2 * H.26L/H.264/AVC/JVT/14496-10/... 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 "libavutil/imgutils.h"
32 #include "mpegvideo.h"
35 #include "h264_mvpred.h"
38 #include "rectangle.h"
40 #include "vdpau_internal.h"
41 #include "libavutil/avassert.h"
48 static const uint8_t rem6[QP_MAX_NUM+1]={
49 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, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3,
52 static const uint8_t div6[QP_MAX_NUM+1]={
53 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, 8, 8, 9, 9, 9, 9, 9, 9,10,10,10,10,
56 static const enum PixelFormat hwaccel_pixfmt_list_h264_jpeg_420[] = {
64 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
66 int ff_h264_check_intra4x4_pred_mode(H264Context *h){
67 MpegEncContext * const s = &h->s;
68 static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
69 static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
72 if(!(h->top_samples_available&0x8000)){
74 int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
76 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);
79 h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
84 if((h->left_samples_available&0x8888)!=0x8888){
85 static const int mask[4]={0x8000,0x2000,0x80,0x20};
87 if(!(h->left_samples_available&mask[i])){
88 int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
90 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);
93 h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
100 } //FIXME cleanup like ff_h264_check_intra_pred_mode
103 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
105 int ff_h264_check_intra_pred_mode(H264Context *h, int mode){
106 MpegEncContext * const s = &h->s;
107 static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
108 static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
111 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);
115 if(!(h->top_samples_available&0x8000)){
118 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);
123 if((h->left_samples_available&0x8080) != 0x8080){
125 if(h->left_samples_available&0x8080){ //mad cow disease mode, aka MBAFF + constrained_intra_pred
126 mode= ALZHEIMER_DC_L0T_PRED8x8 + (!(h->left_samples_available&0x8000)) + 2*(mode == DC_128_PRED8x8);
129 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);
137 const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src, int *dst_length, int *consumed, int length){
142 // src[0]&0x80; //forbidden bit
143 h->nal_ref_idc= src[0]>>5;
144 h->nal_unit_type= src[0]&0x1F;
148 #if HAVE_FAST_UNALIGNED
151 for(i=0; i+1<length; i+=9){
152 if(!((~AV_RN64A(src+i) & (AV_RN64A(src+i) - 0x0100010001000101ULL)) & 0x8000800080008080ULL))
155 for(i=0; i+1<length; i+=5){
156 if(!((~AV_RN32A(src+i) & (AV_RN32A(src+i) - 0x01000101U)) & 0x80008080U))
159 if(i>0 && !src[i]) i--;
163 for(i=0; i+1<length; i+=2){
165 if(i>0 && src[i-1]==0) i--;
167 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
169 /* startcode, so we must be past the end */
177 bufidx = h->nal_unit_type == NAL_DPC ? 1 : 0; // use second escape buffer for inter data
178 si=h->rbsp_buffer_size[bufidx];
179 av_fast_malloc(&h->rbsp_buffer[bufidx], &h->rbsp_buffer_size[bufidx], length+FF_INPUT_BUFFER_PADDING_SIZE+MAX_MBPAIR_SIZE);
180 dst= h->rbsp_buffer[bufidx];
181 if(si != h->rbsp_buffer_size[bufidx])
182 memset(dst + length, 0, FF_INPUT_BUFFER_PADDING_SIZE+MAX_MBPAIR_SIZE);
188 if(i>=length-1){ //no escaped 0
190 *consumed= length+1; //+1 for the header
191 if(h->s.avctx->flags2 & CODEC_FLAG2_FAST){
194 memcpy(dst, src, length);
199 //printf("decoding esc\n");
203 //remove escapes (very rare 1:2^22)
205 dst[di++]= src[si++];
206 dst[di++]= src[si++];
207 }else if(src[si]==0 && src[si+1]==0){
208 if(src[si+2]==3){ //escape
213 }else //next start code
217 dst[di++]= src[si++];
220 dst[di++]= src[si++];
223 memset(dst+di, 0, FF_INPUT_BUFFER_PADDING_SIZE);
226 *consumed= si + 1;//+1 for the header
227 //FIXME store exact number of bits in the getbitcontext (it is needed for decoding)
232 * Identify the exact end of the bitstream
233 * @return the length of the trailing, or 0 if damaged
235 static int ff_h264_decode_rbsp_trailing(H264Context *h, const uint8_t *src){
239 tprintf(h->s.avctx, "rbsp trailing %X\n", v);
248 static inline int get_lowest_part_list_y(H264Context *h, Picture *pic, int n, int height,
249 int y_offset, int list){
250 int raw_my= h->mv_cache[list][ scan8[n] ][1];
251 int filter_height= (raw_my&3) ? 2 : 0;
252 int full_my= (raw_my>>2) + y_offset;
253 int top = full_my - filter_height, bottom = full_my + height + filter_height;
255 return FFMAX(abs(top), bottom);
258 static inline void get_lowest_part_y(H264Context *h, int refs[2][48], int n, int height,
259 int y_offset, int list0, int list1, int *nrefs){
260 MpegEncContext * const s = &h->s;
263 y_offset += 16*(s->mb_y >> MB_FIELD);
266 int ref_n = h->ref_cache[0][ scan8[n] ];
267 Picture *ref= &h->ref_list[0][ref_n];
269 // Error resilience puts the current picture in the ref list.
270 // Don't try to wait on these as it will cause a deadlock.
271 // Fields can wait on each other, though.
272 if (ref->f.thread_opaque != s->current_picture.f.thread_opaque ||
273 (ref->f.reference & 3) != s->picture_structure) {
274 my = get_lowest_part_list_y(h, ref, n, height, y_offset, 0);
275 if (refs[0][ref_n] < 0) nrefs[0] += 1;
276 refs[0][ref_n] = FFMAX(refs[0][ref_n], my);
281 int ref_n = h->ref_cache[1][ scan8[n] ];
282 Picture *ref= &h->ref_list[1][ref_n];
284 if (ref->f.thread_opaque != s->current_picture.f.thread_opaque ||
285 (ref->f.reference & 3) != s->picture_structure) {
286 my = get_lowest_part_list_y(h, ref, n, height, y_offset, 1);
287 if (refs[1][ref_n] < 0) nrefs[1] += 1;
288 refs[1][ref_n] = FFMAX(refs[1][ref_n], my);
294 * Wait until all reference frames are available for MC operations.
296 * @param h the H264 context
298 static void await_references(H264Context *h){
299 MpegEncContext * const s = &h->s;
300 const int mb_xy= h->mb_xy;
301 const int mb_type = s->current_picture.f.mb_type[mb_xy];
306 memset(refs, -1, sizeof(refs));
308 if(IS_16X16(mb_type)){
309 get_lowest_part_y(h, refs, 0, 16, 0,
310 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs);
311 }else if(IS_16X8(mb_type)){
312 get_lowest_part_y(h, refs, 0, 8, 0,
313 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs);
314 get_lowest_part_y(h, refs, 8, 8, 8,
315 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1), nrefs);
316 }else if(IS_8X16(mb_type)){
317 get_lowest_part_y(h, refs, 0, 16, 0,
318 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs);
319 get_lowest_part_y(h, refs, 4, 16, 0,
320 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1), nrefs);
324 assert(IS_8X8(mb_type));
327 const int sub_mb_type= h->sub_mb_type[i];
329 int y_offset= (i&2)<<2;
331 if(IS_SUB_8X8(sub_mb_type)){
332 get_lowest_part_y(h, refs, n , 8, y_offset,
333 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs);
334 }else if(IS_SUB_8X4(sub_mb_type)){
335 get_lowest_part_y(h, refs, n , 4, y_offset,
336 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs);
337 get_lowest_part_y(h, refs, n+2, 4, y_offset+4,
338 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs);
339 }else if(IS_SUB_4X8(sub_mb_type)){
340 get_lowest_part_y(h, refs, n , 8, y_offset,
341 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs);
342 get_lowest_part_y(h, refs, n+1, 8, y_offset,
343 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs);
346 assert(IS_SUB_4X4(sub_mb_type));
348 int sub_y_offset= y_offset + 2*(j&2);
349 get_lowest_part_y(h, refs, n+j, 4, sub_y_offset,
350 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs);
356 for(list=h->list_count-1; list>=0; list--){
357 for(ref=0; ref<48 && nrefs[list]; ref++){
358 int row = refs[list][ref];
360 Picture *ref_pic = &h->ref_list[list][ref];
361 int ref_field = ref_pic->f.reference - 1;
362 int ref_field_picture = ref_pic->field_picture;
363 int pic_height = 16*s->mb_height >> ref_field_picture;
368 if(!FIELD_PICTURE && ref_field_picture){ // frame referencing two fields
369 ff_thread_await_progress((AVFrame*)ref_pic, FFMIN((row >> 1) - !(row&1), pic_height-1), 1);
370 ff_thread_await_progress((AVFrame*)ref_pic, FFMIN((row >> 1) , pic_height-1), 0);
371 }else if(FIELD_PICTURE && !ref_field_picture){ // field referencing one field of a frame
372 ff_thread_await_progress((AVFrame*)ref_pic, FFMIN(row*2 + ref_field , pic_height-1), 0);
373 }else if(FIELD_PICTURE){
374 ff_thread_await_progress((AVFrame*)ref_pic, FFMIN(row, pic_height-1), ref_field);
376 ff_thread_await_progress((AVFrame*)ref_pic, FFMIN(row, pic_height-1), 0);
385 * DCT transforms the 16 dc values.
386 * @param qp quantization parameter ??? FIXME
388 static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
389 // const int qmul= dequant_coeff[qp][0];
391 int temp[16]; //FIXME check if this is a good idea
392 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
393 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
396 const int offset= y_offset[i];
397 const int z0= block[offset+stride*0] + block[offset+stride*4];
398 const int z1= block[offset+stride*0] - block[offset+stride*4];
399 const int z2= block[offset+stride*1] - block[offset+stride*5];
400 const int z3= block[offset+stride*1] + block[offset+stride*5];
409 const int offset= x_offset[i];
410 const int z0= temp[4*0+i] + temp[4*2+i];
411 const int z1= temp[4*0+i] - temp[4*2+i];
412 const int z2= temp[4*1+i] - temp[4*3+i];
413 const int z3= temp[4*1+i] + temp[4*3+i];
415 block[stride*0 +offset]= (z0 + z3)>>1;
416 block[stride*2 +offset]= (z1 + z2)>>1;
417 block[stride*8 +offset]= (z1 - z2)>>1;
418 block[stride*10+offset]= (z0 - z3)>>1;
427 static void chroma_dc_dct_c(DCTELEM *block){
428 const int stride= 16*2;
429 const int xStride= 16;
432 a= block[stride*0 + xStride*0];
433 b= block[stride*0 + xStride*1];
434 c= block[stride*1 + xStride*0];
435 d= block[stride*1 + xStride*1];
442 block[stride*0 + xStride*0]= (a+c);
443 block[stride*0 + xStride*1]= (e+b);
444 block[stride*1 + xStride*0]= (a-c);
445 block[stride*1 + xStride*1]= (e-b);
449 static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
450 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
451 int src_x_offset, int src_y_offset,
452 qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op,
453 int pixel_shift, int chroma444){
454 MpegEncContext * const s = &h->s;
455 const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
456 int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
457 const int luma_xy= (mx&3) + ((my&3)<<2);
458 int offset = ((mx>>2) << pixel_shift) + (my>>2)*h->mb_linesize;
459 uint8_t * src_y = pic->f.data[0] + offset;
460 uint8_t * src_cb, * src_cr;
461 int extra_width= h->emu_edge_width;
462 int extra_height= h->emu_edge_height;
464 const int full_mx= mx>>2;
465 const int full_my= my>>2;
466 const int pic_width = 16*s->mb_width;
467 const int pic_height = 16*s->mb_height >> MB_FIELD;
469 if(mx&7) extra_width -= 3;
470 if(my&7) extra_height -= 3;
472 if( full_mx < 0-extra_width
473 || full_my < 0-extra_height
474 || full_mx + 16/*FIXME*/ > pic_width + extra_width
475 || full_my + 16/*FIXME*/ > pic_height + extra_height){
476 s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_y - (2 << pixel_shift) - 2*h->mb_linesize, h->mb_linesize, 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
477 src_y= s->edge_emu_buffer + (2 << pixel_shift) + 2*h->mb_linesize;
481 qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); //FIXME try variable height perhaps?
483 qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
486 if(CONFIG_GRAY && s->flags&CODEC_FLAG_GRAY) return;
489 src_cb = pic->f.data[1] + offset;
491 s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cb - (2 << pixel_shift) - 2*h->mb_linesize, h->mb_linesize,
492 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
493 src_cb= s->edge_emu_buffer + (2 << pixel_shift) + 2*h->mb_linesize;
495 qpix_op[luma_xy](dest_cb, src_cb, h->mb_linesize); //FIXME try variable height perhaps?
497 qpix_op[luma_xy](dest_cb + delta, src_cb + delta, h->mb_linesize);
500 src_cr = pic->f.data[2] + offset;
502 s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cr - (2 << pixel_shift) - 2*h->mb_linesize, h->mb_linesize,
503 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
504 src_cr= s->edge_emu_buffer + (2 << pixel_shift) + 2*h->mb_linesize;
506 qpix_op[luma_xy](dest_cr, src_cr, h->mb_linesize); //FIXME try variable height perhaps?
508 qpix_op[luma_xy](dest_cr + delta, src_cr + delta, h->mb_linesize);
514 // chroma offset when predicting from a field of opposite parity
515 my += 2 * ((s->mb_y & 1) - (pic->f.reference - 1));
516 emu |= (my>>3) < 0 || (my>>3) + 8 >= (pic_height>>1);
518 src_cb = pic->f.data[1] + ((mx >> 3) << pixel_shift) + (my >> 3) * h->mb_uvlinesize;
519 src_cr = pic->f.data[2] + ((mx >> 3) << pixel_shift) + (my >> 3) * h->mb_uvlinesize;
522 s->dsp.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);
523 src_cb= s->edge_emu_buffer;
525 chroma_op(dest_cb, src_cb, h->mb_uvlinesize, chroma_height, mx&7, my&7);
528 s->dsp.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);
529 src_cr= s->edge_emu_buffer;
531 chroma_op(dest_cr, src_cr, h->mb_uvlinesize, chroma_height, mx&7, my&7);
534 static inline void mc_part_std(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 int list0, int list1, int pixel_shift, int chroma444){
540 MpegEncContext * const s = &h->s;
541 qpel_mc_func *qpix_op= qpix_put;
542 h264_chroma_mc_func chroma_op= chroma_put;
544 dest_y += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
546 dest_cb += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
547 dest_cr += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
549 dest_cb += ( x_offset << pixel_shift) + y_offset*h->mb_uvlinesize;
550 dest_cr += ( x_offset << pixel_shift) + y_offset*h->mb_uvlinesize;
552 x_offset += 8*s->mb_x;
553 y_offset += 8*(s->mb_y >> MB_FIELD);
556 Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
557 mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
558 dest_y, dest_cb, dest_cr, x_offset, y_offset,
559 qpix_op, chroma_op, pixel_shift, chroma444);
562 chroma_op= chroma_avg;
566 Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
567 mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
568 dest_y, dest_cb, dest_cr, x_offset, y_offset,
569 qpix_op, chroma_op, pixel_shift, chroma444);
573 static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
574 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
575 int x_offset, int y_offset,
576 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
577 h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
578 h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
579 int list0, int list1, int pixel_shift, int chroma444){
580 MpegEncContext * const s = &h->s;
582 dest_y += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
584 chroma_weight_avg = luma_weight_avg;
585 chroma_weight_op = luma_weight_op;
586 dest_cb += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
587 dest_cr += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
589 dest_cb += ( x_offset << pixel_shift) + y_offset*h->mb_uvlinesize;
590 dest_cr += ( x_offset << pixel_shift) + y_offset*h->mb_uvlinesize;
592 x_offset += 8*s->mb_x;
593 y_offset += 8*(s->mb_y >> MB_FIELD);
596 /* don't optimize for luma-only case, since B-frames usually
597 * use implicit weights => chroma too. */
598 uint8_t *tmp_cb = s->obmc_scratchpad;
599 uint8_t *tmp_cr = s->obmc_scratchpad + (16 << pixel_shift);
600 uint8_t *tmp_y = s->obmc_scratchpad + 16*h->mb_uvlinesize;
601 int refn0 = h->ref_cache[0][ scan8[n] ];
602 int refn1 = h->ref_cache[1][ scan8[n] ];
604 mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
605 dest_y, dest_cb, dest_cr,
606 x_offset, y_offset, qpix_put, chroma_put, pixel_shift, chroma444);
607 mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
608 tmp_y, tmp_cb, tmp_cr,
609 x_offset, y_offset, qpix_put, chroma_put, pixel_shift, chroma444);
611 if(h->use_weight == 2){
612 int weight0 = h->implicit_weight[refn0][refn1][s->mb_y&1];
613 int weight1 = 64 - weight0;
614 luma_weight_avg( dest_y, tmp_y, h-> mb_linesize, 5, weight0, weight1, 0);
615 chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, 5, weight0, weight1, 0);
616 chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, 5, weight0, weight1, 0);
618 luma_weight_avg(dest_y, tmp_y, h->mb_linesize, h->luma_log2_weight_denom,
619 h->luma_weight[refn0][0][0] , h->luma_weight[refn1][1][0],
620 h->luma_weight[refn0][0][1] + h->luma_weight[refn1][1][1]);
621 chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
622 h->chroma_weight[refn0][0][0][0] , h->chroma_weight[refn1][1][0][0],
623 h->chroma_weight[refn0][0][0][1] + h->chroma_weight[refn1][1][0][1]);
624 chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
625 h->chroma_weight[refn0][0][1][0] , h->chroma_weight[refn1][1][1][0],
626 h->chroma_weight[refn0][0][1][1] + h->chroma_weight[refn1][1][1][1]);
629 int list = list1 ? 1 : 0;
630 int refn = h->ref_cache[list][ scan8[n] ];
631 Picture *ref= &h->ref_list[list][refn];
632 mc_dir_part(h, ref, n, square, chroma_height, delta, list,
633 dest_y, dest_cb, dest_cr, x_offset, y_offset,
634 qpix_put, chroma_put, pixel_shift, chroma444);
636 luma_weight_op(dest_y, h->mb_linesize, h->luma_log2_weight_denom,
637 h->luma_weight[refn][list][0], h->luma_weight[refn][list][1]);
638 if(h->use_weight_chroma){
639 chroma_weight_op(dest_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
640 h->chroma_weight[refn][list][0][0], h->chroma_weight[refn][list][0][1]);
641 chroma_weight_op(dest_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
642 h->chroma_weight[refn][list][1][0], h->chroma_weight[refn][list][1][1]);
647 static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
648 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
649 int x_offset, int y_offset,
650 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
651 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
652 h264_weight_func *weight_op, h264_biweight_func *weight_avg,
653 int list0, int list1, int pixel_shift, int chroma444){
654 if((h->use_weight==2 && list0 && list1
655 && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ][h->s.mb_y&1] != 32))
657 mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
658 x_offset, y_offset, qpix_put, chroma_put,
659 weight_op[0], weight_op[3], weight_avg[0],
660 weight_avg[3], list0, list1, pixel_shift, chroma444);
662 mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
663 x_offset, y_offset, qpix_put, chroma_put, qpix_avg,
664 chroma_avg, list0, list1, pixel_shift, chroma444);
667 static inline void prefetch_motion(H264Context *h, int list, int pixel_shift, int chroma444){
668 /* fetch pixels for estimated mv 4 macroblocks ahead
669 * optimized for 64byte cache lines */
670 MpegEncContext * const s = &h->s;
671 const int refn = h->ref_cache[list][scan8[0]];
673 const int mx= (h->mv_cache[list][scan8[0]][0]>>2) + 16*s->mb_x + 8;
674 const int my= (h->mv_cache[list][scan8[0]][1]>>2) + 16*s->mb_y;
675 uint8_t **src = h->ref_list[list][refn].f.data;
676 int off= (mx << pixel_shift) + (my + (s->mb_x&3)*4)*h->mb_linesize + (64 << pixel_shift);
677 s->dsp.prefetch(src[0]+off, s->linesize, 4);
679 s->dsp.prefetch(src[1]+off, s->linesize, 4);
680 s->dsp.prefetch(src[2]+off, s->linesize, 4);
682 off= (((mx>>1)+64)<<pixel_shift) + ((my>>1) + (s->mb_x&7))*s->uvlinesize;
683 s->dsp.prefetch(src[1]+off, src[2]-src[1], 2);
688 static av_always_inline void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
689 qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
690 qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
691 h264_weight_func *weight_op, h264_biweight_func *weight_avg,
692 int pixel_shift, int chroma444){
693 MpegEncContext * const s = &h->s;
694 const int mb_xy= h->mb_xy;
695 const int mb_type = s->current_picture.f.mb_type[mb_xy];
697 assert(IS_INTER(mb_type));
699 if(HAVE_PTHREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME))
701 prefetch_motion(h, 0, pixel_shift, chroma444);
703 if(IS_16X16(mb_type)){
704 mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
705 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
706 weight_op, weight_avg,
707 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1),
708 pixel_shift, chroma444);
709 }else if(IS_16X8(mb_type)){
710 mc_part(h, 0, 0, 4, 8 << pixel_shift, dest_y, dest_cb, dest_cr, 0, 0,
711 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
712 &weight_op[1], &weight_avg[1],
713 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1),
714 pixel_shift, chroma444);
715 mc_part(h, 8, 0, 4, 8 << pixel_shift, dest_y, dest_cb, dest_cr, 0, 4,
716 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
717 &weight_op[1], &weight_avg[1],
718 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1),
719 pixel_shift, chroma444);
720 }else if(IS_8X16(mb_type)){
721 mc_part(h, 0, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 0, 0,
722 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
723 &weight_op[2], &weight_avg[2],
724 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1),
725 pixel_shift, chroma444);
726 mc_part(h, 4, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 4, 0,
727 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
728 &weight_op[2], &weight_avg[2],
729 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1),
730 pixel_shift, chroma444);
734 assert(IS_8X8(mb_type));
737 const int sub_mb_type= h->sub_mb_type[i];
739 int x_offset= (i&1)<<2;
740 int y_offset= (i&2)<<1;
742 if(IS_SUB_8X8(sub_mb_type)){
743 mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
744 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
745 &weight_op[3], &weight_avg[3],
746 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
747 pixel_shift, chroma444);
748 }else if(IS_SUB_8X4(sub_mb_type)){
749 mc_part(h, n , 0, 2, 4 << pixel_shift, dest_y, dest_cb, dest_cr, x_offset, y_offset,
750 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
751 &weight_op[4], &weight_avg[4],
752 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
753 pixel_shift, chroma444);
754 mc_part(h, n+2, 0, 2, 4 << pixel_shift, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
755 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
756 &weight_op[4], &weight_avg[4],
757 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
758 pixel_shift, chroma444);
759 }else if(IS_SUB_4X8(sub_mb_type)){
760 mc_part(h, n , 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
761 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
762 &weight_op[5], &weight_avg[5],
763 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
764 pixel_shift, chroma444);
765 mc_part(h, n+1, 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
766 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
767 &weight_op[5], &weight_avg[5],
768 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
769 pixel_shift, chroma444);
772 assert(IS_SUB_4X4(sub_mb_type));
774 int sub_x_offset= x_offset + 2*(j&1);
775 int sub_y_offset= y_offset + (j&2);
776 mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
777 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
778 &weight_op[6], &weight_avg[6],
779 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
780 pixel_shift, chroma444);
786 prefetch_motion(h, 1, pixel_shift, chroma444);
789 static void free_tables(H264Context *h, int free_rbsp){
793 av_freep(&h->intra4x4_pred_mode);
794 av_freep(&h->chroma_pred_mode_table);
795 av_freep(&h->cbp_table);
796 av_freep(&h->mvd_table[0]);
797 av_freep(&h->mvd_table[1]);
798 av_freep(&h->direct_table);
799 av_freep(&h->non_zero_count);
800 av_freep(&h->slice_table_base);
801 h->slice_table= NULL;
802 av_freep(&h->list_counts);
804 av_freep(&h->mb2b_xy);
805 av_freep(&h->mb2br_xy);
807 for(i = 0; i < MAX_THREADS; i++) {
808 hx = h->thread_context[i];
810 av_freep(&hx->top_borders[1]);
811 av_freep(&hx->top_borders[0]);
812 av_freep(&hx->s.obmc_scratchpad);
814 av_freep(&hx->rbsp_buffer[1]);
815 av_freep(&hx->rbsp_buffer[0]);
816 hx->rbsp_buffer_size[0] = 0;
817 hx->rbsp_buffer_size[1] = 0;
819 if (i) av_freep(&h->thread_context[i]);
823 static void init_dequant8_coeff_table(H264Context *h){
825 const int max_qp = 51 + 6*(h->sps.bit_depth_luma-8);
828 h->dequant8_coeff[i] = h->dequant8_buffer[i];
830 if(!memcmp(h->pps.scaling_matrix8[j], h->pps.scaling_matrix8[i], 64*sizeof(uint8_t))){
831 h->dequant8_coeff[i] = h->dequant8_buffer[j];
838 for(q=0; q<max_qp+1; q++){
842 h->dequant8_coeff[i][q][(x>>3)|((x&7)<<3)] =
843 ((uint32_t)dequant8_coeff_init[idx][ dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] *
844 h->pps.scaling_matrix8[i][x]) << shift;
849 static void init_dequant4_coeff_table(H264Context *h){
851 const int max_qp = 51 + 6*(h->sps.bit_depth_luma-8);
853 h->dequant4_coeff[i] = h->dequant4_buffer[i];
855 if(!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i], 16*sizeof(uint8_t))){
856 h->dequant4_coeff[i] = h->dequant4_buffer[j];
863 for(q=0; q<max_qp+1; q++){
864 int shift = div6[q] + 2;
867 h->dequant4_coeff[i][q][(x>>2)|((x<<2)&0xF)] =
868 ((uint32_t)dequant4_coeff_init[idx][(x&1) + ((x>>2)&1)] *
869 h->pps.scaling_matrix4[i][x]) << shift;
874 static void init_dequant_tables(H264Context *h){
876 init_dequant4_coeff_table(h);
877 if(h->pps.transform_8x8_mode)
878 init_dequant8_coeff_table(h);
879 if(h->sps.transform_bypass){
882 h->dequant4_coeff[i][0][x] = 1<<6;
883 if(h->pps.transform_8x8_mode)
886 h->dequant8_coeff[i][0][x] = 1<<6;
891 int ff_h264_alloc_tables(H264Context *h){
892 MpegEncContext * const s = &h->s;
893 const int big_mb_num= s->mb_stride * (s->mb_height+1);
894 const int row_mb_num= 2*s->mb_stride*s->avctx->thread_count;
897 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->intra4x4_pred_mode, row_mb_num * 8 * sizeof(uint8_t), fail)
899 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->non_zero_count , big_mb_num * 48 * sizeof(uint8_t), fail)
900 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->slice_table_base , (big_mb_num+s->mb_stride) * sizeof(*h->slice_table_base), fail)
901 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->cbp_table, big_mb_num * sizeof(uint16_t), fail)
903 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t), fail)
904 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[0], 16*row_mb_num * sizeof(uint8_t), fail);
905 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[1], 16*row_mb_num * sizeof(uint8_t), fail);
906 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->direct_table, 4*big_mb_num * sizeof(uint8_t) , fail);
907 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->list_counts, big_mb_num * sizeof(uint8_t), fail)
909 memset(h->slice_table_base, -1, (big_mb_num+s->mb_stride) * sizeof(*h->slice_table_base));
910 h->slice_table= h->slice_table_base + s->mb_stride*2 + 1;
912 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2b_xy , big_mb_num * sizeof(uint32_t), fail);
913 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2br_xy , big_mb_num * sizeof(uint32_t), fail);
914 for(y=0; y<s->mb_height; y++){
915 for(x=0; x<s->mb_width; x++){
916 const int mb_xy= x + y*s->mb_stride;
917 const int b_xy = 4*x + 4*y*h->b_stride;
919 h->mb2b_xy [mb_xy]= b_xy;
920 h->mb2br_xy[mb_xy]= 8*(FMO ? mb_xy : (mb_xy % (2*s->mb_stride)));
924 s->obmc_scratchpad = NULL;
926 if(!h->dequant4_coeff[0])
927 init_dequant_tables(h);
936 * Mimic alloc_tables(), but for every context thread.
938 static void clone_tables(H264Context *dst, H264Context *src, int i){
939 MpegEncContext * const s = &src->s;
940 dst->intra4x4_pred_mode = src->intra4x4_pred_mode + i*8*2*s->mb_stride;
941 dst->non_zero_count = src->non_zero_count;
942 dst->slice_table = src->slice_table;
943 dst->cbp_table = src->cbp_table;
944 dst->mb2b_xy = src->mb2b_xy;
945 dst->mb2br_xy = src->mb2br_xy;
946 dst->chroma_pred_mode_table = src->chroma_pred_mode_table;
947 dst->mvd_table[0] = src->mvd_table[0] + i*8*2*s->mb_stride;
948 dst->mvd_table[1] = src->mvd_table[1] + i*8*2*s->mb_stride;
949 dst->direct_table = src->direct_table;
950 dst->list_counts = src->list_counts;
952 dst->s.obmc_scratchpad = NULL;
953 ff_h264_pred_init(&dst->hpc, src->s.codec_id, src->sps.bit_depth_luma, src->sps.chroma_format_idc);
958 * Allocate buffers which are not shared amongst multiple threads.
960 static int context_init(H264Context *h){
961 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[0], h->s.mb_width * 16*3 * sizeof(uint8_t)*2, fail)
962 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[1], h->s.mb_width * 16*3 * sizeof(uint8_t)*2, fail)
964 h->ref_cache[0][scan8[5 ]+1] = h->ref_cache[0][scan8[7 ]+1] = h->ref_cache[0][scan8[13]+1] =
965 h->ref_cache[1][scan8[5 ]+1] = h->ref_cache[1][scan8[7 ]+1] = h->ref_cache[1][scan8[13]+1] = PART_NOT_AVAILABLE;
969 return -1; // free_tables will clean up for us
972 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size);
974 static av_cold void common_init(H264Context *h){
975 MpegEncContext * const s = &h->s;
977 s->width = s->avctx->width;
978 s->height = s->avctx->height;
979 s->codec_id= s->avctx->codec->id;
981 ff_h264dsp_init(&h->h264dsp, 8, 1);
982 ff_h264_pred_init(&h->hpc, s->codec_id, 8, 1);
984 h->dequant_coeff_pps= -1;
985 s->unrestricted_mv=1;
988 dsputil_init(&s->dsp, s->avctx); // needed so that idct permutation is known early
990 memset(h->pps.scaling_matrix4, 16, 6*16*sizeof(uint8_t));
991 memset(h->pps.scaling_matrix8, 16, 2*64*sizeof(uint8_t));
994 int ff_h264_decode_extradata(H264Context *h)
996 AVCodecContext *avctx = h->s.avctx;
998 if(avctx->extradata[0] == 1){
1000 unsigned char *p = avctx->extradata;
1004 if(avctx->extradata_size < 7) {
1005 av_log(avctx, AV_LOG_ERROR, "avcC too short\n");
1008 /* sps and pps in the avcC always have length coded with 2 bytes,
1009 so put a fake nal_length_size = 2 while parsing them */
1010 h->nal_length_size = 2;
1011 // Decode sps from avcC
1012 cnt = *(p+5) & 0x1f; // Number of sps
1014 for (i = 0; i < cnt; i++) {
1015 nalsize = AV_RB16(p) + 2;
1016 if(decode_nal_units(h, p, nalsize) < 0) {
1017 av_log(avctx, AV_LOG_ERROR, "Decoding sps %d from avcC failed\n", i);
1022 // Decode pps from avcC
1023 cnt = *(p++); // Number of pps
1024 for (i = 0; i < cnt; i++) {
1025 nalsize = AV_RB16(p) + 2;
1026 if (decode_nal_units(h, p, nalsize) < 0) {
1027 av_log(avctx, AV_LOG_ERROR, "Decoding pps %d from avcC failed\n", i);
1032 // Now store right nal length size, that will be use to parse all other nals
1033 h->nal_length_size = (avctx->extradata[4] & 0x03) + 1;
1036 if(decode_nal_units(h, avctx->extradata, avctx->extradata_size) < 0)
1042 av_cold int ff_h264_decode_init(AVCodecContext *avctx){
1043 H264Context *h= avctx->priv_data;
1044 MpegEncContext * const s = &h->s;
1046 MPV_decode_defaults(s);
1051 s->out_format = FMT_H264;
1052 s->workaround_bugs= avctx->workaround_bugs;
1055 // s->decode_mb= ff_h263_decode_mb;
1056 s->quarter_sample = 1;
1057 if(!avctx->has_b_frames)
1060 avctx->chroma_sample_location = AVCHROMA_LOC_LEFT;
1062 ff_h264_decode_init_vlc();
1065 h->sps.bit_depth_luma = avctx->bits_per_raw_sample = 8;
1067 h->thread_context[0] = h;
1068 h->outputed_poc = h->next_outputed_poc = INT_MIN;
1069 h->prev_poc_msb= 1<<16;
1071 ff_h264_reset_sei(h);
1072 if(avctx->codec_id == CODEC_ID_H264){
1073 if(avctx->ticks_per_frame == 1){
1074 s->avctx->time_base.den *=2;
1076 avctx->ticks_per_frame = 2;
1079 if(avctx->extradata_size > 0 && avctx->extradata &&
1080 ff_h264_decode_extradata(h))
1083 if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames < h->sps.num_reorder_frames){
1084 s->avctx->has_b_frames = h->sps.num_reorder_frames;
1091 #define IN_RANGE(a, b, size) (((a) >= (b)) && ((a) < ((b)+(size))))
1092 static void copy_picture_range(Picture **to, Picture **from, int count, MpegEncContext *new_base, MpegEncContext *old_base)
1096 for (i=0; i<count; i++){
1097 assert((IN_RANGE(from[i], old_base, sizeof(*old_base)) ||
1098 IN_RANGE(from[i], old_base->picture, sizeof(Picture) * old_base->picture_count) ||
1100 to[i] = REBASE_PICTURE(from[i], new_base, old_base);
1104 static void copy_parameter_set(void **to, void **from, int count, int size)
1108 for (i=0; i<count; i++){
1109 if (to[i] && !from[i]) av_freep(&to[i]);
1110 else if (from[i] && !to[i]) to[i] = av_malloc(size);
1112 if (from[i]) memcpy(to[i], from[i], size);
1116 static int decode_init_thread_copy(AVCodecContext *avctx){
1117 H264Context *h= avctx->priv_data;
1119 if (!avctx->is_copy) return 0;
1120 memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
1121 memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
1126 #define copy_fields(to, from, start_field, end_field) memcpy(&to->start_field, &from->start_field, (char*)&to->end_field - (char*)&to->start_field)
1127 static int decode_update_thread_context(AVCodecContext *dst, const AVCodecContext *src){
1128 H264Context *h= dst->priv_data, *h1= src->priv_data;
1129 MpegEncContext * const s = &h->s, * const s1 = &h1->s;
1130 int inited = s->context_initialized, err;
1133 if(dst == src || !s1->context_initialized) return 0;
1135 err = ff_mpeg_update_thread_context(dst, src);
1138 //FIXME handle width/height changing
1140 for(i = 0; i < MAX_SPS_COUNT; i++)
1141 av_freep(h->sps_buffers + i);
1143 for(i = 0; i < MAX_PPS_COUNT; i++)
1144 av_freep(h->pps_buffers + i);
1146 memcpy(&h->s + 1, &h1->s + 1, sizeof(H264Context) - sizeof(MpegEncContext)); //copy all fields after MpegEnc
1147 memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
1148 memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
1149 if (ff_h264_alloc_tables(h) < 0) {
1150 av_log(dst, AV_LOG_ERROR, "Could not allocate memory for h264\n");
1151 return AVERROR(ENOMEM);
1156 h->rbsp_buffer[i] = NULL;
1157 h->rbsp_buffer_size[i] = 0;
1160 h->thread_context[0] = h;
1162 // frame_start may not be called for the next thread (if it's decoding a bottom field)
1163 // so this has to be allocated here
1164 h->s.obmc_scratchpad = av_malloc(16*6*s->linesize);
1166 s->dsp.clear_blocks(h->mb);
1167 s->dsp.clear_blocks(h->mb+(24*16<<h->pixel_shift));
1170 //extradata/NAL handling
1171 h->is_avc = h1->is_avc;
1174 copy_parameter_set((void**)h->sps_buffers, (void**)h1->sps_buffers, MAX_SPS_COUNT, sizeof(SPS));
1176 copy_parameter_set((void**)h->pps_buffers, (void**)h1->pps_buffers, MAX_PPS_COUNT, sizeof(PPS));
1179 //Dequantization matrices
1180 //FIXME these are big - can they be only copied when PPS changes?
1181 copy_fields(h, h1, dequant4_buffer, dequant4_coeff);
1184 h->dequant4_coeff[i] = h->dequant4_buffer[0] + (h1->dequant4_coeff[i] - h1->dequant4_buffer[0]);
1187 h->dequant8_coeff[i] = h->dequant8_buffer[0] + (h1->dequant8_coeff[i] - h1->dequant8_buffer[0]);
1189 h->dequant_coeff_pps = h1->dequant_coeff_pps;
1192 copy_fields(h, h1, poc_lsb, redundant_pic_count);
1195 copy_fields(h, h1, ref_count, list_count);
1196 copy_fields(h, h1, ref_list, intra_gb);
1197 copy_fields(h, h1, short_ref, cabac_init_idc);
1199 copy_picture_range(h->short_ref, h1->short_ref, 32, s, s1);
1200 copy_picture_range(h->long_ref, h1->long_ref, 32, s, s1);
1201 copy_picture_range(h->delayed_pic, h1->delayed_pic, MAX_DELAYED_PIC_COUNT+2, s, s1);
1203 h->last_slice_type = h1->last_slice_type;
1205 if(!s->current_picture_ptr) return 0;
1208 err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1209 h->prev_poc_msb = h->poc_msb;
1210 h->prev_poc_lsb = h->poc_lsb;
1212 h->prev_frame_num_offset= h->frame_num_offset;
1213 h->prev_frame_num = h->frame_num;
1214 h->outputed_poc = h->next_outputed_poc;
1219 int ff_h264_frame_start(H264Context *h){
1220 MpegEncContext * const s = &h->s;
1222 const int pixel_shift = h->pixel_shift;
1223 int thread_count = (s->avctx->active_thread_type & FF_THREAD_SLICE) ? s->avctx->thread_count : 1;
1225 if(MPV_frame_start(s, s->avctx) < 0)
1227 ff_er_frame_start(s);
1229 * MPV_frame_start uses pict_type to derive key_frame.
1230 * This is incorrect for H.264; IDR markings must be used.
1231 * Zero here; IDR markings per slice in frame or fields are ORed in later.
1232 * See decode_nal_units().
1234 s->current_picture_ptr->f.key_frame = 0;
1235 s->current_picture_ptr->mmco_reset= 0;
1237 assert(s->linesize && s->uvlinesize);
1239 for(i=0; i<16; i++){
1240 h->block_offset[i]= (4*((scan8[i] - scan8[0])&7) << pixel_shift) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
1241 h->block_offset[48+i]= (4*((scan8[i] - scan8[0])&7) << pixel_shift) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
1243 for(i=0; i<16; i++){
1244 h->block_offset[16+i]=
1245 h->block_offset[32+i]= (4*((scan8[i] - scan8[0])&7) << pixel_shift) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
1246 h->block_offset[48+16+i]=
1247 h->block_offset[48+32+i]= (4*((scan8[i] - scan8[0])&7) << pixel_shift) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
1250 /* can't be in alloc_tables because linesize isn't known there.
1251 * FIXME: redo bipred weight to not require extra buffer? */
1252 for(i = 0; i < thread_count; i++)
1253 if(h->thread_context[i] && !h->thread_context[i]->s.obmc_scratchpad)
1254 h->thread_context[i]->s.obmc_scratchpad = av_malloc(16*6*s->linesize);
1256 /* some macroblocks can be accessed before they're available in case of lost slices, mbaff or threading*/
1257 memset(h->slice_table, -1, (s->mb_height*s->mb_stride-1) * sizeof(*h->slice_table));
1259 // s->decode = (s->flags & CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.f.reference /*|| h->contains_intra*/ || 1;
1261 // We mark the current picture as non-reference after allocating it, so
1262 // that if we break out due to an error it can be released automatically
1263 // in the next MPV_frame_start().
1264 // SVQ3 as well as most other codecs have only last/next/current and thus
1265 // get released even with set reference, besides SVQ3 and others do not
1266 // mark frames as reference later "naturally".
1267 if(s->codec_id != CODEC_ID_SVQ3)
1268 s->current_picture_ptr->f.reference = 0;
1270 s->current_picture_ptr->field_poc[0]=
1271 s->current_picture_ptr->field_poc[1]= INT_MAX;
1273 h->next_output_pic = NULL;
1275 assert(s->current_picture_ptr->long_ref==0);
1281 * Run setup operations that must be run after slice header decoding.
1282 * This includes finding the next displayed frame.
1284 * @param h h264 master context
1285 * @param setup_finished enough NALs have been read that we can call
1286 * ff_thread_finish_setup()
1288 static void decode_postinit(H264Context *h, int setup_finished){
1289 MpegEncContext * const s = &h->s;
1290 Picture *out = s->current_picture_ptr;
1291 Picture *cur = s->current_picture_ptr;
1292 int i, pics, out_of_order, out_idx;
1294 s->current_picture_ptr->f.qscale_type = FF_QSCALE_TYPE_H264;
1295 s->current_picture_ptr->f.pict_type = s->pict_type;
1297 if (h->next_output_pic) return;
1299 if (cur->field_poc[0]==INT_MAX || cur->field_poc[1]==INT_MAX) {
1300 //FIXME: if we have two PAFF fields in one packet, we can't start the next thread here.
1301 //If we have one field per packet, we can. The check in decode_nal_units() is not good enough
1302 //to find this yet, so we assume the worst for now.
1303 //if (setup_finished)
1304 // ff_thread_finish_setup(s->avctx);
1308 cur->f.interlaced_frame = 0;
1309 cur->f.repeat_pict = 0;
1311 /* Signal interlacing information externally. */
1312 /* Prioritize picture timing SEI information over used decoding process if it exists. */
1314 if(h->sps.pic_struct_present_flag){
1315 switch (h->sei_pic_struct)
1317 case SEI_PIC_STRUCT_FRAME:
1319 case SEI_PIC_STRUCT_TOP_FIELD:
1320 case SEI_PIC_STRUCT_BOTTOM_FIELD:
1321 cur->f.interlaced_frame = 1;
1323 case SEI_PIC_STRUCT_TOP_BOTTOM:
1324 case SEI_PIC_STRUCT_BOTTOM_TOP:
1325 if (FIELD_OR_MBAFF_PICTURE)
1326 cur->f.interlaced_frame = 1;
1328 // try to flag soft telecine progressive
1329 cur->f.interlaced_frame = h->prev_interlaced_frame;
1331 case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
1332 case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
1333 // Signal the possibility of telecined film externally (pic_struct 5,6)
1334 // From these hints, let the applications decide if they apply deinterlacing.
1335 cur->f.repeat_pict = 1;
1337 case SEI_PIC_STRUCT_FRAME_DOUBLING:
1338 // Force progressive here, as doubling interlaced frame is a bad idea.
1339 cur->f.repeat_pict = 2;
1341 case SEI_PIC_STRUCT_FRAME_TRIPLING:
1342 cur->f.repeat_pict = 4;
1346 if ((h->sei_ct_type & 3) && h->sei_pic_struct <= SEI_PIC_STRUCT_BOTTOM_TOP)
1347 cur->f.interlaced_frame = (h->sei_ct_type & (1 << 1)) != 0;
1349 /* Derive interlacing flag from used decoding process. */
1350 cur->f.interlaced_frame = FIELD_OR_MBAFF_PICTURE;
1352 h->prev_interlaced_frame = cur->f.interlaced_frame;
1354 if (cur->field_poc[0] != cur->field_poc[1]){
1355 /* Derive top_field_first from field pocs. */
1356 cur->f.top_field_first = cur->field_poc[0] < cur->field_poc[1];
1358 if (cur->f.interlaced_frame || h->sps.pic_struct_present_flag) {
1359 /* Use picture timing SEI information. Even if it is a information of a past frame, better than nothing. */
1360 if(h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM
1361 || h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM_TOP)
1362 cur->f.top_field_first = 1;
1364 cur->f.top_field_first = 0;
1366 /* Most likely progressive */
1367 cur->f.top_field_first = 0;
1371 //FIXME do something with unavailable reference frames
1373 /* Sort B-frames into display order */
1375 if(h->sps.bitstream_restriction_flag
1376 && s->avctx->has_b_frames < h->sps.num_reorder_frames){
1377 s->avctx->has_b_frames = h->sps.num_reorder_frames;
1381 if( s->avctx->strict_std_compliance >= FF_COMPLIANCE_STRICT
1382 && !h->sps.bitstream_restriction_flag){
1383 s->avctx->has_b_frames= MAX_DELAYED_PIC_COUNT;
1388 while(h->delayed_pic[pics]) pics++;
1390 av_assert0(pics <= MAX_DELAYED_PIC_COUNT);
1392 h->delayed_pic[pics++] = cur;
1393 if (cur->f.reference == 0)
1394 cur->f.reference = DELAYED_PIC_REF;
1396 out = h->delayed_pic[0];
1398 for (i = 1; h->delayed_pic[i] && !h->delayed_pic[i]->f.key_frame && !h->delayed_pic[i]->mmco_reset; i++)
1399 if(h->delayed_pic[i]->poc < out->poc){
1400 out = h->delayed_pic[i];
1403 if (s->avctx->has_b_frames == 0 && (h->delayed_pic[0]->f.key_frame || h->delayed_pic[0]->mmco_reset))
1404 h->next_outputed_poc= INT_MIN;
1405 out_of_order = out->poc < h->next_outputed_poc;
1407 if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames >= h->sps.num_reorder_frames)
1409 else if((out_of_order && pics-1 == s->avctx->has_b_frames && s->avctx->has_b_frames < MAX_DELAYED_PIC_COUNT)
1411 ((h->next_outputed_poc != INT_MIN && out->poc > h->next_outputed_poc + 2)
1412 || cur->f.pict_type == AV_PICTURE_TYPE_B)))
1415 s->avctx->has_b_frames++;
1418 if(out_of_order || pics > s->avctx->has_b_frames){
1419 out->f.reference &= ~DELAYED_PIC_REF;
1420 out->owner2 = s; // for frame threading, the owner must be the second field's thread
1421 // or else the first thread can release the picture and reuse it unsafely
1422 for(i=out_idx; h->delayed_pic[i]; i++)
1423 h->delayed_pic[i] = h->delayed_pic[i+1];
1425 if(!out_of_order && pics > s->avctx->has_b_frames){
1426 h->next_output_pic = out;
1427 if (out_idx == 0 && h->delayed_pic[0] && (h->delayed_pic[0]->f.key_frame || h->delayed_pic[0]->mmco_reset)) {
1428 h->next_outputed_poc = INT_MIN;
1430 h->next_outputed_poc = out->poc;
1432 av_log(s->avctx, AV_LOG_DEBUG, "no picture\n");
1436 ff_thread_finish_setup(s->avctx);
1439 static av_always_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){
1440 MpegEncContext * const s = &h->s;
1441 uint8_t *top_border;
1443 const int pixel_shift = h->pixel_shift;
1444 int chroma444 = CHROMA444;
1445 int chroma422 = CHROMA422;
1448 src_cb -= uvlinesize;
1449 src_cr -= uvlinesize;
1451 if(!simple && FRAME_MBAFF){
1454 top_border = h->top_borders[0][s->mb_x];
1455 AV_COPY128(top_border, src_y + 15*linesize);
1457 AV_COPY128(top_border+16, src_y+15*linesize+16);
1458 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1461 AV_COPY128(top_border+32, src_cb + 15*uvlinesize);
1462 AV_COPY128(top_border+48, src_cb + 15*uvlinesize+16);
1463 AV_COPY128(top_border+64, src_cr + 15*uvlinesize);
1464 AV_COPY128(top_border+80, src_cr + 15*uvlinesize+16);
1466 AV_COPY128(top_border+16, src_cb + 15*uvlinesize);
1467 AV_COPY128(top_border+32, src_cr + 15*uvlinesize);
1469 } else if(chroma422){
1471 AV_COPY128(top_border+32, src_cb + 15*uvlinesize);
1472 AV_COPY128(top_border+48, src_cr + 15*uvlinesize);
1474 AV_COPY64(top_border+16, src_cb + 15*uvlinesize);
1475 AV_COPY64(top_border+24, src_cr + 15*uvlinesize);
1479 AV_COPY128(top_border+32, src_cb+7*uvlinesize);
1480 AV_COPY128(top_border+48, src_cr+7*uvlinesize);
1482 AV_COPY64(top_border+16, src_cb+7*uvlinesize);
1483 AV_COPY64(top_border+24, src_cr+7*uvlinesize);
1494 top_border = h->top_borders[top_idx][s->mb_x];
1495 // There are two lines saved, the line above the the top macroblock of a pair,
1496 // and the line above the bottom macroblock
1497 AV_COPY128(top_border, src_y + 16*linesize);
1499 AV_COPY128(top_border+16, src_y+16*linesize+16);
1501 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1504 AV_COPY128(top_border+32, src_cb + 16*linesize);
1505 AV_COPY128(top_border+48, src_cb + 16*linesize+16);
1506 AV_COPY128(top_border+64, src_cr + 16*linesize);
1507 AV_COPY128(top_border+80, src_cr + 16*linesize+16);
1509 AV_COPY128(top_border+16, src_cb + 16*linesize);
1510 AV_COPY128(top_border+32, src_cr + 16*linesize);
1512 } else if(chroma422) {
1514 AV_COPY128(top_border+32, src_cb+16*uvlinesize);
1515 AV_COPY128(top_border+48, src_cr+16*uvlinesize);
1517 AV_COPY64(top_border+16, src_cb+16*uvlinesize);
1518 AV_COPY64(top_border+24, src_cr+16*uvlinesize);
1522 AV_COPY128(top_border+32, src_cb+8*uvlinesize);
1523 AV_COPY128(top_border+48, src_cr+8*uvlinesize);
1525 AV_COPY64(top_border+16, src_cb+8*uvlinesize);
1526 AV_COPY64(top_border+24, src_cr+8*uvlinesize);
1532 static av_always_inline void xchg_mb_border(H264Context *h, uint8_t *src_y,
1533 uint8_t *src_cb, uint8_t *src_cr,
1534 int linesize, int uvlinesize,
1535 int xchg, int chroma444,
1536 int simple, int pixel_shift){
1537 MpegEncContext * const s = &h->s;
1538 int deblock_topleft;
1541 uint8_t *top_border_m1;
1542 uint8_t *top_border;
1544 if(!simple && FRAME_MBAFF){
1549 top_idx = MB_MBAFF ? 0 : 1;
1553 if(h->deblocking_filter == 2) {
1554 deblock_topleft = h->slice_table[h->mb_xy - 1 - s->mb_stride] == h->slice_num;
1555 deblock_top = h->top_type;
1557 deblock_topleft = (s->mb_x > 0);
1558 deblock_top = (s->mb_y > !!MB_FIELD);
1561 src_y -= linesize + 1 + pixel_shift;
1562 src_cb -= uvlinesize + 1 + pixel_shift;
1563 src_cr -= uvlinesize + 1 + pixel_shift;
1565 top_border_m1 = h->top_borders[top_idx][s->mb_x-1];
1566 top_border = h->top_borders[top_idx][s->mb_x];
1568 #define XCHG(a,b,xchg)\
1571 AV_SWAP64(b+0,a+0);\
1572 AV_SWAP64(b+8,a+8);\
1577 if (xchg) AV_SWAP64(b,a);\
1578 else AV_COPY64(b,a);
1581 if(deblock_topleft){
1582 XCHG(top_border_m1 + (8 << pixel_shift), src_y - (7 << pixel_shift), 1);
1584 XCHG(top_border + (0 << pixel_shift), src_y + (1 << pixel_shift), xchg);
1585 XCHG(top_border + (8 << pixel_shift), src_y + (9 << pixel_shift), 1);
1586 if(s->mb_x+1 < s->mb_width){
1587 XCHG(h->top_borders[top_idx][s->mb_x+1], src_y + (17 << pixel_shift), 1);
1590 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1592 if(deblock_topleft){
1593 XCHG(top_border_m1 + (24 << pixel_shift), src_cb - (7 << pixel_shift), 1);
1594 XCHG(top_border_m1 + (40 << pixel_shift), src_cr - (7 << pixel_shift), 1);
1596 XCHG(top_border + (16 << pixel_shift), src_cb + (1 << pixel_shift), xchg);
1597 XCHG(top_border + (24 << pixel_shift), src_cb + (9 << pixel_shift), 1);
1598 XCHG(top_border + (32 << pixel_shift), src_cr + (1 << pixel_shift), xchg);
1599 XCHG(top_border + (40 << pixel_shift), src_cr + (9 << pixel_shift), 1);
1600 if(s->mb_x+1 < s->mb_width){
1601 XCHG(h->top_borders[top_idx][s->mb_x+1] + (16 << pixel_shift), src_cb + (17 << pixel_shift), 1);
1602 XCHG(h->top_borders[top_idx][s->mb_x+1] + (32 << pixel_shift), src_cr + (17 << pixel_shift), 1);
1606 if(deblock_topleft){
1607 XCHG(top_border_m1 + (16 << pixel_shift), src_cb - (7 << pixel_shift), 1);
1608 XCHG(top_border_m1 + (24 << pixel_shift), src_cr - (7 << pixel_shift), 1);
1610 XCHG(top_border + (16 << pixel_shift), src_cb+1+pixel_shift, 1);
1611 XCHG(top_border + (24 << pixel_shift), src_cr+1+pixel_shift, 1);
1617 static av_always_inline int dctcoef_get(DCTELEM *mb, int high_bit_depth, int index) {
1618 if (high_bit_depth) {
1619 return AV_RN32A(((int32_t*)mb) + index);
1621 return AV_RN16A(mb + index);
1624 static av_always_inline void dctcoef_set(DCTELEM *mb, int high_bit_depth, int index, int value) {
1625 if (high_bit_depth) {
1626 AV_WN32A(((int32_t*)mb) + index, value);
1628 AV_WN16A(mb + index, value);
1631 static av_always_inline void hl_decode_mb_predict_luma(H264Context *h, int mb_type, int is_h264, int simple, int transform_bypass,
1632 int pixel_shift, int *block_offset, int linesize, uint8_t *dest_y, int p)
1634 MpegEncContext * const s = &h->s;
1635 void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1636 void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
1638 int qscale = p == 0 ? s->qscale : h->chroma_qp[p-1];
1639 block_offset += 16*p;
1640 if(IS_INTRA4x4(mb_type)){
1641 if(simple || !s->encoding){
1642 if(IS_8x8DCT(mb_type)){
1643 if(transform_bypass){
1645 idct_add = s->dsp.add_pixels8;
1647 idct_dc_add = h->h264dsp.h264_idct8_dc_add;
1648 idct_add = h->h264dsp.h264_idct8_add;
1650 for(i=0; i<16; i+=4){
1651 uint8_t * const ptr= dest_y + block_offset[i];
1652 const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1653 if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1654 h->hpc.pred8x8l_add[dir](ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1656 const int nnz = h->non_zero_count_cache[ scan8[i+p*16] ];
1657 h->hpc.pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
1658 (h->topright_samples_available<<i)&0x4000, linesize);
1660 if(nnz == 1 && dctcoef_get(h->mb, pixel_shift, i*16+p*256))
1661 idct_dc_add(ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1663 idct_add (ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1668 if(transform_bypass){
1670 idct_add = s->dsp.add_pixels4;
1672 idct_dc_add = h->h264dsp.h264_idct_dc_add;
1673 idct_add = h->h264dsp.h264_idct_add;
1675 for(i=0; i<16; i++){
1676 uint8_t * const ptr= dest_y + block_offset[i];
1677 const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1679 if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1680 h->hpc.pred4x4_add[dir](ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1685 if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
1686 const int topright_avail= (h->topright_samples_available<<i)&0x8000;
1687 assert(s->mb_y || linesize <= block_offset[i]);
1688 if(!topright_avail){
1690 tr_high= ((uint16_t*)ptr)[3 - linesize/2]*0x0001000100010001ULL;
1691 topright= (uint8_t*) &tr_high;
1693 tr= ptr[3 - linesize]*0x01010101;
1694 topright= (uint8_t*) &tr;
1697 topright= ptr + (4 << pixel_shift) - linesize;
1701 h->hpc.pred4x4[ dir ](ptr, topright, linesize);
1702 nnz = h->non_zero_count_cache[ scan8[i+p*16] ];
1705 if(nnz == 1 && dctcoef_get(h->mb, pixel_shift, i*16+p*256))
1706 idct_dc_add(ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1708 idct_add (ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1710 ff_svq3_add_idct_c(ptr, h->mb + i*16+p*256, linesize, qscale, 0);
1717 h->hpc.pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
1719 if(h->non_zero_count_cache[ scan8[LUMA_DC_BLOCK_INDEX+p] ]){
1720 if(!transform_bypass)
1721 h->h264dsp.h264_luma_dc_dequant_idct(h->mb+(p*256 << pixel_shift), h->mb_luma_dc[p], h->dequant4_coeff[p][qscale][0]);
1723 static const uint8_t dc_mapping[16] = { 0*16, 1*16, 4*16, 5*16, 2*16, 3*16, 6*16, 7*16,
1724 8*16, 9*16,12*16,13*16,10*16,11*16,14*16,15*16};
1725 for(i = 0; i < 16; i++)
1726 dctcoef_set(h->mb+p*256, pixel_shift, dc_mapping[i], dctcoef_get(h->mb_luma_dc[p], pixel_shift, i));
1730 ff_svq3_luma_dc_dequant_idct_c(h->mb+p*256, h->mb_luma_dc[p], qscale);
1734 static av_always_inline void hl_decode_mb_idct_luma(H264Context *h, int mb_type, int is_h264, int simple, int transform_bypass,
1735 int pixel_shift, int *block_offset, int linesize, uint8_t *dest_y, int p)
1737 MpegEncContext * const s = &h->s;
1738 void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1740 block_offset += 16*p;
1741 if(!IS_INTRA4x4(mb_type)){
1743 if(IS_INTRA16x16(mb_type)){
1744 if(transform_bypass){
1745 if(h->sps.profile_idc==244 && (h->intra16x16_pred_mode==VERT_PRED8x8 || h->intra16x16_pred_mode==HOR_PRED8x8)){
1746 h->hpc.pred16x16_add[h->intra16x16_pred_mode](dest_y, block_offset, h->mb + (p*256 << pixel_shift), linesize);
1748 for(i=0; i<16; i++){
1749 if(h->non_zero_count_cache[ scan8[i+p*16] ] || dctcoef_get(h->mb, pixel_shift, i*16+p*256))
1750 s->dsp.add_pixels4(dest_y + block_offset[i], h->mb + (i*16+p*256 << pixel_shift), linesize);
1754 h->h264dsp.h264_idct_add16intra(dest_y, block_offset, h->mb + (p*256 << pixel_shift), linesize, h->non_zero_count_cache+p*5*8);
1756 }else if(h->cbp&15){
1757 if(transform_bypass){
1758 const int di = IS_8x8DCT(mb_type) ? 4 : 1;
1759 idct_add= IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4;
1760 for(i=0; i<16; i+=di){
1761 if(h->non_zero_count_cache[ scan8[i+p*16] ]){
1762 idct_add(dest_y + block_offset[i], h->mb + (i*16+p*256 << pixel_shift), linesize);
1766 if(IS_8x8DCT(mb_type)){
1767 h->h264dsp.h264_idct8_add4(dest_y, block_offset, h->mb + (p*256 << pixel_shift), linesize, h->non_zero_count_cache+p*5*8);
1769 h->h264dsp.h264_idct_add16(dest_y, block_offset, h->mb + (p*256 << pixel_shift), linesize, h->non_zero_count_cache+p*5*8);
1774 for(i=0; i<16; i++){
1775 if(h->non_zero_count_cache[ scan8[i+p*16] ] || h->mb[i*16+p*256]){ //FIXME benchmark weird rule, & below
1776 uint8_t * const ptr= dest_y + block_offset[i];
1777 ff_svq3_add_idct_c(ptr, h->mb + i*16 + p*256, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
1784 static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple, int pixel_shift){
1785 MpegEncContext * const s = &h->s;
1786 const int mb_x= s->mb_x;
1787 const int mb_y= s->mb_y;
1788 const int mb_xy= h->mb_xy;
1789 const int mb_type = s->current_picture.f.mb_type[mb_xy];
1790 uint8_t *dest_y, *dest_cb, *dest_cr;
1791 int linesize, uvlinesize /*dct_offset*/;
1793 int *block_offset = &h->block_offset[0];
1794 const int transform_bypass = !simple && (s->qscale == 0 && h->sps.transform_bypass);
1795 /* is_h264 should always be true if SVQ3 is disabled. */
1796 const int is_h264 = !CONFIG_SVQ3_DECODER || simple || s->codec_id == CODEC_ID_H264;
1797 void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1798 const int block_h = 16>>s->chroma_y_shift;
1800 dest_y = s->current_picture.f.data[0] + ((mb_x << pixel_shift) + mb_y * s->linesize ) * 16;
1801 dest_cb = s->current_picture.f.data[1] + (mb_x << pixel_shift)*8 + mb_y * s->uvlinesize * block_h;
1802 dest_cr = s->current_picture.f.data[2] + (mb_x << pixel_shift)*8 + mb_y * s->uvlinesize * block_h;
1804 s->dsp.prefetch(dest_y + (s->mb_x&3)*4*s->linesize + (64 << pixel_shift), s->linesize, 4);
1805 s->dsp.prefetch(dest_cb + (s->mb_x&7)*s->uvlinesize + (64 << pixel_shift), dest_cr - dest_cb, 2);
1807 h->list_counts[mb_xy]= h->list_count;
1809 if (!simple && MB_FIELD) {
1810 linesize = h->mb_linesize = s->linesize * 2;
1811 uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
1812 block_offset = &h->block_offset[48];
1813 if(mb_y&1){ //FIXME move out of this function?
1814 dest_y -= s->linesize*15;
1815 dest_cb-= s->uvlinesize*(block_h-1);
1816 dest_cr-= s->uvlinesize*(block_h-1);
1820 for(list=0; list<h->list_count; list++){
1821 if(!USES_LIST(mb_type, list))
1823 if(IS_16X16(mb_type)){
1824 int8_t *ref = &h->ref_cache[list][scan8[0]];
1825 fill_rectangle(ref, 4, 4, 8, (16+*ref)^(s->mb_y&1), 1);
1827 for(i=0; i<16; i+=4){
1828 int ref = h->ref_cache[list][scan8[i]];
1830 fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, (16+ref)^(s->mb_y&1), 1);
1836 linesize = h->mb_linesize = s->linesize;
1837 uvlinesize = h->mb_uvlinesize = s->uvlinesize;
1838 // dct_offset = s->linesize * 16;
1841 if (!simple && IS_INTRA_PCM(mb_type)) {
1842 const int bit_depth = h->sps.bit_depth_luma;
1846 init_get_bits(&gb, (uint8_t*)h->mb, 384*bit_depth);
1848 for (i = 0; i < 16; i++) {
1849 uint16_t *tmp_y = (uint16_t*)(dest_y + i*linesize);
1850 for (j = 0; j < 16; j++)
1851 tmp_y[j] = get_bits(&gb, bit_depth);
1853 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1854 if (!h->sps.chroma_format_idc) {
1855 for (i = 0; i < 8; i++) {
1856 uint16_t *tmp_cb = (uint16_t*)(dest_cb + i*uvlinesize);
1857 uint16_t *tmp_cr = (uint16_t*)(dest_cr + i*uvlinesize);
1858 for (j = 0; j < 8; j++) {
1859 tmp_cb[j] = tmp_cr[j] = 1 << (bit_depth - 1);
1863 for (i = 0; i < block_h; i++) {
1864 uint16_t *tmp_cb = (uint16_t*)(dest_cb + i*uvlinesize);
1865 for (j = 0; j < 8; j++)
1866 tmp_cb[j] = get_bits(&gb, bit_depth);
1868 for (i = 0; i < block_h; i++) {
1869 uint16_t *tmp_cr = (uint16_t*)(dest_cr + i*uvlinesize);
1870 for (j = 0; j < 8; j++)
1871 tmp_cr[j] = get_bits(&gb, bit_depth);
1876 for (i=0; i<16; i++) {
1877 memcpy(dest_y + i* linesize, h->mb + i*8, 16);
1879 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1880 if (!h->sps.chroma_format_idc) {
1881 for (i=0; i<8; i++) {
1882 memset(dest_cb+ i*uvlinesize, 1 << (bit_depth - 1), 8);
1883 memset(dest_cr+ i*uvlinesize, 1 << (bit_depth - 1), 8);
1886 for (i=0; i<block_h; i++) {
1887 memcpy(dest_cb+ i*uvlinesize, h->mb + 128 + i*4, 8);
1888 memcpy(dest_cr+ i*uvlinesize, h->mb + 160 + i*4, 8);
1894 if(IS_INTRA(mb_type)){
1895 if(h->deblocking_filter)
1896 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1, 0, simple, pixel_shift);
1898 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1899 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
1900 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
1903 hl_decode_mb_predict_luma(h, mb_type, is_h264, simple, transform_bypass, pixel_shift, block_offset, linesize, dest_y, 0);
1905 if(h->deblocking_filter)
1906 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0, 0, simple, pixel_shift);
1908 hl_motion(h, dest_y, dest_cb, dest_cr,
1909 s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
1910 s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
1911 h->h264dsp.weight_h264_pixels_tab,
1912 h->h264dsp.biweight_h264_pixels_tab, pixel_shift, 0);
1915 hl_decode_mb_idct_luma(h, mb_type, is_h264, simple, transform_bypass, pixel_shift, block_offset, linesize, dest_y, 0);
1917 if((simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)) && (h->cbp&0x30)){
1918 uint8_t *dest[2] = {dest_cb, dest_cr};
1919 if(transform_bypass){
1920 if(IS_INTRA(mb_type) && h->sps.profile_idc==244 && (h->chroma_pred_mode==VERT_PRED8x8 || h->chroma_pred_mode==HOR_PRED8x8)){
1921 h->hpc.pred8x8_add[h->chroma_pred_mode](dest[0], block_offset + 16, h->mb + (16*16*1 << pixel_shift), uvlinesize);
1922 h->hpc.pred8x8_add[h->chroma_pred_mode](dest[1], block_offset + 32, h->mb + (16*16*2 << pixel_shift), uvlinesize);
1924 idct_add = s->dsp.add_pixels4;
1926 for(i=j*16; i<j*16+4; i++){
1927 if(h->non_zero_count_cache[ scan8[i] ] || dctcoef_get(h->mb, pixel_shift, i*16))
1928 idct_add (dest[j-1] + block_offset[i], h->mb + (i*16 << pixel_shift), uvlinesize);
1936 qp[0] = h->chroma_qp[0]+3;
1937 qp[1] = h->chroma_qp[1]+3;
1939 qp[0] = h->chroma_qp[0];
1940 qp[1] = h->chroma_qp[1];
1942 if(h->non_zero_count_cache[ scan8[CHROMA_DC_BLOCK_INDEX+0] ])
1943 h->h264dsp.h264_chroma_dc_dequant_idct(h->mb + (16*16*1 << pixel_shift), h->dequant4_coeff[IS_INTRA(mb_type) ? 1:4][qp[0]][0]);
1944 if(h->non_zero_count_cache[ scan8[CHROMA_DC_BLOCK_INDEX+1] ])
1945 h->h264dsp.h264_chroma_dc_dequant_idct(h->mb + (16*16*2 << pixel_shift), h->dequant4_coeff[IS_INTRA(mb_type) ? 2:5][qp[1]][0]);
1946 h->h264dsp.h264_idct_add8(dest, block_offset,
1948 h->non_zero_count_cache);
1950 #if CONFIG_SVQ3_DECODER
1952 h->h264dsp.h264_chroma_dc_dequant_idct(h->mb + 16*16*1, h->dequant4_coeff[IS_INTRA(mb_type) ? 1:4][h->chroma_qp[0]][0]);
1953 h->h264dsp.h264_chroma_dc_dequant_idct(h->mb + 16*16*2, h->dequant4_coeff[IS_INTRA(mb_type) ? 2:5][h->chroma_qp[1]][0]);
1955 for(i=j*16; i<j*16+4; i++){
1956 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
1957 uint8_t * const ptr= dest[j-1] + block_offset[i];
1958 ff_svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, ff_h264_chroma_qp[0][s->qscale + 12] - 12, 2);
1967 if(h->cbp || IS_INTRA(mb_type))
1969 s->dsp.clear_blocks(h->mb);
1970 s->dsp.clear_blocks(h->mb+(24*16<<pixel_shift));
1974 static av_always_inline void hl_decode_mb_444_internal(H264Context *h, int simple, int pixel_shift){
1975 MpegEncContext * const s = &h->s;
1976 const int mb_x= s->mb_x;
1977 const int mb_y= s->mb_y;
1978 const int mb_xy= h->mb_xy;
1979 const int mb_type = s->current_picture.f.mb_type[mb_xy];
1983 int *block_offset = &h->block_offset[0];
1984 const int transform_bypass = !simple && (s->qscale == 0 && h->sps.transform_bypass);
1985 const int plane_count = (simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)) ? 3 : 1;
1987 for (p = 0; p < plane_count; p++)
1989 dest[p] = s->current_picture.f.data[p] + ((mb_x << pixel_shift) + mb_y * s->linesize) * 16;
1990 s->dsp.prefetch(dest[p] + (s->mb_x&3)*4*s->linesize + (64 << pixel_shift), s->linesize, 4);
1993 h->list_counts[mb_xy]= h->list_count;
1995 if (!simple && MB_FIELD) {
1996 linesize = h->mb_linesize = h->mb_uvlinesize = s->linesize * 2;
1997 block_offset = &h->block_offset[48];
1998 if(mb_y&1) //FIXME move out of this function?
1999 for (p = 0; p < 3; p++)
2000 dest[p] -= s->linesize*15;
2003 for(list=0; list<h->list_count; list++){
2004 if(!USES_LIST(mb_type, list))
2006 if(IS_16X16(mb_type)){
2007 int8_t *ref = &h->ref_cache[list][scan8[0]];
2008 fill_rectangle(ref, 4, 4, 8, (16+*ref)^(s->mb_y&1), 1);
2010 for(i=0; i<16; i+=4){
2011 int ref = h->ref_cache[list][scan8[i]];
2013 fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, (16+ref)^(s->mb_y&1), 1);
2019 linesize = h->mb_linesize = h->mb_uvlinesize = s->linesize;
2022 if (!simple && IS_INTRA_PCM(mb_type)) {
2024 const int bit_depth = h->sps.bit_depth_luma;
2026 init_get_bits(&gb, (uint8_t*)h->mb, 768*bit_depth);
2028 for (p = 0; p < plane_count; p++) {
2029 for (i = 0; i < 16; i++) {
2030 uint16_t *tmp = (uint16_t*)(dest[p] + i*linesize);
2031 for (j = 0; j < 16; j++)
2032 tmp[j] = get_bits(&gb, bit_depth);
2036 for (p = 0; p < plane_count; p++) {
2037 for (i = 0; i < 16; i++) {
2038 memcpy(dest[p] + i*linesize, h->mb + p*128 + i*8, 16);
2043 if(IS_INTRA(mb_type)){
2044 if(h->deblocking_filter)
2045 xchg_mb_border(h, dest[0], dest[1], dest[2], linesize, linesize, 1, 1, simple, pixel_shift);
2047 for (p = 0; p < plane_count; p++)
2048 hl_decode_mb_predict_luma(h, mb_type, 1, simple, transform_bypass, pixel_shift, block_offset, linesize, dest[p], p);
2050 if(h->deblocking_filter)
2051 xchg_mb_border(h, dest[0], dest[1], dest[2], linesize, linesize, 0, 1, simple, pixel_shift);
2053 hl_motion(h, dest[0], dest[1], dest[2],
2054 s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
2055 s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
2056 h->h264dsp.weight_h264_pixels_tab,
2057 h->h264dsp.biweight_h264_pixels_tab, pixel_shift, 1);
2060 for (p = 0; p < plane_count; p++)
2061 hl_decode_mb_idct_luma(h, mb_type, 1, simple, transform_bypass, pixel_shift, block_offset, linesize, dest[p], p);
2063 if(h->cbp || IS_INTRA(mb_type))
2065 s->dsp.clear_blocks(h->mb);
2066 s->dsp.clear_blocks(h->mb+(24*16<<pixel_shift));
2071 * Process a macroblock; this case avoids checks for expensive uncommon cases.
2073 #define hl_decode_mb_simple(sh, bits) \
2074 static void hl_decode_mb_simple_ ## bits(H264Context *h){ \
2075 hl_decode_mb_internal(h, 1, sh); \
2077 hl_decode_mb_simple(0, 8);
2078 hl_decode_mb_simple(1, 16);
2081 * Process a macroblock; this handles edge cases, such as interlacing.
2083 static void av_noinline hl_decode_mb_complex(H264Context *h){
2084 hl_decode_mb_internal(h, 0, h->pixel_shift);
2087 static void av_noinline hl_decode_mb_444_complex(H264Context *h){
2088 hl_decode_mb_444_internal(h, 0, h->pixel_shift);
2091 static void av_noinline hl_decode_mb_444_simple(H264Context *h){
2092 hl_decode_mb_444_internal(h, 1, 0);
2095 void ff_h264_hl_decode_mb(H264Context *h){
2096 MpegEncContext * const s = &h->s;
2097 const int mb_xy= h->mb_xy;
2098 const int mb_type = s->current_picture.f.mb_type[mb_xy];
2099 int is_complex = CONFIG_SMALL || h->is_complex || IS_INTRA_PCM(mb_type) || s->qscale == 0;
2102 if(is_complex || h->pixel_shift)
2103 hl_decode_mb_444_complex(h);
2105 hl_decode_mb_444_simple(h);
2106 } else if (is_complex) {
2107 hl_decode_mb_complex(h);
2108 } else if (h->pixel_shift) {
2109 hl_decode_mb_simple_16(h);
2111 hl_decode_mb_simple_8(h);
2114 static int pred_weight_table(H264Context *h){
2115 MpegEncContext * const s = &h->s;
2117 int luma_def, chroma_def;
2120 h->use_weight_chroma= 0;
2121 h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
2122 if(h->sps.chroma_format_idc)
2123 h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
2124 luma_def = 1<<h->luma_log2_weight_denom;
2125 chroma_def = 1<<h->chroma_log2_weight_denom;
2127 for(list=0; list<2; list++){
2128 h->luma_weight_flag[list] = 0;
2129 h->chroma_weight_flag[list] = 0;
2130 for(i=0; i<h->ref_count[list]; i++){
2131 int luma_weight_flag, chroma_weight_flag;
2133 luma_weight_flag= get_bits1(&s->gb);
2134 if(luma_weight_flag){
2135 h->luma_weight[i][list][0]= get_se_golomb(&s->gb);
2136 h->luma_weight[i][list][1]= get_se_golomb(&s->gb);
2137 if( h->luma_weight[i][list][0] != luma_def
2138 || h->luma_weight[i][list][1] != 0) {
2140 h->luma_weight_flag[list]= 1;
2143 h->luma_weight[i][list][0]= luma_def;
2144 h->luma_weight[i][list][1]= 0;
2147 if(h->sps.chroma_format_idc){
2148 chroma_weight_flag= get_bits1(&s->gb);
2149 if(chroma_weight_flag){
2152 h->chroma_weight[i][list][j][0]= get_se_golomb(&s->gb);
2153 h->chroma_weight[i][list][j][1]= get_se_golomb(&s->gb);
2154 if( h->chroma_weight[i][list][j][0] != chroma_def
2155 || h->chroma_weight[i][list][j][1] != 0) {
2156 h->use_weight_chroma= 1;
2157 h->chroma_weight_flag[list]= 1;
2163 h->chroma_weight[i][list][j][0]= chroma_def;
2164 h->chroma_weight[i][list][j][1]= 0;
2169 if(h->slice_type_nos != AV_PICTURE_TYPE_B) break;
2171 h->use_weight= h->use_weight || h->use_weight_chroma;
2176 * Initialize implicit_weight table.
2177 * @param field 0/1 initialize the weight for interlaced MBAFF
2178 * -1 initializes the rest
2180 static void implicit_weight_table(H264Context *h, int field){
2181 MpegEncContext * const s = &h->s;
2182 int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
2184 for (i = 0; i < 2; i++) {
2185 h->luma_weight_flag[i] = 0;
2186 h->chroma_weight_flag[i] = 0;
2190 cur_poc = s->current_picture_ptr->poc;
2191 if( h->ref_count[0] == 1 && h->ref_count[1] == 1 && !FRAME_MBAFF
2192 && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
2194 h->use_weight_chroma= 0;
2198 ref_count0= h->ref_count[0];
2199 ref_count1= h->ref_count[1];
2201 cur_poc = s->current_picture_ptr->field_poc[field];
2203 ref_count0= 16+2*h->ref_count[0];
2204 ref_count1= 16+2*h->ref_count[1];
2208 h->use_weight_chroma= 2;
2209 h->luma_log2_weight_denom= 5;
2210 h->chroma_log2_weight_denom= 5;
2212 for(ref0=ref_start; ref0 < ref_count0; ref0++){
2213 int poc0 = h->ref_list[0][ref0].poc;
2214 for(ref1=ref_start; ref1 < ref_count1; ref1++){
2216 if (!h->ref_list[0][ref0].long_ref && !h->ref_list[1][ref1].long_ref) {
2217 int poc1 = h->ref_list[1][ref1].poc;
2218 int td = av_clip(poc1 - poc0, -128, 127);
2220 int tb = av_clip(cur_poc - poc0, -128, 127);
2221 int tx = (16384 + (FFABS(td) >> 1)) / td;
2222 int dist_scale_factor = (tb*tx + 32) >> 8;
2223 if(dist_scale_factor >= -64 && dist_scale_factor <= 128)
2224 w = 64 - dist_scale_factor;
2228 h->implicit_weight[ref0][ref1][0]=
2229 h->implicit_weight[ref0][ref1][1]= w;
2231 h->implicit_weight[ref0][ref1][field]=w;
2238 * instantaneous decoder refresh.
2240 static void idr(H264Context *h){
2241 ff_h264_remove_all_refs(h);
2242 h->prev_frame_num= 0;
2243 h->prev_frame_num_offset= 0;
2248 /* forget old pics after a seek */
2249 static void flush_dpb(AVCodecContext *avctx){
2250 H264Context *h= avctx->priv_data;
2252 for(i=0; i<=MAX_DELAYED_PIC_COUNT; i++) {
2253 if(h->delayed_pic[i])
2254 h->delayed_pic[i]->f.reference = 0;
2255 h->delayed_pic[i]= NULL;
2257 h->outputed_poc=h->next_outputed_poc= INT_MIN;
2258 h->prev_interlaced_frame = 1;
2260 if(h->s.current_picture_ptr)
2261 h->s.current_picture_ptr->f.reference = 0;
2262 h->s.first_field= 0;
2263 ff_h264_reset_sei(h);
2264 ff_mpeg_flush(avctx);
2268 static int init_poc(H264Context *h){
2269 MpegEncContext * const s = &h->s;
2270 const int max_frame_num= 1<<h->sps.log2_max_frame_num;
2272 Picture *cur = s->current_picture_ptr;
2274 h->frame_num_offset= h->prev_frame_num_offset;
2275 if(h->frame_num < h->prev_frame_num)
2276 h->frame_num_offset += max_frame_num;
2278 if(h->sps.poc_type==0){
2279 const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
2281 if (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
2282 h->poc_msb = h->prev_poc_msb + max_poc_lsb;
2283 else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
2284 h->poc_msb = h->prev_poc_msb - max_poc_lsb;
2286 h->poc_msb = h->prev_poc_msb;
2287 //printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
2289 field_poc[1] = h->poc_msb + h->poc_lsb;
2290 if(s->picture_structure == PICT_FRAME)
2291 field_poc[1] += h->delta_poc_bottom;
2292 }else if(h->sps.poc_type==1){
2293 int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
2296 if(h->sps.poc_cycle_length != 0)
2297 abs_frame_num = h->frame_num_offset + h->frame_num;
2301 if(h->nal_ref_idc==0 && abs_frame_num > 0)
2304 expected_delta_per_poc_cycle = 0;
2305 for(i=0; i < h->sps.poc_cycle_length; i++)
2306 expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
2308 if(abs_frame_num > 0){
2309 int poc_cycle_cnt = (abs_frame_num - 1) / h->sps.poc_cycle_length;
2310 int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
2312 expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
2313 for(i = 0; i <= frame_num_in_poc_cycle; i++)
2314 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
2318 if(h->nal_ref_idc == 0)
2319 expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
2321 field_poc[0] = expectedpoc + h->delta_poc[0];
2322 field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
2324 if(s->picture_structure == PICT_FRAME)
2325 field_poc[1] += h->delta_poc[1];
2327 int poc= 2*(h->frame_num_offset + h->frame_num);
2336 if(s->picture_structure != PICT_BOTTOM_FIELD)
2337 s->current_picture_ptr->field_poc[0]= field_poc[0];
2338 if(s->picture_structure != PICT_TOP_FIELD)
2339 s->current_picture_ptr->field_poc[1]= field_poc[1];
2340 cur->poc= FFMIN(cur->field_poc[0], cur->field_poc[1]);
2347 * initialize scan tables
2349 static void init_scan_tables(H264Context *h){
2351 for(i=0; i<16; i++){
2352 #define T(x) (x>>2) | ((x<<2) & 0xF)
2353 h->zigzag_scan[i] = T(zigzag_scan[i]);
2354 h-> field_scan[i] = T( field_scan[i]);
2357 for(i=0; i<64; i++){
2358 #define T(x) (x>>3) | ((x&7)<<3)
2359 h->zigzag_scan8x8[i] = T(ff_zigzag_direct[i]);
2360 h->zigzag_scan8x8_cavlc[i] = T(zigzag_scan8x8_cavlc[i]);
2361 h->field_scan8x8[i] = T(field_scan8x8[i]);
2362 h->field_scan8x8_cavlc[i] = T(field_scan8x8_cavlc[i]);
2365 if(h->sps.transform_bypass){ //FIXME same ugly
2366 h->zigzag_scan_q0 = zigzag_scan;
2367 h->zigzag_scan8x8_q0 = ff_zigzag_direct;
2368 h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc;
2369 h->field_scan_q0 = field_scan;
2370 h->field_scan8x8_q0 = field_scan8x8;
2371 h->field_scan8x8_cavlc_q0 = field_scan8x8_cavlc;
2373 h->zigzag_scan_q0 = h->zigzag_scan;
2374 h->zigzag_scan8x8_q0 = h->zigzag_scan8x8;
2375 h->zigzag_scan8x8_cavlc_q0 = h->zigzag_scan8x8_cavlc;
2376 h->field_scan_q0 = h->field_scan;
2377 h->field_scan8x8_q0 = h->field_scan8x8;
2378 h->field_scan8x8_cavlc_q0 = h->field_scan8x8_cavlc;
2382 static int field_end(H264Context *h, int in_setup){
2383 MpegEncContext * const s = &h->s;
2384 AVCodecContext * const avctx= s->avctx;
2388 if (!in_setup && !s->dropable)
2389 ff_thread_report_progress((AVFrame*)s->current_picture_ptr, (16*s->mb_height >> FIELD_PICTURE) - 1,
2390 s->picture_structure==PICT_BOTTOM_FIELD);
2392 if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
2393 ff_vdpau_h264_set_reference_frames(s);
2395 if(in_setup || !(avctx->active_thread_type&FF_THREAD_FRAME)){
2397 err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
2398 h->prev_poc_msb= h->poc_msb;
2399 h->prev_poc_lsb= h->poc_lsb;
2401 h->prev_frame_num_offset= h->frame_num_offset;
2402 h->prev_frame_num= h->frame_num;
2403 h->outputed_poc = h->next_outputed_poc;
2406 if (avctx->hwaccel) {
2407 if (avctx->hwaccel->end_frame(avctx) < 0)
2408 av_log(avctx, AV_LOG_ERROR, "hardware accelerator failed to decode picture\n");
2411 if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
2412 ff_vdpau_h264_picture_complete(s);
2415 * FIXME: Error handling code does not seem to support interlaced
2416 * when slices span multiple rows
2417 * The ff_er_add_slice calls don't work right for bottom
2418 * fields; they cause massive erroneous error concealing
2419 * Error marking covers both fields (top and bottom).
2420 * This causes a mismatched s->error_count
2421 * and a bad error table. Further, the error count goes to
2422 * INT_MAX when called for bottom field, because mb_y is
2423 * past end by one (callers fault) and resync_mb_y != 0
2424 * causes problems for the first MB line, too.
2437 * Replicate H264 "master" context to thread contexts.
2439 static void clone_slice(H264Context *dst, H264Context *src)
2441 memcpy(dst->block_offset, src->block_offset, sizeof(dst->block_offset));
2442 dst->s.current_picture_ptr = src->s.current_picture_ptr;
2443 dst->s.current_picture = src->s.current_picture;
2444 dst->s.linesize = src->s.linesize;
2445 dst->s.uvlinesize = src->s.uvlinesize;
2446 dst->s.first_field = src->s.first_field;
2448 dst->prev_poc_msb = src->prev_poc_msb;
2449 dst->prev_poc_lsb = src->prev_poc_lsb;
2450 dst->prev_frame_num_offset = src->prev_frame_num_offset;
2451 dst->prev_frame_num = src->prev_frame_num;
2452 dst->short_ref_count = src->short_ref_count;
2454 memcpy(dst->short_ref, src->short_ref, sizeof(dst->short_ref));
2455 memcpy(dst->long_ref, src->long_ref, sizeof(dst->long_ref));
2456 memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
2457 memcpy(dst->ref_list, src->ref_list, sizeof(dst->ref_list));
2459 memcpy(dst->dequant4_coeff, src->dequant4_coeff, sizeof(src->dequant4_coeff));
2460 memcpy(dst->dequant8_coeff, src->dequant8_coeff, sizeof(src->dequant8_coeff));
2464 * computes profile from profile_idc and constraint_set?_flags
2468 * @return profile as defined by FF_PROFILE_H264_*
2470 int ff_h264_get_profile(SPS *sps)
2472 int profile = sps->profile_idc;
2474 switch(sps->profile_idc) {
2475 case FF_PROFILE_H264_BASELINE:
2476 // constraint_set1_flag set to 1
2477 profile |= (sps->constraint_set_flags & 1<<1) ? FF_PROFILE_H264_CONSTRAINED : 0;
2479 case FF_PROFILE_H264_HIGH_10:
2480 case FF_PROFILE_H264_HIGH_422:
2481 case FF_PROFILE_H264_HIGH_444_PREDICTIVE:
2482 // constraint_set3_flag set to 1
2483 profile |= (sps->constraint_set_flags & 1<<3) ? FF_PROFILE_H264_INTRA : 0;
2491 * decodes a slice header.
2492 * This will also call MPV_common_init() and frame_start() as needed.
2494 * @param h h264context
2495 * @param h0 h264 master context (differs from 'h' when doing sliced based parallel decoding)
2497 * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
2499 static int decode_slice_header(H264Context *h, H264Context *h0){
2500 MpegEncContext * const s = &h->s;
2501 MpegEncContext * const s0 = &h0->s;
2502 unsigned int first_mb_in_slice;
2503 unsigned int pps_id;
2504 int num_ref_idx_active_override_flag;
2505 unsigned int slice_type, tmp, i, j;
2506 int default_ref_list_done = 0;
2507 int last_pic_structure;
2509 s->dropable= h->nal_ref_idc == 0;
2511 /* FIXME: 2tap qpel isn't implemented for high bit depth. */
2512 if((s->avctx->flags2 & CODEC_FLAG2_FAST) && !h->nal_ref_idc && !h->pixel_shift){
2513 s->me.qpel_put= s->dsp.put_2tap_qpel_pixels_tab;
2514 s->me.qpel_avg= s->dsp.avg_2tap_qpel_pixels_tab;
2516 s->me.qpel_put= s->dsp.put_h264_qpel_pixels_tab;
2517 s->me.qpel_avg= s->dsp.avg_h264_qpel_pixels_tab;
2520 first_mb_in_slice= get_ue_golomb(&s->gb);
2522 if(first_mb_in_slice == 0){ //FIXME better field boundary detection
2523 if(h0->current_slice && FIELD_PICTURE){
2527 h0->current_slice = 0;
2528 if (!s0->first_field)
2529 s->current_picture_ptr= NULL;
2532 slice_type= get_ue_golomb_31(&s->gb);
2534 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);
2539 h->slice_type_fixed=1;
2541 h->slice_type_fixed=0;
2543 slice_type= golomb_to_pict_type[ slice_type ];
2544 if (slice_type == AV_PICTURE_TYPE_I
2545 || (h0->current_slice != 0 && slice_type == h0->last_slice_type) ) {
2546 default_ref_list_done = 1;
2548 h->slice_type= slice_type;
2549 h->slice_type_nos= slice_type & 3;
2551 s->pict_type= h->slice_type; // to make a few old functions happy, it's wrong though
2553 pps_id= get_ue_golomb(&s->gb);
2554 if(pps_id>=MAX_PPS_COUNT){
2555 av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
2558 if(!h0->pps_buffers[pps_id]) {
2559 av_log(h->s.avctx, AV_LOG_ERROR, "non-existing PPS %u referenced\n", pps_id);
2562 h->pps= *h0->pps_buffers[pps_id];
2564 if(!h0->sps_buffers[h->pps.sps_id]) {
2565 av_log(h->s.avctx, AV_LOG_ERROR, "non-existing SPS %u referenced\n", h->pps.sps_id);
2568 h->sps = *h0->sps_buffers[h->pps.sps_id];
2570 s->avctx->profile = ff_h264_get_profile(&h->sps);
2571 s->avctx->level = h->sps.level_idc;
2572 s->avctx->refs = h->sps.ref_frame_count;
2574 if(h == h0 && h->dequant_coeff_pps != pps_id){
2575 h->dequant_coeff_pps = pps_id;
2576 init_dequant_tables(h);
2579 s->mb_width= h->sps.mb_width;
2580 s->mb_height= h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
2582 h->b_stride= s->mb_width*4;
2584 s->chroma_y_shift = h->sps.chroma_format_idc <= 1; // 400 uses yuv420p
2586 s->width = 16*s->mb_width - (2>>CHROMA444)*FFMIN(h->sps.crop_right, (8<<CHROMA444)-1);
2587 if(h->sps.frame_mbs_only_flag)
2588 s->height= 16*s->mb_height - (1<<s->chroma_y_shift)*FFMIN(h->sps.crop_bottom, (16>>s->chroma_y_shift)-1);
2590 s->height= 16*s->mb_height - (2<<s->chroma_y_shift)*FFMIN(h->sps.crop_bottom, (16>>s->chroma_y_shift)-1);
2592 if (s->context_initialized
2593 && ( s->width != s->avctx->width || s->height != s->avctx->height
2594 || av_cmp_q(h->sps.sar, s->avctx->sample_aspect_ratio))) {
2596 av_log_missing_feature(s->avctx, "Width/height changing with threads is", 0);
2597 return -1; // width / height changed during parallelized decoding
2600 flush_dpb(s->avctx);
2603 if (!s->context_initialized) {
2605 av_log(h->s.avctx, AV_LOG_ERROR, "Cannot (re-)initialize context during parallel decoding.\n");
2609 avcodec_set_dimensions(s->avctx, s->width, s->height);
2610 s->avctx->sample_aspect_ratio= h->sps.sar;
2611 av_assert0(s->avctx->sample_aspect_ratio.den);
2613 if(h->sps.video_signal_type_present_flag){
2614 s->avctx->color_range = h->sps.full_range ? AVCOL_RANGE_JPEG : AVCOL_RANGE_MPEG;
2615 if(h->sps.colour_description_present_flag){
2616 s->avctx->color_primaries = h->sps.color_primaries;
2617 s->avctx->color_trc = h->sps.color_trc;
2618 s->avctx->colorspace = h->sps.colorspace;
2622 if(h->sps.timing_info_present_flag){
2623 int64_t den= h->sps.time_scale;
2624 if(h->x264_build < 44U)
2626 av_reduce(&s->avctx->time_base.num, &s->avctx->time_base.den,
2627 h->sps.num_units_in_tick, den, 1<<30);
2630 switch (h->sps.bit_depth_luma) {
2633 s->avctx->pix_fmt = PIX_FMT_YUV444P9;
2635 s->avctx->pix_fmt = PIX_FMT_YUV420P9;
2639 s->avctx->pix_fmt = PIX_FMT_YUV444P10;
2641 s->avctx->pix_fmt = PIX_FMT_YUV422P10;
2643 s->avctx->pix_fmt = PIX_FMT_YUV420P10;
2647 s->avctx->pix_fmt = s->avctx->color_range == AVCOL_RANGE_JPEG ? PIX_FMT_YUVJ444P : PIX_FMT_YUV444P;
2648 }else if (CHROMA422) {
2649 s->avctx->pix_fmt = s->avctx->color_range == AVCOL_RANGE_JPEG ? PIX_FMT_YUVJ422P : PIX_FMT_YUV422P;
2651 s->avctx->pix_fmt = s->avctx->get_format(s->avctx,
2652 s->avctx->codec->pix_fmts ?
2653 s->avctx->codec->pix_fmts :
2654 s->avctx->color_range == AVCOL_RANGE_JPEG ?
2655 hwaccel_pixfmt_list_h264_jpeg_420 :
2656 ff_hwaccel_pixfmt_list_420);
2660 s->avctx->hwaccel = ff_find_hwaccel(s->avctx->codec->id, s->avctx->pix_fmt);
2662 if (MPV_common_init(s) < 0) {
2663 av_log(h->s.avctx, AV_LOG_ERROR, "MPV_common_init() failed.\n");
2667 h->prev_interlaced_frame = 1;
2669 init_scan_tables(h);
2670 if (ff_h264_alloc_tables(h) < 0) {
2671 av_log(h->s.avctx, AV_LOG_ERROR, "Could not allocate memory for h264\n");
2672 return AVERROR(ENOMEM);
2675 if (!HAVE_THREADS || !(s->avctx->active_thread_type&FF_THREAD_SLICE)) {
2676 if (context_init(h) < 0) {
2677 av_log(h->s.avctx, AV_LOG_ERROR, "context_init() failed.\n");
2681 for(i = 1; i < s->avctx->thread_count; i++) {
2683 c = h->thread_context[i] = av_malloc(sizeof(H264Context));
2684 memcpy(c, h->s.thread_context[i], sizeof(MpegEncContext));
2685 memset(&c->s + 1, 0, sizeof(H264Context) - sizeof(MpegEncContext));
2686 c->h264dsp = h->h264dsp;
2689 c->pixel_shift = h->pixel_shift;
2690 init_scan_tables(c);
2691 clone_tables(c, h, i);
2694 for(i = 0; i < s->avctx->thread_count; i++)
2695 if (context_init(h->thread_context[i]) < 0) {
2696 av_log(h->s.avctx, AV_LOG_ERROR, "context_init() failed.\n");
2702 h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
2705 h->mb_aff_frame = 0;
2706 last_pic_structure = s0->picture_structure;
2707 if(h->sps.frame_mbs_only_flag){
2708 s->picture_structure= PICT_FRAME;
2710 if(get_bits1(&s->gb)) { //field_pic_flag
2711 s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
2713 s->picture_structure= PICT_FRAME;
2714 h->mb_aff_frame = h->sps.mb_aff;
2717 h->mb_field_decoding_flag= s->picture_structure != PICT_FRAME;
2719 if(h0->current_slice == 0){
2720 // Shorten frame num gaps so we don't have to allocate reference frames just to throw them away
2721 if(h->frame_num != h->prev_frame_num) {
2722 int unwrap_prev_frame_num = h->prev_frame_num, max_frame_num = 1<<h->sps.log2_max_frame_num;
2724 if (unwrap_prev_frame_num > h->frame_num) unwrap_prev_frame_num -= max_frame_num;
2726 if ((h->frame_num - unwrap_prev_frame_num) > h->sps.ref_frame_count) {
2727 unwrap_prev_frame_num = (h->frame_num - h->sps.ref_frame_count) - 1;
2728 if (unwrap_prev_frame_num < 0)
2729 unwrap_prev_frame_num += max_frame_num;
2731 h->prev_frame_num = unwrap_prev_frame_num;
2735 while(h->frame_num != h->prev_frame_num &&
2736 h->frame_num != (h->prev_frame_num+1)%(1<<h->sps.log2_max_frame_num)){
2737 Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
2738 av_log(h->s.avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n", h->frame_num, h->prev_frame_num);
2739 if (ff_h264_frame_start(h) < 0)
2741 h->prev_frame_num++;
2742 h->prev_frame_num %= 1<<h->sps.log2_max_frame_num;
2743 s->current_picture_ptr->frame_num= h->prev_frame_num;
2744 ff_thread_report_progress((AVFrame*)s->current_picture_ptr, INT_MAX, 0);
2745 ff_thread_report_progress((AVFrame*)s->current_picture_ptr, INT_MAX, 1);
2746 ff_generate_sliding_window_mmcos(h);
2747 if (ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index) < 0 &&
2748 s->avctx->error_recognition >= FF_ER_EXPLODE)
2749 return AVERROR_INVALIDDATA;
2750 /* Error concealment: if a ref is missing, copy the previous ref in its place.
2751 * FIXME: avoiding a memcpy would be nice, but ref handling makes many assumptions
2752 * about there being no actual duplicates.
2753 * FIXME: this doesn't copy padding for out-of-frame motion vectors. Given we're
2754 * concealing a lost frame, this probably isn't noticable by comparison, but it should
2756 if (h->short_ref_count) {
2758 av_image_copy(h->short_ref[0]->f.data, h->short_ref[0]->f.linesize,
2759 (const uint8_t**)prev->f.data, prev->f.linesize,
2760 s->avctx->pix_fmt, s->mb_width*16, s->mb_height*16);
2761 h->short_ref[0]->poc = prev->poc+2;
2763 h->short_ref[0]->frame_num = h->prev_frame_num;
2767 /* See if we have a decoded first field looking for a pair... */
2768 if (s0->first_field) {
2769 assert(s0->current_picture_ptr);
2770 assert(s0->current_picture_ptr->f.data[0]);
2771 assert(s0->current_picture_ptr->reference != DELAYED_PIC_REF);
2773 /* figure out if we have a complementary field pair */
2774 if (!FIELD_PICTURE || s->picture_structure == last_pic_structure) {
2776 * Previous field is unmatched. Don't display it, but let it
2777 * remain for reference if marked as such.
2779 s0->current_picture_ptr = NULL;
2780 s0->first_field = FIELD_PICTURE;
2783 if (h->nal_ref_idc &&
2784 s0->current_picture_ptr->f.reference &&
2785 s0->current_picture_ptr->frame_num != h->frame_num) {
2787 * This and previous field were reference, but had
2788 * different frame_nums. Consider this field first in
2789 * pair. Throw away previous field except for reference
2792 s0->first_field = 1;
2793 s0->current_picture_ptr = NULL;
2796 /* Second field in complementary pair */
2797 s0->first_field = 0;
2802 /* Frame or first field in a potentially complementary pair */
2803 assert(!s0->current_picture_ptr);
2804 s0->first_field = FIELD_PICTURE;
2807 if(!FIELD_PICTURE || s0->first_field) {
2808 if (ff_h264_frame_start(h) < 0) {
2809 s0->first_field = 0;
2813 ff_release_unused_pictures(s, 0);
2819 s->current_picture_ptr->frame_num= h->frame_num; //FIXME frame_num cleanup
2821 assert(s->mb_num == s->mb_width * s->mb_height);
2822 if(first_mb_in_slice << FIELD_OR_MBAFF_PICTURE >= s->mb_num ||
2823 first_mb_in_slice >= s->mb_num){
2824 av_log(h->s.avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
2827 s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
2828 s->resync_mb_y = s->mb_y = (first_mb_in_slice / s->mb_width) << FIELD_OR_MBAFF_PICTURE;
2829 if (s->picture_structure == PICT_BOTTOM_FIELD)
2830 s->resync_mb_y = s->mb_y = s->mb_y + 1;
2831 assert(s->mb_y < s->mb_height);
2833 if(s->picture_structure==PICT_FRAME){
2834 h->curr_pic_num= h->frame_num;
2835 h->max_pic_num= 1<< h->sps.log2_max_frame_num;
2837 h->curr_pic_num= 2*h->frame_num + 1;
2838 h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
2841 if(h->nal_unit_type == NAL_IDR_SLICE){
2842 get_ue_golomb(&s->gb); /* idr_pic_id */
2845 if(h->sps.poc_type==0){
2846 h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
2848 if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
2849 h->delta_poc_bottom= get_se_golomb(&s->gb);
2853 if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
2854 h->delta_poc[0]= get_se_golomb(&s->gb);
2856 if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
2857 h->delta_poc[1]= get_se_golomb(&s->gb);
2862 if(h->pps.redundant_pic_cnt_present){
2863 h->redundant_pic_count= get_ue_golomb(&s->gb);
2866 //set defaults, might be overridden a few lines later
2867 h->ref_count[0]= h->pps.ref_count[0];
2868 h->ref_count[1]= h->pps.ref_count[1];
2870 if(h->slice_type_nos != AV_PICTURE_TYPE_I){
2871 if(h->slice_type_nos == AV_PICTURE_TYPE_B){
2872 h->direct_spatial_mv_pred= get_bits1(&s->gb);
2874 num_ref_idx_active_override_flag= get_bits1(&s->gb);
2876 if(num_ref_idx_active_override_flag){
2877 h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
2878 if(h->slice_type_nos==AV_PICTURE_TYPE_B)
2879 h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
2881 if(h->ref_count[0]-1 > 32-1 || h->ref_count[1]-1 > 32-1){
2882 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
2883 h->ref_count[0]= h->ref_count[1]= 1;
2887 if(h->slice_type_nos == AV_PICTURE_TYPE_B)
2894 if(!default_ref_list_done){
2895 ff_h264_fill_default_ref_list(h);
2898 if(h->slice_type_nos!=AV_PICTURE_TYPE_I && ff_h264_decode_ref_pic_list_reordering(h) < 0)
2901 if(h->slice_type_nos!=AV_PICTURE_TYPE_I){
2902 s->last_picture_ptr= &h->ref_list[0][0];
2903 ff_copy_picture(&s->last_picture, s->last_picture_ptr);
2905 if(h->slice_type_nos==AV_PICTURE_TYPE_B){
2906 s->next_picture_ptr= &h->ref_list[1][0];
2907 ff_copy_picture(&s->next_picture, s->next_picture_ptr);
2910 if( (h->pps.weighted_pred && h->slice_type_nos == AV_PICTURE_TYPE_P )
2911 || (h->pps.weighted_bipred_idc==1 && h->slice_type_nos== AV_PICTURE_TYPE_B ) )
2912 pred_weight_table(h);
2913 else if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== AV_PICTURE_TYPE_B){
2914 implicit_weight_table(h, -1);
2917 for (i = 0; i < 2; i++) {
2918 h->luma_weight_flag[i] = 0;
2919 h->chroma_weight_flag[i] = 0;
2923 if(h->nal_ref_idc && ff_h264_decode_ref_pic_marking(h0, &s->gb) < 0 &&
2924 s->avctx->error_recognition >= FF_ER_EXPLODE)
2925 return AVERROR_INVALIDDATA;
2928 ff_h264_fill_mbaff_ref_list(h);
2930 if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== AV_PICTURE_TYPE_B){
2931 implicit_weight_table(h, 0);
2932 implicit_weight_table(h, 1);
2936 if(h->slice_type_nos==AV_PICTURE_TYPE_B && !h->direct_spatial_mv_pred)
2937 ff_h264_direct_dist_scale_factor(h);
2938 ff_h264_direct_ref_list_init(h);
2940 if( h->slice_type_nos != AV_PICTURE_TYPE_I && h->pps.cabac ){
2941 tmp = get_ue_golomb_31(&s->gb);
2943 av_log(s->avctx, AV_LOG_ERROR, "cabac_init_idc overflow\n");
2946 h->cabac_init_idc= tmp;
2949 h->last_qscale_diff = 0;
2950 tmp = h->pps.init_qp + get_se_golomb(&s->gb);
2951 if(tmp>51+6*(h->sps.bit_depth_luma-8)){
2952 av_log(s->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
2956 h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
2957 h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
2958 //FIXME qscale / qp ... stuff
2959 if(h->slice_type == AV_PICTURE_TYPE_SP){
2960 get_bits1(&s->gb); /* sp_for_switch_flag */
2962 if(h->slice_type==AV_PICTURE_TYPE_SP || h->slice_type == AV_PICTURE_TYPE_SI){
2963 get_se_golomb(&s->gb); /* slice_qs_delta */
2966 h->deblocking_filter = 1;
2967 h->slice_alpha_c0_offset = 52;
2968 h->slice_beta_offset = 52;
2969 if( h->pps.deblocking_filter_parameters_present ) {
2970 tmp= get_ue_golomb_31(&s->gb);
2972 av_log(s->avctx, AV_LOG_ERROR, "deblocking_filter_idc %u out of range\n", tmp);
2975 h->deblocking_filter= tmp;
2976 if(h->deblocking_filter < 2)
2977 h->deblocking_filter^= 1; // 1<->0
2979 if( h->deblocking_filter ) {
2980 h->slice_alpha_c0_offset += get_se_golomb(&s->gb) << 1;
2981 h->slice_beta_offset += get_se_golomb(&s->gb) << 1;
2982 if( h->slice_alpha_c0_offset > 104U
2983 || h->slice_beta_offset > 104U){
2984 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);
2990 if( s->avctx->skip_loop_filter >= AVDISCARD_ALL
2991 ||(s->avctx->skip_loop_filter >= AVDISCARD_NONKEY && h->slice_type_nos != AV_PICTURE_TYPE_I)
2992 ||(s->avctx->skip_loop_filter >= AVDISCARD_BIDIR && h->slice_type_nos == AV_PICTURE_TYPE_B)
2993 ||(s->avctx->skip_loop_filter >= AVDISCARD_NONREF && h->nal_ref_idc == 0))
2994 h->deblocking_filter= 0;
2996 if(h->deblocking_filter == 1 && h0->max_contexts > 1) {
2997 if(s->avctx->flags2 & CODEC_FLAG2_FAST) {
2998 /* Cheat slightly for speed:
2999 Do not bother to deblock across slices. */
3000 h->deblocking_filter = 2;
3002 h0->max_contexts = 1;
3003 if(!h0->single_decode_warning) {
3004 av_log(s->avctx, AV_LOG_INFO, "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
3005 h0->single_decode_warning = 1;
3008 av_log(h->s.avctx, AV_LOG_ERROR, "Deblocking switched inside frame.\n");
3013 h->qp_thresh = 15 + 52 - FFMIN(h->slice_alpha_c0_offset, h->slice_beta_offset)
3014 - FFMAX3(0, h->pps.chroma_qp_index_offset[0], h->pps.chroma_qp_index_offset[1])
3015 + 6 * (h->sps.bit_depth_luma - 8);
3018 if( h->pps.num_slice_groups > 1 && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
3019 slice_group_change_cycle= get_bits(&s->gb, ?);
3022 h0->last_slice_type = slice_type;
3023 h->slice_num = ++h0->current_slice;
3026 h0->slice_row[(h->slice_num-1)&(MAX_SLICES-1)]= s->resync_mb_y;
3027 if ( h0->slice_row[h->slice_num&(MAX_SLICES-1)] + 3 >= s->resync_mb_y
3028 && h0->slice_row[h->slice_num&(MAX_SLICES-1)] <= s->resync_mb_y
3029 && h->slice_num >= MAX_SLICES) {
3030 //in case of ASO this check needs to be updated depending on how we decide to assign slice numbers in this case
3031 av_log(s->avctx, AV_LOG_WARNING, "Possibly too many slices (%d >= %d), increase MAX_SLICES and recompile if there are artifacts\n", h->slice_num, MAX_SLICES);
3036 int *ref2frm= h->ref2frm[h->slice_num&(MAX_SLICES-1)][j];
3037 for(i=0; i<16; i++){
3039 if (h->ref_list[j][i].f.data[0]) {
3041 uint8_t *base = h->ref_list[j][i].f.base[0];
3042 for(k=0; k<h->short_ref_count; k++)
3043 if (h->short_ref[k]->f.base[0] == base) {
3047 for(k=0; k<h->long_ref_count; k++)
3048 if (h->long_ref[k] && h->long_ref[k]->f.base[0] == base) {
3049 id_list[i]= h->short_ref_count + k;
3058 ref2frm[i+2]= 4*id_list[i]
3059 + (h->ref_list[j][i].f.reference & 3);
3062 for(i=16; i<48; i++)
3063 ref2frm[i+4]= 4*id_list[(i-16)>>1]
3064 + (h->ref_list[j][i].f.reference & 3);
3067 //FIXME: fix draw_edges+PAFF+frame threads
3068 h->emu_edge_width= (s->flags&CODEC_FLAG_EMU_EDGE || (!h->sps.frame_mbs_only_flag && s->avctx->active_thread_type)) ? 0 : 16;
3069 h->emu_edge_height= (FRAME_MBAFF || FIELD_PICTURE) ? 0 : h->emu_edge_width;
3071 if(s->avctx->debug&FF_DEBUG_PICT_INFO){
3072 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",
3074 (s->picture_structure==PICT_FRAME ? "F" : s->picture_structure==PICT_TOP_FIELD ? "T" : "B"),
3076 av_get_picture_type_char(h->slice_type), h->slice_type_fixed ? " fix" : "", h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
3077 pps_id, h->frame_num,
3078 s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
3079 h->ref_count[0], h->ref_count[1],
3081 h->deblocking_filter, h->slice_alpha_c0_offset/2-26, h->slice_beta_offset/2-26,
3083 h->use_weight==1 && h->use_weight_chroma ? "c" : "",
3084 h->slice_type == AV_PICTURE_TYPE_B ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : ""
3091 int ff_h264_get_slice_type(const H264Context *h)
3093 switch (h->slice_type) {
3094 case AV_PICTURE_TYPE_P: return 0;
3095 case AV_PICTURE_TYPE_B: return 1;
3096 case AV_PICTURE_TYPE_I: return 2;
3097 case AV_PICTURE_TYPE_SP: return 3;
3098 case AV_PICTURE_TYPE_SI: return 4;
3103 static av_always_inline void fill_filter_caches_inter(H264Context *h, MpegEncContext * const s, int mb_type, int top_xy,
3104 int left_xy[LEFT_MBS], int top_type, int left_type[LEFT_MBS], int mb_xy, int list)
3106 int b_stride = h->b_stride;
3107 int16_t (*mv_dst)[2] = &h->mv_cache[list][scan8[0]];
3108 int8_t *ref_cache = &h->ref_cache[list][scan8[0]];
3109 if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
3110 if(USES_LIST(top_type, list)){
3111 const int b_xy= h->mb2b_xy[top_xy] + 3*b_stride;
3112 const int b8_xy= 4*top_xy + 2;
3113 int (*ref2frm)[64] = h->ref2frm[ h->slice_table[top_xy]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
3114 AV_COPY128(mv_dst - 1*8, s->current_picture.f.motion_val[list][b_xy + 0]);
3116 ref_cache[1 - 1*8]= ref2frm[list][s->current_picture.f.ref_index[list][b8_xy + 0]];
3118 ref_cache[3 - 1*8]= ref2frm[list][s->current_picture.f.ref_index[list][b8_xy + 1]];
3120 AV_ZERO128(mv_dst - 1*8);
3121 AV_WN32A(&ref_cache[0 - 1*8], ((LIST_NOT_USED)&0xFF)*0x01010101u);
3124 if(!IS_INTERLACED(mb_type^left_type[LTOP])){
3125 if(USES_LIST(left_type[LTOP], list)){
3126 const int b_xy= h->mb2b_xy[left_xy[LTOP]] + 3;
3127 const int b8_xy= 4*left_xy[LTOP] + 1;
3128 int (*ref2frm)[64] = h->ref2frm[ h->slice_table[left_xy[LTOP]]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
3129 AV_COPY32(mv_dst - 1 + 0, s->current_picture.f.motion_val[list][b_xy + b_stride*0]);
3130 AV_COPY32(mv_dst - 1 + 8, s->current_picture.f.motion_val[list][b_xy + b_stride*1]);
3131 AV_COPY32(mv_dst - 1 + 16, s->current_picture.f.motion_val[list][b_xy + b_stride*2]);
3132 AV_COPY32(mv_dst - 1 + 24, s->current_picture.f.motion_val[list][b_xy + b_stride*3]);
3134 ref_cache[-1 + 8]= ref2frm[list][s->current_picture.f.ref_index[list][b8_xy + 2*0]];
3136 ref_cache[-1 + 24]= ref2frm[list][s->current_picture.f.ref_index[list][b8_xy + 2*1]];
3138 AV_ZERO32(mv_dst - 1 + 0);
3139 AV_ZERO32(mv_dst - 1 + 8);
3140 AV_ZERO32(mv_dst - 1 +16);
3141 AV_ZERO32(mv_dst - 1 +24);
3145 ref_cache[-1 + 24]= LIST_NOT_USED;
3150 if(!USES_LIST(mb_type, list)){
3151 fill_rectangle(mv_dst, 4, 4, 8, pack16to32(0,0), 4);
3152 AV_WN32A(&ref_cache[0*8], ((LIST_NOT_USED)&0xFF)*0x01010101u);
3153 AV_WN32A(&ref_cache[1*8], ((LIST_NOT_USED)&0xFF)*0x01010101u);
3154 AV_WN32A(&ref_cache[2*8], ((LIST_NOT_USED)&0xFF)*0x01010101u);
3155 AV_WN32A(&ref_cache[3*8], ((LIST_NOT_USED)&0xFF)*0x01010101u);
3160 int8_t *ref = &s->current_picture.f.ref_index[list][4*mb_xy];
3161 int (*ref2frm)[64] = h->ref2frm[ h->slice_num&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
3162 uint32_t ref01 = (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101;
3163 uint32_t ref23 = (pack16to32(ref2frm[list][ref[2]],ref2frm[list][ref[3]])&0x00FF00FF)*0x0101;
3164 AV_WN32A(&ref_cache[0*8], ref01);
3165 AV_WN32A(&ref_cache[1*8], ref01);
3166 AV_WN32A(&ref_cache[2*8], ref23);
3167 AV_WN32A(&ref_cache[3*8], ref23);
3171 int16_t (*mv_src)[2] = &s->current_picture.f.motion_val[list][4*s->mb_x + 4*s->mb_y*b_stride];
3172 AV_COPY128(mv_dst + 8*0, mv_src + 0*b_stride);
3173 AV_COPY128(mv_dst + 8*1, mv_src + 1*b_stride);
3174 AV_COPY128(mv_dst + 8*2, mv_src + 2*b_stride);
3175 AV_COPY128(mv_dst + 8*3, mv_src + 3*b_stride);
3181 * @return non zero if the loop filter can be skiped
3183 static int fill_filter_caches(H264Context *h, int mb_type){
3184 MpegEncContext * const s = &h->s;
3185 const int mb_xy= h->mb_xy;
3186 int top_xy, left_xy[LEFT_MBS];
3187 int top_type, left_type[LEFT_MBS];
3191 top_xy = mb_xy - (s->mb_stride << MB_FIELD);
3193 /* Wow, what a mess, why didn't they simplify the interlacing & intra
3194 * stuff, I can't imagine that these complex rules are worth it. */
3196 left_xy[LBOT] = left_xy[LTOP] = mb_xy-1;
3198 const int left_mb_field_flag = IS_INTERLACED(s->current_picture.f.mb_type[mb_xy - 1]);
3199 const int curr_mb_field_flag = IS_INTERLACED(mb_type);
3201 if (left_mb_field_flag != curr_mb_field_flag) {
3202 left_xy[LTOP] -= s->mb_stride;
3205 if(curr_mb_field_flag){
3206 top_xy += s->mb_stride & (((s->current_picture.f.mb_type[top_xy] >> 7) & 1) - 1);
3208 if (left_mb_field_flag != curr_mb_field_flag) {
3209 left_xy[LBOT] += s->mb_stride;
3214 h->top_mb_xy = top_xy;
3215 h->left_mb_xy[LTOP] = left_xy[LTOP];
3216 h->left_mb_xy[LBOT] = left_xy[LBOT];
3218 //for sufficiently low qp, filtering wouldn't do anything
3219 //this is a conservative estimate: could also check beta_offset and more accurate chroma_qp
3220 int qp_thresh = h->qp_thresh; //FIXME strictly we should store qp_thresh for each mb of a slice
3221 int qp = s->current_picture.f.qscale_table[mb_xy];
3223 && (left_xy[LTOP] < 0 || ((qp + s->current_picture.f.qscale_table[left_xy[LTOP]] + 1) >> 1) <= qp_thresh)
3224 && (top_xy < 0 || ((qp + s->current_picture.f.qscale_table[top_xy ] + 1) >> 1) <= qp_thresh)) {
3227 if ((left_xy[LTOP] < 0 || ((qp + s->current_picture.f.qscale_table[left_xy[LBOT] ] + 1) >> 1) <= qp_thresh) &&
3228 (top_xy < s->mb_stride || ((qp + s->current_picture.f.qscale_table[top_xy - s->mb_stride] + 1) >> 1) <= qp_thresh))
3233 top_type = s->current_picture.f.mb_type[top_xy];
3234 left_type[LTOP] = s->current_picture.f.mb_type[left_xy[LTOP]];
3235 left_type[LBOT] = s->current_picture.f.mb_type[left_xy[LBOT]];
3236 if(h->deblocking_filter == 2){
3237 if(h->slice_table[top_xy ] != h->slice_num) top_type= 0;
3238 if(h->slice_table[left_xy[LBOT]] != h->slice_num) left_type[LTOP]= left_type[LBOT]= 0;
3240 if(h->slice_table[top_xy ] == 0xFFFF) top_type= 0;
3241 if(h->slice_table[left_xy[LBOT]] == 0xFFFF) left_type[LTOP]= left_type[LBOT] =0;
3243 h->top_type = top_type;
3244 h->left_type[LTOP]= left_type[LTOP];
3245 h->left_type[LBOT]= left_type[LBOT];
3247 if(IS_INTRA(mb_type))
3250 fill_filter_caches_inter(h, s, mb_type, top_xy, left_xy, top_type, left_type, mb_xy, 0);
3251 if(h->list_count == 2)
3252 fill_filter_caches_inter(h, s, mb_type, top_xy, left_xy, top_type, left_type, mb_xy, 1);
3254 nnz = h->non_zero_count[mb_xy];
3255 nnz_cache = h->non_zero_count_cache;
3256 AV_COPY32(&nnz_cache[4+8*1], &nnz[ 0]);
3257 AV_COPY32(&nnz_cache[4+8*2], &nnz[ 4]);
3258 AV_COPY32(&nnz_cache[4+8*3], &nnz[ 8]);
3259 AV_COPY32(&nnz_cache[4+8*4], &nnz[12]);
3260 h->cbp= h->cbp_table[mb_xy];
3263 nnz = h->non_zero_count[top_xy];
3264 AV_COPY32(&nnz_cache[4+8*0], &nnz[3*4]);
3267 if(left_type[LTOP]){
3268 nnz = h->non_zero_count[left_xy[LTOP]];
3269 nnz_cache[3+8*1]= nnz[3+0*4];
3270 nnz_cache[3+8*2]= nnz[3+1*4];
3271 nnz_cache[3+8*3]= nnz[3+2*4];
3272 nnz_cache[3+8*4]= nnz[3+3*4];
3275 // CAVLC 8x8dct requires NNZ values for residual decoding that differ from what the loop filter needs
3276 if(!CABAC && h->pps.transform_8x8_mode){
3277 if(IS_8x8DCT(top_type)){
3279 nnz_cache[5+8*0]= (h->cbp_table[top_xy] & 0x4000) >> 12;
3281 nnz_cache[7+8*0]= (h->cbp_table[top_xy] & 0x8000) >> 12;
3283 if(IS_8x8DCT(left_type[LTOP])){
3285 nnz_cache[3+8*2]= (h->cbp_table[left_xy[LTOP]]&0x2000) >> 12; //FIXME check MBAFF
3287 if(IS_8x8DCT(left_type[LBOT])){
3289 nnz_cache[3+8*4]= (h->cbp_table[left_xy[LBOT]]&0x8000) >> 12; //FIXME check MBAFF
3292 if(IS_8x8DCT(mb_type)){
3293 nnz_cache[scan8[0 ]]= nnz_cache[scan8[1 ]]=
3294 nnz_cache[scan8[2 ]]= nnz_cache[scan8[3 ]]= (h->cbp & 0x1000) >> 12;
3296 nnz_cache[scan8[0+ 4]]= nnz_cache[scan8[1+ 4]]=
3297 nnz_cache[scan8[2+ 4]]= nnz_cache[scan8[3+ 4]]= (h->cbp & 0x2000) >> 12;
3299 nnz_cache[scan8[0+ 8]]= nnz_cache[scan8[1+ 8]]=
3300 nnz_cache[scan8[2+ 8]]= nnz_cache[scan8[3+ 8]]= (h->cbp & 0x4000) >> 12;
3302 nnz_cache[scan8[0+12]]= nnz_cache[scan8[1+12]]=
3303 nnz_cache[scan8[2+12]]= nnz_cache[scan8[3+12]]= (h->cbp & 0x8000) >> 12;
3310 static void loop_filter(H264Context *h, int start_x, int end_x){
3311 MpegEncContext * const s = &h->s;
3312 uint8_t *dest_y, *dest_cb, *dest_cr;
3313 int linesize, uvlinesize, mb_x, mb_y;
3314 const int end_mb_y= s->mb_y + FRAME_MBAFF;
3315 const int old_slice_type= h->slice_type;
3316 const int pixel_shift = h->pixel_shift;
3317 const int block_h = 16>>s->chroma_y_shift;
3319 if(h->deblocking_filter) {
3320 for(mb_x= start_x; mb_x<end_x; mb_x++){
3321 for(mb_y=end_mb_y - FRAME_MBAFF; mb_y<= end_mb_y; mb_y++){
3323 mb_xy = h->mb_xy = mb_x + mb_y*s->mb_stride;
3324 h->slice_num= h->slice_table[mb_xy];
3325 mb_type = s->current_picture.f.mb_type[mb_xy];
3326 h->list_count= h->list_counts[mb_xy];
3329 h->mb_mbaff = h->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
3333 dest_y = s->current_picture.f.data[0] + ((mb_x << pixel_shift) + mb_y * s->linesize ) * 16;
3334 dest_cb = s->current_picture.f.data[1] + (mb_x << pixel_shift)*(8<<CHROMA444) + mb_y * s->uvlinesize * block_h;
3335 dest_cr = s->current_picture.f.data[2] + (mb_x << pixel_shift)*(8<<CHROMA444) + mb_y * s->uvlinesize * block_h;
3336 //FIXME simplify above
3339 linesize = h->mb_linesize = s->linesize * 2;
3340 uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
3341 if(mb_y&1){ //FIXME move out of this function?
3342 dest_y -= s->linesize*15;
3343 dest_cb-= s->uvlinesize*(block_h-1);
3344 dest_cr-= s->uvlinesize*(block_h-1);
3347 linesize = h->mb_linesize = s->linesize;
3348 uvlinesize = h->mb_uvlinesize = s->uvlinesize;
3350 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);
3351 if(fill_filter_caches(h, mb_type))
3353 h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.f.qscale_table[mb_xy]);
3354 h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.f.qscale_table[mb_xy]);
3357 ff_h264_filter_mb (h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
3359 ff_h264_filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
3364 h->slice_type= old_slice_type;
3366 s->mb_y= end_mb_y - FRAME_MBAFF;
3367 h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
3368 h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
3371 static void predict_field_decoding_flag(H264Context *h){
3372 MpegEncContext * const s = &h->s;
3373 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
3374 int mb_type = (h->slice_table[mb_xy-1] == h->slice_num)
3375 ? s->current_picture.f.mb_type[mb_xy - 1]
3376 : (h->slice_table[mb_xy-s->mb_stride] == h->slice_num)
3377 ? s->current_picture.f.mb_type[mb_xy - s->mb_stride]
3379 h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
3383 * Draw edges and report progress for the last MB row.
3385 static void decode_finish_row(H264Context *h){
3386 MpegEncContext * const s = &h->s;
3387 int top = 16*(s->mb_y >> FIELD_PICTURE);
3388 int height = 16 << FRAME_MBAFF;
3389 int deblock_border = (16 + 4) << FRAME_MBAFF;
3390 int pic_height = 16*s->mb_height >> FIELD_PICTURE;
3392 if (h->deblocking_filter) {
3393 if((top + height) >= pic_height)
3394 height += deblock_border;
3396 top -= deblock_border;
3399 if (top >= pic_height || (top + height) < h->emu_edge_height)
3402 height = FFMIN(height, pic_height - top);
3403 if (top < h->emu_edge_height) {
3404 height = top+height;
3408 ff_draw_horiz_band(s, top, height);
3410 if (s->dropable) return;
3412 ff_thread_report_progress((AVFrame*)s->current_picture_ptr, top + height - 1,
3413 s->picture_structure==PICT_BOTTOM_FIELD);
3416 static int decode_slice(struct AVCodecContext *avctx, void *arg){
3417 H264Context *h = *(void**)arg;
3418 MpegEncContext * const s = &h->s;
3419 const int part_mask= s->partitioned_frame ? (AC_END|AC_ERROR) : 0x7F;
3420 int lf_x_start = s->mb_x;
3424 h->is_complex = FRAME_MBAFF || s->picture_structure != PICT_FRAME || s->codec_id != CODEC_ID_H264 ||
3425 (CONFIG_GRAY && (s->flags&CODEC_FLAG_GRAY));
3427 if( h->pps.cabac ) {
3429 align_get_bits( &s->gb );
3432 ff_init_cabac_states( &h->cabac);
3433 ff_init_cabac_decoder( &h->cabac,
3434 s->gb.buffer + get_bits_count(&s->gb)/8,
3435 (get_bits_left(&s->gb) + 7)/8);
3437 ff_h264_init_cabac_states(h);
3441 int ret = ff_h264_decode_mb_cabac(h);
3443 //STOP_TIMER("decode_mb_cabac")
3445 if(ret>=0) ff_h264_hl_decode_mb(h);
3447 if( ret >= 0 && FRAME_MBAFF ) { //FIXME optimal? or let mb_decode decode 16x32 ?
3450 ret = ff_h264_decode_mb_cabac(h);
3452 if(ret>=0) ff_h264_hl_decode_mb(h);
3455 eos = get_cabac_terminate( &h->cabac );
3457 if((s->workaround_bugs & FF_BUG_TRUNCATED) && h->cabac.bytestream > h->cabac.bytestream_end + 2){
3458 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);
3459 if (s->mb_x >= lf_x_start) loop_filter(h, lf_x_start, s->mb_x + 1);
3462 if( ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 2) {
3463 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);
3464 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);
3468 if( ++s->mb_x >= s->mb_width ) {
3469 loop_filter(h, lf_x_start, s->mb_x);
3470 s->mb_x = lf_x_start = 0;
3471 decode_finish_row(h);
3473 if(FIELD_OR_MBAFF_PICTURE) {
3475 if(FRAME_MBAFF && s->mb_y < s->mb_height)
3476 predict_field_decoding_flag(h);
3480 if( eos || s->mb_y >= s->mb_height ) {
3481 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
3482 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);
3483 if (s->mb_x > lf_x_start) loop_filter(h, lf_x_start, s->mb_x);
3490 int ret = ff_h264_decode_mb_cavlc(h);
3492 if(ret>=0) ff_h264_hl_decode_mb(h);
3494 if(ret>=0 && FRAME_MBAFF){ //FIXME optimal? or let mb_decode decode 16x32 ?
3496 ret = ff_h264_decode_mb_cavlc(h);
3498 if(ret>=0) ff_h264_hl_decode_mb(h);
3503 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
3504 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);
3508 if(++s->mb_x >= s->mb_width){
3509 loop_filter(h, lf_x_start, s->mb_x);
3510 s->mb_x = lf_x_start = 0;
3511 decode_finish_row(h);
3513 if(FIELD_OR_MBAFF_PICTURE) {
3515 if(FRAME_MBAFF && s->mb_y < s->mb_height)
3516 predict_field_decoding_flag(h);
3518 if(s->mb_y >= s->mb_height){
3519 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
3521 if( get_bits_count(&s->gb) == s->gb.size_in_bits
3522 || get_bits_count(&s->gb) < s->gb.size_in_bits && s->avctx->error_recognition < FF_ER_AGGRESSIVE) {
3523 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);
3527 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);
3534 if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->mb_skip_run<=0){
3535 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
3536 if(get_bits_count(&s->gb) == s->gb.size_in_bits ){
3537 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);
3538 if (s->mb_x > lf_x_start) loop_filter(h, lf_x_start, s->mb_x);
3542 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);
3552 * Call decode_slice() for each context.
3554 * @param h h264 master context
3555 * @param context_count number of contexts to execute
3557 static int execute_decode_slices(H264Context *h, int context_count){
3558 MpegEncContext * const s = &h->s;
3559 AVCodecContext * const avctx= s->avctx;
3563 if (s->avctx->hwaccel || s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
3565 if(context_count == 1) {
3566 return decode_slice(avctx, &h);
3568 for(i = 1; i < context_count; i++) {
3569 hx = h->thread_context[i];
3570 hx->s.error_recognition = avctx->error_recognition;
3571 hx->s.error_count = 0;
3572 hx->x264_build= h->x264_build;
3575 avctx->execute(avctx, (void *)decode_slice,
3576 h->thread_context, NULL, context_count, sizeof(void*));
3578 /* pull back stuff from slices to master context */
3579 hx = h->thread_context[context_count - 1];
3580 s->mb_x = hx->s.mb_x;
3581 s->mb_y = hx->s.mb_y;
3582 s->dropable = hx->s.dropable;
3583 s->picture_structure = hx->s.picture_structure;
3584 for(i = 1; i < context_count; i++)
3585 h->s.error_count += h->thread_context[i]->s.error_count;
3592 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size){
3593 MpegEncContext * const s = &h->s;
3594 AVCodecContext * const avctx= s->avctx;
3595 H264Context *hx; ///< thread context
3599 int pass = !(avctx->active_thread_type & FF_THREAD_FRAME);
3600 int nals_needed=0; ///< number of NALs that need decoding before the next frame thread starts
3603 h->max_contexts = (HAVE_THREADS && (s->avctx->active_thread_type&FF_THREAD_SLICE)) ? avctx->thread_count : 1;
3604 if(!(s->flags2 & CODEC_FLAG2_CHUNKS)){
3605 h->current_slice = 0;
3606 if (!s->first_field)
3607 s->current_picture_ptr= NULL;
3608 ff_h264_reset_sei(h);
3611 for(;pass <= 1;pass++){
3614 next_avc = h->is_avc ? 0 : buf_size;
3624 if(buf_index >= next_avc) {
3625 if(buf_index >= buf_size) break;
3627 for(i = 0; i < h->nal_length_size; i++)
3628 nalsize = (nalsize << 8) | buf[buf_index++];
3629 if(nalsize <= 0 || nalsize > buf_size - buf_index){
3630 av_log(h->s.avctx, AV_LOG_ERROR, "AVC: nal size %d\n", nalsize);
3633 next_avc= buf_index + nalsize;
3635 // start code prefix search
3636 for(; buf_index + 3 < next_avc; buf_index++){
3637 // This should always succeed in the first iteration.
3638 if(buf[buf_index] == 0 && buf[buf_index+1] == 0 && buf[buf_index+2] == 1)
3642 if(buf_index+3 >= buf_size) break;
3645 if(buf_index >= next_avc) continue;
3648 hx = h->thread_context[context_count];
3650 ptr= ff_h264_decode_nal(hx, buf + buf_index, &dst_length, &consumed, next_avc - buf_index);
3651 if (ptr==NULL || dst_length < 0){
3654 i= buf_index + consumed;
3655 if((s->workaround_bugs & FF_BUG_AUTODETECT) && i+3<next_avc &&
3656 buf[i]==0x00 && buf[i+1]==0x00 && buf[i+2]==0x01 && buf[i+3]==0xE0)
3657 s->workaround_bugs |= FF_BUG_TRUNCATED;
3659 if(!(s->workaround_bugs & FF_BUG_TRUNCATED)){
3660 while(dst_length > 0 && ptr[dst_length - 1] == 0)
3663 bit_length= !dst_length ? 0 : (8*dst_length - ff_h264_decode_rbsp_trailing(h, ptr + dst_length - 1));
3665 if(s->avctx->debug&FF_DEBUG_STARTCODE){
3666 av_log(h->s.avctx, AV_LOG_DEBUG, "NAL %d/%d at %d/%d length %d\n", hx->nal_unit_type, hx->nal_ref_idc, buf_index, buf_size, dst_length);
3669 if (h->is_avc && (nalsize != consumed) && nalsize){
3670 av_log(h->s.avctx, AV_LOG_DEBUG, "AVC: Consumed only %d bytes instead of %d\n", consumed, nalsize);
3673 buf_index += consumed;
3677 // packets can sometimes contain multiple PPS/SPS
3678 // e.g. two PAFF field pictures in one packet, or a demuxer which splits NALs strangely
3679 // if so, when frame threading we can't start the next thread until we've read all of them
3680 switch (hx->nal_unit_type) {
3685 nals_needed = nal_index;
3690 //FIXME do not discard SEI id
3691 if(avctx->skip_frame >= AVDISCARD_NONREF && h->nal_ref_idc == 0)
3696 switch(hx->nal_unit_type){
3698 if (h->nal_unit_type != NAL_IDR_SLICE) {
3699 av_log(h->s.avctx, AV_LOG_ERROR, "Invalid mix of idr and non-idr slices");
3702 idr(h); //FIXME ensure we don't loose some frames if there is reordering
3704 init_get_bits(&hx->s.gb, ptr, bit_length);
3706 hx->inter_gb_ptr= &hx->s.gb;
3707 hx->s.data_partitioning = 0;
3709 if((err = decode_slice_header(hx, h)))
3712 s->current_picture_ptr->f.key_frame |=
3713 (hx->nal_unit_type == NAL_IDR_SLICE) ||
3714 (h->sei_recovery_frame_cnt >= 0);
3715 h->sync |= !!s->current_picture_ptr->f.key_frame;
3716 s->current_picture_ptr->sync = h->sync;
3718 if (h->current_slice == 1) {
3719 if(!(s->flags2 & CODEC_FLAG2_CHUNKS)) {
3720 decode_postinit(h, nal_index >= nals_needed);
3723 if (s->avctx->hwaccel && s->avctx->hwaccel->start_frame(s->avctx, NULL, 0) < 0)
3725 if(CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
3726 ff_vdpau_h264_picture_start(s);
3729 if(hx->redundant_pic_count==0
3730 && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
3731 && (avctx->skip_frame < AVDISCARD_BIDIR || hx->slice_type_nos!=AV_PICTURE_TYPE_B)
3732 && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==AV_PICTURE_TYPE_I)
3733 && avctx->skip_frame < AVDISCARD_ALL){
3734 if(avctx->hwaccel) {
3735 if (avctx->hwaccel->decode_slice(avctx, &buf[buf_index - consumed], consumed) < 0)
3738 if(CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU){
3739 static const uint8_t start_code[] = {0x00, 0x00, 0x01};
3740 ff_vdpau_add_data_chunk(s, start_code, sizeof(start_code));
3741 ff_vdpau_add_data_chunk(s, &buf[buf_index - consumed], consumed );
3747 init_get_bits(&hx->s.gb, ptr, bit_length);
3749 hx->inter_gb_ptr= NULL;
3751 if ((err = decode_slice_header(hx, h)) < 0)
3754 hx->s.data_partitioning = 1;
3758 init_get_bits(&hx->intra_gb, ptr, bit_length);
3759 hx->intra_gb_ptr= &hx->intra_gb;
3762 init_get_bits(&hx->inter_gb, ptr, bit_length);
3763 hx->inter_gb_ptr= &hx->inter_gb;
3765 if(hx->redundant_pic_count==0 && hx->intra_gb_ptr && hx->s.data_partitioning
3766 && s->context_initialized
3767 && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
3768 && (avctx->skip_frame < AVDISCARD_BIDIR || hx->slice_type_nos!=AV_PICTURE_TYPE_B)
3769 && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==AV_PICTURE_TYPE_I)
3770 && avctx->skip_frame < AVDISCARD_ALL)
3774 init_get_bits(&s->gb, ptr, bit_length);
3775 ff_h264_decode_sei(h);
3778 init_get_bits(&s->gb, ptr, bit_length);
3779 ff_h264_decode_seq_parameter_set(h);
3781 if (s->flags& CODEC_FLAG_LOW_DELAY ||
3782 (h->sps.bitstream_restriction_flag && !h->sps.num_reorder_frames))
3785 if(avctx->has_b_frames < 2)
3786 avctx->has_b_frames= !s->low_delay;
3788 if (avctx->bits_per_raw_sample != h->sps.bit_depth_luma ||
3789 h->cur_chroma_format_idc != h->sps.chroma_format_idc) {
3790 if (h->sps.bit_depth_luma >= 8 && h->sps.bit_depth_luma <= 10) {
3791 avctx->bits_per_raw_sample = h->sps.bit_depth_luma;
3792 h->cur_chroma_format_idc = h->sps.chroma_format_idc;
3793 h->pixel_shift = h->sps.bit_depth_luma > 8;
3795 ff_h264dsp_init(&h->h264dsp, h->sps.bit_depth_luma, h->sps.chroma_format_idc);
3796 ff_h264_pred_init(&h->hpc, s->codec_id, h->sps.bit_depth_luma, h->sps.chroma_format_idc);
3797 s->dsp.dct_bits = h->sps.bit_depth_luma > 8 ? 32 : 16;
3798 dsputil_init(&s->dsp, s->avctx);
3800 av_log(avctx, AV_LOG_DEBUG, "Unsupported bit depth: %d\n", h->sps.bit_depth_luma);
3806 init_get_bits(&s->gb, ptr, bit_length);
3808 ff_h264_decode_picture_parameter_set(h, bit_length);
3812 case NAL_END_SEQUENCE:
3813 case NAL_END_STREAM:
3814 case NAL_FILLER_DATA:
3816 case NAL_AUXILIARY_SLICE:
3819 av_log(avctx, AV_LOG_DEBUG, "Unknown NAL code: %d (%d bits)\n", hx->nal_unit_type, bit_length);
3822 if(context_count == h->max_contexts) {
3823 execute_decode_slices(h, context_count);
3828 av_log(h->s.avctx, AV_LOG_ERROR, "decode_slice_header error\n");
3830 /* Slice could not be decoded in parallel mode, copy down
3831 * NAL unit stuff to context 0 and restart. Note that
3832 * rbsp_buffer is not transferred, but since we no longer
3833 * run in parallel mode this should not be an issue. */
3834 h->nal_unit_type = hx->nal_unit_type;
3835 h->nal_ref_idc = hx->nal_ref_idc;
3842 execute_decode_slices(h, context_count);
3847 * returns the number of bytes consumed for building the current frame
3849 static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size){
3850 if(pos==0) pos=1; //avoid infinite loops (i doubt that is needed but ...)
3851 if(pos+10>buf_size) pos=buf_size; // oops ;)
3856 static int decode_frame(AVCodecContext *avctx,
3857 void *data, int *data_size,
3860 const uint8_t *buf = avpkt->data;
3861 int buf_size = avpkt->size;
3862 H264Context *h = avctx->priv_data;
3863 MpegEncContext *s = &h->s;
3864 AVFrame *pict = data;
3867 s->flags= avctx->flags;
3868 s->flags2= avctx->flags2;
3870 /* end of stream, output what is still in the buffers */
3872 if (buf_size == 0) {
3876 s->current_picture_ptr = NULL;
3878 //FIXME factorize this with the output code below
3879 out = h->delayed_pic[0];
3881 for (i = 1; h->delayed_pic[i] && !h->delayed_pic[i]->f.key_frame && !h->delayed_pic[i]->mmco_reset; i++)
3882 if(h->delayed_pic[i]->poc < out->poc){
3883 out = h->delayed_pic[i];
3887 for(i=out_idx; h->delayed_pic[i]; i++)
3888 h->delayed_pic[i] = h->delayed_pic[i+1];
3891 *data_size = sizeof(AVFrame);
3892 *pict= *(AVFrame*)out;
3898 buf_index=decode_nal_units(h, buf, buf_size);
3902 if (!s->current_picture_ptr && h->nal_unit_type == NAL_END_SEQUENCE) {
3907 if(!(s->flags2 & CODEC_FLAG2_CHUNKS) && !s->current_picture_ptr){
3908 if (avctx->skip_frame >= AVDISCARD_NONREF)
3910 av_log(avctx, AV_LOG_ERROR, "no frame!\n");
3914 if(!(s->flags2 & CODEC_FLAG2_CHUNKS) || (s->mb_y >= s->mb_height && s->mb_height)){
3916 if(s->flags2 & CODEC_FLAG2_CHUNKS) decode_postinit(h, 1);
3920 *data_size = 0; /* Wait for second field. */
3921 if (h->next_output_pic && h->next_output_pic->sync) {
3922 h->sync |= 2*!!h->next_output_pic->f.key_frame;
3923 if(h->sync>1 || h->next_output_pic->f.pict_type != AV_PICTURE_TYPE_B){
3924 *data_size = sizeof(AVFrame);
3925 *pict = *(AVFrame*)h->next_output_pic;
3930 assert(pict->data[0] || !*data_size);
3931 ff_print_debug_info(s, pict);
3932 //printf("out %d\n", (int)pict->data[0]);
3934 return get_consumed_bytes(s, buf_index, buf_size);
3937 static inline void fill_mb_avail(H264Context *h){
3938 MpegEncContext * const s = &h->s;
3939 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
3942 h->mb_avail[0]= s->mb_x && h->slice_table[mb_xy - s->mb_stride - 1] == h->slice_num;
3943 h->mb_avail[1]= h->slice_table[mb_xy - s->mb_stride ] == h->slice_num;
3944 h->mb_avail[2]= s->mb_x+1 < s->mb_width && h->slice_table[mb_xy - s->mb_stride + 1] == h->slice_num;
3950 h->mb_avail[3]= s->mb_x && h->slice_table[mb_xy - 1] == h->slice_num;
3951 h->mb_avail[4]= 1; //FIXME move out
3952 h->mb_avail[5]= 0; //FIXME move out
3960 #define SIZE (COUNT*40)
3961 extern AVCodec ff_h264_decoder;
3967 // int int_temp[10000];
3969 AVCodecContext avctx;
3971 avcodec_get_context_defaults3(&avctx, &ff_h264_decoder);
3973 dsputil_init(&dsp, &avctx);
3975 init_put_bits(&pb, temp, SIZE);
3976 printf("testing unsigned exp golomb\n");
3977 for(i=0; i<COUNT; i++){
3979 set_ue_golomb(&pb, i);
3980 STOP_TIMER("set_ue_golomb");
3982 flush_put_bits(&pb);
3984 init_get_bits(&gb, temp, 8*SIZE);
3985 for(i=0; i<COUNT; i++){
3988 s= show_bits(&gb, 24);
3991 j= get_ue_golomb(&gb);
3993 printf("mismatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
3996 STOP_TIMER("get_ue_golomb");
4000 init_put_bits(&pb, temp, SIZE);
4001 printf("testing signed exp golomb\n");
4002 for(i=0; i<COUNT; i++){
4004 set_se_golomb(&pb, i - COUNT/2);
4005 STOP_TIMER("set_se_golomb");
4007 flush_put_bits(&pb);
4009 init_get_bits(&gb, temp, 8*SIZE);
4010 for(i=0; i<COUNT; i++){
4013 s= show_bits(&gb, 24);
4016 j= get_se_golomb(&gb);
4017 if(j != i - COUNT/2){
4018 printf("mismatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
4021 STOP_TIMER("get_se_golomb");
4024 printf("Testing RBSP\n");
4032 av_cold void ff_h264_free_context(H264Context *h)
4036 free_tables(h, 1); //FIXME cleanup init stuff perhaps
4038 for(i = 0; i < MAX_SPS_COUNT; i++)
4039 av_freep(h->sps_buffers + i);
4041 for(i = 0; i < MAX_PPS_COUNT; i++)
4042 av_freep(h->pps_buffers + i);
4045 av_cold int ff_h264_decode_end(AVCodecContext *avctx)
4047 H264Context *h = avctx->priv_data;
4048 MpegEncContext *s = &h->s;
4050 ff_h264_free_context(h);
4054 // memset(h, 0, sizeof(H264Context));
4059 static const AVProfile profiles[] = {
4060 { FF_PROFILE_H264_BASELINE, "Baseline" },
4061 { FF_PROFILE_H264_CONSTRAINED_BASELINE, "Constrained Baseline" },
4062 { FF_PROFILE_H264_MAIN, "Main" },
4063 { FF_PROFILE_H264_EXTENDED, "Extended" },
4064 { FF_PROFILE_H264_HIGH, "High" },
4065 { FF_PROFILE_H264_HIGH_10, "High 10" },
4066 { FF_PROFILE_H264_HIGH_10_INTRA, "High 10 Intra" },
4067 { FF_PROFILE_H264_HIGH_422, "High 4:2:2" },
4068 { FF_PROFILE_H264_HIGH_422_INTRA, "High 4:2:2 Intra" },
4069 { FF_PROFILE_H264_HIGH_444, "High 4:4:4" },
4070 { FF_PROFILE_H264_HIGH_444_PREDICTIVE, "High 4:4:4 Predictive" },
4071 { FF_PROFILE_H264_HIGH_444_INTRA, "High 4:4:4 Intra" },
4072 { FF_PROFILE_H264_CAVLC_444, "CAVLC 4:4:4" },
4073 { FF_PROFILE_UNKNOWN },
4076 AVCodec ff_h264_decoder = {
4078 .type = AVMEDIA_TYPE_VIDEO,
4079 .id = CODEC_ID_H264,
4080 .priv_data_size = sizeof(H264Context),
4081 .init = ff_h264_decode_init,
4082 .close = ff_h264_decode_end,
4083 .decode = decode_frame,
4084 .capabilities = /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 | CODEC_CAP_DELAY |
4085 CODEC_CAP_SLICE_THREADS | CODEC_CAP_FRAME_THREADS,
4087 .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10"),
4088 .init_thread_copy = ONLY_IF_THREADS_ENABLED(decode_init_thread_copy),
4089 .update_thread_context = ONLY_IF_THREADS_ENABLED(decode_update_thread_context),
4090 .profiles = NULL_IF_CONFIG_SMALL(profiles),
4093 #if CONFIG_H264_VDPAU_DECODER
4094 AVCodec ff_h264_vdpau_decoder = {
4095 .name = "h264_vdpau",
4096 .type = AVMEDIA_TYPE_VIDEO,
4097 .id = CODEC_ID_H264,
4098 .priv_data_size = sizeof(H264Context),
4099 .init = ff_h264_decode_init,
4100 .close = ff_h264_decode_end,
4101 .decode = decode_frame,
4102 .capabilities = CODEC_CAP_DR1 | CODEC_CAP_DELAY | CODEC_CAP_HWACCEL_VDPAU,
4104 .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10 (VDPAU acceleration)"),
4105 .pix_fmts = (const enum PixelFormat[]){PIX_FMT_VDPAU_H264, PIX_FMT_NONE},
4106 .profiles = NULL_IF_CONFIG_SMALL(profiles),