077304514651a266a38fe329617f23e5a7b9cc7f
[libav.git] / libavcodec / cavs.c
1 /*
2 * Chinese AVS video (AVS1-P2, JiZhun profile) decoder.
3 * Copyright (c) 2006 Stefan Gehrer <stefan.gehrer@gmx.de>
4 *
5 * This library is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU Lesser General Public
7 * License as published by the Free Software Foundation; either
8 * version 2 of the License, or (at your option) any later version.
9 *
10 * This library is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * Lesser General Public License for more details.
14 *
15 * You should have received a copy of the GNU Lesser General Public
16 * License along with this library; if not, write to the Free Software
17 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
18 */
19
20 #include "avcodec.h"
21 #include "bitstream.h"
22 #include "golomb.h"
23 #include "mpegvideo.h"
24 #include "cavsdata.h"
25
26 typedef struct {
27 MpegEncContext s;
28 Picture picture; //currently decoded frame
29 Picture DPB[2]; //reference frames
30 int dist[2]; //temporal distances from current frame to ref frames
31 int profile, level;
32 int aspect_ratio;
33 int mb_width, mb_height;
34 int pic_type;
35 int progressive;
36 int pic_structure;
37 int skip_mode_flag;
38 int loop_filter_disable;
39 int alpha_offset, beta_offset;
40 int ref_flag;
41 int mbx, mby;
42 int flags;
43 int stc;
44 uint8_t *cy, *cu, *cv;
45 int left_qp;
46 uint8_t *top_qp;
47
48 /* mv motion vector cache
49 0: D3 B2 B3 C2
50 4: A1 X0 X1 -
51 8: A3 X2 X3 -
52
53 X are the vectors in the current macroblock (5,6,9,10)
54 A is the macroblock to the left (4,8)
55 B is the macroblock to the top (1,2)
56 C is the macroblock to the top-right (3)
57 D is the macroblock to the top-left (0)
58
59 the same is repeated for backward motion vectors */
60 vector_t mv[2*4*3];
61 vector_t *top_mv[2];
62 vector_t *col_mv;
63
64 /* luma pred mode cache
65 0: -- B2 B3
66 3: A1 X0 X1
67 6: A3 X2 X3 */
68 int pred_mode_Y[3*3];
69 int *top_pred_Y;
70 int l_stride, c_stride;
71 int luma_scan[4];
72 int qp;
73 int qp_fixed;
74 int cbp;
75
76 /* intra prediction is done with un-deblocked samples
77 they are saved here before deblocking the MB */
78 uint8_t *top_border_y, *top_border_u, *top_border_v;
79 uint8_t left_border_y[16], left_border_u[10], left_border_v[10];
80 uint8_t topleft_border_y, topleft_border_u, topleft_border_v;
81
82 void (*intra_pred_l[8])(uint8_t *d,uint8_t *top,uint8_t *left,int stride);
83 void (*intra_pred_c[7])(uint8_t *d,uint8_t *top,uint8_t *left,int stride);
84 uint8_t *col_type_base;
85 uint8_t *col_type;
86 int sym_factor;
87 int direct_den[2];
88 int scale_den[2];
89 int got_keyframe;
90 } AVSContext;
91
92 /*****************************************************************************
93 *
94 * in-loop deblocking filter
95 *
96 ****************************************************************************/
97
98 static inline int get_bs_p(vector_t *mvP, vector_t *mvQ) {
99 if((mvP->ref == REF_INTRA) || (mvQ->ref == REF_INTRA))
100 return 2;
101 if(mvP->ref != mvQ->ref)
102 return 1;
103 if( (abs(mvP->x - mvQ->x) >= 4) || (abs(mvP->y - mvQ->y) >= 4) )
104 return 1;
105 return 0;
106 }
107
108 static inline int get_bs_b(vector_t *mvP, vector_t *mvQ) {
109 if((mvP->ref == REF_INTRA) || (mvQ->ref == REF_INTRA)) {
110 return 2;
111 } else {
112 vector_t *mvPbw = mvP + MV_BWD_OFFS;
113 vector_t *mvQbw = mvQ + MV_BWD_OFFS;
114 if( (abs( mvP->x - mvQ->x) >= 4) ||
115 (abs( mvP->y - mvQ->y) >= 4) ||
116 (abs(mvPbw->x - mvQbw->x) >= 4) ||
117 (abs(mvPbw->y - mvQbw->y) >= 4) )
118 return 1;
119 }
120 return 0;
121 }
122
123 /* boundary strength (bs) mapping:
124 *
125 * --4---5--
126 * 0 2 |
127 * | 6 | 7 |
128 * 1 3 |
129 * ---------
130 *
131 */
132
133 #define SET_PARAMS \
134 alpha = alpha_tab[clip(qp_avg + h->alpha_offset,0,63)]; \
135 beta = beta_tab[clip(qp_avg + h->beta_offset, 0,63)]; \
136 tc = tc_tab[clip(qp_avg + h->alpha_offset,0,63)];
137
138 static void filter_mb(AVSContext *h, enum mb_t mb_type) {
139 DECLARE_ALIGNED_8(uint8_t, bs[8]);
140 int qp_avg, alpha, beta, tc;
141 int i;
142
143 /* save un-deblocked lines */
144 h->topleft_border_y = h->top_border_y[h->mbx*16+15];
145 h->topleft_border_u = h->top_border_u[h->mbx*10+8];
146 h->topleft_border_v = h->top_border_v[h->mbx*10+8];
147 memcpy(&h->top_border_y[h->mbx*16], h->cy + 15* h->l_stride,16);
148 memcpy(&h->top_border_u[h->mbx*10+1], h->cu + 7* h->c_stride,8);
149 memcpy(&h->top_border_v[h->mbx*10+1], h->cv + 7* h->c_stride,8);
150 for(i=0;i<8;i++) {
151 h->left_border_y[i*2+0] = *(h->cy + 15 + (i*2+0)*h->l_stride);
152 h->left_border_y[i*2+1] = *(h->cy + 15 + (i*2+1)*h->l_stride);
153 h->left_border_u[i+1] = *(h->cu + 7 + i*h->c_stride);
154 h->left_border_v[i+1] = *(h->cv + 7 + i*h->c_stride);
155 }
156 if(!h->loop_filter_disable) {
157 /* clear bs */
158 *((uint64_t *)bs) = 0;
159 /* determine bs */
160 switch(mb_type) {
161 case I_8X8:
162 *((uint64_t *)bs) = 0x0202020202020202ULL;
163 break;
164 case P_8X8:
165 case P_8X16:
166 bs[2] = get_bs_p(&h->mv[MV_FWD_X0], &h->mv[MV_FWD_X1]);
167 bs[3] = get_bs_p(&h->mv[MV_FWD_X2], &h->mv[MV_FWD_X3]);
168 case P_16X8:
169 bs[6] = get_bs_p(&h->mv[MV_FWD_X0], &h->mv[MV_FWD_X2]);
170 bs[7] = get_bs_p(&h->mv[MV_FWD_X1], &h->mv[MV_FWD_X3]);
171 case P_16X16:
172 case P_SKIP:
173 bs[0] = get_bs_p(&h->mv[MV_FWD_A1], &h->mv[MV_FWD_X0]);
174 bs[1] = get_bs_p(&h->mv[MV_FWD_A3], &h->mv[MV_FWD_X2]);
175 bs[4] = get_bs_p(&h->mv[MV_FWD_B2], &h->mv[MV_FWD_X0]);
176 bs[5] = get_bs_p(&h->mv[MV_FWD_B3], &h->mv[MV_FWD_X1]);
177 break;
178 case B_SKIP:
179 case B_DIRECT:
180 case B_8X8:
181 bs[2] = get_bs_b(&h->mv[MV_FWD_X0], &h->mv[MV_FWD_X1]);
182 bs[3] = get_bs_b(&h->mv[MV_FWD_X2], &h->mv[MV_FWD_X3]);
183 bs[6] = get_bs_b(&h->mv[MV_FWD_X0], &h->mv[MV_FWD_X2]);
184 bs[7] = get_bs_b(&h->mv[MV_FWD_X1], &h->mv[MV_FWD_X3]);
185 case B_FWD_16X16:
186 case B_BWD_16X16:
187 case B_SYM_16X16:
188 bs[0] = get_bs_b(&h->mv[MV_FWD_A1], &h->mv[MV_FWD_X0]);
189 bs[1] = get_bs_b(&h->mv[MV_FWD_A3], &h->mv[MV_FWD_X2]);
190 bs[4] = get_bs_b(&h->mv[MV_FWD_B2], &h->mv[MV_FWD_X0]);
191 bs[5] = get_bs_b(&h->mv[MV_FWD_B3], &h->mv[MV_FWD_X1]);
192 break;
193 default:
194 if(mb_type & 1) { //16X8
195 bs[6] = bs[7] = get_bs_b(&h->mv[MV_FWD_X0], &h->mv[MV_FWD_X2]);
196 } else { //8X16
197 bs[2] = bs[3] = get_bs_b(&h->mv[MV_FWD_X0], &h->mv[MV_FWD_X1]);
198 }
199 bs[0] = get_bs_b(&h->mv[MV_FWD_A1], &h->mv[MV_FWD_X0]);
200 bs[1] = get_bs_b(&h->mv[MV_FWD_A3], &h->mv[MV_FWD_X2]);
201 bs[4] = get_bs_b(&h->mv[MV_FWD_B2], &h->mv[MV_FWD_X0]);
202 bs[5] = get_bs_b(&h->mv[MV_FWD_B3], &h->mv[MV_FWD_X1]);
203 }
204 if( *((uint64_t *)bs) ) {
205 if(h->flags & A_AVAIL) {
206 qp_avg = (h->qp + h->left_qp + 1) >> 1;
207 SET_PARAMS;
208 h->s.dsp.cavs_filter_lv(h->cy,h->l_stride,alpha,beta,tc,bs[0],bs[1]);
209 h->s.dsp.cavs_filter_cv(h->cu,h->c_stride,alpha,beta,tc,bs[0],bs[1]);
210 h->s.dsp.cavs_filter_cv(h->cv,h->c_stride,alpha,beta,tc,bs[0],bs[1]);
211 }
212 qp_avg = h->qp;
213 SET_PARAMS;
214 h->s.dsp.cavs_filter_lv(h->cy + 8,h->l_stride,alpha,beta,tc,bs[2],bs[3]);
215 h->s.dsp.cavs_filter_lh(h->cy + 8*h->l_stride,h->l_stride,alpha,beta,tc,
216 bs[6],bs[7]);
217
218 if(h->flags & B_AVAIL) {
219 qp_avg = (h->qp + h->top_qp[h->mbx] + 1) >> 1;
220 SET_PARAMS;
221 h->s.dsp.cavs_filter_lh(h->cy,h->l_stride,alpha,beta,tc,bs[4],bs[5]);
222 h->s.dsp.cavs_filter_ch(h->cu,h->c_stride,alpha,beta,tc,bs[4],bs[5]);
223 h->s.dsp.cavs_filter_ch(h->cv,h->c_stride,alpha,beta,tc,bs[4],bs[5]);
224 }
225 }
226 }
227 h->left_qp = h->qp;
228 h->top_qp[h->mbx] = h->qp;
229 }
230
231 #undef SET_PARAMS
232
233 /*****************************************************************************
234 *
235 * spatial intra prediction
236 *
237 ****************************************************************************/
238
239 static inline void load_intra_pred_luma(AVSContext *h, uint8_t *top,
240 uint8_t *left, int block) {
241 int i;
242
243 switch(block) {
244 case 0:
245 memcpy(&left[1],h->left_border_y,16);
246 left[0] = left[1];
247 left[17] = left[16];
248 memcpy(&top[1],&h->top_border_y[h->mbx*16],16);
249 top[17] = top[16];
250 top[0] = top[1];
251 if((h->flags & A_AVAIL) && (h->flags & B_AVAIL))
252 left[0] = top[0] = h->topleft_border_y;
253 break;
254 case 1:
255 for(i=0;i<8;i++)
256 left[i+1] = *(h->cy + 7 + i*h->l_stride);
257 memset(&left[9],left[8],9);
258 left[0] = left[1];
259 memcpy(&top[1],&h->top_border_y[h->mbx*16+8],8);
260 if(h->flags & C_AVAIL)
261 memcpy(&top[9],&h->top_border_y[(h->mbx + 1)*16],8);
262 else
263 memset(&top[9],top[8],9);
264 top[17] = top[16];
265 top[0] = top[1];
266 if(h->flags & B_AVAIL)
267 left[0] = top[0] = h->top_border_y[h->mbx*16+7];
268 break;
269 case 2:
270 memcpy(&left[1],&h->left_border_y[8],8);
271 memset(&left[9],left[8],9);
272 memcpy(&top[1],h->cy + 7*h->l_stride,16);
273 top[17] = top[16];
274 left[0] = h->left_border_y[7];
275 top[0] = top[1];
276 if(h->flags & A_AVAIL)
277 top[0] = left[0];
278 break;
279 case 3:
280 for(i=0;i<9;i++)
281 left[i] = *(h->cy + 7 + (i+7)*h->l_stride);
282 memset(&left[9],left[8],9);
283 memcpy(&top[0],h->cy + 7 + 7*h->l_stride,9);
284 memset(&top[9],top[8],9);
285 break;
286 }
287 }
288
289 static void intra_pred_vert(uint8_t *d,uint8_t *top,uint8_t *left,int stride) {
290 int y;
291 uint64_t a = *((uint64_t *)(&top[1]));
292 for(y=0;y<8;y++) {
293 *((uint64_t *)(d+y*stride)) = a;
294 }
295 }
296
297 static void intra_pred_horiz(uint8_t *d,uint8_t *top,uint8_t *left,int stride) {
298 int y;
299 uint64_t a;
300 for(y=0;y<8;y++) {
301 a = left[y+1] * 0x0101010101010101ULL;
302 *((uint64_t *)(d+y*stride)) = a;
303 }
304 }
305
306 static void intra_pred_dc_128(uint8_t *d,uint8_t *top,uint8_t *left,int stride) {
307 int y;
308 uint64_t a = 0x8080808080808080ULL;
309 for(y=0;y<8;y++)
310 *((uint64_t *)(d+y*stride)) = a;
311 }
312
313 static void intra_pred_plane(uint8_t *d,uint8_t *top,uint8_t *left,int stride) {
314 int x,y,ia;
315 int ih = 0;
316 int iv = 0;
317 uint8_t *cm = cropTbl + MAX_NEG_CROP;
318
319 for(x=0; x<4; x++) {
320 ih += (x+1)*(top[5+x]-top[3-x]);
321 iv += (x+1)*(left[5+x]-left[3-x]);
322 }
323 ia = (top[8]+left[8])<<4;
324 ih = (17*ih+16)>>5;
325 iv = (17*iv+16)>>5;
326 for(y=0; y<8; y++)
327 for(x=0; x<8; x++)
328 d[y*stride+x] = cm[(ia+(x-3)*ih+(y-3)*iv+16)>>5];
329 }
330
331 #define LOWPASS(ARRAY,INDEX) \
332 (( ARRAY[(INDEX)-1] + 2*ARRAY[(INDEX)] + ARRAY[(INDEX)+1] + 2) >> 2)
333
334 static void intra_pred_lp(uint8_t *d,uint8_t *top,uint8_t *left,int stride) {
335 int x,y;
336 for(y=0; y<8; y++)
337 for(x=0; x<8; x++)
338 d[y*stride+x] = (LOWPASS(top,x+1) + LOWPASS(left,y+1)) >> 1;
339 }
340
341 static void intra_pred_down_left(uint8_t *d,uint8_t *top,uint8_t *left,int stride) {
342 int x,y;
343 for(y=0; y<8; y++)
344 for(x=0; x<8; x++)
345 d[y*stride+x] = (LOWPASS(top,x+y+2) + LOWPASS(left,x+y+2)) >> 1;
346 }
347
348 static void intra_pred_down_right(uint8_t *d,uint8_t *top,uint8_t *left,int stride) {
349 int x,y;
350 for(y=0; y<8; y++)
351 for(x=0; x<8; x++)
352 if(x==y)
353 d[y*stride+x] = (left[1]+2*top[0]+top[1]+2)>>2;
354 else if(x>y)
355 d[y*stride+x] = LOWPASS(top,x-y);
356 else
357 d[y*stride+x] = LOWPASS(left,y-x);
358 }
359
360 static void intra_pred_lp_left(uint8_t *d,uint8_t *top,uint8_t *left,int stride) {
361 int x,y;
362 for(y=0; y<8; y++)
363 for(x=0; x<8; x++)
364 d[y*stride+x] = LOWPASS(left,y+1);
365 }
366
367 static void intra_pred_lp_top(uint8_t *d,uint8_t *top,uint8_t *left,int stride) {
368 int x,y;
369 for(y=0; y<8; y++)
370 for(x=0; x<8; x++)
371 d[y*stride+x] = LOWPASS(top,x+1);
372 }
373
374 #undef LOWPASS
375
376 static inline void modify_pred(const int_fast8_t *mod_table, int *mode) {
377 int newmode = mod_table[*mode];
378 if(newmode < 0) {
379 av_log(NULL, AV_LOG_ERROR, "Illegal intra prediction mode\n");
380 *mode = 0;
381 } else {
382 *mode = newmode;
383 }
384 }
385
386 /*****************************************************************************
387 *
388 * motion compensation
389 *
390 ****************************************************************************/
391
392 static inline void mc_dir_part(AVSContext *h,Picture *pic,int square,
393 int chroma_height,int delta,int list,uint8_t *dest_y,
394 uint8_t *dest_cb,uint8_t *dest_cr,int src_x_offset,
395 int src_y_offset,qpel_mc_func *qpix_op,
396 h264_chroma_mc_func chroma_op,vector_t *mv){
397 MpegEncContext * const s = &h->s;
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; //(s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16;
405 int extra_height= extra_width;
406 int emu=0;
407 const int full_mx= mx>>2;
408 const int full_my= my>>2;
409 const int pic_width = 16*h->mb_width;
410 const int pic_height = 16*h->mb_height;
411
412 if(!pic->data[0])
413 return;
414 if(mx&7) extra_width -= 3;
415 if(my&7) extra_height -= 3;
416
417 if( full_mx < 0-extra_width
418 || full_my < 0-extra_height
419 || full_mx + 16/*FIXME*/ > pic_width + extra_width
420 || full_my + 16/*FIXME*/ > pic_height + extra_height){
421 ff_emulated_edge_mc(s->edge_emu_buffer, src_y - 2 - 2*h->l_stride, h->l_stride,
422 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
423 src_y= s->edge_emu_buffer + 2 + 2*h->l_stride;
424 emu=1;
425 }
426
427 qpix_op[luma_xy](dest_y, src_y, h->l_stride); //FIXME try variable height perhaps?
428 if(!square){
429 qpix_op[luma_xy](dest_y + delta, src_y + delta, h->l_stride);
430 }
431
432 if(emu){
433 ff_emulated_edge_mc(s->edge_emu_buffer, src_cb, h->c_stride,
434 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
435 src_cb= s->edge_emu_buffer;
436 }
437 chroma_op(dest_cb, src_cb, h->c_stride, chroma_height, mx&7, my&7);
438
439 if(emu){
440 ff_emulated_edge_mc(s->edge_emu_buffer, src_cr, h->c_stride,
441 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
442 src_cr= s->edge_emu_buffer;
443 }
444 chroma_op(dest_cr, src_cr, h->c_stride, chroma_height, mx&7, my&7);
445 }
446
447 static inline void mc_part_std(AVSContext *h,int square,int chroma_height,int delta,
448 uint8_t *dest_y,uint8_t *dest_cb,uint8_t *dest_cr,
449 int x_offset, int y_offset,qpel_mc_func *qpix_put,
450 h264_chroma_mc_func chroma_put,qpel_mc_func *qpix_avg,
451 h264_chroma_mc_func chroma_avg, vector_t *mv){
452 qpel_mc_func *qpix_op= qpix_put;
453 h264_chroma_mc_func chroma_op= chroma_put;
454
455 dest_y += 2*x_offset + 2*y_offset*h->l_stride;
456 dest_cb += x_offset + y_offset*h->c_stride;
457 dest_cr += x_offset + y_offset*h->c_stride;
458 x_offset += 8*h->mbx;
459 y_offset += 8*h->mby;
460
461 if(mv->ref >= 0){
462 Picture *ref= &h->DPB[mv->ref];
463 mc_dir_part(h, ref, square, chroma_height, delta, 0,
464 dest_y, dest_cb, dest_cr, x_offset, y_offset,
465 qpix_op, chroma_op, mv);
466
467 qpix_op= qpix_avg;
468 chroma_op= chroma_avg;
469 }
470
471 if((mv+MV_BWD_OFFS)->ref >= 0){
472 Picture *ref= &h->DPB[0];
473 mc_dir_part(h, ref, square, chroma_height, delta, 1,
474 dest_y, dest_cb, dest_cr, x_offset, y_offset,
475 qpix_op, chroma_op, mv+MV_BWD_OFFS);
476 }
477 }
478
479 static void inter_pred(AVSContext *h) {
480 /* always do 8x8 blocks TODO: are larger blocks worth it? */
481 mc_part_std(h, 1, 4, 0, h->cy, h->cu, h->cv, 0, 0,
482 h->s.dsp.put_cavs_qpel_pixels_tab[1],
483 h->s.dsp.put_h264_chroma_pixels_tab[1],
484 h->s.dsp.avg_cavs_qpel_pixels_tab[1],
485 h->s.dsp.avg_h264_chroma_pixels_tab[1],&h->mv[MV_FWD_X0]);
486 mc_part_std(h, 1, 4, 0, h->cy, h->cu, h->cv, 4, 0,
487 h->s.dsp.put_cavs_qpel_pixels_tab[1],
488 h->s.dsp.put_h264_chroma_pixels_tab[1],
489 h->s.dsp.avg_cavs_qpel_pixels_tab[1],
490 h->s.dsp.avg_h264_chroma_pixels_tab[1],&h->mv[MV_FWD_X1]);
491 mc_part_std(h, 1, 4, 0, h->cy, h->cu, h->cv, 0, 4,
492 h->s.dsp.put_cavs_qpel_pixels_tab[1],
493 h->s.dsp.put_h264_chroma_pixels_tab[1],
494 h->s.dsp.avg_cavs_qpel_pixels_tab[1],
495 h->s.dsp.avg_h264_chroma_pixels_tab[1],&h->mv[MV_FWD_X2]);
496 mc_part_std(h, 1, 4, 0, h->cy, h->cu, h->cv, 4, 4,
497 h->s.dsp.put_cavs_qpel_pixels_tab[1],
498 h->s.dsp.put_h264_chroma_pixels_tab[1],
499 h->s.dsp.avg_cavs_qpel_pixels_tab[1],
500 h->s.dsp.avg_h264_chroma_pixels_tab[1],&h->mv[MV_FWD_X3]);
501 /* set intra prediction modes to default values */
502 h->pred_mode_Y[3] = h->pred_mode_Y[6] = INTRA_L_LP;
503 h->top_pred_Y[h->mbx*2+0] = h->top_pred_Y[h->mbx*2+1] = INTRA_L_LP;
504 }
505
506 /*****************************************************************************
507 *
508 * motion vector prediction
509 *
510 ****************************************************************************/
511
512 static inline void set_mvs(vector_t *mv, enum block_t size) {
513 switch(size) {
514 case BLK_16X16:
515 mv[MV_STRIDE ] = mv[0];
516 mv[MV_STRIDE+1] = mv[0];
517 case BLK_16X8:
518 mv[1] = mv[0];
519 break;
520 case BLK_8X16:
521 mv[MV_STRIDE] = mv[0];
522 break;
523 }
524 }
525
526 static inline void store_mvs(AVSContext *h) {
527 h->col_mv[(h->mby*h->mb_width + h->mbx)*4 + 0] = h->mv[MV_FWD_X0];
528 h->col_mv[(h->mby*h->mb_width + h->mbx)*4 + 1] = h->mv[MV_FWD_X1];
529 h->col_mv[(h->mby*h->mb_width + h->mbx)*4 + 2] = h->mv[MV_FWD_X2];
530 h->col_mv[(h->mby*h->mb_width + h->mbx)*4 + 3] = h->mv[MV_FWD_X3];
531 }
532
533 static inline void scale_mv(AVSContext *h, int *d_x, int *d_y, vector_t *src, int distp) {
534 int den = h->scale_den[src->ref];
535
536 *d_x = (src->x*distp*den + 256 + (src->x>>31)) >> 9;
537 *d_y = (src->y*distp*den + 256 + (src->y>>31)) >> 9;
538 }
539
540 static inline void mv_pred_median(AVSContext *h, vector_t *mvP, vector_t *mvA, vector_t *mvB, vector_t *mvC) {
541 int ax, ay, bx, by, cx, cy;
542 int len_ab, len_bc, len_ca, len_mid;
543
544 /* scale candidates according to their temporal span */
545 scale_mv(h, &ax, &ay, mvA, mvP->dist);
546 scale_mv(h, &bx, &by, mvB, mvP->dist);
547 scale_mv(h, &cx, &cy, mvC, mvP->dist);
548 /* find the geometrical median of the three candidates */
549 len_ab = abs(ax - bx) + abs(ay - by);
550 len_bc = abs(bx - cx) + abs(by - cy);
551 len_ca = abs(cx - ax) + abs(cy - ay);
552 len_mid = mid_pred(len_ab, len_bc, len_ca);
553 if(len_mid == len_ab) {
554 mvP->x = cx;
555 mvP->y = cy;
556 } else if(len_mid == len_bc) {
557 mvP->x = ax;
558 mvP->y = ay;
559 } else {
560 mvP->x = bx;
561 mvP->y = by;
562 }
563 }
564
565 static inline void mv_pred_direct(AVSContext *h, vector_t *pmv_fw,
566 vector_t *pmv_bw, vector_t *col_mv) {
567 int den = h->direct_den[col_mv->ref];
568 int m = col_mv->x >> 31;
569
570 pmv_fw->dist = h->dist[1];
571 pmv_bw->dist = h->dist[0];
572 pmv_fw->ref = 1;
573 pmv_bw->ref = 0;
574 /* scale the co-located motion vector according to its temporal span */
575 pmv_fw->x = (((den+(den*col_mv->x*pmv_fw->dist^m)-m-1)>>14)^m)-m;
576 pmv_bw->x = m-(((den+(den*col_mv->x*pmv_bw->dist^m)-m-1)>>14)^m);
577 m = col_mv->y >> 31;
578 pmv_fw->y = (((den+(den*col_mv->y*pmv_fw->dist^m)-m-1)>>14)^m)-m;
579 pmv_bw->y = m-(((den+(den*col_mv->y*pmv_bw->dist^m)-m-1)>>14)^m);
580 }
581
582 static inline void mv_pred_sym(AVSContext *h, vector_t *src, enum block_t size) {
583 vector_t *dst = src + MV_BWD_OFFS;
584
585 /* backward mv is the scaled and negated forward mv */
586 dst->x = -((src->x * h->sym_factor + 256) >> 9);
587 dst->y = -((src->y * h->sym_factor + 256) >> 9);
588 dst->ref = 0;
589 dst->dist = h->dist[0];
590 set_mvs(dst, size);
591 }
592
593 static void mv_pred(AVSContext *h, enum mv_loc_t nP, enum mv_loc_t nC,
594 enum mv_pred_t mode, enum block_t size, int ref) {
595 vector_t *mvP = &h->mv[nP];
596 vector_t *mvA = &h->mv[nP-1];
597 vector_t *mvB = &h->mv[nP-4];
598 vector_t *mvC = &h->mv[nC];
599 int mvAref = mvA->ref;
600 int mvBref = mvB->ref;
601 int mvCref;
602
603 mvP->ref = ref;
604 mvP->dist = h->dist[mvP->ref];
605 if(mvC->ref == NOT_AVAIL)
606 mvC = &h->mv[nP-5]; // set to top-left (mvD)
607 mvCref = mvC->ref;
608 if(mode == MV_PRED_PSKIP) {
609 if((mvAref == NOT_AVAIL) || (mvBref == NOT_AVAIL) ||
610 ((mvA->x | mvA->y | mvA->ref) == 0) ||
611 ((mvB->x | mvB->y | mvB->ref) == 0) ) {
612 mvP->x = mvP->y = 0;
613 set_mvs(mvP,size);
614 return;
615 }
616 }
617 /* if there is only one suitable candidate, take it */
618 if((mvAref >= 0) && (mvBref < 0) && (mvCref < 0)) {
619 mvP->x = mvA->x;
620 mvP->y = mvA->y;
621 } else if((mvAref < 0) && (mvBref >= 0) && (mvCref < 0)) {
622 mvP->x = mvB->x;
623 mvP->y = mvB->y;
624 } else if((mvAref < 0) && (mvBref < 0) && (mvCref >= 0)) {
625 mvP->x = mvC->x;
626 mvP->y = mvC->y;
627 } else {
628 switch(mode) {
629 case MV_PRED_LEFT:
630 if(mvAref == mvP->ref) {
631 mvP->x = mvA->x;
632 mvP->y = mvA->y;
633 } else
634 mv_pred_median(h, mvP, mvA, mvB, mvC);
635 break;
636 case MV_PRED_TOP:
637 if(mvBref == mvP->ref) {
638 mvP->x = mvB->x;
639 mvP->y = mvB->y;
640 } else
641 mv_pred_median(h, mvP, mvA, mvB, mvC);
642 break;
643 case MV_PRED_TOPRIGHT:
644 if(mvCref == mvP->ref) {
645 mvP->x = mvC->x;
646 mvP->y = mvC->y;
647 } else
648 mv_pred_median(h, mvP, mvA, mvB, mvC);
649 break;
650 default:
651 mv_pred_median(h, mvP, mvA, mvB, mvC);
652 break;
653 }
654 }
655 if(mode < MV_PRED_PSKIP) {
656 mvP->x += get_se_golomb(&h->s.gb);
657 mvP->y += get_se_golomb(&h->s.gb);
658 }
659 set_mvs(mvP,size);
660 }
661
662 /*****************************************************************************
663 *
664 * residual data decoding
665 *
666 ****************************************************************************/
667
668 /* kth-order exponential golomb code */
669 static inline int get_ue_code(GetBitContext *gb, int order) {
670 if(order) {
671 int ret = get_ue_golomb(gb) << order;
672 return ret + get_bits(gb,order);
673 }
674 return get_ue_golomb(gb);
675 }
676
677 static int decode_residual_block(AVSContext *h, GetBitContext *gb,
678 const residual_vlc_t *r, int esc_golomb_order,
679 int qp, uint8_t *dst, int stride) {
680 int i,pos = -1;
681 int level_code, esc_code, level, run, mask;
682 int level_buf[64];
683 int run_buf[64];
684 int dqm = dequant_mul[qp];
685 int dqs = dequant_shift[qp];
686 int dqa = 1 << (dqs - 1);
687 const uint8_t *scantab = ff_zigzag_direct;
688 DCTELEM block[64];
689
690 memset(block,0,64*sizeof(DCTELEM));
691 for(i=0;i<65;i++) {
692 level_code = get_ue_code(gb,r->golomb_order);
693 if(level_code >= ESCAPE_CODE) {
694 run = (level_code - ESCAPE_CODE) >> 1;
695 esc_code = get_ue_code(gb,esc_golomb_order);
696 level = esc_code + (run > r->max_run ? 1 : r->level_add[run]);
697 while(level > r->inc_limit)
698 r++;
699 mask = -(level_code & 1);
700 level = (level^mask) - mask;
701 } else {
702 if(level_code < 0)
703 return -1;
704 level = r->rltab[level_code][0];
705 if(!level) //end of block signal
706 break;
707 run = r->rltab[level_code][1];
708 r += r->rltab[level_code][2];
709 }
710 level_buf[i] = level;
711 run_buf[i] = run;
712 }
713 /* inverse scan and dequantization */
714 while(--i >= 0){
715 pos += 1 + run_buf[i];
716 if(pos > 63) {
717 av_log(h->s.avctx, AV_LOG_ERROR,
718 "position out of block bounds at pic %d MB(%d,%d)\n",
719 h->picture.poc, h->mbx, h->mby);
720 return -1;
721 }
722 block[scantab[pos]] = (level_buf[i]*dqm + dqa) >> dqs;
723 }
724 h->s.dsp.cavs_idct8_add(dst,block,stride);
725 return 0;
726 }
727
728
729 static inline void decode_residual_chroma(AVSContext *h) {
730 if(h->cbp & (1<<4))
731 decode_residual_block(h,&h->s.gb,chroma_2dvlc,0, chroma_qp[h->qp],
732 h->cu,h->c_stride);
733 if(h->cbp & (1<<5))
734 decode_residual_block(h,&h->s.gb,chroma_2dvlc,0, chroma_qp[h->qp],
735 h->cv,h->c_stride);
736 }
737
738 static inline void decode_residual_inter(AVSContext *h) {
739 int block;
740
741 /* get coded block pattern */
742 h->cbp = cbp_tab[get_ue_golomb(&h->s.gb)][1];
743 /* get quantizer */
744 if(h->cbp && !h->qp_fixed)
745 h->qp += get_se_golomb(&h->s.gb);
746 for(block=0;block<4;block++)
747 if(h->cbp & (1<<block))
748 decode_residual_block(h,&h->s.gb,inter_2dvlc,0,h->qp,
749 h->cy + h->luma_scan[block], h->l_stride);
750 decode_residual_chroma(h);
751 }
752
753 /*****************************************************************************
754 *
755 * macroblock level
756 *
757 ****************************************************************************/
758
759 static inline void init_mb(AVSContext *h) {
760 int i;
761
762 /* copy predictors from top line (MB B and C) into cache */
763 for(i=0;i<3;i++) {
764 h->mv[MV_FWD_B2+i] = h->top_mv[0][h->mbx*2+i];
765 h->mv[MV_BWD_B2+i] = h->top_mv[1][h->mbx*2+i];
766 }
767 h->pred_mode_Y[1] = h->top_pred_Y[h->mbx*2+0];
768 h->pred_mode_Y[2] = h->top_pred_Y[h->mbx*2+1];
769 /* clear top predictors if MB B is not available */
770 if(!(h->flags & B_AVAIL)) {
771 h->mv[MV_FWD_B2] = un_mv;
772 h->mv[MV_FWD_B3] = un_mv;
773 h->mv[MV_BWD_B2] = un_mv;
774 h->mv[MV_BWD_B3] = un_mv;
775 h->pred_mode_Y[1] = h->pred_mode_Y[2] = NOT_AVAIL;
776 h->flags &= ~(C_AVAIL|D_AVAIL);
777 } else if(h->mbx) {
778 h->flags |= D_AVAIL;
779 }
780 if(h->mbx == h->mb_width-1) //MB C not available
781 h->flags &= ~C_AVAIL;
782 /* clear top-right predictors if MB C is not available */
783 if(!(h->flags & C_AVAIL)) {
784 h->mv[MV_FWD_C2] = un_mv;
785 h->mv[MV_BWD_C2] = un_mv;
786 }
787 /* clear top-left predictors if MB D is not available */
788 if(!(h->flags & D_AVAIL)) {
789 h->mv[MV_FWD_D3] = un_mv;
790 h->mv[MV_BWD_D3] = un_mv;
791 }
792 /* set pointer for co-located macroblock type */
793 h->col_type = &h->col_type_base[h->mby*h->mb_width + h->mbx];
794 }
795
796 static inline void check_for_slice(AVSContext *h);
797
798 static inline int next_mb(AVSContext *h) {
799 int i;
800
801 h->flags |= A_AVAIL;
802 h->cy += 16;
803 h->cu += 8;
804 h->cv += 8;
805 /* copy mvs as predictors to the left */
806 for(i=0;i<=20;i+=4)
807 h->mv[i] = h->mv[i+2];
808 /* copy bottom mvs from cache to top line */
809 h->top_mv[0][h->mbx*2+0] = h->mv[MV_FWD_X2];
810 h->top_mv[0][h->mbx*2+1] = h->mv[MV_FWD_X3];
811 h->top_mv[1][h->mbx*2+0] = h->mv[MV_BWD_X2];
812 h->top_mv[1][h->mbx*2+1] = h->mv[MV_BWD_X3];
813 /* next MB address */
814 h->mbx++;
815 if(h->mbx == h->mb_width) { //new mb line
816 h->flags = B_AVAIL|C_AVAIL;
817 /* clear left pred_modes */
818 h->pred_mode_Y[3] = h->pred_mode_Y[6] = NOT_AVAIL;
819 /* clear left mv predictors */
820 for(i=0;i<=20;i+=4)
821 h->mv[i] = un_mv;
822 h->mbx = 0;
823 h->mby++;
824 /* re-calculate sample pointers */
825 h->cy = h->picture.data[0] + h->mby*16*h->l_stride;
826 h->cu = h->picture.data[1] + h->mby*8*h->c_stride;
827 h->cv = h->picture.data[2] + h->mby*8*h->c_stride;
828 if(h->mby == h->mb_height) { //frame end
829 return 0;
830 } else {
831 //check_for_slice(h);
832 }
833 }
834 return 1;
835 }
836
837 static void decode_mb_i(AVSContext *h, int is_i_pic) {
838 GetBitContext *gb = &h->s.gb;
839 int block, pred_mode_uv;
840 uint8_t top[18];
841 uint8_t left[18];
842 uint8_t *d;
843
844 /* get intra prediction modes from stream */
845 for(block=0;block<4;block++) {
846 int nA,nB,predpred;
847 int pos = scan3x3[block];
848
849 nA = h->pred_mode_Y[pos-1];
850 nB = h->pred_mode_Y[pos-3];
851 if((nA == NOT_AVAIL) || (nB == NOT_AVAIL))
852 predpred = 2;
853 else
854 predpred = FFMIN(nA,nB);
855 if(get_bits1(gb))
856 h->pred_mode_Y[pos] = predpred;
857 else {
858 h->pred_mode_Y[pos] = get_bits(gb,2);
859 if(h->pred_mode_Y[pos] >= predpred)
860 h->pred_mode_Y[pos]++;
861 }
862 }
863 pred_mode_uv = get_ue_golomb(gb);
864 if(pred_mode_uv > 6) {
865 av_log(h->s.avctx, AV_LOG_ERROR, "illegal intra chroma pred mode\n");
866 pred_mode_uv = 0;
867 }
868
869 /* save pred modes before they get modified */
870 h->pred_mode_Y[3] = h->pred_mode_Y[5];
871 h->pred_mode_Y[6] = h->pred_mode_Y[8];
872 h->top_pred_Y[h->mbx*2+0] = h->pred_mode_Y[7];
873 h->top_pred_Y[h->mbx*2+1] = h->pred_mode_Y[8];
874
875 /* modify pred modes according to availability of neighbour samples */
876 if(!(h->flags & A_AVAIL)) {
877 modify_pred(left_modifier_l, &h->pred_mode_Y[4] );
878 modify_pred(left_modifier_l, &h->pred_mode_Y[7] );
879 modify_pred(left_modifier_c, &pred_mode_uv );
880 }
881 if(!(h->flags & B_AVAIL)) {
882 modify_pred(top_modifier_l, &h->pred_mode_Y[4] );
883 modify_pred(top_modifier_l, &h->pred_mode_Y[5] );
884 modify_pred(top_modifier_c, &pred_mode_uv );
885 }
886
887 /* get coded block pattern */
888 if(is_i_pic)
889 h->cbp = cbp_tab[get_ue_golomb(gb)][0];
890 if(h->cbp && !h->qp_fixed)
891 h->qp += get_se_golomb(gb); //qp_delta
892
893 /* luma intra prediction interleaved with residual decode/transform/add */
894 for(block=0;block<4;block++) {
895 d = h->cy + h->luma_scan[block];
896 load_intra_pred_luma(h, top, left, block);
897 h->intra_pred_l[h->pred_mode_Y[scan3x3[block]]]
898 (d, top, left, h->l_stride);
899 if(h->cbp & (1<<block))
900 decode_residual_block(h,gb,intra_2dvlc,1,h->qp,d,h->l_stride);
901 }
902
903 /* chroma intra prediction */
904 /* extend borders by one pixel */
905 h->left_border_u[9] = h->left_border_u[8];
906 h->left_border_v[9] = h->left_border_v[8];
907 h->top_border_u[h->mbx*10+9] = h->top_border_u[h->mbx*10+8];
908 h->top_border_v[h->mbx*10+9] = h->top_border_v[h->mbx*10+8];
909 if(h->mbx && h->mby) {
910 h->top_border_u[h->mbx*10] = h->left_border_u[0] = h->topleft_border_u;
911 h->top_border_v[h->mbx*10] = h->left_border_v[0] = h->topleft_border_v;
912 } else {
913 h->left_border_u[0] = h->left_border_u[1];
914 h->left_border_v[0] = h->left_border_v[1];
915 h->top_border_u[h->mbx*10] = h->top_border_u[h->mbx*10+1];
916 h->top_border_v[h->mbx*10] = h->top_border_v[h->mbx*10+1];
917 }
918 h->intra_pred_c[pred_mode_uv](h->cu, &h->top_border_u[h->mbx*10],
919 h->left_border_u, h->c_stride);
920 h->intra_pred_c[pred_mode_uv](h->cv, &h->top_border_v[h->mbx*10],
921 h->left_border_v, h->c_stride);
922
923 decode_residual_chroma(h);
924 filter_mb(h,I_8X8);
925
926 /* mark motion vectors as intra */
927 h->mv[MV_FWD_X0] = intra_mv;
928 set_mvs(&h->mv[MV_FWD_X0], BLK_16X16);
929 h->mv[MV_BWD_X0] = intra_mv;
930 set_mvs(&h->mv[MV_BWD_X0], BLK_16X16);
931 if(h->pic_type != FF_B_TYPE)
932 *h->col_type = I_8X8;
933 }
934
935 static void mb_skip_p(AVSContext *h) {
936 mv_pred(h, MV_FWD_X0, MV_FWD_C2, MV_PRED_PSKIP, BLK_16X16, 0);
937 inter_pred(h);
938 store_mvs(h);
939 filter_mb(h,P_SKIP);
940 *h->col_type = P_SKIP;
941 }
942
943
944 static void mb_skip_b(AVSContext *h) {
945 int i;
946
947 if(!(*h->col_type)) {
948 /* intra MB at co-location, do in-plane prediction */
949 mv_pred(h, MV_FWD_X0, MV_FWD_C2, MV_PRED_BSKIP, BLK_16X16, 1);
950 mv_pred(h, MV_BWD_X0, MV_BWD_C2, MV_PRED_BSKIP, BLK_16X16, 0);
951 } else {
952 /* direct prediction from co-located P MB, block-wise */
953 for(i=0;i<4;i++)
954 mv_pred_direct(h,&h->mv[mv_scan[i]],
955 &h->mv[mv_scan[i]+MV_BWD_OFFS],
956 &h->col_mv[(h->mby*h->mb_width + h->mbx)*4 + i]);
957 }
958 }
959
960 static void decode_mb_p(AVSContext *h, enum mb_t mb_type) {
961 GetBitContext *gb = &h->s.gb;
962 int ref[4];
963
964 switch(mb_type) {
965 case P_SKIP:
966 mb_skip_p(h);
967 return;
968 case P_16X16:
969 ref[0] = h->ref_flag ? 0 : get_bits1(gb);
970 mv_pred(h, MV_FWD_X0, MV_FWD_C2, MV_PRED_MEDIAN, BLK_16X16,ref[0]);
971 break;
972 case P_16X8:
973 ref[0] = h->ref_flag ? 0 : get_bits1(gb);
974 ref[2] = h->ref_flag ? 0 : get_bits1(gb);
975 mv_pred(h, MV_FWD_X0, MV_FWD_C2, MV_PRED_TOP, BLK_16X8, ref[0]);
976 mv_pred(h, MV_FWD_X2, MV_FWD_A1, MV_PRED_LEFT, BLK_16X8, ref[2]);
977 break;
978 case P_8X16:
979 ref[0] = h->ref_flag ? 0 : get_bits1(gb);
980 ref[1] = h->ref_flag ? 0 : get_bits1(gb);
981 mv_pred(h, MV_FWD_X0, MV_FWD_B3, MV_PRED_LEFT, BLK_8X16, ref[0]);
982 mv_pred(h, MV_FWD_X1, MV_FWD_C2, MV_PRED_TOPRIGHT, BLK_8X16, ref[1]);
983 break;
984 case P_8X8:
985 ref[0] = h->ref_flag ? 0 : get_bits1(gb);
986 ref[1] = h->ref_flag ? 0 : get_bits1(gb);
987 ref[2] = h->ref_flag ? 0 : get_bits1(gb);
988 ref[3] = h->ref_flag ? 0 : get_bits1(gb);
989 mv_pred(h, MV_FWD_X0, MV_FWD_B3, MV_PRED_MEDIAN, BLK_8X8, ref[0]);
990 mv_pred(h, MV_FWD_X1, MV_FWD_C2, MV_PRED_MEDIAN, BLK_8X8, ref[1]);
991 mv_pred(h, MV_FWD_X2, MV_FWD_X1, MV_PRED_MEDIAN, BLK_8X8, ref[2]);
992 mv_pred(h, MV_FWD_X3, MV_FWD_X0, MV_PRED_MEDIAN, BLK_8X8, ref[3]);
993 }
994 inter_pred(h);
995 store_mvs(h);
996 decode_residual_inter(h);
997 filter_mb(h,mb_type);
998 *h->col_type = mb_type;
999 }
1000
1001 static void decode_mb_b(AVSContext *h, enum mb_t mb_type) {
1002 int block;
1003 enum sub_mb_t sub_type[4];
1004 int flags;
1005
1006 /* reset all MVs */
1007 h->mv[MV_FWD_X0] = dir_mv;
1008 set_mvs(&h->mv[MV_FWD_X0], BLK_16X16);
1009 h->mv[MV_BWD_X0] = dir_mv;
1010 set_mvs(&h->mv[MV_BWD_X0], BLK_16X16);
1011 switch(mb_type) {
1012 case B_SKIP:
1013 mb_skip_b(h);
1014 inter_pred(h);
1015 filter_mb(h,B_SKIP);
1016 return;
1017 case B_DIRECT:
1018 mb_skip_b(h);
1019 break;
1020 case B_FWD_16X16:
1021 mv_pred(h, MV_FWD_X0, MV_FWD_C2, MV_PRED_MEDIAN, BLK_16X16, 1);
1022 break;
1023 case B_SYM_16X16:
1024 mv_pred(h, MV_FWD_X0, MV_FWD_C2, MV_PRED_MEDIAN, BLK_16X16, 1);
1025 mv_pred_sym(h, &h->mv[MV_FWD_X0], BLK_16X16);
1026 break;
1027 case B_BWD_16X16:
1028 mv_pred(h, MV_BWD_X0, MV_BWD_C2, MV_PRED_MEDIAN, BLK_16X16, 0);
1029 break;
1030 case B_8X8:
1031 for(block=0;block<4;block++)
1032 sub_type[block] = get_bits(&h->s.gb,2);
1033 for(block=0;block<4;block++) {
1034 switch(sub_type[block]) {
1035 case B_SUB_DIRECT:
1036 if(!(*h->col_type)) {
1037 /* intra MB at co-location, do in-plane prediction */
1038 mv_pred(h, mv_scan[block], mv_scan[block]-3,
1039 MV_PRED_BSKIP, BLK_8X8, 1);
1040 mv_pred(h, mv_scan[block]+MV_BWD_OFFS,
1041 mv_scan[block]-3+MV_BWD_OFFS,
1042 MV_PRED_BSKIP, BLK_8X8, 0);
1043 } else
1044 mv_pred_direct(h,&h->mv[mv_scan[block]],
1045 &h->mv[mv_scan[block]+MV_BWD_OFFS],
1046 &h->col_mv[(h->mby*h->mb_width + h->mbx)*4 + block]);
1047 break;
1048 case B_SUB_FWD:
1049 mv_pred(h, mv_scan[block], mv_scan[block]-3,
1050 MV_PRED_MEDIAN, BLK_8X8, 1);
1051 break;
1052 case B_SUB_SYM:
1053 mv_pred(h, mv_scan[block], mv_scan[block]-3,
1054 MV_PRED_MEDIAN, BLK_8X8, 1);
1055 mv_pred_sym(h, &h->mv[mv_scan[block]], BLK_8X8);
1056 break;
1057 }
1058 }
1059 for(block=0;block<4;block++) {
1060 if(sub_type[block] == B_SUB_BWD)
1061 mv_pred(h, mv_scan[block]+MV_BWD_OFFS,
1062 mv_scan[block]+MV_BWD_OFFS-3,
1063 MV_PRED_MEDIAN, BLK_8X8, 0);
1064 }
1065 break;
1066 default:
1067 assert((mb_type > B_SYM_16X16) && (mb_type < B_8X8));
1068 flags = b_partition_flags[(mb_type-1)>>1];
1069 if(mb_type & 1) { /* 16x8 macroblock types */
1070 if(flags & FWD0)
1071 mv_pred(h, MV_FWD_X0, MV_FWD_C2, MV_PRED_TOP, BLK_16X8, 1);
1072 if(flags & SYM0) {
1073 mv_pred(h, MV_FWD_X0, MV_FWD_C2, MV_PRED_TOP, BLK_16X8, 1);
1074 mv_pred_sym(h, &h->mv[MV_FWD_X0], BLK_16X8);
1075 }
1076 if(flags & FWD1)
1077 mv_pred(h, MV_FWD_X2, MV_FWD_A1, MV_PRED_LEFT, BLK_16X8, 1);
1078 if(flags & SYM1) {
1079 mv_pred(h, MV_FWD_X2, MV_FWD_A1, MV_PRED_LEFT, BLK_16X8, 1);
1080 mv_pred_sym(h, &h->mv[9], BLK_16X8);
1081 }
1082 if(flags & BWD0)
1083 mv_pred(h, MV_BWD_X0, MV_BWD_C2, MV_PRED_TOP, BLK_16X8, 0);
1084 if(flags & BWD1)
1085 mv_pred(h, MV_BWD_X2, MV_BWD_A1, MV_PRED_LEFT, BLK_16X8, 0);
1086 } else { /* 8x16 macroblock types */
1087 if(flags & FWD0)
1088 mv_pred(h, MV_FWD_X0, MV_FWD_B3, MV_PRED_LEFT, BLK_8X16, 1);
1089 if(flags & SYM0) {
1090 mv_pred(h, MV_FWD_X0, MV_FWD_B3, MV_PRED_LEFT, BLK_8X16, 1);
1091 mv_pred_sym(h, &h->mv[MV_FWD_X0], BLK_8X16);
1092 }
1093 if(flags & FWD1)
1094 mv_pred(h, MV_FWD_X1, MV_FWD_C2, MV_PRED_TOPRIGHT,BLK_8X16, 1);
1095 if(flags & SYM1) {
1096 mv_pred(h, MV_FWD_X1, MV_FWD_C2, MV_PRED_TOPRIGHT,BLK_8X16, 1);
1097 mv_pred_sym(h, &h->mv[6], BLK_8X16);
1098 }
1099 if(flags & BWD0)
1100 mv_pred(h, MV_BWD_X0, MV_BWD_B3, MV_PRED_LEFT, BLK_8X16, 0);
1101 if(flags & BWD1)
1102 mv_pred(h, MV_BWD_X1, MV_BWD_C2, MV_PRED_TOPRIGHT,BLK_8X16, 0);
1103 }
1104 }
1105 inter_pred(h);
1106 decode_residual_inter(h);
1107 filter_mb(h,mb_type);
1108 }
1109
1110 /*****************************************************************************
1111 *
1112 * slice level
1113 *
1114 ****************************************************************************/
1115
1116 static inline int decode_slice_header(AVSContext *h, GetBitContext *gb) {
1117 if(h->stc > 0xAF)
1118 av_log(h->s.avctx, AV_LOG_ERROR, "unexpected start code 0x%02x\n", h->stc);
1119 h->mby = h->stc;
1120 if((h->mby == 0) && (!h->qp_fixed)){
1121 h->qp_fixed = get_bits1(gb);
1122 h->qp = get_bits(gb,6);
1123 }
1124 /* inter frame or second slice can have weighting params */
1125 if((h->pic_type != FF_I_TYPE) || (!h->pic_structure && h->mby >= h->mb_width/2))
1126 if(get_bits1(gb)) { //slice_weighting_flag
1127 av_log(h->s.avctx, AV_LOG_ERROR,
1128 "weighted prediction not yet supported\n");
1129 }
1130 return 0;
1131 }
1132
1133 static inline void check_for_slice(AVSContext *h) {
1134 GetBitContext *gb = &h->s.gb;
1135 int align;
1136 align = (-get_bits_count(gb)) & 7;
1137 if((show_bits_long(gb,24+align) & 0xFFFFFF) == 0x000001) {
1138 get_bits_long(gb,24+align);
1139 h->stc = get_bits(gb,8);
1140 decode_slice_header(h,gb);
1141 }
1142 }
1143
1144 /*****************************************************************************
1145 *
1146 * frame level
1147 *
1148 ****************************************************************************/
1149
1150 static void init_pic(AVSContext *h) {
1151 int i;
1152
1153 /* clear some predictors */
1154 for(i=0;i<=20;i+=4)
1155 h->mv[i] = un_mv;
1156 h->mv[MV_BWD_X0] = dir_mv;
1157 set_mvs(&h->mv[MV_BWD_X0], BLK_16X16);
1158 h->mv[MV_FWD_X0] = dir_mv;
1159 set_mvs(&h->mv[MV_FWD_X0], BLK_16X16);
1160 h->pred_mode_Y[3] = h->pred_mode_Y[6] = NOT_AVAIL;
1161 h->cy = h->picture.data[0];
1162 h->cu = h->picture.data[1];
1163 h->cv = h->picture.data[2];
1164 h->l_stride = h->picture.linesize[0];
1165 h->c_stride = h->picture.linesize[1];
1166 h->luma_scan[2] = 8*h->l_stride;
1167 h->luma_scan[3] = 8*h->l_stride+8;
1168 h->mbx = h->mby = 0;
1169 h->flags = 0;
1170 }
1171
1172 static int decode_pic(AVSContext *h) {
1173 MpegEncContext *s = &h->s;
1174 int i,skip_count;
1175 enum mb_t mb_type;
1176
1177 if (!s->context_initialized) {
1178 if (MPV_common_init(s) < 0)
1179 return -1;
1180 }
1181 get_bits(&s->gb,16);//bbv_dwlay
1182 if(h->stc == PIC_PB_START_CODE) {
1183 h->pic_type = get_bits(&s->gb,2) + FF_I_TYPE;
1184 /* make sure we have the reference frames we need */
1185 if(!h->DPB[0].data[0] ||
1186 (!h->DPB[1].data[0] && h->pic_type == FF_B_TYPE))
1187 return -1;
1188 } else {
1189 h->pic_type = FF_I_TYPE;
1190 if(get_bits1(&s->gb))
1191 get_bits(&s->gb,16);//time_code
1192 }
1193 /* release last B frame */
1194 if(h->picture.data[0])
1195 s->avctx->release_buffer(s->avctx, (AVFrame *)&h->picture);
1196
1197 s->avctx->get_buffer(s->avctx, (AVFrame *)&h->picture);
1198 init_pic(h);
1199 h->picture.poc = get_bits(&s->gb,8)*2;
1200
1201 /* get temporal distances and MV scaling factors */
1202 if(h->pic_type != FF_B_TYPE) {
1203 h->dist[0] = (h->picture.poc - h->DPB[0].poc + 512) % 512;
1204 } else {
1205 h->dist[0] = (h->DPB[0].poc - h->picture.poc + 512) % 512;
1206 }
1207 h->dist[1] = (h->picture.poc - h->DPB[1].poc + 512) % 512;
1208 h->scale_den[0] = h->dist[0] ? 512/h->dist[0] : 0;
1209 h->scale_den[1] = h->dist[1] ? 512/h->dist[1] : 0;
1210 if(h->pic_type == FF_B_TYPE) {
1211 h->sym_factor = h->dist[0]*h->scale_den[1];
1212 } else {
1213 h->direct_den[0] = h->dist[0] ? 16384/h->dist[0] : 0;
1214 h->direct_den[1] = h->dist[1] ? 16384/h->dist[1] : 0;
1215 }
1216
1217 if(s->low_delay)
1218 get_ue_golomb(&s->gb); //bbv_check_times
1219 h->progressive = get_bits1(&s->gb);
1220 if(h->progressive)
1221 h->pic_structure = 1;
1222 else if(!(h->pic_structure = get_bits1(&s->gb) && (h->stc == PIC_PB_START_CODE)) )
1223 get_bits1(&s->gb); //advanced_pred_mode_disable
1224 skip_bits1(&s->gb); //top_field_first
1225 skip_bits1(&s->gb); //repeat_first_field
1226 h->qp_fixed = get_bits1(&s->gb);
1227 h->qp = get_bits(&s->gb,6);
1228 if(h->pic_type == FF_I_TYPE) {
1229 if(!h->progressive && !h->pic_structure)
1230 skip_bits1(&s->gb);//what is this?
1231 skip_bits(&s->gb,4); //reserved bits
1232 } else {
1233 if(!(h->pic_type == FF_B_TYPE && h->pic_structure == 1))
1234 h->ref_flag = get_bits1(&s->gb);
1235 skip_bits(&s->gb,4); //reserved bits
1236 h->skip_mode_flag = get_bits1(&s->gb);
1237 }
1238 h->loop_filter_disable = get_bits1(&s->gb);
1239 if(!h->loop_filter_disable && get_bits1(&s->gb)) {
1240 h->alpha_offset = get_se_golomb(&s->gb);
1241 h->beta_offset = get_se_golomb(&s->gb);
1242 } else {
1243 h->alpha_offset = h->beta_offset = 0;
1244 }
1245 check_for_slice(h);
1246 if(h->pic_type == FF_I_TYPE) {
1247 do {
1248 init_mb(h);
1249 decode_mb_i(h,1);
1250 } while(next_mb(h));
1251 } else if(h->pic_type == FF_P_TYPE) {
1252 do {
1253 if(h->skip_mode_flag) {
1254 skip_count = get_ue_golomb(&s->gb);
1255 for(i=0;i<skip_count;i++) {
1256 init_mb(h);
1257 mb_skip_p(h);
1258 if(!next_mb(h))
1259 goto done;
1260 }
1261 mb_type = get_ue_golomb(&s->gb) + P_16X16;
1262 } else {
1263 mb_type = get_ue_golomb(&s->gb) + P_SKIP;
1264 }
1265 init_mb(h);
1266 if(mb_type > P_8X8) {
1267 h->cbp = cbp_tab[mb_type - P_8X8 - 1][0];
1268 decode_mb_i(h,0);
1269 } else {
1270 decode_mb_p(h,mb_type);
1271 }
1272 } while(next_mb(h));
1273 } else { //FF_B_TYPE
1274 do {
1275 if(h->skip_mode_flag) {
1276 skip_count = get_ue_golomb(&s->gb);
1277 for(i=0;i<skip_count;i++) {
1278 init_mb(h);
1279 mb_skip_b(h);
1280 inter_pred(h);
1281 filter_mb(h,B_SKIP);
1282 if(!next_mb(h))
1283 goto done;
1284 }
1285 mb_type = get_ue_golomb(&s->gb) + B_DIRECT;
1286 } else {
1287 mb_type = get_ue_golomb(&s->gb) + B_SKIP;
1288 }
1289 init_mb(h);
1290 if(mb_type > B_8X8) {
1291 h->cbp = cbp_tab[mb_type - B_8X8 - 1][0];
1292 decode_mb_i(h,0);
1293 } else {
1294 decode_mb_b(h,mb_type);
1295 }
1296 } while(next_mb(h));
1297 }
1298 done:
1299 if(h->pic_type != FF_B_TYPE) {
1300 if(h->DPB[1].data[0])
1301 s->avctx->release_buffer(s->avctx, (AVFrame *)&h->DPB[1]);
1302 memcpy(&h->DPB[1], &h->DPB[0], sizeof(Picture));
1303 memcpy(&h->DPB[0], &h->picture, sizeof(Picture));
1304 memset(&h->picture,0,sizeof(Picture));
1305 }
1306 return 0;
1307 }
1308
1309 /*****************************************************************************
1310 *
1311 * headers and interface
1312 *
1313 ****************************************************************************/
1314
1315 static void init_top_lines(AVSContext *h) {
1316 /* alloc top line of predictors */
1317 h->top_qp = av_malloc( h->mb_width);
1318 h->top_mv[0] = av_malloc((h->mb_width*2+1)*sizeof(vector_t));
1319 h->top_mv[1] = av_malloc((h->mb_width*2+1)*sizeof(vector_t));
1320 h->top_pred_Y = av_malloc( h->mb_width*2*sizeof(*h->top_pred_Y));
1321 h->top_border_y = av_malloc((h->mb_width+1)*16);
1322 h->top_border_u = av_malloc((h->mb_width)*10);
1323 h->top_border_v = av_malloc((h->mb_width)*10);
1324
1325 /* alloc space for co-located MVs and types */
1326 h->col_mv = av_malloc( h->mb_width*h->mb_height*4*sizeof(vector_t));
1327 h->col_type_base = av_malloc(h->mb_width*h->mb_height);
1328 }
1329
1330 static int decode_seq_header(AVSContext *h) {
1331 MpegEncContext *s = &h->s;
1332 extern const AVRational ff_frame_rate_tab[];
1333 int frame_rate_code;
1334
1335 h->profile = get_bits(&s->gb,8);
1336 h->level = get_bits(&s->gb,8);
1337 skip_bits1(&s->gb); //progressive sequence
1338 s->width = get_bits(&s->gb,14);
1339 s->height = get_bits(&s->gb,14);
1340 skip_bits(&s->gb,2); //chroma format
1341 skip_bits(&s->gb,3); //sample_precision
1342 h->aspect_ratio = get_bits(&s->gb,4);
1343 frame_rate_code = get_bits(&s->gb,4);
1344 skip_bits(&s->gb,18);//bit_rate_lower
1345 skip_bits1(&s->gb); //marker_bit
1346 skip_bits(&s->gb,12);//bit_rate_upper
1347 s->low_delay = get_bits1(&s->gb);
1348 h->mb_width = (s->width + 15) >> 4;
1349 h->mb_height = (s->height + 15) >> 4;
1350 h->s.avctx->time_base.den = ff_frame_rate_tab[frame_rate_code].num;
1351 h->s.avctx->time_base.num = ff_frame_rate_tab[frame_rate_code].den;
1352 h->s.avctx->width = s->width;
1353 h->s.avctx->height = s->height;
1354 if(!h->top_qp)
1355 init_top_lines(h);
1356 return 0;
1357 }
1358
1359 /**
1360 * finds the end of the current frame in the bitstream.
1361 * @return the position of the first byte of the next frame, or -1
1362 */
1363 int ff_cavs_find_frame_end(ParseContext *pc, const uint8_t *buf, int buf_size) {
1364 int pic_found, i;
1365 uint32_t state;
1366
1367 pic_found= pc->frame_start_found;
1368 state= pc->state;
1369
1370 i=0;
1371 if(!pic_found){
1372 for(i=0; i<buf_size; i++){
1373 state= (state<<8) | buf[i];
1374 if(state == PIC_I_START_CODE || state == PIC_PB_START_CODE){
1375 i++;
1376 pic_found=1;
1377 break;
1378 }
1379 }
1380 }
1381
1382 if(pic_found){
1383 /* EOF considered as end of frame */
1384 if (buf_size == 0)
1385 return 0;
1386 for(; i<buf_size; i++){
1387 state= (state<<8) | buf[i];
1388 if((state&0xFFFFFF00) == 0x100){
1389 if(state < SLICE_MIN_START_CODE || state > SLICE_MAX_START_CODE){
1390 pc->frame_start_found=0;
1391 pc->state=-1;
1392 return i-3;
1393 }
1394 }
1395 }
1396 }
1397 pc->frame_start_found= pic_found;
1398 pc->state= state;
1399 return END_NOT_FOUND;
1400 }
1401
1402 void ff_cavs_flush(AVCodecContext * avctx) {
1403 AVSContext *h = avctx->priv_data;
1404 h->got_keyframe = 0;
1405 }
1406
1407 static int cavs_decode_frame(AVCodecContext * avctx,void *data, int *data_size,
1408 uint8_t * buf, int buf_size) {
1409 AVSContext *h = avctx->priv_data;
1410 MpegEncContext *s = &h->s;
1411 int input_size;
1412 const uint8_t *buf_end;
1413 const uint8_t *buf_ptr;
1414 AVFrame *picture = data;
1415 uint32_t stc;
1416
1417 s->avctx = avctx;
1418
1419 if (buf_size == 0) {
1420 if(!s->low_delay && h->DPB[0].data[0]) {
1421 *data_size = sizeof(AVPicture);
1422 *picture = *(AVFrame *) &h->DPB[0];
1423 }
1424 return 0;
1425 }
1426
1427 buf_ptr = buf;
1428 buf_end = buf + buf_size;
1429 for(;;) {
1430 buf_ptr = ff_find_start_code(buf_ptr,buf_end, &stc);
1431 if(stc & 0xFFFFFE00)
1432 return FFMAX(0, buf_ptr - buf - s->parse_context.last_index);
1433 input_size = (buf_end - buf_ptr)*8;
1434 switch(stc) {
1435 case SEQ_START_CODE:
1436 init_get_bits(&s->gb, buf_ptr, input_size);
1437 decode_seq_header(h);
1438 break;
1439 case PIC_I_START_CODE:
1440 if(!h->got_keyframe) {
1441 if(h->DPB[0].data[0])
1442 avctx->release_buffer(avctx, (AVFrame *)&h->DPB[0]);
1443 if(h->DPB[1].data[0])
1444 avctx->release_buffer(avctx, (AVFrame *)&h->DPB[1]);
1445 h->got_keyframe = 1;
1446 }
1447 case PIC_PB_START_CODE:
1448 *data_size = 0;
1449 if(!h->got_keyframe)
1450 break;
1451 init_get_bits(&s->gb, buf_ptr, input_size);
1452 h->stc = stc;
1453 if(decode_pic(h))
1454 break;
1455 *data_size = sizeof(AVPicture);
1456 if(h->pic_type != FF_B_TYPE) {
1457 if(h->DPB[1].data[0]) {
1458 *picture = *(AVFrame *) &h->DPB[1];
1459 } else {
1460 *data_size = 0;
1461 }
1462 } else
1463 *picture = *(AVFrame *) &h->picture;
1464 break;
1465 case EXT_START_CODE:
1466 //mpeg_decode_extension(avctx,buf_ptr, input_size);
1467 break;
1468 case USER_START_CODE:
1469 //mpeg_decode_user_data(avctx,buf_ptr, input_size);
1470 break;
1471 default:
1472 if (stc >= SLICE_MIN_START_CODE &&
1473 stc <= SLICE_MAX_START_CODE) {
1474 init_get_bits(&s->gb, buf_ptr, input_size);
1475 decode_slice_header(h, &s->gb);
1476 }
1477 break;
1478 }
1479 }
1480 }
1481
1482 static int cavs_decode_init(AVCodecContext * avctx) {
1483 AVSContext *h = avctx->priv_data;
1484 MpegEncContext * const s = &h->s;
1485
1486 MPV_decode_defaults(s);
1487 s->avctx = avctx;
1488
1489 avctx->pix_fmt= PIX_FMT_YUV420P;
1490
1491 h->luma_scan[0] = 0;
1492 h->luma_scan[1] = 8;
1493 h->intra_pred_l[ INTRA_L_VERT] = intra_pred_vert;
1494 h->intra_pred_l[ INTRA_L_HORIZ] = intra_pred_horiz;
1495 h->intra_pred_l[ INTRA_L_LP] = intra_pred_lp;
1496 h->intra_pred_l[ INTRA_L_DOWN_LEFT] = intra_pred_down_left;
1497 h->intra_pred_l[INTRA_L_DOWN_RIGHT] = intra_pred_down_right;
1498 h->intra_pred_l[ INTRA_L_LP_LEFT] = intra_pred_lp_left;
1499 h->intra_pred_l[ INTRA_L_LP_TOP] = intra_pred_lp_top;
1500 h->intra_pred_l[ INTRA_L_DC_128] = intra_pred_dc_128;
1501 h->intra_pred_c[ INTRA_C_LP] = intra_pred_lp;
1502 h->intra_pred_c[ INTRA_C_HORIZ] = intra_pred_horiz;
1503 h->intra_pred_c[ INTRA_C_VERT] = intra_pred_vert;
1504 h->intra_pred_c[ INTRA_C_PLANE] = intra_pred_plane;
1505 h->intra_pred_c[ INTRA_C_LP_LEFT] = intra_pred_lp_left;
1506 h->intra_pred_c[ INTRA_C_LP_TOP] = intra_pred_lp_top;
1507 h->intra_pred_c[ INTRA_C_DC_128] = intra_pred_dc_128;
1508 h->mv[ 7] = un_mv;
1509 h->mv[19] = un_mv;
1510 return 0;
1511 }
1512
1513 static int cavs_decode_end(AVCodecContext * avctx) {
1514 AVSContext *h = avctx->priv_data;
1515
1516 av_free(h->top_qp);
1517 av_free(h->top_mv[0]);
1518 av_free(h->top_mv[1]);
1519 av_free(h->top_pred_Y);
1520 av_free(h->top_border_y);
1521 av_free(h->top_border_u);
1522 av_free(h->top_border_v);
1523 av_free(h->col_mv);
1524 av_free(h->col_type_base);
1525 return 0;
1526 }
1527
1528 AVCodec cavs_decoder = {
1529 "cavs",
1530 CODEC_TYPE_VIDEO,
1531 CODEC_ID_CAVS,
1532 sizeof(AVSContext),
1533 cavs_decode_init,
1534 NULL,
1535 cavs_decode_end,
1536 cavs_decode_frame,
1537 CODEC_CAP_DR1 | CODEC_CAP_DELAY,
1538 .flush= ff_cavs_flush,
1539 };