avoid redundant local variables
[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
580 mvP->ref = ref;
581 mvP->dist = h->dist[mvP->ref];
582 if(mvC->ref == NOT_AVAIL)
583 mvC = &h->mv[nP-5]; // set to top-left (mvD)
584 if(mode == MV_PRED_PSKIP) {
585 if((mvA->ref == NOT_AVAIL) || (mvB->ref == NOT_AVAIL) ||
586 ((mvA->x | mvA->y | mvA->ref) == 0) ||
587 ((mvB->x | mvB->y | mvB->ref) == 0) ) {
588 mvP->x = mvP->y = 0;
589 set_mvs(mvP,size);
590 return;
591 }
592 }
593 /* if there is only one suitable candidate, take it */
594 if((mvA->ref >= 0) && (mvB->ref < 0) && (mvC->ref < 0)) {
595 mvP->x = mvA->x;
596 mvP->y = mvA->y;
597 } else if((mvA->ref < 0) && (mvB->ref >= 0) && (mvC->ref < 0)) {
598 mvP->x = mvB->x;
599 mvP->y = mvB->y;
600 } else if((mvA->ref < 0) && (mvB->ref < 0) && (mvC->ref >= 0)) {
601 mvP->x = mvC->x;
602 mvP->y = mvC->y;
603 } else {
604 switch(mode) {
605 case MV_PRED_LEFT:
606 if(mvA->ref == mvP->ref) {
607 mvP->x = mvA->x;
608 mvP->y = mvA->y;
609 } else
610 mv_pred_median(h, mvP, mvA, mvB, mvC);
611 break;
612 case MV_PRED_TOP:
613 if(mvB->ref == mvP->ref) {
614 mvP->x = mvB->x;
615 mvP->y = mvB->y;
616 } else
617 mv_pred_median(h, mvP, mvA, mvB, mvC);
618 break;
619 case MV_PRED_TOPRIGHT:
620 if(mvC->ref == mvP->ref) {
621 mvP->x = mvC->x;
622 mvP->y = mvC->y;
623 } else
624 mv_pred_median(h, mvP, mvA, mvB, mvC);
625 break;
626 default:
627 mv_pred_median(h, mvP, mvA, mvB, mvC);
628 break;
629 }
630 }
631 if(mode < MV_PRED_PSKIP) {
632 mvP->x += get_se_golomb(&h->s.gb);
633 mvP->y += get_se_golomb(&h->s.gb);
634 }
635 set_mvs(mvP,size);
636 }
637
638 /*****************************************************************************
639 *
640 * residual data decoding
641 *
642 ****************************************************************************/
643
644 /** kth-order exponential golomb code */
645 static inline int get_ue_code(GetBitContext *gb, int order) {
646 if(order) {
647 int ret = get_ue_golomb(gb) << order;
648 return ret + get_bits(gb,order);
649 }
650 return get_ue_golomb(gb);
651 }
652
653 /**
654 * decode coefficients from one 8x8 block, dequantize, inverse transform
655 * and add them to sample block
656 * @param r pointer to 2D VLC table
657 * @param esc_golomb_order escape codes are k-golomb with this order k
658 * @param qp quantizer
659 * @param dst location of sample block
660 * @param stride line stride in frame buffer
661 */
662 static int decode_residual_block(AVSContext *h, GetBitContext *gb,
663 const residual_vlc_t *r, int esc_golomb_order,
664 int qp, uint8_t *dst, int stride) {
665 int i,pos = -1;
666 int level_code, esc_code, level, run, mask;
667 int level_buf[64];
668 int run_buf[64];
669 int dqm = dequant_mul[qp];
670 int dqs = dequant_shift[qp];
671 int dqa = 1 << (dqs - 1);
672 const uint8_t *scantab = ff_zigzag_direct;
673 DCTELEM block[64];
674
675 memset(block,0,64*sizeof(DCTELEM));
676 for(i=0;i<65;i++) {
677 level_code = get_ue_code(gb,r->golomb_order);
678 if(level_code >= ESCAPE_CODE) {
679 run = (level_code - ESCAPE_CODE) >> 1;
680 esc_code = get_ue_code(gb,esc_golomb_order);
681 level = esc_code + (run > r->max_run ? 1 : r->level_add[run]);
682 while(level > r->inc_limit)
683 r++;
684 mask = -(level_code & 1);
685 level = (level^mask) - mask;
686 } else {
687 if(level_code < 0)
688 return -1;
689 level = r->rltab[level_code][0];
690 if(!level) //end of block signal
691 break;
692 run = r->rltab[level_code][1];
693 r += r->rltab[level_code][2];
694 }
695 level_buf[i] = level;
696 run_buf[i] = run;
697 }
698 /* inverse scan and dequantization */
699 while(--i >= 0){
700 pos += 1 + run_buf[i];
701 if(pos > 63) {
702 av_log(h->s.avctx, AV_LOG_ERROR,
703 "position out of block bounds at pic %d MB(%d,%d)\n",
704 h->picture.poc, h->mbx, h->mby);
705 return -1;
706 }
707 block[scantab[pos]] = (level_buf[i]*dqm + dqa) >> dqs;
708 }
709 h->s.dsp.cavs_idct8_add(dst,block,stride);
710 return 0;
711 }
712
713
714 static inline void decode_residual_chroma(AVSContext *h) {
715 if(h->cbp & (1<<4))
716 decode_residual_block(h,&h->s.gb,chroma_2dvlc,0, chroma_qp[h->qp],
717 h->cu,h->c_stride);
718 if(h->cbp & (1<<5))
719 decode_residual_block(h,&h->s.gb,chroma_2dvlc,0, chroma_qp[h->qp],
720 h->cv,h->c_stride);
721 }
722
723 static inline int decode_residual_inter(AVSContext *h) {
724 int block;
725
726 /* get coded block pattern */
727 int cbp= get_ue_golomb(&h->s.gb);
728 if(cbp > 63){
729 av_log(h->s.avctx, AV_LOG_ERROR, "illegal inter cbp\n");
730 return -1;
731 }
732 h->cbp = cbp_tab[cbp][1];
733
734 /* get quantizer */
735 if(h->cbp && !h->qp_fixed)
736 h->qp += get_se_golomb(&h->s.gb);
737 for(block=0;block<4;block++)
738 if(h->cbp & (1<<block))
739 decode_residual_block(h,&h->s.gb,inter_2dvlc,0,h->qp,
740 h->cy + h->luma_scan[block], h->l_stride);
741 decode_residual_chroma(h);
742
743 return 0;
744 }
745
746 /*****************************************************************************
747 *
748 * macroblock level
749 *
750 ****************************************************************************/
751
752 /**
753 * initialise predictors for motion vectors and intra prediction
754 */
755 static inline void init_mb(AVSContext *h) {
756 int i;
757
758 /* copy predictors from top line (MB B and C) into cache */
759 for(i=0;i<3;i++) {
760 h->mv[MV_FWD_B2+i] = h->top_mv[0][h->mbx*2+i];
761 h->mv[MV_BWD_B2+i] = h->top_mv[1][h->mbx*2+i];
762 }
763 h->pred_mode_Y[1] = h->top_pred_Y[h->mbx*2+0];
764 h->pred_mode_Y[2] = h->top_pred_Y[h->mbx*2+1];
765 /* clear top predictors if MB B is not available */
766 if(!(h->flags & B_AVAIL)) {
767 h->mv[MV_FWD_B2] = un_mv;
768 h->mv[MV_FWD_B3] = un_mv;
769 h->mv[MV_BWD_B2] = un_mv;
770 h->mv[MV_BWD_B3] = un_mv;
771 h->pred_mode_Y[1] = h->pred_mode_Y[2] = NOT_AVAIL;
772 h->flags &= ~(C_AVAIL|D_AVAIL);
773 } else if(h->mbx) {
774 h->flags |= D_AVAIL;
775 }
776 if(h->mbx == h->mb_width-1) //MB C not available
777 h->flags &= ~C_AVAIL;
778 /* clear top-right predictors if MB C is not available */
779 if(!(h->flags & C_AVAIL)) {
780 h->mv[MV_FWD_C2] = un_mv;
781 h->mv[MV_BWD_C2] = un_mv;
782 }
783 /* clear top-left predictors if MB D is not available */
784 if(!(h->flags & D_AVAIL)) {
785 h->mv[MV_FWD_D3] = un_mv;
786 h->mv[MV_BWD_D3] = un_mv;
787 }
788 /* set pointer for co-located macroblock type */
789 h->col_type = &h->col_type_base[h->mby*h->mb_width + h->mbx];
790 }
791
792 static inline void check_for_slice(AVSContext *h);
793
794 /**
795 * save predictors for later macroblocks and increase
796 * macroblock address
797 * @returns 0 if end of frame is reached, 1 otherwise
798 */
799 static inline int next_mb(AVSContext *h) {
800 int i;
801
802 h->flags |= A_AVAIL;
803 h->cy += 16;
804 h->cu += 8;
805 h->cv += 8;
806 /* copy mvs as predictors to the left */
807 for(i=0;i<=20;i+=4)
808 h->mv[i] = h->mv[i+2];
809 /* copy bottom mvs from cache to top line */
810 h->top_mv[0][h->mbx*2+0] = h->mv[MV_FWD_X2];
811 h->top_mv[0][h->mbx*2+1] = h->mv[MV_FWD_X3];
812 h->top_mv[1][h->mbx*2+0] = h->mv[MV_BWD_X2];
813 h->top_mv[1][h->mbx*2+1] = h->mv[MV_BWD_X3];
814 /* next MB address */
815 h->mbx++;
816 if(h->mbx == h->mb_width) { //new mb line
817 h->flags = B_AVAIL|C_AVAIL;
818 /* clear left pred_modes */
819 h->pred_mode_Y[3] = h->pred_mode_Y[6] = NOT_AVAIL;
820 /* clear left mv predictors */
821 for(i=0;i<=20;i+=4)
822 h->mv[i] = un_mv;
823 h->mbx = 0;
824 h->mby++;
825 /* re-calculate sample pointers */
826 h->cy = h->picture.data[0] + h->mby*16*h->l_stride;
827 h->cu = h->picture.data[1] + h->mby*8*h->c_stride;
828 h->cv = h->picture.data[2] + h->mby*8*h->c_stride;
829 if(h->mby == h->mb_height) { //frame end
830 return 0;
831 } else {
832 //check_for_slice(h);
833 }
834 }
835 return 1;
836 }
837
838 static int decode_mb_i(AVSContext *h, int cbp_code) {
839 GetBitContext *gb = &h->s.gb;
840 int block, pred_mode_uv;
841 uint8_t top[18];
842 uint8_t left[18];
843 uint8_t *d;
844
845 init_mb(h);
846
847 /* get intra prediction modes from stream */
848 for(block=0;block<4;block++) {
849 int nA,nB,predpred;
850 int pos = scan3x3[block];
851
852 nA = h->pred_mode_Y[pos-1];
853 nB = h->pred_mode_Y[pos-3];
854 predpred = FFMIN(nA,nB);
855 if(predpred == NOT_AVAIL) // if either is not available
856 predpred = INTRA_L_LP;
857 if(!get_bits1(gb)){
858 int rem_mode= get_bits(gb, 2);
859 predpred = rem_mode + (rem_mode >= predpred);
860 }
861 h->pred_mode_Y[pos] = predpred;
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 return -1;
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(h->pic_type == FF_I_TYPE)
889 cbp_code = get_ue_golomb(gb);
890 if(cbp_code > 63){
891 av_log(h->s.avctx, AV_LOG_ERROR, "illegal intra cbp\n");
892 return -1;
893 }
894 h->cbp = cbp_tab[cbp_code][0];
895 if(h->cbp && !h->qp_fixed)
896 h->qp += get_se_golomb(gb); //qp_delta
897
898 /* luma intra prediction interleaved with residual decode/transform/add */
899 for(block=0;block<4;block++) {
900 d = h->cy + h->luma_scan[block];
901 load_intra_pred_luma(h, top, left, block);
902 h->intra_pred_l[h->pred_mode_Y[scan3x3[block]]]
903 (d, top, left, h->l_stride);
904 if(h->cbp & (1<<block))
905 decode_residual_block(h,gb,intra_2dvlc,1,h->qp,d,h->l_stride);
906 }
907
908 /* chroma intra prediction */
909 /* extend borders by one pixel */
910 h->left_border_u[9] = h->left_border_u[8];
911 h->left_border_v[9] = h->left_border_v[8];
912 h->top_border_u[h->mbx*10+9] = h->top_border_u[h->mbx*10+8];
913 h->top_border_v[h->mbx*10+9] = h->top_border_v[h->mbx*10+8];
914 if(h->mbx && h->mby) {
915 h->top_border_u[h->mbx*10] = h->left_border_u[0] = h->topleft_border_u;
916 h->top_border_v[h->mbx*10] = h->left_border_v[0] = h->topleft_border_v;
917 } else {
918 h->left_border_u[0] = h->left_border_u[1];
919 h->left_border_v[0] = h->left_border_v[1];
920 h->top_border_u[h->mbx*10] = h->top_border_u[h->mbx*10+1];
921 h->top_border_v[h->mbx*10] = h->top_border_v[h->mbx*10+1];
922 }
923 h->intra_pred_c[pred_mode_uv](h->cu, &h->top_border_u[h->mbx*10],
924 h->left_border_u, h->c_stride);
925 h->intra_pred_c[pred_mode_uv](h->cv, &h->top_border_v[h->mbx*10],
926 h->left_border_v, h->c_stride);
927
928 decode_residual_chroma(h);
929 filter_mb(h,I_8X8);
930
931 /* mark motion vectors as intra */
932 h->mv[MV_FWD_X0] = intra_mv;
933 set_mvs(&h->mv[MV_FWD_X0], BLK_16X16);
934 h->mv[MV_BWD_X0] = intra_mv;
935 set_mvs(&h->mv[MV_BWD_X0], BLK_16X16);
936 if(h->pic_type != FF_B_TYPE)
937 *h->col_type = I_8X8;
938
939 return 0;
940 }
941
942 static void decode_mb_p(AVSContext *h, enum mb_t mb_type) {
943 GetBitContext *gb = &h->s.gb;
944 int ref[4];
945
946 init_mb(h);
947 switch(mb_type) {
948 case P_SKIP:
949 mv_pred(h, MV_FWD_X0, MV_FWD_C2, MV_PRED_PSKIP, BLK_16X16, 0);
950 break;
951 case P_16X16:
952 ref[0] = h->ref_flag ? 0 : get_bits1(gb);
953 mv_pred(h, MV_FWD_X0, MV_FWD_C2, MV_PRED_MEDIAN, BLK_16X16,ref[0]);
954 break;
955 case P_16X8:
956 ref[0] = h->ref_flag ? 0 : get_bits1(gb);
957 ref[2] = h->ref_flag ? 0 : get_bits1(gb);
958 mv_pred(h, MV_FWD_X0, MV_FWD_C2, MV_PRED_TOP, BLK_16X8, ref[0]);
959 mv_pred(h, MV_FWD_X2, MV_FWD_A1, MV_PRED_LEFT, BLK_16X8, ref[2]);
960 break;
961 case P_8X16:
962 ref[0] = h->ref_flag ? 0 : get_bits1(gb);
963 ref[1] = h->ref_flag ? 0 : get_bits1(gb);
964 mv_pred(h, MV_FWD_X0, MV_FWD_B3, MV_PRED_LEFT, BLK_8X16, ref[0]);
965 mv_pred(h, MV_FWD_X1, MV_FWD_C2, MV_PRED_TOPRIGHT, BLK_8X16, ref[1]);
966 break;
967 case P_8X8:
968 ref[0] = h->ref_flag ? 0 : get_bits1(gb);
969 ref[1] = h->ref_flag ? 0 : get_bits1(gb);
970 ref[2] = h->ref_flag ? 0 : get_bits1(gb);
971 ref[3] = h->ref_flag ? 0 : get_bits1(gb);
972 mv_pred(h, MV_FWD_X0, MV_FWD_B3, MV_PRED_MEDIAN, BLK_8X8, ref[0]);
973 mv_pred(h, MV_FWD_X1, MV_FWD_C2, MV_PRED_MEDIAN, BLK_8X8, ref[1]);
974 mv_pred(h, MV_FWD_X2, MV_FWD_X1, MV_PRED_MEDIAN, BLK_8X8, ref[2]);
975 mv_pred(h, MV_FWD_X3, MV_FWD_X0, MV_PRED_MEDIAN, BLK_8X8, ref[3]);
976 }
977 inter_pred(h, mb_type);
978 store_mvs(h);
979 if(mb_type != P_SKIP)
980 decode_residual_inter(h);
981 filter_mb(h,mb_type);
982 *h->col_type = mb_type;
983 }
984
985 static void decode_mb_b(AVSContext *h, enum mb_t mb_type) {
986 int block;
987 enum sub_mb_t sub_type[4];
988 int flags;
989
990 init_mb(h);
991
992 /* reset all MVs */
993 h->mv[MV_FWD_X0] = dir_mv;
994 set_mvs(&h->mv[MV_FWD_X0], BLK_16X16);
995 h->mv[MV_BWD_X0] = dir_mv;
996 set_mvs(&h->mv[MV_BWD_X0], BLK_16X16);
997 switch(mb_type) {
998 case B_SKIP:
999 case B_DIRECT:
1000 if(!(*h->col_type)) {
1001 /* intra MB at co-location, do in-plane prediction */
1002 mv_pred(h, MV_FWD_X0, MV_FWD_C2, MV_PRED_BSKIP, BLK_16X16, 1);
1003 mv_pred(h, MV_BWD_X0, MV_BWD_C2, MV_PRED_BSKIP, BLK_16X16, 0);
1004 } else
1005 /* direct prediction from co-located P MB, block-wise */
1006 for(block=0;block<4;block++)
1007 mv_pred_direct(h,&h->mv[mv_scan[block]],
1008 &h->col_mv[(h->mby*h->mb_width+h->mbx)*4 + block]);
1009 break;
1010 case B_FWD_16X16:
1011 mv_pred(h, MV_FWD_X0, MV_FWD_C2, MV_PRED_MEDIAN, BLK_16X16, 1);
1012 break;
1013 case B_SYM_16X16:
1014 mv_pred(h, MV_FWD_X0, MV_FWD_C2, MV_PRED_MEDIAN, BLK_16X16, 1);
1015 mv_pred_sym(h, &h->mv[MV_FWD_X0], BLK_16X16);
1016 break;
1017 case B_BWD_16X16:
1018 mv_pred(h, MV_BWD_X0, MV_BWD_C2, MV_PRED_MEDIAN, BLK_16X16, 0);
1019 break;
1020 case B_8X8:
1021 for(block=0;block<4;block++)
1022 sub_type[block] = get_bits(&h->s.gb,2);
1023 for(block=0;block<4;block++) {
1024 switch(sub_type[block]) {
1025 case B_SUB_DIRECT:
1026 if(!(*h->col_type)) {
1027 /* intra MB at co-location, do in-plane prediction */
1028 mv_pred(h, mv_scan[block], mv_scan[block]-3,
1029 MV_PRED_BSKIP, BLK_8X8, 1);
1030 mv_pred(h, mv_scan[block]+MV_BWD_OFFS,
1031 mv_scan[block]-3+MV_BWD_OFFS,
1032 MV_PRED_BSKIP, BLK_8X8, 0);
1033 } else
1034 mv_pred_direct(h,&h->mv[mv_scan[block]],
1035 &h->col_mv[(h->mby*h->mb_width + h->mbx)*4 + block]);
1036 break;
1037 case B_SUB_FWD:
1038 mv_pred(h, mv_scan[block], mv_scan[block]-3,
1039 MV_PRED_MEDIAN, BLK_8X8, 1);
1040 break;
1041 case B_SUB_SYM:
1042 mv_pred(h, mv_scan[block], mv_scan[block]-3,
1043 MV_PRED_MEDIAN, BLK_8X8, 1);
1044 mv_pred_sym(h, &h->mv[mv_scan[block]], BLK_8X8);
1045 break;
1046 }
1047 }
1048 for(block=0;block<4;block++) {
1049 if(sub_type[block] == B_SUB_BWD)
1050 mv_pred(h, mv_scan[block]+MV_BWD_OFFS,
1051 mv_scan[block]+MV_BWD_OFFS-3,
1052 MV_PRED_MEDIAN, BLK_8X8, 0);
1053 }
1054 break;
1055 default:
1056 assert((mb_type > B_SYM_16X16) && (mb_type < B_8X8));
1057 flags = partition_flags[mb_type];
1058 if(mb_type & 1) { /* 16x8 macroblock types */
1059 if(flags & FWD0)
1060 mv_pred(h, MV_FWD_X0, MV_FWD_C2, MV_PRED_TOP, BLK_16X8, 1);
1061 if(flags & SYM0) {
1062 mv_pred(h, MV_FWD_X0, MV_FWD_C2, MV_PRED_TOP, BLK_16X8, 1);
1063 mv_pred_sym(h, &h->mv[MV_FWD_X0], BLK_16X8);
1064 }
1065 if(flags & FWD1)
1066 mv_pred(h, MV_FWD_X2, MV_FWD_A1, MV_PRED_LEFT, BLK_16X8, 1);
1067 if(flags & SYM1) {
1068 mv_pred(h, MV_FWD_X2, MV_FWD_A1, MV_PRED_LEFT, BLK_16X8, 1);
1069 mv_pred_sym(h, &h->mv[9], BLK_16X8);
1070 }
1071 if(flags & BWD0)
1072 mv_pred(h, MV_BWD_X0, MV_BWD_C2, MV_PRED_TOP, BLK_16X8, 0);
1073 if(flags & BWD1)
1074 mv_pred(h, MV_BWD_X2, MV_BWD_A1, MV_PRED_LEFT, BLK_16X8, 0);
1075 } else { /* 8x16 macroblock types */
1076 if(flags & FWD0)
1077 mv_pred(h, MV_FWD_X0, MV_FWD_B3, MV_PRED_LEFT, BLK_8X16, 1);
1078 if(flags & SYM0) {
1079 mv_pred(h, MV_FWD_X0, MV_FWD_B3, MV_PRED_LEFT, BLK_8X16, 1);
1080 mv_pred_sym(h, &h->mv[MV_FWD_X0], BLK_8X16);
1081 }
1082 if(flags & FWD1)
1083 mv_pred(h, MV_FWD_X1, MV_FWD_C2, MV_PRED_TOPRIGHT,BLK_8X16, 1);
1084 if(flags & SYM1) {
1085 mv_pred(h, MV_FWD_X1, MV_FWD_C2, MV_PRED_TOPRIGHT,BLK_8X16, 1);
1086 mv_pred_sym(h, &h->mv[6], BLK_8X16);
1087 }
1088 if(flags & BWD0)
1089 mv_pred(h, MV_BWD_X0, MV_BWD_B3, MV_PRED_LEFT, BLK_8X16, 0);
1090 if(flags & BWD1)
1091 mv_pred(h, MV_BWD_X1, MV_BWD_C2, MV_PRED_TOPRIGHT,BLK_8X16, 0);
1092 }
1093 }
1094 inter_pred(h, mb_type);
1095 if(mb_type != B_SKIP)
1096 decode_residual_inter(h);
1097 filter_mb(h,mb_type);
1098 }
1099
1100 /*****************************************************************************
1101 *
1102 * slice level
1103 *
1104 ****************************************************************************/
1105
1106 static inline int decode_slice_header(AVSContext *h, GetBitContext *gb) {
1107 if(h->stc > 0xAF)
1108 av_log(h->s.avctx, AV_LOG_ERROR, "unexpected start code 0x%02x\n", h->stc);
1109 h->mby = h->stc;
1110 if((h->mby == 0) && (!h->qp_fixed)){
1111 h->qp_fixed = get_bits1(gb);
1112 h->qp = get_bits(gb,6);
1113 }
1114 /* inter frame or second slice can have weighting params */
1115 if((h->pic_type != FF_I_TYPE) || (!h->pic_structure && h->mby >= h->mb_width/2))
1116 if(get_bits1(gb)) { //slice_weighting_flag
1117 av_log(h->s.avctx, AV_LOG_ERROR,
1118 "weighted prediction not yet supported\n");
1119 }
1120 return 0;
1121 }
1122
1123 static inline void check_for_slice(AVSContext *h) {
1124 GetBitContext *gb = &h->s.gb;
1125 int align;
1126 align = (-get_bits_count(gb)) & 7;
1127 if((show_bits_long(gb,24+align) & 0xFFFFFF) == 0x000001) {
1128 get_bits_long(gb,24+align);
1129 h->stc = get_bits(gb,8);
1130 decode_slice_header(h,gb);
1131 }
1132 }
1133
1134 /*****************************************************************************
1135 *
1136 * frame level
1137 *
1138 ****************************************************************************/
1139
1140 static void init_pic(AVSContext *h) {
1141 int i;
1142
1143 /* clear some predictors */
1144 for(i=0;i<=20;i+=4)
1145 h->mv[i] = un_mv;
1146 h->mv[MV_BWD_X0] = dir_mv;
1147 set_mvs(&h->mv[MV_BWD_X0], BLK_16X16);
1148 h->mv[MV_FWD_X0] = dir_mv;
1149 set_mvs(&h->mv[MV_FWD_X0], BLK_16X16);
1150 h->pred_mode_Y[3] = h->pred_mode_Y[6] = NOT_AVAIL;
1151 h->cy = h->picture.data[0];
1152 h->cu = h->picture.data[1];
1153 h->cv = h->picture.data[2];
1154 h->l_stride = h->picture.linesize[0];
1155 h->c_stride = h->picture.linesize[1];
1156 h->luma_scan[2] = 8*h->l_stride;
1157 h->luma_scan[3] = 8*h->l_stride+8;
1158 h->mbx = h->mby = 0;
1159 h->flags = 0;
1160 }
1161
1162 static int decode_pic(AVSContext *h) {
1163 MpegEncContext *s = &h->s;
1164 int skip_count;
1165 enum mb_t mb_type;
1166
1167 if (!s->context_initialized) {
1168 if (MPV_common_init(s) < 0)
1169 return -1;
1170 }
1171 get_bits(&s->gb,16);//bbv_dwlay
1172 if(h->stc == PIC_PB_START_CODE) {
1173 h->pic_type = get_bits(&s->gb,2) + FF_I_TYPE;
1174 /* make sure we have the reference frames we need */
1175 if(!h->DPB[0].data[0] ||
1176 (!h->DPB[1].data[0] && h->pic_type == FF_B_TYPE))
1177 return -1;
1178 } else {
1179 h->pic_type = FF_I_TYPE;
1180 if(get_bits1(&s->gb))
1181 get_bits(&s->gb,16);//time_code
1182 }
1183 /* release last B frame */
1184 if(h->picture.data[0])
1185 s->avctx->release_buffer(s->avctx, (AVFrame *)&h->picture);
1186
1187 s->avctx->get_buffer(s->avctx, (AVFrame *)&h->picture);
1188 init_pic(h);
1189 h->picture.poc = get_bits(&s->gb,8)*2;
1190
1191 /* get temporal distances and MV scaling factors */
1192 if(h->pic_type != FF_B_TYPE) {
1193 h->dist[0] = (h->picture.poc - h->DPB[0].poc + 512) % 512;
1194 } else {
1195 h->dist[0] = (h->DPB[0].poc - h->picture.poc + 512) % 512;
1196 }
1197 h->dist[1] = (h->picture.poc - h->DPB[1].poc + 512) % 512;
1198 h->scale_den[0] = h->dist[0] ? 512/h->dist[0] : 0;
1199 h->scale_den[1] = h->dist[1] ? 512/h->dist[1] : 0;
1200 if(h->pic_type == FF_B_TYPE) {
1201 h->sym_factor = h->dist[0]*h->scale_den[1];
1202 } else {
1203 h->direct_den[0] = h->dist[0] ? 16384/h->dist[0] : 0;
1204 h->direct_den[1] = h->dist[1] ? 16384/h->dist[1] : 0;
1205 }
1206
1207 if(s->low_delay)
1208 get_ue_golomb(&s->gb); //bbv_check_times
1209 h->progressive = get_bits1(&s->gb);
1210 if(h->progressive)
1211 h->pic_structure = 1;
1212 else if(!(h->pic_structure = get_bits1(&s->gb) && (h->stc == PIC_PB_START_CODE)) )
1213 get_bits1(&s->gb); //advanced_pred_mode_disable
1214 skip_bits1(&s->gb); //top_field_first
1215 skip_bits1(&s->gb); //repeat_first_field
1216 h->qp_fixed = get_bits1(&s->gb);
1217 h->qp = get_bits(&s->gb,6);
1218 if(h->pic_type == FF_I_TYPE) {
1219 if(!h->progressive && !h->pic_structure)
1220 skip_bits1(&s->gb);//what is this?
1221 skip_bits(&s->gb,4); //reserved bits
1222 } else {
1223 if(!(h->pic_type == FF_B_TYPE && h->pic_structure == 1))
1224 h->ref_flag = get_bits1(&s->gb);
1225 skip_bits(&s->gb,4); //reserved bits
1226 h->skip_mode_flag = get_bits1(&s->gb);
1227 }
1228 h->loop_filter_disable = get_bits1(&s->gb);
1229 if(!h->loop_filter_disable && get_bits1(&s->gb)) {
1230 h->alpha_offset = get_se_golomb(&s->gb);
1231 h->beta_offset = get_se_golomb(&s->gb);
1232 } else {
1233 h->alpha_offset = h->beta_offset = 0;
1234 }
1235 check_for_slice(h);
1236 if(h->pic_type == FF_I_TYPE) {
1237 do {
1238 decode_mb_i(h, 0);
1239 } while(next_mb(h));
1240 } else if(h->pic_type == FF_P_TYPE) {
1241 do {
1242 if(h->skip_mode_flag) {
1243 skip_count = get_ue_golomb(&s->gb);
1244 while(skip_count--) {
1245 decode_mb_p(h,P_SKIP);
1246 if(!next_mb(h))
1247 goto done;
1248 }
1249 mb_type = get_ue_golomb(&s->gb) + P_16X16;
1250 } else
1251 mb_type = get_ue_golomb(&s->gb) + P_SKIP;
1252 if(mb_type > P_8X8) {
1253 decode_mb_i(h, mb_type - P_8X8 - 1);
1254 } else
1255 decode_mb_p(h,mb_type);
1256 } while(next_mb(h));
1257 } else { /* FF_B_TYPE */
1258 do {
1259 if(h->skip_mode_flag) {
1260 skip_count = get_ue_golomb(&s->gb);
1261 while(skip_count--) {
1262 decode_mb_b(h,B_SKIP);
1263 if(!next_mb(h))
1264 goto done;
1265 }
1266 mb_type = get_ue_golomb(&s->gb) + B_DIRECT;
1267 } else
1268 mb_type = get_ue_golomb(&s->gb) + B_SKIP;
1269 if(mb_type > B_8X8) {
1270 decode_mb_i(h, mb_type - B_8X8 - 1);
1271 } else
1272 decode_mb_b(h,mb_type);
1273 } while(next_mb(h));
1274 }
1275 done:
1276 if(h->pic_type != FF_B_TYPE) {
1277 if(h->DPB[1].data[0])
1278 s->avctx->release_buffer(s->avctx, (AVFrame *)&h->DPB[1]);
1279 memcpy(&h->DPB[1], &h->DPB[0], sizeof(Picture));
1280 memcpy(&h->DPB[0], &h->picture, sizeof(Picture));
1281 memset(&h->picture,0,sizeof(Picture));
1282 }
1283 return 0;
1284 }
1285
1286 /*****************************************************************************
1287 *
1288 * headers and interface
1289 *
1290 ****************************************************************************/
1291
1292 /**
1293 * some predictions require data from the top-neighbouring macroblock.
1294 * this data has to be stored for one complete row of macroblocks
1295 * and this storage space is allocated here
1296 */
1297 static void init_top_lines(AVSContext *h) {
1298 /* alloc top line of predictors */
1299 h->top_qp = av_malloc( h->mb_width);
1300 h->top_mv[0] = av_malloc((h->mb_width*2+1)*sizeof(vector_t));
1301 h->top_mv[1] = av_malloc((h->mb_width*2+1)*sizeof(vector_t));
1302 h->top_pred_Y = av_malloc( h->mb_width*2*sizeof(*h->top_pred_Y));
1303 h->top_border_y = av_malloc((h->mb_width+1)*16);
1304 h->top_border_u = av_malloc((h->mb_width)*10);
1305 h->top_border_v = av_malloc((h->mb_width)*10);
1306
1307 /* alloc space for co-located MVs and types */
1308 h->col_mv = av_malloc( h->mb_width*h->mb_height*4*sizeof(vector_t));
1309 h->col_type_base = av_malloc(h->mb_width*h->mb_height);
1310 }
1311
1312 static int decode_seq_header(AVSContext *h) {
1313 MpegEncContext *s = &h->s;
1314 extern const AVRational ff_frame_rate_tab[];
1315 int frame_rate_code;
1316
1317 h->profile = get_bits(&s->gb,8);
1318 h->level = get_bits(&s->gb,8);
1319 skip_bits1(&s->gb); //progressive sequence
1320 s->width = get_bits(&s->gb,14);
1321 s->height = get_bits(&s->gb,14);
1322 skip_bits(&s->gb,2); //chroma format
1323 skip_bits(&s->gb,3); //sample_precision
1324 h->aspect_ratio = get_bits(&s->gb,4);
1325 frame_rate_code = get_bits(&s->gb,4);
1326 skip_bits(&s->gb,18);//bit_rate_lower
1327 skip_bits1(&s->gb); //marker_bit
1328 skip_bits(&s->gb,12);//bit_rate_upper
1329 s->low_delay = get_bits1(&s->gb);
1330 h->mb_width = (s->width + 15) >> 4;
1331 h->mb_height = (s->height + 15) >> 4;
1332 h->s.avctx->time_base.den = ff_frame_rate_tab[frame_rate_code].num;
1333 h->s.avctx->time_base.num = ff_frame_rate_tab[frame_rate_code].den;
1334 h->s.avctx->width = s->width;
1335 h->s.avctx->height = s->height;
1336 if(!h->top_qp)
1337 init_top_lines(h);
1338 return 0;
1339 }
1340
1341 /**
1342 * finds the end of the current frame in the bitstream.
1343 * @return the position of the first byte of the next frame, or -1
1344 */
1345 int ff_cavs_find_frame_end(ParseContext *pc, const uint8_t *buf, int buf_size) {
1346 int pic_found, i;
1347 uint32_t state;
1348
1349 pic_found= pc->frame_start_found;
1350 state= pc->state;
1351
1352 i=0;
1353 if(!pic_found){
1354 for(i=0; i<buf_size; i++){
1355 state= (state<<8) | buf[i];
1356 if(state == PIC_I_START_CODE || state == PIC_PB_START_CODE){
1357 i++;
1358 pic_found=1;
1359 break;
1360 }
1361 }
1362 }
1363
1364 if(pic_found){
1365 /* EOF considered as end of frame */
1366 if (buf_size == 0)
1367 return 0;
1368 for(; i<buf_size; i++){
1369 state= (state<<8) | buf[i];
1370 if((state&0xFFFFFF00) == 0x100){
1371 if(state < SLICE_MIN_START_CODE || state > SLICE_MAX_START_CODE){
1372 pc->frame_start_found=0;
1373 pc->state=-1;
1374 return i-3;
1375 }
1376 }
1377 }
1378 }
1379 pc->frame_start_found= pic_found;
1380 pc->state= state;
1381 return END_NOT_FOUND;
1382 }
1383
1384 void ff_cavs_flush(AVCodecContext * avctx) {
1385 AVSContext *h = avctx->priv_data;
1386 h->got_keyframe = 0;
1387 }
1388
1389 static int cavs_decode_frame(AVCodecContext * avctx,void *data, int *data_size,
1390 uint8_t * buf, int buf_size) {
1391 AVSContext *h = avctx->priv_data;
1392 MpegEncContext *s = &h->s;
1393 int input_size;
1394 const uint8_t *buf_end;
1395 const uint8_t *buf_ptr;
1396 AVFrame *picture = data;
1397 uint32_t stc;
1398
1399 s->avctx = avctx;
1400
1401 if (buf_size == 0) {
1402 if(!s->low_delay && h->DPB[0].data[0]) {
1403 *data_size = sizeof(AVPicture);
1404 *picture = *(AVFrame *) &h->DPB[0];
1405 }
1406 return 0;
1407 }
1408
1409 buf_ptr = buf;
1410 buf_end = buf + buf_size;
1411 for(;;) {
1412 buf_ptr = ff_find_start_code(buf_ptr,buf_end, &stc);
1413 if(stc & 0xFFFFFE00)
1414 return FFMAX(0, buf_ptr - buf - s->parse_context.last_index);
1415 input_size = (buf_end - buf_ptr)*8;
1416 switch(stc) {
1417 case SEQ_START_CODE:
1418 init_get_bits(&s->gb, buf_ptr, input_size);
1419 decode_seq_header(h);
1420 break;
1421 case PIC_I_START_CODE:
1422 if(!h->got_keyframe) {
1423 if(h->DPB[0].data[0])
1424 avctx->release_buffer(avctx, (AVFrame *)&h->DPB[0]);
1425 if(h->DPB[1].data[0])
1426 avctx->release_buffer(avctx, (AVFrame *)&h->DPB[1]);
1427 h->got_keyframe = 1;
1428 }
1429 case PIC_PB_START_CODE:
1430 *data_size = 0;
1431 if(!h->got_keyframe)
1432 break;
1433 init_get_bits(&s->gb, buf_ptr, input_size);
1434 h->stc = stc;
1435 if(decode_pic(h))
1436 break;
1437 *data_size = sizeof(AVPicture);
1438 if(h->pic_type != FF_B_TYPE) {
1439 if(h->DPB[1].data[0]) {
1440 *picture = *(AVFrame *) &h->DPB[1];
1441 } else {
1442 *data_size = 0;
1443 }
1444 } else
1445 *picture = *(AVFrame *) &h->picture;
1446 break;
1447 case EXT_START_CODE:
1448 //mpeg_decode_extension(avctx,buf_ptr, input_size);
1449 break;
1450 case USER_START_CODE:
1451 //mpeg_decode_user_data(avctx,buf_ptr, input_size);
1452 break;
1453 default:
1454 if (stc >= SLICE_MIN_START_CODE &&
1455 stc <= SLICE_MAX_START_CODE) {
1456 init_get_bits(&s->gb, buf_ptr, input_size);
1457 decode_slice_header(h, &s->gb);
1458 }
1459 break;
1460 }
1461 }
1462 }
1463
1464 static int cavs_decode_init(AVCodecContext * avctx) {
1465 AVSContext *h = avctx->priv_data;
1466 MpegEncContext * const s = &h->s;
1467
1468 MPV_decode_defaults(s);
1469 s->avctx = avctx;
1470
1471 avctx->pix_fmt= PIX_FMT_YUV420P;
1472
1473 h->luma_scan[0] = 0;
1474 h->luma_scan[1] = 8;
1475 h->intra_pred_l[ INTRA_L_VERT] = intra_pred_vert;
1476 h->intra_pred_l[ INTRA_L_HORIZ] = intra_pred_horiz;
1477 h->intra_pred_l[ INTRA_L_LP] = intra_pred_lp;
1478 h->intra_pred_l[ INTRA_L_DOWN_LEFT] = intra_pred_down_left;
1479 h->intra_pred_l[INTRA_L_DOWN_RIGHT] = intra_pred_down_right;
1480 h->intra_pred_l[ INTRA_L_LP_LEFT] = intra_pred_lp_left;
1481 h->intra_pred_l[ INTRA_L_LP_TOP] = intra_pred_lp_top;
1482 h->intra_pred_l[ INTRA_L_DC_128] = intra_pred_dc_128;
1483 h->intra_pred_c[ INTRA_C_LP] = intra_pred_lp;
1484 h->intra_pred_c[ INTRA_C_HORIZ] = intra_pred_horiz;
1485 h->intra_pred_c[ INTRA_C_VERT] = intra_pred_vert;
1486 h->intra_pred_c[ INTRA_C_PLANE] = intra_pred_plane;
1487 h->intra_pred_c[ INTRA_C_LP_LEFT] = intra_pred_lp_left;
1488 h->intra_pred_c[ INTRA_C_LP_TOP] = intra_pred_lp_top;
1489 h->intra_pred_c[ INTRA_C_DC_128] = intra_pred_dc_128;
1490 h->mv[ 7] = un_mv;
1491 h->mv[19] = un_mv;
1492 return 0;
1493 }
1494
1495 static int cavs_decode_end(AVCodecContext * avctx) {
1496 AVSContext *h = avctx->priv_data;
1497
1498 av_free(h->top_qp);
1499 av_free(h->top_mv[0]);
1500 av_free(h->top_mv[1]);
1501 av_free(h->top_pred_Y);
1502 av_free(h->top_border_y);
1503 av_free(h->top_border_u);
1504 av_free(h->top_border_v);
1505 av_free(h->col_mv);
1506 av_free(h->col_type_base);
1507 return 0;
1508 }
1509
1510 AVCodec cavs_decoder = {
1511 "cavs",
1512 CODEC_TYPE_VIDEO,
1513 CODEC_ID_CAVS,
1514 sizeof(AVSContext),
1515 cavs_decode_init,
1516 NULL,
1517 cavs_decode_end,
1518 cavs_decode_frame,
1519 CODEC_CAP_DR1 | CODEC_CAP_DELAY,
1520 .flush= ff_cavs_flush,
1521 };