h264_mvpred.h 30.9 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22
/*
 * H.26L/H.264/AVC/JVT/14496-10/... motion vector predicion
 * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
 *
 * This file is part of FFmpeg.
 *
 * FFmpeg is free software; you can redistribute it and/or
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
 * version 2.1 of the License, or (at your option) any later version.
 *
 * FFmpeg 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
 * Lesser General Public License for more details.
 *
 * You should have received a copy of the GNU Lesser General Public
 * License along with FFmpeg; if not, write to the Free Software
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 */

/**
23
 * @file
24 25 26 27
 * H.264 / AVC / MPEG4 part10 motion vector predicion.
 * @author Michael Niedermayer <michaelni@gmx.at>
 */

28 29 30
#ifndef AVCODEC_H264_MVPRED_H
#define AVCODEC_H264_MVPRED_H

31 32 33 34 35 36 37
#include "internal.h"
#include "avcodec.h"
#include "h264.h"

//#undef NDEBUG
#include <assert.h>

38
static av_always_inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
39 40 41 42 43 44 45
    const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];
    MpegEncContext *s = &h->s;

    /* there is no consistent mapping of mvs to neighboring locations that will
     * make mbaff happy, so we can't move all this logic to fill_caches */
    if(FRAME_MBAFF){

46 47 48
#define SET_DIAG_MV(MV_OP, REF_OP, XY, Y4)\
                const int xy = XY, y4 = Y4;\
                const int mb_type = mb_types[xy+(y4>>2)*s->mb_stride];\
49 50
                if(!USES_LIST(mb_type,list))\
                    return LIST_NOT_USED;\
51
                mv = s->current_picture_ptr->f.motion_val[list][h->mb2b_xy[xy] + 3 + y4*h->b_stride];\
52 53
                h->mv_cache[list][scan8[0]-2][0] = mv[0];\
                h->mv_cache[list][scan8[0]-2][1] = mv[1] MV_OP;\
54
                return s->current_picture_ptr->f.ref_index[list][4*xy + 1 + (y4 & ~1)] REF_OP;
55 56

        if(topright_ref == PART_NOT_AVAILABLE
57
           && i >= scan8[0]+8 && (i&7)==4
58
           && h->ref_cache[list][scan8[0]-1] != PART_NOT_AVAILABLE){
59
            const uint32_t *mb_types = s->current_picture_ptr->f.mb_type;
60
            const int16_t *mv;
61
            AV_ZERO32(h->mv_cache[list][scan8[0]-2]);
62 63
            *C = h->mv_cache[list][scan8[0]-2];

64
            if(!MB_FIELD
65 66
               && IS_INTERLACED(h->left_type[0])){
                SET_DIAG_MV(*2, >>1, h->left_mb_xy[0]+s->mb_stride, (s->mb_y&1)*2+(i>>5));
67 68
            }
            if(MB_FIELD
69
               && !IS_INTERLACED(h->left_type[0])){
70
                // left shift will turn LIST_NOT_USED into PART_NOT_AVAILABLE, but that's OK.
71
                SET_DIAG_MV(/2, <<1, h->left_mb_xy[i>=36], ((i>>2))&3);
72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94
            }
        }
#undef SET_DIAG_MV
    }

    if(topright_ref != PART_NOT_AVAILABLE){
        *C= h->mv_cache[list][ i - 8 + part_width ];
        return topright_ref;
    }else{
        tprintf(s->avctx, "topright MV not available\n");

        *C= h->mv_cache[list][ i - 8 - 1 ];
        return h->ref_cache[list][ i - 8 - 1 ];
    }
}

/**
 * gets the predicted MV.
 * @param n the block index
 * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)
 * @param mx the x component of the predicted motion vector
 * @param my the y component of the predicted motion vector
 */
95
static av_always_inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
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
    const int index8= scan8[n];
    const int top_ref=      h->ref_cache[list][ index8 - 8 ];
    const int left_ref=     h->ref_cache[list][ index8 - 1 ];
    const int16_t * const A= h->mv_cache[list][ index8 - 1 ];
    const int16_t * const B= h->mv_cache[list][ index8 - 8 ];
    const int16_t * C;
    int diagonal_ref, match_count;

    assert(part_width==1 || part_width==2 || part_width==4);

/* mv_cache
  B . . A T T T T
  U . . L . . , .
  U . . L . . . .
  U . . L . . , .
  . . . L . . . .
*/

    diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);
    match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);
    tprintf(h->s.avctx, "pred_motion match_count=%d\n", match_count);
    if(match_count > 1){ //most common
        *mx= mid_pred(A[0], B[0], C[0]);
        *my= mid_pred(A[1], B[1], C[1]);
    }else if(match_count==1){
        if(left_ref==ref){
            *mx= A[0];
            *my= A[1];
        }else if(top_ref==ref){
            *mx= B[0];
            *my= B[1];
        }else{
            *mx= C[0];
            *my= C[1];
        }
    }else{
        if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
            *mx= A[0];
            *my= A[1];
        }else{
            *mx= mid_pred(A[0], B[0], C[0]);
            *my= mid_pred(A[1], B[1], C[1]);
        }
    }

    tprintf(h->s.avctx, "pred_motion (%2d %2d %2d) (%2d %2d %2d) (%2d %2d %2d) -> (%2d %2d %2d) at %2d %2d %d list %d\n", top_ref, B[0], B[1],                    diagonal_ref, C[0], C[1], left_ref, A[0], A[1], ref, *mx, *my, h->s.mb_x, h->s.mb_y, n, list);
}

/**
 * gets the directionally predicted 16x8 MV.
 * @param n the block index
 * @param mx the x component of the predicted motion vector
 * @param my the y component of the predicted motion vector
 */
150
static av_always_inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184
    if(n==0){
        const int top_ref=      h->ref_cache[list][ scan8[0] - 8 ];
        const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];

        tprintf(h->s.avctx, "pred_16x8: (%2d %2d %2d) at %2d %2d %d list %d\n", top_ref, B[0], B[1], h->s.mb_x, h->s.mb_y, n, list);

        if(top_ref == ref){
            *mx= B[0];
            *my= B[1];
            return;
        }
    }else{
        const int left_ref=     h->ref_cache[list][ scan8[8] - 1 ];
        const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];

        tprintf(h->s.avctx, "pred_16x8: (%2d %2d %2d) at %2d %2d %d list %d\n", left_ref, A[0], A[1], h->s.mb_x, h->s.mb_y, n, list);

        if(left_ref == ref){
            *mx= A[0];
            *my= A[1];
            return;
        }
    }

    //RARE
    pred_motion(h, n, 4, list, ref, mx, my);
}

/**
 * gets the directionally predicted 8x16 MV.
 * @param n the block index
 * @param mx the x component of the predicted motion vector
 * @param my the y component of the predicted motion vector
 */
185
static av_always_inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215
    if(n==0){
        const int left_ref=      h->ref_cache[list][ scan8[0] - 1 ];
        const int16_t * const A=  h->mv_cache[list][ scan8[0] - 1 ];

        tprintf(h->s.avctx, "pred_8x16: (%2d %2d %2d) at %2d %2d %d list %d\n", left_ref, A[0], A[1], h->s.mb_x, h->s.mb_y, n, list);

        if(left_ref == ref){
            *mx= A[0];
            *my= A[1];
            return;
        }
    }else{
        const int16_t * C;
        int diagonal_ref;

        diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);

        tprintf(h->s.avctx, "pred_8x16: (%2d %2d %2d) at %2d %2d %d list %d\n", diagonal_ref, C[0], C[1], h->s.mb_x, h->s.mb_y, n, list);

        if(diagonal_ref == ref){
            *mx= C[0];
            *my= C[1];
            return;
        }
    }

    //RARE
    pred_motion(h, n, 2, list, ref, mx, my);
}

216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233
#define FIX_MV_MBAFF(type, refn, mvn, idx)\
    if(FRAME_MBAFF){\
        if(MB_FIELD){\
            if(!IS_INTERLACED(type)){\
                refn <<= 1;\
                AV_COPY32(mvbuf[idx], mvn);\
                mvbuf[idx][1] /= 2;\
                mvn = mvbuf[idx];\
            }\
        }else{\
            if(IS_INTERLACED(type)){\
                refn >>= 1;\
                AV_COPY32(mvbuf[idx], mvn);\
                mvbuf[idx][1] <<= 1;\
                mvn = mvbuf[idx];\
            }\
        }\
    }
234

235
static av_always_inline void pred_pskip_motion(H264Context * const h){
236 237 238
    DECLARE_ALIGNED(4, static const int16_t, zeromv)[2] = {0};
    DECLARE_ALIGNED(4, int16_t, mvbuf)[3][2];
    MpegEncContext * const s = &h->s;
239 240
    int8_t *ref      = s->current_picture.f.ref_index[0];
    int16_t (*mv)[2] = s->current_picture.f.motion_val[0];
241
    int top_ref, left_ref, diagonal_ref, match_count, mx, my;
242 243 244
    const int16_t *A, *B, *C;
    int b_stride = h->b_stride;

245 246
    fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);

247 248 249 250 251 252 253 254 255
    /* To avoid doing an entire fill_decode_caches, we inline the relevant parts here.
     * FIXME: this is a partial duplicate of the logic in fill_decode_caches, but it's
     * faster this way.  Is there a way to avoid this duplication?
     */
    if(USES_LIST(h->left_type[LTOP], 0)){
        left_ref = ref[4*h->left_mb_xy[LTOP] + 1 + (h->left_block[0]&~1)];
        A = mv[h->mb2b_xy[h->left_mb_xy[LTOP]] + 3 + b_stride*h->left_block[0]];
        FIX_MV_MBAFF(h->left_type[LTOP], left_ref, A, 0);
        if(!(left_ref | AV_RN32A(A))){
256
            goto zeromv;
257 258 259 260 261
        }
    }else if(h->left_type[LTOP]){
        left_ref = LIST_NOT_USED;
        A = zeromv;
    }else{
262
        goto zeromv;
263
    }
264

265 266 267 268 269
    if(USES_LIST(h->top_type, 0)){
        top_ref = ref[4*h->top_mb_xy + 2];
        B = mv[h->mb2b_xy[h->top_mb_xy] + 3*b_stride];
        FIX_MV_MBAFF(h->top_type, top_ref, B, 1);
        if(!(top_ref | AV_RN32A(B))){
270
            goto zeromv;
271 272 273 274 275
        }
    }else if(h->top_type){
        top_ref = LIST_NOT_USED;
        B = zeromv;
    }else{
276
        goto zeromv;
277 278
    }

279
    tprintf(h->s.avctx, "pred_pskip: (%d) (%d) at %2d %2d\n", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
280

281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299
    if(USES_LIST(h->topright_type, 0)){
        diagonal_ref = ref[4*h->topright_mb_xy + 2];
        C = mv[h->mb2b_xy[h->topright_mb_xy] + 3*b_stride];
        FIX_MV_MBAFF(h->topright_type, diagonal_ref, C, 2);
    }else if(h->topright_type){
        diagonal_ref = LIST_NOT_USED;
        C = zeromv;
    }else{
        if(USES_LIST(h->topleft_type, 0)){
            diagonal_ref = ref[4*h->topleft_mb_xy + 1 + (h->topleft_partition & 2)];
            C = mv[h->mb2b_xy[h->topleft_mb_xy] + 3 + b_stride + (h->topleft_partition & 2*b_stride)];
            FIX_MV_MBAFF(h->topleft_type, diagonal_ref, C, 2);
        }else if(h->topleft_type){
            diagonal_ref = LIST_NOT_USED;
            C = zeromv;
        }else{
            diagonal_ref = PART_NOT_AVAILABLE;
            C = zeromv;
        }
300 301
    }

302 303 304
    match_count= !diagonal_ref + !top_ref + !left_ref;
    tprintf(h->s.avctx, "pred_pskip_motion match_count=%d\n", match_count);
    if(match_count > 1){
305 306
        mx = mid_pred(A[0], B[0], C[0]);
        my = mid_pred(A[1], B[1], C[1]);
307 308
    }else if(match_count==1){
        if(!left_ref){
309 310
            mx = A[0];
            my = A[1];
311
        }else if(!top_ref){
312 313
            mx = B[0];
            my = B[1];
314
        }else{
315 316
            mx = C[0];
            my = C[1];
317 318
        }
    }else{
319 320
        mx = mid_pred(A[0], B[0], C[0]);
        my = mid_pred(A[1], B[1], C[1]);
321
    }
322

323 324 325 326
    fill_rectangle( h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4);
    return;
zeromv:
    fill_rectangle( h->mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
327 328
    return;
}
329

330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512
static void fill_decode_neighbors(H264Context *h, int mb_type){
    MpegEncContext * const s = &h->s;
    const int mb_xy= h->mb_xy;
    int topleft_xy, top_xy, topright_xy, left_xy[LEFT_MBS];
    static const uint8_t left_block_options[4][32]={
        {0,1,2,3,7,10,8,11,3+0*4, 3+1*4, 3+2*4, 3+3*4, 1+4*4, 1+8*4, 1+5*4, 1+9*4},
        {2,2,3,3,8,11,8,11,3+2*4, 3+2*4, 3+3*4, 3+3*4, 1+5*4, 1+9*4, 1+5*4, 1+9*4},
        {0,0,1,1,7,10,7,10,3+0*4, 3+0*4, 3+1*4, 3+1*4, 1+4*4, 1+8*4, 1+4*4, 1+8*4},
        {0,2,0,2,7,10,7,10,3+0*4, 3+2*4, 3+0*4, 3+2*4, 1+4*4, 1+8*4, 1+4*4, 1+8*4}
    };

    h->topleft_partition= -1;

    top_xy     = mb_xy  - (s->mb_stride << MB_FIELD);

    /* Wow, what a mess, why didn't they simplify the interlacing & intra
     * stuff, I can't imagine that these complex rules are worth it. */

    topleft_xy = top_xy - 1;
    topright_xy= top_xy + 1;
    left_xy[LBOT] = left_xy[LTOP] = mb_xy-1;
    h->left_block = left_block_options[0];
    if(FRAME_MBAFF){
        const int left_mb_field_flag = IS_INTERLACED(s->current_picture.f.mb_type[mb_xy - 1]);
        const int curr_mb_field_flag = IS_INTERLACED(mb_type);
        if(s->mb_y&1){
            if (left_mb_field_flag != curr_mb_field_flag) {
                left_xy[LBOT] = left_xy[LTOP] = mb_xy - s->mb_stride - 1;
                if (curr_mb_field_flag) {
                    left_xy[LBOT] += s->mb_stride;
                    h->left_block = left_block_options[3];
                } else {
                    topleft_xy += s->mb_stride;
                    // take top left mv from the middle of the mb, as opposed to all other modes which use the bottom right partition
                    h->topleft_partition = 0;
                    h->left_block = left_block_options[1];
                }
            }
        }else{
            if(curr_mb_field_flag){
                topleft_xy  += s->mb_stride & (((s->current_picture.f.mb_type[top_xy - 1] >> 7) & 1) - 1);
                topright_xy += s->mb_stride & (((s->current_picture.f.mb_type[top_xy + 1] >> 7) & 1) - 1);
                top_xy      += s->mb_stride & (((s->current_picture.f.mb_type[top_xy    ] >> 7) & 1) - 1);
            }
            if (left_mb_field_flag != curr_mb_field_flag) {
                if (curr_mb_field_flag) {
                    left_xy[LBOT] += s->mb_stride;
                    h->left_block = left_block_options[3];
                } else {
                    h->left_block = left_block_options[2];
                }
            }
        }
    }

    h->topleft_mb_xy = topleft_xy;
    h->top_mb_xy     = top_xy;
    h->topright_mb_xy= topright_xy;
    h->left_mb_xy[LTOP] = left_xy[LTOP];
    h->left_mb_xy[LBOT] = left_xy[LBOT];
    //FIXME do we need all in the context?

    h->topleft_type    = s->current_picture.f.mb_type[topleft_xy];
    h->top_type        = s->current_picture.f.mb_type[top_xy];
    h->topright_type   = s->current_picture.f.mb_type[topright_xy];
    h->left_type[LTOP] = s->current_picture.f.mb_type[left_xy[LTOP]];
    h->left_type[LBOT] = s->current_picture.f.mb_type[left_xy[LBOT]];

    if(FMO){
    if(h->slice_table[topleft_xy    ] != h->slice_num) h->topleft_type = 0;
    if(h->slice_table[top_xy        ] != h->slice_num) h->top_type     = 0;
    if(h->slice_table[left_xy[LTOP] ] != h->slice_num) h->left_type[LTOP] = h->left_type[LBOT] = 0;
    }else{
        if(h->slice_table[topleft_xy ] != h->slice_num){
            h->topleft_type = 0;
            if(h->slice_table[top_xy        ] != h->slice_num) h->top_type     = 0;
            if(h->slice_table[left_xy[LTOP] ] != h->slice_num) h->left_type[LTOP] = h->left_type[LBOT] = 0;
        }
    }
    if(h->slice_table[topright_xy] != h->slice_num) h->topright_type= 0;
}

static void fill_decode_caches(H264Context *h, int mb_type){
    MpegEncContext * const s = &h->s;
    int topleft_xy, top_xy, topright_xy, left_xy[LEFT_MBS];
    int topleft_type, top_type, topright_type, left_type[LEFT_MBS];
    const uint8_t * left_block= h->left_block;
    int i;
    uint8_t *nnz;
    uint8_t *nnz_cache;

    topleft_xy     = h->topleft_mb_xy;
    top_xy         = h->top_mb_xy;
    topright_xy    = h->topright_mb_xy;
    left_xy[LTOP]  = h->left_mb_xy[LTOP];
    left_xy[LBOT]  = h->left_mb_xy[LBOT];
    topleft_type   = h->topleft_type;
    top_type       = h->top_type;
    topright_type  = h->topright_type;
    left_type[LTOP]= h->left_type[LTOP];
    left_type[LBOT]= h->left_type[LBOT];

    if(!IS_SKIP(mb_type)){
        if(IS_INTRA(mb_type)){
            int type_mask= h->pps.constrained_intra_pred ? IS_INTRA(-1) : -1;
            h->topleft_samples_available=
            h->top_samples_available=
            h->left_samples_available= 0xFFFF;
            h->topright_samples_available= 0xEEEA;

            if(!(top_type & type_mask)){
                h->topleft_samples_available= 0xB3FF;
                h->top_samples_available= 0x33FF;
                h->topright_samples_available= 0x26EA;
            }
            if(IS_INTERLACED(mb_type) != IS_INTERLACED(left_type[LTOP])){
                if(IS_INTERLACED(mb_type)){
                    if(!(left_type[LTOP] & type_mask)){
                        h->topleft_samples_available&= 0xDFFF;
                        h->left_samples_available&= 0x5FFF;
                    }
                    if(!(left_type[LBOT] & type_mask)){
                        h->topleft_samples_available&= 0xFF5F;
                        h->left_samples_available&= 0xFF5F;
                    }
                }else{
                    int left_typei = s->current_picture.f.mb_type[left_xy[LTOP] + s->mb_stride];

                    assert(left_xy[LTOP] == left_xy[LBOT]);
                    if(!((left_typei & type_mask) && (left_type[LTOP] & type_mask))){
                        h->topleft_samples_available&= 0xDF5F;
                        h->left_samples_available&= 0x5F5F;
                    }
                }
            }else{
                if(!(left_type[LTOP] & type_mask)){
                    h->topleft_samples_available&= 0xDF5F;
                    h->left_samples_available&= 0x5F5F;
                }
            }

            if(!(topleft_type & type_mask))
                h->topleft_samples_available&= 0x7FFF;

            if(!(topright_type & type_mask))
                h->topright_samples_available&= 0xFBFF;

            if(IS_INTRA4x4(mb_type)){
                if(IS_INTRA4x4(top_type)){
                    AV_COPY32(h->intra4x4_pred_mode_cache+4+8*0, h->intra4x4_pred_mode + h->mb2br_xy[top_xy]);
                }else{
                    h->intra4x4_pred_mode_cache[4+8*0]=
                    h->intra4x4_pred_mode_cache[5+8*0]=
                    h->intra4x4_pred_mode_cache[6+8*0]=
                    h->intra4x4_pred_mode_cache[7+8*0]= 2 - 3*!(top_type & type_mask);
                }
                for(i=0; i<2; i++){
                    if(IS_INTRA4x4(left_type[LEFT(i)])){
                        int8_t *mode= h->intra4x4_pred_mode + h->mb2br_xy[left_xy[LEFT(i)]];
                        h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= mode[6-left_block[0+2*i]];
                        h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= mode[6-left_block[1+2*i]];
                    }else{
                        h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
                        h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= 2 - 3*!(left_type[LEFT(i)] & type_mask);
                    }
                }
            }
        }


/*
0 . T T. T T T T
1 L . .L . . . .
2 L . .L . . . .
3 . T TL . . . .
4 L . .L . . . .
5 L . .. . . . .
*/
//FIXME constraint_intra_pred & partitioning & nnz (let us hope this is just a typo in the spec)
    nnz_cache = h->non_zero_count_cache;
    if(top_type){
        nnz = h->non_zero_count[top_xy];
        AV_COPY32(&nnz_cache[4+8* 0], &nnz[4*3]);
513
        if(!s->chroma_y_shift){
514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536
            AV_COPY32(&nnz_cache[4+8* 5], &nnz[4* 7]);
            AV_COPY32(&nnz_cache[4+8*10], &nnz[4*11]);
        }else{
            AV_COPY32(&nnz_cache[4+8* 5], &nnz[4* 5]);
            AV_COPY32(&nnz_cache[4+8*10], &nnz[4* 9]);
        }
    }else{
        uint32_t top_empty = CABAC && !IS_INTRA(mb_type) ? 0 : 0x40404040;
        AV_WN32A(&nnz_cache[4+8* 0], top_empty);
        AV_WN32A(&nnz_cache[4+8* 5], top_empty);
        AV_WN32A(&nnz_cache[4+8*10], top_empty);
    }

    for (i=0; i<2; i++) {
        if(left_type[LEFT(i)]){
            nnz = h->non_zero_count[left_xy[LEFT(i)]];
            nnz_cache[3+8* 1 + 2*8*i]= nnz[left_block[8+0+2*i]];
            nnz_cache[3+8* 2 + 2*8*i]= nnz[left_block[8+1+2*i]];
            if(CHROMA444){
                nnz_cache[3+8* 6 + 2*8*i]= nnz[left_block[8+0+2*i]+4*4];
                nnz_cache[3+8* 7 + 2*8*i]= nnz[left_block[8+1+2*i]+4*4];
                nnz_cache[3+8*11 + 2*8*i]= nnz[left_block[8+0+2*i]+8*4];
                nnz_cache[3+8*12 + 2*8*i]= nnz[left_block[8+1+2*i]+8*4];
537 538 539 540 541
            }else if(CHROMA422) {
                nnz_cache[3+8* 6 + 2*8*i]= nnz[left_block[8+0+2*i]-2+4*4];
                nnz_cache[3+8* 7 + 2*8*i]= nnz[left_block[8+1+2*i]-2+4*4];
                nnz_cache[3+8*11 + 2*8*i]= nnz[left_block[8+0+2*i]-2+8*4];
                nnz_cache[3+8*12 + 2*8*i]= nnz[left_block[8+1+2*i]-2+8*4];
542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752
            }else{
                nnz_cache[3+8* 6 +   8*i]= nnz[left_block[8+4+2*i]];
                nnz_cache[3+8*11 +   8*i]= nnz[left_block[8+5+2*i]];
            }
        }else{
            nnz_cache[3+8* 1 + 2*8*i]=
            nnz_cache[3+8* 2 + 2*8*i]=
            nnz_cache[3+8* 6 + 2*8*i]=
            nnz_cache[3+8* 7 + 2*8*i]=
            nnz_cache[3+8*11 + 2*8*i]=
            nnz_cache[3+8*12 + 2*8*i]= CABAC && !IS_INTRA(mb_type) ? 0 : 64;
        }
    }

    if( CABAC ) {
        // top_cbp
        if(top_type) {
            h->top_cbp = h->cbp_table[top_xy];
        } else {
            h->top_cbp = IS_INTRA(mb_type) ? 0x7CF : 0x00F;
        }
        // left_cbp
        if (left_type[LTOP]) {
            h->left_cbp =   (h->cbp_table[left_xy[LTOP]] & 0x7F0)
                        |  ((h->cbp_table[left_xy[LTOP]]>>(left_block[0]&(~1)))&2)
                        | (((h->cbp_table[left_xy[LBOT]]>>(left_block[2]&(~1)))&2) << 2);
        } else {
            h->left_cbp = IS_INTRA(mb_type) ? 0x7CF : 0x00F;
        }
    }
    }

    if(IS_INTER(mb_type) || (IS_DIRECT(mb_type) && h->direct_spatial_mv_pred)){
        int list;
        int b_stride = h->b_stride;
        for(list=0; list<h->list_count; list++){
            int8_t *ref_cache = &h->ref_cache[list][scan8[0]];
            int8_t *ref       = s->current_picture.f.ref_index[list];
            int16_t (*mv_cache)[2] = &h->mv_cache[list][scan8[0]];
            int16_t (*mv)[2]       = s->current_picture.f.motion_val[list];
            if(!USES_LIST(mb_type, list)){
                continue;
            }
            assert(!(IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred));

            if(USES_LIST(top_type, list)){
                const int b_xy= h->mb2b_xy[top_xy] + 3*b_stride;
                AV_COPY128(mv_cache[0 - 1*8], mv[b_xy + 0]);
                ref_cache[0 - 1*8]=
                ref_cache[1 - 1*8]= ref[4*top_xy + 2];
                ref_cache[2 - 1*8]=
                ref_cache[3 - 1*8]= ref[4*top_xy + 3];
            }else{
                AV_ZERO128(mv_cache[0 - 1*8]);
                AV_WN32A(&ref_cache[0 - 1*8], ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101);
            }

            if(mb_type & (MB_TYPE_16x8|MB_TYPE_8x8)){
            for(i=0; i<2; i++){
                int cache_idx = -1 + i*2*8;
                if(USES_LIST(left_type[LEFT(i)], list)){
                    const int b_xy= h->mb2b_xy[left_xy[LEFT(i)]] + 3;
                    const int b8_xy= 4*left_xy[LEFT(i)] + 1;
                    AV_COPY32(mv_cache[cache_idx  ], mv[b_xy + b_stride*left_block[0+i*2]]);
                    AV_COPY32(mv_cache[cache_idx+8], mv[b_xy + b_stride*left_block[1+i*2]]);
                    ref_cache[cache_idx  ]= ref[b8_xy + (left_block[0+i*2]&~1)];
                    ref_cache[cache_idx+8]= ref[b8_xy + (left_block[1+i*2]&~1)];
                }else{
                    AV_ZERO32(mv_cache[cache_idx  ]);
                    AV_ZERO32(mv_cache[cache_idx+8]);
                    ref_cache[cache_idx  ]=
                    ref_cache[cache_idx+8]= (left_type[LEFT(i)]) ? LIST_NOT_USED : PART_NOT_AVAILABLE;
                }
            }
            }else{
                if(USES_LIST(left_type[LTOP], list)){
                    const int b_xy= h->mb2b_xy[left_xy[LTOP]] + 3;
                    const int b8_xy= 4*left_xy[LTOP] + 1;
                    AV_COPY32(mv_cache[-1], mv[b_xy + b_stride*left_block[0]]);
                    ref_cache[-1]= ref[b8_xy + (left_block[0]&~1)];
                }else{
                    AV_ZERO32(mv_cache[-1]);
                    ref_cache[-1]= left_type[LTOP] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
                }
            }

            if(USES_LIST(topright_type, list)){
                const int b_xy= h->mb2b_xy[topright_xy] + 3*b_stride;
                AV_COPY32(mv_cache[4 - 1*8], mv[b_xy]);
                ref_cache[4 - 1*8]= ref[4*topright_xy + 2];
            }else{
                AV_ZERO32(mv_cache[4 - 1*8]);
                ref_cache[4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
            }
            if(ref_cache[4 - 1*8] < 0){
                if(USES_LIST(topleft_type, list)){
                    const int b_xy = h->mb2b_xy[topleft_xy] + 3 + b_stride + (h->topleft_partition & 2*b_stride);
                    const int b8_xy= 4*topleft_xy + 1 + (h->topleft_partition & 2);
                    AV_COPY32(mv_cache[-1 - 1*8], mv[b_xy]);
                    ref_cache[-1 - 1*8]= ref[b8_xy];
                }else{
                    AV_ZERO32(mv_cache[-1 - 1*8]);
                    ref_cache[-1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
                }
            }

            if((mb_type&(MB_TYPE_SKIP|MB_TYPE_DIRECT2)) && !FRAME_MBAFF)
                continue;

            if(!(mb_type&(MB_TYPE_SKIP|MB_TYPE_DIRECT2))){
                uint8_t (*mvd_cache)[2] = &h->mvd_cache[list][scan8[0]];
                uint8_t (*mvd)[2] = h->mvd_table[list];
                ref_cache[2+8*0] =
                ref_cache[2+8*2] = PART_NOT_AVAILABLE;
                AV_ZERO32(mv_cache[2+8*0]);
                AV_ZERO32(mv_cache[2+8*2]);

                if( CABAC ) {
                    if(USES_LIST(top_type, list)){
                        const int b_xy= h->mb2br_xy[top_xy];
                        AV_COPY64(mvd_cache[0 - 1*8], mvd[b_xy + 0]);
                    }else{
                        AV_ZERO64(mvd_cache[0 - 1*8]);
                    }
                    if(USES_LIST(left_type[LTOP], list)){
                        const int b_xy= h->mb2br_xy[left_xy[LTOP]] + 6;
                        AV_COPY16(mvd_cache[-1 + 0*8], mvd[b_xy - left_block[0]]);
                        AV_COPY16(mvd_cache[-1 + 1*8], mvd[b_xy - left_block[1]]);
                    }else{
                        AV_ZERO16(mvd_cache[-1 + 0*8]);
                        AV_ZERO16(mvd_cache[-1 + 1*8]);
                    }
                    if(USES_LIST(left_type[LBOT], list)){
                        const int b_xy= h->mb2br_xy[left_xy[LBOT]] + 6;
                        AV_COPY16(mvd_cache[-1 + 2*8], mvd[b_xy - left_block[2]]);
                        AV_COPY16(mvd_cache[-1 + 3*8], mvd[b_xy - left_block[3]]);
                    }else{
                        AV_ZERO16(mvd_cache[-1 + 2*8]);
                        AV_ZERO16(mvd_cache[-1 + 3*8]);
                    }
                    AV_ZERO16(mvd_cache[2+8*0]);
                    AV_ZERO16(mvd_cache[2+8*2]);
                    if(h->slice_type_nos == AV_PICTURE_TYPE_B){
                        uint8_t *direct_cache = &h->direct_cache[scan8[0]];
                        uint8_t *direct_table = h->direct_table;
                        fill_rectangle(direct_cache, 4, 4, 8, MB_TYPE_16x16>>1, 1);

                        if(IS_DIRECT(top_type)){
                            AV_WN32A(&direct_cache[-1*8], 0x01010101u*(MB_TYPE_DIRECT2>>1));
                        }else if(IS_8X8(top_type)){
                            int b8_xy = 4*top_xy;
                            direct_cache[0 - 1*8]= direct_table[b8_xy + 2];
                            direct_cache[2 - 1*8]= direct_table[b8_xy + 3];
                        }else{
                            AV_WN32A(&direct_cache[-1*8], 0x01010101*(MB_TYPE_16x16>>1));
                        }

                        if(IS_DIRECT(left_type[LTOP]))
                            direct_cache[-1 + 0*8]= MB_TYPE_DIRECT2>>1;
                        else if(IS_8X8(left_type[LTOP]))
                            direct_cache[-1 + 0*8]= direct_table[4*left_xy[LTOP] + 1 + (left_block[0]&~1)];
                        else
                            direct_cache[-1 + 0*8]= MB_TYPE_16x16>>1;

                        if(IS_DIRECT(left_type[LBOT]))
                            direct_cache[-1 + 2*8]= MB_TYPE_DIRECT2>>1;
                        else if(IS_8X8(left_type[LBOT]))
                            direct_cache[-1 + 2*8]= direct_table[4*left_xy[LBOT] + 1 + (left_block[2]&~1)];
                        else
                            direct_cache[-1 + 2*8]= MB_TYPE_16x16>>1;
                    }
                }
            }
            if(FRAME_MBAFF){
#define MAP_MVS\
                    MAP_F2F(scan8[0] - 1 - 1*8, topleft_type)\
                    MAP_F2F(scan8[0] + 0 - 1*8, top_type)\
                    MAP_F2F(scan8[0] + 1 - 1*8, top_type)\
                    MAP_F2F(scan8[0] + 2 - 1*8, top_type)\
                    MAP_F2F(scan8[0] + 3 - 1*8, top_type)\
                    MAP_F2F(scan8[0] + 4 - 1*8, topright_type)\
                    MAP_F2F(scan8[0] - 1 + 0*8, left_type[LTOP])\
                    MAP_F2F(scan8[0] - 1 + 1*8, left_type[LTOP])\
                    MAP_F2F(scan8[0] - 1 + 2*8, left_type[LBOT])\
                    MAP_F2F(scan8[0] - 1 + 3*8, left_type[LBOT])
                if(MB_FIELD){
#define MAP_F2F(idx, mb_type)\
                    if(!IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
                        h->ref_cache[list][idx] <<= 1;\
                        h->mv_cache[list][idx][1] /= 2;\
                        h->mvd_cache[list][idx][1] >>=1;\
                    }
                    MAP_MVS
#undef MAP_F2F
                }else{
#define MAP_F2F(idx, mb_type)\
                    if(IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
                        h->ref_cache[list][idx] >>= 1;\
                        h->mv_cache[list][idx][1] <<= 1;\
                        h->mvd_cache[list][idx][1] <<= 1;\
                    }
                    MAP_MVS
#undef MAP_F2F
                }
            }
        }
    }

        h->neighbor_transform_size= !!IS_8x8DCT(top_type) + !!IS_8x8DCT(left_type[LTOP]);
}

753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791
/**
 * decodes a P_SKIP or B_SKIP macroblock
 */
static void av_unused decode_mb_skip(H264Context *h){
    MpegEncContext * const s = &h->s;
    const int mb_xy= h->mb_xy;
    int mb_type=0;

    memset(h->non_zero_count[mb_xy], 0, 48);

    if(MB_FIELD)
        mb_type|= MB_TYPE_INTERLACED;

    if( h->slice_type_nos == AV_PICTURE_TYPE_B )
    {
        // just for fill_caches. pred_direct_motion will set the real mb_type
        mb_type|= MB_TYPE_L0L1|MB_TYPE_DIRECT2|MB_TYPE_SKIP;
        if(h->direct_spatial_mv_pred){
            fill_decode_neighbors(h, mb_type);
        fill_decode_caches(h, mb_type); //FIXME check what is needed and what not ...
        }
        ff_h264_pred_direct_motion(h, &mb_type);
        mb_type|= MB_TYPE_SKIP;
    }
    else
    {
        mb_type|= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0|MB_TYPE_SKIP;

        fill_decode_neighbors(h, mb_type);
        pred_pskip_motion(h);
    }

    write_back_motion(h, mb_type);
    s->current_picture.f.mb_type[mb_xy]      = mb_type;
    s->current_picture.f.qscale_table[mb_xy] = s->qscale;
    h->slice_table[ mb_xy ]= h->slice_num;
    h->prev_mb_skipped= 1;
}

792
#endif /* AVCODEC_H264_MVPRED_H */