1/*
2 * H.26L/H.264/AVC/JVT/14496-10/... direct mb/block decoding
3 * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
4 *
5 * This file is part of FFmpeg.
6 *
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
11 *
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
15 * Lesser General Public License for more details.
16 *
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20 */
21
22/**
23 * @file
24 * H.264 / AVC / MPEG-4 part10 direct mb/block decoding.
25 * @author Michael Niedermayer <michaelni@gmx.at>
26 */
27
28#include "avcodec.h"
29#include "h264dec.h"
30#include "h264_ps.h"
31#include "mpegutils.h"
32#include "rectangle.h"
33#include "threadframe.h"
34
35#include <assert.h>
36
37static int get_scale_factor(H264SliceContext *sl,
38                            int poc, int poc1, int i)
39{
40    int poc0 = sl->ref_list[0][i].poc;
41    int64_t pocdiff = poc1 - (int64_t)poc0;
42    int td = av_clip_int8(pocdiff);
43
44    if (pocdiff != (int)pocdiff)
45        avpriv_request_sample(sl->h264->avctx, "pocdiff overflow");
46
47    if (td == 0 || sl->ref_list[0][i].parent->long_ref) {
48        return 256;
49    } else {
50        int64_t pocdiff0 = poc - (int64_t)poc0;
51        int tb = av_clip_int8(pocdiff0);
52        int tx = (16384 + (FFABS(td) >> 1)) / td;
53
54        if (pocdiff0 != (int)pocdiff0)
55            av_log(sl->h264->avctx, AV_LOG_DEBUG, "pocdiff0 overflow\n");
56
57        return av_clip_intp2((tb * tx + 32) >> 6, 10);
58    }
59}
60
61void ff_h264_direct_dist_scale_factor(const H264Context *const h,
62                                      H264SliceContext *sl)
63{
64    const int poc  = FIELD_PICTURE(h) ? h->cur_pic_ptr->field_poc[h->picture_structure == PICT_BOTTOM_FIELD]
65                                      : h->cur_pic_ptr->poc;
66    const int poc1 = sl->ref_list[1][0].poc;
67    int i, field;
68
69    if (FRAME_MBAFF(h))
70        for (field = 0; field < 2; field++) {
71            const int poc  = h->cur_pic_ptr->field_poc[field];
72            const int poc1 = sl->ref_list[1][0].parent->field_poc[field];
73            for (i = 0; i < 2 * sl->ref_count[0]; i++)
74                sl->dist_scale_factor_field[field][i ^ field] =
75                    get_scale_factor(sl, poc, poc1, i + 16);
76        }
77
78    for (i = 0; i < sl->ref_count[0]; i++)
79        sl->dist_scale_factor[i] = get_scale_factor(sl, poc, poc1, i);
80}
81
82static void fill_colmap(const H264Context *h, H264SliceContext *sl,
83                        int map[2][16 + 32], int list,
84                        int field, int colfield, int mbafi)
85{
86    H264Picture *const ref1 = sl->ref_list[1][0].parent;
87    int j, old_ref, rfield;
88    int start  = mbafi ? 16                       : 0;
89    int end    = mbafi ? 16 + 2 * sl->ref_count[0] : sl->ref_count[0];
90    int interl = mbafi || h->picture_structure != PICT_FRAME;
91
92    /* bogus; fills in for missing frames */
93    memset(map[list], 0, sizeof(map[list]));
94
95    for (rfield = 0; rfield < 2; rfield++) {
96        for (old_ref = 0; old_ref < ref1->ref_count[colfield][list]; old_ref++) {
97            int poc = ref1->ref_poc[colfield][list][old_ref];
98
99            if (!interl)
100                poc |= 3;
101            // FIXME: store all MBAFF references so this is not needed
102            else if (interl && (poc & 3) == 3)
103                poc = (poc & ~3) + rfield + 1;
104
105            for (j = start; j < end; j++) {
106                if (4 * sl->ref_list[0][j].parent->frame_num +
107                    (sl->ref_list[0][j].reference & 3) == poc) {
108                    int cur_ref = mbafi ? (j - 16) ^ field : j;
109                    if (ref1->mbaff)
110                        map[list][2 * old_ref + (rfield ^ field) + 16] = cur_ref;
111                    if (rfield == field || !interl)
112                        map[list][old_ref] = cur_ref;
113                    break;
114                }
115            }
116        }
117    }
118}
119
120void ff_h264_direct_ref_list_init(const H264Context *const h, H264SliceContext *sl)
121{
122    H264Ref *const ref1 = &sl->ref_list[1][0];
123    H264Picture *const cur = h->cur_pic_ptr;
124    int list, j, field;
125    int sidx     = (h->picture_structure & 1) ^ 1;
126    int ref1sidx = (ref1->reference      & 1) ^ 1;
127
128    for (list = 0; list < sl->list_count; list++) {
129        cur->ref_count[sidx][list] = sl->ref_count[list];
130        for (j = 0; j < sl->ref_count[list]; j++)
131            cur->ref_poc[sidx][list][j] = 4 * sl->ref_list[list][j].parent->frame_num +
132                                          (sl->ref_list[list][j].reference & 3);
133    }
134
135    if (h->picture_structure == PICT_FRAME) {
136        memcpy(cur->ref_count[1], cur->ref_count[0], sizeof(cur->ref_count[0]));
137        memcpy(cur->ref_poc[1],   cur->ref_poc[0],   sizeof(cur->ref_poc[0]));
138    }
139
140    if (h->current_slice == 0) {
141        cur->mbaff = FRAME_MBAFF(h);
142    } else {
143        av_assert0(cur->mbaff == FRAME_MBAFF(h));
144    }
145
146    sl->col_fieldoff = 0;
147
148    if (sl->list_count != 2 || !sl->ref_count[1])
149        return;
150
151    if (h->picture_structure == PICT_FRAME) {
152        int cur_poc  = h->cur_pic_ptr->poc;
153        int *col_poc = sl->ref_list[1][0].parent->field_poc;
154        if (col_poc[0] == INT_MAX && col_poc[1] == INT_MAX) {
155            av_log(h->avctx, AV_LOG_ERROR, "co located POCs unavailable\n");
156            sl->col_parity = 1;
157        } else
158            sl->col_parity = (FFABS(col_poc[0] - (int64_t)cur_poc) >=
159                              FFABS(col_poc[1] - (int64_t)cur_poc));
160        ref1sidx =
161        sidx     = sl->col_parity;
162    // FL -> FL & differ parity
163    } else if (!(h->picture_structure & sl->ref_list[1][0].reference) &&
164               !sl->ref_list[1][0].parent->mbaff) {
165        sl->col_fieldoff = 2 * sl->ref_list[1][0].reference - 3;
166    }
167
168    if (sl->slice_type_nos != AV_PICTURE_TYPE_B || sl->direct_spatial_mv_pred)
169        return;
170
171    for (list = 0; list < 2; list++) {
172        fill_colmap(h, sl, sl->map_col_to_list0, list, sidx, ref1sidx, 0);
173        if (FRAME_MBAFF(h))
174            for (field = 0; field < 2; field++)
175                fill_colmap(h, sl, sl->map_col_to_list0_field[field], list, field,
176                            field, 1);
177    }
178}
179
180static void await_reference_mb_row(const H264Context *const h, H264Ref *ref,
181                                   int mb_y)
182{
183    int ref_field         = ref->reference - 1;
184    int ref_field_picture = ref->parent->field_picture;
185    int ref_height        = 16 * h->mb_height >> ref_field_picture;
186
187    if (!HAVE_THREADS || !(h->avctx->active_thread_type & FF_THREAD_FRAME))
188        return;
189
190    /* FIXME: It can be safe to access mb stuff
191     * even if pixels aren't deblocked yet. */
192
193    ff_thread_await_progress(&ref->parent->tf,
194                             FFMIN(16 * mb_y >> ref_field_picture,
195                                   ref_height - 1),
196                             ref_field_picture && ref_field);
197}
198
199static void pred_spatial_direct_motion(const H264Context *const h, H264SliceContext *sl,
200                                       int *mb_type)
201{
202    int b8_stride = 2;
203    int b4_stride = h->b_stride;
204    int mb_xy = sl->mb_xy, mb_y = sl->mb_y;
205    int mb_type_col[2];
206    const int16_t (*l1mv0)[2], (*l1mv1)[2];
207    const int8_t *l1ref0, *l1ref1;
208    const int is_b8x8 = IS_8X8(*mb_type);
209    unsigned int sub_mb_type = MB_TYPE_L0L1;
210    int i8, i4;
211    int ref[2];
212    int mv[2];
213    int list;
214
215    assert(sl->ref_list[1][0].reference & 3);
216
217    await_reference_mb_row(h, &sl->ref_list[1][0],
218                           sl->mb_y + !!IS_INTERLACED(*mb_type));
219
220#define MB_TYPE_16x16_OR_INTRA (MB_TYPE_16x16 | MB_TYPE_INTRA4x4 | \
221                                MB_TYPE_INTRA16x16 | MB_TYPE_INTRA_PCM)
222
223    /* ref = min(neighbors) */
224    for (list = 0; list < 2; list++) {
225        int left_ref     = sl->ref_cache[list][scan8[0] - 1];
226        int top_ref      = sl->ref_cache[list][scan8[0] - 8];
227        int refc         = sl->ref_cache[list][scan8[0] - 8 + 4];
228        const int16_t *C = sl->mv_cache[list][scan8[0]  - 8 + 4];
229        if (refc == PART_NOT_AVAILABLE) {
230            refc = sl->ref_cache[list][scan8[0] - 8 - 1];
231            C    = sl->mv_cache[list][scan8[0]  - 8 - 1];
232        }
233        ref[list] = FFMIN3((unsigned)left_ref,
234                           (unsigned)top_ref,
235                           (unsigned)refc);
236        if (ref[list] >= 0) {
237            /* This is just pred_motion() but with the cases removed that
238             * cannot happen for direct blocks. */
239            const int16_t *const A = sl->mv_cache[list][scan8[0] - 1];
240            const int16_t *const B = sl->mv_cache[list][scan8[0] - 8];
241
242            int match_count = (left_ref == ref[list]) +
243                              (top_ref  == ref[list]) +
244                              (refc     == ref[list]);
245
246            if (match_count > 1) { // most common
247                mv[list] = pack16to32(mid_pred(A[0], B[0], C[0]),
248                                      mid_pred(A[1], B[1], C[1]));
249            } else {
250                assert(match_count == 1);
251                if (left_ref == ref[list])
252                    mv[list] = AV_RN32A(A);
253                else if (top_ref == ref[list])
254                    mv[list] = AV_RN32A(B);
255                else
256                    mv[list] = AV_RN32A(C);
257            }
258            av_assert2(ref[list] < (sl->ref_count[list] << !!FRAME_MBAFF(h)));
259        } else {
260            int mask = ~(MB_TYPE_L0 << (2 * list));
261            mv[list]  = 0;
262            ref[list] = -1;
263            if (!is_b8x8)
264                *mb_type &= mask;
265            sub_mb_type &= mask;
266        }
267    }
268    if (ref[0] < 0 && ref[1] < 0) {
269        ref[0] = ref[1] = 0;
270        if (!is_b8x8)
271            *mb_type |= MB_TYPE_L0L1;
272        sub_mb_type |= MB_TYPE_L0L1;
273    }
274
275    if (!(is_b8x8 | mv[0] | mv[1])) {
276        fill_rectangle(&sl->ref_cache[0][scan8[0]], 4, 4, 8, (uint8_t)ref[0], 1);
277        fill_rectangle(&sl->ref_cache[1][scan8[0]], 4, 4, 8, (uint8_t)ref[1], 1);
278        fill_rectangle(&sl->mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
279        fill_rectangle(&sl->mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
280        *mb_type = (*mb_type & ~(MB_TYPE_8x8 | MB_TYPE_16x8 | MB_TYPE_8x16 |
281                                 MB_TYPE_P1L0 | MB_TYPE_P1L1)) |
282                   MB_TYPE_16x16 | MB_TYPE_DIRECT2;
283        return;
284    }
285
286    if (IS_INTERLACED(sl->ref_list[1][0].parent->mb_type[mb_xy])) { // AFL/AFR/FR/FL -> AFL/FL
287        if (!IS_INTERLACED(*mb_type)) {                    //     AFR/FR    -> AFL/FL
288            mb_y  = (sl->mb_y & ~1) + sl->col_parity;
289            mb_xy = sl->mb_x +
290                    ((sl->mb_y & ~1) + sl->col_parity) * h->mb_stride;
291            b8_stride = 0;
292        } else {
293            mb_y  += sl->col_fieldoff;
294            mb_xy += h->mb_stride * sl->col_fieldoff; // non-zero for FL -> FL & differ parity
295        }
296        goto single_col;
297    } else {                                             // AFL/AFR/FR/FL -> AFR/FR
298        if (IS_INTERLACED(*mb_type)) {                   // AFL       /FL -> AFR/FR
299            mb_y           =  sl->mb_y & ~1;
300            mb_xy          = (sl->mb_y & ~1) * h->mb_stride + sl->mb_x;
301            mb_type_col[0] = sl->ref_list[1][0].parent->mb_type[mb_xy];
302            mb_type_col[1] = sl->ref_list[1][0].parent->mb_type[mb_xy + h->mb_stride];
303            b8_stride      = 2 + 4 * h->mb_stride;
304            b4_stride     *= 6;
305            if (IS_INTERLACED(mb_type_col[0]) !=
306                IS_INTERLACED(mb_type_col[1])) {
307                mb_type_col[0] &= ~MB_TYPE_INTERLACED;
308                mb_type_col[1] &= ~MB_TYPE_INTERLACED;
309            }
310
311            sub_mb_type |= MB_TYPE_16x16 | MB_TYPE_DIRECT2; /* B_SUB_8x8 */
312            if ((mb_type_col[0] & MB_TYPE_16x16_OR_INTRA) &&
313                (mb_type_col[1] & MB_TYPE_16x16_OR_INTRA) &&
314                !is_b8x8) {
315                *mb_type |= MB_TYPE_16x8 | MB_TYPE_DIRECT2;  /* B_16x8 */
316            } else {
317                *mb_type |= MB_TYPE_8x8;
318            }
319        } else {                                         //     AFR/FR    -> AFR/FR
320single_col:
321            mb_type_col[0] =
322            mb_type_col[1] = sl->ref_list[1][0].parent->mb_type[mb_xy];
323
324            sub_mb_type |= MB_TYPE_16x16 | MB_TYPE_DIRECT2; /* B_SUB_8x8 */
325            if (!is_b8x8 && (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)) {
326                *mb_type |= MB_TYPE_16x16 | MB_TYPE_DIRECT2; /* B_16x16 */
327            } else if (!is_b8x8 &&
328                       (mb_type_col[0] & (MB_TYPE_16x8 | MB_TYPE_8x16))) {
329                *mb_type |= MB_TYPE_DIRECT2 |
330                            (mb_type_col[0] & (MB_TYPE_16x8 | MB_TYPE_8x16));
331            } else {
332                if (!h->ps.sps->direct_8x8_inference_flag) {
333                    /* FIXME: Save sub mb types from previous frames (or derive
334                     * from MVs) so we know exactly what block size to use. */
335                    sub_mb_type += (MB_TYPE_8x8 - MB_TYPE_16x16); /* B_SUB_4x4 */
336                }
337                *mb_type |= MB_TYPE_8x8;
338            }
339        }
340    }
341
342    await_reference_mb_row(h, &sl->ref_list[1][0], mb_y);
343
344    l1mv0  = (void*)&sl->ref_list[1][0].parent->motion_val[0][h->mb2b_xy[mb_xy]];
345    l1mv1  = (void*)&sl->ref_list[1][0].parent->motion_val[1][h->mb2b_xy[mb_xy]];
346    l1ref0 = &sl->ref_list[1][0].parent->ref_index[0][4 * mb_xy];
347    l1ref1 = &sl->ref_list[1][0].parent->ref_index[1][4 * mb_xy];
348    if (!b8_stride) {
349        if (sl->mb_y & 1) {
350            l1ref0 += 2;
351            l1ref1 += 2;
352            l1mv0  += 2 * b4_stride;
353            l1mv1  += 2 * b4_stride;
354        }
355    }
356
357    if (IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col[0])) {
358        int n = 0;
359        for (i8 = 0; i8 < 4; i8++) {
360            int x8  = i8 & 1;
361            int y8  = i8 >> 1;
362            int xy8 = x8     + y8 * b8_stride;
363            int xy4 = x8 * 3 + y8 * b4_stride;
364            int a, b;
365
366            if (is_b8x8 && !IS_DIRECT(sl->sub_mb_type[i8]))
367                continue;
368            sl->sub_mb_type[i8] = sub_mb_type;
369
370            fill_rectangle(&sl->ref_cache[0][scan8[i8 * 4]], 2, 2, 8,
371                           (uint8_t)ref[0], 1);
372            fill_rectangle(&sl->ref_cache[1][scan8[i8 * 4]], 2, 2, 8,
373                           (uint8_t)ref[1], 1);
374            if (!IS_INTRA(mb_type_col[y8]) && !sl->ref_list[1][0].parent->long_ref &&
375                ((l1ref0[xy8] == 0 &&
376                  FFABS(l1mv0[xy4][0]) <= 1 &&
377                  FFABS(l1mv0[xy4][1]) <= 1) ||
378                 (l1ref0[xy8] < 0 &&
379                  l1ref1[xy8] == 0 &&
380                  FFABS(l1mv1[xy4][0]) <= 1 &&
381                  FFABS(l1mv1[xy4][1]) <= 1))) {
382                a =
383                b = 0;
384                if (ref[0] > 0)
385                    a = mv[0];
386                if (ref[1] > 0)
387                    b = mv[1];
388                n++;
389            } else {
390                a = mv[0];
391                b = mv[1];
392            }
393            fill_rectangle(&sl->mv_cache[0][scan8[i8 * 4]], 2, 2, 8, a, 4);
394            fill_rectangle(&sl->mv_cache[1][scan8[i8 * 4]], 2, 2, 8, b, 4);
395        }
396        if (!is_b8x8 && !(n & 3))
397            *mb_type = (*mb_type & ~(MB_TYPE_8x8 | MB_TYPE_16x8 | MB_TYPE_8x16 |
398                                     MB_TYPE_P1L0 | MB_TYPE_P1L1)) |
399                       MB_TYPE_16x16 | MB_TYPE_DIRECT2;
400    } else if (IS_16X16(*mb_type)) {
401        int a, b;
402
403        fill_rectangle(&sl->ref_cache[0][scan8[0]], 4, 4, 8, (uint8_t)ref[0], 1);
404        fill_rectangle(&sl->ref_cache[1][scan8[0]], 4, 4, 8, (uint8_t)ref[1], 1);
405        if (!IS_INTRA(mb_type_col[0]) && !sl->ref_list[1][0].parent->long_ref &&
406            ((l1ref0[0] == 0 &&
407              FFABS(l1mv0[0][0]) <= 1 &&
408              FFABS(l1mv0[0][1]) <= 1) ||
409             (l1ref0[0] < 0 && !l1ref1[0] &&
410              FFABS(l1mv1[0][0]) <= 1 &&
411              FFABS(l1mv1[0][1]) <= 1 &&
412              h->x264_build > 33U))) {
413            a = b = 0;
414            if (ref[0] > 0)
415                a = mv[0];
416            if (ref[1] > 0)
417                b = mv[1];
418        } else {
419            a = mv[0];
420            b = mv[1];
421        }
422        fill_rectangle(&sl->mv_cache[0][scan8[0]], 4, 4, 8, a, 4);
423        fill_rectangle(&sl->mv_cache[1][scan8[0]], 4, 4, 8, b, 4);
424    } else {
425        int n = 0;
426        for (i8 = 0; i8 < 4; i8++) {
427            const int x8 = i8 & 1;
428            const int y8 = i8 >> 1;
429
430            if (is_b8x8 && !IS_DIRECT(sl->sub_mb_type[i8]))
431                continue;
432            sl->sub_mb_type[i8] = sub_mb_type;
433
434            fill_rectangle(&sl->mv_cache[0][scan8[i8 * 4]], 2, 2, 8, mv[0], 4);
435            fill_rectangle(&sl->mv_cache[1][scan8[i8 * 4]], 2, 2, 8, mv[1], 4);
436            fill_rectangle(&sl->ref_cache[0][scan8[i8 * 4]], 2, 2, 8,
437                           (uint8_t)ref[0], 1);
438            fill_rectangle(&sl->ref_cache[1][scan8[i8 * 4]], 2, 2, 8,
439                           (uint8_t)ref[1], 1);
440
441            assert(b8_stride == 2);
442            /* col_zero_flag */
443            if (!IS_INTRA(mb_type_col[0]) && !sl->ref_list[1][0].parent->long_ref &&
444                (l1ref0[i8] == 0 ||
445                 (l1ref0[i8] < 0 &&
446                  l1ref1[i8] == 0 &&
447                  h->x264_build > 33U))) {
448                const int16_t (*l1mv)[2] = l1ref0[i8] == 0 ? l1mv0 : l1mv1;
449                if (IS_SUB_8X8(sub_mb_type)) {
450                    const int16_t *mv_col = l1mv[x8 * 3 + y8 * 3 * b4_stride];
451                    if (FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1) {
452                        if (ref[0] == 0)
453                            fill_rectangle(&sl->mv_cache[0][scan8[i8 * 4]], 2, 2,
454                                           8, 0, 4);
455                        if (ref[1] == 0)
456                            fill_rectangle(&sl->mv_cache[1][scan8[i8 * 4]], 2, 2,
457                                           8, 0, 4);
458                        n += 4;
459                    }
460                } else {
461                    int m = 0;
462                    for (i4 = 0; i4 < 4; i4++) {
463                        const int16_t *mv_col = l1mv[x8 * 2 + (i4 & 1) +
464                                                     (y8 * 2 + (i4 >> 1)) * b4_stride];
465                        if (FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1) {
466                            if (ref[0] == 0)
467                                AV_ZERO32(sl->mv_cache[0][scan8[i8 * 4 + i4]]);
468                            if (ref[1] == 0)
469                                AV_ZERO32(sl->mv_cache[1][scan8[i8 * 4 + i4]]);
470                            m++;
471                        }
472                    }
473                    if (!(m & 3))
474                        sl->sub_mb_type[i8] += MB_TYPE_16x16 - MB_TYPE_8x8;
475                    n += m;
476                }
477            }
478        }
479        if (!is_b8x8 && !(n & 15))
480            *mb_type = (*mb_type & ~(MB_TYPE_8x8 | MB_TYPE_16x8 | MB_TYPE_8x16 |
481                                     MB_TYPE_P1L0 | MB_TYPE_P1L1)) |
482                       MB_TYPE_16x16 | MB_TYPE_DIRECT2;
483    }
484}
485
486static void pred_temp_direct_motion(const H264Context *const h, H264SliceContext *sl,
487                                    int *mb_type)
488{
489    int b8_stride = 2;
490    int b4_stride = h->b_stride;
491    int mb_xy = sl->mb_xy, mb_y = sl->mb_y;
492    int mb_type_col[2];
493    const int16_t (*l1mv0)[2], (*l1mv1)[2];
494    const int8_t *l1ref0, *l1ref1;
495    const int is_b8x8 = IS_8X8(*mb_type);
496    unsigned int sub_mb_type;
497    int i8, i4;
498
499    assert(sl->ref_list[1][0].reference & 3);
500
501    await_reference_mb_row(h, &sl->ref_list[1][0],
502                           sl->mb_y + !!IS_INTERLACED(*mb_type));
503
504    if (IS_INTERLACED(sl->ref_list[1][0].parent->mb_type[mb_xy])) { // AFL/AFR/FR/FL -> AFL/FL
505        if (!IS_INTERLACED(*mb_type)) {                    //     AFR/FR    -> AFL/FL
506            mb_y  = (sl->mb_y & ~1) + sl->col_parity;
507            mb_xy = sl->mb_x +
508                    ((sl->mb_y & ~1) + sl->col_parity) * h->mb_stride;
509            b8_stride = 0;
510        } else {
511            mb_y  += sl->col_fieldoff;
512            mb_xy += h->mb_stride * sl->col_fieldoff; // non-zero for FL -> FL & differ parity
513        }
514        goto single_col;
515    } else {                                        // AFL/AFR/FR/FL -> AFR/FR
516        if (IS_INTERLACED(*mb_type)) {              // AFL       /FL -> AFR/FR
517            mb_y           = sl->mb_y & ~1;
518            mb_xy          = sl->mb_x + (sl->mb_y & ~1) * h->mb_stride;
519            mb_type_col[0] = sl->ref_list[1][0].parent->mb_type[mb_xy];
520            mb_type_col[1] = sl->ref_list[1][0].parent->mb_type[mb_xy + h->mb_stride];
521            b8_stride      = 2 + 4 * h->mb_stride;
522            b4_stride     *= 6;
523            if (IS_INTERLACED(mb_type_col[0]) !=
524                IS_INTERLACED(mb_type_col[1])) {
525                mb_type_col[0] &= ~MB_TYPE_INTERLACED;
526                mb_type_col[1] &= ~MB_TYPE_INTERLACED;
527            }
528
529            sub_mb_type = MB_TYPE_16x16 | MB_TYPE_P0L0 | MB_TYPE_P0L1 |
530                          MB_TYPE_DIRECT2;                  /* B_SUB_8x8 */
531
532            if ((mb_type_col[0] & MB_TYPE_16x16_OR_INTRA) &&
533                (mb_type_col[1] & MB_TYPE_16x16_OR_INTRA) &&
534                !is_b8x8) {
535                *mb_type |= MB_TYPE_16x8 | MB_TYPE_L0L1 |
536                            MB_TYPE_DIRECT2;                /* B_16x8 */
537            } else {
538                *mb_type |= MB_TYPE_8x8 | MB_TYPE_L0L1;
539            }
540        } else {                                    //     AFR/FR    -> AFR/FR
541single_col:
542            mb_type_col[0]     =
543                mb_type_col[1] = sl->ref_list[1][0].parent->mb_type[mb_xy];
544
545            sub_mb_type = MB_TYPE_16x16 | MB_TYPE_P0L0 | MB_TYPE_P0L1 |
546                          MB_TYPE_DIRECT2;                  /* B_SUB_8x8 */
547            if (!is_b8x8 && (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)) {
548                *mb_type |= MB_TYPE_16x16 | MB_TYPE_P0L0 | MB_TYPE_P0L1 |
549                            MB_TYPE_DIRECT2;                /* B_16x16 */
550            } else if (!is_b8x8 &&
551                       (mb_type_col[0] & (MB_TYPE_16x8 | MB_TYPE_8x16))) {
552                *mb_type |= MB_TYPE_L0L1 | MB_TYPE_DIRECT2 |
553                            (mb_type_col[0] & (MB_TYPE_16x8 | MB_TYPE_8x16));
554            } else {
555                if (!h->ps.sps->direct_8x8_inference_flag) {
556                    /* FIXME: save sub mb types from previous frames (or derive
557                     * from MVs) so we know exactly what block size to use */
558                    sub_mb_type = MB_TYPE_8x8 | MB_TYPE_P0L0 | MB_TYPE_P0L1 |
559                                  MB_TYPE_DIRECT2;          /* B_SUB_4x4 */
560                }
561                *mb_type |= MB_TYPE_8x8 | MB_TYPE_L0L1;
562            }
563        }
564    }
565
566    await_reference_mb_row(h, &sl->ref_list[1][0], mb_y);
567
568    l1mv0  = (void*)&sl->ref_list[1][0].parent->motion_val[0][h->mb2b_xy[mb_xy]];
569    l1mv1  = (void*)&sl->ref_list[1][0].parent->motion_val[1][h->mb2b_xy[mb_xy]];
570    l1ref0 = &sl->ref_list[1][0].parent->ref_index[0][4 * mb_xy];
571    l1ref1 = &sl->ref_list[1][0].parent->ref_index[1][4 * mb_xy];
572    if (!b8_stride) {
573        if (sl->mb_y & 1) {
574            l1ref0 += 2;
575            l1ref1 += 2;
576            l1mv0  += 2 * b4_stride;
577            l1mv1  += 2 * b4_stride;
578        }
579    }
580
581    {
582        const int *map_col_to_list0[2] = { sl->map_col_to_list0[0],
583                                           sl->map_col_to_list0[1] };
584        const int *dist_scale_factor = sl->dist_scale_factor;
585        int ref_offset;
586
587        if (FRAME_MBAFF(h) && IS_INTERLACED(*mb_type)) {
588            map_col_to_list0[0] = sl->map_col_to_list0_field[sl->mb_y & 1][0];
589            map_col_to_list0[1] = sl->map_col_to_list0_field[sl->mb_y & 1][1];
590            dist_scale_factor   = sl->dist_scale_factor_field[sl->mb_y & 1];
591        }
592        ref_offset = (sl->ref_list[1][0].parent->mbaff << 4) & (mb_type_col[0] >> 3);
593
594        if (IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col[0])) {
595            int y_shift = 2 * !IS_INTERLACED(*mb_type);
596            assert(h->ps.sps->direct_8x8_inference_flag);
597
598            for (i8 = 0; i8 < 4; i8++) {
599                const int x8 = i8 & 1;
600                const int y8 = i8 >> 1;
601                int ref0, scale;
602                const int16_t (*l1mv)[2] = l1mv0;
603
604                if (is_b8x8 && !IS_DIRECT(sl->sub_mb_type[i8]))
605                    continue;
606                sl->sub_mb_type[i8] = sub_mb_type;
607
608                fill_rectangle(&sl->ref_cache[1][scan8[i8 * 4]], 2, 2, 8, 0, 1);
609                if (IS_INTRA(mb_type_col[y8])) {
610                    fill_rectangle(&sl->ref_cache[0][scan8[i8 * 4]], 2, 2, 8, 0, 1);
611                    fill_rectangle(&sl->mv_cache[0][scan8[i8 * 4]], 2, 2, 8, 0, 4);
612                    fill_rectangle(&sl->mv_cache[1][scan8[i8 * 4]], 2, 2, 8, 0, 4);
613                    continue;
614                }
615
616                ref0 = l1ref0[x8 + y8 * b8_stride];
617                if (ref0 >= 0)
618                    ref0 = map_col_to_list0[0][ref0 + ref_offset];
619                else {
620                    ref0 = map_col_to_list0[1][l1ref1[x8 + y8 * b8_stride] +
621                                               ref_offset];
622                    l1mv = l1mv1;
623                }
624                scale = dist_scale_factor[ref0];
625                fill_rectangle(&sl->ref_cache[0][scan8[i8 * 4]], 2, 2, 8,
626                               ref0, 1);
627
628                {
629                    const int16_t *mv_col = l1mv[x8 * 3 + y8 * b4_stride];
630                    int my_col            = (mv_col[1] * (1 << y_shift)) / 2;
631                    int mx                = (scale * mv_col[0] + 128) >> 8;
632                    int my                = (scale * my_col    + 128) >> 8;
633                    fill_rectangle(&sl->mv_cache[0][scan8[i8 * 4]], 2, 2, 8,
634                                   pack16to32(mx, my), 4);
635                    fill_rectangle(&sl->mv_cache[1][scan8[i8 * 4]], 2, 2, 8,
636                                   pack16to32(mx - mv_col[0], my - my_col), 4);
637                }
638            }
639            return;
640        }
641
642        /* one-to-one mv scaling */
643
644        if (IS_16X16(*mb_type)) {
645            int ref, mv0, mv1;
646
647            fill_rectangle(&sl->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1);
648            if (IS_INTRA(mb_type_col[0])) {
649                ref = mv0 = mv1 = 0;
650            } else {
651                const int ref0 = l1ref0[0] >= 0 ? map_col_to_list0[0][l1ref0[0] + ref_offset]
652                                                : map_col_to_list0[1][l1ref1[0] + ref_offset];
653                const int scale = dist_scale_factor[ref0];
654                const int16_t *mv_col = l1ref0[0] >= 0 ? l1mv0[0] : l1mv1[0];
655                int mv_l0[2];
656                mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
657                mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
658                ref      = ref0;
659                mv0      = pack16to32(mv_l0[0], mv_l0[1]);
660                mv1      = pack16to32(mv_l0[0] - mv_col[0], mv_l0[1] - mv_col[1]);
661            }
662            fill_rectangle(&sl->ref_cache[0][scan8[0]], 4, 4, 8, ref, 1);
663            fill_rectangle(&sl->mv_cache[0][scan8[0]], 4, 4, 8, mv0, 4);
664            fill_rectangle(&sl->mv_cache[1][scan8[0]], 4, 4, 8, mv1, 4);
665        } else {
666            for (i8 = 0; i8 < 4; i8++) {
667                const int x8 = i8 & 1;
668                const int y8 = i8 >> 1;
669                int ref0, scale;
670                const int16_t (*l1mv)[2] = l1mv0;
671
672                if (is_b8x8 && !IS_DIRECT(sl->sub_mb_type[i8]))
673                    continue;
674                sl->sub_mb_type[i8] = sub_mb_type;
675                fill_rectangle(&sl->ref_cache[1][scan8[i8 * 4]], 2, 2, 8, 0, 1);
676                if (IS_INTRA(mb_type_col[0])) {
677                    fill_rectangle(&sl->ref_cache[0][scan8[i8 * 4]], 2, 2, 8, 0, 1);
678                    fill_rectangle(&sl->mv_cache[0][scan8[i8 * 4]], 2, 2, 8, 0, 4);
679                    fill_rectangle(&sl->mv_cache[1][scan8[i8 * 4]], 2, 2, 8, 0, 4);
680                    continue;
681                }
682
683                assert(b8_stride == 2);
684                ref0 = l1ref0[i8];
685                if (ref0 >= 0)
686                    ref0 = map_col_to_list0[0][ref0 + ref_offset];
687                else {
688                    ref0 = map_col_to_list0[1][l1ref1[i8] + ref_offset];
689                    l1mv = l1mv1;
690                }
691                scale = dist_scale_factor[ref0];
692
693                fill_rectangle(&sl->ref_cache[0][scan8[i8 * 4]], 2, 2, 8,
694                               ref0, 1);
695                if (IS_SUB_8X8(sub_mb_type)) {
696                    const int16_t *mv_col = l1mv[x8 * 3 + y8 * 3 * b4_stride];
697                    int mx                = (scale * mv_col[0] + 128) >> 8;
698                    int my                = (scale * mv_col[1] + 128) >> 8;
699                    fill_rectangle(&sl->mv_cache[0][scan8[i8 * 4]], 2, 2, 8,
700                                   pack16to32(mx, my), 4);
701                    fill_rectangle(&sl->mv_cache[1][scan8[i8 * 4]], 2, 2, 8,
702                                   pack16to32(mx - mv_col[0], my - mv_col[1]), 4);
703                } else {
704                    for (i4 = 0; i4 < 4; i4++) {
705                        const int16_t *mv_col = l1mv[x8 * 2 + (i4 & 1) +
706                                                     (y8 * 2 + (i4 >> 1)) * b4_stride];
707                        int16_t *mv_l0 = sl->mv_cache[0][scan8[i8 * 4 + i4]];
708                        mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
709                        mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
710                        AV_WN32A(sl->mv_cache[1][scan8[i8 * 4 + i4]],
711                                 pack16to32(mv_l0[0] - mv_col[0],
712                                            mv_l0[1] - mv_col[1]));
713                    }
714                }
715            }
716        }
717    }
718}
719
720void ff_h264_pred_direct_motion(const H264Context *const h, H264SliceContext *sl,
721                                int *mb_type)
722{
723    if (sl->direct_spatial_mv_pred)
724        pred_spatial_direct_motion(h, sl, mb_type);
725    else
726        pred_temp_direct_motion(h, sl, mb_type);
727}
728