simplify
[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 /**
21 * @file cavs.c
22 * Chinese AVS video (AVS1-P2, JiZhun profile) decoder
23 * @author Stefan Gehrer <stefan.gehrer@gmx.de>
24 */
25
26 #include "avcodec.h"
27 #include "bitstream.h"
28 #include "golomb.h"
29 #include "mpegvideo.h"
30 #include "cavsdata.h"
31
32 typedef struct {
33 MpegEncContext s;
34 Picture picture; ///< currently decoded frame
35 Picture DPB[2]; ///< reference frames
36 int dist[2]; ///< temporal distances from current frame to ref frames
37 int profile, level;
38 int aspect_ratio;
39 int mb_width, mb_height;
40 int pic_type;
41 int progressive;
42 int pic_structure;
43 int skip_mode_flag; ///< select between skip_count or one skip_flag per MB
44 int loop_filter_disable;
45 int alpha_offset, beta_offset;
46 int ref_flag;
47 int mbx, mby; ///< macroblock coordinates
48 int flags; ///< availability flags of neighbouring macroblocks
49 int stc; ///< last start code
50 uint8_t *cy, *cu, *cv; ///< current MB sample pointers
51 int left_qp;
52 uint8_t *top_qp;
53
54 /** mv motion vector cache
55 0: D3 B2 B3 C2
56 4: A1 X0 X1 -
57 8: A3 X2 X3 -
58
59 X are the vectors in the current macroblock (5,6,9,10)
60 A is the macroblock to the left (4,8)
61 B is the macroblock to the top (1,2)
62 C is the macroblock to the top-right (3)
63 D is the macroblock to the top-left (0)
64
65 the same is repeated for backward motion vectors */
66 vector_t mv[2*4*3];
67 vector_t *top_mv[2];
68 vector_t *col_mv;
69
70 /** luma pred mode cache
71 0: -- B2 B3
72 3: A1 X0 X1
73 6: A3 X2 X3 */
74 int pred_mode_Y[3*3];
75 int *top_pred_Y;
76 int l_stride, c_stride;
77 int luma_scan[4];
78 int qp;
79 int qp_fixed;
80 int cbp;
81
82 /** intra prediction is done with un-deblocked samples
83 they are saved here before deblocking the MB */
84 uint8_t *top_border_y, *top_border_u, *top_border_v;
85 uint8_t left_border_y[16], left_border_u[10], left_border_v[10];
86 uint8_t topleft_border_y, topleft_border_u, topleft_border_v;
87
88 void (*intra_pred_l[8])(uint8_t *d,uint8_t *top,uint8_t *left,int stride);
89 void (*intra_pred_c[7])(uint8_t *d,uint8_t *top,uint8_t *left,int stride);
90 uint8_t *col_type_base;
91 uint8_t *col_type;
92
93 /* scaling factors for MV prediction */
94 int sym_factor; ///< for scaling in symmetrical B block
95 int direct_den[2]; ///< for scaling in direct B block
96 int scale_den[2]; ///< for scaling neighbouring MVs
97
98 int got_keyframe;
99 } AVSContext;
100
101 /*****************************************************************************
102 *
103 * in-loop deblocking filter
104 *
105 ****************************************************************************/
106
107 static inline int get_bs(vector_t *mvP, vector_t *mvQ, int b) {
108 if((mvP->ref == REF_INTRA) || (mvQ->ref == REF_INTRA))
109 return 2;
110 if( (abs(mvP->x - mvQ->x) >= 4) || (abs(mvP->y - mvQ->y) >= 4) )
111 return 1;
112 if(b){
113 mvP += MV_BWD_OFFS;
114 mvQ += MV_BWD_OFFS;
115 if( (abs(mvP->x - mvQ->x) >= 4) || (abs(mvP->y - mvQ->y) >= 4) )
116 return 1;
117 }else{
118 if(mvP->ref != mvQ->ref)
119 return 1;
120 }
121 return 0;
122 }
123
124 #define SET_PARAMS \
125 alpha = alpha_tab[clip(qp_avg + h->alpha_offset,0,63)]; \
126 beta = beta_tab[clip(qp_avg + h->beta_offset, 0,63)]; \
127 tc = tc_tab[clip(qp_avg + h->alpha_offset,0,63)];
128
129 /**
130 * in-loop deblocking filter for a single macroblock
131 *
132 * boundary strength (bs) mapping:
133 *
134 * --4---5--
135 * 0 2 |
136 * | 6 | 7 |
137 * 1 3 |
138 * ---------
139 *
140 */
141 static void filter_mb(AVSContext *h, enum mb_t mb_type) {
142 DECLARE_ALIGNED_8(uint8_t, bs[8]);
143 int qp_avg, alpha, beta, tc;
144 int i;
145
146 /* save un-deblocked lines */
147 h->topleft_border_y = h->top_border_y[h->mbx*16+15];
148 h->topleft_border_u = h->top_border_u[h->mbx*10+8];
149 h->topleft_border_v = h->top_border_v[h->mbx*10+8];
150 memcpy(&h->top_border_y[h->mbx*16], h->cy + 15* h->l_stride,16);
151 memcpy(&h->top_border_u[h->mbx*10+1], h->cu + 7* h->c_stride,8);
152 memcpy(&h->top_border_v[h->mbx*10+1], h->cv + 7* h->c_stride,8);
153 for(i=0;i<8;i++) {
154 h->left_border_y[i*2+0] = *(h->cy + 15 + (i*2+0)*h->l_stride);
155 h->left_border_y[i*2+1] = *(h->cy + 15 + (i*2+1)*h->l_stride);
156 h->left_border_u[i+1] = *(h->cu + 7 + i*h->c_stride);
157 h->left_border_v[i+1] = *(h->cv + 7 + i*h->c_stride);
158 }
159 if(!h->loop_filter_disable) {
160 /* determine bs */
161 if(mb_type == I_8X8)
162 *((uint64_t *)bs) = 0x0202020202020202ULL;
163 else{
164 *((uint64_t *)bs) = 0;
165 if(partition_flags[mb_type] & SPLITV){
166 bs[2] = get_bs(&h->mv[MV_FWD_X0], &h->mv[MV_FWD_X1], mb_type > P_8X8);
167 bs[3] = get_bs(&h->mv[MV_FWD_X2], &h->mv[MV_FWD_X3], mb_type > P_8X8);
168 }
169 if(partition_flags[mb_type] & SPLITH){
170 bs[6] = get_bs(&h->mv[MV_FWD_X0], &h->mv[MV_FWD_X2], mb_type > P_8X8);
171 bs[7] = get_bs(&h->mv[MV_FWD_X1], &h->mv[MV_FWD_X3], mb_type > P_8X8);
172 }
173 bs[0] = get_bs(&h->mv[MV_FWD_A1], &h->mv[MV_FWD_X0], mb_type > P_8X8);
174 bs[1] = get_bs(&h->mv[MV_FWD_A3], &h->mv[MV_FWD_X2], mb_type > P_8X8);
175 bs[4] = get_bs(&h->mv[MV_FWD_B2], &h->mv[MV_FWD_X0], mb_type > P_8X8);
176 bs[5] = get_bs(&h->mv[MV_FWD_B3], &h->mv[MV_FWD_X1], mb_type > P_8X8);
177 }
178 if( *((uint64_t *)bs) ) {
179 if(h->flags & A_AVAIL) {
180 qp_avg = (h->qp + h->left_qp + 1) >> 1;
181 SET_PARAMS;
182 h->s.dsp.cavs_filter_lv(h->cy,h->l_stride,alpha,beta,tc,bs[0],bs[1]);
183 h->s.dsp.cavs_filter_cv(h->cu,h->c_stride,alpha,beta,tc,bs[0],bs[1]);
184 h->s.dsp.cavs_filter_cv(h->cv,h->c_stride,alpha,beta,tc,bs[0],bs[1]);
185 }
186 qp_avg = h->qp;
187 SET_PARAMS;
188 h->s.dsp.cavs_filter_lv(h->cy + 8,h->l_stride,alpha,beta,tc,bs[2],bs[3]);
189 h->s.dsp.cavs_filter_lh(h->cy + 8*h->l_stride,h->l_stride,alpha,beta,tc,
190 bs[6],bs[7]);
191
192 if(h->flags & B_AVAIL) {
193 qp_avg = (h->qp + h->top_qp[h->mbx] + 1) >> 1;
194 SET_PARAMS;
195 h->s.dsp.cavs_filter_lh(h->cy,h->l_stride,alpha,beta,tc,bs[4],bs[5]);
196 h->s.dsp.cavs_filter_ch(h->cu,h->c_stride,alpha,beta,tc,bs[4],bs[5]);
197 h->s.dsp.cavs_filter_ch(h->cv,h->c_stride,alpha,beta,tc,bs[4],bs[5]);
198 }
199 }
200 }
201 h->left_qp = h->qp;
202 h->top_qp[h->mbx] = h->qp;
203 }
204
205 #undef SET_PARAMS
206
207 /*****************************************************************************
208 *
209 * spatial intra prediction
210 *
211 ****************************************************************************/
212
213 static inline void load_intra_pred_luma(AVSContext *h, uint8_t *top,
214 uint8_t *left, int block) {
215 int i;
216
217 switch(block) {
218 case 0:
219 memcpy(&left[1],h->left_border_y,16);
220 left[0] = left[1];
221 left[17] = left[16];
222 memcpy(&top[1],&h->top_border_y[h->mbx*16],16);
223 top[17] = top[16];
224 top[0] = top[1];
225 if((h->flags & A_AVAIL) && (h->flags & B_AVAIL))
226 left[0] = top[0] = h->topleft_border_y;
227 break;
228 case 1:
229 for(i=0;i<8;i++)
230 left[i+1] = *(h->cy + 7 + i*h->l_stride);
231 memset(&left[9],left[8],9);
232 left[0] = left[1];
233 memcpy(&top[1],&h->top_border_y[h->mbx*16+8],8);
234 if(h->flags & C_AVAIL)
235 memcpy(&top[9],&h->top_border_y[(h->mbx + 1)*16],8);
236 else
237 memset(&top[9],top[8],9);
238 top[17] = top[16];
239 top[0] = top[1];
240 if(h->flags & B_AVAIL)
241 left[0] = top[0] = h->top_border_y[h->mbx*16+7];
242 break;
243 case 2:
244 memcpy(&left[1],&h->left_border_y[8],8);
245 memset(&left[9],left[8],9);
246 memcpy(&top[1],h->cy + 7*h->l_stride,16);
247 top[17] = top[16];
248 left[0] = h->left_border_y[7];
249 top[0] = top[1];
250 if(h->flags & A_AVAIL)
251 top[0] = left[0];
252 break;
253 case 3:
254 for(i=0;i<9;i++)
255 left[i] = *(h->cy + 7 + (i+7)*h->l_stride);
256 memset(&left[9],left[8],9);
257 memcpy(&top[0],h->cy + 7 + 7*h->l_stride,9);
258 memset(&top[9],top[8],9);
259 break;
260 }
261 }
262
263 static void intra_pred_vert(uint8_t *d,uint8_t *top,uint8_t *left,int stride) {
264 int y;
265 uint64_t a = unaligned64(&top[1]);
266 for(y=0;y<8;y++) {
267 *((uint64_t *)(d+y*stride)) = a;
268 }
269 }
270
271 static void intra_pred_horiz(uint8_t *d,uint8_t *top,uint8_t *left,int stride) {
272 int y;
273 uint64_t a;
274 for(y=0;y<8;y++) {
275 a = left[y+1] * 0x0101010101010101ULL;
276 *((uint64_t *)(d+y*stride)) = a;
277 }
278 }
279
280 static void intra_pred_dc_128(uint8_t *d,uint8_t *top,uint8_t *left,int stride) {
281 int y;
282 uint64_t a = 0x8080808080808080ULL;
283 for(y=0;y<8;y++)
284 *((uint64_t *)(d+y*stride)) = a;
285 }
286
287 static void intra_pred_plane(uint8_t *d,uint8_t *top,uint8_t *left,int stride) {
288 int x,y,ia;
289 int ih = 0;
290 int iv = 0;
291 uint8_t *cm = cropTbl + 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 int x,y;
310 for(y=0; y<8; y++)
311 for(x=0; x<8; x++)
312 d[y*stride+x] = (LOWPASS(top,x+1) + LOWPASS(left,y+1)) >> 1;
313 }
314
315 static void intra_pred_down_left(uint8_t *d,uint8_t *top,uint8_t *left,int stride) {
316 int x,y;
317 for(y=0; y<8; y++)
318 for(x=0; x<8; x++)
319 d[y*stride+x] = (LOWPASS(top,x+y+2) + LOWPASS(left,x+y+2)) >> 1;
320 }
321
322 static void intra_pred_down_right(uint8_t *d,uint8_t *top,uint8_t *left,int stride) {
323 int x,y;
324 for(y=0; y<8; y++)
325 for(x=0; x<8; x++)
326 if(x==y)
327 d[y*stride+x] = (left[1]+2*top[0]+top[1]+2)>>2;
328 else if(x>y)
329 d[y*stride+x] = LOWPASS(top,x-y);
330 else
331 d[y*stride+x] = LOWPASS(left,y-x);
332 }
333
334 static void intra_pred_lp_left(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(left,y+1);
339 }
340
341 static void intra_pred_lp_top(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+1);
346 }
347
348 #undef LOWPASS
349
350 static inline void modify_pred(const int_fast8_t *mod_table, int *mode) {
351 *mode = mod_table[*mode];
352 if(*mode < 0) {
353 av_log(NULL, AV_LOG_ERROR, "Illegal intra prediction mode\n");
354 *mode = 0;
355 }
356 }
357
358 /*****************************************************************************
359 *
360 * motion compensation
361 *
362 ****************************************************************************/
363
364 static inline void mc_dir_part(AVSContext *h,Picture *pic,int square,
365 int chroma_height,int delta,int list,uint8_t *dest_y,
366 uint8_t *dest_cb,uint8_t *dest_cr,int src_x_offset,
367 int src_y_offset,qpel_mc_func *qpix_op,
368 h264_chroma_mc_func chroma_op,vector_t *mv){
369 MpegEncContext * const s = &h->s;
370 const int mx= mv->x + src_x_offset*8;
371 const int my= mv->y + src_y_offset*8;
372 const int luma_xy= (mx&3) + ((my&3)<<2);
373 uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*h->l_stride;
374 uint8_t * src_cb= pic->data[1] + (mx>>3) + (my>>3)*h->c_stride;
375 uint8_t * src_cr= pic->data[2] + (mx>>3) + (my>>3)*h->c_stride;
376 int extra_width= 0; //(s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16;
377 int extra_height= extra_width;
378 int emu=0;
379 const int full_mx= mx>>2;
380 const int full_my= my>>2;
381 const int pic_width = 16*h->mb_width;
382 const int pic_height = 16*h->mb_height;
383
384 if(!pic->data[0])
385 return;
386 if(mx&7) extra_width -= 3;
387 if(my&7) extra_height -= 3;
388
389 if( full_mx < 0-extra_width
390 || full_my < 0-extra_height
391 || full_mx + 16/*FIXME*/ > pic_width + extra_width
392 || full_my + 16/*FIXME*/ > pic_height + extra_height){
393 ff_emulated_edge_mc(s->edge_emu_buffer, src_y - 2 - 2*h->l_stride, h->l_stride,
394 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
395 src_y= s->edge_emu_buffer + 2 + 2*h->l_stride;
396 emu=1;
397 }
398
399 qpix_op[luma_xy](dest_y, src_y, h->l_stride); //FIXME try variable height perhaps?
400 if(!square){
401 qpix_op[luma_xy](dest_y + delta, src_y + delta, h->l_stride);
402 }
403
404 if(emu){
405 ff_emulated_edge_mc(s->edge_emu_buffer, src_cb, h->c_stride,
406 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
407 src_cb= s->edge_emu_buffer;
408 }
409 chroma_op(dest_cb, src_cb, h->c_stride, chroma_height, mx&7, my&7);
410
411 if(emu){
412 ff_emulated_edge_mc(s->edge_emu_buffer, src_cr, h->c_stride,
413 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
414 src_cr= s->edge_emu_buffer;
415 }
416 chroma_op(dest_cr, src_cr, h->c_stride, chroma_height, mx&7, my&7);
417 }
418
419 static inline void mc_part_std(AVSContext *h,int square,int chroma_height,int delta,
420 uint8_t *dest_y,uint8_t *dest_cb,uint8_t *dest_cr,
421 int x_offset, int y_offset,qpel_mc_func *qpix_put,
422 h264_chroma_mc_func chroma_put,qpel_mc_func *qpix_avg,
423 h264_chroma_mc_func chroma_avg, vector_t *mv){
424 qpel_mc_func *qpix_op= qpix_put;
425 h264_chroma_mc_func chroma_op= chroma_put;
426
427 dest_y += 2*x_offset + 2*y_offset*h->l_stride;
428 dest_cb += x_offset + y_offset*h->c_stride;
429 dest_cr += x_offset + y_offset*h->c_stride;
430 x_offset += 8*h->mbx;
431 y_offset += 8*h->mby;
432
433 if(mv->ref >= 0){
434 Picture *ref= &h->DPB[mv->ref];
435 mc_dir_part(h, ref, square, chroma_height, delta, 0,
436 dest_y, dest_cb, dest_cr, x_offset, y_offset,
437 qpix_op, chroma_op, mv);
438
439 qpix_op= qpix_avg;
440 chroma_op= chroma_avg;
441 }
442
443 if((mv+MV_BWD_OFFS)->ref >= 0){
444 Picture *ref= &h->DPB[0];
445 mc_dir_part(h, ref, square, chroma_height, delta, 1,
446 dest_y, dest_cb, dest_cr, x_offset, y_offset,
447 qpix_op, chroma_op, mv+MV_BWD_OFFS);
448 }
449 }
450
451 static void inter_pred(AVSContext *h, enum mb_t mb_type) {
452 if(partition_flags[mb_type] == 0){ // 16x16
453 mc_part_std(h, 1, 8, 0, h->cy, h->cu, h->cv, 0, 0,
454 h->s.dsp.put_cavs_qpel_pixels_tab[0],
455 h->s.dsp.put_h264_chroma_pixels_tab[0],
456 h->s.dsp.avg_cavs_qpel_pixels_tab[0],
457 h->s.dsp.avg_h264_chroma_pixels_tab[0],&h->mv[MV_FWD_X0]);
458 }else{
459 mc_part_std(h, 1, 4, 0, h->cy, h->cu, h->cv, 0, 0,
460 h->s.dsp.put_cavs_qpel_pixels_tab[1],
461 h->s.dsp.put_h264_chroma_pixels_tab[1],
462 h->s.dsp.avg_cavs_qpel_pixels_tab[1],
463 h->s.dsp.avg_h264_chroma_pixels_tab[1],&h->mv[MV_FWD_X0]);
464 mc_part_std(h, 1, 4, 0, h->cy, h->cu, h->cv, 4, 0,
465 h->s.dsp.put_cavs_qpel_pixels_tab[1],
466 h->s.dsp.put_h264_chroma_pixels_tab[1],
467 h->s.dsp.avg_cavs_qpel_pixels_tab[1],
468 h->s.dsp.avg_h264_chroma_pixels_tab[1],&h->mv[MV_FWD_X1]);
469 mc_part_std(h, 1, 4, 0, h->cy, h->cu, h->cv, 0, 4,
470 h->s.dsp.put_cavs_qpel_pixels_tab[1],
471 h->s.dsp.put_h264_chroma_pixels_tab[1],
472 h->s.dsp.avg_cavs_qpel_pixels_tab[1],
473 h->s.dsp.avg_h264_chroma_pixels_tab[1],&h->mv[MV_FWD_X2]);
474 mc_part_std(h, 1, 4, 0, h->cy, h->cu, h->cv, 4, 4,
475 h->s.dsp.put_cavs_qpel_pixels_tab[1],
476 h->s.dsp.put_h264_chroma_pixels_tab[1],
477 h->s.dsp.avg_cavs_qpel_pixels_tab[1],
478 h->s.dsp.avg_h264_chroma_pixels_tab[1],&h->mv[MV_FWD_X3]);
479 }
480 /* set intra prediction modes to default values */
481 h->pred_mode_Y[3] = h->pred_mode_Y[6] = INTRA_L_LP;
482 h->top_pred_Y[h->mbx*2+0] = h->top_pred_Y[h->mbx*2+1] = INTRA_L_LP;
483 }
484
485 /*****************************************************************************
486 *
487 * motion vector prediction
488 *
489 ****************************************************************************/
490
491 static inline void set_mvs(vector_t *mv, enum block_t size) {
492 switch(size) {
493 case BLK_16X16:
494 mv[MV_STRIDE ] = mv[0];
495 mv[MV_STRIDE+1] = mv[0];
496 case BLK_16X8:
497 mv[1] = mv[0];
498 break;
499 case BLK_8X16:
500 mv[MV_STRIDE] = mv[0];
501 break;
502 }
503 }
504
505 static inline void store_mvs(AVSContext *h) {
506 h->col_mv[(h->mby*h->mb_width + h->mbx)*4 + 0] = h->mv[MV_FWD_X0];
507 h->col_mv[(h->mby*h->mb_width + h->mbx)*4 + 1] = h->mv[MV_FWD_X1];
508 h->col_mv[(h->mby*h->mb_width + h->mbx)*4 + 2] = h->mv[MV_FWD_X2];
509 h->col_mv[(h->mby*h->mb_width + h->mbx)*4 + 3] = h->mv[MV_FWD_X3];
510 }
511
512 static inline void scale_mv(AVSContext *h, int *d_x, int *d_y, vector_t *src, int distp) {
513 int den = h->scale_den[src->ref];
514
515 *d_x = (src->x*distp*den + 256 + (src->x>>31)) >> 9;
516 *d_y = (src->y*distp*den + 256 + (src->y>>31)) >> 9;
517 }
518
519 static inline void mv_pred_median(AVSContext *h, vector_t *mvP, vector_t *mvA, vector_t *mvB, vector_t *mvC) {
520 int ax, ay, bx, by, cx, cy;
521 int len_ab, len_bc, len_ca, len_mid;
522
523 /* scale candidates according to their temporal span */
524 scale_mv(h, &ax, &ay, mvA, mvP->dist);
525 scale_mv(h, &bx, &by, mvB, mvP->dist);
526 scale_mv(h, &cx, &cy, mvC, mvP->dist);
527 /* find the geometrical median of the three candidates */
528 len_ab = abs(ax - bx) + abs(ay - by);
529 len_bc = abs(bx - cx) + abs(by - cy);
530 len_ca = abs(cx - ax) + abs(cy - ay);
531 len_mid = mid_pred(len_ab, len_bc, len_ca);
532 if(len_mid == len_ab) {
533 mvP->x = cx;
534 mvP->y = cy;
535 } else if(len_mid == len_bc) {
536 mvP->x = ax;
537 mvP->y = ay;
538 } else {
539 mvP->x = bx;
540 mvP->y = by;
541 }
542 }
543
544 static inline void mv_pred_direct(AVSContext *h, vector_t *pmv_fw,
545 vector_t *col_mv) {
546 vector_t *pmv_bw = pmv_fw + MV_BWD_OFFS;
547 int den = h->direct_den[col_mv->ref];
548 int m = col_mv->x >> 31;
549
550 pmv_fw->dist = h->dist[1];
551 pmv_bw->dist = h->dist[0];
552 pmv_fw->ref = 1;
553 pmv_bw->ref = 0;
554 /* scale the co-located motion vector according to its temporal span */
555 pmv_fw->x = (((den+(den*col_mv->x*pmv_fw->dist^m)-m-1)>>14)^m)-m;
556 pmv_bw->x = m-(((den+(den*col_mv->x*pmv_bw->dist^m)-m-1)>>14)^m);
557 m = col_mv->y >> 31;
558 pmv_fw->y = (((den+(den*col_mv->y*pmv_fw->dist^m)-m-1)>>14)^m)-m;
559 pmv_bw->y = m-(((den+(den*col_mv->y*pmv_bw->dist^m)-m-1)>>14)^m);
560 }
561
562 static inline void mv_pred_sym(AVSContext *h, vector_t *src, enum block_t size) {
563 vector_t *dst = src + MV_BWD_OFFS;
564
565 /* backward mv is the scaled and negated forward mv */
566 dst->x = -((src->x * h->sym_factor + 256) >> 9);
567 dst->y = -((src->y * h->sym_factor + 256) >> 9);
568 dst->ref = 0;
569 dst->dist = h->dist[0];
570 set_mvs(dst, size);
571 }
572
573 static void mv_pred(AVSContext *h, enum mv_loc_t nP, enum mv_loc_t nC,
574 enum mv_pred_t mode, enum block_t size, int ref) {
575 vector_t *mvP = &h->mv[nP];
576 vector_t *mvA = &h->mv[nP-1];
577 vector_t *mvB = &h->mv[nP-4];
578 vector_t *mvC = &h->mv[nC];
579 int mvAref = mvA->ref;
580 int mvBref = mvB->ref;
581 int mvCref;
582
583 mvP->ref = ref;
584 mvP->dist = h->dist[mvP->ref];
585 if(mvC->ref == NOT_AVAIL)
586 mvC = &h->mv[nP-5]; // set to top-left (mvD)
587 mvCref = mvC->ref;
588 if(mode == MV_PRED_PSKIP) {
589 if((mvAref == NOT_AVAIL) || (mvBref == NOT_AVAIL) ||
590 ((mvA->x | mvA->y | mvA->ref) == 0) ||
591 ((mvB->x | mvB->y | mvB->ref) == 0) ) {
592 mvP->x = mvP->y = 0;
593 set_mvs(mvP,size);
594 return;
595 }
596 }
597 /* if there is only one suitable candidate, take it */
598 if((mvAref >= 0) && (mvBref < 0) && (mvCref < 0)) {
599 mvP->x = mvA->x;
600 mvP->y = mvA->y;
601 } else if((mvAref < 0) && (mvBref >= 0) && (mvCref < 0)) {
602 mvP->x = mvB->x;
603 mvP->y = mvB->y;
604 } else if((mvAref < 0) && (mvBref < 0) && (mvCref >= 0)) {
605 mvP->x = mvC->x;
606 mvP->y = mvC->y;
607 } else {
608 switch(mode) {
609 case MV_PRED_LEFT:
610 if(mvAref == mvP->ref) {
611 mvP->x = mvA->x;
612 mvP->y = mvA->y;
613 } else
614 mv_pred_median(h, mvP, mvA, mvB, mvC);
615 break;
616 case MV_PRED_TOP:
617 if(mvBref == mvP->ref) {
618 mvP->x = mvB->x;
619 mvP->y = mvB->y;
620 } else
621 mv_pred_median(h, mvP, mvA, mvB, mvC);
622 break;
623 case MV_PRED_TOPRIGHT:
624 if(mvCref == mvP->ref) {
625 mvP->x = mvC->x;
626 mvP->y = mvC->y;
627 } else
628 mv_pred_median(h, mvP, mvA, mvB, mvC);
629 break;
630 default:
631 mv_pred_median(h, mvP, mvA, mvB, mvC);
632 break;
633 }
634 }
635 if(mode < MV_PRED_PSKIP) {
636 mvP->x += get_se_golomb(&h->s.gb);
637 mvP->y += get_se_golomb(&h->s.gb);
638 }
639 set_mvs(mvP,size);
640 }
641
642 /*****************************************************************************
643 *
644 * residual data decoding
645 *
646 ****************************************************************************/
647
648 /** kth-order exponential golomb code */
649 static inline int get_ue_code(GetBitContext *gb, int order) {
650 if(order) {
651 int ret = get_ue_golomb(gb) << order;
652 return ret + get_bits(gb,order);
653 }
654 return get_ue_golomb(gb);
655 }
656
657 /**
658 * decode coefficients from one 8x8 block, dequantize, inverse transform
659 * and add them to sample block
660 * @param r pointer to 2D VLC table
661 * @param esc_golomb_order escape codes are k-golomb with this order k
662 * @param qp quantizer
663 * @param dst location of sample block
664 * @param stride line stride in frame buffer
665 */
666 static int decode_residual_block(AVSContext *h, GetBitContext *gb,
667 const residual_vlc_t *r, int esc_golomb_order,
668 int qp, uint8_t *dst, int stride) {
669 int i,pos = -1;
670 int level_code, esc_code, level, run, mask;
671 int level_buf[64];
672 int run_buf[64];
673 int dqm = dequant_mul[qp];
674 int dqs = dequant_shift[qp];
675 int dqa = 1 << (dqs - 1);
676 const uint8_t *scantab = ff_zigzag_direct;
677 DCTELEM block[64];
678
679 memset(block,0,64*sizeof(DCTELEM));
680 for(i=0;i<65;i++) {
681 level_code = get_ue_code(gb,r->golomb_order);
682 if(level_code >= ESCAPE_CODE) {
683 run = (level_code - ESCAPE_CODE) >> 1;
684 esc_code = get_ue_code(gb,esc_golomb_order);
685 level = esc_code + (run > r->max_run ? 1 : r->level_add[run]);
686 while(level > r->inc_limit)
687 r++;
688 mask = -(level_code & 1);
689 level = (level^mask) - mask;
690 } else {
691 if(level_code < 0)
692 return -1;
693 level = r->rltab[level_code][0];
694 if(!level) //end of block signal
695 break;
696 run = r->rltab[level_code][1];
697 r += r->rltab[level_code][2];
698 }
699 level_buf[i] = level;
700 run_buf[i] = run;
701 }
702 /* inverse scan and dequantization */
703 while(--i >= 0){
704 pos += 1 + run_buf[i];
705 if(pos > 63) {
706 av_log(h->s.avctx, AV_LOG_ERROR,
707 "position out of block bounds at pic %d MB(%d,%d)\n",
708 h->picture.poc, h->mbx, h->mby);
709 return -1;
710 }
711 block[scantab[pos]] = (level_buf[i]*dqm + dqa) >> dqs;
712 }
713 h->s.dsp.cavs_idct8_add(dst,block,stride);
714 return 0;
715 }
716
717
718 static inline void decode_residual_chroma(AVSContext *h) {
719 if(h->cbp & (1<<4))
720 decode_residual_block(h,&h->s.gb,chroma_2dvlc,0, chroma_qp[h->qp],
721 h->cu,h->c_stride);
722 if(h->cbp & (1<<5))
723 decode_residual_block(h,&h->s.gb,chroma_2dvlc,0, chroma_qp[h->qp],
724 h->cv,h->c_stride);
725 }
726
727 static inline int decode_residual_inter(AVSContext *h) {
728 int block;
729
730 /* get coded block pattern */
731 int cbp= get_ue_golomb(&h->s.gb);
732 if(cbp > 63){
733 av_log(h->s.avctx, AV_LOG_ERROR, "illegal inter cbp\n");
734 return -1;
735 }
736 h->cbp = cbp_tab[cbp][1];
737
738 /* get quantizer */
739 if(h->cbp && !h->qp_fixed)
740 h->qp += get_se_golomb(&h->s.gb);
741 for(block=0;block<4;block++)
742 if(h->cbp & (1<<block))
743 decode_residual_block(h,&h->s.gb,inter_2dvlc,0,h->qp,
744 h->cy + h->luma_scan[block], h->l_stride);
745 decode_residual_chroma(h);
746
747 return 0;
748 }
749
750 /*****************************************************************************
751 *
752 * macroblock level
753 *
754 ****************************************************************************/
755
756 /**
757 * initialise predictors for motion vectors and intra prediction
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 /**
799 * save predictors for later macroblocks and increase
800 * macroblock address
801 * @returns 0 if end of frame is reached, 1 otherwise
802 */
803 static inline int next_mb(AVSContext *h) {
804 int i;
805
806 h->flags |= A_AVAIL;
807 h->cy += 16;
808 h->cu += 8;
809 h->cv += 8;
810 /* copy mvs as predictors to the left */
811 for(i=0;i<=20;i+=4)
812 h->mv[i] = h->mv[i+2];
813 /* copy bottom mvs from cache to top line */
814 h->top_mv[0][h->mbx*2+0] = h->mv[MV_FWD_X2];
815 h->top_mv[0][h->mbx*2+1] = h->mv[MV_FWD_X3];
816 h->top_mv[1][h->mbx*2+0] = h->mv[MV_BWD_X2];
817 h->top_mv[1][h->mbx*2+1] = h->mv[MV_BWD_X3];
818 /* next MB address */
819 h->mbx++;
820 if(h->mbx == h->mb_width) { //new mb line
821 h->flags = B_AVAIL|C_AVAIL;
822 /* clear left pred_modes */
823 h->pred_mode_Y[3] = h->pred_mode_Y[6] = NOT_AVAIL;
824 /* clear left mv predictors */
825 for(i=0;i<=20;i+=4)
826 h->mv[i] = un_mv;
827 h->mbx = 0;
828 h->mby++;
829 /* re-calculate sample pointers */
830 h->cy = h->picture.data[0] + h->mby*16*h->l_stride;
831 h->cu = h->picture.data[1] + h->mby*8*h->c_stride;
832 h->cv = h->picture.data[2] + h->mby*8*h->c_stride;
833 if(h->mby == h->mb_height) { //frame end
834 return 0;
835 } else {
836 //check_for_slice(h);
837 }
838 }
839 return 1;
840 }
841
842 static int decode_mb_i(AVSContext *h, int cbp_code) {
843 GetBitContext *gb = &h->s.gb;
844 int block, pred_mode_uv;
845 uint8_t top[18];
846 uint8_t left[18];
847 uint8_t *d;
848
849 init_mb(h);
850
851 /* get intra prediction modes from stream */
852 for(block=0;block<4;block++) {
853 int nA,nB,predpred;
854 int pos = scan3x3[block];
855
856 nA = h->pred_mode_Y[pos-1];
857 nB = h->pred_mode_Y[pos-3];
858 predpred = FFMIN(nA,nB);
859 if(predpred == NOT_AVAIL) // if either is not available
860 predpred = INTRA_L_LP;
861 if(!get_bits1(gb)){
862 int rem_mode= get_bits(gb, 2);
863 predpred = rem_mode + (rem_mode >= predpred);
864 }
865 h->pred_mode_Y[pos] = predpred;
866 }
867 pred_mode_uv = get_ue_golomb(gb);
868 if(pred_mode_uv > 6) {
869 av_log(h->s.avctx, AV_LOG_ERROR, "illegal intra chroma pred mode\n");
870 return -1;
871 }
872
873 /* save pred modes before they get modified */
874 h->pred_mode_Y[3] = h->pred_mode_Y[5];
875 h->pred_mode_Y[6] = h->pred_mode_Y[8];
876 h->top_pred_Y[h->mbx*2+0] = h->pred_mode_Y[7];
877 h->top_pred_Y[h->mbx*2+1] = h->pred_mode_Y[8];
878
879 /* modify pred modes according to availability of neighbour samples */
880 if(!(h->flags & A_AVAIL)) {
881 modify_pred(left_modifier_l, &h->pred_mode_Y[4] );
882 modify_pred(left_modifier_l, &h->pred_mode_Y[7] );
883 modify_pred(left_modifier_c, &pred_mode_uv );
884 }
885 if(!(h->flags & B_AVAIL)) {
886 modify_pred(top_modifier_l, &h->pred_mode_Y[4] );
887 modify_pred(top_modifier_l, &h->pred_mode_Y[5] );
888 modify_pred(top_modifier_c, &pred_mode_uv );
889 }
890
891 /* get coded block pattern */
892 if(h->pic_type == FF_I_TYPE)
893 cbp_code = get_ue_golomb(gb);
894 if(cbp_code > 63){
895 av_log(h->s.avctx, AV_LOG_ERROR, "illegal intra cbp\n");
896 return -1;
897 }
898 h->cbp = cbp_tab[cbp_code][0];
899 if(h->cbp && !h->qp_fixed)
900 h->qp += get_se_golomb(gb); //qp_delta
901
902 /* luma intra prediction interleaved with residual decode/transform/add */
903 for(block=0;block<4;block++) {
904 d = h->cy + h->luma_scan[block];
905 load_intra_pred_luma(h, top, left, block);
906 h->intra_pred_l[h->pred_mode_Y[scan3x3[block]]]
907 (d, top, left, h->l_stride);
908 if(h->cbp & (1<<block))
909 decode_residual_block(h,gb,intra_2dvlc,1,h->qp,d,h->l_stride);
910 }
911
912 /* chroma intra prediction */
913 /* extend borders by one pixel */
914 h->left_border_u[9] = h->left_border_u[8];
915 h->left_border_v[9] = h->left_border_v[8];
916 h->top_border_u[h->mbx*10+9] = h->top_border_u[h->mbx*10+8];
917 h->top_border_v[h->mbx*10+9] = h->top_border_v[h->mbx*10+8];
918 if(h->mbx && h->mby) {
919 h->top_border_u[h->mbx*10] = h->left_border_u[0] = h->topleft_border_u;
920 h->top_border_v[h->mbx*10] = h->left_border_v[0] = h->topleft_border_v;
921 } else {
922 h->left_border_u[0] = h->left_border_u[1];
923 h->left_border_v[0] = h->left_border_v[1];
924 h->top_border_u[h->mbx*10] = h->top_border_u[h->mbx*10+1];
925 h->top_border_v[h->mbx*10] = h->top_border_v[h->mbx*10+1];
926 }
927 h->intra_pred_c[pred_mode_uv](h->cu, &h->top_border_u[h->mbx*10],
928 h->left_border_u, h->c_stride);
929 h->intra_pred_c[pred_mode_uv](h->cv, &h->top_border_v[h->mbx*10],
930 h->left_border_v, h->c_stride);
931
932 decode_residual_chroma(h);
933 filter_mb(h,I_8X8);
934
935 /* mark motion vectors as intra */
936 h->mv[MV_FWD_X0] = intra_mv;
937 set_mvs(&h->mv[MV_FWD_X0], BLK_16X16);
938 h->mv[MV_BWD_X0] = intra_mv;
939 set_mvs(&h->mv[MV_BWD_X0], BLK_16X16);
940 if(h->pic_type != FF_B_TYPE)
941 *h->col_type = I_8X8;
942
943 return 0;
944 }
945
946 static void decode_mb_p(AVSContext *h, enum mb_t mb_type) {
947 GetBitContext *gb = &h->s.gb;
948 int ref[4];
949
950 init_mb(h);
951 switch(mb_type) {
952 case P_SKIP:
953 mv_pred(h, MV_FWD_X0, MV_FWD_C2, MV_PRED_PSKIP, BLK_16X16, 0);
954 break;
955 case P_16X16:
956 ref[0] = h->ref_flag ? 0 : get_bits1(gb);
957 mv_pred(h, MV_FWD_X0, MV_FWD_C2, MV_PRED_MEDIAN, BLK_16X16,ref[0]);
958 break;
959 case P_16X8:
960 ref[0] = h->ref_flag ? 0 : get_bits1(gb);
961 ref[2] = h->ref_flag ? 0 : get_bits1(gb);
962 mv_pred(h, MV_FWD_X0, MV_FWD_C2, MV_PRED_TOP, BLK_16X8, ref[0]);
963 mv_pred(h, MV_FWD_X2, MV_FWD_A1, MV_PRED_LEFT, BLK_16X8, ref[2]);
964 break;
965 case P_8X16:
966 ref[0] = h->ref_flag ? 0 : get_bits1(gb);
967 ref[1] = h->ref_flag ? 0 : get_bits1(gb);
968 mv_pred(h, MV_FWD_X0, MV_FWD_B3, MV_PRED_LEFT, BLK_8X16, ref[0]);
969 mv_pred(h, MV_FWD_X1, MV_FWD_C2, MV_PRED_TOPRIGHT, BLK_8X16, ref[1]);
970 break;
971 case P_8X8:
972 ref[0] = h->ref_flag ? 0 : get_bits1(gb);
973 ref[1] = h->ref_flag ? 0 : get_bits1(gb);
974 ref[2] = h->ref_flag ? 0 : get_bits1(gb);
975 ref[3] = h->ref_flag ? 0 : get_bits1(gb);
976 mv_pred(h, MV_FWD_X0, MV_FWD_B3, MV_PRED_MEDIAN, BLK_8X8, ref[0]);
977 mv_pred(h, MV_FWD_X1, MV_FWD_C2, MV_PRED_MEDIAN, BLK_8X8, ref[1]);
978 mv_pred(h, MV_FWD_X2, MV_FWD_X1, MV_PRED_MEDIAN, BLK_8X8, ref[2]);
979 mv_pred(h, MV_FWD_X3, MV_FWD_X0, MV_PRED_MEDIAN, BLK_8X8, ref[3]);
980 }
981 inter_pred(h, mb_type);
982 store_mvs(h);
983 if(mb_type != P_SKIP)
984 decode_residual_inter(h);
985 filter_mb(h,mb_type);
986 *h->col_type = mb_type;
987 }
988
989 static void decode_mb_b(AVSContext *h, enum mb_t mb_type) {
990 int block;
991 enum sub_mb_t sub_type[4];
992 int flags;
993
994 init_mb(h);
995
996 /* reset all MVs */
997 h->mv[MV_FWD_X0] = dir_mv;
998 set_mvs(&h->mv[MV_FWD_X0], BLK_16X16);
999 h->mv[MV_BWD_X0] = dir_mv;
1000 set_mvs(&h->mv[MV_BWD_X0], BLK_16X16);
1001 switch(mb_type) {
1002 case B_SKIP:
1003 case B_DIRECT:
1004 if(!(*h->col_type)) {
1005 /* intra MB at co-location, do in-plane prediction */
1006 mv_pred(h, MV_FWD_X0, MV_FWD_C2, MV_PRED_BSKIP, BLK_16X16, 1);
1007 mv_pred(h, MV_BWD_X0, MV_BWD_C2, MV_PRED_BSKIP, BLK_16X16, 0);
1008 } else
1009 /* direct prediction from co-located P MB, block-wise */
1010 for(block=0;block<4;block++)
1011 mv_pred_direct(h,&h->mv[mv_scan[block]],
1012 &h->col_mv[(h->mby*h->mb_width+h->mbx)*4 + block]);
1013 break;
1014 case B_FWD_16X16:
1015 mv_pred(h, MV_FWD_X0, MV_FWD_C2, MV_PRED_MEDIAN, BLK_16X16, 1);
1016 break;
1017 case B_SYM_16X16:
1018 mv_pred(h, MV_FWD_X0, MV_FWD_C2, MV_PRED_MEDIAN, BLK_16X16, 1);
1019 mv_pred_sym(h, &h->mv[MV_FWD_X0], BLK_16X16);
1020 break;
1021 case B_BWD_16X16:
1022 mv_pred(h, MV_BWD_X0, MV_BWD_C2, MV_PRED_MEDIAN, BLK_16X16, 0);
1023 break;
1024 case B_8X8:
1025 for(block=0;block<4;block++)
1026 sub_type[block] = get_bits(&h->s.gb,2);
1027 for(block=0;block<4;block++) {
1028 switch(sub_type[block]) {
1029 case B_SUB_DIRECT:
1030 if(!(*h->col_type)) {
1031 /* intra MB at co-location, do in-plane prediction */
1032 mv_pred(h, mv_scan[block], mv_scan[block]-3,
1033 MV_PRED_BSKIP, BLK_8X8, 1);
1034 mv_pred(h, mv_scan[block]+MV_BWD_OFFS,
1035 mv_scan[block]-3+MV_BWD_OFFS,
1036 MV_PRED_BSKIP, BLK_8X8, 0);
1037 } else
1038 mv_pred_direct(h,&h->mv[mv_scan[block]],
1039 &h->col_mv[(h->mby*h->mb_width + h->mbx)*4 + block]);
1040 break;
1041 case B_SUB_FWD:
1042 mv_pred(h, mv_scan[block], mv_scan[block]-3,
1043 MV_PRED_MEDIAN, BLK_8X8, 1);
1044 break;
1045 case B_SUB_SYM:
1046 mv_pred(h, mv_scan[block], mv_scan[block]-3,
1047 MV_PRED_MEDIAN, BLK_8X8, 1);
1048 mv_pred_sym(h, &h->mv[mv_scan[block]], BLK_8X8);
1049 break;
1050 }
1051 }
1052 for(block=0;block<4;block++) {
1053 if(sub_type[block] == B_SUB_BWD)
1054 mv_pred(h, mv_scan[block]+MV_BWD_OFFS,
1055 mv_scan[block]+MV_BWD_OFFS-3,
1056 MV_PRED_MEDIAN, BLK_8X8, 0);
1057 }
1058 break;
1059 default:
1060 assert((mb_type > B_SYM_16X16) && (mb_type < B_8X8));
1061 flags = partition_flags[mb_type];
1062 if(mb_type & 1) { /* 16x8 macroblock types */
1063 if(flags & FWD0)
1064 mv_pred(h, MV_FWD_X0, MV_FWD_C2, MV_PRED_TOP, BLK_16X8, 1);
1065 if(flags & SYM0) {
1066 mv_pred(h, MV_FWD_X0, MV_FWD_C2, MV_PRED_TOP, BLK_16X8, 1);
1067 mv_pred_sym(h, &h->mv[MV_FWD_X0], BLK_16X8);
1068 }
1069 if(flags & FWD1)
1070 mv_pred(h, MV_FWD_X2, MV_FWD_A1, MV_PRED_LEFT, BLK_16X8, 1);
1071 if(flags & SYM1) {
1072 mv_pred(h, MV_FWD_X2, MV_FWD_A1, MV_PRED_LEFT, BLK_16X8, 1);
1073 mv_pred_sym(h, &h->mv[9], BLK_16X8);
1074 }
1075 if(flags & BWD0)
1076 mv_pred(h, MV_BWD_X0, MV_BWD_C2, MV_PRED_TOP, BLK_16X8, 0);
1077 if(flags & BWD1)
1078 mv_pred(h, MV_BWD_X2, MV_BWD_A1, MV_PRED_LEFT, BLK_16X8, 0);
1079 } else { /* 8x16 macroblock types */
1080 if(flags & FWD0)
1081 mv_pred(h, MV_FWD_X0, MV_FWD_B3, MV_PRED_LEFT, BLK_8X16, 1);
1082 if(flags & SYM0) {
1083 mv_pred(h, MV_FWD_X0, MV_FWD_B3, MV_PRED_LEFT, BLK_8X16, 1);
1084 mv_pred_sym(h, &h->mv[MV_FWD_X0], BLK_8X16);
1085 }
1086 if(flags & FWD1)
1087 mv_pred(h, MV_FWD_X1, MV_FWD_C2, MV_PRED_TOPRIGHT,BLK_8X16, 1);
1088 if(flags & SYM1) {
1089 mv_pred(h, MV_FWD_X1, MV_FWD_C2, MV_PRED_TOPRIGHT,BLK_8X16, 1);
1090 mv_pred_sym(h, &h->mv[6], BLK_8X16);
1091 }
1092 if(flags & BWD0)
1093 mv_pred(h, MV_BWD_X0, MV_BWD_B3, MV_PRED_LEFT, BLK_8X16, 0);
1094 if(flags & BWD1)
1095 mv_pred(h, MV_BWD_X1, MV_BWD_C2, MV_PRED_TOPRIGHT,BLK_8X16, 0);
1096 }
1097 }
1098 inter_pred(h, mb_type);
1099 if(mb_type != B_SKIP)
1100 decode_residual_inter(h);
1101 filter_mb(h,mb_type);
1102 }
1103
1104 /*****************************************************************************
1105 *
1106 * slice level
1107 *
1108 ****************************************************************************/
1109
1110 static inline int decode_slice_header(AVSContext *h, GetBitContext *gb) {
1111 if(h->stc > 0xAF)
1112 av_log(h->s.avctx, AV_LOG_ERROR, "unexpected start code 0x%02x\n", h->stc);
1113 h->mby = h->stc;
1114 if((h->mby == 0) && (!h->qp_fixed)){
1115 h->qp_fixed = get_bits1(gb);
1116 h->qp = get_bits(gb,6);
1117 }
1118 /* inter frame or second slice can have weighting params */
1119 if((h->pic_type != FF_I_TYPE) || (!h->pic_structure && h->mby >= h->mb_width/2))
1120 if(get_bits1(gb)) { //slice_weighting_flag
1121 av_log(h->s.avctx, AV_LOG_ERROR,
1122 "weighted prediction not yet supported\n");
1123 }
1124 return 0;
1125 }
1126
1127 static inline void check_for_slice(AVSContext *h) {
1128 GetBitContext *gb = &h->s.gb;
1129 int align;
1130 align = (-get_bits_count(gb)) & 7;
1131 if((show_bits_long(gb,24+align) & 0xFFFFFF) == 0x000001) {
1132 get_bits_long(gb,24+align);
1133 h->stc = get_bits(gb,8);
1134 decode_slice_header(h,gb);
1135 }
1136 }
1137
1138 /*****************************************************************************
1139 *
1140 * frame level
1141 *
1142 ****************************************************************************/
1143
1144 static void init_pic(AVSContext *h) {
1145 int i;
1146
1147 /* clear some predictors */
1148 for(i=0;i<=20;i+=4)
1149 h->mv[i] = un_mv;
1150 h->mv[MV_BWD_X0] = dir_mv;
1151 set_mvs(&h->mv[MV_BWD_X0], BLK_16X16);
1152 h->mv[MV_FWD_X0] = dir_mv;
1153 set_mvs(&h->mv[MV_FWD_X0], BLK_16X16);
1154 h->pred_mode_Y[3] = h->pred_mode_Y[6] = NOT_AVAIL;
1155 h->cy = h->picture.data[0];
1156 h->cu = h->picture.data[1];
1157 h->cv = h->picture.data[2];
1158 h->l_stride = h->picture.linesize[0];
1159 h->c_stride = h->picture.linesize[1];
1160 h->luma_scan[2] = 8*h->l_stride;
1161 h->luma_scan[3] = 8*h->l_stride+8;
1162 h->mbx = h->mby = 0;
1163 h->flags = 0;
1164 }
1165
1166 static int decode_pic(AVSContext *h) {
1167 MpegEncContext *s = &h->s;
1168 int skip_count;
1169 enum mb_t mb_type;
1170
1171 if (!s->context_initialized) {
1172 if (MPV_common_init(s) < 0)
1173 return -1;
1174 }
1175 get_bits(&s->gb,16);//bbv_dwlay
1176 if(h->stc == PIC_PB_START_CODE) {
1177 h->pic_type = get_bits(&s->gb,2) + FF_I_TYPE;
1178 /* make sure we have the reference frames we need */
1179 if(!h->DPB[0].data[0] ||
1180 (!h->DPB[1].data[0] && h->pic_type == FF_B_TYPE))
1181 return -1;
1182 } else {
1183 h->pic_type = FF_I_TYPE;
1184 if(get_bits1(&s->gb))
1185 get_bits(&s->gb,16);//time_code
1186 }
1187 /* release last B frame */
1188 if(h->picture.data[0])
1189 s->avctx->release_buffer(s->avctx, (AVFrame *)&h->picture);
1190
1191 s->avctx->get_buffer(s->avctx, (AVFrame *)&h->picture);
1192 init_pic(h);
1193 h->picture.poc = get_bits(&s->gb,8)*2;
1194
1195 /* get temporal distances and MV scaling factors */
1196 if(h->pic_type != FF_B_TYPE) {
1197 h->dist[0] = (h->picture.poc - h->DPB[0].poc + 512) % 512;
1198 } else {
1199 h->dist[0] = (h->DPB[0].poc - h->picture.poc + 512) % 512;
1200 }
1201 h->dist[1] = (h->picture.poc - h->DPB[1].poc + 512) % 512;
1202 h->scale_den[0] = h->dist[0] ? 512/h->dist[0] : 0;
1203 h->scale_den[1] = h->dist[1] ? 512/h->dist[1] : 0;
1204 if(h->pic_type == FF_B_TYPE) {
1205 h->sym_factor = h->dist[0]*h->scale_den[1];
1206 } else {
1207 h->direct_den[0] = h->dist[0] ? 16384/h->dist[0] : 0;
1208 h->direct_den[1] = h->dist[1] ? 16384/h->dist[1] : 0;
1209 }
1210
1211 if(s->low_delay)
1212 get_ue_golomb(&s->gb); //bbv_check_times
1213 h->progressive = get_bits1(&s->gb);
1214 if(h->progressive)
1215 h->pic_structure = 1;
1216 else if(!(h->pic_structure = get_bits1(&s->gb) && (h->stc == PIC_PB_START_CODE)) )
1217 get_bits1(&s->gb); //advanced_pred_mode_disable
1218 skip_bits1(&s->gb); //top_field_first
1219 skip_bits1(&s->gb); //repeat_first_field
1220 h->qp_fixed = get_bits1(&s->gb);
1221 h->qp = get_bits(&s->gb,6);
1222 if(h->pic_type == FF_I_TYPE) {
1223 if(!h->progressive && !h->pic_structure)
1224 skip_bits1(&s->gb);//what is this?
1225 skip_bits(&s->gb,4); //reserved bits
1226 } else {
1227 if(!(h->pic_type == FF_B_TYPE && h->pic_structure == 1))
1228 h->ref_flag = get_bits1(&s->gb);
1229 skip_bits(&s->gb,4); //reserved bits
1230 h->skip_mode_flag = get_bits1(&s->gb);
1231 }
1232 h->loop_filter_disable = get_bits1(&s->gb);
1233 if(!h->loop_filter_disable && get_bits1(&s->gb)) {
1234 h->alpha_offset = get_se_golomb(&s->gb);
1235 h->beta_offset = get_se_golomb(&s->gb);
1236 } else {
1237 h->alpha_offset = h->beta_offset = 0;
1238 }
1239 check_for_slice(h);
1240 if(h->pic_type == FF_I_TYPE) {
1241 do {
1242 decode_mb_i(h, 0);
1243 } while(next_mb(h));
1244 } else if(h->pic_type == FF_P_TYPE) {
1245 do {
1246 if(h->skip_mode_flag) {
1247 skip_count = get_ue_golomb(&s->gb);
1248 while(skip_count--) {
1249 decode_mb_p(h,P_SKIP);
1250 if(!next_mb(h))
1251 goto done;
1252 }
1253 mb_type = get_ue_golomb(&s->gb) + P_16X16;
1254 } else
1255 mb_type = get_ue_golomb(&s->gb) + P_SKIP;
1256 if(mb_type > P_8X8) {
1257 decode_mb_i(h, mb_type - P_8X8 - 1);
1258 } else
1259 decode_mb_p(h,mb_type);
1260 } while(next_mb(h));
1261 } else { /* FF_B_TYPE */
1262 do {
1263 if(h->skip_mode_flag) {
1264 skip_count = get_ue_golomb(&s->gb);
1265 while(skip_count--) {
1266 decode_mb_b(h,B_SKIP);
1267 if(!next_mb(h))
1268 goto done;
1269 }
1270 mb_type = get_ue_golomb(&s->gb) + B_DIRECT;
1271 } else
1272 mb_type = get_ue_golomb(&s->gb) + B_SKIP;
1273 if(mb_type > B_8X8) {
1274 decode_mb_i(h, mb_type - B_8X8 - 1);
1275 } else
1276 decode_mb_b(h,mb_type);
1277 } while(next_mb(h));
1278 }
1279 done:
1280 if(h->pic_type != FF_B_TYPE) {
1281 if(h->DPB[1].data[0])
1282 s->avctx->release_buffer(s->avctx, (AVFrame *)&h->DPB[1]);
1283 memcpy(&h->DPB[1], &h->DPB[0], sizeof(Picture));
1284 memcpy(&h->DPB[0], &h->picture, sizeof(Picture));
1285 memset(&h->picture,0,sizeof(Picture));
1286 }
1287 return 0;
1288 }
1289
1290 /*****************************************************************************
1291 *
1292 * headers and interface
1293 *
1294 ****************************************************************************/
1295
1296 /**
1297 * some predictions require data from the top-neighbouring macroblock.
1298 * this data has to be stored for one complete row of macroblocks
1299 * and this storage space is allocated here
1300 */
1301 static void init_top_lines(AVSContext *h) {
1302 /* alloc top line of predictors */
1303 h->top_qp = av_malloc( h->mb_width);
1304 h->top_mv[0] = av_malloc((h->mb_width*2+1)*sizeof(vector_t));
1305 h->top_mv[1] = av_malloc((h->mb_width*2+1)*sizeof(vector_t));
1306 h->top_pred_Y = av_malloc( h->mb_width*2*sizeof(*h->top_pred_Y));
1307 h->top_border_y = av_malloc((h->mb_width+1)*16);
1308 h->top_border_u = av_malloc((h->mb_width)*10);
1309 h->top_border_v = av_malloc((h->mb_width)*10);
1310
1311 /* alloc space for co-located MVs and types */
1312 h->col_mv = av_malloc( h->mb_width*h->mb_height*4*sizeof(vector_t));
1313 h->col_type_base = av_malloc(h->mb_width*h->mb_height);
1314 }
1315
1316 static int decode_seq_header(AVSContext *h) {
1317 MpegEncContext *s = &h->s;
1318 extern const AVRational ff_frame_rate_tab[];
1319 int frame_rate_code;
1320
1321 h->profile = get_bits(&s->gb,8);
1322 h->level = get_bits(&s->gb,8);
1323 skip_bits1(&s->gb); //progressive sequence
1324 s->width = get_bits(&s->gb,14);
1325 s->height = get_bits(&s->gb,14);
1326 skip_bits(&s->gb,2); //chroma format
1327 skip_bits(&s->gb,3); //sample_precision
1328 h->aspect_ratio = get_bits(&s->gb,4);
1329 frame_rate_code = get_bits(&s->gb,4);
1330 skip_bits(&s->gb,18);//bit_rate_lower
1331 skip_bits1(&s->gb); //marker_bit
1332 skip_bits(&s->gb,12);//bit_rate_upper
1333 s->low_delay = get_bits1(&s->gb);
1334 h->mb_width = (s->width + 15) >> 4;
1335 h->mb_height = (s->height + 15) >> 4;
1336 h->s.avctx->time_base.den = ff_frame_rate_tab[frame_rate_code].num;
1337 h->s.avctx->time_base.num = ff_frame_rate_tab[frame_rate_code].den;
1338 h->s.avctx->width = s->width;
1339 h->s.avctx->height = s->height;
1340 if(!h->top_qp)
1341 init_top_lines(h);
1342 return 0;
1343 }
1344
1345 /**
1346 * finds the end of the current frame in the bitstream.
1347 * @return the position of the first byte of the next frame, or -1
1348 */
1349 int ff_cavs_find_frame_end(ParseContext *pc, const uint8_t *buf, int buf_size) {
1350 int pic_found, i;
1351 uint32_t state;
1352
1353 pic_found= pc->frame_start_found;
1354 state= pc->state;
1355
1356 i=0;
1357 if(!pic_found){
1358 for(i=0; i<buf_size; i++){
1359 state= (state<<8) | buf[i];
1360 if(state == PIC_I_START_CODE || state == PIC_PB_START_CODE){
1361 i++;
1362 pic_found=1;
1363 break;
1364 }
1365 }
1366 }
1367
1368 if(pic_found){
1369 /* EOF considered as end of frame */
1370 if (buf_size == 0)
1371 return 0;
1372 for(; i<buf_size; i++){
1373 state= (state<<8) | buf[i];
1374 if((state&0xFFFFFF00) == 0x100){
1375 if(state < SLICE_MIN_START_CODE || state > SLICE_MAX_START_CODE){
1376 pc->frame_start_found=0;
1377 pc->state=-1;
1378 return i-3;
1379 }
1380 }
1381 }
1382 }
1383 pc->frame_start_found= pic_found;
1384 pc->state= state;
1385 return END_NOT_FOUND;
1386 }
1387
1388 void ff_cavs_flush(AVCodecContext * avctx) {
1389 AVSContext *h = avctx->priv_data;
1390 h->got_keyframe = 0;
1391 }
1392
1393 static int cavs_decode_frame(AVCodecContext * avctx,void *data, int *data_size,
1394 uint8_t * buf, int buf_size) {
1395 AVSContext *h = avctx->priv_data;
1396 MpegEncContext *s = &h->s;
1397 int input_size;
1398 const uint8_t *buf_end;
1399 const uint8_t *buf_ptr;
1400 AVFrame *picture = data;
1401 uint32_t stc;
1402
1403 s->avctx = avctx;
1404
1405 if (buf_size == 0) {
1406 if(!s->low_delay && h->DPB[0].data[0]) {
1407 *data_size = sizeof(AVPicture);
1408 *picture = *(AVFrame *) &h->DPB[0];
1409 }
1410 return 0;
1411 }
1412
1413 buf_ptr = buf;
1414 buf_end = buf + buf_size;
1415 for(;;) {
1416 buf_ptr = ff_find_start_code(buf_ptr,buf_end, &stc);
1417 if(stc & 0xFFFFFE00)
1418 return FFMAX(0, buf_ptr - buf - s->parse_context.last_index);
1419 input_size = (buf_end - buf_ptr)*8;
1420 switch(stc) {
1421 case SEQ_START_CODE:
1422 init_get_bits(&s->gb, buf_ptr, input_size);
1423 decode_seq_header(h);
1424 break;
1425 case PIC_I_START_CODE:
1426 if(!h->got_keyframe) {
1427 if(h->DPB[0].data[0])
1428 avctx->release_buffer(avctx, (AVFrame *)&h->DPB[0]);
1429 if(h->DPB[1].data[0])
1430 avctx->release_buffer(avctx, (AVFrame *)&h->DPB[1]);
1431 h->got_keyframe = 1;
1432 }
1433 case PIC_PB_START_CODE:
1434 *data_size = 0;
1435 if(!h->got_keyframe)
1436 break;
1437 init_get_bits(&s->gb, buf_ptr, input_size);
1438 h->stc = stc;
1439 if(decode_pic(h))
1440 break;
1441 *data_size = sizeof(AVPicture);
1442 if(h->pic_type != FF_B_TYPE) {
1443 if(h->DPB[1].data[0]) {
1444 *picture = *(AVFrame *) &h->DPB[1];
1445 } else {
1446 *data_size = 0;
1447 }
1448 } else
1449 *picture = *(AVFrame *) &h->picture;
1450 break;
1451 case EXT_START_CODE:
1452 //mpeg_decode_extension(avctx,buf_ptr, input_size);
1453 break;
1454 case USER_START_CODE:
1455 //mpeg_decode_user_data(avctx,buf_ptr, input_size);
1456 break;
1457 default:
1458 if (stc >= SLICE_MIN_START_CODE &&
1459 stc <= SLICE_MAX_START_CODE) {
1460 init_get_bits(&s->gb, buf_ptr, input_size);
1461 decode_slice_header(h, &s->gb);
1462 }
1463 break;
1464 }
1465 }
1466 }
1467
1468 static int cavs_decode_init(AVCodecContext * avctx) {
1469 AVSContext *h = avctx->priv_data;
1470 MpegEncContext * const s = &h->s;
1471
1472 MPV_decode_defaults(s);
1473 s->avctx = avctx;
1474
1475 avctx->pix_fmt= PIX_FMT_YUV420P;
1476
1477 h->luma_scan[0] = 0;
1478 h->luma_scan[1] = 8;
1479 h->intra_pred_l[ INTRA_L_VERT] = intra_pred_vert;
1480 h->intra_pred_l[ INTRA_L_HORIZ] = intra_pred_horiz;
1481 h->intra_pred_l[ INTRA_L_LP] = intra_pred_lp;
1482 h->intra_pred_l[ INTRA_L_DOWN_LEFT] = intra_pred_down_left;
1483 h->intra_pred_l[INTRA_L_DOWN_RIGHT] = intra_pred_down_right;
1484 h->intra_pred_l[ INTRA_L_LP_LEFT] = intra_pred_lp_left;
1485 h->intra_pred_l[ INTRA_L_LP_TOP] = intra_pred_lp_top;
1486 h->intra_pred_l[ INTRA_L_DC_128] = intra_pred_dc_128;
1487 h->intra_pred_c[ INTRA_C_LP] = intra_pred_lp;
1488 h->intra_pred_c[ INTRA_C_HORIZ] = intra_pred_horiz;
1489 h->intra_pred_c[ INTRA_C_VERT] = intra_pred_vert;
1490 h->intra_pred_c[ INTRA_C_PLANE] = intra_pred_plane;
1491 h->intra_pred_c[ INTRA_C_LP_LEFT] = intra_pred_lp_left;
1492 h->intra_pred_c[ INTRA_C_LP_TOP] = intra_pred_lp_top;
1493 h->intra_pred_c[ INTRA_C_DC_128] = intra_pred_dc_128;
1494 h->mv[ 7] = un_mv;
1495 h->mv[19] = un_mv;
1496 return 0;
1497 }
1498
1499 static int cavs_decode_end(AVCodecContext * avctx) {
1500 AVSContext *h = avctx->priv_data;
1501
1502 av_free(h->top_qp);
1503 av_free(h->top_mv[0]);
1504 av_free(h->top_mv[1]);
1505 av_free(h->top_pred_Y);
1506 av_free(h->top_border_y);
1507 av_free(h->top_border_u);
1508 av_free(h->top_border_v);
1509 av_free(h->col_mv);
1510 av_free(h->col_type_base);
1511 return 0;
1512 }
1513
1514 AVCodec cavs_decoder = {
1515 "cavs",
1516 CODEC_TYPE_VIDEO,
1517 CODEC_ID_CAVS,
1518 sizeof(AVSContext),
1519 cavs_decode_init,
1520 NULL,
1521 cavs_decode_end,
1522 cavs_decode_frame,
1523 CODEC_CAP_DR1 | CODEC_CAP_DELAY,
1524 .flush= ff_cavs_flush,
1525 };