2 * H.26L/H.264/AVC/JVT/14496-10/... encoder/decoder
3 * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
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
23 * @file libavcodec/h264.c
24 * H.264 / AVC / MPEG4 part10 codec.
25 * @author Michael Niedermayer <michaelni@gmx.at>
31 #include "mpegvideo.h"
34 #include "h264_mvpred.h"
35 #include "h264_parser.h"
38 #include "rectangle.h"
39 #include "vdpau_internal.h"
46 static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
47 static void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
49 static const uint8_t rem6[52]={
50 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3,
53 static const uint8_t div6[52]={
54 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 4, 4, 4, 4, 4, 4, 5, 5, 5, 5, 5, 5, 6, 6, 6, 6, 6, 6, 7, 7, 7, 7, 7, 7, 8, 8, 8, 8,
57 void ff_h264_write_back_intra_pred_mode(H264Context *h){
58 const int mb_xy= h->mb_xy;
60 h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
61 h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
62 h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
63 h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
64 h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
65 h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
66 h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
70 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
72 int ff_h264_check_intra4x4_pred_mode(H264Context *h){
73 MpegEncContext * const s = &h->s;
74 static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
75 static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
78 if(!(h->top_samples_available&0x8000)){
80 int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
82 av_log(h->s.avctx, AV_LOG_ERROR, "top block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y);
85 h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
90 if((h->left_samples_available&0x8888)!=0x8888){
91 static const int mask[4]={0x8000,0x2000,0x80,0x20};
93 if(!(h->left_samples_available&mask[i])){
94 int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
96 av_log(h->s.avctx, AV_LOG_ERROR, "left block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y);
99 h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
106 } //FIXME cleanup like ff_h264_check_intra_pred_mode
109 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
111 int ff_h264_check_intra_pred_mode(H264Context *h, int mode){
112 MpegEncContext * const s = &h->s;
113 static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
114 static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
117 av_log(h->s.avctx, AV_LOG_ERROR, "out of range intra chroma pred mode at %d %d\n", s->mb_x, s->mb_y);
121 if(!(h->top_samples_available&0x8000)){
124 av_log(h->s.avctx, AV_LOG_ERROR, "top block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);
129 if((h->left_samples_available&0x8080) != 0x8080){
131 if(h->left_samples_available&0x8080){ //mad cow disease mode, aka MBAFF + constrained_intra_pred
132 mode= ALZHEIMER_DC_L0T_PRED8x8 + (!(h->left_samples_available&0x8000)) + 2*(mode == DC_128_PRED8x8);
135 av_log(h->s.avctx, AV_LOG_ERROR, "left block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);
143 const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src, int *dst_length, int *consumed, int length){
148 // src[0]&0x80; //forbidden bit
149 h->nal_ref_idc= src[0]>>5;
150 h->nal_unit_type= src[0]&0x1F;
154 for(i=0; i<length; i++)
155 printf("%2X ", src[i]);
158 #if HAVE_FAST_UNALIGNED
161 for(i=0; i+1<length; i+=9){
162 if(!((~*(const uint64_t*)(src+i) & (*(const uint64_t*)(src+i) - 0x0100010001000101ULL)) & 0x8000800080008080ULL))
165 for(i=0; i+1<length; i+=5){
166 if(!((~*(const uint32_t*)(src+i) & (*(const uint32_t*)(src+i) - 0x01000101U)) & 0x80008080U))
169 if(i>0 && !src[i]) i--;
173 for(i=0; i+1<length; i+=2){
175 if(i>0 && src[i-1]==0) i--;
177 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
179 /* startcode, so we must be past the end */
187 if(i>=length-1){ //no escaped 0
189 *consumed= length+1; //+1 for the header
193 bufidx = h->nal_unit_type == NAL_DPC ? 1 : 0; // use second escape buffer for inter data
194 av_fast_malloc(&h->rbsp_buffer[bufidx], &h->rbsp_buffer_size[bufidx], length+FF_INPUT_BUFFER_PADDING_SIZE);
195 dst= h->rbsp_buffer[bufidx];
201 //printf("decoding esc\n");
205 //remove escapes (very rare 1:2^22)
207 dst[di++]= src[si++];
208 dst[di++]= src[si++];
209 }else if(src[si]==0 && src[si+1]==0){
210 if(src[si+2]==3){ //escape
215 }else //next start code
219 dst[di++]= src[si++];
222 dst[di++]= src[si++];
225 memset(dst+di, 0, FF_INPUT_BUFFER_PADDING_SIZE);
228 *consumed= si + 1;//+1 for the header
229 //FIXME store exact number of bits in the getbitcontext (it is needed for decoding)
233 int ff_h264_decode_rbsp_trailing(H264Context *h, const uint8_t *src){
237 tprintf(h->s.avctx, "rbsp trailing %X\n", v);
247 * IDCT transforms the 16 dc values and dequantizes them.
248 * @param qp quantization parameter
250 static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
253 int temp[16]; //FIXME check if this is a good idea
254 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
255 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
257 //memset(block, 64, 2*256);
260 const int offset= y_offset[i];
261 const int z0= block[offset+stride*0] + block[offset+stride*4];
262 const int z1= block[offset+stride*0] - block[offset+stride*4];
263 const int z2= block[offset+stride*1] - block[offset+stride*5];
264 const int z3= block[offset+stride*1] + block[offset+stride*5];
273 const int offset= x_offset[i];
274 const int z0= temp[4*0+i] + temp[4*2+i];
275 const int z1= temp[4*0+i] - temp[4*2+i];
276 const int z2= temp[4*1+i] - temp[4*3+i];
277 const int z3= temp[4*1+i] + temp[4*3+i];
279 block[stride*0 +offset]= ((((z0 + z3)*qmul + 128 ) >> 8)); //FIXME think about merging this into decode_residual
280 block[stride*2 +offset]= ((((z1 + z2)*qmul + 128 ) >> 8));
281 block[stride*8 +offset]= ((((z1 - z2)*qmul + 128 ) >> 8));
282 block[stride*10+offset]= ((((z0 - z3)*qmul + 128 ) >> 8));
288 * DCT transforms the 16 dc values.
289 * @param qp quantization parameter ??? FIXME
291 static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
292 // const int qmul= dequant_coeff[qp][0];
294 int temp[16]; //FIXME check if this is a good idea
295 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
296 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
299 const int offset= y_offset[i];
300 const int z0= block[offset+stride*0] + block[offset+stride*4];
301 const int z1= block[offset+stride*0] - block[offset+stride*4];
302 const int z2= block[offset+stride*1] - block[offset+stride*5];
303 const int z3= block[offset+stride*1] + block[offset+stride*5];
312 const int offset= x_offset[i];
313 const int z0= temp[4*0+i] + temp[4*2+i];
314 const int z1= temp[4*0+i] - temp[4*2+i];
315 const int z2= temp[4*1+i] - temp[4*3+i];
316 const int z3= temp[4*1+i] + temp[4*3+i];
318 block[stride*0 +offset]= (z0 + z3)>>1;
319 block[stride*2 +offset]= (z1 + z2)>>1;
320 block[stride*8 +offset]= (z1 - z2)>>1;
321 block[stride*10+offset]= (z0 - z3)>>1;
329 static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
330 const int stride= 16*2;
331 const int xStride= 16;
334 a= block[stride*0 + xStride*0];
335 b= block[stride*0 + xStride*1];
336 c= block[stride*1 + xStride*0];
337 d= block[stride*1 + xStride*1];
344 block[stride*0 + xStride*0]= ((a+c)*qmul) >> 7;
345 block[stride*0 + xStride*1]= ((e+b)*qmul) >> 7;
346 block[stride*1 + xStride*0]= ((a-c)*qmul) >> 7;
347 block[stride*1 + xStride*1]= ((e-b)*qmul) >> 7;
351 static void chroma_dc_dct_c(DCTELEM *block){
352 const int stride= 16*2;
353 const int xStride= 16;
356 a= block[stride*0 + xStride*0];
357 b= block[stride*0 + xStride*1];
358 c= block[stride*1 + xStride*0];
359 d= block[stride*1 + xStride*1];
366 block[stride*0 + xStride*0]= (a+c);
367 block[stride*0 + xStride*1]= (e+b);
368 block[stride*1 + xStride*0]= (a-c);
369 block[stride*1 + xStride*1]= (e-b);
373 static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
374 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
375 int src_x_offset, int src_y_offset,
376 qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
377 MpegEncContext * const s = &h->s;
378 const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
379 int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
380 const int luma_xy= (mx&3) + ((my&3)<<2);
381 uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*h->mb_linesize;
382 uint8_t * src_cb, * src_cr;
383 int extra_width= h->emu_edge_width;
384 int extra_height= h->emu_edge_height;
386 const int full_mx= mx>>2;
387 const int full_my= my>>2;
388 const int pic_width = 16*s->mb_width;
389 const int pic_height = 16*s->mb_height >> MB_FIELD;
391 if(mx&7) extra_width -= 3;
392 if(my&7) extra_height -= 3;
394 if( full_mx < 0-extra_width
395 || full_my < 0-extra_height
396 || full_mx + 16/*FIXME*/ > pic_width + extra_width
397 || full_my + 16/*FIXME*/ > pic_height + extra_height){
398 ff_emulated_edge_mc(s->edge_emu_buffer, src_y - 2 - 2*h->mb_linesize, h->mb_linesize, 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
399 src_y= s->edge_emu_buffer + 2 + 2*h->mb_linesize;
403 qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); //FIXME try variable height perhaps?
405 qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
408 if(CONFIG_GRAY && s->flags&CODEC_FLAG_GRAY) return;
411 // chroma offset when predicting from a field of opposite parity
412 my += 2 * ((s->mb_y & 1) - (pic->reference - 1));
413 emu |= (my>>3) < 0 || (my>>3) + 8 >= (pic_height>>1);
415 src_cb= pic->data[1] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
416 src_cr= pic->data[2] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
419 ff_emulated_edge_mc(s->edge_emu_buffer, src_cb, h->mb_uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
420 src_cb= s->edge_emu_buffer;
422 chroma_op(dest_cb, src_cb, h->mb_uvlinesize, chroma_height, mx&7, my&7);
425 ff_emulated_edge_mc(s->edge_emu_buffer, src_cr, h->mb_uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
426 src_cr= s->edge_emu_buffer;
428 chroma_op(dest_cr, src_cr, h->mb_uvlinesize, chroma_height, mx&7, my&7);
431 static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
432 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
433 int x_offset, int y_offset,
434 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
435 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
436 int list0, int list1){
437 MpegEncContext * const s = &h->s;
438 qpel_mc_func *qpix_op= qpix_put;
439 h264_chroma_mc_func chroma_op= chroma_put;
441 dest_y += 2*x_offset + 2*y_offset*h-> mb_linesize;
442 dest_cb += x_offset + y_offset*h->mb_uvlinesize;
443 dest_cr += x_offset + y_offset*h->mb_uvlinesize;
444 x_offset += 8*s->mb_x;
445 y_offset += 8*(s->mb_y >> MB_FIELD);
448 Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
449 mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
450 dest_y, dest_cb, dest_cr, x_offset, y_offset,
454 chroma_op= chroma_avg;
458 Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
459 mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
460 dest_y, dest_cb, dest_cr, x_offset, y_offset,
465 static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
466 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
467 int x_offset, int y_offset,
468 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
469 h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
470 h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
471 int list0, int list1){
472 MpegEncContext * const s = &h->s;
474 dest_y += 2*x_offset + 2*y_offset*h-> mb_linesize;
475 dest_cb += x_offset + y_offset*h->mb_uvlinesize;
476 dest_cr += x_offset + y_offset*h->mb_uvlinesize;
477 x_offset += 8*s->mb_x;
478 y_offset += 8*(s->mb_y >> MB_FIELD);
481 /* don't optimize for luma-only case, since B-frames usually
482 * use implicit weights => chroma too. */
483 uint8_t *tmp_cb = s->obmc_scratchpad;
484 uint8_t *tmp_cr = s->obmc_scratchpad + 8;
485 uint8_t *tmp_y = s->obmc_scratchpad + 8*h->mb_uvlinesize;
486 int refn0 = h->ref_cache[0][ scan8[n] ];
487 int refn1 = h->ref_cache[1][ scan8[n] ];
489 mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
490 dest_y, dest_cb, dest_cr,
491 x_offset, y_offset, qpix_put, chroma_put);
492 mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
493 tmp_y, tmp_cb, tmp_cr,
494 x_offset, y_offset, qpix_put, chroma_put);
496 if(h->use_weight == 2){
497 int weight0 = h->implicit_weight[refn0][refn1];
498 int weight1 = 64 - weight0;
499 luma_weight_avg( dest_y, tmp_y, h-> mb_linesize, 5, weight0, weight1, 0);
500 chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, 5, weight0, weight1, 0);
501 chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, 5, weight0, weight1, 0);
503 luma_weight_avg(dest_y, tmp_y, h->mb_linesize, h->luma_log2_weight_denom,
504 h->luma_weight[0][refn0], h->luma_weight[1][refn1],
505 h->luma_offset[0][refn0] + h->luma_offset[1][refn1]);
506 chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
507 h->chroma_weight[0][refn0][0], h->chroma_weight[1][refn1][0],
508 h->chroma_offset[0][refn0][0] + h->chroma_offset[1][refn1][0]);
509 chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
510 h->chroma_weight[0][refn0][1], h->chroma_weight[1][refn1][1],
511 h->chroma_offset[0][refn0][1] + h->chroma_offset[1][refn1][1]);
514 int list = list1 ? 1 : 0;
515 int refn = h->ref_cache[list][ scan8[n] ];
516 Picture *ref= &h->ref_list[list][refn];
517 mc_dir_part(h, ref, n, square, chroma_height, delta, list,
518 dest_y, dest_cb, dest_cr, x_offset, y_offset,
519 qpix_put, chroma_put);
521 luma_weight_op(dest_y, h->mb_linesize, h->luma_log2_weight_denom,
522 h->luma_weight[list][refn], h->luma_offset[list][refn]);
523 if(h->use_weight_chroma){
524 chroma_weight_op(dest_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
525 h->chroma_weight[list][refn][0], h->chroma_offset[list][refn][0]);
526 chroma_weight_op(dest_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
527 h->chroma_weight[list][refn][1], h->chroma_offset[list][refn][1]);
532 static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
533 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
534 int x_offset, int y_offset,
535 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
536 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
537 h264_weight_func *weight_op, h264_biweight_func *weight_avg,
538 int list0, int list1){
539 if((h->use_weight==2 && list0 && list1
540 && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ] != 32))
542 mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
543 x_offset, y_offset, qpix_put, chroma_put,
544 weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
546 mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
547 x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
550 static inline void prefetch_motion(H264Context *h, int list){
551 /* fetch pixels for estimated mv 4 macroblocks ahead
552 * optimized for 64byte cache lines */
553 MpegEncContext * const s = &h->s;
554 const int refn = h->ref_cache[list][scan8[0]];
556 const int mx= (h->mv_cache[list][scan8[0]][0]>>2) + 16*s->mb_x + 8;
557 const int my= (h->mv_cache[list][scan8[0]][1]>>2) + 16*s->mb_y;
558 uint8_t **src= h->ref_list[list][refn].data;
559 int off= mx + (my + (s->mb_x&3)*4)*h->mb_linesize + 64;
560 s->dsp.prefetch(src[0]+off, s->linesize, 4);
561 off= (mx>>1) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + 64;
562 s->dsp.prefetch(src[1]+off, src[2]-src[1], 2);
566 static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
567 qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
568 qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
569 h264_weight_func *weight_op, h264_biweight_func *weight_avg){
570 MpegEncContext * const s = &h->s;
571 const int mb_xy= h->mb_xy;
572 const int mb_type= s->current_picture.mb_type[mb_xy];
574 assert(IS_INTER(mb_type));
576 prefetch_motion(h, 0);
578 if(IS_16X16(mb_type)){
579 mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
580 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
581 weight_op, weight_avg,
582 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
583 }else if(IS_16X8(mb_type)){
584 mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
585 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
586 &weight_op[1], &weight_avg[1],
587 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
588 mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
589 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
590 &weight_op[1], &weight_avg[1],
591 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
592 }else if(IS_8X16(mb_type)){
593 mc_part(h, 0, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 0, 0,
594 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
595 &weight_op[2], &weight_avg[2],
596 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
597 mc_part(h, 4, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 4, 0,
598 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
599 &weight_op[2], &weight_avg[2],
600 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
604 assert(IS_8X8(mb_type));
607 const int sub_mb_type= h->sub_mb_type[i];
609 int x_offset= (i&1)<<2;
610 int y_offset= (i&2)<<1;
612 if(IS_SUB_8X8(sub_mb_type)){
613 mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
614 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
615 &weight_op[3], &weight_avg[3],
616 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
617 }else if(IS_SUB_8X4(sub_mb_type)){
618 mc_part(h, n , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
619 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
620 &weight_op[4], &weight_avg[4],
621 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
622 mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
623 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
624 &weight_op[4], &weight_avg[4],
625 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
626 }else if(IS_SUB_4X8(sub_mb_type)){
627 mc_part(h, n , 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
628 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
629 &weight_op[5], &weight_avg[5],
630 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
631 mc_part(h, n+1, 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
632 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
633 &weight_op[5], &weight_avg[5],
634 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
637 assert(IS_SUB_4X4(sub_mb_type));
639 int sub_x_offset= x_offset + 2*(j&1);
640 int sub_y_offset= y_offset + (j&2);
641 mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
642 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
643 &weight_op[6], &weight_avg[6],
644 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
650 prefetch_motion(h, 1);
654 static void free_tables(H264Context *h){
657 av_freep(&h->intra4x4_pred_mode);
658 av_freep(&h->chroma_pred_mode_table);
659 av_freep(&h->cbp_table);
660 av_freep(&h->mvd_table[0]);
661 av_freep(&h->mvd_table[1]);
662 av_freep(&h->direct_table);
663 av_freep(&h->non_zero_count);
664 av_freep(&h->slice_table_base);
665 h->slice_table= NULL;
666 av_freep(&h->list_counts);
668 av_freep(&h->mb2b_xy);
669 av_freep(&h->mb2b8_xy);
671 for(i = 0; i < MAX_THREADS; i++) {
672 hx = h->thread_context[i];
674 av_freep(&hx->top_borders[1]);
675 av_freep(&hx->top_borders[0]);
676 av_freep(&hx->s.obmc_scratchpad);
677 av_freep(&hx->rbsp_buffer[1]);
678 av_freep(&hx->rbsp_buffer[0]);
679 hx->rbsp_buffer_size[0] = 0;
680 hx->rbsp_buffer_size[1] = 0;
681 if (i) av_freep(&h->thread_context[i]);
685 static void init_dequant8_coeff_table(H264Context *h){
687 const int transpose = (h->s.dsp.h264_idct8_add != ff_h264_idct8_add_c); //FIXME ugly
688 h->dequant8_coeff[0] = h->dequant8_buffer[0];
689 h->dequant8_coeff[1] = h->dequant8_buffer[1];
692 if(i && !memcmp(h->pps.scaling_matrix8[0], h->pps.scaling_matrix8[1], 64*sizeof(uint8_t))){
693 h->dequant8_coeff[1] = h->dequant8_buffer[0];
701 h->dequant8_coeff[i][q][transpose ? (x>>3)|((x&7)<<3) : x] =
702 ((uint32_t)dequant8_coeff_init[idx][ dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] *
703 h->pps.scaling_matrix8[i][x]) << shift;
708 static void init_dequant4_coeff_table(H264Context *h){
710 const int transpose = (h->s.dsp.h264_idct_add != ff_h264_idct_add_c); //FIXME ugly
712 h->dequant4_coeff[i] = h->dequant4_buffer[i];
714 if(!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i], 16*sizeof(uint8_t))){
715 h->dequant4_coeff[i] = h->dequant4_buffer[j];
723 int shift = div6[q] + 2;
726 h->dequant4_coeff[i][q][transpose ? (x>>2)|((x<<2)&0xF) : x] =
727 ((uint32_t)dequant4_coeff_init[idx][(x&1) + ((x>>2)&1)] *
728 h->pps.scaling_matrix4[i][x]) << shift;
733 static void init_dequant_tables(H264Context *h){
735 init_dequant4_coeff_table(h);
736 if(h->pps.transform_8x8_mode)
737 init_dequant8_coeff_table(h);
738 if(h->sps.transform_bypass){
741 h->dequant4_coeff[i][0][x] = 1<<6;
742 if(h->pps.transform_8x8_mode)
745 h->dequant8_coeff[i][0][x] = 1<<6;
750 int ff_h264_alloc_tables(H264Context *h){
751 MpegEncContext * const s = &h->s;
752 const int big_mb_num= s->mb_stride * (s->mb_height+1);
755 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->intra4x4_pred_mode, big_mb_num * 8 * sizeof(uint8_t), fail)
757 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->non_zero_count , big_mb_num * 32 * sizeof(uint8_t), fail)
758 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->slice_table_base , (big_mb_num+s->mb_stride) * sizeof(*h->slice_table_base), fail)
759 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->cbp_table, big_mb_num * sizeof(uint16_t), fail)
761 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t), fail)
762 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[0], 32*big_mb_num * sizeof(uint16_t), fail);
763 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[1], 32*big_mb_num * sizeof(uint16_t), fail);
764 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->direct_table, 32*big_mb_num * sizeof(uint8_t) , fail);
765 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->list_counts, big_mb_num * sizeof(uint8_t), fail)
767 memset(h->slice_table_base, -1, (big_mb_num+s->mb_stride) * sizeof(*h->slice_table_base));
768 h->slice_table= h->slice_table_base + s->mb_stride*2 + 1;
770 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2b_xy , big_mb_num * sizeof(uint32_t), fail);
771 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2b8_xy , big_mb_num * sizeof(uint32_t), fail);
772 for(y=0; y<s->mb_height; y++){
773 for(x=0; x<s->mb_width; x++){
774 const int mb_xy= x + y*s->mb_stride;
775 const int b_xy = 4*x + 4*y*h->b_stride;
776 const int b8_xy= 2*x + 2*y*h->b8_stride;
778 h->mb2b_xy [mb_xy]= b_xy;
779 h->mb2b8_xy[mb_xy]= b8_xy;
783 s->obmc_scratchpad = NULL;
785 if(!h->dequant4_coeff[0])
786 init_dequant_tables(h);
795 * Mimic alloc_tables(), but for every context thread.
797 static void clone_tables(H264Context *dst, H264Context *src){
798 dst->intra4x4_pred_mode = src->intra4x4_pred_mode;
799 dst->non_zero_count = src->non_zero_count;
800 dst->slice_table = src->slice_table;
801 dst->cbp_table = src->cbp_table;
802 dst->mb2b_xy = src->mb2b_xy;
803 dst->mb2b8_xy = src->mb2b8_xy;
804 dst->chroma_pred_mode_table = src->chroma_pred_mode_table;
805 dst->mvd_table[0] = src->mvd_table[0];
806 dst->mvd_table[1] = src->mvd_table[1];
807 dst->direct_table = src->direct_table;
808 dst->list_counts = src->list_counts;
810 dst->s.obmc_scratchpad = NULL;
811 ff_h264_pred_init(&dst->hpc, src->s.codec_id);
816 * Allocate buffers which are not shared amongst multiple threads.
818 static int context_init(H264Context *h){
819 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[0], h->s.mb_width * (16+8+8) * sizeof(uint8_t), fail)
820 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[1], h->s.mb_width * (16+8+8) * sizeof(uint8_t), fail)
824 return -1; // free_tables will clean up for us
827 static av_cold void common_init(H264Context *h){
828 MpegEncContext * const s = &h->s;
830 s->width = s->avctx->width;
831 s->height = s->avctx->height;
832 s->codec_id= s->avctx->codec->id;
834 ff_h264_pred_init(&h->hpc, s->codec_id);
836 h->dequant_coeff_pps= -1;
837 s->unrestricted_mv=1;
840 dsputil_init(&s->dsp, s->avctx); // needed so that idct permutation is known early
842 memset(h->pps.scaling_matrix4, 16, 6*16*sizeof(uint8_t));
843 memset(h->pps.scaling_matrix8, 16, 2*64*sizeof(uint8_t));
846 av_cold int ff_h264_decode_init(AVCodecContext *avctx){
847 H264Context *h= avctx->priv_data;
848 MpegEncContext * const s = &h->s;
850 MPV_decode_defaults(s);
855 s->out_format = FMT_H264;
856 s->workaround_bugs= avctx->workaround_bugs;
859 // s->decode_mb= ff_h263_decode_mb;
860 s->quarter_sample = 1;
861 if(!avctx->has_b_frames)
864 avctx->chroma_sample_location = AVCHROMA_LOC_LEFT;
866 ff_h264_decode_init_vlc();
868 if(avctx->extradata_size > 0 && avctx->extradata &&
869 *(char *)avctx->extradata == 1){
876 h->thread_context[0] = h;
877 h->outputed_poc = INT_MIN;
878 h->prev_poc_msb= 1<<16;
879 ff_h264_reset_sei(h);
880 if(avctx->codec_id == CODEC_ID_H264){
881 if(avctx->ticks_per_frame == 1){
882 s->avctx->time_base.den *=2;
884 avctx->ticks_per_frame = 2;
889 int ff_h264_frame_start(H264Context *h){
890 MpegEncContext * const s = &h->s;
893 if(MPV_frame_start(s, s->avctx) < 0)
895 ff_er_frame_start(s);
897 * MPV_frame_start uses pict_type to derive key_frame.
898 * This is incorrect for H.264; IDR markings must be used.
899 * Zero here; IDR markings per slice in frame or fields are ORed in later.
900 * See decode_nal_units().
902 s->current_picture_ptr->key_frame= 0;
903 s->current_picture_ptr->mmco_reset= 0;
905 assert(s->linesize && s->uvlinesize);
908 h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
909 h->block_offset[24+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
912 h->block_offset[16+i]=
913 h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
914 h->block_offset[24+16+i]=
915 h->block_offset[24+20+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
918 /* can't be in alloc_tables because linesize isn't known there.
919 * FIXME: redo bipred weight to not require extra buffer? */
920 for(i = 0; i < s->avctx->thread_count; i++)
921 if(!h->thread_context[i]->s.obmc_scratchpad)
922 h->thread_context[i]->s.obmc_scratchpad = av_malloc(16*2*s->linesize + 8*2*s->uvlinesize);
924 /* some macroblocks will be accessed before they're available */
925 if(FRAME_MBAFF || s->avctx->thread_count > 1)
926 memset(h->slice_table, -1, (s->mb_height*s->mb_stride-1) * sizeof(*h->slice_table));
928 // s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
930 // We mark the current picture as non-reference after allocating it, so
931 // that if we break out due to an error it can be released automatically
932 // in the next MPV_frame_start().
933 // SVQ3 as well as most other codecs have only last/next/current and thus
934 // get released even with set reference, besides SVQ3 and others do not
935 // mark frames as reference later "naturally".
936 if(s->codec_id != CODEC_ID_SVQ3)
937 s->current_picture_ptr->reference= 0;
939 s->current_picture_ptr->field_poc[0]=
940 s->current_picture_ptr->field_poc[1]= INT_MAX;
941 assert(s->current_picture_ptr->long_ref==0);
946 static inline void backup_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize, int simple){
947 MpegEncContext * const s = &h->s;
952 src_cb -= uvlinesize;
953 src_cr -= uvlinesize;
955 if(!simple && FRAME_MBAFF){
958 top_border = h->top_borders[0][s->mb_x];
959 AV_COPY128(top_border, src_y + 15*linesize);
960 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
961 AV_COPY64(top_border+16, src_cb+7*uvlinesize);
962 AV_COPY64(top_border+24, src_cr+7*uvlinesize);
971 top_border = h->top_borders[top_idx][s->mb_x];
972 // There are two lines saved, the line above the the top macroblock of a pair,
973 // and the line above the bottom macroblock
974 AV_COPY128(top_border, src_y + 16*linesize);
976 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
977 AV_COPY64(top_border+16, src_cb+8*uvlinesize);
978 AV_COPY64(top_border+24, src_cr+8*uvlinesize);
982 static inline void xchg_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize, int xchg, int simple){
983 MpegEncContext * const s = &h->s;
990 uint8_t *top_border_m1;
993 if(!simple && FRAME_MBAFF){
998 top_idx = MB_MBAFF ? 0 : 1;
1002 if(h->deblocking_filter == 2) {
1004 deblock_left = h->slice_table[mb_xy] == h->slice_table[mb_xy - 1];
1005 deblock_top = h->slice_table[mb_xy] == h->slice_table[h->top_mb_xy];
1007 deblock_left = (s->mb_x > 0);
1008 deblock_top = (s->mb_y > !!MB_FIELD);
1011 src_y -= linesize + 1;
1012 src_cb -= uvlinesize + 1;
1013 src_cr -= uvlinesize + 1;
1015 top_border_m1 = h->top_borders[top_idx][s->mb_x-1];
1016 top_border = h->top_borders[top_idx][s->mb_x];
1018 #define XCHG(a,b,xchg)\
1019 if (xchg) AV_SWAP64(b,a);\
1020 else AV_COPY64(b,a);
1024 XCHG(top_border_m1+8, src_y -7, 1);
1026 XCHG(top_border+0, src_y +1, xchg);
1027 XCHG(top_border+8, src_y +9, 1);
1028 if(s->mb_x+1 < s->mb_width){
1029 XCHG(h->top_borders[top_idx][s->mb_x+1], src_y +17, 1);
1033 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1036 XCHG(top_border_m1+16, src_cb -7, 1);
1037 XCHG(top_border_m1+24, src_cr -7, 1);
1039 XCHG(top_border+16, src_cb+1, 1);
1040 XCHG(top_border+24, src_cr+1, 1);
1045 static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
1046 MpegEncContext * const s = &h->s;
1047 const int mb_x= s->mb_x;
1048 const int mb_y= s->mb_y;
1049 const int mb_xy= h->mb_xy;
1050 const int mb_type= s->current_picture.mb_type[mb_xy];
1051 uint8_t *dest_y, *dest_cb, *dest_cr;
1052 int linesize, uvlinesize /*dct_offset*/;
1054 int *block_offset = &h->block_offset[0];
1055 const int transform_bypass = !simple && (s->qscale == 0 && h->sps.transform_bypass);
1056 /* is_h264 should always be true if SVQ3 is disabled. */
1057 const int is_h264 = !CONFIG_SVQ3_DECODER || simple || s->codec_id == CODEC_ID_H264;
1058 void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1059 void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
1061 dest_y = s->current_picture.data[0] + (mb_x + mb_y * s->linesize ) * 16;
1062 dest_cb = s->current_picture.data[1] + (mb_x + mb_y * s->uvlinesize) * 8;
1063 dest_cr = s->current_picture.data[2] + (mb_x + mb_y * s->uvlinesize) * 8;
1065 s->dsp.prefetch(dest_y + (s->mb_x&3)*4*s->linesize + 64, s->linesize, 4);
1066 s->dsp.prefetch(dest_cb + (s->mb_x&7)*s->uvlinesize + 64, dest_cr - dest_cb, 2);
1068 h->list_counts[mb_xy]= h->list_count;
1070 if (!simple && MB_FIELD) {
1071 linesize = h->mb_linesize = s->linesize * 2;
1072 uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
1073 block_offset = &h->block_offset[24];
1074 if(mb_y&1){ //FIXME move out of this function?
1075 dest_y -= s->linesize*15;
1076 dest_cb-= s->uvlinesize*7;
1077 dest_cr-= s->uvlinesize*7;
1081 for(list=0; list<h->list_count; list++){
1082 if(!USES_LIST(mb_type, list))
1084 if(IS_16X16(mb_type)){
1085 int8_t *ref = &h->ref_cache[list][scan8[0]];
1086 fill_rectangle(ref, 4, 4, 8, (16+*ref)^(s->mb_y&1), 1);
1088 for(i=0; i<16; i+=4){
1089 int ref = h->ref_cache[list][scan8[i]];
1091 fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, (16+ref)^(s->mb_y&1), 1);
1097 linesize = h->mb_linesize = s->linesize;
1098 uvlinesize = h->mb_uvlinesize = s->uvlinesize;
1099 // dct_offset = s->linesize * 16;
1102 if (!simple && IS_INTRA_PCM(mb_type)) {
1103 for (i=0; i<16; i++) {
1104 memcpy(dest_y + i* linesize, h->mb + i*8, 16);
1106 for (i=0; i<8; i++) {
1107 memcpy(dest_cb+ i*uvlinesize, h->mb + 128 + i*4, 8);
1108 memcpy(dest_cr+ i*uvlinesize, h->mb + 160 + i*4, 8);
1111 if(IS_INTRA(mb_type)){
1112 if(h->deblocking_filter)
1113 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1, simple);
1115 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1116 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
1117 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
1120 if(IS_INTRA4x4(mb_type)){
1121 if(simple || !s->encoding){
1122 if(IS_8x8DCT(mb_type)){
1123 if(transform_bypass){
1125 idct_add = s->dsp.add_pixels8;
1127 idct_dc_add = s->dsp.h264_idct8_dc_add;
1128 idct_add = s->dsp.h264_idct8_add;
1130 for(i=0; i<16; i+=4){
1131 uint8_t * const ptr= dest_y + block_offset[i];
1132 const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1133 if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1134 h->hpc.pred8x8l_add[dir](ptr, h->mb + i*16, linesize);
1136 const int nnz = h->non_zero_count_cache[ scan8[i] ];
1137 h->hpc.pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
1138 (h->topright_samples_available<<i)&0x4000, linesize);
1140 if(nnz == 1 && h->mb[i*16])
1141 idct_dc_add(ptr, h->mb + i*16, linesize);
1143 idct_add (ptr, h->mb + i*16, linesize);
1148 if(transform_bypass){
1150 idct_add = s->dsp.add_pixels4;
1152 idct_dc_add = s->dsp.h264_idct_dc_add;
1153 idct_add = s->dsp.h264_idct_add;
1155 for(i=0; i<16; i++){
1156 uint8_t * const ptr= dest_y + block_offset[i];
1157 const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1159 if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1160 h->hpc.pred4x4_add[dir](ptr, h->mb + i*16, linesize);
1164 if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
1165 const int topright_avail= (h->topright_samples_available<<i)&0x8000;
1166 assert(mb_y || linesize <= block_offset[i]);
1167 if(!topright_avail){
1168 tr= ptr[3 - linesize]*0x01010101;
1169 topright= (uint8_t*) &tr;
1171 topright= ptr + 4 - linesize;
1175 h->hpc.pred4x4[ dir ](ptr, topright, linesize);
1176 nnz = h->non_zero_count_cache[ scan8[i] ];
1179 if(nnz == 1 && h->mb[i*16])
1180 idct_dc_add(ptr, h->mb + i*16, linesize);
1182 idct_add (ptr, h->mb + i*16, linesize);
1184 svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
1191 h->hpc.pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
1193 if(!transform_bypass)
1194 h264_luma_dc_dequant_idct_c(h->mb, s->qscale, h->dequant4_coeff[0][s->qscale][0]);
1196 svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
1198 if(h->deblocking_filter)
1199 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0, simple);
1201 hl_motion(h, dest_y, dest_cb, dest_cr,
1202 s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
1203 s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
1204 s->dsp.weight_h264_pixels_tab, s->dsp.biweight_h264_pixels_tab);
1208 if(!IS_INTRA4x4(mb_type)){
1210 if(IS_INTRA16x16(mb_type)){
1211 if(transform_bypass){
1212 if(h->sps.profile_idc==244 && (h->intra16x16_pred_mode==VERT_PRED8x8 || h->intra16x16_pred_mode==HOR_PRED8x8)){
1213 h->hpc.pred16x16_add[h->intra16x16_pred_mode](dest_y, block_offset, h->mb, linesize);
1215 for(i=0; i<16; i++){
1216 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16])
1217 s->dsp.add_pixels4(dest_y + block_offset[i], h->mb + i*16, linesize);
1221 s->dsp.h264_idct_add16intra(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1223 }else if(h->cbp&15){
1224 if(transform_bypass){
1225 const int di = IS_8x8DCT(mb_type) ? 4 : 1;
1226 idct_add= IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4;
1227 for(i=0; i<16; i+=di){
1228 if(h->non_zero_count_cache[ scan8[i] ]){
1229 idct_add(dest_y + block_offset[i], h->mb + i*16, linesize);
1233 if(IS_8x8DCT(mb_type)){
1234 s->dsp.h264_idct8_add4(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1236 s->dsp.h264_idct_add16(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1241 for(i=0; i<16; i++){
1242 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
1243 uint8_t * const ptr= dest_y + block_offset[i];
1244 svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
1250 if((simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)) && (h->cbp&0x30)){
1251 uint8_t *dest[2] = {dest_cb, dest_cr};
1252 if(transform_bypass){
1253 if(IS_INTRA(mb_type) && h->sps.profile_idc==244 && (h->chroma_pred_mode==VERT_PRED8x8 || h->chroma_pred_mode==HOR_PRED8x8)){
1254 h->hpc.pred8x8_add[h->chroma_pred_mode](dest[0], block_offset + 16, h->mb + 16*16, uvlinesize);
1255 h->hpc.pred8x8_add[h->chroma_pred_mode](dest[1], block_offset + 20, h->mb + 20*16, uvlinesize);
1257 idct_add = s->dsp.add_pixels4;
1258 for(i=16; i<16+8; i++){
1259 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16])
1260 idct_add (dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
1264 chroma_dc_dequant_idct_c(h->mb + 16*16, h->chroma_qp[0], h->dequant4_coeff[IS_INTRA(mb_type) ? 1:4][h->chroma_qp[0]][0]);
1265 chroma_dc_dequant_idct_c(h->mb + 16*16+4*16, h->chroma_qp[1], h->dequant4_coeff[IS_INTRA(mb_type) ? 2:5][h->chroma_qp[1]][0]);
1267 idct_add = s->dsp.h264_idct_add;
1268 idct_dc_add = s->dsp.h264_idct_dc_add;
1269 for(i=16; i<16+8; i++){
1270 if(h->non_zero_count_cache[ scan8[i] ])
1271 idct_add (dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
1272 else if(h->mb[i*16])
1273 idct_dc_add(dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
1276 for(i=16; i<16+8; i++){
1277 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
1278 uint8_t * const ptr= dest[(i&4)>>2] + block_offset[i];
1279 svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, ff_h264_chroma_qp[s->qscale + 12] - 12, 2);
1286 if(h->cbp || IS_INTRA(mb_type))
1287 s->dsp.clear_blocks(h->mb);
1289 if(h->deblocking_filter && 0) {
1290 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, simple);
1291 fill_filter_caches(h, mb_type); //FIXME don't fill stuff which isn't used by filter_mb
1292 h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.qscale_table[mb_xy]);
1293 h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.qscale_table[mb_xy]);
1294 if (!simple && FRAME_MBAFF) {
1295 ff_h264_filter_mb (h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
1297 ff_h264_filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
1303 * Process a macroblock; this case avoids checks for expensive uncommon cases.
1305 static void hl_decode_mb_simple(H264Context *h){
1306 hl_decode_mb_internal(h, 1);
1310 * Process a macroblock; this handles edge cases, such as interlacing.
1312 static void av_noinline hl_decode_mb_complex(H264Context *h){
1313 hl_decode_mb_internal(h, 0);
1316 void ff_h264_hl_decode_mb(H264Context *h){
1317 MpegEncContext * const s = &h->s;
1318 const int mb_xy= h->mb_xy;
1319 const int mb_type= s->current_picture.mb_type[mb_xy];
1320 int is_complex = CONFIG_SMALL || h->is_complex || IS_INTRA_PCM(mb_type) || s->qscale == 0;
1323 hl_decode_mb_complex(h);
1324 else hl_decode_mb_simple(h);
1327 static int pred_weight_table(H264Context *h){
1328 MpegEncContext * const s = &h->s;
1330 int luma_def, chroma_def;
1333 h->use_weight_chroma= 0;
1334 h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
1335 h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
1336 luma_def = 1<<h->luma_log2_weight_denom;
1337 chroma_def = 1<<h->chroma_log2_weight_denom;
1339 for(list=0; list<2; list++){
1340 h->luma_weight_flag[list] = 0;
1341 h->chroma_weight_flag[list] = 0;
1342 for(i=0; i<h->ref_count[list]; i++){
1343 int luma_weight_flag, chroma_weight_flag;
1345 luma_weight_flag= get_bits1(&s->gb);
1346 if(luma_weight_flag){
1347 h->luma_weight[list][i]= get_se_golomb(&s->gb);
1348 h->luma_offset[list][i]= get_se_golomb(&s->gb);
1349 if( h->luma_weight[list][i] != luma_def
1350 || h->luma_offset[list][i] != 0) {
1352 h->luma_weight_flag[list]= 1;
1355 h->luma_weight[list][i]= luma_def;
1356 h->luma_offset[list][i]= 0;
1360 chroma_weight_flag= get_bits1(&s->gb);
1361 if(chroma_weight_flag){
1364 h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);
1365 h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);
1366 if( h->chroma_weight[list][i][j] != chroma_def
1367 || h->chroma_offset[list][i][j] != 0) {
1368 h->use_weight_chroma= 1;
1369 h->chroma_weight_flag[list]= 1;
1375 h->chroma_weight[list][i][j]= chroma_def;
1376 h->chroma_offset[list][i][j]= 0;
1381 if(h->slice_type_nos != FF_B_TYPE) break;
1383 h->use_weight= h->use_weight || h->use_weight_chroma;
1387 static void implicit_weight_table(H264Context *h){
1388 MpegEncContext * const s = &h->s;
1390 int cur_poc = s->current_picture_ptr->poc;
1392 for (i = 0; i < 2; i++) {
1393 h->luma_weight_flag[i] = 0;
1394 h->chroma_weight_flag[i] = 0;
1397 if( h->ref_count[0] == 1 && h->ref_count[1] == 1
1398 && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
1400 h->use_weight_chroma= 0;
1405 h->use_weight_chroma= 2;
1406 h->luma_log2_weight_denom= 5;
1407 h->chroma_log2_weight_denom= 5;
1409 for(ref0=0; ref0 < h->ref_count[0]; ref0++){
1410 int poc0 = h->ref_list[0][ref0].poc;
1411 for(ref1=0; ref1 < h->ref_count[1]; ref1++){
1412 int poc1 = h->ref_list[1][ref1].poc;
1413 int td = av_clip(poc1 - poc0, -128, 127);
1415 int tb = av_clip(cur_poc - poc0, -128, 127);
1416 int tx = (16384 + (FFABS(td) >> 1)) / td;
1417 int dist_scale_factor = av_clip((tb*tx + 32) >> 6, -1024, 1023) >> 2;
1418 if(dist_scale_factor < -64 || dist_scale_factor > 128)
1419 h->implicit_weight[ref0][ref1] = 32;
1421 h->implicit_weight[ref0][ref1] = 64 - dist_scale_factor;
1423 h->implicit_weight[ref0][ref1] = 32;
1429 * instantaneous decoder refresh.
1431 static void idr(H264Context *h){
1432 ff_h264_remove_all_refs(h);
1433 h->prev_frame_num= 0;
1434 h->prev_frame_num_offset= 0;
1439 /* forget old pics after a seek */
1440 static void flush_dpb(AVCodecContext *avctx){
1441 H264Context *h= avctx->priv_data;
1443 for(i=0; i<MAX_DELAYED_PIC_COUNT; i++) {
1444 if(h->delayed_pic[i])
1445 h->delayed_pic[i]->reference= 0;
1446 h->delayed_pic[i]= NULL;
1448 h->outputed_poc= INT_MIN;
1449 h->prev_interlaced_frame = 1;
1451 if(h->s.current_picture_ptr)
1452 h->s.current_picture_ptr->reference= 0;
1453 h->s.first_field= 0;
1454 ff_h264_reset_sei(h);
1455 ff_mpeg_flush(avctx);
1458 static int init_poc(H264Context *h){
1459 MpegEncContext * const s = &h->s;
1460 const int max_frame_num= 1<<h->sps.log2_max_frame_num;
1462 Picture *cur = s->current_picture_ptr;
1464 h->frame_num_offset= h->prev_frame_num_offset;
1465 if(h->frame_num < h->prev_frame_num)
1466 h->frame_num_offset += max_frame_num;
1468 if(h->sps.poc_type==0){
1469 const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
1471 if (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
1472 h->poc_msb = h->prev_poc_msb + max_poc_lsb;
1473 else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
1474 h->poc_msb = h->prev_poc_msb - max_poc_lsb;
1476 h->poc_msb = h->prev_poc_msb;
1477 //printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
1479 field_poc[1] = h->poc_msb + h->poc_lsb;
1480 if(s->picture_structure == PICT_FRAME)
1481 field_poc[1] += h->delta_poc_bottom;
1482 }else if(h->sps.poc_type==1){
1483 int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
1486 if(h->sps.poc_cycle_length != 0)
1487 abs_frame_num = h->frame_num_offset + h->frame_num;
1491 if(h->nal_ref_idc==0 && abs_frame_num > 0)
1494 expected_delta_per_poc_cycle = 0;
1495 for(i=0; i < h->sps.poc_cycle_length; i++)
1496 expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
1498 if(abs_frame_num > 0){
1499 int poc_cycle_cnt = (abs_frame_num - 1) / h->sps.poc_cycle_length;
1500 int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
1502 expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
1503 for(i = 0; i <= frame_num_in_poc_cycle; i++)
1504 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
1508 if(h->nal_ref_idc == 0)
1509 expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
1511 field_poc[0] = expectedpoc + h->delta_poc[0];
1512 field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
1514 if(s->picture_structure == PICT_FRAME)
1515 field_poc[1] += h->delta_poc[1];
1517 int poc= 2*(h->frame_num_offset + h->frame_num);
1526 if(s->picture_structure != PICT_BOTTOM_FIELD)
1527 s->current_picture_ptr->field_poc[0]= field_poc[0];
1528 if(s->picture_structure != PICT_TOP_FIELD)
1529 s->current_picture_ptr->field_poc[1]= field_poc[1];
1530 cur->poc= FFMIN(cur->field_poc[0], cur->field_poc[1]);
1537 * initialize scan tables
1539 static void init_scan_tables(H264Context *h){
1540 MpegEncContext * const s = &h->s;
1542 if(s->dsp.h264_idct_add == ff_h264_idct_add_c){ //FIXME little ugly
1543 memcpy(h->zigzag_scan, zigzag_scan, 16*sizeof(uint8_t));
1544 memcpy(h-> field_scan, field_scan, 16*sizeof(uint8_t));
1546 for(i=0; i<16; i++){
1547 #define T(x) (x>>2) | ((x<<2) & 0xF)
1548 h->zigzag_scan[i] = T(zigzag_scan[i]);
1549 h-> field_scan[i] = T( field_scan[i]);
1553 if(s->dsp.h264_idct8_add == ff_h264_idct8_add_c){
1554 memcpy(h->zigzag_scan8x8, ff_zigzag_direct, 64*sizeof(uint8_t));
1555 memcpy(h->zigzag_scan8x8_cavlc, zigzag_scan8x8_cavlc, 64*sizeof(uint8_t));
1556 memcpy(h->field_scan8x8, field_scan8x8, 64*sizeof(uint8_t));
1557 memcpy(h->field_scan8x8_cavlc, field_scan8x8_cavlc, 64*sizeof(uint8_t));
1559 for(i=0; i<64; i++){
1560 #define T(x) (x>>3) | ((x&7)<<3)
1561 h->zigzag_scan8x8[i] = T(ff_zigzag_direct[i]);
1562 h->zigzag_scan8x8_cavlc[i] = T(zigzag_scan8x8_cavlc[i]);
1563 h->field_scan8x8[i] = T(field_scan8x8[i]);
1564 h->field_scan8x8_cavlc[i] = T(field_scan8x8_cavlc[i]);
1568 if(h->sps.transform_bypass){ //FIXME same ugly
1569 h->zigzag_scan_q0 = zigzag_scan;
1570 h->zigzag_scan8x8_q0 = ff_zigzag_direct;
1571 h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc;
1572 h->field_scan_q0 = field_scan;
1573 h->field_scan8x8_q0 = field_scan8x8;
1574 h->field_scan8x8_cavlc_q0 = field_scan8x8_cavlc;
1576 h->zigzag_scan_q0 = h->zigzag_scan;
1577 h->zigzag_scan8x8_q0 = h->zigzag_scan8x8;
1578 h->zigzag_scan8x8_cavlc_q0 = h->zigzag_scan8x8_cavlc;
1579 h->field_scan_q0 = h->field_scan;
1580 h->field_scan8x8_q0 = h->field_scan8x8;
1581 h->field_scan8x8_cavlc_q0 = h->field_scan8x8_cavlc;
1585 static void field_end(H264Context *h){
1586 MpegEncContext * const s = &h->s;
1587 AVCodecContext * const avctx= s->avctx;
1590 s->current_picture_ptr->qscale_type= FF_QSCALE_TYPE_H264;
1591 s->current_picture_ptr->pict_type= s->pict_type;
1593 if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
1594 ff_vdpau_h264_set_reference_frames(s);
1597 ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1598 h->prev_poc_msb= h->poc_msb;
1599 h->prev_poc_lsb= h->poc_lsb;
1601 h->prev_frame_num_offset= h->frame_num_offset;
1602 h->prev_frame_num= h->frame_num;
1604 if (avctx->hwaccel) {
1605 if (avctx->hwaccel->end_frame(avctx) < 0)
1606 av_log(avctx, AV_LOG_ERROR, "hardware accelerator failed to decode picture\n");
1609 if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
1610 ff_vdpau_h264_picture_complete(s);
1613 * FIXME: Error handling code does not seem to support interlaced
1614 * when slices span multiple rows
1615 * The ff_er_add_slice calls don't work right for bottom
1616 * fields; they cause massive erroneous error concealing
1617 * Error marking covers both fields (top and bottom).
1618 * This causes a mismatched s->error_count
1619 * and a bad error table. Further, the error count goes to
1620 * INT_MAX when called for bottom field, because mb_y is
1621 * past end by one (callers fault) and resync_mb_y != 0
1622 * causes problems for the first MB line, too.
1633 * Replicates H264 "master" context to thread contexts.
1635 static void clone_slice(H264Context *dst, H264Context *src)
1637 memcpy(dst->block_offset, src->block_offset, sizeof(dst->block_offset));
1638 dst->s.current_picture_ptr = src->s.current_picture_ptr;
1639 dst->s.current_picture = src->s.current_picture;
1640 dst->s.linesize = src->s.linesize;
1641 dst->s.uvlinesize = src->s.uvlinesize;
1642 dst->s.first_field = src->s.first_field;
1644 dst->prev_poc_msb = src->prev_poc_msb;
1645 dst->prev_poc_lsb = src->prev_poc_lsb;
1646 dst->prev_frame_num_offset = src->prev_frame_num_offset;
1647 dst->prev_frame_num = src->prev_frame_num;
1648 dst->short_ref_count = src->short_ref_count;
1650 memcpy(dst->short_ref, src->short_ref, sizeof(dst->short_ref));
1651 memcpy(dst->long_ref, src->long_ref, sizeof(dst->long_ref));
1652 memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
1653 memcpy(dst->ref_list, src->ref_list, sizeof(dst->ref_list));
1655 memcpy(dst->dequant4_coeff, src->dequant4_coeff, sizeof(src->dequant4_coeff));
1656 memcpy(dst->dequant8_coeff, src->dequant8_coeff, sizeof(src->dequant8_coeff));
1660 * decodes a slice header.
1661 * This will also call MPV_common_init() and frame_start() as needed.
1663 * @param h h264context
1664 * @param h0 h264 master context (differs from 'h' when doing sliced based parallel decoding)
1666 * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
1668 static int decode_slice_header(H264Context *h, H264Context *h0){
1669 MpegEncContext * const s = &h->s;
1670 MpegEncContext * const s0 = &h0->s;
1671 unsigned int first_mb_in_slice;
1672 unsigned int pps_id;
1673 int num_ref_idx_active_override_flag;
1674 unsigned int slice_type, tmp, i, j;
1675 int default_ref_list_done = 0;
1676 int last_pic_structure;
1678 s->dropable= h->nal_ref_idc == 0;
1680 if((s->avctx->flags2 & CODEC_FLAG2_FAST) && !h->nal_ref_idc){
1681 s->me.qpel_put= s->dsp.put_2tap_qpel_pixels_tab;
1682 s->me.qpel_avg= s->dsp.avg_2tap_qpel_pixels_tab;
1684 s->me.qpel_put= s->dsp.put_h264_qpel_pixels_tab;
1685 s->me.qpel_avg= s->dsp.avg_h264_qpel_pixels_tab;
1688 first_mb_in_slice= get_ue_golomb(&s->gb);
1690 if(first_mb_in_slice == 0){ //FIXME better field boundary detection
1691 if(h0->current_slice && FIELD_PICTURE){
1695 h0->current_slice = 0;
1696 if (!s0->first_field)
1697 s->current_picture_ptr= NULL;
1700 slice_type= get_ue_golomb_31(&s->gb);
1702 av_log(h->s.avctx, AV_LOG_ERROR, "slice type too large (%d) at %d %d\n", h->slice_type, s->mb_x, s->mb_y);
1707 h->slice_type_fixed=1;
1709 h->slice_type_fixed=0;
1711 slice_type= golomb_to_pict_type[ slice_type ];
1712 if (slice_type == FF_I_TYPE
1713 || (h0->current_slice != 0 && slice_type == h0->last_slice_type) ) {
1714 default_ref_list_done = 1;
1716 h->slice_type= slice_type;
1717 h->slice_type_nos= slice_type & 3;
1719 s->pict_type= h->slice_type; // to make a few old functions happy, it's wrong though
1720 if (s->pict_type == FF_B_TYPE && s0->last_picture_ptr == NULL) {
1721 av_log(h->s.avctx, AV_LOG_ERROR,
1722 "B picture before any references, skipping\n");
1726 pps_id= get_ue_golomb(&s->gb);
1727 if(pps_id>=MAX_PPS_COUNT){
1728 av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
1731 if(!h0->pps_buffers[pps_id]) {
1732 av_log(h->s.avctx, AV_LOG_ERROR, "non-existing PPS %u referenced\n", pps_id);
1735 h->pps= *h0->pps_buffers[pps_id];
1737 if(!h0->sps_buffers[h->pps.sps_id]) {
1738 av_log(h->s.avctx, AV_LOG_ERROR, "non-existing SPS %u referenced\n", h->pps.sps_id);
1741 h->sps = *h0->sps_buffers[h->pps.sps_id];
1743 if(h == h0 && h->dequant_coeff_pps != pps_id){
1744 h->dequant_coeff_pps = pps_id;
1745 init_dequant_tables(h);
1748 s->mb_width= h->sps.mb_width;
1749 s->mb_height= h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
1751 h->b_stride= s->mb_width*4;
1752 h->b8_stride= s->mb_width*2;
1754 s->width = 16*s->mb_width - 2*FFMIN(h->sps.crop_right, 7);
1755 if(h->sps.frame_mbs_only_flag)
1756 s->height= 16*s->mb_height - 2*FFMIN(h->sps.crop_bottom, 7);
1758 s->height= 16*s->mb_height - 4*FFMIN(h->sps.crop_bottom, 3);
1760 if (s->context_initialized
1761 && ( s->width != s->avctx->width || s->height != s->avctx->height)) {
1763 return -1; // width / height changed during parallelized decoding
1765 flush_dpb(s->avctx);
1768 if (!s->context_initialized) {
1770 return -1; // we cant (re-)initialize context during parallel decoding
1772 avcodec_set_dimensions(s->avctx, s->width, s->height);
1773 s->avctx->sample_aspect_ratio= h->sps.sar;
1774 if(!s->avctx->sample_aspect_ratio.den)
1775 s->avctx->sample_aspect_ratio.den = 1;
1777 if(h->sps.video_signal_type_present_flag){
1778 s->avctx->color_range = h->sps.full_range ? AVCOL_RANGE_JPEG : AVCOL_RANGE_MPEG;
1779 if(h->sps.colour_description_present_flag){
1780 s->avctx->color_primaries = h->sps.color_primaries;
1781 s->avctx->color_trc = h->sps.color_trc;
1782 s->avctx->colorspace = h->sps.colorspace;
1786 if(h->sps.timing_info_present_flag){
1787 s->avctx->time_base= (AVRational){h->sps.num_units_in_tick, h->sps.time_scale};
1788 if(h->x264_build > 0 && h->x264_build < 44)
1789 s->avctx->time_base.den *= 2;
1790 av_reduce(&s->avctx->time_base.num, &s->avctx->time_base.den,
1791 s->avctx->time_base.num, s->avctx->time_base.den, 1<<30);
1793 s->avctx->pix_fmt = s->avctx->get_format(s->avctx, s->avctx->codec->pix_fmts);
1794 s->avctx->hwaccel = ff_find_hwaccel(s->avctx->codec->id, s->avctx->pix_fmt);
1796 if (MPV_common_init(s) < 0)
1799 h->prev_interlaced_frame = 1;
1801 init_scan_tables(h);
1802 ff_h264_alloc_tables(h);
1804 for(i = 1; i < s->avctx->thread_count; i++) {
1806 c = h->thread_context[i] = av_malloc(sizeof(H264Context));
1807 memcpy(c, h->s.thread_context[i], sizeof(MpegEncContext));
1808 memset(&c->s + 1, 0, sizeof(H264Context) - sizeof(MpegEncContext));
1811 init_scan_tables(c);
1815 for(i = 0; i < s->avctx->thread_count; i++)
1816 if(context_init(h->thread_context[i]) < 0)
1820 h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
1823 h->mb_aff_frame = 0;
1824 last_pic_structure = s0->picture_structure;
1825 if(h->sps.frame_mbs_only_flag){
1826 s->picture_structure= PICT_FRAME;
1828 if(get_bits1(&s->gb)) { //field_pic_flag
1829 s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
1831 s->picture_structure= PICT_FRAME;
1832 h->mb_aff_frame = h->sps.mb_aff;
1835 h->mb_field_decoding_flag= s->picture_structure != PICT_FRAME;
1837 if(h0->current_slice == 0){
1838 while(h->frame_num != h->prev_frame_num &&
1839 h->frame_num != (h->prev_frame_num+1)%(1<<h->sps.log2_max_frame_num)){
1840 av_log(NULL, AV_LOG_DEBUG, "Frame num gap %d %d\n", h->frame_num, h->prev_frame_num);
1841 if (ff_h264_frame_start(h) < 0)
1843 h->prev_frame_num++;
1844 h->prev_frame_num %= 1<<h->sps.log2_max_frame_num;
1845 s->current_picture_ptr->frame_num= h->prev_frame_num;
1846 ff_h264_execute_ref_pic_marking(h, NULL, 0);
1849 /* See if we have a decoded first field looking for a pair... */
1850 if (s0->first_field) {
1851 assert(s0->current_picture_ptr);
1852 assert(s0->current_picture_ptr->data[0]);
1853 assert(s0->current_picture_ptr->reference != DELAYED_PIC_REF);
1855 /* figure out if we have a complementary field pair */
1856 if (!FIELD_PICTURE || s->picture_structure == last_pic_structure) {
1858 * Previous field is unmatched. Don't display it, but let it
1859 * remain for reference if marked as such.
1861 s0->current_picture_ptr = NULL;
1862 s0->first_field = FIELD_PICTURE;
1865 if (h->nal_ref_idc &&
1866 s0->current_picture_ptr->reference &&
1867 s0->current_picture_ptr->frame_num != h->frame_num) {
1869 * This and previous field were reference, but had
1870 * different frame_nums. Consider this field first in
1871 * pair. Throw away previous field except for reference
1874 s0->first_field = 1;
1875 s0->current_picture_ptr = NULL;
1878 /* Second field in complementary pair */
1879 s0->first_field = 0;
1884 /* Frame or first field in a potentially complementary pair */
1885 assert(!s0->current_picture_ptr);
1886 s0->first_field = FIELD_PICTURE;
1889 if((!FIELD_PICTURE || s0->first_field) && ff_h264_frame_start(h) < 0) {
1890 s0->first_field = 0;
1897 s->current_picture_ptr->frame_num= h->frame_num; //FIXME frame_num cleanup
1899 assert(s->mb_num == s->mb_width * s->mb_height);
1900 if(first_mb_in_slice << FIELD_OR_MBAFF_PICTURE >= s->mb_num ||
1901 first_mb_in_slice >= s->mb_num){
1902 av_log(h->s.avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
1905 s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
1906 s->resync_mb_y = s->mb_y = (first_mb_in_slice / s->mb_width) << FIELD_OR_MBAFF_PICTURE;
1907 if (s->picture_structure == PICT_BOTTOM_FIELD)
1908 s->resync_mb_y = s->mb_y = s->mb_y + 1;
1909 assert(s->mb_y < s->mb_height);
1911 if(s->picture_structure==PICT_FRAME){
1912 h->curr_pic_num= h->frame_num;
1913 h->max_pic_num= 1<< h->sps.log2_max_frame_num;
1915 h->curr_pic_num= 2*h->frame_num + 1;
1916 h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
1919 if(h->nal_unit_type == NAL_IDR_SLICE){
1920 get_ue_golomb(&s->gb); /* idr_pic_id */
1923 if(h->sps.poc_type==0){
1924 h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
1926 if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
1927 h->delta_poc_bottom= get_se_golomb(&s->gb);
1931 if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
1932 h->delta_poc[0]= get_se_golomb(&s->gb);
1934 if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
1935 h->delta_poc[1]= get_se_golomb(&s->gb);
1940 if(h->pps.redundant_pic_cnt_present){
1941 h->redundant_pic_count= get_ue_golomb(&s->gb);
1944 //set defaults, might be overridden a few lines later
1945 h->ref_count[0]= h->pps.ref_count[0];
1946 h->ref_count[1]= h->pps.ref_count[1];
1948 if(h->slice_type_nos != FF_I_TYPE){
1949 if(h->slice_type_nos == FF_B_TYPE){
1950 h->direct_spatial_mv_pred= get_bits1(&s->gb);
1952 num_ref_idx_active_override_flag= get_bits1(&s->gb);
1954 if(num_ref_idx_active_override_flag){
1955 h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
1956 if(h->slice_type_nos==FF_B_TYPE)
1957 h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
1959 if(h->ref_count[0]-1 > 32-1 || h->ref_count[1]-1 > 32-1){
1960 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
1961 h->ref_count[0]= h->ref_count[1]= 1;
1965 if(h->slice_type_nos == FF_B_TYPE)
1972 if(!default_ref_list_done){
1973 ff_h264_fill_default_ref_list(h);
1976 if(h->slice_type_nos!=FF_I_TYPE && ff_h264_decode_ref_pic_list_reordering(h) < 0)
1979 if(h->slice_type_nos!=FF_I_TYPE){
1980 s->last_picture_ptr= &h->ref_list[0][0];
1981 ff_copy_picture(&s->last_picture, s->last_picture_ptr);
1983 if(h->slice_type_nos==FF_B_TYPE){
1984 s->next_picture_ptr= &h->ref_list[1][0];
1985 ff_copy_picture(&s->next_picture, s->next_picture_ptr);
1988 if( (h->pps.weighted_pred && h->slice_type_nos == FF_P_TYPE )
1989 || (h->pps.weighted_bipred_idc==1 && h->slice_type_nos== FF_B_TYPE ) )
1990 pred_weight_table(h);
1991 else if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== FF_B_TYPE)
1992 implicit_weight_table(h);
1995 for (i = 0; i < 2; i++) {
1996 h->luma_weight_flag[i] = 0;
1997 h->chroma_weight_flag[i] = 0;
2002 ff_h264_decode_ref_pic_marking(h0, &s->gb);
2005 ff_h264_fill_mbaff_ref_list(h);
2007 if(h->slice_type_nos==FF_B_TYPE && !h->direct_spatial_mv_pred)
2008 ff_h264_direct_dist_scale_factor(h);
2009 ff_h264_direct_ref_list_init(h);
2011 if( h->slice_type_nos != FF_I_TYPE && h->pps.cabac ){
2012 tmp = get_ue_golomb_31(&s->gb);
2014 av_log(s->avctx, AV_LOG_ERROR, "cabac_init_idc overflow\n");
2017 h->cabac_init_idc= tmp;
2020 h->last_qscale_diff = 0;
2021 tmp = h->pps.init_qp + get_se_golomb(&s->gb);
2023 av_log(s->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
2027 h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
2028 h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
2029 //FIXME qscale / qp ... stuff
2030 if(h->slice_type == FF_SP_TYPE){
2031 get_bits1(&s->gb); /* sp_for_switch_flag */
2033 if(h->slice_type==FF_SP_TYPE || h->slice_type == FF_SI_TYPE){
2034 get_se_golomb(&s->gb); /* slice_qs_delta */
2037 h->deblocking_filter = 1;
2038 h->slice_alpha_c0_offset = 52;
2039 h->slice_beta_offset = 52;
2040 if( h->pps.deblocking_filter_parameters_present ) {
2041 tmp= get_ue_golomb_31(&s->gb);
2043 av_log(s->avctx, AV_LOG_ERROR, "deblocking_filter_idc %u out of range\n", tmp);
2046 h->deblocking_filter= tmp;
2047 if(h->deblocking_filter < 2)
2048 h->deblocking_filter^= 1; // 1<->0
2050 if( h->deblocking_filter ) {
2051 h->slice_alpha_c0_offset += get_se_golomb(&s->gb) << 1;
2052 h->slice_beta_offset += get_se_golomb(&s->gb) << 1;
2053 if( h->slice_alpha_c0_offset > 104U
2054 || h->slice_beta_offset > 104U){
2055 av_log(s->avctx, AV_LOG_ERROR, "deblocking filter parameters %d %d out of range\n", h->slice_alpha_c0_offset, h->slice_beta_offset);
2061 if( s->avctx->skip_loop_filter >= AVDISCARD_ALL
2062 ||(s->avctx->skip_loop_filter >= AVDISCARD_NONKEY && h->slice_type_nos != FF_I_TYPE)
2063 ||(s->avctx->skip_loop_filter >= AVDISCARD_BIDIR && h->slice_type_nos == FF_B_TYPE)
2064 ||(s->avctx->skip_loop_filter >= AVDISCARD_NONREF && h->nal_ref_idc == 0))
2065 h->deblocking_filter= 0;
2067 if(h->deblocking_filter == 1 && h0->max_contexts > 1) {
2068 if(s->avctx->flags2 & CODEC_FLAG2_FAST) {
2069 /* Cheat slightly for speed:
2070 Do not bother to deblock across slices. */
2071 h->deblocking_filter = 2;
2073 h0->max_contexts = 1;
2074 if(!h0->single_decode_warning) {
2075 av_log(s->avctx, AV_LOG_INFO, "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
2076 h0->single_decode_warning = 1;
2079 return 1; // deblocking switched inside frame
2082 h->qp_thresh= 15 + 52 - FFMIN(h->slice_alpha_c0_offset, h->slice_beta_offset) - FFMAX3(0, h->pps.chroma_qp_index_offset[0], h->pps.chroma_qp_index_offset[1]);
2085 if( h->pps.num_slice_groups > 1 && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
2086 slice_group_change_cycle= get_bits(&s->gb, ?);
2089 h0->last_slice_type = slice_type;
2090 h->slice_num = ++h0->current_slice;
2091 if(h->slice_num >= MAX_SLICES){
2092 av_log(s->avctx, AV_LOG_ERROR, "Too many slices, increase MAX_SLICES and recompile\n");
2097 int *ref2frm= h->ref2frm[h->slice_num&(MAX_SLICES-1)][j];
2098 for(i=0; i<16; i++){
2100 if(h->ref_list[j][i].data[0]){
2102 uint8_t *base= h->ref_list[j][i].base[0];
2103 for(k=0; k<h->short_ref_count; k++)
2104 if(h->short_ref[k]->base[0] == base){
2108 for(k=0; k<h->long_ref_count; k++)
2109 if(h->long_ref[k] && h->long_ref[k]->base[0] == base){
2110 id_list[i]= h->short_ref_count + k;
2119 ref2frm[i+2]= 4*id_list[i]
2120 +(h->ref_list[j][i].reference&3);
2123 for(i=16; i<48; i++)
2124 ref2frm[i+4]= 4*id_list[(i-16)>>1]
2125 +(h->ref_list[j][i].reference&3);
2128 h->emu_edge_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16;
2129 h->emu_edge_height= (FRAME_MBAFF || FIELD_PICTURE) ? 0 : h->emu_edge_width;
2131 s->avctx->refs= h->sps.ref_frame_count;
2133 if(s->avctx->debug&FF_DEBUG_PICT_INFO){
2134 av_log(h->s.avctx, AV_LOG_DEBUG, "slice:%d %s mb:%d %c%s%s pps:%u frame:%d poc:%d/%d ref:%d/%d qp:%d loop:%d:%d:%d weight:%d%s %s\n",
2136 (s->picture_structure==PICT_FRAME ? "F" : s->picture_structure==PICT_TOP_FIELD ? "T" : "B"),
2138 av_get_pict_type_char(h->slice_type), h->slice_type_fixed ? " fix" : "", h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
2139 pps_id, h->frame_num,
2140 s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
2141 h->ref_count[0], h->ref_count[1],
2143 h->deblocking_filter, h->slice_alpha_c0_offset/2-26, h->slice_beta_offset/2-26,
2145 h->use_weight==1 && h->use_weight_chroma ? "c" : "",
2146 h->slice_type == FF_B_TYPE ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : ""
2153 int ff_h264_get_slice_type(const H264Context *h)
2155 switch (h->slice_type) {
2156 case FF_P_TYPE: return 0;
2157 case FF_B_TYPE: return 1;
2158 case FF_I_TYPE: return 2;
2159 case FF_SP_TYPE: return 3;
2160 case FF_SI_TYPE: return 4;
2165 static void loop_filter(H264Context *h){
2166 MpegEncContext * const s = &h->s;
2167 uint8_t *dest_y, *dest_cb, *dest_cr;
2168 int linesize, uvlinesize, mb_x, mb_y;
2169 const int end_mb_y= s->mb_y + FRAME_MBAFF;
2170 const int old_slice_type= h->slice_type;
2172 if(h->deblocking_filter) {
2173 for(mb_x= 0; mb_x<s->mb_width; mb_x++){
2174 for(mb_y=end_mb_y - FRAME_MBAFF; mb_y<= end_mb_y; mb_y++){
2175 int list, mb_xy, mb_type;
2176 mb_xy = h->mb_xy = mb_x + mb_y*s->mb_stride;
2177 h->slice_num= h->slice_table[mb_xy];
2178 mb_type= s->current_picture.mb_type[mb_xy];
2179 h->list_count= h->list_counts[mb_xy];
2182 h->mb_mbaff = h->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
2186 dest_y = s->current_picture.data[0] + (mb_x + mb_y * s->linesize ) * 16;
2187 dest_cb = s->current_picture.data[1] + (mb_x + mb_y * s->uvlinesize) * 8;
2188 dest_cr = s->current_picture.data[2] + (mb_x + mb_y * s->uvlinesize) * 8;
2189 //FIXME simplify above
2192 linesize = h->mb_linesize = s->linesize * 2;
2193 uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
2194 if(mb_y&1){ //FIXME move out of this function?
2195 dest_y -= s->linesize*15;
2196 dest_cb-= s->uvlinesize*7;
2197 dest_cr-= s->uvlinesize*7;
2200 linesize = h->mb_linesize = s->linesize;
2201 uvlinesize = h->mb_uvlinesize = s->uvlinesize;
2203 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);
2204 if(fill_filter_caches(h, mb_type))
2206 h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.qscale_table[mb_xy]);
2207 h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.qscale_table[mb_xy]);
2210 ff_h264_filter_mb (h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2212 ff_h264_filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2217 h->slice_type= old_slice_type;
2219 s->mb_y= end_mb_y - FRAME_MBAFF;
2222 static int decode_slice(struct AVCodecContext *avctx, void *arg){
2223 H264Context *h = *(void**)arg;
2224 MpegEncContext * const s = &h->s;
2225 const int part_mask= s->partitioned_frame ? (AC_END|AC_ERROR) : 0x7F;
2229 h->is_complex = FRAME_MBAFF || s->picture_structure != PICT_FRAME || s->codec_id != CODEC_ID_H264 ||
2230 (CONFIG_GRAY && (s->flags&CODEC_FLAG_GRAY));
2232 if( h->pps.cabac ) {
2234 align_get_bits( &s->gb );
2237 ff_init_cabac_states( &h->cabac);
2238 ff_init_cabac_decoder( &h->cabac,
2239 s->gb.buffer + get_bits_count(&s->gb)/8,
2240 (get_bits_left(&s->gb) + 7)/8);
2242 ff_h264_init_cabac_states(h);
2246 int ret = ff_h264_decode_mb_cabac(h);
2248 //STOP_TIMER("decode_mb_cabac")
2250 if(ret>=0) ff_h264_hl_decode_mb(h);
2252 if( ret >= 0 && FRAME_MBAFF ) { //FIXME optimal? or let mb_decode decode 16x32 ?
2255 ret = ff_h264_decode_mb_cabac(h);
2257 if(ret>=0) ff_h264_hl_decode_mb(h);
2260 eos = get_cabac_terminate( &h->cabac );
2262 if( ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 2) {
2263 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d, bytestream (%td)\n", s->mb_x, s->mb_y, h->cabac.bytestream_end - h->cabac.bytestream);
2264 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask);
2268 if( ++s->mb_x >= s->mb_width ) {
2271 ff_draw_horiz_band(s, 16*s->mb_y, 16);
2273 if(FIELD_OR_MBAFF_PICTURE) {
2278 if( eos || s->mb_y >= s->mb_height ) {
2279 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2280 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
2287 int ret = ff_h264_decode_mb_cavlc(h);
2289 if(ret>=0) ff_h264_hl_decode_mb(h);
2291 if(ret>=0 && FRAME_MBAFF){ //FIXME optimal? or let mb_decode decode 16x32 ?
2293 ret = ff_h264_decode_mb_cavlc(h);
2295 if(ret>=0) ff_h264_hl_decode_mb(h);
2300 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
2301 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask);
2306 if(++s->mb_x >= s->mb_width){
2309 ff_draw_horiz_band(s, 16*s->mb_y, 16);
2311 if(FIELD_OR_MBAFF_PICTURE) {
2314 if(s->mb_y >= s->mb_height){
2315 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2317 if(get_bits_count(&s->gb) == s->gb.size_in_bits ) {
2318 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
2322 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
2329 if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->mb_skip_run<=0){
2330 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2331 if(get_bits_count(&s->gb) == s->gb.size_in_bits ){
2332 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
2336 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask);
2345 for(;s->mb_y < s->mb_height; s->mb_y++){
2346 for(;s->mb_x < s->mb_width; s->mb_x++){
2347 int ret= decode_mb(h);
2349 ff_h264_hl_decode_mb(h);
2352 av_log(s->avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
2353 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask);
2358 if(++s->mb_x >= s->mb_width){
2360 if(++s->mb_y >= s->mb_height){
2361 if(get_bits_count(s->gb) == s->gb.size_in_bits){
2362 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
2366 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
2373 if(get_bits_count(s->?gb) >= s->gb?.size_in_bits){
2374 if(get_bits_count(s->gb) == s->gb.size_in_bits){
2375 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
2379 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask);
2386 ff_draw_horiz_band(s, 16*s->mb_y, 16);
2389 return -1; //not reached
2393 * Call decode_slice() for each context.
2395 * @param h h264 master context
2396 * @param context_count number of contexts to execute
2398 static void execute_decode_slices(H264Context *h, int context_count){
2399 MpegEncContext * const s = &h->s;
2400 AVCodecContext * const avctx= s->avctx;
2404 if (s->avctx->hwaccel)
2406 if(s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
2408 if(context_count == 1) {
2409 decode_slice(avctx, &h);
2411 for(i = 1; i < context_count; i++) {
2412 hx = h->thread_context[i];
2413 hx->s.error_recognition = avctx->error_recognition;
2414 hx->s.error_count = 0;
2417 avctx->execute(avctx, (void *)decode_slice,
2418 h->thread_context, NULL, context_count, sizeof(void*));
2420 /* pull back stuff from slices to master context */
2421 hx = h->thread_context[context_count - 1];
2422 s->mb_x = hx->s.mb_x;
2423 s->mb_y = hx->s.mb_y;
2424 s->dropable = hx->s.dropable;
2425 s->picture_structure = hx->s.picture_structure;
2426 for(i = 1; i < context_count; i++)
2427 h->s.error_count += h->thread_context[i]->s.error_count;
2432 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size){
2433 MpegEncContext * const s = &h->s;
2434 AVCodecContext * const avctx= s->avctx;
2436 H264Context *hx; ///< thread context
2437 int context_count = 0;
2438 int next_avc= h->is_avc ? 0 : buf_size;
2440 h->max_contexts = avctx->thread_count;
2443 for(i=0; i<50; i++){
2444 av_log(NULL, AV_LOG_ERROR,"%02X ", buf[i]);
2447 if(!(s->flags2 & CODEC_FLAG2_CHUNKS)){
2448 h->current_slice = 0;
2449 if (!s->first_field)
2450 s->current_picture_ptr= NULL;
2451 ff_h264_reset_sei(h);
2462 if(buf_index >= next_avc) {
2463 if(buf_index >= buf_size) break;
2465 for(i = 0; i < h->nal_length_size; i++)
2466 nalsize = (nalsize << 8) | buf[buf_index++];
2467 if(nalsize <= 1 || nalsize > buf_size - buf_index){
2472 av_log(h->s.avctx, AV_LOG_ERROR, "AVC: nal size %d\n", nalsize);
2476 next_avc= buf_index + nalsize;
2478 // start code prefix search
2479 for(; buf_index + 3 < next_avc; buf_index++){
2480 // This should always succeed in the first iteration.
2481 if(buf[buf_index] == 0 && buf[buf_index+1] == 0 && buf[buf_index+2] == 1)
2485 if(buf_index+3 >= buf_size) break;
2488 if(buf_index >= next_avc) continue;
2491 hx = h->thread_context[context_count];
2493 ptr= ff_h264_decode_nal(hx, buf + buf_index, &dst_length, &consumed, next_avc - buf_index);
2494 if (ptr==NULL || dst_length < 0){
2497 while(ptr[dst_length - 1] == 0 && dst_length > 0)
2499 bit_length= !dst_length ? 0 : (8*dst_length - ff_h264_decode_rbsp_trailing(h, ptr + dst_length - 1));
2501 if(s->avctx->debug&FF_DEBUG_STARTCODE){
2502 av_log(h->s.avctx, AV_LOG_DEBUG, "NAL %d at %d/%d length %d\n", hx->nal_unit_type, buf_index, buf_size, dst_length);
2505 if (h->is_avc && (nalsize != consumed) && nalsize){
2506 av_log(h->s.avctx, AV_LOG_DEBUG, "AVC: Consumed only %d bytes instead of %d\n", consumed, nalsize);
2509 buf_index += consumed;
2511 if( (s->hurry_up == 1 && h->nal_ref_idc == 0) //FIXME do not discard SEI id
2512 ||(avctx->skip_frame >= AVDISCARD_NONREF && h->nal_ref_idc == 0))
2517 switch(hx->nal_unit_type){
2519 if (h->nal_unit_type != NAL_IDR_SLICE) {
2520 av_log(h->s.avctx, AV_LOG_ERROR, "Invalid mix of idr and non-idr slices");
2523 idr(h); //FIXME ensure we don't loose some frames if there is reordering
2525 init_get_bits(&hx->s.gb, ptr, bit_length);
2527 hx->inter_gb_ptr= &hx->s.gb;
2528 hx->s.data_partitioning = 0;
2530 if((err = decode_slice_header(hx, h)))
2533 if (s->avctx->hwaccel && h->current_slice == 1) {
2534 if (s->avctx->hwaccel->start_frame(s->avctx, NULL, 0) < 0)
2538 s->current_picture_ptr->key_frame |=
2539 (hx->nal_unit_type == NAL_IDR_SLICE) ||
2540 (h->sei_recovery_frame_cnt >= 0);
2541 if(hx->redundant_pic_count==0 && hx->s.hurry_up < 5
2542 && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
2543 && (avctx->skip_frame < AVDISCARD_BIDIR || hx->slice_type_nos!=FF_B_TYPE)
2544 && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==FF_I_TYPE)
2545 && avctx->skip_frame < AVDISCARD_ALL){
2546 if(avctx->hwaccel) {
2547 if (avctx->hwaccel->decode_slice(avctx, &buf[buf_index - consumed], consumed) < 0)
2550 if(CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU){
2551 static const uint8_t start_code[] = {0x00, 0x00, 0x01};
2552 ff_vdpau_add_data_chunk(s, start_code, sizeof(start_code));
2553 ff_vdpau_add_data_chunk(s, &buf[buf_index - consumed], consumed );
2559 init_get_bits(&hx->s.gb, ptr, bit_length);
2561 hx->inter_gb_ptr= NULL;
2563 if ((err = decode_slice_header(hx, h)) < 0)
2566 hx->s.data_partitioning = 1;
2570 init_get_bits(&hx->intra_gb, ptr, bit_length);
2571 hx->intra_gb_ptr= &hx->intra_gb;
2574 init_get_bits(&hx->inter_gb, ptr, bit_length);
2575 hx->inter_gb_ptr= &hx->inter_gb;
2577 if(hx->redundant_pic_count==0 && hx->intra_gb_ptr && hx->s.data_partitioning
2578 && s->context_initialized
2580 && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
2581 && (avctx->skip_frame < AVDISCARD_BIDIR || hx->slice_type_nos!=FF_B_TYPE)
2582 && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==FF_I_TYPE)
2583 && avctx->skip_frame < AVDISCARD_ALL)
2587 init_get_bits(&s->gb, ptr, bit_length);
2588 ff_h264_decode_sei(h);
2591 init_get_bits(&s->gb, ptr, bit_length);
2592 ff_h264_decode_seq_parameter_set(h);
2594 if(s->flags& CODEC_FLAG_LOW_DELAY)
2597 if(avctx->has_b_frames < 2)
2598 avctx->has_b_frames= !s->low_delay;
2601 init_get_bits(&s->gb, ptr, bit_length);
2603 ff_h264_decode_picture_parameter_set(h, bit_length);
2607 case NAL_END_SEQUENCE:
2608 case NAL_END_STREAM:
2609 case NAL_FILLER_DATA:
2611 case NAL_AUXILIARY_SLICE:
2614 av_log(avctx, AV_LOG_DEBUG, "Unknown NAL code: %d (%d bits)\n", hx->nal_unit_type, bit_length);
2617 if(context_count == h->max_contexts) {
2618 execute_decode_slices(h, context_count);
2623 av_log(h->s.avctx, AV_LOG_ERROR, "decode_slice_header error\n");
2625 /* Slice could not be decoded in parallel mode, copy down
2626 * NAL unit stuff to context 0 and restart. Note that
2627 * rbsp_buffer is not transferred, but since we no longer
2628 * run in parallel mode this should not be an issue. */
2629 h->nal_unit_type = hx->nal_unit_type;
2630 h->nal_ref_idc = hx->nal_ref_idc;
2636 execute_decode_slices(h, context_count);
2641 * returns the number of bytes consumed for building the current frame
2643 static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size){
2644 if(pos==0) pos=1; //avoid infinite loops (i doubt that is needed but ...)
2645 if(pos+10>buf_size) pos=buf_size; // oops ;)
2650 static int decode_frame(AVCodecContext *avctx,
2651 void *data, int *data_size,
2654 const uint8_t *buf = avpkt->data;
2655 int buf_size = avpkt->size;
2656 H264Context *h = avctx->priv_data;
2657 MpegEncContext *s = &h->s;
2658 AVFrame *pict = data;
2661 s->flags= avctx->flags;
2662 s->flags2= avctx->flags2;
2664 /* end of stream, output what is still in the buffers */
2665 if (buf_size == 0) {
2669 //FIXME factorize this with the output code below
2670 out = h->delayed_pic[0];
2672 for(i=1; h->delayed_pic[i] && !h->delayed_pic[i]->key_frame && !h->delayed_pic[i]->mmco_reset; i++)
2673 if(h->delayed_pic[i]->poc < out->poc){
2674 out = h->delayed_pic[i];
2678 for(i=out_idx; h->delayed_pic[i]; i++)
2679 h->delayed_pic[i] = h->delayed_pic[i+1];
2682 *data_size = sizeof(AVFrame);
2683 *pict= *(AVFrame*)out;
2689 if(h->is_avc && !h->got_avcC) {
2690 int i, cnt, nalsize;
2691 unsigned char *p = avctx->extradata;
2692 if(avctx->extradata_size < 7) {
2693 av_log(avctx, AV_LOG_ERROR, "avcC too short\n");
2697 av_log(avctx, AV_LOG_ERROR, "Unknown avcC version %d\n", *p);
2700 /* sps and pps in the avcC always have length coded with 2 bytes,
2701 so put a fake nal_length_size = 2 while parsing them */
2702 h->nal_length_size = 2;
2703 // Decode sps from avcC
2704 cnt = *(p+5) & 0x1f; // Number of sps
2706 for (i = 0; i < cnt; i++) {
2707 nalsize = AV_RB16(p) + 2;
2708 if(decode_nal_units(h, p, nalsize) < 0) {
2709 av_log(avctx, AV_LOG_ERROR, "Decoding sps %d from avcC failed\n", i);
2714 // Decode pps from avcC
2715 cnt = *(p++); // Number of pps
2716 for (i = 0; i < cnt; i++) {
2717 nalsize = AV_RB16(p) + 2;
2718 if(decode_nal_units(h, p, nalsize) != nalsize) {
2719 av_log(avctx, AV_LOG_ERROR, "Decoding pps %d from avcC failed\n", i);
2724 // Now store right nal length size, that will be use to parse all other nals
2725 h->nal_length_size = ((*(((char*)(avctx->extradata))+4))&0x03)+1;
2726 // Do not reparse avcC
2730 if(!h->got_avcC && !h->is_avc && s->avctx->extradata_size){
2731 if(decode_nal_units(h, s->avctx->extradata, s->avctx->extradata_size) < 0)
2736 buf_index=decode_nal_units(h, buf, buf_size);
2740 if(!(s->flags2 & CODEC_FLAG2_CHUNKS) && !s->current_picture_ptr){
2741 if (avctx->skip_frame >= AVDISCARD_NONREF || s->hurry_up) return 0;
2742 av_log(avctx, AV_LOG_ERROR, "no frame!\n");
2746 if(!(s->flags2 & CODEC_FLAG2_CHUNKS) || (s->mb_y >= s->mb_height && s->mb_height)){
2747 Picture *out = s->current_picture_ptr;
2748 Picture *cur = s->current_picture_ptr;
2749 int i, pics, out_of_order, out_idx;
2753 if (cur->field_poc[0]==INT_MAX || cur->field_poc[1]==INT_MAX) {
2754 /* Wait for second field. */
2758 cur->interlaced_frame = 0;
2759 cur->repeat_pict = 0;
2761 /* Signal interlacing information externally. */
2762 /* Prioritize picture timing SEI information over used decoding process if it exists. */
2764 if(h->sps.pic_struct_present_flag){
2765 switch (h->sei_pic_struct)
2767 case SEI_PIC_STRUCT_FRAME:
2769 case SEI_PIC_STRUCT_TOP_FIELD:
2770 case SEI_PIC_STRUCT_BOTTOM_FIELD:
2771 cur->interlaced_frame = 1;
2773 case SEI_PIC_STRUCT_TOP_BOTTOM:
2774 case SEI_PIC_STRUCT_BOTTOM_TOP:
2775 if (FIELD_OR_MBAFF_PICTURE)
2776 cur->interlaced_frame = 1;
2778 // try to flag soft telecine progressive
2779 cur->interlaced_frame = h->prev_interlaced_frame;
2781 case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
2782 case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
2783 // Signal the possibility of telecined film externally (pic_struct 5,6)
2784 // From these hints, let the applications decide if they apply deinterlacing.
2785 cur->repeat_pict = 1;
2787 case SEI_PIC_STRUCT_FRAME_DOUBLING:
2788 // Force progressive here, as doubling interlaced frame is a bad idea.
2789 cur->repeat_pict = 2;
2791 case SEI_PIC_STRUCT_FRAME_TRIPLING:
2792 cur->repeat_pict = 4;
2796 if ((h->sei_ct_type & 3) && h->sei_pic_struct <= SEI_PIC_STRUCT_BOTTOM_TOP)
2797 cur->interlaced_frame = (h->sei_ct_type & (1<<1)) != 0;
2799 /* Derive interlacing flag from used decoding process. */
2800 cur->interlaced_frame = FIELD_OR_MBAFF_PICTURE;
2802 h->prev_interlaced_frame = cur->interlaced_frame;
2804 if (cur->field_poc[0] != cur->field_poc[1]){
2805 /* Derive top_field_first from field pocs. */
2806 cur->top_field_first = cur->field_poc[0] < cur->field_poc[1];
2808 if(cur->interlaced_frame || h->sps.pic_struct_present_flag){
2809 /* Use picture timing SEI information. Even if it is a information of a past frame, better than nothing. */
2810 if(h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM
2811 || h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM_TOP)
2812 cur->top_field_first = 1;
2814 cur->top_field_first = 0;
2816 /* Most likely progressive */
2817 cur->top_field_first = 0;
2821 //FIXME do something with unavailable reference frames
2823 /* Sort B-frames into display order */
2825 if(h->sps.bitstream_restriction_flag
2826 && s->avctx->has_b_frames < h->sps.num_reorder_frames){
2827 s->avctx->has_b_frames = h->sps.num_reorder_frames;
2831 if( s->avctx->strict_std_compliance >= FF_COMPLIANCE_STRICT
2832 && !h->sps.bitstream_restriction_flag){
2833 s->avctx->has_b_frames= MAX_DELAYED_PIC_COUNT;
2838 while(h->delayed_pic[pics]) pics++;
2840 assert(pics <= MAX_DELAYED_PIC_COUNT);
2842 h->delayed_pic[pics++] = cur;
2843 if(cur->reference == 0)
2844 cur->reference = DELAYED_PIC_REF;
2846 out = h->delayed_pic[0];
2848 for(i=1; h->delayed_pic[i] && !h->delayed_pic[i]->key_frame && !h->delayed_pic[i]->mmco_reset; i++)
2849 if(h->delayed_pic[i]->poc < out->poc){
2850 out = h->delayed_pic[i];
2853 if(s->avctx->has_b_frames == 0 && (h->delayed_pic[0]->key_frame || h->delayed_pic[0]->mmco_reset))
2854 h->outputed_poc= INT_MIN;
2855 out_of_order = out->poc < h->outputed_poc;
2857 if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames >= h->sps.num_reorder_frames)
2859 else if((out_of_order && pics-1 == s->avctx->has_b_frames && s->avctx->has_b_frames < MAX_DELAYED_PIC_COUNT)
2861 ((h->outputed_poc != INT_MIN && out->poc > h->outputed_poc + 2)
2862 || cur->pict_type == FF_B_TYPE)))
2865 s->avctx->has_b_frames++;
2868 if(out_of_order || pics > s->avctx->has_b_frames){
2869 out->reference &= ~DELAYED_PIC_REF;
2870 for(i=out_idx; h->delayed_pic[i]; i++)
2871 h->delayed_pic[i] = h->delayed_pic[i+1];
2873 if(!out_of_order && pics > s->avctx->has_b_frames){
2874 *data_size = sizeof(AVFrame);
2876 if(out_idx==0 && h->delayed_pic[0] && (h->delayed_pic[0]->key_frame || h->delayed_pic[0]->mmco_reset)) {
2877 h->outputed_poc = INT_MIN;
2879 h->outputed_poc = out->poc;
2880 *pict= *(AVFrame*)out;
2882 av_log(avctx, AV_LOG_DEBUG, "no picture\n");
2887 assert(pict->data[0] || !*data_size);
2888 ff_print_debug_info(s, pict);
2889 //printf("out %d\n", (int)pict->data[0]);
2891 return get_consumed_bytes(s, buf_index, buf_size);
2894 static inline void fill_mb_avail(H264Context *h){
2895 MpegEncContext * const s = &h->s;
2896 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
2899 h->mb_avail[0]= s->mb_x && h->slice_table[mb_xy - s->mb_stride - 1] == h->slice_num;
2900 h->mb_avail[1]= h->slice_table[mb_xy - s->mb_stride ] == h->slice_num;
2901 h->mb_avail[2]= s->mb_x+1 < s->mb_width && h->slice_table[mb_xy - s->mb_stride + 1] == h->slice_num;
2907 h->mb_avail[3]= s->mb_x && h->slice_table[mb_xy - 1] == h->slice_num;
2908 h->mb_avail[4]= 1; //FIXME move out
2909 h->mb_avail[5]= 0; //FIXME move out
2917 #define SIZE (COUNT*40)
2923 // int int_temp[10000];
2925 AVCodecContext avctx;
2927 dsputil_init(&dsp, &avctx);
2929 init_put_bits(&pb, temp, SIZE);
2930 printf("testing unsigned exp golomb\n");
2931 for(i=0; i<COUNT; i++){
2933 set_ue_golomb(&pb, i);
2934 STOP_TIMER("set_ue_golomb");
2936 flush_put_bits(&pb);
2938 init_get_bits(&gb, temp, 8*SIZE);
2939 for(i=0; i<COUNT; i++){
2942 s= show_bits(&gb, 24);
2945 j= get_ue_golomb(&gb);
2947 printf("mismatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
2950 STOP_TIMER("get_ue_golomb");
2954 init_put_bits(&pb, temp, SIZE);
2955 printf("testing signed exp golomb\n");
2956 for(i=0; i<COUNT; i++){
2958 set_se_golomb(&pb, i - COUNT/2);
2959 STOP_TIMER("set_se_golomb");
2961 flush_put_bits(&pb);
2963 init_get_bits(&gb, temp, 8*SIZE);
2964 for(i=0; i<COUNT; i++){
2967 s= show_bits(&gb, 24);
2970 j= get_se_golomb(&gb);
2971 if(j != i - COUNT/2){
2972 printf("mismatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
2975 STOP_TIMER("get_se_golomb");
2979 printf("testing 4x4 (I)DCT\n");
2982 uint8_t src[16], ref[16];
2983 uint64_t error= 0, max_error=0;
2985 for(i=0; i<COUNT; i++){
2987 // printf("%d %d %d\n", r1, r2, (r2-r1)*16);
2988 for(j=0; j<16; j++){
2989 ref[j]= random()%255;
2990 src[j]= random()%255;
2993 h264_diff_dct_c(block, src, ref, 4);
2996 for(j=0; j<16; j++){
2997 // printf("%d ", block[j]);
2998 block[j]= block[j]*4;
2999 if(j&1) block[j]= (block[j]*4 + 2)/5;
3000 if(j&4) block[j]= (block[j]*4 + 2)/5;
3004 s->dsp.h264_idct_add(ref, block, 4);
3005 /* for(j=0; j<16; j++){
3006 printf("%d ", ref[j]);
3010 for(j=0; j<16; j++){
3011 int diff= FFABS(src[j] - ref[j]);
3014 max_error= FFMAX(max_error, diff);
3017 printf("error=%f max_error=%d\n", ((float)error)/COUNT/16, (int)max_error );
3018 printf("testing quantizer\n");
3019 for(qp=0; qp<52; qp++){
3021 src1_block[i]= src2_block[i]= random()%255;
3024 printf("Testing NAL layer\n");
3026 uint8_t bitstream[COUNT];
3027 uint8_t nal[COUNT*2];
3029 memset(&h, 0, sizeof(H264Context));
3031 for(i=0; i<COUNT; i++){
3039 for(j=0; j<COUNT; j++){
3040 bitstream[j]= (random() % 255) + 1;
3043 for(j=0; j<zeros; j++){
3044 int pos= random() % COUNT;
3045 while(bitstream[pos] == 0){
3054 nal_length= encode_nal(&h, nal, bitstream, COUNT, COUNT*2);
3056 printf("encoding failed\n");
3060 out= ff_h264_decode_nal(&h, nal, &out_length, &consumed, nal_length);
3064 if(out_length != COUNT){
3065 printf("incorrect length %d %d\n", out_length, COUNT);
3069 if(consumed != nal_length){
3070 printf("incorrect consumed length %d %d\n", nal_length, consumed);
3074 if(memcmp(bitstream, out, COUNT)){
3075 printf("mismatch\n");
3081 printf("Testing RBSP\n");
3089 av_cold void ff_h264_free_context(H264Context *h)
3093 free_tables(h); //FIXME cleanup init stuff perhaps
3095 for(i = 0; i < MAX_SPS_COUNT; i++)
3096 av_freep(h->sps_buffers + i);
3098 for(i = 0; i < MAX_PPS_COUNT; i++)
3099 av_freep(h->pps_buffers + i);
3102 av_cold int ff_h264_decode_end(AVCodecContext *avctx)
3104 H264Context *h = avctx->priv_data;
3105 MpegEncContext *s = &h->s;
3107 ff_h264_free_context(h);
3111 // memset(h, 0, sizeof(H264Context));
3117 AVCodec h264_decoder = {
3121 sizeof(H264Context),
3122 ff_h264_decode_init,
3126 /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 | CODEC_CAP_DELAY,
3128 .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10"),
3129 .pix_fmts= ff_hwaccel_pixfmt_list_420,
3132 #if CONFIG_H264_VDPAU_DECODER
3133 AVCodec h264_vdpau_decoder = {
3137 sizeof(H264Context),
3138 ff_h264_decode_init,
3142 CODEC_CAP_DR1 | CODEC_CAP_DELAY | CODEC_CAP_HWACCEL_VDPAU,
3144 .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10 (VDPAU acceleration)"),
3145 .pix_fmts = (const enum PixelFormat[]){PIX_FMT_VDPAU_H264, PIX_FMT_NONE},