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
get_scale_factor(H264SliceContext *sl, int poc, int poc1, int i)37 static 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
ff_h264_direct_dist_scale_factor(const H264Context *const h, H264SliceContext *sl)61 void 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
fill_colmap(const H264Context *h, H264SliceContext *sl, int map[2][16 + 32], int list, int field, int colfield, int mbafi)82 static 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
ff_h264_direct_ref_list_init(const H264Context *const h, H264SliceContext *sl)120 void 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
await_reference_mb_row(const H264Context *const h, H264Ref *ref, int mb_y)180 static 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
pred_spatial_direct_motion(const H264Context *const h, H264SliceContext *sl, int *mb_type)199 static 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
320 single_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
pred_temp_direct_motion(const H264Context *const h, H264SliceContext *sl, int *mb_type)486 static 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
541 single_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
ff_h264_pred_direct_motion(const H264Context *const h, H264SliceContext *sl, int *mb_type)720 void 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