| GCC Code Coverage Report | |||||||||||||||||||||
|
|||||||||||||||||||||
| Line | Branch | Exec | Source |
1 |
/* |
||
2 |
* H.26L/H.264/AVC/JVT/14496-10/... motion vector prediction |
||
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 motion vector prediction. |
||
25 |
* @author Michael Niedermayer <michaelni@gmx.at> |
||
26 |
*/ |
||
27 |
|||
28 |
#ifndef AVCODEC_H264_MVPRED_H |
||
29 |
#define AVCODEC_H264_MVPRED_H |
||
30 |
|||
31 |
#include "internal.h" |
||
32 |
#include "avcodec.h" |
||
33 |
#include "h264dec.h" |
||
34 |
#include "mpegutils.h" |
||
35 |
#include "libavutil/avassert.h" |
||
36 |
#include "libavutil/mem_internal.h" |
||
37 |
|||
38 |
|||
39 |
17648015 |
static av_always_inline int fetch_diagonal_mv(const H264Context *h, H264SliceContext *sl, |
|
40 |
const int16_t **C, |
||
41 |
int i, int list, int part_width) |
||
42 |
{ |
||
43 |
17648015 |
const int topright_ref = sl->ref_cache[list][i - 8 + part_width]; |
|
44 |
|||
45 |
/* there is no consistent mapping of mvs to neighboring locations that will |
||
46 |
* make mbaff happy, so we can't move all this logic to fill_caches */ |
||
47 |
✓✓ | 17648015 |
if (FRAME_MBAFF(h)) { |
48 |
#define SET_DIAG_MV(MV_OP, REF_OP, XY, Y4) \ |
||
49 |
const int xy = XY, y4 = Y4; \ |
||
50 |
const int mb_type = mb_types[xy + (y4 >> 2) * h->mb_stride]; \ |
||
51 |
if (!USES_LIST(mb_type, list)) \ |
||
52 |
return LIST_NOT_USED; \ |
||
53 |
mv = h->cur_pic_ptr->motion_val[list][h->mb2b_xy[xy] + 3 + y4 * h->b_stride]; \ |
||
54 |
sl->mv_cache[list][scan8[0] - 2][0] = mv[0]; \ |
||
55 |
sl->mv_cache[list][scan8[0] - 2][1] = mv[1] MV_OP; \ |
||
56 |
return h->cur_pic_ptr->ref_index[list][4 * xy + 1 + (y4 & ~1)] REF_OP; |
||
57 |
|||
58 |
✓✓ | 2891570 |
if (topright_ref == PART_NOT_AVAILABLE |
59 |
✓✓✓✓ |
1265392 |
&& i >= scan8[0] + 8 && (i & 7) == 4 |
60 |
✓✓ | 196840 |
&& sl->ref_cache[list][scan8[0] - 1] != PART_NOT_AVAILABLE) { |
61 |
188678 |
const uint32_t *mb_types = h->cur_pic_ptr->mb_type; |
|
62 |
const int16_t *mv; |
||
63 |
188678 |
AV_ZERO32(sl->mv_cache[list][scan8[0] - 2]); |
|
64 |
188678 |
*C = sl->mv_cache[list][scan8[0] - 2]; |
|
65 |
|||
66 |
✓✓✓✓ |
188678 |
if (!MB_FIELD(sl) && IS_INTERLACED(sl->left_type[0])) { |
67 |
✓✓ | 26640 |
SET_DIAG_MV(* 2, >> 1, sl->left_mb_xy[0] + h->mb_stride, |
68 |
(sl->mb_y & 1) * 2 + (i >> 5)); |
||
69 |
} |
||
70 |
✓✓✓✓ |
162038 |
if (MB_FIELD(sl) && !IS_INTERLACED(sl->left_type[0])) { |
71 |
// left shift will turn LIST_NOT_USED into PART_NOT_AVAILABLE, but that's OK. |
||
72 |
✓✓ | 35582 |
SET_DIAG_MV(/ 2, *2, sl->left_mb_xy[i >= 36], ((i >> 2)) & 3); |
73 |
} |
||
74 |
} |
||
75 |
#undef SET_DIAG_MV |
||
76 |
} |
||
77 |
|||
78 |
✓✓ | 17585793 |
if (topright_ref != PART_NOT_AVAILABLE) { |
79 |
11057418 |
*C = sl->mv_cache[list][i - 8 + part_width]; |
|
80 |
11057418 |
return topright_ref; |
|
81 |
} else { |
||
82 |
ff_tlog(h->avctx, "topright MV not available\n"); |
||
83 |
|||
84 |
6528375 |
*C = sl->mv_cache[list][i - 8 - 1]; |
|
85 |
6528375 |
return sl->ref_cache[list][i - 8 - 1]; |
|
86 |
} |
||
87 |
} |
||
88 |
|||
89 |
/** |
||
90 |
* Get the predicted MV. |
||
91 |
* @param n the block index |
||
92 |
* @param part_width the width of the partition (4, 8,16) -> (1, 2, 4) |
||
93 |
* @param mx the x component of the predicted motion vector |
||
94 |
* @param my the y component of the predicted motion vector |
||
95 |
*/ |
||
96 |
17055649 |
static av_always_inline void pred_motion(const H264Context *const h, |
|
97 |
H264SliceContext *sl, |
||
98 |
int n, |
||
99 |
int part_width, int list, int ref, |
||
100 |
int *const mx, int *const my) |
||
101 |
{ |
||
102 |
17055649 |
const int index8 = scan8[n]; |
|
103 |
17055649 |
const int top_ref = sl->ref_cache[list][index8 - 8]; |
|
104 |
17055649 |
const int left_ref = sl->ref_cache[list][index8 - 1]; |
|
105 |
17055649 |
const int16_t *const A = sl->mv_cache[list][index8 - 1]; |
|
106 |
17055649 |
const int16_t *const B = sl->mv_cache[list][index8 - 8]; |
|
107 |
const int16_t *C; |
||
108 |
int diagonal_ref, match_count; |
||
109 |
|||
110 |
av_assert2(part_width == 1 || part_width == 2 || part_width == 4); |
||
111 |
|||
112 |
/* mv_cache |
||
113 |
* B . . A T T T T |
||
114 |
* U . . L . . , . |
||
115 |
* U . . L . . . . |
||
116 |
* U . . L . . , . |
||
117 |
* . . . L . . . . |
||
118 |
*/ |
||
119 |
|||
120 |
17055649 |
diagonal_ref = fetch_diagonal_mv(h, sl, &C, index8, list, part_width); |
|
121 |
17055649 |
match_count = (diagonal_ref == ref) + (top_ref == ref) + (left_ref == ref); |
|
122 |
ff_tlog(h->avctx, "pred_motion match_count=%d\n", match_count); |
||
123 |
✓✓ | 17055649 |
if (match_count > 1) { //most common |
124 |
12589498 |
*mx = mid_pred(A[0], B[0], C[0]); |
|
125 |
12589498 |
*my = mid_pred(A[1], B[1], C[1]); |
|
126 |
✓✓ | 4466151 |
} else if (match_count == 1) { |
127 |
✓✓ | 3035177 |
if (left_ref == ref) { |
128 |
1910046 |
*mx = A[0]; |
|
129 |
1910046 |
*my = A[1]; |
|
130 |
✓✓ | 1125131 |
} else if (top_ref == ref) { |
131 |
808806 |
*mx = B[0]; |
|
132 |
808806 |
*my = B[1]; |
|
133 |
} else { |
||
134 |
316325 |
*mx = C[0]; |
|
135 |
316325 |
*my = C[1]; |
|
136 |
} |
||
137 |
} else { |
||
138 |
✓✓✓✓ |
1430974 |
if (top_ref == PART_NOT_AVAILABLE && |
139 |
✓✓ | 232958 |
diagonal_ref == PART_NOT_AVAILABLE && |
140 |
left_ref != PART_NOT_AVAILABLE) { |
||
141 |
207954 |
*mx = A[0]; |
|
142 |
207954 |
*my = A[1]; |
|
143 |
} else { |
||
144 |
1223020 |
*mx = mid_pred(A[0], B[0], C[0]); |
|
145 |
1223020 |
*my = mid_pred(A[1], B[1], C[1]); |
|
146 |
} |
||
147 |
} |
||
148 |
|||
149 |
ff_tlog(h->avctx, |
||
150 |
"pred_motion (%2d %2d %2d) (%2d %2d %2d) (%2d %2d %2d) -> (%2d %2d %2d) at %2d %2d %d list %d\n", |
||
151 |
top_ref, B[0], B[1], diagonal_ref, C[0], C[1], left_ref, |
||
152 |
A[0], A[1], ref, *mx, *my, sl->mb_x, sl->mb_y, n, list); |
||
153 |
17055649 |
} |
|
154 |
|||
155 |
/** |
||
156 |
* Get the directionally predicted 16x8 MV. |
||
157 |
* @param n the block index |
||
158 |
* @param mx the x component of the predicted motion vector |
||
159 |
* @param my the y component of the predicted motion vector |
||
160 |
*/ |
||
161 |
1341285 |
static av_always_inline void pred_16x8_motion(const H264Context *const h, |
|
162 |
H264SliceContext *sl, |
||
163 |
int n, int list, int ref, |
||
164 |
int *const mx, int *const my) |
||
165 |
{ |
||
166 |
✓✓ | 1341285 |
if (n == 0) { |
167 |
669802 |
const int top_ref = sl->ref_cache[list][scan8[0] - 8]; |
|
168 |
669802 |
const int16_t *const B = sl->mv_cache[list][scan8[0] - 8]; |
|
169 |
|||
170 |
ff_tlog(h->avctx, "pred_16x8: (%2d %2d %2d) at %2d %2d %d list %d\n", |
||
171 |
top_ref, B[0], B[1], sl->mb_x, sl->mb_y, n, list); |
||
172 |
|||
173 |
✓✓ | 669802 |
if (top_ref == ref) { |
174 |
378948 |
*mx = B[0]; |
|
175 |
378948 |
*my = B[1]; |
|
176 |
378948 |
return; |
|
177 |
} |
||
178 |
} else { |
||
179 |
671483 |
const int left_ref = sl->ref_cache[list][scan8[8] - 1]; |
|
180 |
671483 |
const int16_t *const A = sl->mv_cache[list][scan8[8] - 1]; |
|
181 |
|||
182 |
ff_tlog(h->avctx, "pred_16x8: (%2d %2d %2d) at %2d %2d %d list %d\n", |
||
183 |
left_ref, A[0], A[1], sl->mb_x, sl->mb_y, n, list); |
||
184 |
|||
185 |
✓✓ | 671483 |
if (left_ref == ref) { |
186 |
449579 |
*mx = A[0]; |
|
187 |
449579 |
*my = A[1]; |
|
188 |
449579 |
return; |
|
189 |
} |
||
190 |
} |
||
191 |
|||
192 |
//RARE |
||
193 |
512758 |
pred_motion(h, sl, n, 4, list, ref, mx, my); |
|
194 |
} |
||
195 |
|||
196 |
/** |
||
197 |
* Get the directionally predicted 8x16 MV. |
||
198 |
* @param n the block index |
||
199 |
* @param mx the x component of the predicted motion vector |
||
200 |
* @param my the y component of the predicted motion vector |
||
201 |
*/ |
||
202 |
1186049 |
static av_always_inline void pred_8x16_motion(const H264Context *const h, |
|
203 |
H264SliceContext *sl, |
||
204 |
int n, int list, int ref, |
||
205 |
int *const mx, int *const my) |
||
206 |
{ |
||
207 |
✓✓ | 1186049 |
if (n == 0) { |
208 |
593683 |
const int left_ref = sl->ref_cache[list][scan8[0] - 1]; |
|
209 |
593683 |
const int16_t *const A = sl->mv_cache[list][scan8[0] - 1]; |
|
210 |
|||
211 |
ff_tlog(h->avctx, "pred_8x16: (%2d %2d %2d) at %2d %2d %d list %d\n", |
||
212 |
left_ref, A[0], A[1], sl->mb_x, sl->mb_y, n, list); |
||
213 |
|||
214 |
✓✓ | 593683 |
if (left_ref == ref) { |
215 |
369982 |
*mx = A[0]; |
|
216 |
369982 |
*my = A[1]; |
|
217 |
369982 |
return; |
|
218 |
} |
||
219 |
} else { |
||
220 |
const int16_t *C; |
||
221 |
int diagonal_ref; |
||
222 |
|||
223 |
592366 |
diagonal_ref = fetch_diagonal_mv(h, sl, &C, scan8[4], list, 2); |
|
224 |
|||
225 |
ff_tlog(h->avctx, "pred_8x16: (%2d %2d %2d) at %2d %2d %d list %d\n", |
||
226 |
diagonal_ref, C[0], C[1], sl->mb_x, sl->mb_y, n, list); |
||
227 |
|||
228 |
✓✓ | 592366 |
if (diagonal_ref == ref) { |
229 |
337481 |
*mx = C[0]; |
|
230 |
337481 |
*my = C[1]; |
|
231 |
337481 |
return; |
|
232 |
} |
||
233 |
} |
||
234 |
|||
235 |
//RARE |
||
236 |
478586 |
pred_motion(h, sl, n, 2, list, ref, mx, my); |
|
237 |
} |
||
238 |
|||
239 |
#define FIX_MV_MBAFF(type, refn, mvn, idx) \ |
||
240 |
if (FRAME_MBAFF(h)) { \ |
||
241 |
if (MB_FIELD(sl)) { \ |
||
242 |
if (!IS_INTERLACED(type)) { \ |
||
243 |
refn <<= 1; \ |
||
244 |
AV_COPY32(mvbuf[idx], mvn); \ |
||
245 |
mvbuf[idx][1] /= 2; \ |
||
246 |
mvn = mvbuf[idx]; \ |
||
247 |
} \ |
||
248 |
} else { \ |
||
249 |
if (IS_INTERLACED(type)) { \ |
||
250 |
refn >>= 1; \ |
||
251 |
AV_COPY32(mvbuf[idx], mvn); \ |
||
252 |
mvbuf[idx][1] *= 2; \ |
||
253 |
mvn = mvbuf[idx]; \ |
||
254 |
} \ |
||
255 |
} \ |
||
256 |
} |
||
257 |
|||
258 |
1392779 |
static av_always_inline void pred_pskip_motion(const H264Context *const h, |
|
259 |
H264SliceContext *sl) |
||
260 |
{ |
||
261 |
DECLARE_ALIGNED(4, static const int16_t, zeromv)[2] = { 0 }; |
||
262 |
DECLARE_ALIGNED(4, int16_t, mvbuf)[3][2]; |
||
263 |
1392779 |
int8_t *ref = h->cur_pic.ref_index[0]; |
|
264 |
1392779 |
int16_t(*mv)[2] = h->cur_pic.motion_val[0]; |
|
265 |
int top_ref, left_ref, diagonal_ref, match_count, mx, my; |
||
266 |
const int16_t *A, *B, *C; |
||
267 |
1392779 |
int b_stride = h->b_stride; |
|
268 |
|||
269 |
1392779 |
fill_rectangle(&sl->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1); |
|
270 |
|||
271 |
/* To avoid doing an entire fill_decode_caches, we inline the relevant |
||
272 |
* parts here. |
||
273 |
* FIXME: this is a partial duplicate of the logic in fill_decode_caches, |
||
274 |
* but it's faster this way. Is there a way to avoid this duplication? |
||
275 |
*/ |
||
276 |
✓✓ | 1392779 |
if (USES_LIST(sl->left_type[LTOP], 0)) { |
277 |
1315143 |
left_ref = ref[4 * sl->left_mb_xy[LTOP] + 1 + (sl->left_block[0] & ~1)]; |
|
278 |
1315143 |
A = mv[h->mb2b_xy[sl->left_mb_xy[LTOP]] + 3 + b_stride * sl->left_block[0]]; |
|
279 |
✓✓✓✓ ✓✓✓✓ |
1315143 |
FIX_MV_MBAFF(sl->left_type[LTOP], left_ref, A, 0); |
280 |
✓✓ | 1315143 |
if (!(left_ref | AV_RN32A(A))) |
281 |
1038558 |
goto zeromv; |
|
282 |
✓✓ | 77636 |
} else if (sl->left_type[LTOP]) { |
283 |
31542 |
left_ref = LIST_NOT_USED; |
|
284 |
31542 |
A = zeromv; |
|
285 |
} else { |
||
286 |
46094 |
goto zeromv; |
|
287 |
} |
||
288 |
|||
289 |
✓✓ | 308127 |
if (USES_LIST(sl->top_type, 0)) { |
290 |
279288 |
top_ref = ref[4 * sl->top_mb_xy + 2]; |
|
291 |
279288 |
B = mv[h->mb2b_xy[sl->top_mb_xy] + 3 * b_stride]; |
|
292 |
✓✓✓✓ ✓✓✓✓ |
279288 |
FIX_MV_MBAFF(sl->top_type, top_ref, B, 1); |
293 |
✓✓ | 279288 |
if (!(top_ref | AV_RN32A(B))) |
294 |
46338 |
goto zeromv; |
|
295 |
✓✓ | 28839 |
} else if (sl->top_type) { |
296 |
19032 |
top_ref = LIST_NOT_USED; |
|
297 |
19032 |
B = zeromv; |
|
298 |
} else { |
||
299 |
9807 |
goto zeromv; |
|
300 |
} |
||
301 |
|||
302 |
ff_tlog(h->avctx, "pred_pskip: (%d) (%d) at %2d %2d\n", |
||
303 |
top_ref, left_ref, sl->mb_x, sl->mb_y); |
||
304 |
|||
305 |
✓✓ | 251982 |
if (USES_LIST(sl->topright_type, 0)) { |
306 |
213404 |
diagonal_ref = ref[4 * sl->topright_mb_xy + 2]; |
|
307 |
213404 |
C = mv[h->mb2b_xy[sl->topright_mb_xy] + 3 * b_stride]; |
|
308 |
✓✓✓✓ ✓✓✓✓ |
213404 |
FIX_MV_MBAFF(sl->topright_type, diagonal_ref, C, 2); |
309 |
✓✓ | 38578 |
} else if (sl->topright_type) { |
310 |
18754 |
diagonal_ref = LIST_NOT_USED; |
|
311 |
18754 |
C = zeromv; |
|
312 |
} else { |
||
313 |
✓✓ | 19824 |
if (USES_LIST(sl->topleft_type, 0)) { |
314 |
18841 |
diagonal_ref = ref[4 * sl->topleft_mb_xy + 1 + |
|
315 |
18841 |
(sl->topleft_partition & 2)]; |
|
316 |
18841 |
C = mv[h->mb2b_xy[sl->topleft_mb_xy] + 3 + b_stride + |
|
317 |
18841 |
(sl->topleft_partition & 2 * b_stride)]; |
|
318 |
✓✓✓✓ ✓✓✓✓ |
18841 |
FIX_MV_MBAFF(sl->topleft_type, diagonal_ref, C, 2); |
319 |
✓✗ | 983 |
} else if (sl->topleft_type) { |
320 |
983 |
diagonal_ref = LIST_NOT_USED; |
|
321 |
983 |
C = zeromv; |
|
322 |
} else { |
||
323 |
diagonal_ref = PART_NOT_AVAILABLE; |
||
324 |
C = zeromv; |
||
325 |
} |
||
326 |
} |
||
327 |
|||
328 |
251982 |
match_count = !diagonal_ref + !top_ref + !left_ref; |
|
329 |
ff_tlog(h->avctx, "pred_pskip_motion match_count=%d\n", match_count); |
||
330 |
✓✓ | 251982 |
if (match_count > 1) { |
331 |
209731 |
mx = mid_pred(A[0], B[0], C[0]); |
|
332 |
209731 |
my = mid_pred(A[1], B[1], C[1]); |
|
333 |
✓✓ | 42251 |
} else if (match_count == 1) { |
334 |
✓✓ | 36162 |
if (!left_ref) { |
335 |
17561 |
mx = A[0]; |
|
336 |
17561 |
my = A[1]; |
|
337 |
✓✓ | 18601 |
} else if (!top_ref) { |
338 |
8127 |
mx = B[0]; |
|
339 |
8127 |
my = B[1]; |
|
340 |
} else { |
||
341 |
10474 |
mx = C[0]; |
|
342 |
10474 |
my = C[1]; |
|
343 |
} |
||
344 |
} else { |
||
345 |
6089 |
mx = mid_pred(A[0], B[0], C[0]); |
|
346 |
6089 |
my = mid_pred(A[1], B[1], C[1]); |
|
347 |
} |
||
348 |
|||
349 |
251982 |
fill_rectangle(sl->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx, my), 4); |
|
350 |
251982 |
return; |
|
351 |
|||
352 |
1140797 |
zeromv: |
|
353 |
1140797 |
fill_rectangle(sl->mv_cache[0][scan8[0]], 4, 4, 8, 0, 4); |
|
354 |
1140797 |
return; |
|
355 |
} |
||
356 |
|||
357 |
12689910 |
static void fill_decode_neighbors(const H264Context *h, H264SliceContext *sl, int mb_type) |
|
358 |
{ |
||
359 |
12689910 |
const int mb_xy = sl->mb_xy; |
|
360 |
int topleft_xy, top_xy, topright_xy, left_xy[LEFT_MBS]; |
||
361 |
static const uint8_t left_block_options[4][32] = { |
||
362 |
{ 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 }, |
||
363 |
{ 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 }, |
||
364 |
{ 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 }, |
||
365 |
{ 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 } |
||
366 |
}; |
||
367 |
|||
368 |
12689910 |
sl->topleft_partition = -1; |
|
369 |
|||
370 |
12689910 |
top_xy = mb_xy - (h->mb_stride << MB_FIELD(sl)); |
|
371 |
|||
372 |
/* Wow, what a mess, why didn't they simplify the interlacing & intra |
||
373 |
* stuff, I can't imagine that these complex rules are worth it. */ |
||
374 |
|||
375 |
12689910 |
topleft_xy = top_xy - 1; |
|
376 |
12689910 |
topright_xy = top_xy + 1; |
|
377 |
12689910 |
left_xy[LBOT] = left_xy[LTOP] = mb_xy - 1; |
|
378 |
12689910 |
sl->left_block = left_block_options[0]; |
|
379 |
✓✓ | 12689910 |
if (FRAME_MBAFF(h)) { |
380 |
2095408 |
const int left_mb_field_flag = IS_INTERLACED(h->cur_pic.mb_type[mb_xy - 1]); |
|
381 |
2095408 |
const int curr_mb_field_flag = IS_INTERLACED(mb_type); |
|
382 |
✓✓ | 2095408 |
if (sl->mb_y & 1) { |
383 |
✓✓ | 1047592 |
if (left_mb_field_flag != curr_mb_field_flag) { |
384 |
247902 |
left_xy[LBOT] = left_xy[LTOP] = mb_xy - h->mb_stride - 1; |
|
385 |
✓✓ | 247902 |
if (curr_mb_field_flag) { |
386 |
127965 |
left_xy[LBOT] += h->mb_stride; |
|
387 |
127965 |
sl->left_block = left_block_options[3]; |
|
388 |
} else { |
||
389 |
119937 |
topleft_xy += h->mb_stride; |
|
390 |
/* take top left mv from the middle of the mb, as opposed |
||
391 |
* to all other modes which use the bottom right partition */ |
||
392 |
119937 |
sl->topleft_partition = 0; |
|
393 |
119937 |
sl->left_block = left_block_options[1]; |
|
394 |
} |
||
395 |
} |
||
396 |
} else { |
||
397 |
✓✓ | 1047816 |
if (curr_mb_field_flag) { |
398 |
305129 |
topleft_xy += h->mb_stride & (((h->cur_pic.mb_type[top_xy - 1] >> 7) & 1) - 1); |
|
399 |
305129 |
topright_xy += h->mb_stride & (((h->cur_pic.mb_type[top_xy + 1] >> 7) & 1) - 1); |
|
400 |
305129 |
top_xy += h->mb_stride & (((h->cur_pic.mb_type[top_xy] >> 7) & 1) - 1); |
|
401 |
} |
||
402 |
✓✓ | 1047816 |
if (left_mb_field_flag != curr_mb_field_flag) { |
403 |
✓✓ | 248076 |
if (curr_mb_field_flag) { |
404 |
127938 |
left_xy[LBOT] += h->mb_stride; |
|
405 |
127938 |
sl->left_block = left_block_options[3]; |
|
406 |
} else { |
||
407 |
120138 |
sl->left_block = left_block_options[2]; |
|
408 |
} |
||
409 |
} |
||
410 |
} |
||
411 |
} |
||
412 |
|||
413 |
12689910 |
sl->topleft_mb_xy = topleft_xy; |
|
414 |
12689910 |
sl->top_mb_xy = top_xy; |
|
415 |
12689910 |
sl->topright_mb_xy = topright_xy; |
|
416 |
12689910 |
sl->left_mb_xy[LTOP] = left_xy[LTOP]; |
|
417 |
12689910 |
sl->left_mb_xy[LBOT] = left_xy[LBOT]; |
|
418 |
//FIXME do we need all in the context? |
||
419 |
|||
420 |
12689910 |
sl->topleft_type = h->cur_pic.mb_type[topleft_xy]; |
|
421 |
12689910 |
sl->top_type = h->cur_pic.mb_type[top_xy]; |
|
422 |
12689910 |
sl->topright_type = h->cur_pic.mb_type[topright_xy]; |
|
423 |
12689910 |
sl->left_type[LTOP] = h->cur_pic.mb_type[left_xy[LTOP]]; |
|
424 |
12689910 |
sl->left_type[LBOT] = h->cur_pic.mb_type[left_xy[LBOT]]; |
|
425 |
|||
426 |
if (FMO) { |
||
427 |
if (h->slice_table[topleft_xy] != sl->slice_num) |
||
428 |
sl->topleft_type = 0; |
||
429 |
if (h->slice_table[top_xy] != sl->slice_num) |
||
430 |
sl->top_type = 0; |
||
431 |
if (h->slice_table[left_xy[LTOP]] != sl->slice_num) |
||
432 |
sl->left_type[LTOP] = sl->left_type[LBOT] = 0; |
||
433 |
} else { |
||
434 |
✓✓ | 12689910 |
if (h->slice_table[topleft_xy] != sl->slice_num) { |
435 |
1319531 |
sl->topleft_type = 0; |
|
436 |
✓✓ | 1319531 |
if (h->slice_table[top_xy] != sl->slice_num) |
437 |
980187 |
sl->top_type = 0; |
|
438 |
✓✓ | 1319531 |
if (h->slice_table[left_xy[LTOP]] != sl->slice_num) |
439 |
373085 |
sl->left_type[LTOP] = sl->left_type[LBOT] = 0; |
|
440 |
} |
||
441 |
} |
||
442 |
✓✓ | 12689910 |
if (h->slice_table[topright_xy] != sl->slice_num) |
443 |
2041199 |
sl->topright_type = 0; |
|
444 |
12689910 |
} |
|
445 |
|||
446 |
11291472 |
static void fill_decode_caches(const H264Context *h, H264SliceContext *sl, int mb_type) |
|
447 |
{ |
||
448 |
int topleft_xy, top_xy, topright_xy, left_xy[LEFT_MBS]; |
||
449 |
int topleft_type, top_type, topright_type, left_type[LEFT_MBS]; |
||
450 |
11291472 |
const uint8_t *left_block = sl->left_block; |
|
451 |
int i; |
||
452 |
uint8_t *nnz; |
||
453 |
uint8_t *nnz_cache; |
||
454 |
|||
455 |
11291472 |
topleft_xy = sl->topleft_mb_xy; |
|
456 |
11291472 |
top_xy = sl->top_mb_xy; |
|
457 |
11291472 |
topright_xy = sl->topright_mb_xy; |
|
458 |
11291472 |
left_xy[LTOP] = sl->left_mb_xy[LTOP]; |
|
459 |
11291472 |
left_xy[LBOT] = sl->left_mb_xy[LBOT]; |
|
460 |
11291472 |
topleft_type = sl->topleft_type; |
|
461 |
11291472 |
top_type = sl->top_type; |
|
462 |
11291472 |
topright_type = sl->topright_type; |
|
463 |
11291472 |
left_type[LTOP] = sl->left_type[LTOP]; |
|
464 |
11291472 |
left_type[LBOT] = sl->left_type[LBOT]; |
|
465 |
|||
466 |
✓✓ | 11291472 |
if (!IS_SKIP(mb_type)) { |
467 |
✓✓ | 9857219 |
if (IS_INTRA(mb_type)) { |
468 |
✓✓ | 3738958 |
int type_mask = h->ps.pps->constrained_intra_pred ? IS_INTRA(-1) : -1; |
469 |
3738958 |
sl->topleft_samples_available = |
|
470 |
3738958 |
sl->top_samples_available = |
|
471 |
3738958 |
sl->left_samples_available = 0xFFFF; |
|
472 |
3738958 |
sl->topright_samples_available = 0xEEEA; |
|
473 |
|||
474 |
✓✓ | 3738958 |
if (!(top_type & type_mask)) { |
475 |
268827 |
sl->topleft_samples_available = 0xB3FF; |
|
476 |
268827 |
sl->top_samples_available = 0x33FF; |
|
477 |
268827 |
sl->topright_samples_available = 0x26EA; |
|
478 |
} |
||
479 |
✓✓ | 3738958 |
if (IS_INTERLACED(mb_type) != IS_INTERLACED(left_type[LTOP])) { |
480 |
✓✓ | 211987 |
if (IS_INTERLACED(mb_type)) { |
481 |
✓✓ | 110440 |
if (!(left_type[LTOP] & type_mask)) { |
482 |
15628 |
sl->topleft_samples_available &= 0xDFFF; |
|
483 |
15628 |
sl->left_samples_available &= 0x5FFF; |
|
484 |
} |
||
485 |
✓✓ | 110440 |
if (!(left_type[LBOT] & type_mask)) { |
486 |
15633 |
sl->topleft_samples_available &= 0xFF5F; |
|
487 |
15633 |
sl->left_samples_available &= 0xFF5F; |
|
488 |
} |
||
489 |
} else { |
||
490 |
101547 |
int left_typei = h->cur_pic.mb_type[left_xy[LTOP] + h->mb_stride]; |
|
491 |
|||
492 |
av_assert2(left_xy[LTOP] == left_xy[LBOT]); |
||
493 |
✓✓✓✓ |
101547 |
if (!((left_typei & type_mask) && (left_type[LTOP] & type_mask))) { |
494 |
53 |
sl->topleft_samples_available &= 0xDF5F; |
|
495 |
53 |
sl->left_samples_available &= 0x5F5F; |
|
496 |
} |
||
497 |
} |
||
498 |
} else { |
||
499 |
✓✓ | 3526971 |
if (!(left_type[LTOP] & type_mask)) { |
500 |
64367 |
sl->topleft_samples_available &= 0xDF5F; |
|
501 |
64367 |
sl->left_samples_available &= 0x5F5F; |
|
502 |
} |
||
503 |
} |
||
504 |
|||
505 |
✓✓ | 3738958 |
if (!(topleft_type & type_mask)) |
506 |
340157 |
sl->topleft_samples_available &= 0x7FFF; |
|
507 |
|||
508 |
✓✓ | 3738958 |
if (!(topright_type & type_mask)) |
509 |
706496 |
sl->topright_samples_available &= 0xFBFF; |
|
510 |
|||
511 |
✓✓ | 3738958 |
if (IS_INTRA4x4(mb_type)) { |
512 |
✓✓ | 2884335 |
if (IS_INTRA4x4(top_type)) { |
513 |
2156403 |
AV_COPY32(sl->intra4x4_pred_mode_cache + 4 + 8 * 0, sl->intra4x4_pred_mode + h->mb2br_xy[top_xy]); |
|
514 |
} else { |
||
515 |
727932 |
sl->intra4x4_pred_mode_cache[4 + 8 * 0] = |
|
516 |
727932 |
sl->intra4x4_pred_mode_cache[5 + 8 * 0] = |
|
517 |
727932 |
sl->intra4x4_pred_mode_cache[6 + 8 * 0] = |
|
518 |
✓✓ | 727932 |
sl->intra4x4_pred_mode_cache[7 + 8 * 0] = 2 - 3 * !(top_type & type_mask); |
519 |
} |
||
520 |
✓✓ | 8653005 |
for (i = 0; i < 2; i++) { |
521 |
✓✓ | 5768670 |
if (IS_INTRA4x4(left_type[LEFT(i)])) { |
522 |
4623036 |
int8_t *mode = sl->intra4x4_pred_mode + h->mb2br_xy[left_xy[LEFT(i)]]; |
|
523 |
4623036 |
sl->intra4x4_pred_mode_cache[3 + 8 * 1 + 2 * 8 * i] = mode[6 - left_block[0 + 2 * i]]; |
|
524 |
4623036 |
sl->intra4x4_pred_mode_cache[3 + 8 * 2 + 2 * 8 * i] = mode[6 - left_block[1 + 2 * i]]; |
|
525 |
} else { |
||
526 |
1145634 |
sl->intra4x4_pred_mode_cache[3 + 8 * 1 + 2 * 8 * i] = |
|
527 |
✓✓ | 1145634 |
sl->intra4x4_pred_mode_cache[3 + 8 * 2 + 2 * 8 * i] = 2 - 3 * !(left_type[LEFT(i)] & type_mask); |
528 |
} |
||
529 |
} |
||
530 |
} |
||
531 |
} |
||
532 |
|||
533 |
/* |
||
534 |
* 0 . T T. T T T T |
||
535 |
* 1 L . .L . . . . |
||
536 |
* 2 L . .L . . . . |
||
537 |
* 3 . T TL . . . . |
||
538 |
* 4 L . .L . . . . |
||
539 |
* 5 L . .. . . . . |
||
540 |
*/ |
||
541 |
/* FIXME: constraint_intra_pred & partitioning & nnz |
||
542 |
* (let us hope this is just a typo in the spec) */ |
||
543 |
9857219 |
nnz_cache = sl->non_zero_count_cache; |
|
544 |
✓✓ | 9857219 |
if (top_type) { |
545 |
9051194 |
nnz = h->non_zero_count[top_xy]; |
|
546 |
9051194 |
AV_COPY32(&nnz_cache[4 + 8 * 0], &nnz[4 * 3]); |
|
547 |
✓✓ | 9051194 |
if (!h->chroma_y_shift) { |
548 |
854774 |
AV_COPY32(&nnz_cache[4 + 8 * 5], &nnz[4 * 7]); |
|
549 |
854774 |
AV_COPY32(&nnz_cache[4 + 8 * 10], &nnz[4 * 11]); |
|
550 |
} else { |
||
551 |
8196420 |
AV_COPY32(&nnz_cache[4 + 8 * 5], &nnz[4 * 5]); |
|
552 |
8196420 |
AV_COPY32(&nnz_cache[4 + 8 * 10], &nnz[4 * 9]); |
|
553 |
} |
||
554 |
} else { |
||
555 |
✓✓✗✗ |
806025 |
uint32_t top_empty = CABAC(h) && !IS_INTRA(mb_type) ? 0 : 0x40404040; |
556 |
806025 |
AV_WN32A(&nnz_cache[4 + 8 * 0], top_empty); |
|
557 |
806025 |
AV_WN32A(&nnz_cache[4 + 8 * 5], top_empty); |
|
558 |
806025 |
AV_WN32A(&nnz_cache[4 + 8 * 10], top_empty); |
|
559 |
} |
||
560 |
|||
561 |
✓✓ | 29571657 |
for (i = 0; i < 2; i++) { |
562 |
✓✓ | 19714438 |
if (left_type[LEFT(i)]) { |
563 |
19126026 |
nnz = h->non_zero_count[left_xy[LEFT(i)]]; |
|
564 |
19126026 |
nnz_cache[3 + 8 * 1 + 2 * 8 * i] = nnz[left_block[8 + 0 + 2 * i]]; |
|
565 |
19126026 |
nnz_cache[3 + 8 * 2 + 2 * 8 * i] = nnz[left_block[8 + 1 + 2 * i]]; |
|
566 |
✓✓ | 19126026 |
if (CHROMA444(h)) { |
567 |
190240 |
nnz_cache[3 + 8 * 6 + 2 * 8 * i] = nnz[left_block[8 + 0 + 2 * i] + 4 * 4]; |
|
568 |
190240 |
nnz_cache[3 + 8 * 7 + 2 * 8 * i] = nnz[left_block[8 + 1 + 2 * i] + 4 * 4]; |
|
569 |
190240 |
nnz_cache[3 + 8 * 11 + 2 * 8 * i] = nnz[left_block[8 + 0 + 2 * i] + 8 * 4]; |
|
570 |
190240 |
nnz_cache[3 + 8 * 12 + 2 * 8 * i] = nnz[left_block[8 + 1 + 2 * i] + 8 * 4]; |
|
571 |
✓✓ | 18935786 |
} else if (CHROMA422(h)) { |
572 |
1619188 |
nnz_cache[3 + 8 * 6 + 2 * 8 * i] = nnz[left_block[8 + 0 + 2 * i] - 2 + 4 * 4]; |
|
573 |
1619188 |
nnz_cache[3 + 8 * 7 + 2 * 8 * i] = nnz[left_block[8 + 1 + 2 * i] - 2 + 4 * 4]; |
|
574 |
1619188 |
nnz_cache[3 + 8 * 11 + 2 * 8 * i] = nnz[left_block[8 + 0 + 2 * i] - 2 + 8 * 4]; |
|
575 |
1619188 |
nnz_cache[3 + 8 * 12 + 2 * 8 * i] = nnz[left_block[8 + 1 + 2 * i] - 2 + 8 * 4]; |
|
576 |
} else { |
||
577 |
17316598 |
nnz_cache[3 + 8 * 6 + 8 * i] = nnz[left_block[8 + 4 + 2 * i]]; |
|
578 |
17316598 |
nnz_cache[3 + 8 * 11 + 8 * i] = nnz[left_block[8 + 5 + 2 * i]]; |
|
579 |
} |
||
580 |
} else { |
||
581 |
588412 |
nnz_cache[3 + 8 * 1 + 2 * 8 * i] = |
|
582 |
588412 |
nnz_cache[3 + 8 * 2 + 2 * 8 * i] = |
|
583 |
588412 |
nnz_cache[3 + 8 * 6 + 2 * 8 * i] = |
|
584 |
588412 |
nnz_cache[3 + 8 * 7 + 2 * 8 * i] = |
|
585 |
588412 |
nnz_cache[3 + 8 * 11 + 2 * 8 * i] = |
|
586 |
✓✓✗✗ |
588412 |
nnz_cache[3 + 8 * 12 + 2 * 8 * i] = CABAC(h) && !IS_INTRA(mb_type) ? 0 : 64; |
587 |
} |
||
588 |
} |
||
589 |
|||
590 |
if (CABAC(h)) { |
||
591 |
// top_cbp |
||
592 |
✓✓ | 6383529 |
if (top_type) |
593 |
5920418 |
sl->top_cbp = h->cbp_table[top_xy]; |
|
594 |
else |
||
595 |
✓✓ | 463111 |
sl->top_cbp = IS_INTRA(mb_type) ? 0x7CF : 0x00F; |
596 |
// left_cbp |
||
597 |
✓✓ | 6383529 |
if (left_type[LTOP]) { |
598 |
6216172 |
sl->left_cbp = (h->cbp_table[left_xy[LTOP]] & 0x7F0) | |
|
599 |
6216172 |
((h->cbp_table[left_xy[LTOP]] >> (left_block[0] & (~1))) & 2) | |
|
600 |
6216172 |
(((h->cbp_table[left_xy[LBOT]] >> (left_block[2] & (~1))) & 2) << 2); |
|
601 |
} else { |
||
602 |
✓✓ | 167357 |
sl->left_cbp = IS_INTRA(mb_type) ? 0x7CF : 0x00F; |
603 |
} |
||
604 |
} |
||
605 |
} |
||
606 |
|||
607 |
✓✓✓✓ ✓✓ |
11291472 |
if (IS_INTER(mb_type) || (IS_DIRECT(mb_type) && sl->direct_spatial_mv_pred)) { |
608 |
int list; |
||
609 |
7488554 |
int b_stride = h->b_stride; |
|
610 |
✓✓ | 19267640 |
for (list = 0; list < sl->list_count; list++) { |
611 |
11779086 |
int8_t *ref_cache = &sl->ref_cache[list][scan8[0]]; |
|
612 |
11779086 |
int8_t *ref = h->cur_pic.ref_index[list]; |
|
613 |
11779086 |
int16_t(*mv_cache)[2] = &sl->mv_cache[list][scan8[0]]; |
|
614 |
11779086 |
int16_t(*mv)[2] = h->cur_pic.motion_val[list]; |
|
615 |
✓✓ | 11779086 |
if (!USES_LIST(mb_type, list)) |
616 |
1234825 |
continue; |
|
617 |
av_assert2(!(IS_DIRECT(mb_type) && !sl->direct_spatial_mv_pred)); |
||
618 |
|||
619 |
✓✓ | 10544261 |
if (USES_LIST(top_type, list)) { |
620 |
8375634 |
const int b_xy = h->mb2b_xy[top_xy] + 3 * b_stride; |
|
621 |
8375634 |
AV_COPY128(mv_cache[0 - 1 * 8], mv[b_xy + 0]); |
|
622 |
8375634 |
ref_cache[0 - 1 * 8] = |
|
623 |
8375634 |
ref_cache[1 - 1 * 8] = ref[4 * top_xy + 2]; |
|
624 |
8375634 |
ref_cache[2 - 1 * 8] = |
|
625 |
8375634 |
ref_cache[3 - 1 * 8] = ref[4 * top_xy + 3]; |
|
626 |
} else { |
||
627 |
2168627 |
AV_ZERO128(mv_cache[0 - 1 * 8]); |
|
628 |
✓✓ | 2168627 |
AV_WN32A(&ref_cache[0 - 1 * 8], |
629 |
((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE) & 0xFF) * 0x01010101u); |
||
630 |
} |
||
631 |
|||
632 |
✓✓ | 10544261 |
if (mb_type & (MB_TYPE_16x8 | MB_TYPE_8x8)) { |
633 |
✓✓ | 8150436 |
for (i = 0; i < 2; i++) { |
634 |
5433624 |
int cache_idx = -1 + i * 2 * 8; |
|
635 |
✓✓ | 5433624 |
if (USES_LIST(left_type[LEFT(i)], list)) { |
636 |
4645692 |
const int b_xy = h->mb2b_xy[left_xy[LEFT(i)]] + 3; |
|
637 |
4645692 |
const int b8_xy = 4 * left_xy[LEFT(i)] + 1; |
|
638 |
4645692 |
AV_COPY32(mv_cache[cache_idx], |
|
639 |
mv[b_xy + b_stride * left_block[0 + i * 2]]); |
||
640 |
4645692 |
AV_COPY32(mv_cache[cache_idx + 8], |
|
641 |
mv[b_xy + b_stride * left_block[1 + i * 2]]); |
||
642 |
4645692 |
ref_cache[cache_idx] = ref[b8_xy + (left_block[0 + i * 2] & ~1)]; |
|
643 |
4645692 |
ref_cache[cache_idx + 8] = ref[b8_xy + (left_block[1 + i * 2] & ~1)]; |
|
644 |
} else { |
||
645 |
787932 |
AV_ZERO32(mv_cache[cache_idx]); |
|
646 |
787932 |
AV_ZERO32(mv_cache[cache_idx + 8]); |
|
647 |
787932 |
ref_cache[cache_idx] = |
|
648 |
✓✓ | 787932 |
ref_cache[cache_idx + 8] = (left_type[LEFT(i)]) ? LIST_NOT_USED |
649 |
: PART_NOT_AVAILABLE; |
||
650 |
} |
||
651 |
} |
||
652 |
} else { |
||
653 |
✓✓ | 7827449 |
if (USES_LIST(left_type[LTOP], list)) { |
654 |
6598846 |
const int b_xy = h->mb2b_xy[left_xy[LTOP]] + 3; |
|
655 |
6598846 |
const int b8_xy = 4 * left_xy[LTOP] + 1; |
|
656 |
6598846 |
AV_COPY32(mv_cache[-1], mv[b_xy + b_stride * left_block[0]]); |
|
657 |
6598846 |
ref_cache[-1] = ref[b8_xy + (left_block[0] & ~1)]; |
|
658 |
} else { |
||
659 |
1228603 |
AV_ZERO32(mv_cache[-1]); |
|
660 |
✓✓ | 1228603 |
ref_cache[-1] = left_type[LTOP] ? LIST_NOT_USED |
661 |
: PART_NOT_AVAILABLE; |
||
662 |
} |
||
663 |
} |
||
664 |
|||
665 |
✓✓ | 10544261 |
if (USES_LIST(topright_type, list)) { |
666 |
7628746 |
const int b_xy = h->mb2b_xy[topright_xy] + 3 * b_stride; |
|
667 |
7628746 |
AV_COPY32(mv_cache[4 - 1 * 8], mv[b_xy]); |
|
668 |
7628746 |
ref_cache[4 - 1 * 8] = ref[4 * topright_xy + 2]; |
|
669 |
} else { |
||
670 |
2915515 |
AV_ZERO32(mv_cache[4 - 1 * 8]); |
|
671 |
✓✓ | 2915515 |
ref_cache[4 - 1 * 8] = topright_type ? LIST_NOT_USED |
672 |
: PART_NOT_AVAILABLE; |
||
673 |
} |
||
674 |
✓✓✓✓ |
10544261 |
if(ref_cache[2 - 1*8] < 0 || ref_cache[4 - 1 * 8] < 0) { |
675 |
✓✓ | 3567062 |
if (USES_LIST(topleft_type, list)) { |
676 |
1638215 |
const int b_xy = h->mb2b_xy[topleft_xy] + 3 + b_stride + |
|
677 |
1638215 |
(sl->topleft_partition & 2 * b_stride); |
|
678 |
1638215 |
const int b8_xy = 4 * topleft_xy + 1 + (sl->topleft_partition & 2); |
|
679 |
1638215 |
AV_COPY32(mv_cache[-1 - 1 * 8], mv[b_xy]); |
|
680 |
1638215 |
ref_cache[-1 - 1 * 8] = ref[b8_xy]; |
|
681 |
} else { |
||
682 |
1928847 |
AV_ZERO32(mv_cache[-1 - 1 * 8]); |
|
683 |
✓✓ | 1928847 |
ref_cache[-1 - 1 * 8] = topleft_type ? LIST_NOT_USED |
684 |
: PART_NOT_AVAILABLE; |
||
685 |
} |
||
686 |
} |
||
687 |
|||
688 |
✓✓✓✓ |
10544261 |
if ((mb_type & (MB_TYPE_SKIP | MB_TYPE_DIRECT2)) && !FRAME_MBAFF(h)) |
689 |
3378464 |
continue; |
|
690 |
|||
691 |
✓✓ | 7165797 |
if (!(mb_type & (MB_TYPE_SKIP | MB_TYPE_DIRECT2))) { |
692 |
6661181 |
uint8_t(*mvd_cache)[2] = &sl->mvd_cache[list][scan8[0]]; |
|
693 |
6661181 |
uint8_t(*mvd)[2] = sl->mvd_table[list]; |
|
694 |
6661181 |
ref_cache[2 + 8 * 0] = |
|
695 |
6661181 |
ref_cache[2 + 8 * 2] = PART_NOT_AVAILABLE; |
|
696 |
6661181 |
AV_ZERO32(mv_cache[2 + 8 * 0]); |
|
697 |
6661181 |
AV_ZERO32(mv_cache[2 + 8 * 2]); |
|
698 |
|||
699 |
if (CABAC(h)) { |
||
700 |
✓✓ | 4348982 |
if (USES_LIST(top_type, list)) { |
701 |
3445005 |
const int b_xy = h->mb2br_xy[top_xy]; |
|
702 |
3445005 |
AV_COPY64(mvd_cache[0 - 1 * 8], mvd[b_xy + 0]); |
|
703 |
} else { |
||
704 |
903977 |
AV_ZERO64(mvd_cache[0 - 1 * 8]); |
|
705 |
} |
||
706 |
✓✓ | 4348982 |
if (USES_LIST(left_type[LTOP], list)) { |
707 |
3658247 |
const int b_xy = h->mb2br_xy[left_xy[LTOP]] + 6; |
|
708 |
3658247 |
AV_COPY16(mvd_cache[-1 + 0 * 8], mvd[b_xy - left_block[0]]); |
|
709 |
3658247 |
AV_COPY16(mvd_cache[-1 + 1 * 8], mvd[b_xy - left_block[1]]); |
|
710 |
} else { |
||
711 |
690735 |
AV_ZERO16(mvd_cache[-1 + 0 * 8]); |
|
712 |
690735 |
AV_ZERO16(mvd_cache[-1 + 1 * 8]); |
|
713 |
} |
||
714 |
✓✓ | 4348982 |
if (USES_LIST(left_type[LBOT], list)) { |
715 |
3658047 |
const int b_xy = h->mb2br_xy[left_xy[LBOT]] + 6; |
|
716 |
3658047 |
AV_COPY16(mvd_cache[-1 + 2 * 8], mvd[b_xy - left_block[2]]); |
|
717 |
3658047 |
AV_COPY16(mvd_cache[-1 + 3 * 8], mvd[b_xy - left_block[3]]); |
|
718 |
} else { |
||
719 |
690935 |
AV_ZERO16(mvd_cache[-1 + 2 * 8]); |
|
720 |
690935 |
AV_ZERO16(mvd_cache[-1 + 3 * 8]); |
|
721 |
} |
||
722 |
4348982 |
AV_ZERO16(mvd_cache[2 + 8 * 0]); |
|
723 |
4348982 |
AV_ZERO16(mvd_cache[2 + 8 * 2]); |
|
724 |
✓✓ | 4348982 |
if (sl->slice_type_nos == AV_PICTURE_TYPE_B) { |
725 |
2518606 |
uint8_t *direct_cache = &sl->direct_cache[scan8[0]]; |
|
726 |
2518606 |
uint8_t *direct_table = h->direct_table; |
|
727 |
2518606 |
fill_rectangle(direct_cache, 4, 4, 8, MB_TYPE_16x16 >> 1, 1); |
|
728 |
|||
729 |
✓✓ | 2518606 |
if (IS_DIRECT(top_type)) { |
730 |
381290 |
AV_WN32A(&direct_cache[-1 * 8], |
|
731 |
0x01010101u * (MB_TYPE_DIRECT2 >> 1)); |
||
732 |
✓✓ | 2137316 |
} else if (IS_8X8(top_type)) { |
733 |
610301 |
int b8_xy = 4 * top_xy; |
|
734 |
610301 |
direct_cache[0 - 1 * 8] = direct_table[b8_xy + 2]; |
|
735 |
610301 |
direct_cache[2 - 1 * 8] = direct_table[b8_xy + 3]; |
|
736 |
} else { |
||
737 |
1527015 |
AV_WN32A(&direct_cache[-1 * 8], |
|
738 |
0x01010101 * (MB_TYPE_16x16 >> 1)); |
||
739 |
} |
||
740 |
|||
741 |
✓✓ | 2518606 |
if (IS_DIRECT(left_type[LTOP])) |
742 |
376882 |
direct_cache[-1 + 0 * 8] = MB_TYPE_DIRECT2 >> 1; |
|
743 |
✓✓ | 2141724 |
else if (IS_8X8(left_type[LTOP])) |
744 |
645260 |
direct_cache[-1 + 0 * 8] = direct_table[4 * left_xy[LTOP] + 1 + (left_block[0] & ~1)]; |
|
745 |
else |
||
746 |
1496464 |
direct_cache[-1 + 0 * 8] = MB_TYPE_16x16 >> 1; |
|
747 |
|||
748 |
✓✓ | 2518606 |
if (IS_DIRECT(left_type[LBOT])) |
749 |
375997 |
direct_cache[-1 + 2 * 8] = MB_TYPE_DIRECT2 >> 1; |
|
750 |
✓✓ | 2142609 |
else if (IS_8X8(left_type[LBOT])) |
751 |
647349 |
direct_cache[-1 + 2 * 8] = direct_table[4 * left_xy[LBOT] + 1 + (left_block[2] & ~1)]; |
|
752 |
else |
||
753 |
1495260 |
direct_cache[-1 + 2 * 8] = MB_TYPE_16x16 >> 1; |
|
754 |
} |
||
755 |
} |
||
756 |
} |
||
757 |
|||
758 |
#define MAP_MVS \ |
||
759 |
MAP_F2F(scan8[0] - 1 - 1 * 8, topleft_type) \ |
||
760 |
MAP_F2F(scan8[0] + 0 - 1 * 8, top_type) \ |
||
761 |
MAP_F2F(scan8[0] + 1 - 1 * 8, top_type) \ |
||
762 |
MAP_F2F(scan8[0] + 2 - 1 * 8, top_type) \ |
||
763 |
MAP_F2F(scan8[0] + 3 - 1 * 8, top_type) \ |
||
764 |
MAP_F2F(scan8[0] + 4 - 1 * 8, topright_type) \ |
||
765 |
MAP_F2F(scan8[0] - 1 + 0 * 8, left_type[LTOP]) \ |
||
766 |
MAP_F2F(scan8[0] - 1 + 1 * 8, left_type[LTOP]) \ |
||
767 |
MAP_F2F(scan8[0] - 1 + 2 * 8, left_type[LBOT]) \ |
||
768 |
MAP_F2F(scan8[0] - 1 + 3 * 8, left_type[LBOT]) |
||
769 |
|||
770 |
✓✓ | 7165797 |
if (FRAME_MBAFF(h)) { |
771 |
✓✓ | 1583802 |
if (MB_FIELD(sl)) { |
772 |
|||
773 |
#define MAP_F2F(idx, mb_type) \ |
||
774 |
if (!IS_INTERLACED(mb_type) && sl->ref_cache[list][idx] >= 0) { \ |
||
775 |
sl->ref_cache[list][idx] *= 2; \ |
||
776 |
sl->mv_cache[list][idx][1] /= 2; \ |
||
777 |
sl->mvd_cache[list][idx][1] >>= 1; \ |
||
778 |
} |
||
779 |
|||
780 |
✓✓✓✓ ✓✓✓✓ ✓✓✓✓ ✓✓✓✓ ✓✓✓✓ ✓✓✓✓ ✓✓✓✓ ✓✓✓✓ ✓✓✓✓ ✓✓✓✓ |
543998 |
MAP_MVS |
781 |
} else { |
||
782 |
|||
783 |
#undef MAP_F2F |
||
784 |
#define MAP_F2F(idx, mb_type) \ |
||
785 |
if (IS_INTERLACED(mb_type) && sl->ref_cache[list][idx] >= 0) { \ |
||
786 |
sl->ref_cache[list][idx] >>= 1; \ |
||
787 |
sl->mv_cache[list][idx][1] *= 2; \ |
||
788 |
sl->mvd_cache[list][idx][1] <<= 1; \ |
||
789 |
} |
||
790 |
|||
791 |
✓✓✓✓ ✓✓✓✓ ✓✓✓✓ ✓✓✓✓ ✓✓✓✓ ✓✓✓✓ ✓✓✓✓ ✓✓✓✓ ✓✓✓✓ ✓✓✓✓ |
1039804 |
MAP_MVS |
792 |
#undef MAP_F2F |
||
793 |
} |
||
794 |
} |
||
795 |
} |
||
796 |
} |
||
797 |
|||
798 |
11291472 |
sl->neighbor_transform_size = !!IS_8x8DCT(top_type) + !!IS_8x8DCT(left_type[LTOP]); |
|
799 |
11291472 |
} |
|
800 |
|||
801 |
/** |
||
802 |
* decodes a P_SKIP or B_SKIP macroblock |
||
803 |
*/ |
||
804 |
2963098 |
static void av_unused decode_mb_skip(const H264Context *h, H264SliceContext *sl) |
|
805 |
{ |
||
806 |
2963098 |
const int mb_xy = sl->mb_xy; |
|
807 |
2963098 |
int mb_type = 0; |
|
808 |
|||
809 |
2963098 |
memset(h->non_zero_count[mb_xy], 0, 48); |
|
810 |
|||
811 |
✓✓ | 2963098 |
if (MB_FIELD(sl)) |
812 |
887678 |
mb_type |= MB_TYPE_INTERLACED; |
|
813 |
|||
814 |
✓✓ | 2963098 |
if (sl->slice_type_nos == AV_PICTURE_TYPE_B) { |
815 |
// just for fill_caches. pred_direct_motion will set the real mb_type |
||
816 |
1570319 |
mb_type |= MB_TYPE_L0L1 | MB_TYPE_DIRECT2 | MB_TYPE_SKIP; |
|
817 |
✓✓ | 1570319 |
if (sl->direct_spatial_mv_pred) { |
818 |
1434253 |
fill_decode_neighbors(h, sl, mb_type); |
|
819 |
1434253 |
fill_decode_caches(h, sl, mb_type); //FIXME check what is needed and what not ... |
|
820 |
} |
||
821 |
1570319 |
ff_h264_pred_direct_motion(h, sl, &mb_type); |
|
822 |
1570319 |
mb_type |= MB_TYPE_SKIP; |
|
823 |
} else { |
||
824 |
1392779 |
mb_type |= MB_TYPE_16x16 | MB_TYPE_P0L0 | MB_TYPE_P1L0 | MB_TYPE_SKIP; |
|
825 |
|||
826 |
1392779 |
fill_decode_neighbors(h, sl, mb_type); |
|
827 |
1392779 |
pred_pskip_motion(h, sl); |
|
828 |
} |
||
829 |
|||
830 |
2963098 |
write_back_motion(h, sl, mb_type); |
|
831 |
2963098 |
h->cur_pic.mb_type[mb_xy] = mb_type; |
|
832 |
2963098 |
h->cur_pic.qscale_table[mb_xy] = sl->qscale; |
|
833 |
2963098 |
h->slice_table[mb_xy] = sl->slice_num; |
|
834 |
2963098 |
sl->prev_mb_skipped = 1; |
|
835 |
2963098 |
} |
|
836 |
|||
837 |
#endif /* AVCODEC_H264_MVPRED_H */ |
| Generated by: GCOVR (Version 4.2) |