decimate_mb >= 6 )
                return 0;
        }
    }

    /* encode chroma */
667
    i_qp = h->mb.i_chroma_qp;
668
    thresh = (x264_lambda2_tab[i_qp] + 32) >> 6;
Laurent Aimar's avatar
Laurent Aimar committed
669 670 671 672 673 674

    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];

675 676
        if( !b_bidir )
        {
677 678
            h->mc.mc_chroma( h->mb.pic.p_fdec[1+ch],       FDEC_STRIDE,
                             h->mb.pic.p_fref[0][0][4+ch], h->mb.pic.i_stride[1+ch],
679
                             mvp[0], mvp[1], 8, 8 );
680
        }
Laurent Aimar's avatar
Laurent Aimar committed
681

682 683 684 685 686
        /* there is almost never a termination during chroma, but we can't avoid the check entirely */
        /* so instead we check SSD and skip the actual check if the score is low enough. */
        if( h->pixf.ssd[PIXEL_8x8]( p_dst, FDEC_STRIDE, p_src, FENC_STRIDE ) < thresh )
            continue;

687
        h->dctf.sub8x8_dct( dct4x4, p_src, p_dst );
Laurent Aimar's avatar
Laurent Aimar committed
688 689 690 691 692 693 694

        /* 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 );
Loren Merritt's avatar
Loren Merritt committed
695
        h->quantf.quant_2x2_dc( dct2x2, h->quant4_mf[CQM_4PC][i_qp][0]>>1, h->quant4_bias[CQM_4PC][i_qp][0]<<1 );
Fiona Glaser's avatar
Fiona Glaser committed
696
        if( array_non_zero(dct2x2) )
Laurent Aimar's avatar
Laurent Aimar committed
697 698 699 700 701
            return 0;

        /* calculate dct coeffs */
        for( i4x4 = 0, i_decimate_mb = 0; i4x4 < 4; i4x4++ )
        {
Loren Merritt's avatar
Loren Merritt committed
702
            h->quantf.quant_4x4( dct4x4[i4x4], h->quant4_mf[CQM_4PC][i_qp], h->quant4_bias[CQM_4PC][i_qp] );
Fiona Glaser's avatar
Fiona Glaser committed
703 704
            if( !array_non_zero(dct4x4[i4x4]) )
                continue;
705 706
            h->zigzagf.scan_4x4( dctscan, dct4x4[i4x4] );
            i_decimate_mb += x264_mb_decimate_score( dctscan+1, 15 );
Laurent Aimar's avatar
Laurent Aimar committed
707 708 709 710 711
            if( i_decimate_mb >= 7 )
                return 0;
        }
    }

712
    h->mb.b_skip_mc = 1;
Laurent Aimar's avatar
Laurent Aimar committed
713 714
    return 1;
}
715 716 717 718 719 720 721 722 723

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

void x264_noise_reduction_update( x264_t *h )
{
    int cat, i;
724
    for( cat = 0; cat < 2; cat++ )
725
    {
726
        int size = cat ? 64 : 16;
727
        const uint16_t *weight = cat ? x264_dct8_weight2_tab : x264_dct4_weight2_tab;
728

729
        if( h->nr_count[cat] > (cat ? (1<<16) : (1<<18)) )
730 731 732 733 734 735 736 737 738 739 740 741 742 743
        {
            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);
    }
}

744 745 746 747 748 749 750 751 752
/*****************************************************************************
 * 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;
753 754
    int b_decimate = h->sh.i_type == SLICE_TYPE_B || h->param.analyse.b_dct_decimate;
    int nnz8x8;
755 756 757 758 759 760
    int ch;

    x264_mb_mc_8x8( h, i8 );

    if( h->mb.b_transform_8x8 )
    {
Loren Merritt's avatar
Loren Merritt committed
761
        DECLARE_ALIGNED_16( int16_t dct8x8[8][8] );
762
        h->dctf.sub8x8_dct8( dct8x8, p_fenc, p_fdec );
Loren Merritt's avatar
Loren Merritt committed
763
        h->quantf.quant_8x8( dct8x8, h->quant8_mf[CQM_8PY][i_qp], h->quant8_bias[CQM_8PY][i_qp] );
764
        h->zigzagf.scan_8x8( h->dct.luma8x8[i8], dct8x8 );
765

766 767 768
        if( b_decimate )
            nnz8x8 = 4 <= x264_mb_decimate_score( h->dct.luma8x8[i8], 64 );
        else
769
            nnz8x8 = array_non_zero( dct8x8 );
770

771 772 773
        if( nnz8x8 )
        {
            h->quantf.dequant_8x8( dct8x8, h->dequant8_mf[CQM_8PY], i_qp );
774
            h->dctf.add8x8_idct8( p_fdec, dct8x8 );
775 776 777 778
        }
    }
    else
    {
779
        int i4;
Loren Merritt's avatar
Loren Merritt committed
780
        DECLARE_ALIGNED_16( int16_t dct4x4[4][4][4] );
781
        h->dctf.sub8x8_dct( dct4x4, p_fenc, p_fdec );
782 783
        for( i4 = 0; i4 < 4; i4++ )
            h->quantf.quant_4x4( dct4x4[i4], h->quant4_mf[CQM_4PY][i_qp], h->quant4_bias[CQM_4PY][i_qp] );
784
        for( i4 = 0; i4 < 4; i4++ )
785
            h->zigzagf.scan_4x4( h->dct.luma4x4[i8*4+i4], dct4x4[i4] );
786

787
        if( b_decimate )
788
        {
789 790
            int i_decimate_8x8 = 0;
            for( i4 = 0; i4 < 4 && i_decimate_8x8 < 4; i4++ )
791
                i_decimate_8x8 += x264_mb_decimate_score( h->dct.luma4x4[i8*4+i4], 16 );
792
            nnz8x8 = 4 <= i_decimate_8x8;
793
        }
794
        else
795
            nnz8x8 = array_non_zero( dct4x4 );
796

797 798 799
        if( nnz8x8 )
        {
            for( i4 = 0; i4 < 4; i4++ )
800
                h->quantf.dequant_4x4( dct4x4[i4], h->dequant4_mf[CQM_4PY], i_qp );
801
            h->dctf.add8x8_idct( p_fdec, dct4x4 );
802 803 804
        }
    }

805
    i_qp = h->mb.i_chroma_qp;
806 807 808

    for( ch = 0; ch < 2; ch++ )
    {
Loren Merritt's avatar
Loren Merritt committed
809
        DECLARE_ALIGNED_16( int16_t dct4x4[4][4] );
810 811 812
        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;

813
        h->dctf.sub4x4_dct( dct4x4, p_fenc, p_fdec );
Loren Merritt's avatar
Loren Merritt committed
814
        h->quantf.quant_4x4( dct4x4, h->quant4_mf[CQM_4PC][i_qp], h->quant4_bias[CQM_4PC][i_qp] );
815
        h->zigzagf.scan_4x4( h->dct.luma4x4[16+i8+ch*4], dct4x4 );
816
        if( array_non_zero( dct4x4 ) )
817 818 819 820
        {
            h->quantf.dequant_4x4( dct4x4, h->dequant4_mf[CQM_4PC], i_qp );
            h->dctf.add4x4_idct( p_fdec, dct4x4 );
        }