2 * Chinese AVS video (AVS1-P2, JiZhun profile) decoder.
3 * Copyright (c) 2006 Stefan Gehrer <stefan.gehrer@gmx.de>
5 * This file is part of FFmpeg.
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
24 * Chinese AVS video (AVS1-P2, JiZhun profile) decoder
25 * @author Stefan Gehrer <stefan.gehrer@gmx.de>
29 #include "bitstream.h"
34 /*****************************************************************************
36 * in-loop deblocking filter
38 ****************************************************************************/
40 static inline int get_bs(vector_t *mvP, vector_t *mvQ, int b) {
41 if((mvP->ref == REF_INTRA) || (mvQ->ref == REF_INTRA))
43 if( (abs(mvP->x - mvQ->x) >= 4) || (abs(mvP->y - mvQ->y) >= 4) )
48 if( (abs(mvP->x - mvQ->x) >= 4) || (abs(mvP->y - mvQ->y) >= 4) )
51 if(mvP->ref != mvQ->ref)
58 alpha = alpha_tab[av_clip(qp_avg + h->alpha_offset,0,63)]; \
59 beta = beta_tab[av_clip(qp_avg + h->beta_offset, 0,63)]; \
60 tc = tc_tab[av_clip(qp_avg + h->alpha_offset,0,63)];
63 * in-loop deblocking filter for a single macroblock
65 * boundary strength (bs) mapping:
74 static void filter_mb(AVSContext *h, enum mb_t mb_type) {
75 DECLARE_ALIGNED_8(uint8_t, bs[8]);
76 int qp_avg, alpha, beta, tc;
79 /* save un-deblocked lines */
80 h->topleft_border_y = h->top_border_y[h->mbx*16+15];
81 h->topleft_border_u = h->top_border_u[h->mbx*10+8];
82 h->topleft_border_v = h->top_border_v[h->mbx*10+8];
83 memcpy(&h->top_border_y[h->mbx*16], h->cy + 15* h->l_stride,16);
84 memcpy(&h->top_border_u[h->mbx*10+1], h->cu + 7* h->c_stride,8);
85 memcpy(&h->top_border_v[h->mbx*10+1], h->cv + 7* h->c_stride,8);
87 h->left_border_y[i*2+1] = *(h->cy + 15 + (i*2+0)*h->l_stride);
88 h->left_border_y[i*2+2] = *(h->cy + 15 + (i*2+1)*h->l_stride);
89 h->left_border_u[i+1] = *(h->cu + 7 + i*h->c_stride);
90 h->left_border_v[i+1] = *(h->cv + 7 + i*h->c_stride);
92 if(!h->loop_filter_disable) {
95 *((uint64_t *)bs) = 0x0202020202020202ULL;
97 *((uint64_t *)bs) = 0;
98 if(partition_flags[mb_type] & SPLITV){
99 bs[2] = get_bs(&h->mv[MV_FWD_X0], &h->mv[MV_FWD_X1], mb_type > P_8X8);
100 bs[3] = get_bs(&h->mv[MV_FWD_X2], &h->mv[MV_FWD_X3], mb_type > P_8X8);
102 if(partition_flags[mb_type] & SPLITH){
103 bs[6] = get_bs(&h->mv[MV_FWD_X0], &h->mv[MV_FWD_X2], mb_type > P_8X8);
104 bs[7] = get_bs(&h->mv[MV_FWD_X1], &h->mv[MV_FWD_X3], mb_type > P_8X8);
106 bs[0] = get_bs(&h->mv[MV_FWD_A1], &h->mv[MV_FWD_X0], mb_type > P_8X8);
107 bs[1] = get_bs(&h->mv[MV_FWD_A3], &h->mv[MV_FWD_X2], mb_type > P_8X8);
108 bs[4] = get_bs(&h->mv[MV_FWD_B2], &h->mv[MV_FWD_X0], mb_type > P_8X8);
109 bs[5] = get_bs(&h->mv[MV_FWD_B3], &h->mv[MV_FWD_X1], mb_type > P_8X8);
111 if( *((uint64_t *)bs) ) {
112 if(h->flags & A_AVAIL) {
113 qp_avg = (h->qp + h->left_qp + 1) >> 1;
115 h->s.dsp.cavs_filter_lv(h->cy,h->l_stride,alpha,beta,tc,bs[0],bs[1]);
116 h->s.dsp.cavs_filter_cv(h->cu,h->c_stride,alpha,beta,tc,bs[0],bs[1]);
117 h->s.dsp.cavs_filter_cv(h->cv,h->c_stride,alpha,beta,tc,bs[0],bs[1]);
121 h->s.dsp.cavs_filter_lv(h->cy + 8,h->l_stride,alpha,beta,tc,bs[2],bs[3]);
122 h->s.dsp.cavs_filter_lh(h->cy + 8*h->l_stride,h->l_stride,alpha,beta,tc,
125 if(h->flags & B_AVAIL) {
126 qp_avg = (h->qp + h->top_qp[h->mbx] + 1) >> 1;
128 h->s.dsp.cavs_filter_lh(h->cy,h->l_stride,alpha,beta,tc,bs[4],bs[5]);
129 h->s.dsp.cavs_filter_ch(h->cu,h->c_stride,alpha,beta,tc,bs[4],bs[5]);
130 h->s.dsp.cavs_filter_ch(h->cv,h->c_stride,alpha,beta,tc,bs[4],bs[5]);
135 h->top_qp[h->mbx] = h->qp;
140 /*****************************************************************************
142 * spatial intra prediction
144 ****************************************************************************/
146 static void intra_pred_vert(uint8_t *d,uint8_t *top,uint8_t *left,int stride) {
148 uint64_t a = unaligned64(&top[1]);
150 *((uint64_t *)(d+y*stride)) = a;
154 static void intra_pred_horiz(uint8_t *d,uint8_t *top,uint8_t *left,int stride) {
158 a = left[y+1] * 0x0101010101010101ULL;
159 *((uint64_t *)(d+y*stride)) = a;
163 static void intra_pred_dc_128(uint8_t *d,uint8_t *top,uint8_t *left,int stride) {
165 uint64_t a = 0x8080808080808080ULL;
167 *((uint64_t *)(d+y*stride)) = a;
170 static void intra_pred_plane(uint8_t *d,uint8_t *top,uint8_t *left,int stride) {
174 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
177 ih += (x+1)*(top[5+x]-top[3-x]);
178 iv += (x+1)*(left[5+x]-left[3-x]);
180 ia = (top[8]+left[8])<<4;
185 d[y*stride+x] = cm[(ia+(x-3)*ih+(y-3)*iv+16)>>5];
188 #define LOWPASS(ARRAY,INDEX) \
189 (( ARRAY[(INDEX)-1] + 2*ARRAY[(INDEX)] + ARRAY[(INDEX)+1] + 2) >> 2)
191 static void intra_pred_lp(uint8_t *d,uint8_t *top,uint8_t *left,int stride) {
195 d[y*stride+x] = (LOWPASS(top,x+1) + LOWPASS(left,y+1)) >> 1;
198 static void intra_pred_down_left(uint8_t *d,uint8_t *top,uint8_t *left,int stride) {
202 d[y*stride+x] = (LOWPASS(top,x+y+2) + LOWPASS(left,x+y+2)) >> 1;
205 static void intra_pred_down_right(uint8_t *d,uint8_t *top,uint8_t *left,int stride) {
210 d[y*stride+x] = (left[1]+2*top[0]+top[1]+2)>>2;
212 d[y*stride+x] = LOWPASS(top,x-y);
214 d[y*stride+x] = LOWPASS(left,y-x);
217 static void intra_pred_lp_left(uint8_t *d,uint8_t *top,uint8_t *left,int stride) {
221 d[y*stride+x] = LOWPASS(left,y+1);
224 static void intra_pred_lp_top(uint8_t *d,uint8_t *top,uint8_t *left,int stride) {
228 d[y*stride+x] = LOWPASS(top,x+1);
233 /*****************************************************************************
235 * motion compensation
237 ****************************************************************************/
239 static inline void mc_dir_part(AVSContext *h,Picture *pic,int square,
240 int chroma_height,int delta,int list,uint8_t *dest_y,
241 uint8_t *dest_cb,uint8_t *dest_cr,int src_x_offset,
242 int src_y_offset,qpel_mc_func *qpix_op,
243 h264_chroma_mc_func chroma_op,vector_t *mv){
244 MpegEncContext * const s = &h->s;
245 const int mx= mv->x + src_x_offset*8;
246 const int my= mv->y + src_y_offset*8;
247 const int luma_xy= (mx&3) + ((my&3)<<2);
248 uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*h->l_stride;
249 uint8_t * src_cb= pic->data[1] + (mx>>3) + (my>>3)*h->c_stride;
250 uint8_t * src_cr= pic->data[2] + (mx>>3) + (my>>3)*h->c_stride;
251 int extra_width= 0; //(s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16;
252 int extra_height= extra_width;
254 const int full_mx= mx>>2;
255 const int full_my= my>>2;
256 const int pic_width = 16*h->mb_width;
257 const int pic_height = 16*h->mb_height;
261 if(mx&7) extra_width -= 3;
262 if(my&7) extra_height -= 3;
264 if( full_mx < 0-extra_width
265 || full_my < 0-extra_height
266 || full_mx + 16/*FIXME*/ > pic_width + extra_width
267 || full_my + 16/*FIXME*/ > pic_height + extra_height){
268 ff_emulated_edge_mc(s->edge_emu_buffer, src_y - 2 - 2*h->l_stride, h->l_stride,
269 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
270 src_y= s->edge_emu_buffer + 2 + 2*h->l_stride;
274 qpix_op[luma_xy](dest_y, src_y, h->l_stride); //FIXME try variable height perhaps?
276 qpix_op[luma_xy](dest_y + delta, src_y + delta, h->l_stride);
280 ff_emulated_edge_mc(s->edge_emu_buffer, src_cb, h->c_stride,
281 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
282 src_cb= s->edge_emu_buffer;
284 chroma_op(dest_cb, src_cb, h->c_stride, chroma_height, mx&7, my&7);
287 ff_emulated_edge_mc(s->edge_emu_buffer, src_cr, h->c_stride,
288 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
289 src_cr= s->edge_emu_buffer;
291 chroma_op(dest_cr, src_cr, h->c_stride, chroma_height, mx&7, my&7);
294 static inline void mc_part_std(AVSContext *h,int square,int chroma_height,int delta,
295 uint8_t *dest_y,uint8_t *dest_cb,uint8_t *dest_cr,
296 int x_offset, int y_offset,qpel_mc_func *qpix_put,
297 h264_chroma_mc_func chroma_put,qpel_mc_func *qpix_avg,
298 h264_chroma_mc_func chroma_avg, vector_t *mv){
299 qpel_mc_func *qpix_op= qpix_put;
300 h264_chroma_mc_func chroma_op= chroma_put;
302 dest_y += 2*x_offset + 2*y_offset*h->l_stride;
303 dest_cb += x_offset + y_offset*h->c_stride;
304 dest_cr += x_offset + y_offset*h->c_stride;
305 x_offset += 8*h->mbx;
306 y_offset += 8*h->mby;
309 Picture *ref= &h->DPB[mv->ref];
310 mc_dir_part(h, ref, square, chroma_height, delta, 0,
311 dest_y, dest_cb, dest_cr, x_offset, y_offset,
312 qpix_op, chroma_op, mv);
315 chroma_op= chroma_avg;
318 if((mv+MV_BWD_OFFS)->ref >= 0){
319 Picture *ref= &h->DPB[0];
320 mc_dir_part(h, ref, square, chroma_height, delta, 1,
321 dest_y, dest_cb, dest_cr, x_offset, y_offset,
322 qpix_op, chroma_op, mv+MV_BWD_OFFS);
326 static void inter_pred(AVSContext *h, enum mb_t mb_type) {
327 if(partition_flags[mb_type] == 0){ // 16x16
328 mc_part_std(h, 1, 8, 0, h->cy, h->cu, h->cv, 0, 0,
329 h->s.dsp.put_cavs_qpel_pixels_tab[0],
330 h->s.dsp.put_h264_chroma_pixels_tab[0],
331 h->s.dsp.avg_cavs_qpel_pixels_tab[0],
332 h->s.dsp.avg_h264_chroma_pixels_tab[0],&h->mv[MV_FWD_X0]);
334 mc_part_std(h, 1, 4, 0, h->cy, h->cu, h->cv, 0, 0,
335 h->s.dsp.put_cavs_qpel_pixels_tab[1],
336 h->s.dsp.put_h264_chroma_pixels_tab[1],
337 h->s.dsp.avg_cavs_qpel_pixels_tab[1],
338 h->s.dsp.avg_h264_chroma_pixels_tab[1],&h->mv[MV_FWD_X0]);
339 mc_part_std(h, 1, 4, 0, h->cy, h->cu, h->cv, 4, 0,
340 h->s.dsp.put_cavs_qpel_pixels_tab[1],
341 h->s.dsp.put_h264_chroma_pixels_tab[1],
342 h->s.dsp.avg_cavs_qpel_pixels_tab[1],
343 h->s.dsp.avg_h264_chroma_pixels_tab[1],&h->mv[MV_FWD_X1]);
344 mc_part_std(h, 1, 4, 0, h->cy, h->cu, h->cv, 0, 4,
345 h->s.dsp.put_cavs_qpel_pixels_tab[1],
346 h->s.dsp.put_h264_chroma_pixels_tab[1],
347 h->s.dsp.avg_cavs_qpel_pixels_tab[1],
348 h->s.dsp.avg_h264_chroma_pixels_tab[1],&h->mv[MV_FWD_X2]);
349 mc_part_std(h, 1, 4, 0, h->cy, h->cu, h->cv, 4, 4,
350 h->s.dsp.put_cavs_qpel_pixels_tab[1],
351 h->s.dsp.put_h264_chroma_pixels_tab[1],
352 h->s.dsp.avg_cavs_qpel_pixels_tab[1],
353 h->s.dsp.avg_h264_chroma_pixels_tab[1],&h->mv[MV_FWD_X3]);
357 /*****************************************************************************
359 * motion vector prediction
361 ****************************************************************************/
363 static inline void store_mvs(AVSContext *h) {
364 h->col_mv[(h->mby*h->mb_width + h->mbx)*4 + 0] = h->mv[MV_FWD_X0];
365 h->col_mv[(h->mby*h->mb_width + h->mbx)*4 + 1] = h->mv[MV_FWD_X1];
366 h->col_mv[(h->mby*h->mb_width + h->mbx)*4 + 2] = h->mv[MV_FWD_X2];
367 h->col_mv[(h->mby*h->mb_width + h->mbx)*4 + 3] = h->mv[MV_FWD_X3];
370 static inline void scale_mv(AVSContext *h, int *d_x, int *d_y, vector_t *src, int distp) {
371 int den = h->scale_den[src->ref];
373 *d_x = (src->x*distp*den + 256 + (src->x>>31)) >> 9;
374 *d_y = (src->y*distp*den + 256 + (src->y>>31)) >> 9;
377 static inline void mv_pred_median(AVSContext *h, vector_t *mvP, vector_t *mvA, vector_t *mvB, vector_t *mvC) {
378 int ax, ay, bx, by, cx, cy;
379 int len_ab, len_bc, len_ca, len_mid;
381 /* scale candidates according to their temporal span */
382 scale_mv(h, &ax, &ay, mvA, mvP->dist);
383 scale_mv(h, &bx, &by, mvB, mvP->dist);
384 scale_mv(h, &cx, &cy, mvC, mvP->dist);
385 /* find the geometrical median of the three candidates */
386 len_ab = abs(ax - bx) + abs(ay - by);
387 len_bc = abs(bx - cx) + abs(by - cy);
388 len_ca = abs(cx - ax) + abs(cy - ay);
389 len_mid = mid_pred(len_ab, len_bc, len_ca);
390 if(len_mid == len_ab) {
393 } else if(len_mid == len_bc) {
402 static inline void mv_pred_direct(AVSContext *h, vector_t *pmv_fw,
404 vector_t *pmv_bw = pmv_fw + MV_BWD_OFFS;
405 int den = h->direct_den[col_mv->ref];
406 int m = col_mv->x >> 31;
408 pmv_fw->dist = h->dist[1];
409 pmv_bw->dist = h->dist[0];
412 /* scale the co-located motion vector according to its temporal span */
413 pmv_fw->x = (((den+(den*col_mv->x*pmv_fw->dist^m)-m-1)>>14)^m)-m;
414 pmv_bw->x = m-(((den+(den*col_mv->x*pmv_bw->dist^m)-m-1)>>14)^m);
416 pmv_fw->y = (((den+(den*col_mv->y*pmv_fw->dist^m)-m-1)>>14)^m)-m;
417 pmv_bw->y = m-(((den+(den*col_mv->y*pmv_bw->dist^m)-m-1)>>14)^m);
420 static inline void mv_pred_sym(AVSContext *h, vector_t *src, enum block_t size) {
421 vector_t *dst = src + MV_BWD_OFFS;
423 /* backward mv is the scaled and negated forward mv */
424 dst->x = -((src->x * h->sym_factor + 256) >> 9);
425 dst->y = -((src->y * h->sym_factor + 256) >> 9);
427 dst->dist = h->dist[0];
431 static void mv_pred(AVSContext *h, enum mv_loc_t nP, enum mv_loc_t nC,
432 enum mv_pred_t mode, enum block_t size, int ref) {
433 vector_t *mvP = &h->mv[nP];
434 vector_t *mvA = &h->mv[nP-1];
435 vector_t *mvB = &h->mv[nP-4];
436 vector_t *mvC = &h->mv[nC];
437 const vector_t *mvP2 = NULL;
440 mvP->dist = h->dist[mvP->ref];
441 if(mvC->ref == NOT_AVAIL)
442 mvC = &h->mv[nP-5]; // set to top-left (mvD)
443 if((mode == MV_PRED_PSKIP) &&
444 ((mvA->ref == NOT_AVAIL) || (mvB->ref == NOT_AVAIL) ||
445 ((mvA->x | mvA->y | mvA->ref) == 0) ||
446 ((mvB->x | mvB->y | mvB->ref) == 0) )) {
447 mvP2 = &ff_cavs_un_mv;
448 /* if there is only one suitable candidate, take it */
449 } else if((mvA->ref >= 0) && (mvB->ref < 0) && (mvC->ref < 0)) {
451 } else if((mvA->ref < 0) && (mvB->ref >= 0) && (mvC->ref < 0)) {
453 } else if((mvA->ref < 0) && (mvB->ref < 0) && (mvC->ref >= 0)) {
455 } else if(mode == MV_PRED_LEFT && mvA->ref == ref){
457 } else if(mode == MV_PRED_TOP && mvB->ref == ref){
459 } else if(mode == MV_PRED_TOPRIGHT && mvC->ref == ref){
466 mv_pred_median(h, mvP, mvA, mvB, mvC);
468 if(mode < MV_PRED_PSKIP) {
469 mvP->x += get_se_golomb(&h->s.gb);
470 mvP->y += get_se_golomb(&h->s.gb);
475 /*****************************************************************************
477 * residual data decoding
479 ****************************************************************************/
481 /** kth-order exponential golomb code */
482 static inline int get_ue_code(GetBitContext *gb, int order) {
484 int ret = get_ue_golomb(gb) << order;
485 return ret + get_bits(gb,order);
487 return get_ue_golomb(gb);
491 * decode coefficients from one 8x8 block, dequantize, inverse transform
492 * and add them to sample block
493 * @param r pointer to 2D VLC table
494 * @param esc_golomb_order escape codes are k-golomb with this order k
495 * @param qp quantizer
496 * @param dst location of sample block
497 * @param stride line stride in frame buffer
499 static int decode_residual_block(AVSContext *h, GetBitContext *gb,
500 const dec_2dvlc_t *r, int esc_golomb_order,
501 int qp, uint8_t *dst, int stride) {
502 int i, level_code, esc_code, level, run, mask;
503 DCTELEM level_buf[64];
505 DCTELEM *block = h->block;
508 level_code = get_ue_code(gb,r->golomb_order);
509 if(level_code >= ESCAPE_CODE) {
510 run = ((level_code - ESCAPE_CODE) >> 1) + 1;
511 esc_code = get_ue_code(gb,esc_golomb_order);
512 level = esc_code + (run > r->max_run ? 1 : r->level_add[run]);
513 while(level > r->inc_limit)
515 mask = -(level_code & 1);
516 level = (level^mask) - mask;
518 level = r->rltab[level_code][0];
519 if(!level) //end of block signal
521 run = r->rltab[level_code][1];
522 r += r->rltab[level_code][2];
524 level_buf[i] = level;
527 if(dequant(h,level_buf, run_buf, block, dequant_mul[qp],
528 dequant_shift[qp], i))
530 h->s.dsp.cavs_idct8_add(dst,block,stride);
535 static inline void decode_residual_chroma(AVSContext *h) {
537 decode_residual_block(h,&h->s.gb,chroma_dec,0, chroma_qp[h->qp],
540 decode_residual_block(h,&h->s.gb,chroma_dec,0, chroma_qp[h->qp],
544 static inline int decode_residual_inter(AVSContext *h) {
547 /* get coded block pattern */
548 int cbp= get_ue_golomb(&h->s.gb);
550 av_log(h->s.avctx, AV_LOG_ERROR, "illegal inter cbp\n");
553 h->cbp = cbp_tab[cbp][1];
556 if(h->cbp && !h->qp_fixed)
557 h->qp = (h->qp + get_se_golomb(&h->s.gb)) & 63;
558 for(block=0;block<4;block++)
559 if(h->cbp & (1<<block))
560 decode_residual_block(h,&h->s.gb,inter_dec,0,h->qp,
561 h->cy + h->luma_scan[block], h->l_stride);
562 decode_residual_chroma(h);
567 /*****************************************************************************
571 ****************************************************************************/
573 static int decode_mb_i(AVSContext *h, int cbp_code) {
574 GetBitContext *gb = &h->s.gb;
575 int block, pred_mode_uv;
577 uint8_t *left = NULL;
582 /* get intra prediction modes from stream */
583 for(block=0;block<4;block++) {
585 int pos = scan3x3[block];
587 nA = h->pred_mode_Y[pos-1];
588 nB = h->pred_mode_Y[pos-3];
589 predpred = FFMIN(nA,nB);
590 if(predpred == NOT_AVAIL) // if either is not available
591 predpred = INTRA_L_LP;
593 int rem_mode= get_bits(gb, 2);
594 predpred = rem_mode + (rem_mode >= predpred);
596 h->pred_mode_Y[pos] = predpred;
598 pred_mode_uv = get_ue_golomb(gb);
599 if(pred_mode_uv > 6) {
600 av_log(h->s.avctx, AV_LOG_ERROR, "illegal intra chroma pred mode\n");
603 modify_mb_i(h, &pred_mode_uv);
605 /* get coded block pattern */
606 if(h->pic_type == FF_I_TYPE)
607 cbp_code = get_ue_golomb(gb);
609 av_log(h->s.avctx, AV_LOG_ERROR, "illegal intra cbp\n");
612 h->cbp = cbp_tab[cbp_code][0];
613 if(h->cbp && !h->qp_fixed)
614 h->qp = (h->qp + get_se_golomb(gb)) & 63; //qp_delta
616 /* luma intra prediction interleaved with residual decode/transform/add */
617 for(block=0;block<4;block++) {
618 d = h->cy + h->luma_scan[block];
619 load_intra_pred_luma(h, top, &left, block);
620 h->intra_pred_l[h->pred_mode_Y[scan3x3[block]]]
621 (d, top, left, h->l_stride);
622 if(h->cbp & (1<<block))
623 decode_residual_block(h,gb,intra_dec,1,h->qp,d,h->l_stride);
626 /* chroma intra prediction */
627 load_intra_pred_chroma(h);
628 h->intra_pred_c[pred_mode_uv](h->cu, &h->top_border_u[h->mbx*10],
629 h->left_border_u, h->c_stride);
630 h->intra_pred_c[pred_mode_uv](h->cv, &h->top_border_v[h->mbx*10],
631 h->left_border_v, h->c_stride);
633 decode_residual_chroma(h);
639 static void decode_mb_p(AVSContext *h, enum mb_t mb_type) {
640 GetBitContext *gb = &h->s.gb;
646 mv_pred(h, MV_FWD_X0, MV_FWD_C2, MV_PRED_PSKIP, BLK_16X16, 0);
649 ref[0] = h->ref_flag ? 0 : get_bits1(gb);
650 mv_pred(h, MV_FWD_X0, MV_FWD_C2, MV_PRED_MEDIAN, BLK_16X16,ref[0]);
653 ref[0] = h->ref_flag ? 0 : get_bits1(gb);
654 ref[2] = h->ref_flag ? 0 : get_bits1(gb);
655 mv_pred(h, MV_FWD_X0, MV_FWD_C2, MV_PRED_TOP, BLK_16X8, ref[0]);
656 mv_pred(h, MV_FWD_X2, MV_FWD_A1, MV_PRED_LEFT, BLK_16X8, ref[2]);
659 ref[0] = h->ref_flag ? 0 : get_bits1(gb);
660 ref[1] = h->ref_flag ? 0 : get_bits1(gb);
661 mv_pred(h, MV_FWD_X0, MV_FWD_B3, MV_PRED_LEFT, BLK_8X16, ref[0]);
662 mv_pred(h, MV_FWD_X1, MV_FWD_C2, MV_PRED_TOPRIGHT, BLK_8X16, ref[1]);
665 ref[0] = h->ref_flag ? 0 : get_bits1(gb);
666 ref[1] = h->ref_flag ? 0 : get_bits1(gb);
667 ref[2] = h->ref_flag ? 0 : get_bits1(gb);
668 ref[3] = h->ref_flag ? 0 : get_bits1(gb);
669 mv_pred(h, MV_FWD_X0, MV_FWD_B3, MV_PRED_MEDIAN, BLK_8X8, ref[0]);
670 mv_pred(h, MV_FWD_X1, MV_FWD_C2, MV_PRED_MEDIAN, BLK_8X8, ref[1]);
671 mv_pred(h, MV_FWD_X2, MV_FWD_X1, MV_PRED_MEDIAN, BLK_8X8, ref[2]);
672 mv_pred(h, MV_FWD_X3, MV_FWD_X0, MV_PRED_MEDIAN, BLK_8X8, ref[3]);
674 inter_pred(h, mb_type);
675 set_intra_mode_default(h);
677 if(mb_type != P_SKIP)
678 decode_residual_inter(h);
679 filter_mb(h,mb_type);
680 *h->col_type = mb_type;
683 static void decode_mb_b(AVSContext *h, enum mb_t mb_type) {
685 enum sub_mb_t sub_type[4];
691 h->mv[MV_FWD_X0] = ff_cavs_dir_mv;
692 set_mvs(&h->mv[MV_FWD_X0], BLK_16X16);
693 h->mv[MV_BWD_X0] = ff_cavs_dir_mv;
694 set_mvs(&h->mv[MV_BWD_X0], BLK_16X16);
698 if(!(*h->col_type)) {
699 /* intra MB at co-location, do in-plane prediction */
700 mv_pred(h, MV_FWD_X0, MV_FWD_C2, MV_PRED_BSKIP, BLK_16X16, 1);
701 mv_pred(h, MV_BWD_X0, MV_BWD_C2, MV_PRED_BSKIP, BLK_16X16, 0);
703 /* direct prediction from co-located P MB, block-wise */
704 for(block=0;block<4;block++)
705 mv_pred_direct(h,&h->mv[mv_scan[block]],
706 &h->col_mv[(h->mby*h->mb_width+h->mbx)*4 + block]);
709 mv_pred(h, MV_FWD_X0, MV_FWD_C2, MV_PRED_MEDIAN, BLK_16X16, 1);
712 mv_pred(h, MV_FWD_X0, MV_FWD_C2, MV_PRED_MEDIAN, BLK_16X16, 1);
713 mv_pred_sym(h, &h->mv[MV_FWD_X0], BLK_16X16);
716 mv_pred(h, MV_BWD_X0, MV_BWD_C2, MV_PRED_MEDIAN, BLK_16X16, 0);
719 for(block=0;block<4;block++)
720 sub_type[block] = get_bits(&h->s.gb,2);
721 for(block=0;block<4;block++) {
722 switch(sub_type[block]) {
724 if(!(*h->col_type)) {
725 /* intra MB at co-location, do in-plane prediction */
726 mv_pred(h, mv_scan[block], mv_scan[block]-3,
727 MV_PRED_BSKIP, BLK_8X8, 1);
728 mv_pred(h, mv_scan[block]+MV_BWD_OFFS,
729 mv_scan[block]-3+MV_BWD_OFFS,
730 MV_PRED_BSKIP, BLK_8X8, 0);
732 mv_pred_direct(h,&h->mv[mv_scan[block]],
733 &h->col_mv[(h->mby*h->mb_width + h->mbx)*4 + block]);
736 mv_pred(h, mv_scan[block], mv_scan[block]-3,
737 MV_PRED_MEDIAN, BLK_8X8, 1);
740 mv_pred(h, mv_scan[block], mv_scan[block]-3,
741 MV_PRED_MEDIAN, BLK_8X8, 1);
742 mv_pred_sym(h, &h->mv[mv_scan[block]], BLK_8X8);
746 for(block=0;block<4;block++) {
747 if(sub_type[block] == B_SUB_BWD)
748 mv_pred(h, mv_scan[block]+MV_BWD_OFFS,
749 mv_scan[block]+MV_BWD_OFFS-3,
750 MV_PRED_MEDIAN, BLK_8X8, 0);
754 assert((mb_type > B_SYM_16X16) && (mb_type < B_8X8));
755 flags = partition_flags[mb_type];
756 if(mb_type & 1) { /* 16x8 macroblock types */
758 mv_pred(h, MV_FWD_X0, MV_FWD_C2, MV_PRED_TOP, BLK_16X8, 1);
760 mv_pred_sym(h, &h->mv[MV_FWD_X0], BLK_16X8);
762 mv_pred(h, MV_FWD_X2, MV_FWD_A1, MV_PRED_LEFT, BLK_16X8, 1);
764 mv_pred_sym(h, &h->mv[MV_FWD_X2], BLK_16X8);
766 mv_pred(h, MV_BWD_X0, MV_BWD_C2, MV_PRED_TOP, BLK_16X8, 0);
768 mv_pred(h, MV_BWD_X2, MV_BWD_A1, MV_PRED_LEFT, BLK_16X8, 0);
769 } else { /* 8x16 macroblock types */
771 mv_pred(h, MV_FWD_X0, MV_FWD_B3, MV_PRED_LEFT, BLK_8X16, 1);
773 mv_pred_sym(h, &h->mv[MV_FWD_X0], BLK_8X16);
775 mv_pred(h, MV_FWD_X1, MV_FWD_C2, MV_PRED_TOPRIGHT,BLK_8X16, 1);
777 mv_pred_sym(h, &h->mv[MV_FWD_X1], BLK_8X16);
779 mv_pred(h, MV_BWD_X0, MV_BWD_B3, MV_PRED_LEFT, BLK_8X16, 0);
781 mv_pred(h, MV_BWD_X1, MV_BWD_C2, MV_PRED_TOPRIGHT,BLK_8X16, 0);
784 inter_pred(h, mb_type);
785 set_intra_mode_default(h);
786 if(mb_type != B_SKIP)
787 decode_residual_inter(h);
788 filter_mb(h,mb_type);
791 /*****************************************************************************
795 ****************************************************************************/
797 static inline int decode_slice_header(AVSContext *h, GetBitContext *gb) {
799 av_log(h->s.avctx, AV_LOG_ERROR, "unexpected start code 0x%02x\n", h->stc);
801 if((h->mby == 0) && (!h->qp_fixed)){
802 h->qp_fixed = get_bits1(gb);
803 h->qp = get_bits(gb,6);
805 /* inter frame or second slice can have weighting params */
806 if((h->pic_type != FF_I_TYPE) || (!h->pic_structure && h->mby >= h->mb_width/2))
807 if(get_bits1(gb)) { //slice_weighting_flag
808 av_log(h->s.avctx, AV_LOG_ERROR,
809 "weighted prediction not yet supported\n");
814 static inline void check_for_slice(AVSContext *h) {
815 GetBitContext *gb = &h->s.gb;
817 align = (-get_bits_count(gb)) & 7;
818 if((show_bits_long(gb,24+align) & 0xFFFFFF) == 0x000001) {
819 get_bits_long(gb,24+align);
820 h->stc = get_bits(gb,8);
821 decode_slice_header(h,gb);
825 /*****************************************************************************
829 ****************************************************************************/
831 static void init_pic(AVSContext *h) {
834 /* clear some predictors */
836 h->mv[i] = ff_cavs_un_mv;
837 h->mv[MV_BWD_X0] = ff_cavs_dir_mv;
838 set_mvs(&h->mv[MV_BWD_X0], BLK_16X16);
839 h->mv[MV_FWD_X0] = ff_cavs_dir_mv;
840 set_mvs(&h->mv[MV_FWD_X0], BLK_16X16);
841 h->pred_mode_Y[3] = h->pred_mode_Y[6] = NOT_AVAIL;
842 h->cy = h->picture.data[0];
843 h->cu = h->picture.data[1];
844 h->cv = h->picture.data[2];
845 h->l_stride = h->picture.linesize[0];
846 h->c_stride = h->picture.linesize[1];
847 h->luma_scan[2] = 8*h->l_stride;
848 h->luma_scan[3] = 8*h->l_stride+8;
853 static int decode_pic(AVSContext *h) {
854 MpegEncContext *s = &h->s;
858 if (!s->context_initialized) {
859 s->avctx->idct_algo = FF_IDCT_CAVS;
860 if (MPV_common_init(s) < 0)
862 ff_init_scantable(s->dsp.idct_permutation,&h->scantable,ff_zigzag_direct);
864 get_bits(&s->gb,16);//bbv_dwlay
865 if(h->stc == PIC_PB_START_CODE) {
866 h->pic_type = get_bits(&s->gb,2) + FF_I_TYPE;
867 if(h->pic_type > FF_B_TYPE) {
868 av_log(s->avctx, AV_LOG_ERROR, "illegal picture type\n");
871 /* make sure we have the reference frames we need */
872 if(!h->DPB[0].data[0] ||
873 (!h->DPB[1].data[0] && h->pic_type == FF_B_TYPE))
876 h->pic_type = FF_I_TYPE;
877 if(get_bits1(&s->gb))
878 get_bits(&s->gb,16);//time_code
880 /* release last B frame */
881 if(h->picture.data[0])
882 s->avctx->release_buffer(s->avctx, (AVFrame *)&h->picture);
884 s->avctx->get_buffer(s->avctx, (AVFrame *)&h->picture);
886 h->picture.poc = get_bits(&s->gb,8)*2;
888 /* get temporal distances and MV scaling factors */
889 if(h->pic_type != FF_B_TYPE) {
890 h->dist[0] = (h->picture.poc - h->DPB[0].poc + 512) % 512;
892 h->dist[0] = (h->DPB[0].poc - h->picture.poc + 512) % 512;
894 h->dist[1] = (h->picture.poc - h->DPB[1].poc + 512) % 512;
895 h->scale_den[0] = h->dist[0] ? 512/h->dist[0] : 0;
896 h->scale_den[1] = h->dist[1] ? 512/h->dist[1] : 0;
897 if(h->pic_type == FF_B_TYPE) {
898 h->sym_factor = h->dist[0]*h->scale_den[1];
900 h->direct_den[0] = h->dist[0] ? 16384/h->dist[0] : 0;
901 h->direct_den[1] = h->dist[1] ? 16384/h->dist[1] : 0;
905 get_ue_golomb(&s->gb); //bbv_check_times
906 h->progressive = get_bits1(&s->gb);
908 h->pic_structure = 1;
909 else if(!(h->pic_structure = get_bits1(&s->gb) && (h->stc == PIC_PB_START_CODE)) )
910 get_bits1(&s->gb); //advanced_pred_mode_disable
911 skip_bits1(&s->gb); //top_field_first
912 skip_bits1(&s->gb); //repeat_first_field
913 h->qp_fixed = get_bits1(&s->gb);
914 h->qp = get_bits(&s->gb,6);
915 if(h->pic_type == FF_I_TYPE) {
916 if(!h->progressive && !h->pic_structure)
917 skip_bits1(&s->gb);//what is this?
918 skip_bits(&s->gb,4); //reserved bits
920 if(!(h->pic_type == FF_B_TYPE && h->pic_structure == 1))
921 h->ref_flag = get_bits1(&s->gb);
922 skip_bits(&s->gb,4); //reserved bits
923 h->skip_mode_flag = get_bits1(&s->gb);
925 h->loop_filter_disable = get_bits1(&s->gb);
926 if(!h->loop_filter_disable && get_bits1(&s->gb)) {
927 h->alpha_offset = get_se_golomb(&s->gb);
928 h->beta_offset = get_se_golomb(&s->gb);
930 h->alpha_offset = h->beta_offset = 0;
933 if(h->pic_type == FF_I_TYPE) {
937 } else if(h->pic_type == FF_P_TYPE) {
939 if(h->skip_mode_flag) {
940 skip_count = get_ue_golomb(&s->gb);
941 while(skip_count--) {
942 decode_mb_p(h,P_SKIP);
946 mb_type = get_ue_golomb(&s->gb) + P_16X16;
948 mb_type = get_ue_golomb(&s->gb) + P_SKIP;
949 if(mb_type > P_8X8) {
950 decode_mb_i(h, mb_type - P_8X8 - 1);
952 decode_mb_p(h,mb_type);
954 } else { /* FF_B_TYPE */
956 if(h->skip_mode_flag) {
957 skip_count = get_ue_golomb(&s->gb);
958 while(skip_count--) {
959 decode_mb_b(h,B_SKIP);
963 mb_type = get_ue_golomb(&s->gb) + B_DIRECT;
965 mb_type = get_ue_golomb(&s->gb) + B_SKIP;
966 if(mb_type > B_8X8) {
967 decode_mb_i(h, mb_type - B_8X8 - 1);
969 decode_mb_b(h,mb_type);
973 if(h->pic_type != FF_B_TYPE) {
974 if(h->DPB[1].data[0])
975 s->avctx->release_buffer(s->avctx, (AVFrame *)&h->DPB[1]);
976 memcpy(&h->DPB[1], &h->DPB[0], sizeof(Picture));
977 memcpy(&h->DPB[0], &h->picture, sizeof(Picture));
978 memset(&h->picture,0,sizeof(Picture));
983 /*****************************************************************************
985 * headers and interface
987 ****************************************************************************/
990 * some predictions require data from the top-neighbouring macroblock.
991 * this data has to be stored for one complete row of macroblocks
992 * and this storage space is allocated here
994 static void init_top_lines(AVSContext *h) {
995 /* alloc top line of predictors */
996 h->top_qp = av_malloc( h->mb_width);
997 h->top_mv[0] = av_malloc((h->mb_width*2+1)*sizeof(vector_t));
998 h->top_mv[1] = av_malloc((h->mb_width*2+1)*sizeof(vector_t));
999 h->top_pred_Y = av_malloc( h->mb_width*2*sizeof(*h->top_pred_Y));
1000 h->top_border_y = av_malloc((h->mb_width+1)*16);
1001 h->top_border_u = av_malloc((h->mb_width)*10);
1002 h->top_border_v = av_malloc((h->mb_width)*10);
1004 /* alloc space for co-located MVs and types */
1005 h->col_mv = av_malloc( h->mb_width*h->mb_height*4*sizeof(vector_t));
1006 h->col_type_base = av_malloc(h->mb_width*h->mb_height);
1007 h->block = av_mallocz(64*sizeof(DCTELEM));
1010 static int decode_seq_header(AVSContext *h) {
1011 MpegEncContext *s = &h->s;
1012 int frame_rate_code;
1014 h->profile = get_bits(&s->gb,8);
1015 h->level = get_bits(&s->gb,8);
1016 skip_bits1(&s->gb); //progressive sequence
1017 s->width = get_bits(&s->gb,14);
1018 s->height = get_bits(&s->gb,14);
1019 skip_bits(&s->gb,2); //chroma format
1020 skip_bits(&s->gb,3); //sample_precision
1021 h->aspect_ratio = get_bits(&s->gb,4);
1022 frame_rate_code = get_bits(&s->gb,4);
1023 skip_bits(&s->gb,18);//bit_rate_lower
1024 skip_bits1(&s->gb); //marker_bit
1025 skip_bits(&s->gb,12);//bit_rate_upper
1026 s->low_delay = get_bits1(&s->gb);
1027 h->mb_width = (s->width + 15) >> 4;
1028 h->mb_height = (s->height + 15) >> 4;
1029 h->s.avctx->time_base.den = ff_frame_rate_tab[frame_rate_code].num;
1030 h->s.avctx->time_base.num = ff_frame_rate_tab[frame_rate_code].den;
1031 h->s.avctx->width = s->width;
1032 h->s.avctx->height = s->height;
1038 static void cavs_flush(AVCodecContext * avctx) {
1039 AVSContext *h = avctx->priv_data;
1040 h->got_keyframe = 0;
1043 static int cavs_decode_frame(AVCodecContext * avctx,void *data, int *data_size,
1044 uint8_t * buf, int buf_size) {
1045 AVSContext *h = avctx->priv_data;
1046 MpegEncContext *s = &h->s;
1048 const uint8_t *buf_end;
1049 const uint8_t *buf_ptr;
1050 AVFrame *picture = data;
1055 if (buf_size == 0) {
1056 if(!s->low_delay && h->DPB[0].data[0]) {
1057 *data_size = sizeof(AVPicture);
1058 *picture = *(AVFrame *) &h->DPB[0];
1064 buf_end = buf + buf_size;
1066 buf_ptr = ff_find_start_code(buf_ptr,buf_end, &stc);
1067 if(stc & 0xFFFFFE00)
1068 return FFMAX(0, buf_ptr - buf - s->parse_context.last_index);
1069 input_size = (buf_end - buf_ptr)*8;
1071 case CAVS_START_CODE:
1072 init_get_bits(&s->gb, buf_ptr, input_size);
1073 decode_seq_header(h);
1075 case PIC_I_START_CODE:
1076 if(!h->got_keyframe) {
1077 if(h->DPB[0].data[0])
1078 avctx->release_buffer(avctx, (AVFrame *)&h->DPB[0]);
1079 if(h->DPB[1].data[0])
1080 avctx->release_buffer(avctx, (AVFrame *)&h->DPB[1]);
1081 h->got_keyframe = 1;
1083 case PIC_PB_START_CODE:
1085 if(!h->got_keyframe)
1087 init_get_bits(&s->gb, buf_ptr, input_size);
1091 *data_size = sizeof(AVPicture);
1092 if(h->pic_type != FF_B_TYPE) {
1093 if(h->DPB[1].data[0]) {
1094 *picture = *(AVFrame *) &h->DPB[1];
1099 *picture = *(AVFrame *) &h->picture;
1101 case EXT_START_CODE:
1102 //mpeg_decode_extension(avctx,buf_ptr, input_size);
1104 case USER_START_CODE:
1105 //mpeg_decode_user_data(avctx,buf_ptr, input_size);
1108 if (stc >= SLICE_MIN_START_CODE &&
1109 stc <= SLICE_MAX_START_CODE) {
1110 init_get_bits(&s->gb, buf_ptr, input_size);
1111 decode_slice_header(h, &s->gb);
1118 static int cavs_decode_init(AVCodecContext * avctx) {
1119 AVSContext *h = avctx->priv_data;
1120 MpegEncContext * const s = &h->s;
1122 MPV_decode_defaults(s);
1125 avctx->pix_fmt= PIX_FMT_YUV420P;
1127 h->luma_scan[0] = 0;
1128 h->luma_scan[1] = 8;
1129 h->intra_pred_l[ INTRA_L_VERT] = intra_pred_vert;
1130 h->intra_pred_l[ INTRA_L_HORIZ] = intra_pred_horiz;
1131 h->intra_pred_l[ INTRA_L_LP] = intra_pred_lp;
1132 h->intra_pred_l[ INTRA_L_DOWN_LEFT] = intra_pred_down_left;
1133 h->intra_pred_l[INTRA_L_DOWN_RIGHT] = intra_pred_down_right;
1134 h->intra_pred_l[ INTRA_L_LP_LEFT] = intra_pred_lp_left;
1135 h->intra_pred_l[ INTRA_L_LP_TOP] = intra_pred_lp_top;
1136 h->intra_pred_l[ INTRA_L_DC_128] = intra_pred_dc_128;
1137 h->intra_pred_c[ INTRA_C_LP] = intra_pred_lp;
1138 h->intra_pred_c[ INTRA_C_HORIZ] = intra_pred_horiz;
1139 h->intra_pred_c[ INTRA_C_VERT] = intra_pred_vert;
1140 h->intra_pred_c[ INTRA_C_PLANE] = intra_pred_plane;
1141 h->intra_pred_c[ INTRA_C_LP_LEFT] = intra_pred_lp_left;
1142 h->intra_pred_c[ INTRA_C_LP_TOP] = intra_pred_lp_top;
1143 h->intra_pred_c[ INTRA_C_DC_128] = intra_pred_dc_128;
1144 h->mv[ 7] = ff_cavs_un_mv;
1145 h->mv[19] = ff_cavs_un_mv;
1149 static int cavs_decode_end(AVCodecContext * avctx) {
1150 AVSContext *h = avctx->priv_data;
1153 av_free(h->top_mv[0]);
1154 av_free(h->top_mv[1]);
1155 av_free(h->top_pred_Y);
1156 av_free(h->top_border_y);
1157 av_free(h->top_border_u);
1158 av_free(h->top_border_v);
1160 av_free(h->col_type_base);
1165 AVCodec cavs_decoder = {
1174 CODEC_CAP_DR1 | CODEC_CAP_DELAY,