Commit | Line | Data |
---|---|---|
2ba45a60 DM |
1 | /* |
2 | * Chinese AVS video (AVS1-P2, JiZhun profile) decoder. | |
3 | * Copyright (c) 2006 Stefan Gehrer <stefan.gehrer@gmx.de> | |
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 | * Chinese AVS video (AVS1-P2, JiZhun profile) decoder | |
25 | * @author Stefan Gehrer <stefan.gehrer@gmx.de> | |
26 | */ | |
27 | ||
28 | #include "avcodec.h" | |
29 | #include "get_bits.h" | |
30 | #include "golomb.h" | |
31 | #include "h264chroma.h" | |
32 | #include "idctdsp.h" | |
f6fa7814 | 33 | #include "internal.h" |
2ba45a60 DM |
34 | #include "mathops.h" |
35 | #include "qpeldsp.h" | |
36 | #include "cavs.h" | |
37 | ||
38 | static const uint8_t alpha_tab[64] = { | |
39 | 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 2, 2, 2, 3, 3, | |
40 | 4, 4, 5, 5, 6, 7, 8, 9, 10, 11, 12, 13, 15, 16, 18, 20, | |
41 | 22, 24, 26, 28, 30, 33, 33, 35, 35, 36, 37, 37, 39, 39, 42, 44, | |
42 | 46, 48, 50, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 64 | |
43 | }; | |
44 | ||
45 | static const uint8_t beta_tab[64] = { | |
46 | 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 2, 2, 2, | |
47 | 2, 2, 3, 3, 3, 3, 4, 4, 4, 4, 5, 5, 5, 5, 6, 6, | |
48 | 6, 7, 7, 7, 8, 8, 8, 9, 9, 10, 10, 11, 11, 12, 13, 14, | |
49 | 15, 16, 17, 18, 19, 20, 21, 22, 23, 23, 24, 24, 25, 25, 26, 27 | |
50 | }; | |
51 | ||
52 | static const uint8_t tc_tab[64] = { | |
53 | 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, | |
54 | 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2, | |
55 | 2, 2, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 3, 4, 4, 4, | |
56 | 5, 5, 5, 6, 6, 6, 7, 7, 7, 7, 8, 8, 8, 9, 9, 9 | |
57 | }; | |
58 | ||
59 | /** mark block as unavailable, i.e. out of picture | |
60 | * or not yet decoded */ | |
61 | static const cavs_vector un_mv = { 0, 0, 1, NOT_AVAIL }; | |
62 | ||
63 | static const int8_t left_modifier_l[8] = { 0, -1, 6, -1, -1, 7, 6, 7 }; | |
64 | static const int8_t top_modifier_l[8] = { -1, 1, 5, -1, -1, 5, 7, 7 }; | |
65 | static const int8_t left_modifier_c[7] = { 5, -1, 2, -1, 6, 5, 6 }; | |
66 | static const int8_t top_modifier_c[7] = { 4, 1, -1, -1, 4, 6, 6 }; | |
67 | ||
68 | /***************************************************************************** | |
69 | * | |
70 | * in-loop deblocking filter | |
71 | * | |
72 | ****************************************************************************/ | |
73 | ||
74 | static inline int get_bs(cavs_vector *mvP, cavs_vector *mvQ, int b) | |
75 | { | |
76 | if ((mvP->ref == REF_INTRA) || (mvQ->ref == REF_INTRA)) | |
77 | return 2; | |
78 | if((abs(mvP->x - mvQ->x) >= 4) || | |
79 | (abs(mvP->y - mvQ->y) >= 4) || | |
80 | (mvP->ref != mvQ->ref)) | |
81 | return 1; | |
82 | if (b) { | |
83 | mvP += MV_BWD_OFFS; | |
84 | mvQ += MV_BWD_OFFS; | |
85 | if((abs(mvP->x - mvQ->x) >= 4) || | |
86 | (abs(mvP->y - mvQ->y) >= 4) || | |
87 | (mvP->ref != mvQ->ref)) | |
88 | return 1; | |
89 | } | |
90 | return 0; | |
91 | } | |
92 | ||
93 | #define SET_PARAMS \ | |
94 | alpha = alpha_tab[av_clip(qp_avg + h->alpha_offset, 0, 63)]; \ | |
95 | beta = beta_tab[av_clip(qp_avg + h->beta_offset, 0, 63)]; \ | |
96 | tc = tc_tab[av_clip(qp_avg + h->alpha_offset, 0, 63)]; | |
97 | ||
98 | /** | |
99 | * in-loop deblocking filter for a single macroblock | |
100 | * | |
101 | * boundary strength (bs) mapping: | |
102 | * | |
103 | * --4---5-- | |
104 | * 0 2 | | |
105 | * | 6 | 7 | | |
106 | * 1 3 | | |
107 | * --------- | |
108 | * | |
109 | */ | |
110 | void ff_cavs_filter(AVSContext *h, enum cavs_mb mb_type) | |
111 | { | |
112 | uint8_t bs[8]; | |
113 | int qp_avg, alpha, beta, tc; | |
114 | int i; | |
115 | ||
116 | /* save un-deblocked lines */ | |
117 | h->topleft_border_y = h->top_border_y[h->mbx * 16 + 15]; | |
118 | h->topleft_border_u = h->top_border_u[h->mbx * 10 + 8]; | |
119 | h->topleft_border_v = h->top_border_v[h->mbx * 10 + 8]; | |
120 | memcpy(&h->top_border_y[h->mbx * 16], h->cy + 15 * h->l_stride, 16); | |
121 | memcpy(&h->top_border_u[h->mbx * 10 + 1], h->cu + 7 * h->c_stride, 8); | |
122 | memcpy(&h->top_border_v[h->mbx * 10 + 1], h->cv + 7 * h->c_stride, 8); | |
123 | for (i = 0; i < 8; i++) { | |
124 | h->left_border_y[i * 2 + 1] = *(h->cy + 15 + (i * 2 + 0) * h->l_stride); | |
125 | h->left_border_y[i * 2 + 2] = *(h->cy + 15 + (i * 2 + 1) * h->l_stride); | |
126 | h->left_border_u[i + 1] = *(h->cu + 7 + i * h->c_stride); | |
127 | h->left_border_v[i + 1] = *(h->cv + 7 + i * h->c_stride); | |
128 | } | |
129 | if (!h->loop_filter_disable) { | |
130 | /* determine bs */ | |
131 | if (mb_type == I_8X8) | |
132 | memset(bs, 2, 8); | |
133 | else { | |
134 | memset(bs, 0, 8); | |
135 | if (ff_cavs_partition_flags[mb_type] & SPLITV) { | |
136 | bs[2] = get_bs(&h->mv[MV_FWD_X0], &h->mv[MV_FWD_X1], mb_type > P_8X8); | |
137 | bs[3] = get_bs(&h->mv[MV_FWD_X2], &h->mv[MV_FWD_X3], mb_type > P_8X8); | |
138 | } | |
139 | if (ff_cavs_partition_flags[mb_type] & SPLITH) { | |
140 | bs[6] = get_bs(&h->mv[MV_FWD_X0], &h->mv[MV_FWD_X2], mb_type > P_8X8); | |
141 | bs[7] = get_bs(&h->mv[MV_FWD_X1], &h->mv[MV_FWD_X3], mb_type > P_8X8); | |
142 | } | |
143 | bs[0] = get_bs(&h->mv[MV_FWD_A1], &h->mv[MV_FWD_X0], mb_type > P_8X8); | |
144 | bs[1] = get_bs(&h->mv[MV_FWD_A3], &h->mv[MV_FWD_X2], mb_type > P_8X8); | |
145 | bs[4] = get_bs(&h->mv[MV_FWD_B2], &h->mv[MV_FWD_X0], mb_type > P_8X8); | |
146 | bs[5] = get_bs(&h->mv[MV_FWD_B3], &h->mv[MV_FWD_X1], mb_type > P_8X8); | |
147 | } | |
148 | if (AV_RN64(bs)) { | |
149 | if (h->flags & A_AVAIL) { | |
150 | qp_avg = (h->qp + h->left_qp + 1) >> 1; | |
151 | SET_PARAMS; | |
152 | h->cdsp.cavs_filter_lv(h->cy, h->l_stride, alpha, beta, tc, bs[0], bs[1]); | |
153 | qp_avg = (ff_cavs_chroma_qp[h->qp] + ff_cavs_chroma_qp[h->left_qp] + 1) >> 1; | |
154 | SET_PARAMS; | |
155 | h->cdsp.cavs_filter_cv(h->cu, h->c_stride, alpha, beta, tc, bs[0], bs[1]); | |
156 | h->cdsp.cavs_filter_cv(h->cv, h->c_stride, alpha, beta, tc, bs[0], bs[1]); | |
157 | } | |
158 | qp_avg = h->qp; | |
159 | SET_PARAMS; | |
160 | h->cdsp.cavs_filter_lv(h->cy + 8, h->l_stride, alpha, beta, tc, bs[2], bs[3]); | |
161 | h->cdsp.cavs_filter_lh(h->cy + 8 * h->l_stride, h->l_stride, alpha, beta, tc, bs[6], bs[7]); | |
162 | ||
163 | if (h->flags & B_AVAIL) { | |
164 | qp_avg = (h->qp + h->top_qp[h->mbx] + 1) >> 1; | |
165 | SET_PARAMS; | |
166 | h->cdsp.cavs_filter_lh(h->cy, h->l_stride, alpha, beta, tc, bs[4], bs[5]); | |
167 | qp_avg = (ff_cavs_chroma_qp[h->qp] + ff_cavs_chroma_qp[h->top_qp[h->mbx]] + 1) >> 1; | |
168 | SET_PARAMS; | |
169 | h->cdsp.cavs_filter_ch(h->cu, h->c_stride, alpha, beta, tc, bs[4], bs[5]); | |
170 | h->cdsp.cavs_filter_ch(h->cv, h->c_stride, alpha, beta, tc, bs[4], bs[5]); | |
171 | } | |
172 | } | |
173 | } | |
174 | h->left_qp = h->qp; | |
175 | h->top_qp[h->mbx] = h->qp; | |
176 | } | |
177 | ||
178 | #undef SET_PARAMS | |
179 | ||
180 | /***************************************************************************** | |
181 | * | |
182 | * spatial intra prediction | |
183 | * | |
184 | ****************************************************************************/ | |
185 | ||
186 | void ff_cavs_load_intra_pred_luma(AVSContext *h, uint8_t *top, | |
187 | uint8_t **left, int block) | |
188 | { | |
189 | int i; | |
190 | ||
191 | switch (block) { | |
192 | case 0: | |
193 | *left = h->left_border_y; | |
194 | h->left_border_y[0] = h->left_border_y[1]; | |
195 | memset(&h->left_border_y[17], h->left_border_y[16], 9); | |
196 | memcpy(&top[1], &h->top_border_y[h->mbx * 16], 16); | |
197 | top[17] = top[16]; | |
198 | top[0] = top[1]; | |
199 | if ((h->flags & A_AVAIL) && (h->flags & B_AVAIL)) | |
200 | h->left_border_y[0] = top[0] = h->topleft_border_y; | |
201 | break; | |
202 | case 1: | |
203 | *left = h->intern_border_y; | |
204 | for (i = 0; i < 8; i++) | |
205 | h->intern_border_y[i + 1] = *(h->cy + 7 + i * h->l_stride); | |
206 | memset(&h->intern_border_y[9], h->intern_border_y[8], 9); | |
207 | h->intern_border_y[0] = h->intern_border_y[1]; | |
208 | memcpy(&top[1], &h->top_border_y[h->mbx * 16 + 8], 8); | |
209 | if (h->flags & C_AVAIL) | |
210 | memcpy(&top[9], &h->top_border_y[(h->mbx + 1) * 16], 8); | |
211 | else | |
212 | memset(&top[9], top[8], 9); | |
213 | top[17] = top[16]; | |
214 | top[0] = top[1]; | |
215 | if (h->flags & B_AVAIL) | |
216 | h->intern_border_y[0] = top[0] = h->top_border_y[h->mbx * 16 + 7]; | |
217 | break; | |
218 | case 2: | |
219 | *left = &h->left_border_y[8]; | |
220 | memcpy(&top[1], h->cy + 7 * h->l_stride, 16); | |
221 | top[17] = top[16]; | |
222 | top[0] = top[1]; | |
223 | if (h->flags & A_AVAIL) | |
224 | top[0] = h->left_border_y[8]; | |
225 | break; | |
226 | case 3: | |
227 | *left = &h->intern_border_y[8]; | |
228 | for (i = 0; i < 8; i++) | |
229 | h->intern_border_y[i + 9] = *(h->cy + 7 + (i + 8) * h->l_stride); | |
230 | memset(&h->intern_border_y[17], h->intern_border_y[16], 9); | |
231 | memcpy(&top[0], h->cy + 7 + 7 * h->l_stride, 9); | |
232 | memset(&top[9], top[8], 9); | |
233 | break; | |
234 | } | |
235 | } | |
236 | ||
237 | void ff_cavs_load_intra_pred_chroma(AVSContext *h) | |
238 | { | |
239 | /* extend borders by one pixel */ | |
240 | h->left_border_u[9] = h->left_border_u[8]; | |
241 | h->left_border_v[9] = h->left_border_v[8]; | |
242 | if(h->flags & C_AVAIL) { | |
243 | h->top_border_u[h->mbx*10 + 9] = h->top_border_u[h->mbx*10 + 11]; | |
244 | h->top_border_v[h->mbx*10 + 9] = h->top_border_v[h->mbx*10 + 11]; | |
245 | } else { | |
246 | h->top_border_u[h->mbx * 10 + 9] = h->top_border_u[h->mbx * 10 + 8]; | |
247 | h->top_border_v[h->mbx * 10 + 9] = h->top_border_v[h->mbx * 10 + 8]; | |
248 | } | |
249 | if((h->flags & A_AVAIL) && (h->flags & B_AVAIL)) { | |
250 | h->top_border_u[h->mbx * 10] = h->left_border_u[0] = h->topleft_border_u; | |
251 | h->top_border_v[h->mbx * 10] = h->left_border_v[0] = h->topleft_border_v; | |
252 | } else { | |
253 | h->left_border_u[0] = h->left_border_u[1]; | |
254 | h->left_border_v[0] = h->left_border_v[1]; | |
255 | h->top_border_u[h->mbx * 10] = h->top_border_u[h->mbx * 10 + 1]; | |
256 | h->top_border_v[h->mbx * 10] = h->top_border_v[h->mbx * 10 + 1]; | |
257 | } | |
258 | } | |
259 | ||
260 | static void intra_pred_vert(uint8_t *d, uint8_t *top, uint8_t *left, int stride) | |
261 | { | |
262 | int y; | |
263 | uint64_t a = AV_RN64(&top[1]); | |
264 | for (y = 0; y < 8; y++) | |
265 | *((uint64_t *)(d + y * stride)) = a; | |
266 | } | |
267 | ||
268 | static void intra_pred_horiz(uint8_t *d, uint8_t *top, uint8_t *left, int stride) | |
269 | { | |
270 | int y; | |
271 | uint64_t a; | |
272 | for (y = 0; y < 8; y++) { | |
273 | a = left[y + 1] * 0x0101010101010101ULL; | |
274 | *((uint64_t *)(d + y * stride)) = a; | |
275 | } | |
276 | } | |
277 | ||
278 | static void intra_pred_dc_128(uint8_t *d, uint8_t *top, uint8_t *left, int stride) | |
279 | { | |
280 | int y; | |
281 | uint64_t a = 0x8080808080808080ULL; | |
282 | for (y = 0; y < 8; y++) | |
283 | *((uint64_t *)(d + y * stride)) = a; | |
284 | } | |
285 | ||
286 | static void intra_pred_plane(uint8_t *d, uint8_t *top, uint8_t *left, int stride) | |
287 | { | |
288 | int x, y, ia; | |
289 | int ih = 0; | |
290 | int iv = 0; | |
291 | const uint8_t *cm = ff_crop_tab + MAX_NEG_CROP; | |
292 | ||
293 | for (x = 0; x < 4; x++) { | |
294 | ih += (x + 1) * (top[5 + x] - top[3 - x]); | |
295 | iv += (x + 1) * (left[5 + x] - left[3 - x]); | |
296 | } | |
297 | ia = (top[8] + left[8]) << 4; | |
298 | ih = (17 * ih + 16) >> 5; | |
299 | iv = (17 * iv + 16) >> 5; | |
300 | for (y = 0; y < 8; y++) | |
301 | for (x = 0; x < 8; x++) | |
302 | d[y * stride + x] = cm[(ia + (x - 3) * ih + (y - 3) * iv + 16) >> 5]; | |
303 | } | |
304 | ||
305 | #define LOWPASS(ARRAY, INDEX) \ | |
306 | ((ARRAY[(INDEX) - 1] + 2 * ARRAY[(INDEX)] + ARRAY[(INDEX) + 1] + 2) >> 2) | |
307 | ||
308 | static void intra_pred_lp(uint8_t *d, uint8_t *top, uint8_t *left, int stride) | |
309 | { | |
310 | int x, y; | |
311 | for (y = 0; y < 8; y++) | |
312 | for (x = 0; x < 8; x++) | |
313 | d[y * stride + x] = (LOWPASS(top, x + 1) + LOWPASS(left, y + 1)) >> 1; | |
314 | } | |
315 | ||
316 | static void intra_pred_down_left(uint8_t *d, uint8_t *top, uint8_t *left, int stride) | |
317 | { | |
318 | int x, y; | |
319 | for (y = 0; y < 8; y++) | |
320 | for (x = 0; x < 8; x++) | |
321 | d[y * stride + x] = (LOWPASS(top, x + y + 2) + LOWPASS(left, x + y + 2)) >> 1; | |
322 | } | |
323 | ||
324 | static void intra_pred_down_right(uint8_t *d, uint8_t *top, uint8_t *left, int stride) | |
325 | { | |
326 | int x, y; | |
327 | for (y = 0; y < 8; y++) | |
328 | for (x = 0; x < 8; x++) | |
329 | if (x == y) | |
330 | d[y * stride + x] = (left[1] + 2 * top[0] + top[1] + 2) >> 2; | |
331 | else if (x > y) | |
332 | d[y * stride + x] = LOWPASS(top, x - y); | |
333 | else | |
334 | d[y * stride + x] = LOWPASS(left, y - x); | |
335 | } | |
336 | ||
337 | static void intra_pred_lp_left(uint8_t *d, uint8_t *top, uint8_t *left, int stride) | |
338 | { | |
339 | int x, y; | |
340 | for (y = 0; y < 8; y++) | |
341 | for (x = 0; x < 8; x++) | |
342 | d[y * stride + x] = LOWPASS(left, y + 1); | |
343 | } | |
344 | ||
345 | static void intra_pred_lp_top(uint8_t *d, uint8_t *top, uint8_t *left, int stride) | |
346 | { | |
347 | int x, y; | |
348 | for (y = 0; y < 8; y++) | |
349 | for (x = 0; x < 8; x++) | |
350 | d[y * stride + x] = LOWPASS(top, x + 1); | |
351 | } | |
352 | ||
353 | #undef LOWPASS | |
354 | ||
355 | static inline void modify_pred(const int8_t *mod_table, int *mode) | |
356 | { | |
357 | *mode = mod_table[*mode]; | |
358 | if (*mode < 0) { | |
359 | av_log(NULL, AV_LOG_ERROR, "Illegal intra prediction mode\n"); | |
360 | *mode = 0; | |
361 | } | |
362 | } | |
363 | ||
364 | void ff_cavs_modify_mb_i(AVSContext *h, int *pred_mode_uv) | |
365 | { | |
366 | /* save pred modes before they get modified */ | |
367 | h->pred_mode_Y[3] = h->pred_mode_Y[5]; | |
368 | h->pred_mode_Y[6] = h->pred_mode_Y[8]; | |
369 | h->top_pred_Y[h->mbx * 2 + 0] = h->pred_mode_Y[7]; | |
370 | h->top_pred_Y[h->mbx * 2 + 1] = h->pred_mode_Y[8]; | |
371 | ||
372 | /* modify pred modes according to availability of neighbour samples */ | |
373 | if (!(h->flags & A_AVAIL)) { | |
374 | modify_pred(left_modifier_l, &h->pred_mode_Y[4]); | |
375 | modify_pred(left_modifier_l, &h->pred_mode_Y[7]); | |
376 | modify_pred(left_modifier_c, pred_mode_uv); | |
377 | } | |
378 | if (!(h->flags & B_AVAIL)) { | |
379 | modify_pred(top_modifier_l, &h->pred_mode_Y[4]); | |
380 | modify_pred(top_modifier_l, &h->pred_mode_Y[5]); | |
381 | modify_pred(top_modifier_c, pred_mode_uv); | |
382 | } | |
383 | } | |
384 | ||
385 | /***************************************************************************** | |
386 | * | |
387 | * motion compensation | |
388 | * | |
389 | ****************************************************************************/ | |
390 | ||
391 | static inline void mc_dir_part(AVSContext *h, AVFrame *pic, int chroma_height, | |
392 | int delta, int list, uint8_t *dest_y, | |
393 | uint8_t *dest_cb, uint8_t *dest_cr, | |
394 | int src_x_offset, int src_y_offset, | |
395 | qpel_mc_func *qpix_op, | |
396 | h264_chroma_mc_func chroma_op, cavs_vector *mv) | |
397 | { | |
398 | const int mx = mv->x + src_x_offset * 8; | |
399 | const int my = mv->y + src_y_offset * 8; | |
400 | const int luma_xy = (mx & 3) + ((my & 3) << 2); | |
401 | uint8_t *src_y = pic->data[0] + (mx >> 2) + (my >> 2) * h->l_stride; | |
402 | uint8_t *src_cb = pic->data[1] + (mx >> 3) + (my >> 3) * h->c_stride; | |
403 | uint8_t *src_cr = pic->data[2] + (mx >> 3) + (my >> 3) * h->c_stride; | |
404 | int extra_width = 0; | |
405 | int extra_height = extra_width; | |
406 | const int full_mx = mx >> 2; | |
407 | const int full_my = my >> 2; | |
408 | const int pic_width = 16 * h->mb_width; | |
409 | const int pic_height = 16 * h->mb_height; | |
410 | int emu = 0; | |
411 | ||
412 | if (!pic->data[0]) | |
413 | return; | |
414 | if (mx & 7) | |
415 | extra_width -= 3; | |
416 | if (my & 7) | |
417 | extra_height -= 3; | |
418 | ||
419 | if (full_mx < 0 - extra_width || | |
420 | full_my < 0 - extra_height || | |
421 | full_mx + 16 /* FIXME */ > pic_width + extra_width || | |
422 | full_my + 16 /* FIXME */ > pic_height + extra_height) { | |
423 | h->vdsp.emulated_edge_mc(h->edge_emu_buffer, | |
424 | src_y - 2 - 2 * h->l_stride, | |
425 | h->l_stride, h->l_stride, | |
426 | 16 + 5, 16 + 5 /* FIXME */, | |
427 | full_mx - 2, full_my - 2, | |
428 | pic_width, pic_height); | |
429 | src_y = h->edge_emu_buffer + 2 + 2 * h->l_stride; | |
430 | emu = 1; | |
431 | } | |
432 | ||
433 | // FIXME try variable height perhaps? | |
434 | qpix_op[luma_xy](dest_y, src_y, h->l_stride); | |
435 | ||
436 | if (emu) { | |
437 | h->vdsp.emulated_edge_mc(h->edge_emu_buffer, src_cb, | |
438 | h->c_stride, h->c_stride, | |
439 | 9, 9 /* FIXME */, | |
440 | mx >> 3, my >> 3, | |
441 | pic_width >> 1, pic_height >> 1); | |
442 | src_cb = h->edge_emu_buffer; | |
443 | } | |
444 | chroma_op(dest_cb, src_cb, h->c_stride, chroma_height, mx & 7, my & 7); | |
445 | ||
446 | if (emu) { | |
447 | h->vdsp.emulated_edge_mc(h->edge_emu_buffer, src_cr, | |
448 | h->c_stride, h->c_stride, | |
449 | 9, 9 /* FIXME */, | |
450 | mx >> 3, my >> 3, | |
451 | pic_width >> 1, pic_height >> 1); | |
452 | src_cr = h->edge_emu_buffer; | |
453 | } | |
454 | chroma_op(dest_cr, src_cr, h->c_stride, chroma_height, mx & 7, my & 7); | |
455 | } | |
456 | ||
457 | static inline void mc_part_std(AVSContext *h, int chroma_height, int delta, | |
458 | uint8_t *dest_y, | |
459 | uint8_t *dest_cb, | |
460 | uint8_t *dest_cr, | |
461 | int x_offset, int y_offset, | |
462 | qpel_mc_func *qpix_put, | |
463 | h264_chroma_mc_func chroma_put, | |
464 | qpel_mc_func *qpix_avg, | |
465 | h264_chroma_mc_func chroma_avg, | |
466 | cavs_vector *mv) | |
467 | { | |
468 | qpel_mc_func *qpix_op = qpix_put; | |
469 | h264_chroma_mc_func chroma_op = chroma_put; | |
470 | ||
471 | dest_y += x_offset * 2 + y_offset * h->l_stride * 2; | |
472 | dest_cb += x_offset + y_offset * h->c_stride; | |
473 | dest_cr += x_offset + y_offset * h->c_stride; | |
474 | x_offset += 8 * h->mbx; | |
475 | y_offset += 8 * h->mby; | |
476 | ||
477 | if (mv->ref >= 0) { | |
478 | AVFrame *ref = h->DPB[mv->ref].f; | |
479 | mc_dir_part(h, ref, chroma_height, delta, 0, | |
480 | dest_y, dest_cb, dest_cr, x_offset, y_offset, | |
481 | qpix_op, chroma_op, mv); | |
482 | ||
483 | qpix_op = qpix_avg; | |
484 | chroma_op = chroma_avg; | |
485 | } | |
486 | ||
487 | if ((mv + MV_BWD_OFFS)->ref >= 0) { | |
488 | AVFrame *ref = h->DPB[0].f; | |
489 | mc_dir_part(h, ref, chroma_height, delta, 1, | |
490 | dest_y, dest_cb, dest_cr, x_offset, y_offset, | |
491 | qpix_op, chroma_op, mv + MV_BWD_OFFS); | |
492 | } | |
493 | } | |
494 | ||
495 | void ff_cavs_inter(AVSContext *h, enum cavs_mb mb_type) | |
496 | { | |
497 | if (ff_cavs_partition_flags[mb_type] == 0) { // 16x16 | |
498 | mc_part_std(h, 8, 0, h->cy, h->cu, h->cv, 0, 0, | |
499 | h->cdsp.put_cavs_qpel_pixels_tab[0], | |
500 | h->h264chroma.put_h264_chroma_pixels_tab[0], | |
501 | h->cdsp.avg_cavs_qpel_pixels_tab[0], | |
502 | h->h264chroma.avg_h264_chroma_pixels_tab[0], | |
503 | &h->mv[MV_FWD_X0]); | |
504 | } else { | |
505 | mc_part_std(h, 4, 0, h->cy, h->cu, h->cv, 0, 0, | |
506 | h->cdsp.put_cavs_qpel_pixels_tab[1], | |
507 | h->h264chroma.put_h264_chroma_pixels_tab[1], | |
508 | h->cdsp.avg_cavs_qpel_pixels_tab[1], | |
509 | h->h264chroma.avg_h264_chroma_pixels_tab[1], | |
510 | &h->mv[MV_FWD_X0]); | |
511 | mc_part_std(h, 4, 0, h->cy, h->cu, h->cv, 4, 0, | |
512 | h->cdsp.put_cavs_qpel_pixels_tab[1], | |
513 | h->h264chroma.put_h264_chroma_pixels_tab[1], | |
514 | h->cdsp.avg_cavs_qpel_pixels_tab[1], | |
515 | h->h264chroma.avg_h264_chroma_pixels_tab[1], | |
516 | &h->mv[MV_FWD_X1]); | |
517 | mc_part_std(h, 4, 0, h->cy, h->cu, h->cv, 0, 4, | |
518 | h->cdsp.put_cavs_qpel_pixels_tab[1], | |
519 | h->h264chroma.put_h264_chroma_pixels_tab[1], | |
520 | h->cdsp.avg_cavs_qpel_pixels_tab[1], | |
521 | h->h264chroma.avg_h264_chroma_pixels_tab[1], | |
522 | &h->mv[MV_FWD_X2]); | |
523 | mc_part_std(h, 4, 0, h->cy, h->cu, h->cv, 4, 4, | |
524 | h->cdsp.put_cavs_qpel_pixels_tab[1], | |
525 | h->h264chroma.put_h264_chroma_pixels_tab[1], | |
526 | h->cdsp.avg_cavs_qpel_pixels_tab[1], | |
527 | h->h264chroma.avg_h264_chroma_pixels_tab[1], | |
528 | &h->mv[MV_FWD_X3]); | |
529 | } | |
530 | } | |
531 | ||
532 | /***************************************************************************** | |
533 | * | |
534 | * motion vector prediction | |
535 | * | |
536 | ****************************************************************************/ | |
537 | ||
538 | static inline void scale_mv(AVSContext *h, int *d_x, int *d_y, | |
539 | cavs_vector *src, int distp) | |
540 | { | |
541 | int den = h->scale_den[FFMAX(src->ref, 0)]; | |
542 | ||
f6fa7814 DM |
543 | *d_x = (src->x * distp * den + 256 + FF_SIGNBIT(src->x)) >> 9; |
544 | *d_y = (src->y * distp * den + 256 + FF_SIGNBIT(src->y)) >> 9; | |
2ba45a60 DM |
545 | } |
546 | ||
547 | static inline void mv_pred_median(AVSContext *h, | |
548 | cavs_vector *mvP, | |
549 | cavs_vector *mvA, | |
550 | cavs_vector *mvB, | |
551 | cavs_vector *mvC) | |
552 | { | |
553 | int ax, ay, bx, by, cx, cy; | |
554 | int len_ab, len_bc, len_ca, len_mid; | |
555 | ||
556 | /* scale candidates according to their temporal span */ | |
557 | scale_mv(h, &ax, &ay, mvA, mvP->dist); | |
558 | scale_mv(h, &bx, &by, mvB, mvP->dist); | |
559 | scale_mv(h, &cx, &cy, mvC, mvP->dist); | |
560 | /* find the geometrical median of the three candidates */ | |
561 | len_ab = abs(ax - bx) + abs(ay - by); | |
562 | len_bc = abs(bx - cx) + abs(by - cy); | |
563 | len_ca = abs(cx - ax) + abs(cy - ay); | |
564 | len_mid = mid_pred(len_ab, len_bc, len_ca); | |
565 | if (len_mid == len_ab) { | |
566 | mvP->x = cx; | |
567 | mvP->y = cy; | |
568 | } else if (len_mid == len_bc) { | |
569 | mvP->x = ax; | |
570 | mvP->y = ay; | |
571 | } else { | |
572 | mvP->x = bx; | |
573 | mvP->y = by; | |
574 | } | |
575 | } | |
576 | ||
577 | void ff_cavs_mv(AVSContext *h, enum cavs_mv_loc nP, enum cavs_mv_loc nC, | |
578 | enum cavs_mv_pred mode, enum cavs_block size, int ref) | |
579 | { | |
580 | cavs_vector *mvP = &h->mv[nP]; | |
581 | cavs_vector *mvA = &h->mv[nP-1]; | |
582 | cavs_vector *mvB = &h->mv[nP-4]; | |
583 | cavs_vector *mvC = &h->mv[nC]; | |
584 | const cavs_vector *mvP2 = NULL; | |
585 | ||
586 | mvP->ref = ref; | |
587 | mvP->dist = h->dist[mvP->ref]; | |
588 | if (mvC->ref == NOT_AVAIL || (nP == MV_FWD_X3) || (nP == MV_BWD_X3 )) | |
589 | mvC = &h->mv[nP - 5]; // set to top-left (mvD) | |
590 | if (mode == MV_PRED_PSKIP && | |
591 | (mvA->ref == NOT_AVAIL || | |
592 | mvB->ref == NOT_AVAIL || | |
593 | (mvA->x | mvA->y | mvA->ref) == 0 || | |
594 | (mvB->x | mvB->y | mvB->ref) == 0)) { | |
595 | mvP2 = &un_mv; | |
596 | /* if there is only one suitable candidate, take it */ | |
597 | } else if (mvA->ref >= 0 && mvB->ref < 0 && mvC->ref < 0) { | |
598 | mvP2 = mvA; | |
599 | } else if (mvA->ref < 0 && mvB->ref >= 0 && mvC->ref < 0) { | |
600 | mvP2 = mvB; | |
601 | } else if (mvA->ref < 0 && mvB->ref < 0 && mvC->ref >= 0) { | |
602 | mvP2 = mvC; | |
603 | } else if (mode == MV_PRED_LEFT && mvA->ref == ref) { | |
604 | mvP2 = mvA; | |
605 | } else if (mode == MV_PRED_TOP && mvB->ref == ref) { | |
606 | mvP2 = mvB; | |
607 | } else if (mode == MV_PRED_TOPRIGHT && mvC->ref == ref) { | |
608 | mvP2 = mvC; | |
609 | } | |
610 | if (mvP2) { | |
611 | mvP->x = mvP2->x; | |
612 | mvP->y = mvP2->y; | |
613 | } else | |
614 | mv_pred_median(h, mvP, mvA, mvB, mvC); | |
615 | ||
616 | if (mode < MV_PRED_PSKIP) { | |
617 | mvP->x += get_se_golomb(&h->gb); | |
618 | mvP->y += get_se_golomb(&h->gb); | |
619 | } | |
620 | set_mvs(mvP, size); | |
621 | } | |
622 | ||
623 | /***************************************************************************** | |
624 | * | |
625 | * macroblock level | |
626 | * | |
627 | ****************************************************************************/ | |
628 | ||
629 | /** | |
630 | * initialise predictors for motion vectors and intra prediction | |
631 | */ | |
632 | void ff_cavs_init_mb(AVSContext *h) | |
633 | { | |
634 | int i; | |
635 | ||
636 | /* copy predictors from top line (MB B and C) into cache */ | |
637 | for (i = 0; i < 3; i++) { | |
638 | h->mv[MV_FWD_B2 + i] = h->top_mv[0][h->mbx * 2 + i]; | |
639 | h->mv[MV_BWD_B2 + i] = h->top_mv[1][h->mbx * 2 + i]; | |
640 | } | |
641 | h->pred_mode_Y[1] = h->top_pred_Y[h->mbx * 2 + 0]; | |
642 | h->pred_mode_Y[2] = h->top_pred_Y[h->mbx * 2 + 1]; | |
643 | /* clear top predictors if MB B is not available */ | |
644 | if (!(h->flags & B_AVAIL)) { | |
645 | h->mv[MV_FWD_B2] = un_mv; | |
646 | h->mv[MV_FWD_B3] = un_mv; | |
647 | h->mv[MV_BWD_B2] = un_mv; | |
648 | h->mv[MV_BWD_B3] = un_mv; | |
649 | h->pred_mode_Y[1] = h->pred_mode_Y[2] = NOT_AVAIL; | |
650 | h->flags &= ~(C_AVAIL | D_AVAIL); | |
651 | } else if (h->mbx) { | |
652 | h->flags |= D_AVAIL; | |
653 | } | |
654 | if (h->mbx == h->mb_width - 1) // MB C not available | |
655 | h->flags &= ~C_AVAIL; | |
656 | /* clear top-right predictors if MB C is not available */ | |
657 | if (!(h->flags & C_AVAIL)) { | |
658 | h->mv[MV_FWD_C2] = un_mv; | |
659 | h->mv[MV_BWD_C2] = un_mv; | |
660 | } | |
661 | /* clear top-left predictors if MB D is not available */ | |
662 | if (!(h->flags & D_AVAIL)) { | |
663 | h->mv[MV_FWD_D3] = un_mv; | |
664 | h->mv[MV_BWD_D3] = un_mv; | |
665 | } | |
666 | } | |
667 | ||
668 | /** | |
669 | * save predictors for later macroblocks and increase | |
670 | * macroblock address | |
671 | * @return 0 if end of frame is reached, 1 otherwise | |
672 | */ | |
673 | int ff_cavs_next_mb(AVSContext *h) | |
674 | { | |
675 | int i; | |
676 | ||
677 | h->flags |= A_AVAIL; | |
678 | h->cy += 16; | |
679 | h->cu += 8; | |
680 | h->cv += 8; | |
681 | /* copy mvs as predictors to the left */ | |
682 | for (i = 0; i <= 20; i += 4) | |
683 | h->mv[i] = h->mv[i + 2]; | |
684 | /* copy bottom mvs from cache to top line */ | |
685 | h->top_mv[0][h->mbx * 2 + 0] = h->mv[MV_FWD_X2]; | |
686 | h->top_mv[0][h->mbx * 2 + 1] = h->mv[MV_FWD_X3]; | |
687 | h->top_mv[1][h->mbx * 2 + 0] = h->mv[MV_BWD_X2]; | |
688 | h->top_mv[1][h->mbx * 2 + 1] = h->mv[MV_BWD_X3]; | |
689 | /* next MB address */ | |
690 | h->mbidx++; | |
691 | h->mbx++; | |
692 | if (h->mbx == h->mb_width) { // New mb line | |
693 | h->flags = B_AVAIL | C_AVAIL; | |
694 | /* clear left pred_modes */ | |
695 | h->pred_mode_Y[3] = h->pred_mode_Y[6] = NOT_AVAIL; | |
696 | /* clear left mv predictors */ | |
697 | for (i = 0; i <= 20; i += 4) | |
698 | h->mv[i] = un_mv; | |
699 | h->mbx = 0; | |
700 | h->mby++; | |
701 | /* re-calculate sample pointers */ | |
702 | h->cy = h->cur.f->data[0] + h->mby * 16 * h->l_stride; | |
703 | h->cu = h->cur.f->data[1] + h->mby * 8 * h->c_stride; | |
704 | h->cv = h->cur.f->data[2] + h->mby * 8 * h->c_stride; | |
705 | if (h->mby == h->mb_height) { // Frame end | |
706 | return 0; | |
707 | } | |
708 | } | |
709 | return 1; | |
710 | } | |
711 | ||
712 | /***************************************************************************** | |
713 | * | |
714 | * frame level | |
715 | * | |
716 | ****************************************************************************/ | |
717 | ||
718 | int ff_cavs_init_pic(AVSContext *h) | |
719 | { | |
720 | int i; | |
721 | ||
722 | /* clear some predictors */ | |
723 | for (i = 0; i <= 20; i += 4) | |
724 | h->mv[i] = un_mv; | |
725 | h->mv[MV_BWD_X0] = ff_cavs_dir_mv; | |
726 | set_mvs(&h->mv[MV_BWD_X0], BLK_16X16); | |
727 | h->mv[MV_FWD_X0] = ff_cavs_dir_mv; | |
728 | set_mvs(&h->mv[MV_FWD_X0], BLK_16X16); | |
729 | h->pred_mode_Y[3] = h->pred_mode_Y[6] = NOT_AVAIL; | |
730 | h->cy = h->cur.f->data[0]; | |
731 | h->cu = h->cur.f->data[1]; | |
732 | h->cv = h->cur.f->data[2]; | |
733 | h->l_stride = h->cur.f->linesize[0]; | |
734 | h->c_stride = h->cur.f->linesize[1]; | |
735 | h->luma_scan[2] = 8 * h->l_stride; | |
736 | h->luma_scan[3] = 8 * h->l_stride + 8; | |
737 | h->mbx = h->mby = h->mbidx = 0; | |
738 | h->flags = 0; | |
739 | ||
740 | return 0; | |
741 | } | |
742 | ||
743 | /***************************************************************************** | |
744 | * | |
745 | * headers and interface | |
746 | * | |
747 | ****************************************************************************/ | |
748 | ||
749 | /** | |
750 | * some predictions require data from the top-neighbouring macroblock. | |
751 | * this data has to be stored for one complete row of macroblocks | |
752 | * and this storage space is allocated here | |
753 | */ | |
754 | void ff_cavs_init_top_lines(AVSContext *h) | |
755 | { | |
756 | /* alloc top line of predictors */ | |
757 | h->top_qp = av_mallocz(h->mb_width); | |
758 | h->top_mv[0] = av_mallocz_array(h->mb_width * 2 + 1, sizeof(cavs_vector)); | |
759 | h->top_mv[1] = av_mallocz_array(h->mb_width * 2 + 1, sizeof(cavs_vector)); | |
760 | h->top_pred_Y = av_mallocz_array(h->mb_width * 2, sizeof(*h->top_pred_Y)); | |
761 | h->top_border_y = av_mallocz_array(h->mb_width + 1, 16); | |
762 | h->top_border_u = av_mallocz_array(h->mb_width, 10); | |
763 | h->top_border_v = av_mallocz_array(h->mb_width, 10); | |
764 | ||
765 | /* alloc space for co-located MVs and types */ | |
766 | h->col_mv = av_mallocz_array(h->mb_width * h->mb_height, | |
767 | 4 * sizeof(cavs_vector)); | |
768 | h->col_type_base = av_mallocz(h->mb_width * h->mb_height); | |
769 | h->block = av_mallocz(64 * sizeof(int16_t)); | |
770 | } | |
771 | ||
772 | av_cold int ff_cavs_init(AVCodecContext *avctx) | |
773 | { | |
774 | AVSContext *h = avctx->priv_data; | |
775 | ||
776 | ff_blockdsp_init(&h->bdsp, avctx); | |
777 | ff_h264chroma_init(&h->h264chroma, 8); | |
778 | ff_idctdsp_init(&h->idsp, avctx); | |
779 | ff_videodsp_init(&h->vdsp, 8); | |
780 | ff_cavsdsp_init(&h->cdsp, avctx); | |
781 | ff_init_scantable_permutation(h->idsp.idct_permutation, | |
782 | h->cdsp.idct_perm); | |
783 | ff_init_scantable(h->idsp.idct_permutation, &h->scantable, ff_zigzag_direct); | |
784 | ||
785 | h->avctx = avctx; | |
786 | avctx->pix_fmt = AV_PIX_FMT_YUV420P; | |
787 | ||
788 | h->cur.f = av_frame_alloc(); | |
789 | h->DPB[0].f = av_frame_alloc(); | |
790 | h->DPB[1].f = av_frame_alloc(); | |
791 | if (!h->cur.f || !h->DPB[0].f || !h->DPB[1].f) { | |
792 | ff_cavs_end(avctx); | |
793 | return AVERROR(ENOMEM); | |
794 | } | |
795 | ||
796 | h->luma_scan[0] = 0; | |
797 | h->luma_scan[1] = 8; | |
798 | h->intra_pred_l[INTRA_L_VERT] = intra_pred_vert; | |
799 | h->intra_pred_l[INTRA_L_HORIZ] = intra_pred_horiz; | |
800 | h->intra_pred_l[INTRA_L_LP] = intra_pred_lp; | |
801 | h->intra_pred_l[INTRA_L_DOWN_LEFT] = intra_pred_down_left; | |
802 | h->intra_pred_l[INTRA_L_DOWN_RIGHT] = intra_pred_down_right; | |
803 | h->intra_pred_l[INTRA_L_LP_LEFT] = intra_pred_lp_left; | |
804 | h->intra_pred_l[INTRA_L_LP_TOP] = intra_pred_lp_top; | |
805 | h->intra_pred_l[INTRA_L_DC_128] = intra_pred_dc_128; | |
806 | h->intra_pred_c[INTRA_C_LP] = intra_pred_lp; | |
807 | h->intra_pred_c[INTRA_C_HORIZ] = intra_pred_horiz; | |
808 | h->intra_pred_c[INTRA_C_VERT] = intra_pred_vert; | |
809 | h->intra_pred_c[INTRA_C_PLANE] = intra_pred_plane; | |
810 | h->intra_pred_c[INTRA_C_LP_LEFT] = intra_pred_lp_left; | |
811 | h->intra_pred_c[INTRA_C_LP_TOP] = intra_pred_lp_top; | |
812 | h->intra_pred_c[INTRA_C_DC_128] = intra_pred_dc_128; | |
813 | h->mv[7] = un_mv; | |
814 | h->mv[19] = un_mv; | |
815 | return 0; | |
816 | } | |
817 | ||
818 | av_cold int ff_cavs_end(AVCodecContext *avctx) | |
819 | { | |
820 | AVSContext *h = avctx->priv_data; | |
821 | ||
822 | av_frame_free(&h->cur.f); | |
823 | av_frame_free(&h->DPB[0].f); | |
824 | av_frame_free(&h->DPB[1].f); | |
825 | ||
f6fa7814 DM |
826 | av_freep(&h->top_qp); |
827 | av_freep(&h->top_mv[0]); | |
828 | av_freep(&h->top_mv[1]); | |
829 | av_freep(&h->top_pred_Y); | |
830 | av_freep(&h->top_border_y); | |
831 | av_freep(&h->top_border_u); | |
832 | av_freep(&h->top_border_v); | |
833 | av_freep(&h->col_mv); | |
834 | av_freep(&h->col_type_base); | |
835 | av_freep(&h->block); | |
2ba45a60 DM |
836 | av_freep(&h->edge_emu_buffer); |
837 | return 0; | |
838 | } |