make use of 16x16 motion comp functions, 16x8 and 8x16 do not seem to be worth it
[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_p(vector_t *mvP, vector_t *mvQ) {
108 if((mvP->ref == REF_INTRA) || (mvQ->ref == REF_INTRA))
109 return 2;
110 if(mvP->ref != mvQ->ref)
111 return 1;
112 if( (abs(mvP->x - mvQ->x) >= 4) || (abs(mvP->y - mvQ->y) >= 4) )
113 return 1;
114 return 0;
115 }
116
117 static inline int get_bs_b(vector_t *mvP, vector_t *mvQ) {
118 if((mvP->ref == REF_INTRA) || (mvQ->ref == REF_INTRA)) {
119 return 2;
120 } else {
121 vector_t *mvPbw = mvP + MV_BWD_OFFS;
122 vector_t *mvQbw = mvQ + MV_BWD_OFFS;
123 if( (abs( mvP->x - mvQ->x) >= 4) ||
124 (abs( mvP->y - mvQ->y) >= 4) ||
125 (abs(mvPbw->x - mvQbw->x) >= 4) ||
126 (abs(mvPbw->y - mvQbw->y) >= 4) )
127 return 1;
128 }
129 return 0;
130 }
131
132 #define SET_PARAMS \
133 alpha = alpha_tab[clip(qp_avg + h->alpha_offset,0,63)]; \
134 beta = beta_tab[clip(qp_avg + h->beta_offset, 0,63)]; \
135 tc = tc_tab[clip(qp_avg + h->alpha_offset,0,63)];
136
137 /**
138 * in-loop deblocking filter for a single macroblock
139 *
140 * boundary strength (bs) mapping:
141 *
142 * --4---5--
143 * 0 2 |
144 * | 6 | 7 |
145 * 1 3 |
146 * ---------
147 *
148 */
149 static void filter_mb(AVSContext *h, enum mb_t mb_type) {
150 DECLARE_ALIGNED_8(uint8_t, bs[8]);
151 int qp_avg, alpha, beta, tc;
152 int i;
153
154 /* save un-deblocked lines */
155 h->topleft_border_y = h->top_border_y[h->mbx*16+15];
156 h->topleft_border_u = h->top_border_u[h->mbx*10+8];
157 h->topleft_border_v = h->top_border_v[h->mbx*10+8];
158 memcpy(&h->top_border_y[h->mbx*16], h->cy + 15* h->l_stride,16);
159 memcpy(&h->top_border_u[h->mbx*10+1], h->cu + 7* h->c_stride,8);
160 memcpy(&h->top_border_v[h->mbx*10+1], h->cv + 7* h->c_stride,8);
161 for(i=0;i<8;i++) {
162 h->left_border_y[i*2+0] = *(h->cy + 15 + (i*2+0)*h->l_stride);
163 h->left_border_y[i*2+1] = *(h->cy + 15 + (i*2+1)*h->l_stride);
164 h->left_border_u[i+1] = *(h->cu + 7 + i*h->c_stride);
165 h->left_border_v[i+1] = *(h->cv + 7 + i*h->c_stride);
166 }
167 if(!h->loop_filter_disable) {
168 /* clear bs */
169 *((uint64_t *)bs) = 0;
170 /* determine bs */
171 switch(mb_type) {
172 case I_8X8:
173 *((uint64_t *)bs) = 0x0202020202020202ULL;
174 break;
175 case P_8X8:
176 case P_8X16:
177 bs[2] = get_bs_p(&h->mv[MV_FWD_X0], &h->mv[MV_FWD_X1]);
178 bs[3] = get_bs_p(&h->mv[MV_FWD_X2], &h->mv[MV_FWD_X3]);
179 case P_16X8:
180 bs[6] = get_bs_p(&h->mv[MV_FWD_X0], &h->mv[MV_FWD_X2]);
181 bs[7] = get_bs_p(&h->mv[MV_FWD_X1], &h->mv[MV_FWD_X3]);
182 case P_16X16:
183 case P_SKIP:
184 bs[0] = get_bs_p(&h->mv[MV_FWD_A1], &h->mv[MV_FWD_X0]);
185 bs[1] = get_bs_p(&h->mv[MV_FWD_A3], &h->mv[MV_FWD_X2]);
186 bs[4] = get_bs_p(&h->mv[MV_FWD_B2], &h->mv[MV_FWD_X0]);
187 bs[5] = get_bs_p(&h->mv[MV_FWD_B3], &h->mv[MV_FWD_X1]);
188 break;
189 case B_SKIP:
190 case B_DIRECT:
191 case B_8X8:
192 bs[2] = get_bs_b(&h->mv[MV_FWD_X0], &h->mv[MV_FWD_X1]);
193 bs[3] = get_bs_b(&h->mv[MV_FWD_X2], &h->mv[MV_FWD_X3]);
194 bs[6] = get_bs_b(&h->mv[MV_FWD_X0], &h->mv[MV_FWD_X2]);
195 bs[7] = get_bs_b(&h->mv[MV_FWD_X1], &h->mv[MV_FWD_X3]);
196 case B_FWD_16X16:
197 case B_BWD_16X16:
198 case B_SYM_16X16:
199 bs[0] = get_bs_b(&h->mv[MV_FWD_A1], &h->mv[MV_FWD_X0]);
200 bs[1] = get_bs_b(&h->mv[MV_FWD_A3], &h->mv[MV_FWD_X2]);
201 bs[4] = get_bs_b(&h->mv[MV_FWD_B2], &h->mv[MV_FWD_X0]);
202 bs[5] = get_bs_b(&h->mv[MV_FWD_B3], &h->mv[MV_FWD_X1]);
203 break;
204 default:
205 if(mb_type & 1) { //16X8
206 bs[6] = bs[7] = get_bs_b(&h->mv[MV_FWD_X0], &h->mv[MV_FWD_X2]);
207 } else { //8X16
208 bs[2] = bs[3] = get_bs_b(&h->mv[MV_FWD_X0], &h->mv[MV_FWD_X1]);
209 }
210 bs[0] = get_bs_b(&h->mv[MV_FWD_A1], &h->mv[MV_FWD_X0]);
211 bs[1] = get_bs_b(&h->mv[MV_FWD_A3], &h->mv[MV_FWD_X2]);
212 bs[4] = get_bs_b(&h->mv[MV_FWD_B2], &h->mv[MV_FWD_X0]);
213 bs[5] = get_bs_b(&h->mv[MV_FWD_B3], &h->mv[MV_FWD_X1]);
214 }
215 if( *((uint64_t *)bs) ) {
216 if(h->flags & A_AVAIL) {
217 qp_avg = (h->qp + h->left_qp + 1) >> 1;
218 SET_PARAMS;
219 h->s.dsp.cavs_filter_lv(h->cy,h->l_stride,alpha,beta,tc,bs[0],bs[1]);
220 h->s.dsp.cavs_filter_cv(h->cu,h->c_stride,alpha,beta,tc,bs[0],bs[1]);
221 h->s.dsp.cavs_filter_cv(h->cv,h->c_stride,alpha,beta,tc,bs[0],bs[1]);
222 }
223 qp_avg = h->qp;
224 SET_PARAMS;
225 h->s.dsp.cavs_filter_lv(h->cy + 8,h->l_stride,alpha,beta,tc,bs[2],bs[3]);
226 h->s.dsp.cavs_filter_lh(h->cy + 8*h->l_stride,h->l_stride,alpha,beta,tc,
227 bs[6],bs[7]);
228
229 if(h->flags & B_AVAIL) {
230 qp_avg = (h->qp + h->top_qp[h->mbx] + 1) >> 1;
231 SET_PARAMS;
232 h->s.dsp.cavs_filter_lh(h->cy,h->l_stride,alpha,beta,tc,bs[4],bs[5]);
233 h->s.dsp.cavs_filter_ch(h->cu,h->c_stride,alpha,beta,tc,bs[4],bs[5]);
234 h->s.dsp.cavs_filter_ch(h->cv,h->c_stride,alpha,beta,tc,bs[4],bs[5]);
235 }
236 }
237 }
238 h->left_qp = h->qp;
239 h->top_qp[h->mbx] = h->qp;
240 }
241
242 #undef SET_PARAMS
243
244 /*****************************************************************************
245 *
246 * spatial intra prediction
247 *
248 ****************************************************************************/
249
250 static inline void load_intra_pred_luma(AVSContext *h, uint8_t *top,
251 uint8_t *left, int block) {
252 int i;
253
254 switch(block) {
255 case 0:
256 memcpy(&left[1],h->left_border_y,16);
257 left[0] = left[1];
258 left[17] = left[16];
259 memcpy(&top[1],&h->top_border_y[h->mbx*16],16);
260 top[17] = top[16];
261 top[0] = top[1];
262 if((h->flags & A_AVAIL) && (h->flags & B_AVAIL))
263 left[0] = top[0] = h->topleft_border_y;
264 break;
265 case 1:
266 for(i=0;i<8;i++)
267 left[i+1] = *(h->cy + 7 + i*h->l_stride);
268 memset(&left[9],left[8],9);
269 left[0] = left[1];
270 memcpy(&top[1],&h->top_border_y[h->mbx*16+8],8);
271 if(h->flags & C_AVAIL)
272 memcpy(&top[9],&h->top_border_y[(h->mbx + 1)*16],8);
273 else
274 memset(&top[9],top[8],9);
275 top[17] = top[16];
276 top[0] = top[1];
277 if(h->flags & B_AVAIL)
278 left[0] = top[0] = h->top_border_y[h->mbx*16+7];
279 break;
280 case 2:
281 memcpy(&left[1],&h->left_border_y[8],8);
282 memset(&left[9],left[8],9);
283 memcpy(&top[1],h->cy + 7*h->l_stride,16);
284 top[17] = top[16];
285 left[0] = h->left_border_y[7];
286 top[0] = top[1];
287 if(h->flags & A_AVAIL)
288 top[0] = left[0];
289 break;
290 case 3:
291 for(i=0;i<9;i++)
292 left[i] = *(h->cy + 7 + (i+7)*h->l_stride);
293 memset(&left[9],left[8],9);
294 memcpy(&top[0],h->cy + 7 + 7*h->l_stride,9);
295 memset(&top[9],top[8],9);
296 break;
297 }
298 }
299
300 static void intra_pred_vert(uint8_t *d,uint8_t *top,uint8_t *left,int stride) {
301 int y;
302 uint64_t a = *((uint64_t *)(&top[1]));
303 for(y=0;y<8;y++) {
304 *((uint64_t *)(d+y*stride)) = a;
305 }
306 }
307
308 static void intra_pred_horiz(uint8_t *d,uint8_t *top,uint8_t *left,int stride) {
309 int y;
310 uint64_t a;
311 for(y=0;y<8;y++) {
312 a = left[y+1] * 0x0101010101010101ULL;
313 *((uint64_t *)(d+y*stride)) = a;
314 }
315 }
316
317 static void intra_pred_dc_128(uint8_t *d,uint8_t *top,uint8_t *left,int stride) {
318 int y;
319 uint64_t a = 0x8080808080808080ULL;
320 for(y=0;y<8;y++)
321 *((uint64_t *)(d+y*stride)) = a;
322 }
323
324 static void intra_pred_plane(uint8_t *d,uint8_t *top,uint8_t *left,int stride) {
325 int x,y,ia;
326 int ih = 0;
327 int iv = 0;
328 uint8_t *cm = cropTbl + MAX_NEG_CROP;
329
330 for(x=0; x<4; x++) {
331 ih += (x+1)*(top[5+x]-top[3-x]);
332 iv += (x+1)*(left[5+x]-left[3-x]);
333 }
334 ia = (top[8]+left[8])<<4;
335 ih = (17*ih+16)>>5;
336 iv = (17*iv+16)>>5;
337 for(y=0; y<8; y++)
338 for(x=0; x<8; x++)
339 d[y*stride+x] = cm[(ia+(x-3)*ih+(y-3)*iv+16)>>5];
340 }
341
342 #define LOWPASS(ARRAY,INDEX) \
343 (( ARRAY[(INDEX)-1] + 2*ARRAY[(INDEX)] + ARRAY[(INDEX)+1] + 2) >> 2)
344
345 static void intra_pred_lp(uint8_t *d,uint8_t *top,uint8_t *left,int stride) {
346 int x,y;
347 for(y=0; y<8; y++)
348 for(x=0; x<8; x++)
349 d[y*stride+x] = (LOWPASS(top,x+1) + LOWPASS(left,y+1)) >> 1;
350 }
351
352 static void intra_pred_down_left(uint8_t *d,uint8_t *top,uint8_t *left,int stride) {
353 int x,y;
354 for(y=0; y<8; y++)
355 for(x=0; x<8; x++)
356 d[y*stride+x] = (LOWPASS(top,x+y+2) + LOWPASS(left,x+y+2)) >> 1;
357 }
358
359 static void intra_pred_down_right(uint8_t *d,uint8_t *top,uint8_t *left,int stride) {
360 int x,y;
361 for(y=0; y<8; y++)
362 for(x=0; x<8; x++)
363 if(x==y)
364 d[y*stride+x] = (left[1]+2*top[0]+top[1]+2)>>2;
365 else if(x>y)
366 d[y*stride+x] = LOWPASS(top,x-y);
367 else
368 d[y*stride+x] = LOWPASS(left,y-x);
369 }
370
371 static void intra_pred_lp_left(uint8_t *d,uint8_t *top,uint8_t *left,int stride) {
372 int x,y;
373 for(y=0; y<8; y++)
374 for(x=0; x<8; x++)
375 d[y*stride+x] = LOWPASS(left,y+1);
376 }
377
378 static void intra_pred_lp_top(uint8_t *d,uint8_t *top,uint8_t *left,int stride) {
379 int x,y;
380 for(y=0; y<8; y++)
381 for(x=0; x<8; x++)
382 d[y*stride+x] = LOWPASS(top,x+1);
383 }
384
385 #undef LOWPASS
386
387 static inline void modify_pred(const int_fast8_t *mod_table, int *mode) {
388 *mode = mod_table[*mode];
389 if(*mode < 0) {
390 av_log(NULL, AV_LOG_ERROR, "Illegal intra prediction mode\n");
391 *mode = 0;
392 }
393 }
394
395 /*****************************************************************************
396 *
397 * motion compensation
398 *
399 ****************************************************************************/
400
401 static inline void mc_dir_part(AVSContext *h,Picture *pic,int square,
402 int chroma_height,int delta,int list,uint8_t *dest_y,
403 uint8_t *dest_cb,uint8_t *dest_cr,int src_x_offset,
404 int src_y_offset,qpel_mc_func *qpix_op,
405 h264_chroma_mc_func chroma_op,vector_t *mv){
406 MpegEncContext * const s = &h->s;
407 const int mx= mv->x + src_x_offset*8;
408 const int my= mv->y + src_y_offset*8;
409 const int luma_xy= (mx&3) + ((my&3)<<2);
410 uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*h->l_stride;
411 uint8_t * src_cb= pic->data[1] + (mx>>3) + (my>>3)*h->c_stride;
412 uint8_t * src_cr= pic->data[2] + (mx>>3) + (my>>3)*h->c_stride;
413 int extra_width= 0; //(s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16;
414 int extra_height= extra_width;
415 int emu=0;
416 const int full_mx= mx>>2;
417 const int full_my= my>>2;
418 const int pic_width = 16*h->mb_width;
419 const int pic_height = 16*h->mb_height;
420
421 if(!pic->data[0])
422 return;
423 if(mx&7) extra_width -= 3;
424 if(my&7) extra_height -= 3;
425
426 if( full_mx < 0-extra_width
427 || full_my < 0-extra_height
428 || full_mx + 16/*FIXME*/ > pic_width + extra_width
429 || full_my + 16/*FIXME*/ > pic_height + extra_height){
430 ff_emulated_edge_mc(s->edge_emu_buffer, src_y - 2 - 2*h->l_stride, h->l_stride,
431 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
432 src_y= s->edge_emu_buffer + 2 + 2*h->l_stride;
433 emu=1;
434 }
435
436 qpix_op[luma_xy](dest_y, src_y, h->l_stride); //FIXME try variable height perhaps?
437 if(!square){
438 qpix_op[luma_xy](dest_y + delta, src_y + delta, h->l_stride);
439 }
440
441 if(emu){
442 ff_emulated_edge_mc(s->edge_emu_buffer, src_cb, h->c_stride,
443 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
444 src_cb= s->edge_emu_buffer;
445 }
446 chroma_op(dest_cb, src_cb, h->c_stride, chroma_height, mx&7, my&7);
447
448 if(emu){
449 ff_emulated_edge_mc(s->edge_emu_buffer, src_cr, h->c_stride,
450 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
451 src_cr= s->edge_emu_buffer;
452 }
453 chroma_op(dest_cr, src_cr, h->c_stride, chroma_height, mx&7, my&7);
454 }
455
456 static inline void mc_part_std(AVSContext *h,int square,int chroma_height,int delta,
457 uint8_t *dest_y,uint8_t *dest_cb,uint8_t *dest_cr,
458 int x_offset, int y_offset,qpel_mc_func *qpix_put,
459 h264_chroma_mc_func chroma_put,qpel_mc_func *qpix_avg,
460 h264_chroma_mc_func chroma_avg, vector_t *mv){
461 qpel_mc_func *qpix_op= qpix_put;
462 h264_chroma_mc_func chroma_op= chroma_put;
463
464 dest_y += 2*x_offset + 2*y_offset*h->l_stride;
465 dest_cb += x_offset + y_offset*h->c_stride;
466 dest_cr += x_offset + y_offset*h->c_stride;
467 x_offset += 8*h->mbx;
468 y_offset += 8*h->mby;
469
470 if(mv->ref >= 0){
471 Picture *ref= &h->DPB[mv->ref];
472 mc_dir_part(h, ref, square, chroma_height, delta, 0,
473 dest_y, dest_cb, dest_cr, x_offset, y_offset,
474 qpix_op, chroma_op, mv);
475
476 qpix_op= qpix_avg;
477 chroma_op= chroma_avg;
478 }
479
480 if((mv+MV_BWD_OFFS)->ref >= 0){
481 Picture *ref= &h->DPB[0];
482 mc_dir_part(h, ref, square, chroma_height, delta, 1,
483 dest_y, dest_cb, dest_cr, x_offset, y_offset,
484 qpix_op, chroma_op, mv+MV_BWD_OFFS);
485 }
486 }
487
488 static void inter_pred(AVSContext *h, enum mb_t mb_type) {
489 switch(mb_type) {
490 case P_SKIP:
491 case P_16X16:
492 case B_FWD_16X16:
493 case B_BWD_16X16:
494 case B_SYM_16X16:
495 mc_part_std(h, 1, 8, 0, h->cy, h->cu, h->cv, 0, 0,
496 h->s.dsp.put_cavs_qpel_pixels_tab[0],
497 h->s.dsp.put_h264_chroma_pixels_tab[0],
498 h->s.dsp.avg_cavs_qpel_pixels_tab[0],
499 h->s.dsp.avg_h264_chroma_pixels_tab[0],&h->mv[MV_FWD_X0]);
500 break;
501 default:
502 mc_part_std(h, 1, 4, 0, h->cy, h->cu, h->cv, 0, 0,
503 h->s.dsp.put_cavs_qpel_pixels_tab[1],
504 h->s.dsp.put_h264_chroma_pixels_tab[1],
505 h->s.dsp.avg_cavs_qpel_pixels_tab[1],
506 h->s.dsp.avg_h264_chroma_pixels_tab[1],&h->mv[MV_FWD_X0]);
507 mc_part_std(h, 1, 4, 0, h->cy, h->cu, h->cv, 4, 0,
508 h->s.dsp.put_cavs_qpel_pixels_tab[1],
509 h->s.dsp.put_h264_chroma_pixels_tab[1],
510 h->s.dsp.avg_cavs_qpel_pixels_tab[1],
511 h->s.dsp.avg_h264_chroma_pixels_tab[1],&h->mv[MV_FWD_X1]);
512 mc_part_std(h, 1, 4, 0, h->cy, h->cu, h->cv, 0, 4,
513 h->s.dsp.put_cavs_qpel_pixels_tab[1],
514 h->s.dsp.put_h264_chroma_pixels_tab[1],
515 h->s.dsp.avg_cavs_qpel_pixels_tab[1],
516 h->s.dsp.avg_h264_chroma_pixels_tab[1],&h->mv[MV_FWD_X2]);
517 mc_part_std(h, 1, 4, 0, h->cy, h->cu, h->cv, 4, 4,
518 h->s.dsp.put_cavs_qpel_pixels_tab[1],
519 h->s.dsp.put_h264_chroma_pixels_tab[1],
520 h->s.dsp.avg_cavs_qpel_pixels_tab[1],
521 h->s.dsp.avg_h264_chroma_pixels_tab[1],&h->mv[MV_FWD_X3]);
522 break;
523 }
524 /* set intra prediction modes to default values */
525 h->pred_mode_Y[3] = h->pred_mode_Y[6] = INTRA_L_LP;
526 h->top_pred_Y[h->mbx*2+0] = h->top_pred_Y[h->mbx*2+1] = INTRA_L_LP;
527 }
528
529 /*****************************************************************************
530 *
531 * motion vector prediction
532 *
533 ****************************************************************************/
534
535 static inline void set_mvs(vector_t *mv, enum block_t size) {
536 switch(size) {
537 case BLK_16X16:
538 mv[MV_STRIDE ] = mv[0];
539 mv[MV_STRIDE+1] = mv[0];
540 case BLK_16X8:
541 mv[1] = mv[0];
542 break;
543 case BLK_8X16:
544 mv[MV_STRIDE] = mv[0];
545 break;
546 }
547 }
548
549 static inline void store_mvs(AVSContext *h) {
550 h->col_mv[(h->mby*h->mb_width + h->mbx)*4 + 0] = h->mv[MV_FWD_X0];
551 h->col_mv[(h->mby*h->mb_width + h->mbx)*4 + 1] = h->mv[MV_FWD_X1];
552 h->col_mv[(h->mby*h->mb_width + h->mbx)*4 + 2] = h->mv[MV_FWD_X2];
553 h->col_mv[(h->mby*h->mb_width + h->mbx)*4 + 3] = h->mv[MV_FWD_X3];
554 }
555
556 static inline void scale_mv(AVSContext *h, int *d_x, int *d_y, vector_t *src, int distp) {
557 int den = h->scale_den[src->ref];
558
559 *d_x = (src->x*distp*den + 256 + (src->x>>31)) >> 9;
560 *d_y = (src->y*distp*den + 256 + (src->y>>31)) >> 9;
561 }
562
563 static inline void mv_pred_median(AVSContext *h, vector_t *mvP, vector_t *mvA, vector_t *mvB, vector_t *mvC) {
564 int ax, ay, bx, by, cx, cy;
565 int len_ab, len_bc, len_ca, len_mid;
566
567 /* scale candidates according to their temporal span */
568 scale_mv(h, &ax, &ay, mvA, mvP->dist);
569 scale_mv(h, &bx, &by, mvB, mvP->dist);
570 scale_mv(h, &cx, &cy, mvC, mvP->dist);
571 /* find the geometrical median of the three candidates */
572 len_ab = abs(ax - bx) + abs(ay - by);
573 len_bc = abs(bx - cx) + abs(by - cy);
574 len_ca = abs(cx - ax) + abs(cy - ay);
575 len_mid = mid_pred(len_ab, len_bc, len_ca);
576 if(len_mid == len_ab) {
577 mvP->x = cx;
578 mvP->y = cy;
579 } else if(len_mid == len_bc) {
580 mvP->x = ax;
581 mvP->y = ay;
582 } else {
583 mvP->x = bx;
584 mvP->y = by;
585 }
586 }
587
588 static inline void mv_pred_direct(AVSContext *h, vector_t *pmv_fw,
589 vector_t *col_mv) {
590 vector_t *pmv_bw = pmv_fw + MV_BWD_OFFS;
591 int den = h->direct_den[col_mv->ref];
592 int m = col_mv->x >> 31;
593
594 pmv_fw->dist = h->dist[1];
595 pmv_bw->dist = h->dist[0];
596 pmv_fw->ref = 1;
597 pmv_bw->ref = 0;
598 /* scale the co-located motion vector according to its temporal span */
599 pmv_fw->x = (((den+(den*col_mv->x*pmv_fw->dist^m)-m-1)>>14)^m)-m;
600 pmv_bw->x = m-(((den+(den*col_mv->x*pmv_bw->dist^m)-m-1)>>14)^m);
601 m = col_mv->y >> 31;
602 pmv_fw->y = (((den+(den*col_mv->y*pmv_fw->dist^m)-m-1)>>14)^m)-m;
603 pmv_bw->y = m-(((den+(den*col_mv->y*pmv_bw->dist^m)-m-1)>>14)^m);
604 }
605
606 static inline void mv_pred_sym(AVSContext *h, vector_t *src, enum block_t size) {
607 vector_t *dst = src + MV_BWD_OFFS;
608
609 /* backward mv is the scaled and negated forward mv */
610 dst->x = -((src->x * h->sym_factor + 256) >> 9);
611 dst->y = -((src->y * h->sym_factor + 256) >> 9);
612 dst->ref = 0;
613 dst->dist = h->dist[0];
614 set_mvs(dst, size);
615 }
616
617 static void mv_pred(AVSContext *h, enum mv_loc_t nP, enum mv_loc_t nC,
618 enum mv_pred_t mode, enum block_t size, int ref) {
619 vector_t *mvP = &h->mv[nP];
620 vector_t *mvA = &h->mv[nP-1];
621 vector_t *mvB = &h->mv[nP-4];
622 vector_t *mvC = &h->mv[nC];
623 int mvAref = mvA->ref;
624 int mvBref = mvB->ref;
625 int mvCref;
626
627 mvP->ref = ref;
628 mvP->dist = h->dist[mvP->ref];
629 if(mvC->ref == NOT_AVAIL)
630 mvC = &h->mv[nP-5]; // set to top-left (mvD)
631 mvCref = mvC->ref;
632 if(mode == MV_PRED_PSKIP) {
633 if((mvAref == NOT_AVAIL) || (mvBref == NOT_AVAIL) ||
634 ((mvA->x | mvA->y | mvA->ref) == 0) ||
635 ((mvB->x | mvB->y | mvB->ref) == 0) ) {
636 mvP->x = mvP->y = 0;
637 set_mvs(mvP,size);
638 return;
639 }
640 }
641 /* if there is only one suitable candidate, take it */
642 if((mvAref >= 0) && (mvBref < 0) && (mvCref < 0)) {
643 mvP->x = mvA->x;
644 mvP->y = mvA->y;
645 } else if((mvAref < 0) && (mvBref >= 0) && (mvCref < 0)) {
646 mvP->x = mvB->x;
647 mvP->y = mvB->y;
648 } else if((mvAref < 0) && (mvBref < 0) && (mvCref >= 0)) {
649 mvP->x = mvC->x;
650 mvP->y = mvC->y;
651 } else {
652 switch(mode) {
653 case MV_PRED_LEFT:
654 if(mvAref == mvP->ref) {
655 mvP->x = mvA->x;
656 mvP->y = mvA->y;
657 } else
658 mv_pred_median(h, mvP, mvA, mvB, mvC);
659 break;
660 case MV_PRED_TOP:
661 if(mvBref == mvP->ref) {
662 mvP->x = mvB->x;
663 mvP->y = mvB->y;
664 } else
665 mv_pred_median(h, mvP, mvA, mvB, mvC);
666 break;
667 case MV_PRED_TOPRIGHT:
668 if(mvCref == mvP->ref) {
669 mvP->x = mvC->x;
670 mvP->y = mvC->y;
671 } else
672 mv_pred_median(h, mvP, mvA, mvB, mvC);
673 break;
674 default:
675 mv_pred_median(h, mvP, mvA, mvB, mvC);
676 break;
677 }
678 }
679 if(mode < MV_PRED_PSKIP) {
680 mvP->x += get_se_golomb(&h->s.gb);
681 mvP->y += get_se_golomb(&h->s.gb);
682 }
683 set_mvs(mvP,size);
684 }
685
686 /*****************************************************************************
687 *
688 * residual data decoding
689 *
690 ****************************************************************************/
691
692 /** kth-order exponential golomb code */
693 static inline int get_ue_code(GetBitContext *gb, int order) {
694 if(order) {
695 int ret = get_ue_golomb(gb) << order;
696 return ret + get_bits(gb,order);
697 }
698 return get_ue_golomb(gb);
699 }
700
701 /**
702 * decode coefficients from one 8x8 block, dequantize, inverse transform
703 * and add them to sample block
704 * @param r pointer to 2D VLC table
705 * @param esc_golomb_order escape codes are k-golomb with this order k
706 * @param qp quantizer
707 * @param dst location of sample block
708 * @param stride line stride in frame buffer
709 */
710 static int decode_residual_block(AVSContext *h, GetBitContext *gb,
711 const residual_vlc_t *r, int esc_golomb_order,
712 int qp, uint8_t *dst, int stride) {
713 int i,pos = -1;
714 int level_code, esc_code, level, run, mask;
715 int level_buf[64];
716 int run_buf[64];
717 int dqm = dequant_mul[qp];
718 int dqs = dequant_shift[qp];
719 int dqa = 1 << (dqs - 1);
720 const uint8_t *scantab = ff_zigzag_direct;
721 DCTELEM block[64];
722
723 memset(block,0,64*sizeof(DCTELEM));
724 for(i=0;i<65;i++) {
725 level_code = get_ue_code(gb,r->golomb_order);
726 if(level_code >= ESCAPE_CODE) {
727 run = (level_code - ESCAPE_CODE) >> 1;
728 esc_code = get_ue_code(gb,esc_golomb_order);
729 level = esc_code + (run > r->max_run ? 1 : r->level_add[run]);
730 while(level > r->inc_limit)
731 r++;
732 mask = -(level_code & 1);
733 level = (level^mask) - mask;
734 } else {
735 if(level_code < 0)
736 return -1;
737 level = r->rltab[level_code][0];
738 if(!level) //end of block signal
739 break;
740 run = r->rltab[level_code][1];
741 r += r->rltab[level_code][2];
742 }
743 level_buf[i] = level;
744 run_buf[i] = run;
745 }
746 /* inverse scan and dequantization */
747 while(--i >= 0){
748 pos += 1 + run_buf[i];
749 if(pos > 63) {
750 av_log(h->s.avctx, AV_LOG_ERROR,
751 "position out of block bounds at pic %d MB(%d,%d)\n",
752 h->picture.poc, h->mbx, h->mby);
753 return -1;
754 }
755 block[scantab[pos]] = (level_buf[i]*dqm + dqa) >> dqs;
756 }
757 h->s.dsp.cavs_idct8_add(dst,block,stride);
758 return 0;
759 }
760
761
762 static inline void decode_residual_chroma(AVSContext *h) {
763 if(h->cbp & (1<<4))
764 decode_residual_block(h,&h->s.gb,chroma_2dvlc,0, chroma_qp[h->qp],
765 h->cu,h->c_stride);
766 if(h->cbp & (1<<5))
767 decode_residual_block(h,&h->s.gb,chroma_2dvlc,0, chroma_qp[h->qp],
768 h->cv,h->c_stride);
769 }
770
771 static inline int decode_residual_inter(AVSContext *h) {
772 int block;
773
774 /* get coded block pattern */
775 int cbp= get_ue_golomb(&h->s.gb);
776 if(cbp > 63){
777 av_log(h->s.avctx, AV_LOG_ERROR, "illegal inter cbp\n");
778 return -1;
779 }
780 h->cbp = cbp_tab[cbp][1];
781
782 /* get quantizer */
783 if(h->cbp && !h->qp_fixed)
784 h->qp += get_se_golomb(&h->s.gb);
785 for(block=0;block<4;block++)
786 if(h->cbp & (1<<block))
787 decode_residual_block(h,&h->s.gb,inter_2dvlc,0,h->qp,
788 h->cy + h->luma_scan[block], h->l_stride);
789 decode_residual_chroma(h);
790
791 return 0;
792 }
793
794 /*****************************************************************************
795 *
796 * macroblock level
797 *
798 ****************************************************************************/
799
800 /**
801 * initialise predictors for motion vectors and intra prediction
802 */
803 static inline void init_mb(AVSContext *h) {
804 int i;
805
806 /* copy predictors from top line (MB B and C) into cache */
807 for(i=0;i<3;i++) {
808 h->mv[MV_FWD_B2+i] = h->top_mv[0][h->mbx*2+i];
809 h->mv[MV_BWD_B2+i] = h->top_mv[1][h->mbx*2+i];
810 }
811 h->pred_mode_Y[1] = h->top_pred_Y[h->mbx*2+0];
812 h->pred_mode_Y[2] = h->top_pred_Y[h->mbx*2+1];
813 /* clear top predictors if MB B is not available */
814 if(!(h->flags & B_AVAIL)) {
815 h->mv[MV_FWD_B2] = un_mv;
816 h->mv[MV_FWD_B3] = un_mv;
817 h->mv[MV_BWD_B2] = un_mv;
818 h->mv[MV_BWD_B3] = un_mv;
819 h->pred_mode_Y[1] = h->pred_mode_Y[2] = NOT_AVAIL;
820 h->flags &= ~(C_AVAIL|D_AVAIL);
821 } else if(h->mbx) {
822 h->flags |= D_AVAIL;
823 }
824 if(h->mbx == h->mb_width-1) //MB C not available
825 h->flags &= ~C_AVAIL;
826 /* clear top-right predictors if MB C is not available */
827 if(!(h->flags & C_AVAIL)) {
828 h->mv[MV_FWD_C2] = un_mv;
829 h->mv[MV_BWD_C2] = un_mv;
830 }
831 /* clear top-left predictors if MB D is not available */
832 if(!(h->flags & D_AVAIL)) {
833 h->mv[MV_FWD_D3] = un_mv;
834 h->mv[MV_BWD_D3] = un_mv;
835 }
836 /* set pointer for co-located macroblock type */
837 h->col_type = &h->col_type_base[h->mby*h->mb_width + h->mbx];
838 }
839
840 static inline void check_for_slice(AVSContext *h);
841
842 /**
843 * save predictors for later macroblocks and increase
844 * macroblock address
845 * @returns 0 if end of frame is reached, 1 otherwise
846 */
847 static inline int next_mb(AVSContext *h) {
848 int i;
849
850 h->flags |= A_AVAIL;
851 h->cy += 16;
852 h->cu += 8;
853 h->cv += 8;
854 /* copy mvs as predictors to the left */
855 for(i=0;i<=20;i+=4)
856 h->mv[i] = h->mv[i+2];
857 /* copy bottom mvs from cache to top line */
858 h->top_mv[0][h->mbx*2+0] = h->mv[MV_FWD_X2];
859 h->top_mv[0][h->mbx*2+1] = h->mv[MV_FWD_X3];
860 h->top_mv[1][h->mbx*2+0] = h->mv[MV_BWD_X2];
861 h->top_mv[1][h->mbx*2+1] = h->mv[MV_BWD_X3];
862 /* next MB address */
863 h->mbx++;
864 if(h->mbx == h->mb_width) { //new mb line
865 h->flags = B_AVAIL|C_AVAIL;
866 /* clear left pred_modes */
867 h->pred_mode_Y[3] = h->pred_mode_Y[6] = NOT_AVAIL;
868 /* clear left mv predictors */
869 for(i=0;i<=20;i+=4)
870 h->mv[i] = un_mv;
871 h->mbx = 0;
872 h->mby++;
873 /* re-calculate sample pointers */
874 h->cy = h->picture.data[0] + h->mby*16*h->l_stride;
875 h->cu = h->picture.data[1] + h->mby*8*h->c_stride;
876 h->cv = h->picture.data[2] + h->mby*8*h->c_stride;
877 if(h->mby == h->mb_height) { //frame end
878 return 0;
879 } else {
880 //check_for_slice(h);
881 }
882 }
883 return 1;
884 }
885
886 static int decode_mb_i(AVSContext *h, int cbp_code) {
887 GetBitContext *gb = &h->s.gb;
888 int block, pred_mode_uv;
889 uint8_t top[18];
890 uint8_t left[18];
891 uint8_t *d;
892
893 init_mb(h);
894
895 /* get intra prediction modes from stream */
896 for(block=0;block<4;block++) {
897 int nA,nB,predpred;
898 int pos = scan3x3[block];
899
900 nA = h->pred_mode_Y[pos-1];
901 nB = h->pred_mode_Y[pos-3];
902 predpred = FFMIN(nA,nB);
903 if(predpred == NOT_AVAIL) // if either is not available
904 predpred = INTRA_L_LP;
905 if(!get_bits1(gb)){
906 int rem_mode= get_bits(gb, 2);
907 predpred = rem_mode + (rem_mode >= predpred);
908 }
909 h->pred_mode_Y[pos] = predpred;
910 }
911 pred_mode_uv = get_ue_golomb(gb);
912 if(pred_mode_uv > 6) {
913 av_log(h->s.avctx, AV_LOG_ERROR, "illegal intra chroma pred mode\n");
914 return -1;
915 }
916
917 /* save pred modes before they get modified */
918 h->pred_mode_Y[3] = h->pred_mode_Y[5];
919 h->pred_mode_Y[6] = h->pred_mode_Y[8];
920 h->top_pred_Y[h->mbx*2+0] = h->pred_mode_Y[7];
921 h->top_pred_Y[h->mbx*2+1] = h->pred_mode_Y[8];
922
923 /* modify pred modes according to availability of neighbour samples */
924 if(!(h->flags & A_AVAIL)) {
925 modify_pred(left_modifier_l, &h->pred_mode_Y[4] );
926 modify_pred(left_modifier_l, &h->pred_mode_Y[7] );
927 modify_pred(left_modifier_c, &pred_mode_uv );
928 }
929 if(!(h->flags & B_AVAIL)) {
930 modify_pred(top_modifier_l, &h->pred_mode_Y[4] );
931 modify_pred(top_modifier_l, &h->pred_mode_Y[5] );
932 modify_pred(top_modifier_c, &pred_mode_uv );
933 }
934
935 /* get coded block pattern */
936 if(h->pic_type == FF_I_TYPE)
937 cbp_code = get_ue_golomb(gb);
938 if(cbp_code > 63){
939 av_log(h->s.avctx, AV_LOG_ERROR, "illegal intra cbp\n");
940 return -1;
941 }
942 h->cbp = cbp_tab[cbp_code][0];
943 if(h->cbp && !h->qp_fixed)
944 h->qp += get_se_golomb(gb); //qp_delta
945
946 /* luma intra prediction interleaved with residual decode/transform/add */
947 for(block=0;block<4;block++) {
948 d = h->cy + h->luma_scan[block];
949 load_intra_pred_luma(h, top, left, block);
950 h->intra_pred_l[h->pred_mode_Y[scan3x3[block]]]
951 (d, top, left, h->l_stride);
952 if(h->cbp & (1<<block))
953 decode_residual_block(h,gb,intra_2dvlc,1,h->qp,d,h->l_stride);
954 }
955
956 /* chroma intra prediction */
957 /* extend borders by one pixel */
958 h->left_border_u[9] = h->left_border_u[8];
959 h->left_border_v[9] = h->left_border_v[8];
960 h->top_border_u[h->mbx*10+9] = h->top_border_u[h->mbx*10+8];
961 h->top_border_v[h->mbx*10+9] = h->top_border_v[h->mbx*10+8];
962 if(h->mbx && h->mby) {
963 h->top_border_u[h->mbx*10] = h->left_border_u[0] = h->topleft_border_u;
964 h->top_border_v[h->mbx*10] = h->left_border_v[0] = h->topleft_border_v;
965 } else {
966 h->left_border_u[0] = h->left_border_u[1];
967 h->left_border_v[0] = h->left_border_v[1];
968 h->top_border_u[h->mbx*10] = h->top_border_u[h->mbx*10+1];
969 h->top_border_v[h->mbx*10] = h->top_border_v[h->mbx*10+1];
970 }
971 h->intra_pred_c[pred_mode_uv](h->cu, &h->top_border_u[h->mbx*10],
972 h->left_border_u, h->c_stride);
973 h->intra_pred_c[pred_mode_uv](h->cv, &h->top_border_v[h->mbx*10],
974 h->left_border_v, h->c_stride);
975
976 decode_residual_chroma(h);
977 filter_mb(h,I_8X8);
978
979 /* mark motion vectors as intra */
980 h->mv[MV_FWD_X0] = intra_mv;
981 set_mvs(&h->mv[MV_FWD_X0], BLK_16X16);
982 h->mv[MV_BWD_X0] = intra_mv;
983 set_mvs(&h->mv[MV_BWD_X0], BLK_16X16);
984 if(h->pic_type != FF_B_TYPE)
985 *h->col_type = I_8X8;
986
987 return 0;
988 }
989
990 static void decode_mb_p(AVSContext *h, enum mb_t mb_type) {
991 GetBitContext *gb = &h->s.gb;
992 int ref[4];
993
994 init_mb(h);
995 switch(mb_type) {
996 case P_SKIP:
997 mv_pred(h, MV_FWD_X0, MV_FWD_C2, MV_PRED_PSKIP, BLK_16X16, 0);
998 break;
999 case P_16X16:
1000 ref[0] = h->ref_flag ? 0 : get_bits1(gb);
1001 mv_pred(h, MV_FWD_X0, MV_FWD_C2, MV_PRED_MEDIAN, BLK_16X16,ref[0]);
1002 break;
1003 case P_16X8:
1004 ref[0] = h->ref_flag ? 0 : get_bits1(gb);
1005 ref[2] = h->ref_flag ? 0 : get_bits1(gb);
1006 mv_pred(h, MV_FWD_X0, MV_FWD_C2, MV_PRED_TOP, BLK_16X8, ref[0]);
1007 mv_pred(h, MV_FWD_X2, MV_FWD_A1, MV_PRED_LEFT, BLK_16X8, ref[2]);
1008 break;
1009 case P_8X16:
1010 ref[0] = h->ref_flag ? 0 : get_bits1(gb);
1011 ref[1] = h->ref_flag ? 0 : get_bits1(gb);
1012 mv_pred(h, MV_FWD_X0, MV_FWD_B3, MV_PRED_LEFT, BLK_8X16, ref[0]);
1013 mv_pred(h, MV_FWD_X1, MV_FWD_C2, MV_PRED_TOPRIGHT, BLK_8X16, ref[1]);
1014 break;
1015 case P_8X8:
1016 ref[0] = h->ref_flag ? 0 : get_bits1(gb);
1017 ref[1] = h->ref_flag ? 0 : get_bits1(gb);
1018 ref[2] = h->ref_flag ? 0 : get_bits1(gb);
1019 ref[3] = h->ref_flag ? 0 : get_bits1(gb);
1020 mv_pred(h, MV_FWD_X0, MV_FWD_B3, MV_PRED_MEDIAN, BLK_8X8, ref[0]);
1021 mv_pred(h, MV_FWD_X1, MV_FWD_C2, MV_PRED_MEDIAN, BLK_8X8, ref[1]);
1022 mv_pred(h, MV_FWD_X2, MV_FWD_X1, MV_PRED_MEDIAN, BLK_8X8, ref[2]);
1023 mv_pred(h, MV_FWD_X3, MV_FWD_X0, MV_PRED_MEDIAN, BLK_8X8, ref[3]);
1024 }
1025 inter_pred(h, mb_type);
1026 store_mvs(h);
1027 if(mb_type != P_SKIP)
1028 decode_residual_inter(h);
1029 filter_mb(h,mb_type);
1030 *h->col_type = mb_type;
1031 }
1032
1033 static void decode_mb_b(AVSContext *h, enum mb_t mb_type) {
1034 int block;
1035 enum sub_mb_t sub_type[4];
1036 int flags;
1037
1038 init_mb(h);
1039
1040 /* reset all MVs */
1041 h->mv[MV_FWD_X0] = dir_mv;
1042 set_mvs(&h->mv[MV_FWD_X0], BLK_16X16);
1043 h->mv[MV_BWD_X0] = dir_mv;
1044 set_mvs(&h->mv[MV_BWD_X0], BLK_16X16);
1045 switch(mb_type) {
1046 case B_SKIP:
1047 case B_DIRECT:
1048 if(!(*h->col_type)) {
1049 /* intra MB at co-location, do in-plane prediction */
1050 mv_pred(h, MV_FWD_X0, MV_FWD_C2, MV_PRED_BSKIP, BLK_16X16, 1);
1051 mv_pred(h, MV_BWD_X0, MV_BWD_C2, MV_PRED_BSKIP, BLK_16X16, 0);
1052 } else
1053 /* direct prediction from co-located P MB, block-wise */
1054 for(block=0;block<4;block++)
1055 mv_pred_direct(h,&h->mv[mv_scan[block]],
1056 &h->col_mv[(h->mby*h->mb_width+h->mbx)*4 + block]);
1057 break;
1058 case B_FWD_16X16:
1059 mv_pred(h, MV_FWD_X0, MV_FWD_C2, MV_PRED_MEDIAN, BLK_16X16, 1);
1060 break;
1061 case B_SYM_16X16:
1062 mv_pred(h, MV_FWD_X0, MV_FWD_C2, MV_PRED_MEDIAN, BLK_16X16, 1);
1063 mv_pred_sym(h, &h->mv[MV_FWD_X0], BLK_16X16);
1064 break;
1065 case B_BWD_16X16:
1066 mv_pred(h, MV_BWD_X0, MV_BWD_C2, MV_PRED_MEDIAN, BLK_16X16, 0);
1067 break;
1068 case B_8X8:
1069 for(block=0;block<4;block++)
1070 sub_type[block] = get_bits(&h->s.gb,2);
1071 for(block=0;block<4;block++) {
1072 switch(sub_type[block]) {
1073 case B_SUB_DIRECT:
1074 if(!(*h->col_type)) {
1075 /* intra MB at co-location, do in-plane prediction */
1076 mv_pred(h, mv_scan[block], mv_scan[block]-3,
1077 MV_PRED_BSKIP, BLK_8X8, 1);
1078 mv_pred(h, mv_scan[block]+MV_BWD_OFFS,
1079 mv_scan[block]-3+MV_BWD_OFFS,
1080 MV_PRED_BSKIP, BLK_8X8, 0);
1081 } else
1082 mv_pred_direct(h,&h->mv[mv_scan[block]],
1083 &h->col_mv[(h->mby*h->mb_width + h->mbx)*4 + block]);
1084 break;
1085 case B_SUB_FWD:
1086 mv_pred(h, mv_scan[block], mv_scan[block]-3,
1087 MV_PRED_MEDIAN, BLK_8X8, 1);
1088 break;
1089 case B_SUB_SYM:
1090 mv_pred(h, mv_scan[block], mv_scan[block]-3,
1091 MV_PRED_MEDIAN, BLK_8X8, 1);
1092 mv_pred_sym(h, &h->mv[mv_scan[block]], BLK_8X8);
1093 break;
1094 }
1095 }
1096 for(block=0;block<4;block++) {
1097 if(sub_type[block] == B_SUB_BWD)
1098 mv_pred(h, mv_scan[block]+MV_BWD_OFFS,
1099 mv_scan[block]+MV_BWD_OFFS-3,
1100 MV_PRED_MEDIAN, BLK_8X8, 0);
1101 }
1102 break;
1103 default:
1104 assert((mb_type > B_SYM_16X16) && (mb_type < B_8X8));
1105 flags = b_partition_flags[(mb_type-1)>>1];
1106 if(mb_type & 1) { /* 16x8 macroblock types */
1107 if(flags & FWD0)
1108 mv_pred(h, MV_FWD_X0, MV_FWD_C2, MV_PRED_TOP, BLK_16X8, 1);
1109 if(flags & SYM0) {
1110 mv_pred(h, MV_FWD_X0, MV_FWD_C2, MV_PRED_TOP, BLK_16X8, 1);
1111 mv_pred_sym(h, &h->mv[MV_FWD_X0], BLK_16X8);
1112 }
1113 if(flags & FWD1)
1114 mv_pred(h, MV_FWD_X2, MV_FWD_A1, MV_PRED_LEFT, BLK_16X8, 1);
1115 if(flags & SYM1) {
1116 mv_pred(h, MV_FWD_X2, MV_FWD_A1, MV_PRED_LEFT, BLK_16X8, 1);
1117 mv_pred_sym(h, &h->mv[9], BLK_16X8);
1118 }
1119 if(flags & BWD0)
1120 mv_pred(h, MV_BWD_X0, MV_BWD_C2, MV_PRED_TOP, BLK_16X8, 0);
1121 if(flags & BWD1)
1122 mv_pred(h, MV_BWD_X2, MV_BWD_A1, MV_PRED_LEFT, BLK_16X8, 0);
1123 } else { /* 8x16 macroblock types */
1124 if(flags & FWD0)
1125 mv_pred(h, MV_FWD_X0, MV_FWD_B3, MV_PRED_LEFT, BLK_8X16, 1);
1126 if(flags & SYM0) {
1127 mv_pred(h, MV_FWD_X0, MV_FWD_B3, MV_PRED_LEFT, BLK_8X16, 1);
1128 mv_pred_sym(h, &h->mv[MV_FWD_X0], BLK_8X16);
1129 }
1130 if(flags & FWD1)
1131 mv_pred(h, MV_FWD_X1, MV_FWD_C2, MV_PRED_TOPRIGHT,BLK_8X16, 1);
1132 if(flags & SYM1) {
1133 mv_pred(h, MV_FWD_X1, MV_FWD_C2, MV_PRED_TOPRIGHT,BLK_8X16, 1);
1134 mv_pred_sym(h, &h->mv[6], BLK_8X16);
1135 }
1136 if(flags & BWD0)
1137 mv_pred(h, MV_BWD_X0, MV_BWD_B3, MV_PRED_LEFT, BLK_8X16, 0);
1138 if(flags & BWD1)
1139 mv_pred(h, MV_BWD_X1, MV_BWD_C2, MV_PRED_TOPRIGHT,BLK_8X16, 0);
1140 }
1141 }
1142 inter_pred(h, mb_type);
1143 if(mb_type != B_SKIP)
1144 decode_residual_inter(h);
1145 filter_mb(h,mb_type);
1146 }
1147
1148 /*****************************************************************************
1149 *
1150 * slice level
1151 *
1152 ****************************************************************************/
1153
1154 static inline int decode_slice_header(AVSContext *h, GetBitContext *gb) {
1155 if(h->stc > 0xAF)
1156 av_log(h->s.avctx, AV_LOG_ERROR, "unexpected start code 0x%02x\n", h->stc);
1157 h->mby = h->stc;
1158 if((h->mby == 0) && (!h->qp_fixed)){
1159 h->qp_fixed = get_bits1(gb);
1160 h->qp = get_bits(gb,6);
1161 }
1162 /* inter frame or second slice can have weighting params */
1163 if((h->pic_type != FF_I_TYPE) || (!h->pic_structure && h->mby >= h->mb_width/2))
1164 if(get_bits1(gb)) { //slice_weighting_flag
1165 av_log(h->s.avctx, AV_LOG_ERROR,
1166 "weighted prediction not yet supported\n");
1167 }
1168 return 0;
1169 }
1170
1171 static inline void check_for_slice(AVSContext *h) {
1172 GetBitContext *gb = &h->s.gb;
1173 int align;
1174 align = (-get_bits_count(gb)) & 7;
1175 if((show_bits_long(gb,24+align) & 0xFFFFFF) == 0x000001) {
1176 get_bits_long(gb,24+align);
1177 h->stc = get_bits(gb,8);
1178 decode_slice_header(h,gb);
1179 }
1180 }
1181
1182 /*****************************************************************************
1183 *
1184 * frame level
1185 *
1186 ****************************************************************************/
1187
1188 static void init_pic(AVSContext *h) {
1189 int i;
1190
1191 /* clear some predictors */
1192 for(i=0;i<=20;i+=4)
1193 h->mv[i] = un_mv;
1194 h->mv[MV_BWD_X0] = dir_mv;
1195 set_mvs(&h->mv[MV_BWD_X0], BLK_16X16);
1196 h->mv[MV_FWD_X0] = dir_mv;
1197 set_mvs(&h->mv[MV_FWD_X0], BLK_16X16);
1198 h->pred_mode_Y[3] = h->pred_mode_Y[6] = NOT_AVAIL;
1199 h->cy = h->picture.data[0];
1200 h->cu = h->picture.data[1];
1201 h->cv = h->picture.data[2];
1202 h->l_stride = h->picture.linesize[0];
1203 h->c_stride = h->picture.linesize[1];
1204 h->luma_scan[2] = 8*h->l_stride;
1205 h->luma_scan[3] = 8*h->l_stride+8;
1206 h->mbx = h->mby = 0;
1207 h->flags = 0;
1208 }
1209
1210 static int decode_pic(AVSContext *h) {
1211 MpegEncContext *s = &h->s;
1212 int skip_count;
1213 enum mb_t mb_type;
1214
1215 if (!s->context_initialized) {
1216 if (MPV_common_init(s) < 0)
1217 return -1;
1218 }
1219 get_bits(&s->gb,16);//bbv_dwlay
1220 if(h->stc == PIC_PB_START_CODE) {
1221 h->pic_type = get_bits(&s->gb,2) + FF_I_TYPE;
1222 /* make sure we have the reference frames we need */
1223 if(!h->DPB[0].data[0] ||
1224 (!h->DPB[1].data[0] && h->pic_type == FF_B_TYPE))
1225 return -1;
1226 } else {
1227 h->pic_type = FF_I_TYPE;
1228 if(get_bits1(&s->gb))
1229 get_bits(&s->gb,16);//time_code
1230 }
1231 /* release last B frame */
1232 if(h->picture.data[0])
1233 s->avctx->release_buffer(s->avctx, (AVFrame *)&h->picture);
1234
1235 s->avctx->get_buffer(s->avctx, (AVFrame *)&h->picture);
1236 init_pic(h);
1237 h->picture.poc = get_bits(&s->gb,8)*2;
1238
1239 /* get temporal distances and MV scaling factors */
1240 if(h->pic_type != FF_B_TYPE) {
1241 h->dist[0] = (h->picture.poc - h->DPB[0].poc + 512) % 512;
1242 } else {
1243 h->dist[0] = (h->DPB[0].poc - h->picture.poc + 512) % 512;
1244 }
1245 h->dist[1] = (h->picture.poc - h->DPB[1].poc + 512) % 512;
1246 h->scale_den[0] = h->dist[0] ? 512/h->dist[0] : 0;
1247 h->scale_den[1] = h->dist[1] ? 512/h->dist[1] : 0;
1248 if(h->pic_type == FF_B_TYPE) {
1249 h->sym_factor = h->dist[0]*h->scale_den[1];
1250 } else {
1251 h->direct_den[0] = h->dist[0] ? 16384/h->dist[0] : 0;
1252 h->direct_den[1] = h->dist[1] ? 16384/h->dist[1] : 0;
1253 }
1254
1255 if(s->low_delay)
1256 get_ue_golomb(&s->gb); //bbv_check_times
1257 h->progressive = get_bits1(&s->gb);
1258 if(h->progressive)
1259 h->pic_structure = 1;
1260 else if(!(h->pic_structure = get_bits1(&s->gb) && (h->stc == PIC_PB_START_CODE)) )
1261 get_bits1(&s->gb); //advanced_pred_mode_disable
1262 skip_bits1(&s->gb); //top_field_first
1263 skip_bits1(&s->gb); //repeat_first_field
1264 h->qp_fixed = get_bits1(&s->gb);
1265 h->qp = get_bits(&s->gb,6);
1266 if(h->pic_type == FF_I_TYPE) {
1267 if(!h->progressive && !h->pic_structure)
1268 skip_bits1(&s->gb);//what is this?
1269 skip_bits(&s->gb,4); //reserved bits
1270 } else {
1271 if(!(h->pic_type == FF_B_TYPE && h->pic_structure == 1))
1272 h->ref_flag = get_bits1(&s->gb);
1273 skip_bits(&s->gb,4); //reserved bits
1274 h->skip_mode_flag = get_bits1(&s->gb);
1275 }
1276 h->loop_filter_disable = get_bits1(&s->gb);
1277 if(!h->loop_filter_disable && get_bits1(&s->gb)) {
1278 h->alpha_offset = get_se_golomb(&s->gb);
1279 h->beta_offset = get_se_golomb(&s->gb);
1280 } else {
1281 h->alpha_offset = h->beta_offset = 0;
1282 }
1283 check_for_slice(h);
1284 if(h->pic_type == FF_I_TYPE) {
1285 do {
1286 decode_mb_i(h, 0);
1287 } while(next_mb(h));
1288 } else if(h->pic_type == FF_P_TYPE) {
1289 do {
1290 if(h->skip_mode_flag) {
1291 skip_count = get_ue_golomb(&s->gb);
1292 while(skip_count--) {
1293 decode_mb_p(h,P_SKIP);
1294 if(!next_mb(h))
1295 goto done;
1296 }
1297 mb_type = get_ue_golomb(&s->gb) + P_16X16;
1298 } else
1299 mb_type = get_ue_golomb(&s->gb) + P_SKIP;
1300 if(mb_type > P_8X8) {
1301 decode_mb_i(h, mb_type - P_8X8 - 1);
1302 } else
1303 decode_mb_p(h,mb_type);
1304 } while(next_mb(h));
1305 } else { /* FF_B_TYPE */
1306 do {
1307 if(h->skip_mode_flag) {
1308 skip_count = get_ue_golomb(&s->gb);
1309 while(skip_count--) {
1310 decode_mb_b(h,B_SKIP);
1311 if(!next_mb(h))
1312 goto done;
1313 }
1314 mb_type = get_ue_golomb(&s->gb) + B_DIRECT;
1315 } else
1316 mb_type = get_ue_golomb(&s->gb) + B_SKIP;
1317 if(mb_type > B_8X8) {
1318 decode_mb_i(h, mb_type - B_8X8 - 1);
1319 } else
1320 decode_mb_b(h,mb_type);
1321 } while(next_mb(h));
1322 }
1323 done:
1324 if(h->pic_type != FF_B_TYPE) {
1325 if(h->DPB[1].data[0])
1326 s->avctx->release_buffer(s->avctx, (AVFrame *)&h->DPB[1]);
1327 memcpy(&h->DPB[1], &h->DPB[0], sizeof(Picture));
1328 memcpy(&h->DPB[0], &h->picture, sizeof(Picture));
1329 memset(&h->picture,0,sizeof(Picture));
1330 }
1331 return 0;
1332 }
1333
1334 /*****************************************************************************
1335 *
1336 * headers and interface
1337 *
1338 ****************************************************************************/
1339
1340 /**
1341 * some predictions require data from the top-neighbouring macroblock.
1342 * this data has to be stored for one complete row of macroblocks
1343 * and this storage space is allocated here
1344 */
1345 static void init_top_lines(AVSContext *h) {
1346 /* alloc top line of predictors */
1347 h->top_qp = av_malloc( h->mb_width);
1348 h->top_mv[0] = av_malloc((h->mb_width*2+1)*sizeof(vector_t));
1349 h->top_mv[1] = av_malloc((h->mb_width*2+1)*sizeof(vector_t));
1350 h->top_pred_Y = av_malloc( h->mb_width*2*sizeof(*h->top_pred_Y));
1351 h->top_border_y = av_malloc((h->mb_width+1)*16);
1352 h->top_border_u = av_malloc((h->mb_width)*10);
1353 h->top_border_v = av_malloc((h->mb_width)*10);
1354
1355 /* alloc space for co-located MVs and types */
1356 h->col_mv = av_malloc( h->mb_width*h->mb_height*4*sizeof(vector_t));
1357 h->col_type_base = av_malloc(h->mb_width*h->mb_height);
1358 }
1359
1360 static int decode_seq_header(AVSContext *h) {
1361 MpegEncContext *s = &h->s;
1362 extern const AVRational ff_frame_rate_tab[];
1363 int frame_rate_code;
1364
1365 h->profile = get_bits(&s->gb,8);
1366 h->level = get_bits(&s->gb,8);
1367 skip_bits1(&s->gb); //progressive sequence
1368 s->width = get_bits(&s->gb,14);
1369 s->height = get_bits(&s->gb,14);
1370 skip_bits(&s->gb,2); //chroma format
1371 skip_bits(&s->gb,3); //sample_precision
1372 h->aspect_ratio = get_bits(&s->gb,4);
1373 frame_rate_code = get_bits(&s->gb,4);
1374 skip_bits(&s->gb,18);//bit_rate_lower
1375 skip_bits1(&s->gb); //marker_bit
1376 skip_bits(&s->gb,12);//bit_rate_upper
1377 s->low_delay = get_bits1(&s->gb);
1378 h->mb_width = (s->width + 15) >> 4;
1379 h->mb_height = (s->height + 15) >> 4;
1380 h->s.avctx->time_base.den = ff_frame_rate_tab[frame_rate_code].num;
1381 h->s.avctx->time_base.num = ff_frame_rate_tab[frame_rate_code].den;
1382 h->s.avctx->width = s->width;
1383 h->s.avctx->height = s->height;
1384 if(!h->top_qp)
1385 init_top_lines(h);
1386 return 0;
1387 }
1388
1389 /**
1390 * finds the end of the current frame in the bitstream.
1391 * @return the position of the first byte of the next frame, or -1
1392 */
1393 int ff_cavs_find_frame_end(ParseContext *pc, const uint8_t *buf, int buf_size) {
1394 int pic_found, i;
1395 uint32_t state;
1396
1397 pic_found= pc->frame_start_found;
1398 state= pc->state;
1399
1400 i=0;
1401 if(!pic_found){
1402 for(i=0; i<buf_size; i++){
1403 state= (state<<8) | buf[i];
1404 if(state == PIC_I_START_CODE || state == PIC_PB_START_CODE){
1405 i++;
1406 pic_found=1;
1407 break;
1408 }
1409 }
1410 }
1411
1412 if(pic_found){
1413 /* EOF considered as end of frame */
1414 if (buf_size == 0)
1415 return 0;
1416 for(; i<buf_size; i++){
1417 state= (state<<8) | buf[i];
1418 if((state&0xFFFFFF00) == 0x100){
1419 if(state < SLICE_MIN_START_CODE || state > SLICE_MAX_START_CODE){
1420 pc->frame_start_found=0;
1421 pc->state=-1;
1422 return i-3;
1423 }
1424 }
1425 }
1426 }
1427 pc->frame_start_found= pic_found;
1428 pc->state= state;
1429 return END_NOT_FOUND;
1430 }
1431
1432 void ff_cavs_flush(AVCodecContext * avctx) {
1433 AVSContext *h = avctx->priv_data;
1434 h->got_keyframe = 0;
1435 }
1436
1437 static int cavs_decode_frame(AVCodecContext * avctx,void *data, int *data_size,
1438 uint8_t * buf, int buf_size) {
1439 AVSContext *h = avctx->priv_data;
1440 MpegEncContext *s = &h->s;
1441 int input_size;
1442 const uint8_t *buf_end;
1443 const uint8_t *buf_ptr;
1444 AVFrame *picture = data;
1445 uint32_t stc;
1446
1447 s->avctx = avctx;
1448
1449 if (buf_size == 0) {
1450 if(!s->low_delay && h->DPB[0].data[0]) {
1451 *data_size = sizeof(AVPicture);
1452 *picture = *(AVFrame *) &h->DPB[0];
1453 }
1454 return 0;
1455 }
1456
1457 buf_ptr = buf;
1458 buf_end = buf + buf_size;
1459 for(;;) {
1460 buf_ptr = ff_find_start_code(buf_ptr,buf_end, &stc);
1461 if(stc & 0xFFFFFE00)
1462 return FFMAX(0, buf_ptr - buf - s->parse_context.last_index);
1463 input_size = (buf_end - buf_ptr)*8;
1464 switch(stc) {
1465 case SEQ_START_CODE:
1466 init_get_bits(&s->gb, buf_ptr, input_size);
1467 decode_seq_header(h);
1468 break;
1469 case PIC_I_START_CODE:
1470 if(!h->got_keyframe) {
1471 if(h->DPB[0].data[0])
1472 avctx->release_buffer(avctx, (AVFrame *)&h->DPB[0]);
1473 if(h->DPB[1].data[0])
1474 avctx->release_buffer(avctx, (AVFrame *)&h->DPB[1]);
1475 h->got_keyframe = 1;
1476 }
1477 case PIC_PB_START_CODE:
1478 *data_size = 0;
1479 if(!h->got_keyframe)
1480 break;
1481 init_get_bits(&s->gb, buf_ptr, input_size);
1482 h->stc = stc;
1483 if(decode_pic(h))
1484 break;
1485 *data_size = sizeof(AVPicture);
1486 if(h->pic_type != FF_B_TYPE) {
1487 if(h->DPB[1].data[0]) {
1488 *picture = *(AVFrame *) &h->DPB[1];
1489 } else {
1490 *data_size = 0;
1491 }
1492 } else
1493 *picture = *(AVFrame *) &h->picture;
1494 break;
1495 case EXT_START_CODE:
1496 //mpeg_decode_extension(avctx,buf_ptr, input_size);
1497 break;
1498 case USER_START_CODE:
1499 //mpeg_decode_user_data(avctx,buf_ptr, input_size);
1500 break;
1501 default:
1502 if (stc >= SLICE_MIN_START_CODE &&
1503 stc <= SLICE_MAX_START_CODE) {
1504 init_get_bits(&s->gb, buf_ptr, input_size);
1505 decode_slice_header(h, &s->gb);
1506 }
1507 break;
1508 }
1509 }
1510 }
1511
1512 static int cavs_decode_init(AVCodecContext * avctx) {
1513 AVSContext *h = avctx->priv_data;
1514 MpegEncContext * const s = &h->s;
1515
1516 MPV_decode_defaults(s);
1517 s->avctx = avctx;
1518
1519 avctx->pix_fmt= PIX_FMT_YUV420P;
1520
1521 h->luma_scan[0] = 0;
1522 h->luma_scan[1] = 8;
1523 h->intra_pred_l[ INTRA_L_VERT] = intra_pred_vert;
1524 h->intra_pred_l[ INTRA_L_HORIZ] = intra_pred_horiz;
1525 h->intra_pred_l[ INTRA_L_LP] = intra_pred_lp;
1526 h->intra_pred_l[ INTRA_L_DOWN_LEFT] = intra_pred_down_left;
1527 h->intra_pred_l[INTRA_L_DOWN_RIGHT] = intra_pred_down_right;
1528 h->intra_pred_l[ INTRA_L_LP_LEFT] = intra_pred_lp_left;
1529 h->intra_pred_l[ INTRA_L_LP_TOP] = intra_pred_lp_top;
1530 h->intra_pred_l[ INTRA_L_DC_128] = intra_pred_dc_128;
1531 h->intra_pred_c[ INTRA_C_LP] = intra_pred_lp;
1532 h->intra_pred_c[ INTRA_C_HORIZ] = intra_pred_horiz;
1533 h->intra_pred_c[ INTRA_C_VERT] = intra_pred_vert;
1534 h->intra_pred_c[ INTRA_C_PLANE] = intra_pred_plane;
1535 h->intra_pred_c[ INTRA_C_LP_LEFT] = intra_pred_lp_left;
1536 h->intra_pred_c[ INTRA_C_LP_TOP] = intra_pred_lp_top;
1537 h->intra_pred_c[ INTRA_C_DC_128] = intra_pred_dc_128;
1538 h->mv[ 7] = un_mv;
1539 h->mv[19] = un_mv;
1540 return 0;
1541 }
1542
1543 static int cavs_decode_end(AVCodecContext * avctx) {
1544 AVSContext *h = avctx->priv_data;
1545
1546 av_free(h->top_qp);
1547 av_free(h->top_mv[0]);
1548 av_free(h->top_mv[1]);
1549 av_free(h->top_pred_Y);
1550 av_free(h->top_border_y);
1551 av_free(h->top_border_u);
1552 av_free(h->top_border_v);
1553 av_free(h->col_mv);
1554 av_free(h->col_type_base);
1555 return 0;
1556 }
1557
1558 AVCodec cavs_decoder = {
1559 "cavs",
1560 CODEC_TYPE_VIDEO,
1561 CODEC_ID_CAVS,
1562 sizeof(AVSContext),
1563 cavs_decode_init,
1564 NULL,
1565 cavs_decode_end,
1566 cavs_decode_frame,
1567 CODEC_CAP_DR1 | CODEC_CAP_DELAY,
1568 .flush= ff_cavs_flush,
1569 };