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
24 * H.264 / AVC / MPEG4 part10 codec.
25 * @author Michael Niedermayer <michaelni@gmx.at>
28 #include "libavcore/imgutils.h"
32 #include "mpegvideo.h"
35 #include "h264_mvpred.h"
36 #include "h264_parser.h"
39 #include "rectangle.h"
40 #include "vdpau_internal.h"
47 static const uint8_t rem6[52]={
48 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,
51 static const uint8_t div6[52]={
52 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,
55 static const enum PixelFormat hwaccel_pixfmt_list_h264_jpeg_420[] = {
62 void ff_h264_write_back_intra_pred_mode(H264Context *h){
63 int8_t *mode= h->intra4x4_pred_mode + h->mb2br_xy[h->mb_xy];
65 AV_COPY32(mode, h->intra4x4_pred_mode_cache + 4 + 8*4);
66 mode[4]= h->intra4x4_pred_mode_cache[7+8*3];
67 mode[5]= h->intra4x4_pred_mode_cache[7+8*2];
68 mode[6]= h->intra4x4_pred_mode_cache[7+8*1];
72 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
74 int ff_h264_check_intra4x4_pred_mode(H264Context *h){
75 MpegEncContext * const s = &h->s;
76 static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
77 static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
80 if(!(h->top_samples_available&0x8000)){
82 int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
84 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);
87 h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
92 if((h->left_samples_available&0x8888)!=0x8888){
93 static const int mask[4]={0x8000,0x2000,0x80,0x20};
95 if(!(h->left_samples_available&mask[i])){
96 int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
98 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);
101 h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
108 } //FIXME cleanup like ff_h264_check_intra_pred_mode
111 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
113 int ff_h264_check_intra_pred_mode(H264Context *h, int mode){
114 MpegEncContext * const s = &h->s;
115 static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
116 static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
119 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);
123 if(!(h->top_samples_available&0x8000)){
126 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);
131 if((h->left_samples_available&0x8080) != 0x8080){
133 if(h->left_samples_available&0x8080){ //mad cow disease mode, aka MBAFF + constrained_intra_pred
134 mode= ALZHEIMER_DC_L0T_PRED8x8 + (!(h->left_samples_available&0x8000)) + 2*(mode == DC_128_PRED8x8);
137 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);
145 const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src, int *dst_length, int *consumed, int length){
150 // src[0]&0x80; //forbidden bit
151 h->nal_ref_idc= src[0]>>5;
152 h->nal_unit_type= src[0]&0x1F;
156 for(i=0; i<length; i++)
157 printf("%2X ", src[i]);
160 #if HAVE_FAST_UNALIGNED
163 for(i=0; i+1<length; i+=9){
164 if(!((~AV_RN64A(src+i) & (AV_RN64A(src+i) - 0x0100010001000101ULL)) & 0x8000800080008080ULL))
167 for(i=0; i+1<length; i+=5){
168 if(!((~AV_RN32A(src+i) & (AV_RN32A(src+i) - 0x01000101U)) & 0x80008080U))
171 if(i>0 && !src[i]) i--;
175 for(i=0; i+1<length; i+=2){
177 if(i>0 && src[i-1]==0) i--;
179 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
181 /* startcode, so we must be past the end */
189 if(i>=length-1){ //no escaped 0
191 *consumed= length+1; //+1 for the header
195 bufidx = h->nal_unit_type == NAL_DPC ? 1 : 0; // use second escape buffer for inter data
196 av_fast_malloc(&h->rbsp_buffer[bufidx], &h->rbsp_buffer_size[bufidx], length+FF_INPUT_BUFFER_PADDING_SIZE);
197 dst= h->rbsp_buffer[bufidx];
203 //printf("decoding esc\n");
207 //remove escapes (very rare 1:2^22)
209 dst[di++]= src[si++];
210 dst[di++]= src[si++];
211 }else if(src[si]==0 && src[si+1]==0){
212 if(src[si+2]==3){ //escape
217 }else //next start code
221 dst[di++]= src[si++];
224 dst[di++]= src[si++];
227 memset(dst+di, 0, FF_INPUT_BUFFER_PADDING_SIZE);
230 *consumed= si + 1;//+1 for the header
231 //FIXME store exact number of bits in the getbitcontext (it is needed for decoding)
235 int ff_h264_decode_rbsp_trailing(H264Context *h, const uint8_t *src){
239 tprintf(h->s.avctx, "rbsp trailing %X\n", v);
249 * IDCT transforms the 16 dc values and dequantizes them.
250 * @param qp quantization parameter
252 static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
255 int temp[16]; //FIXME check if this is a good idea
256 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
257 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
259 //memset(block, 64, 2*256);
262 const int offset= y_offset[i];
263 const int z0= block[offset+stride*0] + block[offset+stride*4];
264 const int z1= block[offset+stride*0] - block[offset+stride*4];
265 const int z2= block[offset+stride*1] - block[offset+stride*5];
266 const int z3= block[offset+stride*1] + block[offset+stride*5];
275 const int offset= x_offset[i];
276 const int z0= temp[4*0+i] + temp[4*2+i];
277 const int z1= temp[4*0+i] - temp[4*2+i];
278 const int z2= temp[4*1+i] - temp[4*3+i];
279 const int z3= temp[4*1+i] + temp[4*3+i];
281 block[stride*0 +offset]= ((((z0 + z3)*qmul + 128 ) >> 8)); //FIXME think about merging this into decode_residual
282 block[stride*2 +offset]= ((((z1 + z2)*qmul + 128 ) >> 8));
283 block[stride*8 +offset]= ((((z1 - z2)*qmul + 128 ) >> 8));
284 block[stride*10+offset]= ((((z0 - z3)*qmul + 128 ) >> 8));
290 * DCT transforms the 16 dc values.
291 * @param qp quantization parameter ??? FIXME
293 static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
294 // const int qmul= dequant_coeff[qp][0];
296 int temp[16]; //FIXME check if this is a good idea
297 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
298 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
301 const int offset= y_offset[i];
302 const int z0= block[offset+stride*0] + block[offset+stride*4];
303 const int z1= block[offset+stride*0] - block[offset+stride*4];
304 const int z2= block[offset+stride*1] - block[offset+stride*5];
305 const int z3= block[offset+stride*1] + block[offset+stride*5];
314 const int offset= x_offset[i];
315 const int z0= temp[4*0+i] + temp[4*2+i];
316 const int z1= temp[4*0+i] - temp[4*2+i];
317 const int z2= temp[4*1+i] - temp[4*3+i];
318 const int z3= temp[4*1+i] + temp[4*3+i];
320 block[stride*0 +offset]= (z0 + z3)>>1;
321 block[stride*2 +offset]= (z1 + z2)>>1;
322 block[stride*8 +offset]= (z1 - z2)>>1;
323 block[stride*10+offset]= (z0 - z3)>>1;
331 static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
332 const int stride= 16*2;
333 const int xStride= 16;
336 a= block[stride*0 + xStride*0];
337 b= block[stride*0 + xStride*1];
338 c= block[stride*1 + xStride*0];
339 d= block[stride*1 + xStride*1];
346 block[stride*0 + xStride*0]= ((a+c)*qmul) >> 7;
347 block[stride*0 + xStride*1]= ((e+b)*qmul) >> 7;
348 block[stride*1 + xStride*0]= ((a-c)*qmul) >> 7;
349 block[stride*1 + xStride*1]= ((e-b)*qmul) >> 7;
353 static void chroma_dc_dct_c(DCTELEM *block){
354 const int stride= 16*2;
355 const int xStride= 16;
358 a= block[stride*0 + xStride*0];
359 b= block[stride*0 + xStride*1];
360 c= block[stride*1 + xStride*0];
361 d= block[stride*1 + xStride*1];
368 block[stride*0 + xStride*0]= (a+c);
369 block[stride*0 + xStride*1]= (e+b);
370 block[stride*1 + xStride*0]= (a-c);
371 block[stride*1 + xStride*1]= (e-b);
375 static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
376 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
377 int src_x_offset, int src_y_offset,
378 qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
379 MpegEncContext * const s = &h->s;
380 const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
381 int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
382 const int luma_xy= (mx&3) + ((my&3)<<2);
383 uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*h->mb_linesize;
384 uint8_t * src_cb, * src_cr;
385 int extra_width= h->emu_edge_width;
386 int extra_height= h->emu_edge_height;
388 const int full_mx= mx>>2;
389 const int full_my= my>>2;
390 const int pic_width = 16*s->mb_width;
391 const int pic_height = 16*s->mb_height >> MB_FIELD;
393 if(mx&7) extra_width -= 3;
394 if(my&7) extra_height -= 3;
396 if( full_mx < 0-extra_width
397 || full_my < 0-extra_height
398 || full_mx + 16/*FIXME*/ > pic_width + extra_width
399 || full_my + 16/*FIXME*/ > pic_height + extra_height){
400 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);
401 src_y= s->edge_emu_buffer + 2 + 2*h->mb_linesize;
405 qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); //FIXME try variable height perhaps?
407 qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
410 if(CONFIG_GRAY && s->flags&CODEC_FLAG_GRAY) return;
413 // chroma offset when predicting from a field of opposite parity
414 my += 2 * ((s->mb_y & 1) - (pic->reference - 1));
415 emu |= (my>>3) < 0 || (my>>3) + 8 >= (pic_height>>1);
417 src_cb= pic->data[1] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
418 src_cr= pic->data[2] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
421 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);
422 src_cb= s->edge_emu_buffer;
424 chroma_op(dest_cb, src_cb, h->mb_uvlinesize, chroma_height, mx&7, my&7);
427 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);
428 src_cr= s->edge_emu_buffer;
430 chroma_op(dest_cr, src_cr, h->mb_uvlinesize, chroma_height, mx&7, my&7);
433 static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
434 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
435 int x_offset, int y_offset,
436 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
437 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
438 int list0, int list1){
439 MpegEncContext * const s = &h->s;
440 qpel_mc_func *qpix_op= qpix_put;
441 h264_chroma_mc_func chroma_op= chroma_put;
443 dest_y += 2*x_offset + 2*y_offset*h-> mb_linesize;
444 dest_cb += x_offset + y_offset*h->mb_uvlinesize;
445 dest_cr += x_offset + y_offset*h->mb_uvlinesize;
446 x_offset += 8*s->mb_x;
447 y_offset += 8*(s->mb_y >> MB_FIELD);
450 Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
451 mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
452 dest_y, dest_cb, dest_cr, x_offset, y_offset,
456 chroma_op= chroma_avg;
460 Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
461 mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
462 dest_y, dest_cb, dest_cr, x_offset, y_offset,
467 static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
468 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
469 int x_offset, int y_offset,
470 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
471 h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
472 h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
473 int list0, int list1){
474 MpegEncContext * const s = &h->s;
476 dest_y += 2*x_offset + 2*y_offset*h-> mb_linesize;
477 dest_cb += x_offset + y_offset*h->mb_uvlinesize;
478 dest_cr += x_offset + y_offset*h->mb_uvlinesize;
479 x_offset += 8*s->mb_x;
480 y_offset += 8*(s->mb_y >> MB_FIELD);
483 /* don't optimize for luma-only case, since B-frames usually
484 * use implicit weights => chroma too. */
485 uint8_t *tmp_cb = s->obmc_scratchpad;
486 uint8_t *tmp_cr = s->obmc_scratchpad + 8;
487 uint8_t *tmp_y = s->obmc_scratchpad + 8*h->mb_uvlinesize;
488 int refn0 = h->ref_cache[0][ scan8[n] ];
489 int refn1 = h->ref_cache[1][ scan8[n] ];
491 mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
492 dest_y, dest_cb, dest_cr,
493 x_offset, y_offset, qpix_put, chroma_put);
494 mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
495 tmp_y, tmp_cb, tmp_cr,
496 x_offset, y_offset, qpix_put, chroma_put);
498 if(h->use_weight == 2){
499 int weight0 = h->implicit_weight[refn0][refn1][s->mb_y&1];
500 int weight1 = 64 - weight0;
501 luma_weight_avg( dest_y, tmp_y, h-> mb_linesize, 5, weight0, weight1, 0);
502 chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, 5, weight0, weight1, 0);
503 chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, 5, weight0, weight1, 0);
505 luma_weight_avg(dest_y, tmp_y, h->mb_linesize, h->luma_log2_weight_denom,
506 h->luma_weight[refn0][0][0] , h->luma_weight[refn1][1][0],
507 h->luma_weight[refn0][0][1] + h->luma_weight[refn1][1][1]);
508 chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
509 h->chroma_weight[refn0][0][0][0] , h->chroma_weight[refn1][1][0][0],
510 h->chroma_weight[refn0][0][0][1] + h->chroma_weight[refn1][1][0][1]);
511 chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
512 h->chroma_weight[refn0][0][1][0] , h->chroma_weight[refn1][1][1][0],
513 h->chroma_weight[refn0][0][1][1] + h->chroma_weight[refn1][1][1][1]);
516 int list = list1 ? 1 : 0;
517 int refn = h->ref_cache[list][ scan8[n] ];
518 Picture *ref= &h->ref_list[list][refn];
519 mc_dir_part(h, ref, n, square, chroma_height, delta, list,
520 dest_y, dest_cb, dest_cr, x_offset, y_offset,
521 qpix_put, chroma_put);
523 luma_weight_op(dest_y, h->mb_linesize, h->luma_log2_weight_denom,
524 h->luma_weight[refn][list][0], h->luma_weight[refn][list][1]);
525 if(h->use_weight_chroma){
526 chroma_weight_op(dest_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
527 h->chroma_weight[refn][list][0][0], h->chroma_weight[refn][list][0][1]);
528 chroma_weight_op(dest_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
529 h->chroma_weight[refn][list][1][0], h->chroma_weight[refn][list][1][1]);
534 static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
535 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
536 int x_offset, int y_offset,
537 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
538 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
539 h264_weight_func *weight_op, h264_biweight_func *weight_avg,
540 int list0, int list1){
541 if((h->use_weight==2 && list0 && list1
542 && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ][h->s.mb_y&1] != 32))
544 mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
545 x_offset, y_offset, qpix_put, chroma_put,
546 weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
548 mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
549 x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
552 static inline void prefetch_motion(H264Context *h, int list){
553 /* fetch pixels for estimated mv 4 macroblocks ahead
554 * optimized for 64byte cache lines */
555 MpegEncContext * const s = &h->s;
556 const int refn = h->ref_cache[list][scan8[0]];
558 const int mx= (h->mv_cache[list][scan8[0]][0]>>2) + 16*s->mb_x + 8;
559 const int my= (h->mv_cache[list][scan8[0]][1]>>2) + 16*s->mb_y;
560 uint8_t **src= h->ref_list[list][refn].data;
561 int off= mx + (my + (s->mb_x&3)*4)*h->mb_linesize + 64;
562 s->dsp.prefetch(src[0]+off, s->linesize, 4);
563 off= (mx>>1) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + 64;
564 s->dsp.prefetch(src[1]+off, src[2]-src[1], 2);
568 static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
569 qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
570 qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
571 h264_weight_func *weight_op, h264_biweight_func *weight_avg){
572 MpegEncContext * const s = &h->s;
573 const int mb_xy= h->mb_xy;
574 const int mb_type= s->current_picture.mb_type[mb_xy];
576 assert(IS_INTER(mb_type));
578 prefetch_motion(h, 0);
580 if(IS_16X16(mb_type)){
581 mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
582 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
583 weight_op, weight_avg,
584 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
585 }else if(IS_16X8(mb_type)){
586 mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
587 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
588 &weight_op[1], &weight_avg[1],
589 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
590 mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
591 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
592 &weight_op[1], &weight_avg[1],
593 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
594 }else if(IS_8X16(mb_type)){
595 mc_part(h, 0, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 0, 0,
596 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
597 &weight_op[2], &weight_avg[2],
598 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
599 mc_part(h, 4, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 4, 0,
600 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
601 &weight_op[2], &weight_avg[2],
602 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
606 assert(IS_8X8(mb_type));
609 const int sub_mb_type= h->sub_mb_type[i];
611 int x_offset= (i&1)<<2;
612 int y_offset= (i&2)<<1;
614 if(IS_SUB_8X8(sub_mb_type)){
615 mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
616 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
617 &weight_op[3], &weight_avg[3],
618 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
619 }else if(IS_SUB_8X4(sub_mb_type)){
620 mc_part(h, n , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
621 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
622 &weight_op[4], &weight_avg[4],
623 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
624 mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
625 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
626 &weight_op[4], &weight_avg[4],
627 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
628 }else if(IS_SUB_4X8(sub_mb_type)){
629 mc_part(h, n , 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
630 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
631 &weight_op[5], &weight_avg[5],
632 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
633 mc_part(h, n+1, 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
634 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
635 &weight_op[5], &weight_avg[5],
636 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
639 assert(IS_SUB_4X4(sub_mb_type));
641 int sub_x_offset= x_offset + 2*(j&1);
642 int sub_y_offset= y_offset + (j&2);
643 mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
644 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
645 &weight_op[6], &weight_avg[6],
646 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
652 prefetch_motion(h, 1);
656 static void free_tables(H264Context *h){
659 av_freep(&h->intra4x4_pred_mode);
660 av_freep(&h->chroma_pred_mode_table);
661 av_freep(&h->cbp_table);
662 av_freep(&h->mvd_table[0]);
663 av_freep(&h->mvd_table[1]);
664 av_freep(&h->direct_table);
665 av_freep(&h->non_zero_count);
666 av_freep(&h->slice_table_base);
667 h->slice_table= NULL;
668 av_freep(&h->list_counts);
670 av_freep(&h->mb2b_xy);
671 av_freep(&h->mb2br_xy);
673 for(i = 0; i < MAX_THREADS; i++) {
674 hx = h->thread_context[i];
676 av_freep(&hx->top_borders[1]);
677 av_freep(&hx->top_borders[0]);
678 av_freep(&hx->s.obmc_scratchpad);
679 av_freep(&hx->rbsp_buffer[1]);
680 av_freep(&hx->rbsp_buffer[0]);
681 hx->rbsp_buffer_size[0] = 0;
682 hx->rbsp_buffer_size[1] = 0;
683 if (i) av_freep(&h->thread_context[i]);
687 static void init_dequant8_coeff_table(H264Context *h){
689 const int transpose = (h->h264dsp.h264_idct8_add != ff_h264_idct8_add_c); //FIXME ugly
690 h->dequant8_coeff[0] = h->dequant8_buffer[0];
691 h->dequant8_coeff[1] = h->dequant8_buffer[1];
694 if(i && !memcmp(h->pps.scaling_matrix8[0], h->pps.scaling_matrix8[1], 64*sizeof(uint8_t))){
695 h->dequant8_coeff[1] = h->dequant8_buffer[0];
703 h->dequant8_coeff[i][q][transpose ? (x>>3)|((x&7)<<3) : x] =
704 ((uint32_t)dequant8_coeff_init[idx][ dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] *
705 h->pps.scaling_matrix8[i][x]) << shift;
710 static void init_dequant4_coeff_table(H264Context *h){
712 const int transpose = (h->h264dsp.h264_idct_add != ff_h264_idct_add_c); //FIXME ugly
714 h->dequant4_coeff[i] = h->dequant4_buffer[i];
716 if(!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i], 16*sizeof(uint8_t))){
717 h->dequant4_coeff[i] = h->dequant4_buffer[j];
725 int shift = div6[q] + 2;
728 h->dequant4_coeff[i][q][transpose ? (x>>2)|((x<<2)&0xF) : x] =
729 ((uint32_t)dequant4_coeff_init[idx][(x&1) + ((x>>2)&1)] *
730 h->pps.scaling_matrix4[i][x]) << shift;
735 static void init_dequant_tables(H264Context *h){
737 init_dequant4_coeff_table(h);
738 if(h->pps.transform_8x8_mode)
739 init_dequant8_coeff_table(h);
740 if(h->sps.transform_bypass){
743 h->dequant4_coeff[i][0][x] = 1<<6;
744 if(h->pps.transform_8x8_mode)
747 h->dequant8_coeff[i][0][x] = 1<<6;
752 int ff_h264_alloc_tables(H264Context *h){
753 MpegEncContext * const s = &h->s;
754 const int big_mb_num= s->mb_stride * (s->mb_height+1);
755 const int row_mb_num= 2*s->mb_stride*s->avctx->thread_count;
758 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->intra4x4_pred_mode, row_mb_num * 8 * sizeof(uint8_t), fail)
760 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->non_zero_count , big_mb_num * 32 * sizeof(uint8_t), fail)
761 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->slice_table_base , (big_mb_num+s->mb_stride) * sizeof(*h->slice_table_base), fail)
762 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->cbp_table, big_mb_num * sizeof(uint16_t), fail)
764 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t), fail)
765 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[0], 16*row_mb_num * sizeof(uint8_t), fail);
766 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[1], 16*row_mb_num * sizeof(uint8_t), fail);
767 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->direct_table, 4*big_mb_num * sizeof(uint8_t) , fail);
768 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->list_counts, big_mb_num * sizeof(uint8_t), fail)
770 memset(h->slice_table_base, -1, (big_mb_num+s->mb_stride) * sizeof(*h->slice_table_base));
771 h->slice_table= h->slice_table_base + s->mb_stride*2 + 1;
773 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2b_xy , big_mb_num * sizeof(uint32_t), fail);
774 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2br_xy , big_mb_num * sizeof(uint32_t), fail);
775 for(y=0; y<s->mb_height; y++){
776 for(x=0; x<s->mb_width; x++){
777 const int mb_xy= x + y*s->mb_stride;
778 const int b_xy = 4*x + 4*y*h->b_stride;
780 h->mb2b_xy [mb_xy]= b_xy;
781 h->mb2br_xy[mb_xy]= 8*(FMO ? mb_xy : (mb_xy % (2*s->mb_stride)));
785 s->obmc_scratchpad = NULL;
787 if(!h->dequant4_coeff[0])
788 init_dequant_tables(h);
797 * Mimic alloc_tables(), but for every context thread.
799 static void clone_tables(H264Context *dst, H264Context *src, int i){
800 MpegEncContext * const s = &src->s;
801 dst->intra4x4_pred_mode = src->intra4x4_pred_mode + i*8*2*s->mb_stride;
802 dst->non_zero_count = src->non_zero_count;
803 dst->slice_table = src->slice_table;
804 dst->cbp_table = src->cbp_table;
805 dst->mb2b_xy = src->mb2b_xy;
806 dst->mb2br_xy = src->mb2br_xy;
807 dst->chroma_pred_mode_table = src->chroma_pred_mode_table;
808 dst->mvd_table[0] = src->mvd_table[0] + i*8*2*s->mb_stride;
809 dst->mvd_table[1] = src->mvd_table[1] + i*8*2*s->mb_stride;
810 dst->direct_table = src->direct_table;
811 dst->list_counts = src->list_counts;
813 dst->s.obmc_scratchpad = NULL;
814 ff_h264_pred_init(&dst->hpc, src->s.codec_id);
819 * Allocate buffers which are not shared amongst multiple threads.
821 static int context_init(H264Context *h){
822 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[0], h->s.mb_width * (16+8+8) * sizeof(uint8_t), fail)
823 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[1], h->s.mb_width * (16+8+8) * sizeof(uint8_t), fail)
825 h->ref_cache[0][scan8[5 ]+1] = h->ref_cache[0][scan8[7 ]+1] = h->ref_cache[0][scan8[13]+1] =
826 h->ref_cache[1][scan8[5 ]+1] = h->ref_cache[1][scan8[7 ]+1] = h->ref_cache[1][scan8[13]+1] = PART_NOT_AVAILABLE;
830 return -1; // free_tables will clean up for us
833 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size);
835 static av_cold void common_init(H264Context *h){
836 MpegEncContext * const s = &h->s;
838 s->width = s->avctx->width;
839 s->height = s->avctx->height;
840 s->codec_id= s->avctx->codec->id;
842 ff_h264dsp_init(&h->h264dsp);
843 ff_h264_pred_init(&h->hpc, s->codec_id);
845 h->dequant_coeff_pps= -1;
846 s->unrestricted_mv=1;
849 dsputil_init(&s->dsp, s->avctx); // needed so that idct permutation is known early
851 memset(h->pps.scaling_matrix4, 16, 6*16*sizeof(uint8_t));
852 memset(h->pps.scaling_matrix8, 16, 2*64*sizeof(uint8_t));
855 int ff_h264_decode_extradata(H264Context *h)
857 AVCodecContext *avctx = h->s.avctx;
859 if(*(char *)avctx->extradata == 1){
861 unsigned char *p = avctx->extradata;
865 if(avctx->extradata_size < 7) {
866 av_log(avctx, AV_LOG_ERROR, "avcC too short\n");
869 /* sps and pps in the avcC always have length coded with 2 bytes,
870 so put a fake nal_length_size = 2 while parsing them */
871 h->nal_length_size = 2;
872 // Decode sps from avcC
873 cnt = *(p+5) & 0x1f; // Number of sps
875 for (i = 0; i < cnt; i++) {
876 nalsize = AV_RB16(p) + 2;
877 if(decode_nal_units(h, p, nalsize) < 0) {
878 av_log(avctx, AV_LOG_ERROR, "Decoding sps %d from avcC failed\n", i);
883 // Decode pps from avcC
884 cnt = *(p++); // Number of pps
885 for (i = 0; i < cnt; i++) {
886 nalsize = AV_RB16(p) + 2;
887 if(decode_nal_units(h, p, nalsize) != nalsize) {
888 av_log(avctx, AV_LOG_ERROR, "Decoding pps %d from avcC failed\n", i);
893 // Now store right nal length size, that will be use to parse all other nals
894 h->nal_length_size = ((*(((char*)(avctx->extradata))+4))&0x03)+1;
897 if(decode_nal_units(h, avctx->extradata, avctx->extradata_size) < 0)
903 av_cold int ff_h264_decode_init(AVCodecContext *avctx){
904 H264Context *h= avctx->priv_data;
905 MpegEncContext * const s = &h->s;
907 MPV_decode_defaults(s);
912 s->out_format = FMT_H264;
913 s->workaround_bugs= avctx->workaround_bugs;
916 // s->decode_mb= ff_h263_decode_mb;
917 s->quarter_sample = 1;
918 if(!avctx->has_b_frames)
921 avctx->chroma_sample_location = AVCHROMA_LOC_LEFT;
923 ff_h264_decode_init_vlc();
925 h->thread_context[0] = h;
926 h->outputed_poc = INT_MIN;
927 h->prev_poc_msb= 1<<16;
929 ff_h264_reset_sei(h);
930 if(avctx->codec_id == CODEC_ID_H264){
931 if(avctx->ticks_per_frame == 1){
932 s->avctx->time_base.den *=2;
934 avctx->ticks_per_frame = 2;
937 if(avctx->extradata_size > 0 && avctx->extradata &&
938 ff_h264_decode_extradata(h))
941 if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames < h->sps.num_reorder_frames){
942 s->avctx->has_b_frames = h->sps.num_reorder_frames;
949 int ff_h264_frame_start(H264Context *h){
950 MpegEncContext * const s = &h->s;
953 if(MPV_frame_start(s, s->avctx) < 0)
955 ff_er_frame_start(s);
957 * MPV_frame_start uses pict_type to derive key_frame.
958 * This is incorrect for H.264; IDR markings must be used.
959 * Zero here; IDR markings per slice in frame or fields are ORed in later.
960 * See decode_nal_units().
962 s->current_picture_ptr->key_frame= 0;
963 s->current_picture_ptr->mmco_reset= 0;
965 assert(s->linesize && s->uvlinesize);
968 h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
969 h->block_offset[24+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
972 h->block_offset[16+i]=
973 h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
974 h->block_offset[24+16+i]=
975 h->block_offset[24+20+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
978 /* can't be in alloc_tables because linesize isn't known there.
979 * FIXME: redo bipred weight to not require extra buffer? */
980 for(i = 0; i < s->avctx->thread_count; i++)
981 if(!h->thread_context[i]->s.obmc_scratchpad)
982 h->thread_context[i]->s.obmc_scratchpad = av_malloc(16*2*s->linesize + 8*2*s->uvlinesize);
984 /* some macroblocks can be accessed before they're available in case of lost slices, mbaff or threading*/
985 memset(h->slice_table, -1, (s->mb_height*s->mb_stride-1) * sizeof(*h->slice_table));
987 // s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
989 // We mark the current picture as non-reference after allocating it, so
990 // that if we break out due to an error it can be released automatically
991 // in the next MPV_frame_start().
992 // SVQ3 as well as most other codecs have only last/next/current and thus
993 // get released even with set reference, besides SVQ3 and others do not
994 // mark frames as reference later "naturally".
995 if(s->codec_id != CODEC_ID_SVQ3)
996 s->current_picture_ptr->reference= 0;
998 s->current_picture_ptr->field_poc[0]=
999 s->current_picture_ptr->field_poc[1]= INT_MAX;
1000 assert(s->current_picture_ptr->long_ref==0);
1005 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){
1006 MpegEncContext * const s = &h->s;
1007 uint8_t *top_border;
1011 src_cb -= uvlinesize;
1012 src_cr -= uvlinesize;
1014 if(!simple && FRAME_MBAFF){
1017 top_border = h->top_borders[0][s->mb_x];
1018 AV_COPY128(top_border, src_y + 15*linesize);
1019 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1020 AV_COPY64(top_border+16, src_cb+7*uvlinesize);
1021 AV_COPY64(top_border+24, src_cr+7*uvlinesize);
1030 top_border = h->top_borders[top_idx][s->mb_x];
1031 // There are two lines saved, the line above the the top macroblock of a pair,
1032 // and the line above the bottom macroblock
1033 AV_COPY128(top_border, src_y + 16*linesize);
1035 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1036 AV_COPY64(top_border+16, src_cb+8*uvlinesize);
1037 AV_COPY64(top_border+24, src_cr+8*uvlinesize);
1041 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){
1042 MpegEncContext * const s = &h->s;
1046 uint8_t *top_border_m1;
1047 uint8_t *top_border;
1049 if(!simple && FRAME_MBAFF){
1054 top_idx = MB_MBAFF ? 0 : 1;
1058 if(h->deblocking_filter == 2) {
1059 deblock_left = h->left_type[0];
1060 deblock_top = h->top_type;
1062 deblock_left = (s->mb_x > 0);
1063 deblock_top = (s->mb_y > !!MB_FIELD);
1066 src_y -= linesize + 1;
1067 src_cb -= uvlinesize + 1;
1068 src_cr -= uvlinesize + 1;
1070 top_border_m1 = h->top_borders[top_idx][s->mb_x-1];
1071 top_border = h->top_borders[top_idx][s->mb_x];
1073 #define XCHG(a,b,xchg)\
1074 if (xchg) AV_SWAP64(b,a);\
1075 else AV_COPY64(b,a);
1079 XCHG(top_border_m1+8, src_y -7, 1);
1081 XCHG(top_border+0, src_y +1, xchg);
1082 XCHG(top_border+8, src_y +9, 1);
1083 if(s->mb_x+1 < s->mb_width){
1084 XCHG(h->top_borders[top_idx][s->mb_x+1], src_y +17, 1);
1088 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1091 XCHG(top_border_m1+16, src_cb -7, 1);
1092 XCHG(top_border_m1+24, src_cr -7, 1);
1094 XCHG(top_border+16, src_cb+1, 1);
1095 XCHG(top_border+24, src_cr+1, 1);
1100 static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
1101 MpegEncContext * const s = &h->s;
1102 const int mb_x= s->mb_x;
1103 const int mb_y= s->mb_y;
1104 const int mb_xy= h->mb_xy;
1105 const int mb_type= s->current_picture.mb_type[mb_xy];
1106 uint8_t *dest_y, *dest_cb, *dest_cr;
1107 int linesize, uvlinesize /*dct_offset*/;
1109 int *block_offset = &h->block_offset[0];
1110 const int transform_bypass = !simple && (s->qscale == 0 && h->sps.transform_bypass);
1111 /* is_h264 should always be true if SVQ3 is disabled. */
1112 const int is_h264 = !CONFIG_SVQ3_DECODER || simple || s->codec_id == CODEC_ID_H264;
1113 void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1114 void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
1116 dest_y = s->current_picture.data[0] + (mb_x + mb_y * s->linesize ) * 16;
1117 dest_cb = s->current_picture.data[1] + (mb_x + mb_y * s->uvlinesize) * 8;
1118 dest_cr = s->current_picture.data[2] + (mb_x + mb_y * s->uvlinesize) * 8;
1120 s->dsp.prefetch(dest_y + (s->mb_x&3)*4*s->linesize + 64, s->linesize, 4);
1121 s->dsp.prefetch(dest_cb + (s->mb_x&7)*s->uvlinesize + 64, dest_cr - dest_cb, 2);
1123 h->list_counts[mb_xy]= h->list_count;
1125 if (!simple && MB_FIELD) {
1126 linesize = h->mb_linesize = s->linesize * 2;
1127 uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
1128 block_offset = &h->block_offset[24];
1129 if(mb_y&1){ //FIXME move out of this function?
1130 dest_y -= s->linesize*15;
1131 dest_cb-= s->uvlinesize*7;
1132 dest_cr-= s->uvlinesize*7;
1136 for(list=0; list<h->list_count; list++){
1137 if(!USES_LIST(mb_type, list))
1139 if(IS_16X16(mb_type)){
1140 int8_t *ref = &h->ref_cache[list][scan8[0]];
1141 fill_rectangle(ref, 4, 4, 8, (16+*ref)^(s->mb_y&1), 1);
1143 for(i=0; i<16; i+=4){
1144 int ref = h->ref_cache[list][scan8[i]];
1146 fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, (16+ref)^(s->mb_y&1), 1);
1152 linesize = h->mb_linesize = s->linesize;
1153 uvlinesize = h->mb_uvlinesize = s->uvlinesize;
1154 // dct_offset = s->linesize * 16;
1157 if (!simple && IS_INTRA_PCM(mb_type)) {
1158 for (i=0; i<16; i++) {
1159 memcpy(dest_y + i* linesize, h->mb + i*8, 16);
1161 for (i=0; i<8; i++) {
1162 memcpy(dest_cb+ i*uvlinesize, h->mb + 128 + i*4, 8);
1163 memcpy(dest_cr+ i*uvlinesize, h->mb + 160 + i*4, 8);
1166 if(IS_INTRA(mb_type)){
1167 if(h->deblocking_filter)
1168 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1, simple);
1170 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1171 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
1172 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
1175 if(IS_INTRA4x4(mb_type)){
1176 if(simple || !s->encoding){
1177 if(IS_8x8DCT(mb_type)){
1178 if(transform_bypass){
1180 idct_add = s->dsp.add_pixels8;
1182 idct_dc_add = h->h264dsp.h264_idct8_dc_add;
1183 idct_add = h->h264dsp.h264_idct8_add;
1185 for(i=0; i<16; i+=4){
1186 uint8_t * const ptr= dest_y + block_offset[i];
1187 const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1188 if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1189 h->hpc.pred8x8l_add[dir](ptr, h->mb + i*16, linesize);
1191 const int nnz = h->non_zero_count_cache[ scan8[i] ];
1192 h->hpc.pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
1193 (h->topright_samples_available<<i)&0x4000, linesize);
1195 if(nnz == 1 && h->mb[i*16])
1196 idct_dc_add(ptr, h->mb + i*16, linesize);
1198 idct_add (ptr, h->mb + i*16, linesize);
1203 if(transform_bypass){
1205 idct_add = s->dsp.add_pixels4;
1207 idct_dc_add = h->h264dsp.h264_idct_dc_add;
1208 idct_add = h->h264dsp.h264_idct_add;
1210 for(i=0; i<16; i++){
1211 uint8_t * const ptr= dest_y + block_offset[i];
1212 const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1214 if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1215 h->hpc.pred4x4_add[dir](ptr, h->mb + i*16, linesize);
1219 if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
1220 const int topright_avail= (h->topright_samples_available<<i)&0x8000;
1221 assert(mb_y || linesize <= block_offset[i]);
1222 if(!topright_avail){
1223 tr= ptr[3 - linesize]*0x01010101;
1224 topright= (uint8_t*) &tr;
1226 topright= ptr + 4 - linesize;
1230 h->hpc.pred4x4[ dir ](ptr, topright, linesize);
1231 nnz = h->non_zero_count_cache[ scan8[i] ];
1234 if(nnz == 1 && h->mb[i*16])
1235 idct_dc_add(ptr, h->mb + i*16, linesize);
1237 idct_add (ptr, h->mb + i*16, linesize);
1239 ff_svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
1246 h->hpc.pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
1248 if(!transform_bypass)
1249 h264_luma_dc_dequant_idct_c(h->mb, s->qscale, h->dequant4_coeff[0][s->qscale][0]);
1251 ff_svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
1253 if(h->deblocking_filter)
1254 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0, simple);
1256 hl_motion(h, dest_y, dest_cb, dest_cr,
1257 s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
1258 s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
1259 h->h264dsp.weight_h264_pixels_tab, h->h264dsp.biweight_h264_pixels_tab);
1263 if(!IS_INTRA4x4(mb_type)){
1265 if(IS_INTRA16x16(mb_type)){
1266 if(transform_bypass){
1267 if(h->sps.profile_idc==244 && (h->intra16x16_pred_mode==VERT_PRED8x8 || h->intra16x16_pred_mode==HOR_PRED8x8)){
1268 h->hpc.pred16x16_add[h->intra16x16_pred_mode](dest_y, block_offset, h->mb, linesize);
1270 for(i=0; i<16; i++){
1271 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16])
1272 s->dsp.add_pixels4(dest_y + block_offset[i], h->mb + i*16, linesize);
1276 h->h264dsp.h264_idct_add16intra(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1278 }else if(h->cbp&15){
1279 if(transform_bypass){
1280 const int di = IS_8x8DCT(mb_type) ? 4 : 1;
1281 idct_add= IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4;
1282 for(i=0; i<16; i+=di){
1283 if(h->non_zero_count_cache[ scan8[i] ]){
1284 idct_add(dest_y + block_offset[i], h->mb + i*16, linesize);
1288 if(IS_8x8DCT(mb_type)){
1289 h->h264dsp.h264_idct8_add4(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1291 h->h264dsp.h264_idct_add16(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1296 for(i=0; i<16; i++){
1297 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
1298 uint8_t * const ptr= dest_y + block_offset[i];
1299 ff_svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
1305 if((simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)) && (h->cbp&0x30)){
1306 uint8_t *dest[2] = {dest_cb, dest_cr};
1307 if(transform_bypass){
1308 if(IS_INTRA(mb_type) && h->sps.profile_idc==244 && (h->chroma_pred_mode==VERT_PRED8x8 || h->chroma_pred_mode==HOR_PRED8x8)){
1309 h->hpc.pred8x8_add[h->chroma_pred_mode](dest[0], block_offset + 16, h->mb + 16*16, uvlinesize);
1310 h->hpc.pred8x8_add[h->chroma_pred_mode](dest[1], block_offset + 20, h->mb + 20*16, uvlinesize);
1312 idct_add = s->dsp.add_pixels4;
1313 for(i=16; i<16+8; i++){
1314 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16])
1315 idct_add (dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
1319 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]);
1320 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]);
1322 h->h264dsp.h264_idct_add8(dest, block_offset,
1324 h->non_zero_count_cache);
1326 for(i=16; i<16+8; i++){
1327 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
1328 uint8_t * const ptr= dest[(i&4)>>2] + block_offset[i];
1329 ff_svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, ff_h264_chroma_qp[s->qscale + 12] - 12, 2);
1336 if(h->cbp || IS_INTRA(mb_type))
1337 s->dsp.clear_blocks(h->mb);
1341 * Process a macroblock; this case avoids checks for expensive uncommon cases.
1343 static void hl_decode_mb_simple(H264Context *h){
1344 hl_decode_mb_internal(h, 1);
1348 * Process a macroblock; this handles edge cases, such as interlacing.
1350 static void av_noinline hl_decode_mb_complex(H264Context *h){
1351 hl_decode_mb_internal(h, 0);
1354 void ff_h264_hl_decode_mb(H264Context *h){
1355 MpegEncContext * const s = &h->s;
1356 const int mb_xy= h->mb_xy;
1357 const int mb_type= s->current_picture.mb_type[mb_xy];
1358 int is_complex = CONFIG_SMALL || h->is_complex || IS_INTRA_PCM(mb_type) || s->qscale == 0;
1361 hl_decode_mb_complex(h);
1362 else hl_decode_mb_simple(h);
1365 static int pred_weight_table(H264Context *h){
1366 MpegEncContext * const s = &h->s;
1368 int luma_def, chroma_def;
1371 h->use_weight_chroma= 0;
1372 h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
1373 h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
1374 luma_def = 1<<h->luma_log2_weight_denom;
1375 chroma_def = 1<<h->chroma_log2_weight_denom;
1377 for(list=0; list<2; list++){
1378 h->luma_weight_flag[list] = 0;
1379 h->chroma_weight_flag[list] = 0;
1380 for(i=0; i<h->ref_count[list]; i++){
1381 int luma_weight_flag, chroma_weight_flag;
1383 luma_weight_flag= get_bits1(&s->gb);
1384 if(luma_weight_flag){
1385 h->luma_weight[i][list][0]= get_se_golomb(&s->gb);
1386 h->luma_weight[i][list][1]= get_se_golomb(&s->gb);
1387 if( h->luma_weight[i][list][0] != luma_def
1388 || h->luma_weight[i][list][1] != 0) {
1390 h->luma_weight_flag[list]= 1;
1393 h->luma_weight[i][list][0]= luma_def;
1394 h->luma_weight[i][list][1]= 0;
1398 chroma_weight_flag= get_bits1(&s->gb);
1399 if(chroma_weight_flag){
1402 h->chroma_weight[i][list][j][0]= get_se_golomb(&s->gb);
1403 h->chroma_weight[i][list][j][1]= get_se_golomb(&s->gb);
1404 if( h->chroma_weight[i][list][j][0] != chroma_def
1405 || h->chroma_weight[i][list][j][1] != 0) {
1406 h->use_weight_chroma= 1;
1407 h->chroma_weight_flag[list]= 1;
1413 h->chroma_weight[i][list][j][0]= chroma_def;
1414 h->chroma_weight[i][list][j][1]= 0;
1419 if(h->slice_type_nos != FF_B_TYPE) break;
1421 h->use_weight= h->use_weight || h->use_weight_chroma;
1426 * Initialize implicit_weight table.
1427 * @param field 0/1 initialize the weight for interlaced MBAFF
1428 * -1 initializes the rest
1430 static void implicit_weight_table(H264Context *h, int field){
1431 MpegEncContext * const s = &h->s;
1432 int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
1434 for (i = 0; i < 2; i++) {
1435 h->luma_weight_flag[i] = 0;
1436 h->chroma_weight_flag[i] = 0;
1440 cur_poc = s->current_picture_ptr->poc;
1441 if( h->ref_count[0] == 1 && h->ref_count[1] == 1 && !FRAME_MBAFF
1442 && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
1444 h->use_weight_chroma= 0;
1448 ref_count0= h->ref_count[0];
1449 ref_count1= h->ref_count[1];
1451 cur_poc = s->current_picture_ptr->field_poc[field];
1453 ref_count0= 16+2*h->ref_count[0];
1454 ref_count1= 16+2*h->ref_count[1];
1458 h->use_weight_chroma= 2;
1459 h->luma_log2_weight_denom= 5;
1460 h->chroma_log2_weight_denom= 5;
1462 for(ref0=ref_start; ref0 < ref_count0; ref0++){
1463 int poc0 = h->ref_list[0][ref0].poc;
1464 for(ref1=ref_start; ref1 < ref_count1; ref1++){
1465 int poc1 = h->ref_list[1][ref1].poc;
1466 int td = av_clip(poc1 - poc0, -128, 127);
1469 int tb = av_clip(cur_poc - poc0, -128, 127);
1470 int tx = (16384 + (FFABS(td) >> 1)) / td;
1471 int dist_scale_factor = (tb*tx + 32) >> 8;
1472 if(dist_scale_factor >= -64 && dist_scale_factor <= 128)
1473 w = 64 - dist_scale_factor;
1476 h->implicit_weight[ref0][ref1][0]=
1477 h->implicit_weight[ref0][ref1][1]= w;
1479 h->implicit_weight[ref0][ref1][field]=w;
1486 * instantaneous decoder refresh.
1488 static void idr(H264Context *h){
1489 ff_h264_remove_all_refs(h);
1490 h->prev_frame_num= 0;
1491 h->prev_frame_num_offset= 0;
1496 /* forget old pics after a seek */
1497 static void flush_dpb(AVCodecContext *avctx){
1498 H264Context *h= avctx->priv_data;
1500 for(i=0; i<MAX_DELAYED_PIC_COUNT; i++) {
1501 if(h->delayed_pic[i])
1502 h->delayed_pic[i]->reference= 0;
1503 h->delayed_pic[i]= NULL;
1505 h->outputed_poc= INT_MIN;
1506 h->prev_interlaced_frame = 1;
1508 if(h->s.current_picture_ptr)
1509 h->s.current_picture_ptr->reference= 0;
1510 h->s.first_field= 0;
1511 ff_h264_reset_sei(h);
1512 ff_mpeg_flush(avctx);
1515 static int init_poc(H264Context *h){
1516 MpegEncContext * const s = &h->s;
1517 const int max_frame_num= 1<<h->sps.log2_max_frame_num;
1519 Picture *cur = s->current_picture_ptr;
1521 h->frame_num_offset= h->prev_frame_num_offset;
1522 if(h->frame_num < h->prev_frame_num)
1523 h->frame_num_offset += max_frame_num;
1525 if(h->sps.poc_type==0){
1526 const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
1528 if (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
1529 h->poc_msb = h->prev_poc_msb + max_poc_lsb;
1530 else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
1531 h->poc_msb = h->prev_poc_msb - max_poc_lsb;
1533 h->poc_msb = h->prev_poc_msb;
1534 //printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
1536 field_poc[1] = h->poc_msb + h->poc_lsb;
1537 if(s->picture_structure == PICT_FRAME)
1538 field_poc[1] += h->delta_poc_bottom;
1539 }else if(h->sps.poc_type==1){
1540 int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
1543 if(h->sps.poc_cycle_length != 0)
1544 abs_frame_num = h->frame_num_offset + h->frame_num;
1548 if(h->nal_ref_idc==0 && abs_frame_num > 0)
1551 expected_delta_per_poc_cycle = 0;
1552 for(i=0; i < h->sps.poc_cycle_length; i++)
1553 expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
1555 if(abs_frame_num > 0){
1556 int poc_cycle_cnt = (abs_frame_num - 1) / h->sps.poc_cycle_length;
1557 int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
1559 expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
1560 for(i = 0; i <= frame_num_in_poc_cycle; i++)
1561 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
1565 if(h->nal_ref_idc == 0)
1566 expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
1568 field_poc[0] = expectedpoc + h->delta_poc[0];
1569 field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
1571 if(s->picture_structure == PICT_FRAME)
1572 field_poc[1] += h->delta_poc[1];
1574 int poc= 2*(h->frame_num_offset + h->frame_num);
1583 if(s->picture_structure != PICT_BOTTOM_FIELD)
1584 s->current_picture_ptr->field_poc[0]= field_poc[0];
1585 if(s->picture_structure != PICT_TOP_FIELD)
1586 s->current_picture_ptr->field_poc[1]= field_poc[1];
1587 cur->poc= FFMIN(cur->field_poc[0], cur->field_poc[1]);
1594 * initialize scan tables
1596 static void init_scan_tables(H264Context *h){
1598 if(h->h264dsp.h264_idct_add == ff_h264_idct_add_c){ //FIXME little ugly
1599 memcpy(h->zigzag_scan, zigzag_scan, 16*sizeof(uint8_t));
1600 memcpy(h-> field_scan, field_scan, 16*sizeof(uint8_t));
1602 for(i=0; i<16; i++){
1603 #define T(x) (x>>2) | ((x<<2) & 0xF)
1604 h->zigzag_scan[i] = T(zigzag_scan[i]);
1605 h-> field_scan[i] = T( field_scan[i]);
1609 if(h->h264dsp.h264_idct8_add == ff_h264_idct8_add_c){
1610 memcpy(h->zigzag_scan8x8, ff_zigzag_direct, 64*sizeof(uint8_t));
1611 memcpy(h->zigzag_scan8x8_cavlc, zigzag_scan8x8_cavlc, 64*sizeof(uint8_t));
1612 memcpy(h->field_scan8x8, field_scan8x8, 64*sizeof(uint8_t));
1613 memcpy(h->field_scan8x8_cavlc, field_scan8x8_cavlc, 64*sizeof(uint8_t));
1615 for(i=0; i<64; i++){
1616 #define T(x) (x>>3) | ((x&7)<<3)
1617 h->zigzag_scan8x8[i] = T(ff_zigzag_direct[i]);
1618 h->zigzag_scan8x8_cavlc[i] = T(zigzag_scan8x8_cavlc[i]);
1619 h->field_scan8x8[i] = T(field_scan8x8[i]);
1620 h->field_scan8x8_cavlc[i] = T(field_scan8x8_cavlc[i]);
1624 if(h->sps.transform_bypass){ //FIXME same ugly
1625 h->zigzag_scan_q0 = zigzag_scan;
1626 h->zigzag_scan8x8_q0 = ff_zigzag_direct;
1627 h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc;
1628 h->field_scan_q0 = field_scan;
1629 h->field_scan8x8_q0 = field_scan8x8;
1630 h->field_scan8x8_cavlc_q0 = field_scan8x8_cavlc;
1632 h->zigzag_scan_q0 = h->zigzag_scan;
1633 h->zigzag_scan8x8_q0 = h->zigzag_scan8x8;
1634 h->zigzag_scan8x8_cavlc_q0 = h->zigzag_scan8x8_cavlc;
1635 h->field_scan_q0 = h->field_scan;
1636 h->field_scan8x8_q0 = h->field_scan8x8;
1637 h->field_scan8x8_cavlc_q0 = h->field_scan8x8_cavlc;
1641 static void field_end(H264Context *h){
1642 MpegEncContext * const s = &h->s;
1643 AVCodecContext * const avctx= s->avctx;
1646 s->current_picture_ptr->qscale_type= FF_QSCALE_TYPE_H264;
1647 s->current_picture_ptr->pict_type= s->pict_type;
1649 if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
1650 ff_vdpau_h264_set_reference_frames(s);
1653 ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1654 h->prev_poc_msb= h->poc_msb;
1655 h->prev_poc_lsb= h->poc_lsb;
1657 h->prev_frame_num_offset= h->frame_num_offset;
1658 h->prev_frame_num= h->frame_num;
1660 if (avctx->hwaccel) {
1661 if (avctx->hwaccel->end_frame(avctx) < 0)
1662 av_log(avctx, AV_LOG_ERROR, "hardware accelerator failed to decode picture\n");
1665 if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
1666 ff_vdpau_h264_picture_complete(s);
1669 * FIXME: Error handling code does not seem to support interlaced
1670 * when slices span multiple rows
1671 * The ff_er_add_slice calls don't work right for bottom
1672 * fields; they cause massive erroneous error concealing
1673 * Error marking covers both fields (top and bottom).
1674 * This causes a mismatched s->error_count
1675 * and a bad error table. Further, the error count goes to
1676 * INT_MAX when called for bottom field, because mb_y is
1677 * past end by one (callers fault) and resync_mb_y != 0
1678 * causes problems for the first MB line, too.
1689 * Replicate H264 "master" context to thread contexts.
1691 static void clone_slice(H264Context *dst, H264Context *src)
1693 memcpy(dst->block_offset, src->block_offset, sizeof(dst->block_offset));
1694 dst->s.current_picture_ptr = src->s.current_picture_ptr;
1695 dst->s.current_picture = src->s.current_picture;
1696 dst->s.linesize = src->s.linesize;
1697 dst->s.uvlinesize = src->s.uvlinesize;
1698 dst->s.first_field = src->s.first_field;
1700 dst->prev_poc_msb = src->prev_poc_msb;
1701 dst->prev_poc_lsb = src->prev_poc_lsb;
1702 dst->prev_frame_num_offset = src->prev_frame_num_offset;
1703 dst->prev_frame_num = src->prev_frame_num;
1704 dst->short_ref_count = src->short_ref_count;
1706 memcpy(dst->short_ref, src->short_ref, sizeof(dst->short_ref));
1707 memcpy(dst->long_ref, src->long_ref, sizeof(dst->long_ref));
1708 memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
1709 memcpy(dst->ref_list, src->ref_list, sizeof(dst->ref_list));
1711 memcpy(dst->dequant4_coeff, src->dequant4_coeff, sizeof(src->dequant4_coeff));
1712 memcpy(dst->dequant8_coeff, src->dequant8_coeff, sizeof(src->dequant8_coeff));
1716 * decodes a slice header.
1717 * This will also call MPV_common_init() and frame_start() as needed.
1719 * @param h h264context
1720 * @param h0 h264 master context (differs from 'h' when doing sliced based parallel decoding)
1722 * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
1724 static int decode_slice_header(H264Context *h, H264Context *h0){
1725 MpegEncContext * const s = &h->s;
1726 MpegEncContext * const s0 = &h0->s;
1727 unsigned int first_mb_in_slice;
1728 unsigned int pps_id;
1729 int num_ref_idx_active_override_flag;
1730 unsigned int slice_type, tmp, i, j;
1731 int default_ref_list_done = 0;
1732 int last_pic_structure;
1734 s->dropable= h->nal_ref_idc == 0;
1736 if((s->avctx->flags2 & CODEC_FLAG2_FAST) && !h->nal_ref_idc){
1737 s->me.qpel_put= s->dsp.put_2tap_qpel_pixels_tab;
1738 s->me.qpel_avg= s->dsp.avg_2tap_qpel_pixels_tab;
1740 s->me.qpel_put= s->dsp.put_h264_qpel_pixels_tab;
1741 s->me.qpel_avg= s->dsp.avg_h264_qpel_pixels_tab;
1744 first_mb_in_slice= get_ue_golomb(&s->gb);
1746 if(first_mb_in_slice == 0){ //FIXME better field boundary detection
1747 if(h0->current_slice && FIELD_PICTURE){
1751 h0->current_slice = 0;
1752 if (!s0->first_field)
1753 s->current_picture_ptr= NULL;
1756 slice_type= get_ue_golomb_31(&s->gb);
1758 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);
1763 h->slice_type_fixed=1;
1765 h->slice_type_fixed=0;
1767 slice_type= golomb_to_pict_type[ slice_type ];
1768 if (slice_type == FF_I_TYPE
1769 || (h0->current_slice != 0 && slice_type == h0->last_slice_type) ) {
1770 default_ref_list_done = 1;
1772 h->slice_type= slice_type;
1773 h->slice_type_nos= slice_type & 3;
1775 s->pict_type= h->slice_type; // to make a few old functions happy, it's wrong though
1777 pps_id= get_ue_golomb(&s->gb);
1778 if(pps_id>=MAX_PPS_COUNT){
1779 av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
1782 if(!h0->pps_buffers[pps_id]) {
1783 av_log(h->s.avctx, AV_LOG_ERROR, "non-existing PPS %u referenced\n", pps_id);
1786 h->pps= *h0->pps_buffers[pps_id];
1788 if(!h0->sps_buffers[h->pps.sps_id]) {
1789 av_log(h->s.avctx, AV_LOG_ERROR, "non-existing SPS %u referenced\n", h->pps.sps_id);
1792 h->sps = *h0->sps_buffers[h->pps.sps_id];
1794 s->avctx->profile = h->sps.profile_idc;
1795 s->avctx->level = h->sps.level_idc;
1796 s->avctx->refs = h->sps.ref_frame_count;
1798 if(h == h0 && h->dequant_coeff_pps != pps_id){
1799 h->dequant_coeff_pps = pps_id;
1800 init_dequant_tables(h);
1803 s->mb_width= h->sps.mb_width;
1804 s->mb_height= h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
1806 h->b_stride= s->mb_width*4;
1808 s->width = 16*s->mb_width - 2*FFMIN(h->sps.crop_right, 7);
1809 if(h->sps.frame_mbs_only_flag)
1810 s->height= 16*s->mb_height - 2*FFMIN(h->sps.crop_bottom, 7);
1812 s->height= 16*s->mb_height - 4*FFMIN(h->sps.crop_bottom, 3);
1814 if (s->context_initialized
1815 && ( s->width != s->avctx->width || s->height != s->avctx->height
1816 || av_cmp_q(h->sps.sar, s->avctx->sample_aspect_ratio))) {
1818 return -1; // width / height changed during parallelized decoding
1820 flush_dpb(s->avctx);
1823 if (!s->context_initialized) {
1825 return -1; // we cant (re-)initialize context during parallel decoding
1827 avcodec_set_dimensions(s->avctx, s->width, s->height);
1828 s->avctx->sample_aspect_ratio= h->sps.sar;
1829 if(!s->avctx->sample_aspect_ratio.den)
1830 s->avctx->sample_aspect_ratio.den = 1;
1832 if(h->sps.video_signal_type_present_flag){
1833 s->avctx->color_range = h->sps.full_range ? AVCOL_RANGE_JPEG : AVCOL_RANGE_MPEG;
1834 if(h->sps.colour_description_present_flag){
1835 s->avctx->color_primaries = h->sps.color_primaries;
1836 s->avctx->color_trc = h->sps.color_trc;
1837 s->avctx->colorspace = h->sps.colorspace;
1841 if(h->sps.timing_info_present_flag){
1842 int64_t den= h->sps.time_scale;
1843 if(h->x264_build < 44U)
1845 av_reduce(&s->avctx->time_base.num, &s->avctx->time_base.den,
1846 h->sps.num_units_in_tick, den, 1<<30);
1848 s->avctx->pix_fmt = s->avctx->get_format(s->avctx,
1849 s->avctx->codec->pix_fmts ?
1850 s->avctx->codec->pix_fmts :
1851 s->avctx->color_range == AVCOL_RANGE_JPEG ?
1852 hwaccel_pixfmt_list_h264_jpeg_420 :
1853 ff_hwaccel_pixfmt_list_420);
1854 s->avctx->hwaccel = ff_find_hwaccel(s->avctx->codec->id, s->avctx->pix_fmt);
1856 if (MPV_common_init(s) < 0)
1859 h->prev_interlaced_frame = 1;
1861 init_scan_tables(h);
1862 ff_h264_alloc_tables(h);
1864 for(i = 1; i < s->avctx->thread_count; i++) {
1866 c = h->thread_context[i] = av_malloc(sizeof(H264Context));
1867 memcpy(c, h->s.thread_context[i], sizeof(MpegEncContext));
1868 memset(&c->s + 1, 0, sizeof(H264Context) - sizeof(MpegEncContext));
1869 c->h264dsp = h->h264dsp;
1872 init_scan_tables(c);
1873 clone_tables(c, h, i);
1876 for(i = 0; i < s->avctx->thread_count; i++)
1877 if(context_init(h->thread_context[i]) < 0)
1881 h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
1884 h->mb_aff_frame = 0;
1885 last_pic_structure = s0->picture_structure;
1886 if(h->sps.frame_mbs_only_flag){
1887 s->picture_structure= PICT_FRAME;
1889 if(get_bits1(&s->gb)) { //field_pic_flag
1890 s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
1892 s->picture_structure= PICT_FRAME;
1893 h->mb_aff_frame = h->sps.mb_aff;
1896 h->mb_field_decoding_flag= s->picture_structure != PICT_FRAME;
1898 if(h0->current_slice == 0){
1899 while(h->frame_num != h->prev_frame_num &&
1900 h->frame_num != (h->prev_frame_num+1)%(1<<h->sps.log2_max_frame_num)){
1901 Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
1902 av_log(h->s.avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n", h->frame_num, h->prev_frame_num);
1903 if (ff_h264_frame_start(h) < 0)
1905 h->prev_frame_num++;
1906 h->prev_frame_num %= 1<<h->sps.log2_max_frame_num;
1907 s->current_picture_ptr->frame_num= h->prev_frame_num;
1908 ff_generate_sliding_window_mmcos(h);
1909 ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1910 /* Error concealment: if a ref is missing, copy the previous ref in its place.
1911 * FIXME: avoiding a memcpy would be nice, but ref handling makes many assumptions
1912 * about there being no actual duplicates.
1913 * FIXME: this doesn't copy padding for out-of-frame motion vectors. Given we're
1914 * concealing a lost frame, this probably isn't noticable by comparison, but it should
1916 if (h->short_ref_count) {
1918 av_image_copy(h->short_ref[0]->data, h->short_ref[0]->linesize,
1919 (const uint8_t**)prev->data, prev->linesize,
1920 PIX_FMT_YUV420P, s->mb_width*16, s->mb_height*16);
1921 h->short_ref[0]->poc = prev->poc+2;
1923 h->short_ref[0]->frame_num = h->prev_frame_num;
1927 /* See if we have a decoded first field looking for a pair... */
1928 if (s0->first_field) {
1929 assert(s0->current_picture_ptr);
1930 assert(s0->current_picture_ptr->data[0]);
1931 assert(s0->current_picture_ptr->reference != DELAYED_PIC_REF);
1933 /* figure out if we have a complementary field pair */
1934 if (!FIELD_PICTURE || s->picture_structure == last_pic_structure) {
1936 * Previous field is unmatched. Don't display it, but let it
1937 * remain for reference if marked as such.
1939 s0->current_picture_ptr = NULL;
1940 s0->first_field = FIELD_PICTURE;
1943 if (h->nal_ref_idc &&
1944 s0->current_picture_ptr->reference &&
1945 s0->current_picture_ptr->frame_num != h->frame_num) {
1947 * This and previous field were reference, but had
1948 * different frame_nums. Consider this field first in
1949 * pair. Throw away previous field except for reference
1952 s0->first_field = 1;
1953 s0->current_picture_ptr = NULL;
1956 /* Second field in complementary pair */
1957 s0->first_field = 0;
1962 /* Frame or first field in a potentially complementary pair */
1963 assert(!s0->current_picture_ptr);
1964 s0->first_field = FIELD_PICTURE;
1967 if((!FIELD_PICTURE || s0->first_field) && ff_h264_frame_start(h) < 0) {
1968 s0->first_field = 0;
1975 s->current_picture_ptr->frame_num= h->frame_num; //FIXME frame_num cleanup
1977 assert(s->mb_num == s->mb_width * s->mb_height);
1978 if(first_mb_in_slice << FIELD_OR_MBAFF_PICTURE >= s->mb_num ||
1979 first_mb_in_slice >= s->mb_num){
1980 av_log(h->s.avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
1983 s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
1984 s->resync_mb_y = s->mb_y = (first_mb_in_slice / s->mb_width) << FIELD_OR_MBAFF_PICTURE;
1985 if (s->picture_structure == PICT_BOTTOM_FIELD)
1986 s->resync_mb_y = s->mb_y = s->mb_y + 1;
1987 assert(s->mb_y < s->mb_height);
1989 if(s->picture_structure==PICT_FRAME){
1990 h->curr_pic_num= h->frame_num;
1991 h->max_pic_num= 1<< h->sps.log2_max_frame_num;
1993 h->curr_pic_num= 2*h->frame_num + 1;
1994 h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
1997 if(h->nal_unit_type == NAL_IDR_SLICE){
1998 get_ue_golomb(&s->gb); /* idr_pic_id */
2001 if(h->sps.poc_type==0){
2002 h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
2004 if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
2005 h->delta_poc_bottom= get_se_golomb(&s->gb);
2009 if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
2010 h->delta_poc[0]= get_se_golomb(&s->gb);
2012 if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
2013 h->delta_poc[1]= get_se_golomb(&s->gb);
2018 if(h->pps.redundant_pic_cnt_present){
2019 h->redundant_pic_count= get_ue_golomb(&s->gb);
2022 //set defaults, might be overridden a few lines later
2023 h->ref_count[0]= h->pps.ref_count[0];
2024 h->ref_count[1]= h->pps.ref_count[1];
2026 if(h->slice_type_nos != FF_I_TYPE){
2027 if(h->slice_type_nos == FF_B_TYPE){
2028 h->direct_spatial_mv_pred= get_bits1(&s->gb);
2030 num_ref_idx_active_override_flag= get_bits1(&s->gb);
2032 if(num_ref_idx_active_override_flag){
2033 h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
2034 if(h->slice_type_nos==FF_B_TYPE)
2035 h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
2037 if(h->ref_count[0]-1 > 32-1 || h->ref_count[1]-1 > 32-1){
2038 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
2039 h->ref_count[0]= h->ref_count[1]= 1;
2043 if(h->slice_type_nos == FF_B_TYPE)
2050 if(!default_ref_list_done){
2051 ff_h264_fill_default_ref_list(h);
2054 if(h->slice_type_nos!=FF_I_TYPE && ff_h264_decode_ref_pic_list_reordering(h) < 0)
2057 if(h->slice_type_nos!=FF_I_TYPE){
2058 s->last_picture_ptr= &h->ref_list[0][0];
2059 ff_copy_picture(&s->last_picture, s->last_picture_ptr);
2061 if(h->slice_type_nos==FF_B_TYPE){
2062 s->next_picture_ptr= &h->ref_list[1][0];
2063 ff_copy_picture(&s->next_picture, s->next_picture_ptr);
2066 if( (h->pps.weighted_pred && h->slice_type_nos == FF_P_TYPE )
2067 || (h->pps.weighted_bipred_idc==1 && h->slice_type_nos== FF_B_TYPE ) )
2068 pred_weight_table(h);
2069 else if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== FF_B_TYPE){
2070 implicit_weight_table(h, -1);
2073 for (i = 0; i < 2; i++) {
2074 h->luma_weight_flag[i] = 0;
2075 h->chroma_weight_flag[i] = 0;
2080 ff_h264_decode_ref_pic_marking(h0, &s->gb);
2083 ff_h264_fill_mbaff_ref_list(h);
2085 if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== FF_B_TYPE){
2086 implicit_weight_table(h, 0);
2087 implicit_weight_table(h, 1);
2091 if(h->slice_type_nos==FF_B_TYPE && !h->direct_spatial_mv_pred)
2092 ff_h264_direct_dist_scale_factor(h);
2093 ff_h264_direct_ref_list_init(h);
2095 if( h->slice_type_nos != FF_I_TYPE && h->pps.cabac ){
2096 tmp = get_ue_golomb_31(&s->gb);
2098 av_log(s->avctx, AV_LOG_ERROR, "cabac_init_idc overflow\n");
2101 h->cabac_init_idc= tmp;
2104 h->last_qscale_diff = 0;
2105 tmp = h->pps.init_qp + get_se_golomb(&s->gb);
2107 av_log(s->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
2111 h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
2112 h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
2113 //FIXME qscale / qp ... stuff
2114 if(h->slice_type == FF_SP_TYPE){
2115 get_bits1(&s->gb); /* sp_for_switch_flag */
2117 if(h->slice_type==FF_SP_TYPE || h->slice_type == FF_SI_TYPE){
2118 get_se_golomb(&s->gb); /* slice_qs_delta */
2121 h->deblocking_filter = 1;
2122 h->slice_alpha_c0_offset = 52;
2123 h->slice_beta_offset = 52;
2124 if( h->pps.deblocking_filter_parameters_present ) {
2125 tmp= get_ue_golomb_31(&s->gb);
2127 av_log(s->avctx, AV_LOG_ERROR, "deblocking_filter_idc %u out of range\n", tmp);
2130 h->deblocking_filter= tmp;
2131 if(h->deblocking_filter < 2)
2132 h->deblocking_filter^= 1; // 1<->0
2134 if( h->deblocking_filter ) {
2135 h->slice_alpha_c0_offset += get_se_golomb(&s->gb) << 1;
2136 h->slice_beta_offset += get_se_golomb(&s->gb) << 1;
2137 if( h->slice_alpha_c0_offset > 104U
2138 || h->slice_beta_offset > 104U){
2139 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);
2145 if( s->avctx->skip_loop_filter >= AVDISCARD_ALL
2146 ||(s->avctx->skip_loop_filter >= AVDISCARD_NONKEY && h->slice_type_nos != FF_I_TYPE)
2147 ||(s->avctx->skip_loop_filter >= AVDISCARD_BIDIR && h->slice_type_nos == FF_B_TYPE)
2148 ||(s->avctx->skip_loop_filter >= AVDISCARD_NONREF && h->nal_ref_idc == 0))
2149 h->deblocking_filter= 0;
2151 if(h->deblocking_filter == 1 && h0->max_contexts > 1) {
2152 if(s->avctx->flags2 & CODEC_FLAG2_FAST) {
2153 /* Cheat slightly for speed:
2154 Do not bother to deblock across slices. */
2155 h->deblocking_filter = 2;
2157 h0->max_contexts = 1;
2158 if(!h0->single_decode_warning) {
2159 av_log(s->avctx, AV_LOG_INFO, "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
2160 h0->single_decode_warning = 1;
2163 return 1; // deblocking switched inside frame
2166 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]);
2169 if( h->pps.num_slice_groups > 1 && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
2170 slice_group_change_cycle= get_bits(&s->gb, ?);
2173 h0->last_slice_type = slice_type;
2174 h->slice_num = ++h0->current_slice;
2175 if(h->slice_num >= MAX_SLICES){
2176 av_log(s->avctx, AV_LOG_ERROR, "Too many slices, increase MAX_SLICES and recompile\n");
2181 int *ref2frm= h->ref2frm[h->slice_num&(MAX_SLICES-1)][j];
2182 for(i=0; i<16; i++){
2184 if(h->ref_list[j][i].data[0]){
2186 uint8_t *base= h->ref_list[j][i].base[0];
2187 for(k=0; k<h->short_ref_count; k++)
2188 if(h->short_ref[k]->base[0] == base){
2192 for(k=0; k<h->long_ref_count; k++)
2193 if(h->long_ref[k] && h->long_ref[k]->base[0] == base){
2194 id_list[i]= h->short_ref_count + k;
2203 ref2frm[i+2]= 4*id_list[i]
2204 +(h->ref_list[j][i].reference&3);
2207 for(i=16; i<48; i++)
2208 ref2frm[i+4]= 4*id_list[(i-16)>>1]
2209 +(h->ref_list[j][i].reference&3);
2212 h->emu_edge_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16;
2213 h->emu_edge_height= (FRAME_MBAFF || FIELD_PICTURE) ? 0 : h->emu_edge_width;
2215 if(s->avctx->debug&FF_DEBUG_PICT_INFO){
2216 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",
2218 (s->picture_structure==PICT_FRAME ? "F" : s->picture_structure==PICT_TOP_FIELD ? "T" : "B"),
2220 av_get_pict_type_char(h->slice_type), h->slice_type_fixed ? " fix" : "", h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
2221 pps_id, h->frame_num,
2222 s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
2223 h->ref_count[0], h->ref_count[1],
2225 h->deblocking_filter, h->slice_alpha_c0_offset/2-26, h->slice_beta_offset/2-26,
2227 h->use_weight==1 && h->use_weight_chroma ? "c" : "",
2228 h->slice_type == FF_B_TYPE ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : ""
2235 int ff_h264_get_slice_type(const H264Context *h)
2237 switch (h->slice_type) {
2238 case FF_P_TYPE: return 0;
2239 case FF_B_TYPE: return 1;
2240 case FF_I_TYPE: return 2;
2241 case FF_SP_TYPE: return 3;
2242 case FF_SI_TYPE: return 4;
2249 * @return non zero if the loop filter can be skiped
2251 static int fill_filter_caches(H264Context *h, int mb_type){
2252 MpegEncContext * const s = &h->s;
2253 const int mb_xy= h->mb_xy;
2254 int top_xy, left_xy[2];
2255 int top_type, left_type[2];
2257 top_xy = mb_xy - (s->mb_stride << MB_FIELD);
2259 //FIXME deblocking could skip the intra and nnz parts.
2261 /* Wow, what a mess, why didn't they simplify the interlacing & intra
2262 * stuff, I can't imagine that these complex rules are worth it. */
2264 left_xy[1] = left_xy[0] = mb_xy-1;
2266 const int left_mb_field_flag = IS_INTERLACED(s->current_picture.mb_type[mb_xy-1]);
2267 const int curr_mb_field_flag = IS_INTERLACED(mb_type);
2269 if (left_mb_field_flag != curr_mb_field_flag) {
2270 left_xy[0] -= s->mb_stride;
2273 if(curr_mb_field_flag){
2274 top_xy += s->mb_stride & (((s->current_picture.mb_type[top_xy ]>>7)&1)-1);
2276 if (left_mb_field_flag != curr_mb_field_flag) {
2277 left_xy[1] += s->mb_stride;
2282 h->top_mb_xy = top_xy;
2283 h->left_mb_xy[0] = left_xy[0];
2284 h->left_mb_xy[1] = left_xy[1];
2286 //for sufficiently low qp, filtering wouldn't do anything
2287 //this is a conservative estimate: could also check beta_offset and more accurate chroma_qp
2288 int qp_thresh = h->qp_thresh; //FIXME strictly we should store qp_thresh for each mb of a slice
2289 int qp = s->current_picture.qscale_table[mb_xy];
2291 && (left_xy[0]<0 || ((qp + s->current_picture.qscale_table[left_xy[0]] + 1)>>1) <= qp_thresh)
2292 && (top_xy < 0 || ((qp + s->current_picture.qscale_table[top_xy ] + 1)>>1) <= qp_thresh)){
2295 if( (left_xy[0]< 0 || ((qp + s->current_picture.qscale_table[left_xy[1] ] + 1)>>1) <= qp_thresh)
2296 && (top_xy < s->mb_stride || ((qp + s->current_picture.qscale_table[top_xy -s->mb_stride] + 1)>>1) <= qp_thresh))
2301 top_type = s->current_picture.mb_type[top_xy] ;
2302 left_type[0] = s->current_picture.mb_type[left_xy[0]];
2303 left_type[1] = s->current_picture.mb_type[left_xy[1]];
2304 if(h->deblocking_filter == 2){
2305 if(h->slice_table[top_xy ] != h->slice_num) top_type= 0;
2306 if(h->slice_table[left_xy[0] ] != h->slice_num) left_type[0]= left_type[1]= 0;
2308 if(h->slice_table[top_xy ] == 0xFFFF) top_type= 0;
2309 if(h->slice_table[left_xy[0] ] == 0xFFFF) left_type[0]= left_type[1] =0;
2311 h->top_type = top_type ;
2312 h->left_type[0]= left_type[0];
2313 h->left_type[1]= left_type[1];
2315 if(IS_INTRA(mb_type))
2318 AV_COPY64(&h->non_zero_count_cache[0+8*1], &h->non_zero_count[mb_xy][ 0]);
2319 AV_COPY64(&h->non_zero_count_cache[0+8*2], &h->non_zero_count[mb_xy][ 8]);
2320 AV_COPY32(&h->non_zero_count_cache[0+8*5], &h->non_zero_count[mb_xy][16]);
2321 AV_COPY32(&h->non_zero_count_cache[4+8*3], &h->non_zero_count[mb_xy][20]);
2322 AV_COPY64(&h->non_zero_count_cache[0+8*4], &h->non_zero_count[mb_xy][24]);
2324 h->cbp= h->cbp_table[mb_xy];
2328 for(list=0; list<h->list_count; list++){
2331 int16_t (*mv_dst)[2];
2332 int16_t (*mv_src)[2];
2334 if(!USES_LIST(mb_type, list)){
2335 fill_rectangle( h->mv_cache[list][scan8[0]], 4, 4, 8, pack16to32(0,0), 4);
2336 AV_WN32A(&h->ref_cache[list][scan8[ 0]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
2337 AV_WN32A(&h->ref_cache[list][scan8[ 2]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
2338 AV_WN32A(&h->ref_cache[list][scan8[ 8]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
2339 AV_WN32A(&h->ref_cache[list][scan8[10]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
2343 ref = &s->current_picture.ref_index[list][4*mb_xy];
2345 int (*ref2frm)[64] = h->ref2frm[ h->slice_num&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
2346 AV_WN32A(&h->ref_cache[list][scan8[ 0]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
2347 AV_WN32A(&h->ref_cache[list][scan8[ 2]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
2349 AV_WN32A(&h->ref_cache[list][scan8[ 8]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
2350 AV_WN32A(&h->ref_cache[list][scan8[10]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
2353 b_stride = h->b_stride;
2354 mv_dst = &h->mv_cache[list][scan8[0]];
2355 mv_src = &s->current_picture.motion_val[list][4*s->mb_x + 4*s->mb_y*b_stride];
2357 AV_COPY128(mv_dst + 8*y, mv_src + y*b_stride);
2372 //FIXME constraint_intra_pred & partitioning & nnz (let us hope this is just a typo in the spec)
2374 AV_COPY32(&h->non_zero_count_cache[4+8*0], &h->non_zero_count[top_xy][4+3*8]);
2378 h->non_zero_count_cache[3+8*1]= h->non_zero_count[left_xy[0]][7+0*8];
2379 h->non_zero_count_cache[3+8*2]= h->non_zero_count[left_xy[0]][7+1*8];
2380 h->non_zero_count_cache[3+8*3]= h->non_zero_count[left_xy[0]][7+2*8];
2381 h->non_zero_count_cache[3+8*4]= h->non_zero_count[left_xy[0]][7+3*8];
2384 // CAVLC 8x8dct requires NNZ values for residual decoding that differ from what the loop filter needs
2385 if(!CABAC && h->pps.transform_8x8_mode){
2386 if(IS_8x8DCT(top_type)){
2387 h->non_zero_count_cache[4+8*0]=
2388 h->non_zero_count_cache[5+8*0]= h->cbp_table[top_xy] & 4;
2389 h->non_zero_count_cache[6+8*0]=
2390 h->non_zero_count_cache[7+8*0]= h->cbp_table[top_xy] & 8;
2392 if(IS_8x8DCT(left_type[0])){
2393 h->non_zero_count_cache[3+8*1]=
2394 h->non_zero_count_cache[3+8*2]= h->cbp_table[left_xy[0]]&2; //FIXME check MBAFF
2396 if(IS_8x8DCT(left_type[1])){
2397 h->non_zero_count_cache[3+8*3]=
2398 h->non_zero_count_cache[3+8*4]= h->cbp_table[left_xy[1]]&8; //FIXME check MBAFF
2401 if(IS_8x8DCT(mb_type)){
2402 h->non_zero_count_cache[scan8[0 ]]= h->non_zero_count_cache[scan8[1 ]]=
2403 h->non_zero_count_cache[scan8[2 ]]= h->non_zero_count_cache[scan8[3 ]]= h->cbp & 1;
2405 h->non_zero_count_cache[scan8[0+ 4]]= h->non_zero_count_cache[scan8[1+ 4]]=
2406 h->non_zero_count_cache[scan8[2+ 4]]= h->non_zero_count_cache[scan8[3+ 4]]= h->cbp & 2;
2408 h->non_zero_count_cache[scan8[0+ 8]]= h->non_zero_count_cache[scan8[1+ 8]]=
2409 h->non_zero_count_cache[scan8[2+ 8]]= h->non_zero_count_cache[scan8[3+ 8]]= h->cbp & 4;
2411 h->non_zero_count_cache[scan8[0+12]]= h->non_zero_count_cache[scan8[1+12]]=
2412 h->non_zero_count_cache[scan8[2+12]]= h->non_zero_count_cache[scan8[3+12]]= h->cbp & 8;
2416 if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
2418 for(list=0; list<h->list_count; list++){
2419 if(USES_LIST(top_type, list)){
2420 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
2421 const int b8_xy= 4*top_xy + 2;
2422 int (*ref2frm)[64] = h->ref2frm[ h->slice_table[top_xy]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
2423 AV_COPY128(h->mv_cache[list][scan8[0] + 0 - 1*8], s->current_picture.motion_val[list][b_xy + 0]);
2424 h->ref_cache[list][scan8[0] + 0 - 1*8]=
2425 h->ref_cache[list][scan8[0] + 1 - 1*8]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 0]];
2426 h->ref_cache[list][scan8[0] + 2 - 1*8]=
2427 h->ref_cache[list][scan8[0] + 3 - 1*8]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 1]];
2429 AV_ZERO128(h->mv_cache[list][scan8[0] + 0 - 1*8]);
2430 AV_WN32A(&h->ref_cache[list][scan8[0] + 0 - 1*8], ((LIST_NOT_USED)&0xFF)*0x01010101u);
2433 if(!IS_INTERLACED(mb_type^left_type[0])){
2434 if(USES_LIST(left_type[0], list)){
2435 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
2436 const int b8_xy= 4*left_xy[0] + 1;
2437 int (*ref2frm)[64] = h->ref2frm[ h->slice_table[left_xy[0]]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
2438 AV_COPY32(h->mv_cache[list][scan8[0] - 1 + 0 ], s->current_picture.motion_val[list][b_xy + h->b_stride*0]);
2439 AV_COPY32(h->mv_cache[list][scan8[0] - 1 + 8 ], s->current_picture.motion_val[list][b_xy + h->b_stride*1]);
2440 AV_COPY32(h->mv_cache[list][scan8[0] - 1 +16 ], s->current_picture.motion_val[list][b_xy + h->b_stride*2]);
2441 AV_COPY32(h->mv_cache[list][scan8[0] - 1 +24 ], s->current_picture.motion_val[list][b_xy + h->b_stride*3]);
2442 h->ref_cache[list][scan8[0] - 1 + 0 ]=
2443 h->ref_cache[list][scan8[0] - 1 + 8 ]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 2*0]];
2444 h->ref_cache[list][scan8[0] - 1 +16 ]=
2445 h->ref_cache[list][scan8[0] - 1 +24 ]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 2*1]];
2447 AV_ZERO32(h->mv_cache [list][scan8[0] - 1 + 0 ]);
2448 AV_ZERO32(h->mv_cache [list][scan8[0] - 1 + 8 ]);
2449 AV_ZERO32(h->mv_cache [list][scan8[0] - 1 +16 ]);
2450 AV_ZERO32(h->mv_cache [list][scan8[0] - 1 +24 ]);
2451 h->ref_cache[list][scan8[0] - 1 + 0 ]=
2452 h->ref_cache[list][scan8[0] - 1 + 8 ]=
2453 h->ref_cache[list][scan8[0] - 1 + 16 ]=
2454 h->ref_cache[list][scan8[0] - 1 + 24 ]= LIST_NOT_USED;
2463 static void loop_filter(H264Context *h){
2464 MpegEncContext * const s = &h->s;
2465 uint8_t *dest_y, *dest_cb, *dest_cr;
2466 int linesize, uvlinesize, mb_x, mb_y;
2467 const int end_mb_y= s->mb_y + FRAME_MBAFF;
2468 const int old_slice_type= h->slice_type;
2470 if(h->deblocking_filter) {
2471 for(mb_x= 0; mb_x<s->mb_width; mb_x++){
2472 for(mb_y=end_mb_y - FRAME_MBAFF; mb_y<= end_mb_y; mb_y++){
2474 mb_xy = h->mb_xy = mb_x + mb_y*s->mb_stride;
2475 h->slice_num= h->slice_table[mb_xy];
2476 mb_type= s->current_picture.mb_type[mb_xy];
2477 h->list_count= h->list_counts[mb_xy];
2480 h->mb_mbaff = h->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
2484 dest_y = s->current_picture.data[0] + (mb_x + mb_y * s->linesize ) * 16;
2485 dest_cb = s->current_picture.data[1] + (mb_x + mb_y * s->uvlinesize) * 8;
2486 dest_cr = s->current_picture.data[2] + (mb_x + mb_y * s->uvlinesize) * 8;
2487 //FIXME simplify above
2490 linesize = h->mb_linesize = s->linesize * 2;
2491 uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
2492 if(mb_y&1){ //FIXME move out of this function?
2493 dest_y -= s->linesize*15;
2494 dest_cb-= s->uvlinesize*7;
2495 dest_cr-= s->uvlinesize*7;
2498 linesize = h->mb_linesize = s->linesize;
2499 uvlinesize = h->mb_uvlinesize = s->uvlinesize;
2501 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);
2502 if(fill_filter_caches(h, mb_type))
2504 h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.qscale_table[mb_xy]);
2505 h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.qscale_table[mb_xy]);
2508 ff_h264_filter_mb (h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2510 ff_h264_filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2515 h->slice_type= old_slice_type;
2517 s->mb_y= end_mb_y - FRAME_MBAFF;
2518 h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
2519 h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
2522 static void predict_field_decoding_flag(H264Context *h){
2523 MpegEncContext * const s = &h->s;
2524 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
2525 int mb_type = (h->slice_table[mb_xy-1] == h->slice_num)
2526 ? s->current_picture.mb_type[mb_xy-1]
2527 : (h->slice_table[mb_xy-s->mb_stride] == h->slice_num)
2528 ? s->current_picture.mb_type[mb_xy-s->mb_stride]
2530 h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
2533 static int decode_slice(struct AVCodecContext *avctx, void *arg){
2534 H264Context *h = *(void**)arg;
2535 MpegEncContext * const s = &h->s;
2536 const int part_mask= s->partitioned_frame ? (AC_END|AC_ERROR) : 0x7F;
2540 h->is_complex = FRAME_MBAFF || s->picture_structure != PICT_FRAME || s->codec_id != CODEC_ID_H264 ||
2541 (CONFIG_GRAY && (s->flags&CODEC_FLAG_GRAY));
2543 if( h->pps.cabac ) {
2545 align_get_bits( &s->gb );
2548 ff_init_cabac_states( &h->cabac);
2549 ff_init_cabac_decoder( &h->cabac,
2550 s->gb.buffer + get_bits_count(&s->gb)/8,
2551 (get_bits_left(&s->gb) + 7)/8);
2553 ff_h264_init_cabac_states(h);
2557 int ret = ff_h264_decode_mb_cabac(h);
2559 //STOP_TIMER("decode_mb_cabac")
2561 if(ret>=0) ff_h264_hl_decode_mb(h);
2563 if( ret >= 0 && FRAME_MBAFF ) { //FIXME optimal? or let mb_decode decode 16x32 ?
2566 ret = ff_h264_decode_mb_cabac(h);
2568 if(ret>=0) ff_h264_hl_decode_mb(h);
2571 eos = get_cabac_terminate( &h->cabac );
2573 if((s->workaround_bugs & FF_BUG_TRUNCATED) && h->cabac.bytestream > h->cabac.bytestream_end + 2){
2574 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);
2577 if( ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 2) {
2578 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);
2579 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);
2583 if( ++s->mb_x >= s->mb_width ) {
2586 ff_draw_horiz_band(s, 16*s->mb_y, 16);
2588 if(FIELD_OR_MBAFF_PICTURE) {
2590 if(FRAME_MBAFF && s->mb_y < s->mb_height)
2591 predict_field_decoding_flag(h);
2595 if( eos || s->mb_y >= s->mb_height ) {
2596 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2597 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);
2604 int ret = ff_h264_decode_mb_cavlc(h);
2606 if(ret>=0) ff_h264_hl_decode_mb(h);
2608 if(ret>=0 && FRAME_MBAFF){ //FIXME optimal? or let mb_decode decode 16x32 ?
2610 ret = ff_h264_decode_mb_cavlc(h);
2612 if(ret>=0) ff_h264_hl_decode_mb(h);
2617 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
2618 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);
2623 if(++s->mb_x >= s->mb_width){
2626 ff_draw_horiz_band(s, 16*s->mb_y, 16);
2628 if(FIELD_OR_MBAFF_PICTURE) {
2630 if(FRAME_MBAFF && s->mb_y < s->mb_height)
2631 predict_field_decoding_flag(h);
2633 if(s->mb_y >= s->mb_height){
2634 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2636 if(get_bits_count(&s->gb) == s->gb.size_in_bits ) {
2637 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);
2641 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);
2648 if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->mb_skip_run<=0){
2649 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2650 if(get_bits_count(&s->gb) == s->gb.size_in_bits ){
2651 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);
2655 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);
2664 for(;s->mb_y < s->mb_height; s->mb_y++){
2665 for(;s->mb_x < s->mb_width; s->mb_x++){
2666 int ret= decode_mb(h);
2668 ff_h264_hl_decode_mb(h);
2671 av_log(s->avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
2672 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);
2677 if(++s->mb_x >= s->mb_width){
2679 if(++s->mb_y >= s->mb_height){
2680 if(get_bits_count(s->gb) == s->gb.size_in_bits){
2681 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);
2685 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);
2692 if(get_bits_count(s->?gb) >= s->gb?.size_in_bits){
2693 if(get_bits_count(s->gb) == s->gb.size_in_bits){
2694 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);
2698 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);
2705 ff_draw_horiz_band(s, 16*s->mb_y, 16);
2708 return -1; //not reached
2712 * Call decode_slice() for each context.
2714 * @param h h264 master context
2715 * @param context_count number of contexts to execute
2717 static void execute_decode_slices(H264Context *h, int context_count){
2718 MpegEncContext * const s = &h->s;
2719 AVCodecContext * const avctx= s->avctx;
2723 if (s->avctx->hwaccel)
2725 if(s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
2727 if(context_count == 1) {
2728 decode_slice(avctx, &h);
2730 for(i = 1; i < context_count; i++) {
2731 hx = h->thread_context[i];
2732 hx->s.error_recognition = avctx->error_recognition;
2733 hx->s.error_count = 0;
2736 avctx->execute(avctx, (void *)decode_slice,
2737 h->thread_context, NULL, context_count, sizeof(void*));
2739 /* pull back stuff from slices to master context */
2740 hx = h->thread_context[context_count - 1];
2741 s->mb_x = hx->s.mb_x;
2742 s->mb_y = hx->s.mb_y;
2743 s->dropable = hx->s.dropable;
2744 s->picture_structure = hx->s.picture_structure;
2745 for(i = 1; i < context_count; i++)
2746 h->s.error_count += h->thread_context[i]->s.error_count;
2751 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size){
2752 MpegEncContext * const s = &h->s;
2753 AVCodecContext * const avctx= s->avctx;
2755 H264Context *hx; ///< thread context
2756 int context_count = 0;
2757 int next_avc= h->is_avc ? 0 : buf_size;
2759 h->max_contexts = avctx->thread_count;
2762 for(i=0; i<50; i++){
2763 av_log(NULL, AV_LOG_ERROR,"%02X ", buf[i]);
2766 if(!(s->flags2 & CODEC_FLAG2_CHUNKS)){
2767 h->current_slice = 0;
2768 if (!s->first_field)
2769 s->current_picture_ptr= NULL;
2770 ff_h264_reset_sei(h);
2781 if(buf_index >= next_avc) {
2782 if(buf_index >= buf_size) break;
2784 for(i = 0; i < h->nal_length_size; i++)
2785 nalsize = (nalsize << 8) | buf[buf_index++];
2786 if(nalsize <= 1 || nalsize > buf_size - buf_index){
2791 av_log(h->s.avctx, AV_LOG_ERROR, "AVC: nal size %d\n", nalsize);
2795 next_avc= buf_index + nalsize;
2797 // start code prefix search
2798 for(; buf_index + 3 < next_avc; buf_index++){
2799 // This should always succeed in the first iteration.
2800 if(buf[buf_index] == 0 && buf[buf_index+1] == 0 && buf[buf_index+2] == 1)
2804 if(buf_index+3 >= buf_size) break;
2807 if(buf_index >= next_avc) continue;
2810 hx = h->thread_context[context_count];
2812 ptr= ff_h264_decode_nal(hx, buf + buf_index, &dst_length, &consumed, next_avc - buf_index);
2813 if (ptr==NULL || dst_length < 0){
2816 i= buf_index + consumed;
2817 if((s->workaround_bugs & FF_BUG_AUTODETECT) && i+3<next_avc &&
2818 buf[i]==0x00 && buf[i+1]==0x00 && buf[i+2]==0x01 && buf[i+3]==0xE0)
2819 s->workaround_bugs |= FF_BUG_TRUNCATED;
2821 if(!(s->workaround_bugs & FF_BUG_TRUNCATED)){
2822 while(ptr[dst_length - 1] == 0 && dst_length > 0)
2825 bit_length= !dst_length ? 0 : (8*dst_length - ff_h264_decode_rbsp_trailing(h, ptr + dst_length - 1));
2827 if(s->avctx->debug&FF_DEBUG_STARTCODE){
2828 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);
2831 if (h->is_avc && (nalsize != consumed) && nalsize){
2832 av_log(h->s.avctx, AV_LOG_DEBUG, "AVC: Consumed only %d bytes instead of %d\n", consumed, nalsize);
2835 buf_index += consumed;
2837 if( (s->hurry_up == 1 && h->nal_ref_idc == 0) //FIXME do not discard SEI id
2838 ||(avctx->skip_frame >= AVDISCARD_NONREF && h->nal_ref_idc == 0))
2843 switch(hx->nal_unit_type){
2845 if (h->nal_unit_type != NAL_IDR_SLICE) {
2846 av_log(h->s.avctx, AV_LOG_ERROR, "Invalid mix of idr and non-idr slices");
2849 idr(h); //FIXME ensure we don't loose some frames if there is reordering
2851 init_get_bits(&hx->s.gb, ptr, bit_length);
2853 hx->inter_gb_ptr= &hx->s.gb;
2854 hx->s.data_partitioning = 0;
2856 if((err = decode_slice_header(hx, h)))
2859 if (h->current_slice == 1) {
2860 if (s->avctx->hwaccel && s->avctx->hwaccel->start_frame(s->avctx, NULL, 0) < 0)
2862 if(CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
2863 ff_vdpau_h264_picture_start(s);
2866 s->current_picture_ptr->key_frame |=
2867 (hx->nal_unit_type == NAL_IDR_SLICE) ||
2868 (h->sei_recovery_frame_cnt >= 0);
2869 if(hx->redundant_pic_count==0 && hx->s.hurry_up < 5
2870 && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
2871 && (avctx->skip_frame < AVDISCARD_BIDIR || hx->slice_type_nos!=FF_B_TYPE)
2872 && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==FF_I_TYPE)
2873 && avctx->skip_frame < AVDISCARD_ALL){
2874 if(avctx->hwaccel) {
2875 if (avctx->hwaccel->decode_slice(avctx, &buf[buf_index - consumed], consumed) < 0)
2878 if(CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU){
2879 static const uint8_t start_code[] = {0x00, 0x00, 0x01};
2880 ff_vdpau_add_data_chunk(s, start_code, sizeof(start_code));
2881 ff_vdpau_add_data_chunk(s, &buf[buf_index - consumed], consumed );
2887 init_get_bits(&hx->s.gb, ptr, bit_length);
2889 hx->inter_gb_ptr= NULL;
2891 if ((err = decode_slice_header(hx, h)) < 0)
2894 hx->s.data_partitioning = 1;
2898 init_get_bits(&hx->intra_gb, ptr, bit_length);
2899 hx->intra_gb_ptr= &hx->intra_gb;
2902 init_get_bits(&hx->inter_gb, ptr, bit_length);
2903 hx->inter_gb_ptr= &hx->inter_gb;
2905 if(hx->redundant_pic_count==0 && hx->intra_gb_ptr && hx->s.data_partitioning
2906 && s->context_initialized
2908 && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
2909 && (avctx->skip_frame < AVDISCARD_BIDIR || hx->slice_type_nos!=FF_B_TYPE)
2910 && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==FF_I_TYPE)
2911 && avctx->skip_frame < AVDISCARD_ALL)
2915 init_get_bits(&s->gb, ptr, bit_length);
2916 ff_h264_decode_sei(h);
2919 init_get_bits(&s->gb, ptr, bit_length);
2920 ff_h264_decode_seq_parameter_set(h);
2922 if(s->flags& CODEC_FLAG_LOW_DELAY)
2925 if(avctx->has_b_frames < 2)
2926 avctx->has_b_frames= !s->low_delay;
2929 init_get_bits(&s->gb, ptr, bit_length);
2931 ff_h264_decode_picture_parameter_set(h, bit_length);
2935 case NAL_END_SEQUENCE:
2936 case NAL_END_STREAM:
2937 case NAL_FILLER_DATA:
2939 case NAL_AUXILIARY_SLICE:
2942 av_log(avctx, AV_LOG_DEBUG, "Unknown NAL code: %d (%d bits)\n", hx->nal_unit_type, bit_length);
2945 if(context_count == h->max_contexts) {
2946 execute_decode_slices(h, context_count);
2951 av_log(h->s.avctx, AV_LOG_ERROR, "decode_slice_header error\n");
2953 /* Slice could not be decoded in parallel mode, copy down
2954 * NAL unit stuff to context 0 and restart. Note that
2955 * rbsp_buffer is not transferred, but since we no longer
2956 * run in parallel mode this should not be an issue. */
2957 h->nal_unit_type = hx->nal_unit_type;
2958 h->nal_ref_idc = hx->nal_ref_idc;
2964 execute_decode_slices(h, context_count);
2969 * returns the number of bytes consumed for building the current frame
2971 static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size){
2972 if(pos==0) pos=1; //avoid infinite loops (i doubt that is needed but ...)
2973 if(pos+10>buf_size) pos=buf_size; // oops ;)
2978 static int decode_frame(AVCodecContext *avctx,
2979 void *data, int *data_size,
2982 const uint8_t *buf = avpkt->data;
2983 int buf_size = avpkt->size;
2984 H264Context *h = avctx->priv_data;
2985 MpegEncContext *s = &h->s;
2986 AVFrame *pict = data;
2989 s->flags= avctx->flags;
2990 s->flags2= avctx->flags2;
2992 /* end of stream, output what is still in the buffers */
2993 if (buf_size == 0) {
2997 //FIXME factorize this with the output code below
2998 out = h->delayed_pic[0];
3000 for(i=1; h->delayed_pic[i] && !h->delayed_pic[i]->key_frame && !h->delayed_pic[i]->mmco_reset; i++)
3001 if(h->delayed_pic[i]->poc < out->poc){
3002 out = h->delayed_pic[i];
3006 for(i=out_idx; h->delayed_pic[i]; i++)
3007 h->delayed_pic[i] = h->delayed_pic[i+1];
3010 *data_size = sizeof(AVFrame);
3011 *pict= *(AVFrame*)out;
3017 buf_index=decode_nal_units(h, buf, buf_size);
3021 if(!(s->flags2 & CODEC_FLAG2_CHUNKS) && !s->current_picture_ptr){
3022 if (avctx->skip_frame >= AVDISCARD_NONREF || s->hurry_up) return 0;
3023 av_log(avctx, AV_LOG_ERROR, "no frame!\n");
3027 if(!(s->flags2 & CODEC_FLAG2_CHUNKS) || (s->mb_y >= s->mb_height && s->mb_height)){
3028 Picture *out = s->current_picture_ptr;
3029 Picture *cur = s->current_picture_ptr;
3030 int i, pics, out_of_order, out_idx;
3034 if (cur->field_poc[0]==INT_MAX || cur->field_poc[1]==INT_MAX) {
3035 /* Wait for second field. */
3039 cur->interlaced_frame = 0;
3040 cur->repeat_pict = 0;
3042 /* Signal interlacing information externally. */
3043 /* Prioritize picture timing SEI information over used decoding process if it exists. */
3045 if(h->sps.pic_struct_present_flag){
3046 switch (h->sei_pic_struct)
3048 case SEI_PIC_STRUCT_FRAME:
3050 case SEI_PIC_STRUCT_TOP_FIELD:
3051 case SEI_PIC_STRUCT_BOTTOM_FIELD:
3052 cur->interlaced_frame = 1;
3054 case SEI_PIC_STRUCT_TOP_BOTTOM:
3055 case SEI_PIC_STRUCT_BOTTOM_TOP:
3056 if (FIELD_OR_MBAFF_PICTURE)
3057 cur->interlaced_frame = 1;
3059 // try to flag soft telecine progressive
3060 cur->interlaced_frame = h->prev_interlaced_frame;
3062 case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
3063 case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
3064 // Signal the possibility of telecined film externally (pic_struct 5,6)
3065 // From these hints, let the applications decide if they apply deinterlacing.
3066 cur->repeat_pict = 1;
3068 case SEI_PIC_STRUCT_FRAME_DOUBLING:
3069 // Force progressive here, as doubling interlaced frame is a bad idea.
3070 cur->repeat_pict = 2;
3072 case SEI_PIC_STRUCT_FRAME_TRIPLING:
3073 cur->repeat_pict = 4;
3077 if ((h->sei_ct_type & 3) && h->sei_pic_struct <= SEI_PIC_STRUCT_BOTTOM_TOP)
3078 cur->interlaced_frame = (h->sei_ct_type & (1<<1)) != 0;
3080 /* Derive interlacing flag from used decoding process. */
3081 cur->interlaced_frame = FIELD_OR_MBAFF_PICTURE;
3083 h->prev_interlaced_frame = cur->interlaced_frame;
3085 if (cur->field_poc[0] != cur->field_poc[1]){
3086 /* Derive top_field_first from field pocs. */
3087 cur->top_field_first = cur->field_poc[0] < cur->field_poc[1];
3089 if(cur->interlaced_frame || h->sps.pic_struct_present_flag){
3090 /* Use picture timing SEI information. Even if it is a information of a past frame, better than nothing. */
3091 if(h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM
3092 || h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM_TOP)
3093 cur->top_field_first = 1;
3095 cur->top_field_first = 0;
3097 /* Most likely progressive */
3098 cur->top_field_first = 0;
3102 //FIXME do something with unavailable reference frames
3104 /* Sort B-frames into display order */
3106 if(h->sps.bitstream_restriction_flag
3107 && s->avctx->has_b_frames < h->sps.num_reorder_frames){
3108 s->avctx->has_b_frames = h->sps.num_reorder_frames;
3112 if( s->avctx->strict_std_compliance >= FF_COMPLIANCE_STRICT
3113 && !h->sps.bitstream_restriction_flag){
3114 s->avctx->has_b_frames= MAX_DELAYED_PIC_COUNT;
3119 while(h->delayed_pic[pics]) pics++;
3121 assert(pics <= MAX_DELAYED_PIC_COUNT);
3123 h->delayed_pic[pics++] = cur;
3124 if(cur->reference == 0)
3125 cur->reference = DELAYED_PIC_REF;
3127 out = h->delayed_pic[0];
3129 for(i=1; h->delayed_pic[i] && !h->delayed_pic[i]->key_frame && !h->delayed_pic[i]->mmco_reset; i++)
3130 if(h->delayed_pic[i]->poc < out->poc){
3131 out = h->delayed_pic[i];
3134 if(s->avctx->has_b_frames == 0 && (h->delayed_pic[0]->key_frame || h->delayed_pic[0]->mmco_reset))
3135 h->outputed_poc= INT_MIN;
3136 out_of_order = out->poc < h->outputed_poc;
3138 if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames >= h->sps.num_reorder_frames)
3140 else if((out_of_order && pics-1 == s->avctx->has_b_frames && s->avctx->has_b_frames < MAX_DELAYED_PIC_COUNT)
3142 ((h->outputed_poc != INT_MIN && out->poc > h->outputed_poc + 2)
3143 || cur->pict_type == FF_B_TYPE)))
3146 s->avctx->has_b_frames++;
3149 if(out_of_order || pics > s->avctx->has_b_frames){
3150 out->reference &= ~DELAYED_PIC_REF;
3151 for(i=out_idx; h->delayed_pic[i]; i++)
3152 h->delayed_pic[i] = h->delayed_pic[i+1];
3154 if(!out_of_order && pics > s->avctx->has_b_frames){
3155 *data_size = sizeof(AVFrame);
3157 if(out_idx==0 && h->delayed_pic[0] && (h->delayed_pic[0]->key_frame || h->delayed_pic[0]->mmco_reset)) {
3158 h->outputed_poc = INT_MIN;
3160 h->outputed_poc = out->poc;
3161 *pict= *(AVFrame*)out;
3163 av_log(avctx, AV_LOG_DEBUG, "no picture\n");
3168 assert(pict->data[0] || !*data_size);
3169 ff_print_debug_info(s, pict);
3170 //printf("out %d\n", (int)pict->data[0]);
3172 return get_consumed_bytes(s, buf_index, buf_size);
3175 static inline void fill_mb_avail(H264Context *h){
3176 MpegEncContext * const s = &h->s;
3177 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
3180 h->mb_avail[0]= s->mb_x && h->slice_table[mb_xy - s->mb_stride - 1] == h->slice_num;
3181 h->mb_avail[1]= h->slice_table[mb_xy - s->mb_stride ] == h->slice_num;
3182 h->mb_avail[2]= s->mb_x+1 < s->mb_width && h->slice_table[mb_xy - s->mb_stride + 1] == h->slice_num;
3188 h->mb_avail[3]= s->mb_x && h->slice_table[mb_xy - 1] == h->slice_num;
3189 h->mb_avail[4]= 1; //FIXME move out
3190 h->mb_avail[5]= 0; //FIXME move out
3198 #define SIZE (COUNT*40)
3204 // int int_temp[10000];
3206 AVCodecContext avctx;
3208 dsputil_init(&dsp, &avctx);
3210 init_put_bits(&pb, temp, SIZE);
3211 printf("testing unsigned exp golomb\n");
3212 for(i=0; i<COUNT; i++){
3214 set_ue_golomb(&pb, i);
3215 STOP_TIMER("set_ue_golomb");
3217 flush_put_bits(&pb);
3219 init_get_bits(&gb, temp, 8*SIZE);
3220 for(i=0; i<COUNT; i++){
3223 s= show_bits(&gb, 24);
3226 j= get_ue_golomb(&gb);
3228 printf("mismatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
3231 STOP_TIMER("get_ue_golomb");
3235 init_put_bits(&pb, temp, SIZE);
3236 printf("testing signed exp golomb\n");
3237 for(i=0; i<COUNT; i++){
3239 set_se_golomb(&pb, i - COUNT/2);
3240 STOP_TIMER("set_se_golomb");
3242 flush_put_bits(&pb);
3244 init_get_bits(&gb, temp, 8*SIZE);
3245 for(i=0; i<COUNT; i++){
3248 s= show_bits(&gb, 24);
3251 j= get_se_golomb(&gb);
3252 if(j != i - COUNT/2){
3253 printf("mismatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
3256 STOP_TIMER("get_se_golomb");
3260 printf("testing 4x4 (I)DCT\n");
3263 uint8_t src[16], ref[16];
3264 uint64_t error= 0, max_error=0;
3266 for(i=0; i<COUNT; i++){
3268 // printf("%d %d %d\n", r1, r2, (r2-r1)*16);
3269 for(j=0; j<16; j++){
3270 ref[j]= random()%255;
3271 src[j]= random()%255;
3274 h264_diff_dct_c(block, src, ref, 4);
3277 for(j=0; j<16; j++){
3278 // printf("%d ", block[j]);
3279 block[j]= block[j]*4;
3280 if(j&1) block[j]= (block[j]*4 + 2)/5;
3281 if(j&4) block[j]= (block[j]*4 + 2)/5;
3285 h->h264dsp.h264_idct_add(ref, block, 4);
3286 /* for(j=0; j<16; j++){
3287 printf("%d ", ref[j]);
3291 for(j=0; j<16; j++){
3292 int diff= FFABS(src[j] - ref[j]);
3295 max_error= FFMAX(max_error, diff);
3298 printf("error=%f max_error=%d\n", ((float)error)/COUNT/16, (int)max_error );
3299 printf("testing quantizer\n");
3300 for(qp=0; qp<52; qp++){
3302 src1_block[i]= src2_block[i]= random()%255;
3305 printf("Testing NAL layer\n");
3307 uint8_t bitstream[COUNT];
3308 uint8_t nal[COUNT*2];
3310 memset(&h, 0, sizeof(H264Context));
3312 for(i=0; i<COUNT; i++){
3320 for(j=0; j<COUNT; j++){
3321 bitstream[j]= (random() % 255) + 1;
3324 for(j=0; j<zeros; j++){
3325 int pos= random() % COUNT;
3326 while(bitstream[pos] == 0){
3335 nal_length= encode_nal(&h, nal, bitstream, COUNT, COUNT*2);
3337 printf("encoding failed\n");
3341 out= ff_h264_decode_nal(&h, nal, &out_length, &consumed, nal_length);
3345 if(out_length != COUNT){
3346 printf("incorrect length %d %d\n", out_length, COUNT);
3350 if(consumed != nal_length){
3351 printf("incorrect consumed length %d %d\n", nal_length, consumed);
3355 if(memcmp(bitstream, out, COUNT)){
3356 printf("mismatch\n");
3362 printf("Testing RBSP\n");
3370 av_cold void ff_h264_free_context(H264Context *h)
3374 free_tables(h); //FIXME cleanup init stuff perhaps
3376 for(i = 0; i < MAX_SPS_COUNT; i++)
3377 av_freep(h->sps_buffers + i);
3379 for(i = 0; i < MAX_PPS_COUNT; i++)
3380 av_freep(h->pps_buffers + i);
3383 av_cold int ff_h264_decode_end(AVCodecContext *avctx)
3385 H264Context *h = avctx->priv_data;
3386 MpegEncContext *s = &h->s;
3388 ff_h264_free_context(h);
3392 // memset(h, 0, sizeof(H264Context));
3398 AVCodec h264_decoder = {
3402 sizeof(H264Context),
3403 ff_h264_decode_init,
3407 /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 | CODEC_CAP_DELAY,
3409 .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10"),
3412 #if CONFIG_H264_VDPAU_DECODER
3413 AVCodec h264_vdpau_decoder = {
3417 sizeof(H264Context),
3418 ff_h264_decode_init,
3422 CODEC_CAP_DR1 | CODEC_CAP_DELAY | CODEC_CAP_HWACCEL_VDPAU,
3424 .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10 (VDPAU acceleration)"),
3425 .pix_fmts = (const enum PixelFormat[]){PIX_FMT_VDPAU_H264, PIX_FMT_NONE},