mvpred.c 22 KB
Newer Older
1
/*****************************************************************************
2
 * mvpred.c: motion vector prediction
3
 *****************************************************************************
Henrik Gramner's avatar
Henrik Gramner committed
4
 * Copyright (C) 2003-2019 x264 project
5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22
 *
 * Authors: Loren Merritt <lorenm@u.washington.edu>
 *          Fiona Glaser <fiona@x264.com>
 *          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., 51 Franklin Street, Fifth Floor, Boston, MA  02111, USA.
23 24 25
 *
 * This program is also available under a commercial proprietary license.
 * For more information, contact us at licensing@x264.com.
26 27 28 29 30 31 32 33 34 35 36 37 38 39 40
 *****************************************************************************/

#include "common.h"

void x264_mb_predict_mv( x264_t *h, int i_list, int idx, int i_width, int16_t mvp[2] )
{
    const int i8 = x264_scan8[idx];
    const int i_ref= h->mb.cache.ref[i_list][i8];
    int     i_refa = h->mb.cache.ref[i_list][i8 - 1];
    int16_t *mv_a  = h->mb.cache.mv[i_list][i8 - 1];
    int     i_refb = h->mb.cache.ref[i_list][i8 - 8];
    int16_t *mv_b  = h->mb.cache.mv[i_list][i8 - 8];
    int     i_refc = h->mb.cache.ref[i_list][i8 - 8 + i_width];
    int16_t *mv_c  = h->mb.cache.mv[i_list][i8 - 8 + i_width];

Simon Horlick's avatar
Simon Horlick committed
41
    // Partitions not yet reached in scan order are unavailable.
42 43 44 45 46
    if( (idx&3) >= 2 + (i_width&1) || i_refc == -2 )
    {
        i_refc = h->mb.cache.ref[i_list][i8 - 8 - 1];
        mv_c   = h->mb.cache.mv[i_list][i8 - 8 - 1];

47
        if( SLICE_MBAFF
Simon Horlick's avatar
Simon Horlick committed
48
            && h->mb.cache.ref[i_list][x264_scan8[0]-1] != -2
49
            && MB_INTERLACED != h->mb.field[h->mb.i_mb_left_xy[0]] )
Simon Horlick's avatar
Simon Horlick committed
50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67
        {
            if( idx == 2 )
            {
                mv_c = h->mb.cache.topright_mv[i_list][0];
                i_refc = h->mb.cache.topright_ref[i_list][0];
            }
            else if( idx == 8 )
            {
                mv_c = h->mb.cache.topright_mv[i_list][1];
                i_refc = h->mb.cache.topright_ref[i_list][1];
            }
            else if( idx == 10 )
            {
                mv_c = h->mb.cache.topright_mv[i_list][2];
                i_refc = h->mb.cache.topright_ref[i_list][2];
            }
        }
    }
68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182
    if( h->mb.i_partition == D_16x8 )
    {
        if( idx == 0 )
        {
            if( i_refb == i_ref )
            {
                CP32( mvp, mv_b );
                return;
            }
        }
        else
        {
            if( i_refa == i_ref )
            {
                CP32( mvp, mv_a );
                return;
            }
        }
    }
    else if( h->mb.i_partition == D_8x16 )
    {
        if( idx == 0 )
        {
            if( i_refa == i_ref )
            {
                CP32( mvp, mv_a );
                return;
            }
        }
        else
        {
            if( i_refc == i_ref )
            {
                CP32( mvp, mv_c );
                return;
            }
        }
    }

    int i_count = (i_refa == i_ref) + (i_refb == i_ref) + (i_refc == i_ref);

    if( i_count > 1 )
    {
median:
        x264_median_mv( mvp, mv_a, mv_b, mv_c );
    }
    else if( i_count == 1 )
    {
        if( i_refa == i_ref )
            CP32( mvp, mv_a );
        else if( i_refb == i_ref )
            CP32( mvp, mv_b );
        else
            CP32( mvp, mv_c );
    }
    else if( i_refb == -2 && i_refc == -2 && i_refa != -2 )
        CP32( mvp, mv_a );
    else
        goto median;
}

void x264_mb_predict_mv_16x16( x264_t *h, int i_list, int i_ref, int16_t mvp[2] )
{
    int     i_refa = h->mb.cache.ref[i_list][X264_SCAN8_0 - 1];
    int16_t *mv_a  = h->mb.cache.mv[i_list][X264_SCAN8_0 - 1];
    int     i_refb = h->mb.cache.ref[i_list][X264_SCAN8_0 - 8];
    int16_t *mv_b  = h->mb.cache.mv[i_list][X264_SCAN8_0 - 8];
    int     i_refc = h->mb.cache.ref[i_list][X264_SCAN8_0 - 8 + 4];
    int16_t *mv_c  = h->mb.cache.mv[i_list][X264_SCAN8_0 - 8 + 4];
    if( i_refc == -2 )
    {
        i_refc = h->mb.cache.ref[i_list][X264_SCAN8_0 - 8 - 1];
        mv_c   = h->mb.cache.mv[i_list][X264_SCAN8_0 - 8 - 1];
    }

    int i_count = (i_refa == i_ref) + (i_refb == i_ref) + (i_refc == i_ref);

    if( i_count > 1 )
    {
median:
        x264_median_mv( mvp, mv_a, mv_b, mv_c );
    }
    else if( i_count == 1 )
    {
        if( i_refa == i_ref )
            CP32( mvp, mv_a );
        else if( i_refb == i_ref )
            CP32( mvp, mv_b );
        else
            CP32( mvp, mv_c );
    }
    else if( i_refb == -2 && i_refc == -2 && i_refa != -2 )
        CP32( mvp, mv_a );
    else
        goto median;
}


void x264_mb_predict_mv_pskip( x264_t *h, int16_t mv[2] )
{
    int     i_refa = h->mb.cache.ref[0][X264_SCAN8_0 - 1];
    int     i_refb = h->mb.cache.ref[0][X264_SCAN8_0 - 8];
    int16_t *mv_a  = h->mb.cache.mv[0][X264_SCAN8_0 - 1];
    int16_t *mv_b  = h->mb.cache.mv[0][X264_SCAN8_0 - 8];

    if( i_refa == -2 || i_refb == -2 ||
        !( i_refa | M32( mv_a ) ) ||
        !( i_refb | M32( mv_b ) ) )
    {
        M32( mv ) = 0;
    }
    else
        x264_mb_predict_mv_16x16( h, 0, 0, mv );
}

183
static int mb_predict_mv_direct16x16_temporal( x264_t *h )
184
{
Simon Horlick's avatar
Simon Horlick committed
185 186 187 188 189
    int mb_x = h->mb.i_mb_x;
    int mb_y = h->mb.i_mb_y;
    int mb_xy = h->mb.i_mb_xy;
    int type_col[2] = { h->fref[1][0]->mb_type[mb_xy], h->fref[1][0]->mb_type[mb_xy] };
    int partition_col[2] = { h->fref[1][0]->mb_partition[mb_xy], h->fref[1][0]->mb_partition[mb_xy] };
190 191
    int preshift = MB_INTERLACED;
    int postshift = MB_INTERLACED;
Simon Horlick's avatar
Simon Horlick committed
192 193 194
    int offset = 1;
    int yshift = 1;
    h->mb.i_partition = partition_col[0];
195
    if( PARAM_INTERLACED && h->fref[1][0]->field[mb_xy] != MB_INTERLACED )
196
    {
197
        if( MB_INTERLACED )
Simon Horlick's avatar
Simon Horlick committed
198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216
        {
            mb_y = h->mb.i_mb_y&~1;
            mb_xy = mb_x + h->mb.i_mb_stride * mb_y;
            type_col[0] = h->fref[1][0]->mb_type[mb_xy];
            type_col[1] = h->fref[1][0]->mb_type[mb_xy + h->mb.i_mb_stride];
            partition_col[0] = h->fref[1][0]->mb_partition[mb_xy];
            partition_col[1] = h->fref[1][0]->mb_partition[mb_xy + h->mb.i_mb_stride];
            preshift = 0;
            yshift = 0;

            if( (IS_INTRA(type_col[0]) || partition_col[0] == D_16x16) &&
                (IS_INTRA(type_col[1]) || partition_col[1] == D_16x16) &&
                partition_col[0] != D_8x8 )
                h->mb.i_partition = D_16x8;
            else
                h->mb.i_partition = D_8x8;
        }
        else
        {
217
            int cur_poc = h->fdec->i_poc + h->fdec->i_delta_poc[MB_INTERLACED&h->mb.i_mb_y&1];
Simon Horlick's avatar
Simon Horlick committed
218 219 220 221 222 223 224 225 226 227 228
            int col_parity = abs(h->fref[1][0]->i_poc + h->fref[1][0]->i_delta_poc[0] - cur_poc)
                          >= abs(h->fref[1][0]->i_poc + h->fref[1][0]->i_delta_poc[1] - cur_poc);
            mb_y = (h->mb.i_mb_y&~1) + col_parity;
            mb_xy = mb_x + h->mb.i_mb_stride * mb_y;
            type_col[0] = type_col[1] = h->fref[1][0]->mb_type[mb_xy];
            partition_col[0] = partition_col[1] = h->fref[1][0]->mb_partition[mb_xy];
            preshift = 1;
            yshift = 2;
            h->mb.i_partition = partition_col[0];
        }
        offset = 0;
229
    }
Simon Horlick's avatar
Simon Horlick committed
230 231 232 233
    int i_mb_4x4 = 16 * h->mb.i_mb_stride * mb_y + 4 * mb_x;
    int i_mb_8x8 =  4 * h->mb.i_mb_stride * mb_y + 2 * mb_x;

    x264_macroblock_cache_ref( h, 0, 0, 4, 4, 1, 0 );
234 235 236 237

    /* Don't do any checks other than the ones we have to, based
     * on the size of the colocated partitions.
     * Depends on the enum order: D_8x8, D_16x8, D_8x16, D_16x16 */
Simon Horlick's avatar
Simon Horlick committed
238 239 240 241
    int max_i8 = (D_16x16 - h->mb.i_partition) + 1;
    int step = (h->mb.i_partition == D_16x8) + 1;
    int width = 4 >> ((D_16x16 - h->mb.i_partition)&1);
    int height = 4 >> ((D_16x16 - h->mb.i_partition)>>1);
242 243 244 245
    for( int i8 = 0; i8 < max_i8; i8 += step )
    {
        int x8 = i8&1;
        int y8 = i8>>1;
246 247
        int ypart = (SLICE_MBAFF && h->fref[1][0]->field[mb_xy] != MB_INTERLACED) ?
                    MB_INTERLACED ? y8*6 : 2*(h->mb.i_mb_y&1) + y8 :
Simon Horlick's avatar
Simon Horlick committed
248 249 250 251 252 253 254 255 256 257 258
                    3*y8;

        if( IS_INTRA( type_col[y8] ) )
        {
            x264_macroblock_cache_ref( h, 2*x8, 2*y8, width, height, 0, 0 );
            x264_macroblock_cache_mv(  h, 2*x8, 2*y8, width, height, 0, 0 );
            x264_macroblock_cache_mv(  h, 2*x8, 2*y8, width, height, 1, 0 );
            continue;
        }

        int i_part_8x8 = i_mb_8x8 + x8 + (ypart>>1) * h->mb.i_b8_stride;
259
        int i_ref1_ref = h->fref[1][0]->ref[0][i_part_8x8];
260
        int i_ref = (map_col_to_list0(i_ref1_ref>>preshift) << postshift) + (offset&i_ref1_ref&MB_INTERLACED);
261 262 263 264

        if( i_ref >= 0 )
        {
            int dist_scale_factor = h->mb.dist_scale_factor[i_ref][0];
Simon Horlick's avatar
Simon Horlick committed
265 266
            int16_t *mv_col = h->fref[1][0]->mv[0][i_mb_4x4 + 3*x8 + ypart * h->mb.i_b4_stride];
            int16_t mv_y = (mv_col[1]<<yshift)/2;
267
            int l0x = ( dist_scale_factor * mv_col[0] + 128 ) >> 8;
Simon Horlick's avatar
Simon Horlick committed
268 269
            int l0y = ( dist_scale_factor * mv_y + 128 ) >> 8;
            if( h->param.i_threads > 1 && (l0y > h->mb.mv_max_spel[1] || l0y-mv_y > h->mb.mv_max_spel[1]) )
270 271 272
                return 0;
            x264_macroblock_cache_ref( h, 2*x8, 2*y8, width, height, 0, i_ref );
            x264_macroblock_cache_mv( h, 2*x8, 2*y8, width, height, 0, pack16to32_mask(l0x, l0y) );
Simon Horlick's avatar
Simon Horlick committed
273
            x264_macroblock_cache_mv( h, 2*x8, 2*y8, width, height, 1, pack16to32_mask(l0x-mv_col[0], l0y-mv_y) );
274 275 276 277 278 279 280 281 282 283 284 285 286 287 288
        }
        else
        {
            /* the collocated ref isn't in the current list0 */
            /* FIXME: we might still be able to use direct_8x8 on some partitions */
            /* FIXME: with B-pyramid + extensive ref list reordering
             *   (not currently used), we would also have to check
             *   l1mv1 like in spatial mode */
            return 0;
        }
    }

    return 1;
}

289
static ALWAYS_INLINE int mb_predict_mv_direct16x16_spatial( x264_t *h, int b_interlaced )
290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336
{
    int8_t ref[2];
    ALIGNED_ARRAY_8( int16_t, mv,[2],[2] );
    for( int i_list = 0; i_list < 2; i_list++ )
    {
        int     i_refa = h->mb.cache.ref[i_list][X264_SCAN8_0 - 1];
        int16_t *mv_a  = h->mb.cache.mv[i_list][X264_SCAN8_0 - 1];
        int     i_refb = h->mb.cache.ref[i_list][X264_SCAN8_0 - 8];
        int16_t *mv_b  = h->mb.cache.mv[i_list][X264_SCAN8_0 - 8];
        int     i_refc = h->mb.cache.ref[i_list][X264_SCAN8_0 - 8 + 4];
        int16_t *mv_c  = h->mb.cache.mv[i_list][X264_SCAN8_0 - 8 + 4];
        if( i_refc == -2 )
        {
            i_refc = h->mb.cache.ref[i_list][X264_SCAN8_0 - 8 - 1];
            mv_c   = h->mb.cache.mv[i_list][X264_SCAN8_0 - 8 - 1];
        }

        int i_ref = X264_MIN3( (unsigned)i_refa, (unsigned)i_refb, (unsigned)i_refc );
        if( i_ref < 0 )
        {
            i_ref = -1;
            M32( mv[i_list] ) = 0;
        }
        else
        {
            /* Same as x264_mb_predict_mv_16x16, but simplified to eliminate cases
             * not relevant to spatial direct. */
            int i_count = (i_refa == i_ref) + (i_refb == i_ref) + (i_refc == i_ref);

            if( i_count > 1 )
                x264_median_mv( mv[i_list], mv_a, mv_b, mv_c );
            else
            {
                if( i_refa == i_ref )
                    CP32( mv[i_list], mv_a );
                else if( i_refb == i_ref )
                    CP32( mv[i_list], mv_b );
                else
                    CP32( mv[i_list], mv_c );
            }
        }

        x264_macroblock_cache_ref( h, 0, 0, 4, 4, i_list, i_ref );
        x264_macroblock_cache_mv_ptr( h, 0, 0, 4, 4, i_list, mv[i_list] );
        ref[i_list] = i_ref;
    }

Simon Horlick's avatar
Simon Horlick committed
337 338 339 340 341 342
    int mb_x = h->mb.i_mb_x;
    int mb_y = h->mb.i_mb_y;
    int mb_xy = h->mb.i_mb_xy;
    int type_col[2] = { h->fref[1][0]->mb_type[mb_xy], h->fref[1][0]->mb_type[mb_xy] };
    int partition_col[2] = { h->fref[1][0]->mb_partition[mb_xy], h->fref[1][0]->mb_partition[mb_xy] };
    h->mb.i_partition = partition_col[0];
343
    if( b_interlaced && h->fref[1][0]->field[mb_xy] != MB_INTERLACED )
Simon Horlick's avatar
Simon Horlick committed
344
    {
345
        if( MB_INTERLACED )
Simon Horlick's avatar
Simon Horlick committed
346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362
        {
            mb_y = h->mb.i_mb_y&~1;
            mb_xy = mb_x + h->mb.i_mb_stride * mb_y;
            type_col[0] = h->fref[1][0]->mb_type[mb_xy];
            type_col[1] = h->fref[1][0]->mb_type[mb_xy + h->mb.i_mb_stride];
            partition_col[0] = h->fref[1][0]->mb_partition[mb_xy];
            partition_col[1] = h->fref[1][0]->mb_partition[mb_xy + h->mb.i_mb_stride];

            if( (IS_INTRA(type_col[0]) || partition_col[0] == D_16x16) &&
                (IS_INTRA(type_col[1]) || partition_col[1] == D_16x16) &&
                partition_col[0] != D_8x8 )
                h->mb.i_partition = D_16x8;
            else
                h->mb.i_partition = D_8x8;
        }
        else
        {
363
            int cur_poc = h->fdec->i_poc + h->fdec->i_delta_poc[MB_INTERLACED&h->mb.i_mb_y&1];
Simon Horlick's avatar
Simon Horlick committed
364 365 366 367 368 369 370 371 372
            int col_parity = abs(h->fref[1][0]->i_poc + h->fref[1][0]->i_delta_poc[0] - cur_poc)
                          >= abs(h->fref[1][0]->i_poc + h->fref[1][0]->i_delta_poc[1] - cur_poc);
            mb_y = (h->mb.i_mb_y&~1) + col_parity;
            mb_xy = mb_x + h->mb.i_mb_stride * mb_y;
            type_col[0] = type_col[1] = h->fref[1][0]->mb_type[mb_xy];
            partition_col[0] = partition_col[1] = h->fref[1][0]->mb_partition[mb_xy];
            h->mb.i_partition = partition_col[0];
        }
    }
373 374
    int i_mb_4x4 = b_interlaced ? 4 * (h->mb.i_b4_stride*mb_y + mb_x) : h->mb.i_b4_xy;
    int i_mb_8x8 = b_interlaced ? 2 * (h->mb.i_b8_stride*mb_y + mb_x) : h->mb.i_b8_xy;
Simon Horlick's avatar
Simon Horlick committed
375 376 377 378 379 380

    int8_t *l1ref0 = &h->fref[1][0]->ref[0][i_mb_8x8];
    int8_t *l1ref1 = &h->fref[1][0]->ref[1][i_mb_8x8];
    int16_t (*l1mv[2])[2] = { (int16_t (*)[2]) &h->fref[1][0]->mv[0][i_mb_4x4],
                              (int16_t (*)[2]) &h->fref[1][0]->mv[1][i_mb_4x4] };

381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399
    if( (M16( ref ) & 0x8080) == 0x8080 ) /* if( ref[0] < 0 && ref[1] < 0 ) */
    {
        x264_macroblock_cache_ref( h, 0, 0, 4, 4, 0, 0 );
        x264_macroblock_cache_ref( h, 0, 0, 4, 4, 1, 0 );
        return 1;
    }

    if( h->param.i_threads > 1
        && ( mv[0][1] > h->mb.mv_max_spel[1]
          || mv[1][1] > h->mb.mv_max_spel[1] ) )
    {
#if 0
        fprintf(stderr, "direct_spatial: (%d,%d) (%d,%d) > %d \n",
                mv[0][0], mv[0][1], mv[1][0], mv[1][1],
                h->mb.mv_max_spel[1]);
#endif
        return 0;
    }

400
    if( !M64( mv ) || (!b_interlaced && IS_INTRA( type_col[0] )) || (ref[0]&&ref[1]) )
401 402 403 404 405
        return 1;

    /* Don't do any checks other than the ones we have to, based
     * on the size of the colocated partitions.
     * Depends on the enum order: D_8x8, D_16x8, D_8x16, D_16x16 */
Simon Horlick's avatar
Simon Horlick committed
406 407 408 409
    int max_i8 = (D_16x16 - h->mb.i_partition) + 1;
    int step = (h->mb.i_partition == D_16x8) + 1;
    int width = 4 >> ((D_16x16 - h->mb.i_partition)&1);
    int height = 4 >> ((D_16x16 - h->mb.i_partition)>>1);
410 411 412 413 414 415

    /* col_zero_flag */
    for( int i8 = 0; i8 < max_i8; i8 += step )
    {
        const int x8 = i8&1;
        const int y8 = i8>>1;
416
        int ypart = (b_interlaced && h->fref[1][0]->field[mb_xy] != MB_INTERLACED) ?
417
                    MB_INTERLACED ? y8*6 : 2*(h->mb.i_mb_y&1) + y8 :
Simon Horlick's avatar
Simon Horlick committed
418 419 420 421
                    3*y8;
        int o8 = x8 + (ypart>>1) * h->mb.i_b8_stride;
        int o4 = 3*x8 + ypart * h->mb.i_b4_stride;

422
        if( b_interlaced && IS_INTRA( type_col[y8] ) )
Simon Horlick's avatar
Simon Horlick committed
423 424
            continue;

425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442
        int idx;
        if( l1ref0[o8] == 0 )
            idx = 0;
        else if( l1ref0[o8] < 0 && l1ref1[o8] == 0 )
            idx = 1;
        else
            continue;

        if( abs( l1mv[idx][o4][0] ) <= 1 && abs( l1mv[idx][o4][1] ) <= 1 )
        {
            if( ref[0] == 0 ) x264_macroblock_cache_mv( h, 2*x8, 2*y8, width, height, 0, 0 );
            if( ref[1] == 0 ) x264_macroblock_cache_mv( h, 2*x8, 2*y8, width, height, 1, 0 );
        }
    }

    return 1;
}

443

444
static int mb_predict_mv_direct16x16_spatial_interlaced( x264_t *h )
445
{
446
    return mb_predict_mv_direct16x16_spatial( h, 1 );
447 448
}

449
static int mb_predict_mv_direct16x16_spatial_progressive( x264_t *h )
450
{
451
    return mb_predict_mv_direct16x16_spatial( h, 0 );
452 453
}

454 455 456 457 458 459
int x264_mb_predict_mv_direct16x16( x264_t *h, int *b_changed )
{
    int b_available;
    if( h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_NONE )
        return 0;
    else if( h->sh.b_direct_spatial_mv_pred )
460 461
    {
        if( SLICE_MBAFF )
462
            b_available = mb_predict_mv_direct16x16_spatial_interlaced( h );
463
        else
464
            b_available = mb_predict_mv_direct16x16_spatial_progressive( h );
465
    }
466
    else
467
        b_available = mb_predict_mv_direct16x16_temporal( h );
468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523

    if( b_changed != NULL && b_available )
    {
        int changed;

        changed  = M32( h->mb.cache.direct_mv[0][0] ) ^ M32( h->mb.cache.mv[0][x264_scan8[0]] );
        changed |= M32( h->mb.cache.direct_mv[1][0] ) ^ M32( h->mb.cache.mv[1][x264_scan8[0]] );
        changed |= h->mb.cache.direct_ref[0][0] ^ h->mb.cache.ref[0][x264_scan8[0]];
        changed |= h->mb.cache.direct_ref[1][0] ^ h->mb.cache.ref[1][x264_scan8[0]];
        if( !changed && h->mb.i_partition != D_16x16 )
        {
            changed |= M32( h->mb.cache.direct_mv[0][3] ) ^ M32( h->mb.cache.mv[0][x264_scan8[12]] );
            changed |= M32( h->mb.cache.direct_mv[1][3] ) ^ M32( h->mb.cache.mv[1][x264_scan8[12]] );
            changed |= h->mb.cache.direct_ref[0][3] ^ h->mb.cache.ref[0][x264_scan8[12]];
            changed |= h->mb.cache.direct_ref[1][3] ^ h->mb.cache.ref[1][x264_scan8[12]];
        }
        if( !changed && h->mb.i_partition == D_8x8 )
        {
            changed |= M32( h->mb.cache.direct_mv[0][1] ) ^ M32( h->mb.cache.mv[0][x264_scan8[4]] );
            changed |= M32( h->mb.cache.direct_mv[1][1] ) ^ M32( h->mb.cache.mv[1][x264_scan8[4]] );
            changed |= M32( h->mb.cache.direct_mv[0][2] ) ^ M32( h->mb.cache.mv[0][x264_scan8[8]] );
            changed |= M32( h->mb.cache.direct_mv[1][2] ) ^ M32( h->mb.cache.mv[1][x264_scan8[8]] );
            changed |= h->mb.cache.direct_ref[0][1] ^ h->mb.cache.ref[0][x264_scan8[4]];
            changed |= h->mb.cache.direct_ref[1][1] ^ h->mb.cache.ref[1][x264_scan8[4]];
            changed |= h->mb.cache.direct_ref[0][2] ^ h->mb.cache.ref[0][x264_scan8[8]];
            changed |= h->mb.cache.direct_ref[1][2] ^ h->mb.cache.ref[1][x264_scan8[8]];
        }
        *b_changed = changed;
        if( !changed )
            return b_available;
    }

    /* cache ref & mv */
    if( b_available )
        for( int l = 0; l < 2; l++ )
        {
            CP32( h->mb.cache.direct_mv[l][0], h->mb.cache.mv[l][x264_scan8[ 0]] );
            CP32( h->mb.cache.direct_mv[l][1], h->mb.cache.mv[l][x264_scan8[ 4]] );
            CP32( h->mb.cache.direct_mv[l][2], h->mb.cache.mv[l][x264_scan8[ 8]] );
            CP32( h->mb.cache.direct_mv[l][3], h->mb.cache.mv[l][x264_scan8[12]] );
            h->mb.cache.direct_ref[l][0] = h->mb.cache.ref[l][x264_scan8[ 0]];
            h->mb.cache.direct_ref[l][1] = h->mb.cache.ref[l][x264_scan8[ 4]];
            h->mb.cache.direct_ref[l][2] = h->mb.cache.ref[l][x264_scan8[ 8]];
            h->mb.cache.direct_ref[l][3] = h->mb.cache.ref[l][x264_scan8[12]];
            h->mb.cache.direct_partition = h->mb.i_partition;
        }

    return b_available;
}

/* This just improves encoder performance, it's not part of the spec */
void x264_mb_predict_mv_ref16x16( x264_t *h, int i_list, int i_ref, int16_t mvc[9][2], int *i_mvc )
{
    int16_t (*mvr)[2] = h->mb.mvr[i_list][i_ref];
    int i = 0;

Anton Mitrofanov's avatar
Anton Mitrofanov committed
524
#define SET_MVP(mvp) \
525 526 527 528 529
    { \
        CP32( mvc[i], mvp ); \
        i++; \
    }

530 531 532 533 534 535 536 537 538 539
#define SET_IMVP(xy) \
    if( xy >= 0 ) \
    { \
        int shift = 1 + MB_INTERLACED - h->mb.field[xy]; \
        int16_t *mvp = h->mb.mvr[i_list][i_ref<<1>>shift][xy]; \
        mvc[i][0] = mvp[0]; \
        mvc[i][1] = mvp[1]<<1>>shift; \
        i++; \
    }

540 541 542 543 544 545 546 547 548
    /* b_direct */
    if( h->sh.i_type == SLICE_TYPE_B
        && h->mb.cache.ref[i_list][x264_scan8[12]] == i_ref )
    {
        SET_MVP( h->mb.cache.mv[i_list][x264_scan8[12]] );
    }

    if( i_ref == 0 && h->frames.b_have_lowres )
    {
549 550
        int idx = i_list ? h->fref[1][0]->i_frame-h->fenc->i_frame-1
                         : h->fenc->i_frame-h->fref[0][0]->i_frame-1;
551
        if( idx <= h->param.i_bframe )
552
        {
553 554 555 556 557 558
            int16_t (*lowres_mv)[2] = h->fenc->lowres_mvs[i_list][idx];
            if( lowres_mv[0][0] != 0x7fff )
            {
                M32( mvc[i] ) = (M32( lowres_mv[h->mb.i_mb_xy] )*2)&0xfffeffff;
                i++;
            }
559 560 561 562
        }
    }

    /* spatial predictors */
563 564 565 566 567 568 569 570 571 572 573 574 575 576 577
    if( SLICE_MBAFF )
    {
        SET_IMVP( h->mb.i_mb_left_xy[0] );
        SET_IMVP( h->mb.i_mb_top_xy );
        SET_IMVP( h->mb.i_mb_topleft_xy );
        SET_IMVP( h->mb.i_mb_topright_xy );
    }
    else
    {
        SET_MVP( mvr[h->mb.i_mb_left_xy[0]] );
        SET_MVP( mvr[h->mb.i_mb_top_xy] );
        SET_MVP( mvr[h->mb.i_mb_topleft_xy] );
        SET_MVP( mvr[h->mb.i_mb_topright_xy] );
    }
#undef SET_IMVP
578 579 580
#undef SET_MVP

    /* temporal predictors */
581
    if( h->fref[0][0]->i_ref[0] > 0 )
582
    {
583
        x264_frame_t *l0 = h->fref[0][0];
584
        int field = h->mb.i_mb_y&1;
585
        int curpoc = h->fdec->i_poc + h->fdec->i_delta_poc[field];
586
        int refpoc = h->fref[i_list][i_ref>>SLICE_MBAFF]->i_poc;
587
        refpoc += l0->i_delta_poc[field^(i_ref&1)];
588

Anton Mitrofanov's avatar
Anton Mitrofanov committed
589
#define SET_TMVP( dx, dy ) \
590 591
        { \
            int mb_index = h->mb.i_mb_xy + dx + dy*h->mb.i_mb_stride; \
592
            int scale = (curpoc - refpoc) * l0->inv_ref_poc[MB_INTERLACED&field]; \
Anton Mitrofanov's avatar
Anton Mitrofanov committed
593 594 595
            mvc[i][0] = (l0->mv16x16[mb_index][0]*scale + 128) >> 8; \
            mvc[i][1] = (l0->mv16x16[mb_index][1]*scale + 128) >> 8; \
            i++; \
596 597 598
        }

        SET_TMVP(0,0);
599
        if( h->mb.i_mb_x < h->mb.i_mb_width-1 )
600
            SET_TMVP(1,0);
601
        if( h->mb.i_mb_y < h->mb.i_mb_height-1 )
602 603 604 605 606 607
            SET_TMVP(0,1);
#undef SET_TMVP
    }

    *i_mvc = i;
}