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 Libav.
7 * Libav 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 * Libav 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 Libav; 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 if(i>=length-1){ //no escaped 0
179 *consumed= length+1; //+1 for the header
183 bufidx = h->nal_unit_type == NAL_DPC ? 1 : 0; // use second escape buffer for inter data
184 av_fast_malloc(&h->rbsp_buffer[bufidx], &h->rbsp_buffer_size[bufidx], length+FF_INPUT_BUFFER_PADDING_SIZE);
185 dst= h->rbsp_buffer[bufidx];
191 //printf("decoding esc\n");
195 //remove escapes (very rare 1:2^22)
197 dst[di++]= src[si++];
198 dst[di++]= src[si++];
199 }else if(src[si]==0 && src[si+1]==0){
200 if(src[si+2]==3){ //escape
205 }else //next start code
209 dst[di++]= src[si++];
212 dst[di++]= src[si++];
215 memset(dst+di, 0, FF_INPUT_BUFFER_PADDING_SIZE);
218 *consumed= si + 1;//+1 for the header
219 //FIXME store exact number of bits in the getbitcontext (it is needed for decoding)
224 * Identify the exact end of the bitstream
225 * @return the length of the trailing, or 0 if damaged
227 static int ff_h264_decode_rbsp_trailing(H264Context *h, const uint8_t *src){
231 tprintf(h->s.avctx, "rbsp trailing %X\n", v);
240 static inline int get_lowest_part_list_y(H264Context *h, Picture *pic, int n, int height,
241 int y_offset, int list){
242 int raw_my= h->mv_cache[list][ scan8[n] ][1];
243 int filter_height= (raw_my&3) ? 2 : 0;
244 int full_my= (raw_my>>2) + y_offset;
245 int top = full_my - filter_height, bottom = full_my + height + filter_height;
247 return FFMAX(abs(top), bottom);
250 static inline void get_lowest_part_y(H264Context *h, int refs[2][48], int n, int height,
251 int y_offset, int list0, int list1, int *nrefs){
252 MpegEncContext * const s = &h->s;
255 y_offset += 16*(s->mb_y >> MB_FIELD);
258 int ref_n = h->ref_cache[0][ scan8[n] ];
259 Picture *ref= &h->ref_list[0][ref_n];
261 // Error resilience puts the current picture in the ref list.
262 // Don't try to wait on these as it will cause a deadlock.
263 // Fields can wait on each other, though.
264 if (ref->f.thread_opaque != s->current_picture.f.thread_opaque ||
265 (ref->f.reference & 3) != s->picture_structure) {
266 my = get_lowest_part_list_y(h, ref, n, height, y_offset, 0);
267 if (refs[0][ref_n] < 0) nrefs[0] += 1;
268 refs[0][ref_n] = FFMAX(refs[0][ref_n], my);
273 int ref_n = h->ref_cache[1][ scan8[n] ];
274 Picture *ref= &h->ref_list[1][ref_n];
276 if (ref->f.thread_opaque != s->current_picture.f.thread_opaque ||
277 (ref->f.reference & 3) != s->picture_structure) {
278 my = get_lowest_part_list_y(h, ref, n, height, y_offset, 1);
279 if (refs[1][ref_n] < 0) nrefs[1] += 1;
280 refs[1][ref_n] = FFMAX(refs[1][ref_n], my);
286 * Wait until all reference frames are available for MC operations.
288 * @param h the H264 context
290 static void await_references(H264Context *h){
291 MpegEncContext * const s = &h->s;
292 const int mb_xy= h->mb_xy;
293 const int mb_type = s->current_picture.f.mb_type[mb_xy];
298 memset(refs, -1, sizeof(refs));
300 if(IS_16X16(mb_type)){
301 get_lowest_part_y(h, refs, 0, 16, 0,
302 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs);
303 }else if(IS_16X8(mb_type)){
304 get_lowest_part_y(h, refs, 0, 8, 0,
305 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs);
306 get_lowest_part_y(h, refs, 8, 8, 8,
307 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1), nrefs);
308 }else if(IS_8X16(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 get_lowest_part_y(h, refs, 4, 16, 0,
312 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1), nrefs);
316 assert(IS_8X8(mb_type));
319 const int sub_mb_type= h->sub_mb_type[i];
321 int y_offset= (i&2)<<2;
323 if(IS_SUB_8X8(sub_mb_type)){
324 get_lowest_part_y(h, refs, n , 8, y_offset,
325 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs);
326 }else if(IS_SUB_8X4(sub_mb_type)){
327 get_lowest_part_y(h, refs, n , 4, y_offset,
328 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs);
329 get_lowest_part_y(h, refs, n+2, 4, y_offset+4,
330 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs);
331 }else if(IS_SUB_4X8(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 get_lowest_part_y(h, refs, n+1, 8, y_offset,
335 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs);
338 assert(IS_SUB_4X4(sub_mb_type));
340 int sub_y_offset= y_offset + 2*(j&2);
341 get_lowest_part_y(h, refs, n+j, 4, sub_y_offset,
342 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs);
348 for(list=h->list_count-1; list>=0; list--){
349 for(ref=0; ref<48 && nrefs[list]; ref++){
350 int row = refs[list][ref];
352 Picture *ref_pic = &h->ref_list[list][ref];
353 int ref_field = ref_pic->f.reference - 1;
354 int ref_field_picture = ref_pic->field_picture;
355 int pic_height = 16*s->mb_height >> ref_field_picture;
360 if(!FIELD_PICTURE && ref_field_picture){ // frame referencing two fields
361 ff_thread_await_progress((AVFrame*)ref_pic, FFMIN((row >> 1) - !(row&1), pic_height-1), 1);
362 ff_thread_await_progress((AVFrame*)ref_pic, FFMIN((row >> 1) , pic_height-1), 0);
363 }else if(FIELD_PICTURE && !ref_field_picture){ // field referencing one field of a frame
364 ff_thread_await_progress((AVFrame*)ref_pic, FFMIN(row*2 + ref_field , pic_height-1), 0);
365 }else if(FIELD_PICTURE){
366 ff_thread_await_progress((AVFrame*)ref_pic, FFMIN(row, pic_height-1), ref_field);
368 ff_thread_await_progress((AVFrame*)ref_pic, FFMIN(row, pic_height-1), 0);
377 * DCT transforms the 16 dc values.
378 * @param qp quantization parameter ??? FIXME
380 static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
381 // const int qmul= dequant_coeff[qp][0];
383 int temp[16]; //FIXME check if this is a good idea
384 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
385 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
388 const int offset= y_offset[i];
389 const int z0= block[offset+stride*0] + block[offset+stride*4];
390 const int z1= block[offset+stride*0] - block[offset+stride*4];
391 const int z2= block[offset+stride*1] - block[offset+stride*5];
392 const int z3= block[offset+stride*1] + block[offset+stride*5];
401 const int offset= x_offset[i];
402 const int z0= temp[4*0+i] + temp[4*2+i];
403 const int z1= temp[4*0+i] - temp[4*2+i];
404 const int z2= temp[4*1+i] - temp[4*3+i];
405 const int z3= temp[4*1+i] + temp[4*3+i];
407 block[stride*0 +offset]= (z0 + z3)>>1;
408 block[stride*2 +offset]= (z1 + z2)>>1;
409 block[stride*8 +offset]= (z1 - z2)>>1;
410 block[stride*10+offset]= (z0 - z3)>>1;
419 static void chroma_dc_dct_c(DCTELEM *block){
420 const int stride= 16*2;
421 const int xStride= 16;
424 a= block[stride*0 + xStride*0];
425 b= block[stride*0 + xStride*1];
426 c= block[stride*1 + xStride*0];
427 d= block[stride*1 + xStride*1];
434 block[stride*0 + xStride*0]= (a+c);
435 block[stride*0 + xStride*1]= (e+b);
436 block[stride*1 + xStride*0]= (a-c);
437 block[stride*1 + xStride*1]= (e-b);
441 static av_always_inline void
442 mc_dir_part(H264Context *h, Picture *pic, int n, int square,
443 int height, int delta, int list,
444 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
445 int src_x_offset, int src_y_offset,
446 qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op,
447 int pixel_shift, int chroma_idc)
449 MpegEncContext * const s = &h->s;
450 const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
451 int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
452 const int luma_xy= (mx&3) + ((my&3)<<2);
453 int offset = ((mx>>2) << pixel_shift) + (my>>2)*h->mb_linesize;
454 uint8_t * src_y = pic->f.data[0] + offset;
455 uint8_t * src_cb, * src_cr;
456 int extra_width= h->emu_edge_width;
457 int extra_height= h->emu_edge_height;
459 const int full_mx= mx>>2;
460 const int full_my= my>>2;
461 const int pic_width = 16*s->mb_width;
462 const int pic_height = 16*s->mb_height >> MB_FIELD;
465 if(mx&7) extra_width -= 3;
466 if(my&7) extra_height -= 3;
468 if( full_mx < 0-extra_width
469 || full_my < 0-extra_height
470 || full_mx + 16/*FIXME*/ > pic_width + extra_width
471 || full_my + 16/*FIXME*/ > pic_height + extra_height){
472 s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_y - (2 << pixel_shift) - 2*h->mb_linesize, h->mb_linesize,
473 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
474 src_y= s->edge_emu_buffer + (2 << pixel_shift) + 2*h->mb_linesize;
478 qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); //FIXME try variable height perhaps?
480 qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
483 if(CONFIG_GRAY && s->flags&CODEC_FLAG_GRAY) return;
485 if(chroma_idc == 3 /* yuv444 */){
486 src_cb = pic->f.data[1] + offset;
488 s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cb - (2 << pixel_shift) - 2*h->mb_linesize, h->mb_linesize,
489 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
490 src_cb= s->edge_emu_buffer + (2 << pixel_shift) + 2*h->mb_linesize;
492 qpix_op[luma_xy](dest_cb, src_cb, h->mb_linesize); //FIXME try variable height perhaps?
494 qpix_op[luma_xy](dest_cb + delta, src_cb + delta, h->mb_linesize);
497 src_cr = pic->f.data[2] + offset;
499 s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cr - (2 << pixel_shift) - 2*h->mb_linesize, h->mb_linesize,
500 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
501 src_cr= s->edge_emu_buffer + (2 << pixel_shift) + 2*h->mb_linesize;
503 qpix_op[luma_xy](dest_cr, src_cr, h->mb_linesize); //FIXME try variable height perhaps?
505 qpix_op[luma_xy](dest_cr + delta, src_cr + delta, h->mb_linesize);
510 ysh = 3 - (chroma_idc == 2 /* yuv422 */);
511 if(chroma_idc == 1 /* yuv420 */ && MB_FIELD){
512 // chroma offset when predicting from a field of opposite parity
513 my += 2 * ((s->mb_y & 1) - (pic->f.reference - 1));
514 emu |= (my>>3) < 0 || (my>>3) + 8 >= (pic_height>>1);
517 src_cb = pic->f.data[1] + ((mx >> 3) << pixel_shift) + (my >> ysh) * h->mb_uvlinesize;
518 src_cr = pic->f.data[2] + ((mx >> 3) << pixel_shift) + (my >> ysh) * h->mb_uvlinesize;
521 s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cb, h->mb_uvlinesize,
522 9, 8 * chroma_idc + 1, (mx >> 3), (my >> ysh),
523 pic_width >> 1, pic_height >> (chroma_idc == 1 /* yuv420 */));
524 src_cb= s->edge_emu_buffer;
526 chroma_op(dest_cb, src_cb, h->mb_uvlinesize, height >> (chroma_idc == 1 /* yuv420 */),
527 mx&7, (my << (chroma_idc == 2 /* yuv422 */)) &7);
530 s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cr, h->mb_uvlinesize,
531 9, 8 * chroma_idc + 1, (mx >> 3), (my >> ysh),
532 pic_width >> 1, pic_height >> (chroma_idc == 1 /* yuv420 */));
533 src_cr= s->edge_emu_buffer;
535 chroma_op(dest_cr, src_cr, h->mb_uvlinesize, height >> (chroma_idc == 1 /* yuv420 */),
536 mx&7, (my << (chroma_idc == 2 /* yuv422 */)) &7);
539 static av_always_inline void
540 mc_part_std(H264Context *h, int n, int square, int height, int delta,
541 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
542 int x_offset, int y_offset,
543 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
544 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
545 int list0, int list1, int pixel_shift, int chroma_idc)
547 MpegEncContext * const s = &h->s;
548 qpel_mc_func *qpix_op= qpix_put;
549 h264_chroma_mc_func chroma_op= chroma_put;
551 dest_y += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
552 if (chroma_idc == 3 /* yuv444 */) {
553 dest_cb += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
554 dest_cr += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
555 } else if (chroma_idc == 2 /* yuv422 */) {
556 dest_cb += ( x_offset << pixel_shift) + 2*y_offset*h->mb_uvlinesize;
557 dest_cr += ( x_offset << pixel_shift) + 2*y_offset*h->mb_uvlinesize;
558 } else /* yuv420 */ {
559 dest_cb += ( x_offset << pixel_shift) + y_offset*h->mb_uvlinesize;
560 dest_cr += ( x_offset << pixel_shift) + y_offset*h->mb_uvlinesize;
562 x_offset += 8*s->mb_x;
563 y_offset += 8*(s->mb_y >> MB_FIELD);
566 Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
567 mc_dir_part(h, ref, n, square, height, delta, 0,
568 dest_y, dest_cb, dest_cr, x_offset, y_offset,
569 qpix_op, chroma_op, pixel_shift, chroma_idc);
572 chroma_op= chroma_avg;
576 Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
577 mc_dir_part(h, ref, n, square, height, delta, 1,
578 dest_y, dest_cb, dest_cr, x_offset, y_offset,
579 qpix_op, chroma_op, pixel_shift, chroma_idc);
583 static av_always_inline void
584 mc_part_weighted(H264Context *h, int n, int square, int height, int delta,
585 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
586 int x_offset, int y_offset,
587 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
588 h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
589 h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
590 int list0, int list1, int pixel_shift, int chroma_idc){
591 MpegEncContext * const s = &h->s;
594 dest_y += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
595 if (chroma_idc == 3 /* yuv444 */) {
596 chroma_height = height;
597 chroma_weight_avg = luma_weight_avg;
598 chroma_weight_op = luma_weight_op;
599 dest_cb += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
600 dest_cr += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
601 } else if (chroma_idc == 2 /* yuv422 */) {
602 chroma_height = height;
603 dest_cb += ( x_offset << pixel_shift) + 2*y_offset*h->mb_uvlinesize;
604 dest_cr += ( x_offset << pixel_shift) + 2*y_offset*h->mb_uvlinesize;
605 } else /* yuv420 */ {
606 chroma_height = height >> 1;
607 dest_cb += ( x_offset << pixel_shift) + y_offset*h->mb_uvlinesize;
608 dest_cr += ( x_offset << pixel_shift) + y_offset*h->mb_uvlinesize;
610 x_offset += 8*s->mb_x;
611 y_offset += 8*(s->mb_y >> MB_FIELD);
614 /* don't optimize for luma-only case, since B-frames usually
615 * use implicit weights => chroma too. */
616 uint8_t *tmp_cb = s->obmc_scratchpad;
617 uint8_t *tmp_cr = s->obmc_scratchpad + (16 << pixel_shift);
618 uint8_t *tmp_y = s->obmc_scratchpad + 16*h->mb_uvlinesize;
619 int refn0 = h->ref_cache[0][ scan8[n] ];
620 int refn1 = h->ref_cache[1][ scan8[n] ];
622 mc_dir_part(h, &h->ref_list[0][refn0], n, square, height, delta, 0,
623 dest_y, dest_cb, dest_cr,
624 x_offset, y_offset, qpix_put, chroma_put,
625 pixel_shift, chroma_idc);
626 mc_dir_part(h, &h->ref_list[1][refn1], n, square, height, delta, 1,
627 tmp_y, tmp_cb, tmp_cr,
628 x_offset, y_offset, qpix_put, chroma_put,
629 pixel_shift, chroma_idc);
631 if(h->use_weight == 2){
632 int weight0 = h->implicit_weight[refn0][refn1][s->mb_y&1];
633 int weight1 = 64 - weight0;
634 luma_weight_avg( dest_y, tmp_y, h-> mb_linesize,
635 height, 5, weight0, weight1, 0);
636 chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize,
637 chroma_height, 5, weight0, weight1, 0);
638 chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize,
639 chroma_height, 5, weight0, weight1, 0);
641 luma_weight_avg(dest_y, tmp_y, h->mb_linesize, height, h->luma_log2_weight_denom,
642 h->luma_weight[refn0][0][0] , h->luma_weight[refn1][1][0],
643 h->luma_weight[refn0][0][1] + h->luma_weight[refn1][1][1]);
644 chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, chroma_height, h->chroma_log2_weight_denom,
645 h->chroma_weight[refn0][0][0][0] , h->chroma_weight[refn1][1][0][0],
646 h->chroma_weight[refn0][0][0][1] + h->chroma_weight[refn1][1][0][1]);
647 chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, chroma_height, h->chroma_log2_weight_denom,
648 h->chroma_weight[refn0][0][1][0] , h->chroma_weight[refn1][1][1][0],
649 h->chroma_weight[refn0][0][1][1] + h->chroma_weight[refn1][1][1][1]);
652 int list = list1 ? 1 : 0;
653 int refn = h->ref_cache[list][ scan8[n] ];
654 Picture *ref= &h->ref_list[list][refn];
655 mc_dir_part(h, ref, n, square, height, delta, list,
656 dest_y, dest_cb, dest_cr, x_offset, y_offset,
657 qpix_put, chroma_put, pixel_shift, chroma_idc);
659 luma_weight_op(dest_y, h->mb_linesize, height, h->luma_log2_weight_denom,
660 h->luma_weight[refn][list][0], h->luma_weight[refn][list][1]);
661 if(h->use_weight_chroma){
662 chroma_weight_op(dest_cb, h->mb_uvlinesize, chroma_height, h->chroma_log2_weight_denom,
663 h->chroma_weight[refn][list][0][0], h->chroma_weight[refn][list][0][1]);
664 chroma_weight_op(dest_cr, h->mb_uvlinesize, chroma_height, h->chroma_log2_weight_denom,
665 h->chroma_weight[refn][list][1][0], h->chroma_weight[refn][list][1][1]);
670 static av_always_inline void
671 mc_part(H264Context *h, int n, int square, int height, int delta,
672 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
673 int x_offset, int y_offset,
674 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
675 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
676 h264_weight_func *weight_op, h264_biweight_func *weight_avg,
677 int list0, int list1, int pixel_shift, int chroma_idc)
679 if((h->use_weight==2 && list0 && list1
680 && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ][h->s.mb_y&1] != 32))
682 mc_part_weighted(h, n, square, height, delta, dest_y, dest_cb, dest_cr,
683 x_offset, y_offset, qpix_put, chroma_put,
684 weight_op[0], weight_op[1], weight_avg[0],
685 weight_avg[1], list0, list1, pixel_shift, chroma_idc);
687 mc_part_std(h, n, square, height, delta, dest_y, dest_cb, dest_cr,
688 x_offset, y_offset, qpix_put, chroma_put, qpix_avg,
689 chroma_avg, list0, list1, pixel_shift, chroma_idc);
692 static av_always_inline void
693 prefetch_motion(H264Context *h, int list, int pixel_shift, int chroma_idc)
695 /* fetch pixels for estimated mv 4 macroblocks ahead
696 * optimized for 64byte cache lines */
697 MpegEncContext * const s = &h->s;
698 const int refn = h->ref_cache[list][scan8[0]];
700 const int mx= (h->mv_cache[list][scan8[0]][0]>>2) + 16*s->mb_x + 8;
701 const int my= (h->mv_cache[list][scan8[0]][1]>>2) + 16*s->mb_y;
702 uint8_t **src = h->ref_list[list][refn].f.data;
703 int off= (mx << pixel_shift) + (my + (s->mb_x&3)*4)*h->mb_linesize + (64 << pixel_shift);
704 s->dsp.prefetch(src[0]+off, s->linesize, 4);
705 if (chroma_idc == 3 /* yuv444 */) {
706 s->dsp.prefetch(src[1]+off, s->linesize, 4);
707 s->dsp.prefetch(src[2]+off, s->linesize, 4);
709 off= ((mx>>1) << pixel_shift) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + (64 << pixel_shift);
710 s->dsp.prefetch(src[1]+off, src[2]-src[1], 2);
715 static av_always_inline void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
716 qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
717 qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
718 h264_weight_func *weight_op, h264_biweight_func *weight_avg,
719 int pixel_shift, int chroma_idc)
721 MpegEncContext * const s = &h->s;
722 const int mb_xy= h->mb_xy;
723 const int mb_type = s->current_picture.f.mb_type[mb_xy];
725 assert(IS_INTER(mb_type));
727 if(HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME))
729 prefetch_motion(h, 0, pixel_shift, chroma_idc);
731 if(IS_16X16(mb_type)){
732 mc_part(h, 0, 1, 16, 0, dest_y, dest_cb, dest_cr, 0, 0,
733 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
734 weight_op, weight_avg,
735 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1),
736 pixel_shift, chroma_idc);
737 }else if(IS_16X8(mb_type)){
738 mc_part(h, 0, 0, 8, 8 << pixel_shift, dest_y, dest_cb, dest_cr, 0, 0,
739 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
740 weight_op, weight_avg,
741 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1),
742 pixel_shift, chroma_idc);
743 mc_part(h, 8, 0, 8, 8 << pixel_shift, dest_y, dest_cb, dest_cr, 0, 4,
744 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
745 weight_op, weight_avg,
746 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1),
747 pixel_shift, chroma_idc);
748 }else if(IS_8X16(mb_type)){
749 mc_part(h, 0, 0, 16, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 0, 0,
750 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
751 &weight_op[1], &weight_avg[1],
752 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1),
753 pixel_shift, chroma_idc);
754 mc_part(h, 4, 0, 16, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 4, 0,
755 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
756 &weight_op[1], &weight_avg[1],
757 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1),
758 pixel_shift, chroma_idc);
762 assert(IS_8X8(mb_type));
765 const int sub_mb_type= h->sub_mb_type[i];
767 int x_offset= (i&1)<<2;
768 int y_offset= (i&2)<<1;
770 if(IS_SUB_8X8(sub_mb_type)){
771 mc_part(h, n, 1, 8, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
772 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
773 &weight_op[1], &weight_avg[1],
774 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
775 pixel_shift, chroma_idc);
776 }else if(IS_SUB_8X4(sub_mb_type)){
777 mc_part(h, n , 0, 4, 4 << pixel_shift, dest_y, dest_cb, dest_cr, x_offset, y_offset,
778 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
779 &weight_op[1], &weight_avg[1],
780 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
781 pixel_shift, chroma_idc);
782 mc_part(h, n+2, 0, 4, 4 << pixel_shift, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
783 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
784 &weight_op[1], &weight_avg[1],
785 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
786 pixel_shift, chroma_idc);
787 }else if(IS_SUB_4X8(sub_mb_type)){
788 mc_part(h, n , 0, 8, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
789 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
790 &weight_op[2], &weight_avg[2],
791 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
792 pixel_shift, chroma_idc);
793 mc_part(h, n+1, 0, 8, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
794 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
795 &weight_op[2], &weight_avg[2],
796 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
797 pixel_shift, chroma_idc);
800 assert(IS_SUB_4X4(sub_mb_type));
802 int sub_x_offset= x_offset + 2*(j&1);
803 int sub_y_offset= y_offset + (j&2);
804 mc_part(h, n+j, 1, 4, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
805 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
806 &weight_op[2], &weight_avg[2],
807 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
808 pixel_shift, chroma_idc);
814 prefetch_motion(h, 1, pixel_shift, chroma_idc);
817 static av_always_inline void
818 hl_motion_420(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
819 qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
820 qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
821 h264_weight_func *weight_op, h264_biweight_func *weight_avg,
824 hl_motion(h, dest_y, dest_cb, dest_cr, qpix_put, chroma_put,
825 qpix_avg, chroma_avg, weight_op, weight_avg, pixel_shift, 1);
828 static av_always_inline void
829 hl_motion_422(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
830 qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
831 qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
832 h264_weight_func *weight_op, h264_biweight_func *weight_avg,
835 hl_motion(h, dest_y, dest_cb, dest_cr, qpix_put, chroma_put,
836 qpix_avg, chroma_avg, weight_op, weight_avg, pixel_shift, 2);
839 static void free_tables(H264Context *h, int free_rbsp){
843 av_freep(&h->intra4x4_pred_mode);
844 av_freep(&h->chroma_pred_mode_table);
845 av_freep(&h->cbp_table);
846 av_freep(&h->mvd_table[0]);
847 av_freep(&h->mvd_table[1]);
848 av_freep(&h->direct_table);
849 av_freep(&h->non_zero_count);
850 av_freep(&h->slice_table_base);
851 h->slice_table= NULL;
852 av_freep(&h->list_counts);
854 av_freep(&h->mb2b_xy);
855 av_freep(&h->mb2br_xy);
857 for(i = 0; i < MAX_THREADS; i++) {
858 hx = h->thread_context[i];
860 av_freep(&hx->top_borders[1]);
861 av_freep(&hx->top_borders[0]);
862 av_freep(&hx->s.obmc_scratchpad);
864 av_freep(&hx->rbsp_buffer[1]);
865 av_freep(&hx->rbsp_buffer[0]);
866 hx->rbsp_buffer_size[0] = 0;
867 hx->rbsp_buffer_size[1] = 0;
869 if (i) av_freep(&h->thread_context[i]);
873 static void init_dequant8_coeff_table(H264Context *h){
875 const int max_qp = 51 + 6*(h->sps.bit_depth_luma-8);
878 h->dequant8_coeff[i] = h->dequant8_buffer[i];
880 if(!memcmp(h->pps.scaling_matrix8[j], h->pps.scaling_matrix8[i], 64*sizeof(uint8_t))){
881 h->dequant8_coeff[i] = h->dequant8_buffer[j];
888 for(q=0; q<max_qp+1; q++){
892 h->dequant8_coeff[i][q][(x>>3)|((x&7)<<3)] =
893 ((uint32_t)dequant8_coeff_init[idx][ dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] *
894 h->pps.scaling_matrix8[i][x]) << shift;
899 static void init_dequant4_coeff_table(H264Context *h){
901 const int max_qp = 51 + 6*(h->sps.bit_depth_luma-8);
903 h->dequant4_coeff[i] = h->dequant4_buffer[i];
905 if(!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i], 16*sizeof(uint8_t))){
906 h->dequant4_coeff[i] = h->dequant4_buffer[j];
913 for(q=0; q<max_qp+1; q++){
914 int shift = div6[q] + 2;
917 h->dequant4_coeff[i][q][(x>>2)|((x<<2)&0xF)] =
918 ((uint32_t)dequant4_coeff_init[idx][(x&1) + ((x>>2)&1)] *
919 h->pps.scaling_matrix4[i][x]) << shift;
924 static void init_dequant_tables(H264Context *h){
926 init_dequant4_coeff_table(h);
927 if(h->pps.transform_8x8_mode)
928 init_dequant8_coeff_table(h);
929 if(h->sps.transform_bypass){
932 h->dequant4_coeff[i][0][x] = 1<<6;
933 if(h->pps.transform_8x8_mode)
936 h->dequant8_coeff[i][0][x] = 1<<6;
941 int ff_h264_alloc_tables(H264Context *h){
942 MpegEncContext * const s = &h->s;
943 const int big_mb_num= s->mb_stride * (s->mb_height+1);
944 const int row_mb_num= 2*s->mb_stride*s->avctx->thread_count;
947 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->intra4x4_pred_mode, row_mb_num * 8 * sizeof(uint8_t), fail)
949 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->non_zero_count , big_mb_num * 48 * sizeof(uint8_t), fail)
950 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->slice_table_base , (big_mb_num+s->mb_stride) * sizeof(*h->slice_table_base), fail)
951 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->cbp_table, big_mb_num * sizeof(uint16_t), fail)
953 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t), fail)
954 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[0], 16*row_mb_num * sizeof(uint8_t), fail);
955 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[1], 16*row_mb_num * sizeof(uint8_t), fail);
956 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->direct_table, 4*big_mb_num * sizeof(uint8_t) , fail);
957 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->list_counts, big_mb_num * sizeof(uint8_t), fail)
959 memset(h->slice_table_base, -1, (big_mb_num+s->mb_stride) * sizeof(*h->slice_table_base));
960 h->slice_table= h->slice_table_base + s->mb_stride*2 + 1;
962 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2b_xy , big_mb_num * sizeof(uint32_t), fail);
963 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2br_xy , big_mb_num * sizeof(uint32_t), fail);
964 for(y=0; y<s->mb_height; y++){
965 for(x=0; x<s->mb_width; x++){
966 const int mb_xy= x + y*s->mb_stride;
967 const int b_xy = 4*x + 4*y*h->b_stride;
969 h->mb2b_xy [mb_xy]= b_xy;
970 h->mb2br_xy[mb_xy]= 8*(FMO ? mb_xy : (mb_xy % (2*s->mb_stride)));
974 s->obmc_scratchpad = NULL;
976 if(!h->dequant4_coeff[0])
977 init_dequant_tables(h);
986 * Mimic alloc_tables(), but for every context thread.
988 static void clone_tables(H264Context *dst, H264Context *src, int i){
989 MpegEncContext * const s = &src->s;
990 dst->intra4x4_pred_mode = src->intra4x4_pred_mode + i*8*2*s->mb_stride;
991 dst->non_zero_count = src->non_zero_count;
992 dst->slice_table = src->slice_table;
993 dst->cbp_table = src->cbp_table;
994 dst->mb2b_xy = src->mb2b_xy;
995 dst->mb2br_xy = src->mb2br_xy;
996 dst->chroma_pred_mode_table = src->chroma_pred_mode_table;
997 dst->mvd_table[0] = src->mvd_table[0] + i*8*2*s->mb_stride;
998 dst->mvd_table[1] = src->mvd_table[1] + i*8*2*s->mb_stride;
999 dst->direct_table = src->direct_table;
1000 dst->list_counts = src->list_counts;
1002 dst->s.obmc_scratchpad = NULL;
1003 ff_h264_pred_init(&dst->hpc, src->s.codec_id, src->sps.bit_depth_luma, src->sps.chroma_format_idc);
1008 * Allocate buffers which are not shared amongst multiple threads.
1010 static int context_init(H264Context *h){
1011 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[0], h->s.mb_width * 16*3 * sizeof(uint8_t)*2, fail)
1012 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[1], h->s.mb_width * 16*3 * sizeof(uint8_t)*2, fail)
1014 h->ref_cache[0][scan8[5 ]+1] = h->ref_cache[0][scan8[7 ]+1] = h->ref_cache[0][scan8[13]+1] =
1015 h->ref_cache[1][scan8[5 ]+1] = h->ref_cache[1][scan8[7 ]+1] = h->ref_cache[1][scan8[13]+1] = PART_NOT_AVAILABLE;
1019 return -1; // free_tables will clean up for us
1022 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size);
1024 static av_cold void common_init(H264Context *h){
1025 MpegEncContext * const s = &h->s;
1027 s->width = s->avctx->width;
1028 s->height = s->avctx->height;
1029 s->codec_id= s->avctx->codec->id;
1031 ff_h264dsp_init(&h->h264dsp, 8, 1);
1032 ff_h264_pred_init(&h->hpc, s->codec_id, 8, 1);
1034 h->dequant_coeff_pps= -1;
1035 s->unrestricted_mv=1;
1037 dsputil_init(&s->dsp, s->avctx); // needed so that idct permutation is known early
1039 memset(h->pps.scaling_matrix4, 16, 6*16*sizeof(uint8_t));
1040 memset(h->pps.scaling_matrix8, 16, 2*64*sizeof(uint8_t));
1043 int ff_h264_decode_extradata(H264Context *h)
1045 AVCodecContext *avctx = h->s.avctx;
1047 if(avctx->extradata[0] == 1){
1048 int i, cnt, nalsize;
1049 unsigned char *p = avctx->extradata;
1053 if(avctx->extradata_size < 7) {
1054 av_log(avctx, AV_LOG_ERROR, "avcC too short\n");
1057 /* sps and pps in the avcC always have length coded with 2 bytes,
1058 so put a fake nal_length_size = 2 while parsing them */
1059 h->nal_length_size = 2;
1060 // Decode sps from avcC
1061 cnt = *(p+5) & 0x1f; // Number of sps
1063 for (i = 0; i < cnt; i++) {
1064 nalsize = AV_RB16(p) + 2;
1065 if (p - avctx->extradata + nalsize > avctx->extradata_size)
1067 if(decode_nal_units(h, p, nalsize) < 0) {
1068 av_log(avctx, AV_LOG_ERROR, "Decoding sps %d from avcC failed\n", i);
1073 // Decode pps from avcC
1074 cnt = *(p++); // Number of pps
1075 for (i = 0; i < cnt; i++) {
1076 nalsize = AV_RB16(p) + 2;
1077 if (p - avctx->extradata + nalsize > avctx->extradata_size)
1079 if (decode_nal_units(h, p, nalsize) < 0) {
1080 av_log(avctx, AV_LOG_ERROR, "Decoding pps %d from avcC failed\n", i);
1085 // Now store right nal length size, that will be use to parse all other nals
1086 h->nal_length_size = (avctx->extradata[4] & 0x03) + 1;
1089 if(decode_nal_units(h, avctx->extradata, avctx->extradata_size) < 0)
1095 av_cold int ff_h264_decode_init(AVCodecContext *avctx){
1096 H264Context *h= avctx->priv_data;
1097 MpegEncContext * const s = &h->s;
1100 MPV_decode_defaults(s);
1105 s->out_format = FMT_H264;
1106 s->workaround_bugs= avctx->workaround_bugs;
1109 // s->decode_mb= ff_h263_decode_mb;
1110 s->quarter_sample = 1;
1111 if(!avctx->has_b_frames)
1114 avctx->chroma_sample_location = AVCHROMA_LOC_LEFT;
1116 ff_h264_decode_init_vlc();
1119 h->sps.bit_depth_luma = avctx->bits_per_raw_sample = 8;
1121 h->thread_context[0] = h;
1122 h->outputed_poc = h->next_outputed_poc = INT_MIN;
1123 for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
1124 h->last_pocs[i] = INT_MIN;
1125 h->prev_poc_msb= 1<<16;
1127 ff_h264_reset_sei(h);
1128 if(avctx->codec_id == CODEC_ID_H264){
1129 if(avctx->ticks_per_frame == 1){
1130 s->avctx->time_base.den *=2;
1132 avctx->ticks_per_frame = 2;
1135 if(avctx->extradata_size > 0 && avctx->extradata &&
1136 ff_h264_decode_extradata(h))
1139 if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames < h->sps.num_reorder_frames){
1140 s->avctx->has_b_frames = h->sps.num_reorder_frames;
1147 #define IN_RANGE(a, b, size) (((a) >= (b)) && ((a) < ((b)+(size))))
1148 static void copy_picture_range(Picture **to, Picture **from, int count, MpegEncContext *new_base, MpegEncContext *old_base)
1152 for (i=0; i<count; i++){
1153 assert((IN_RANGE(from[i], old_base, sizeof(*old_base)) ||
1154 IN_RANGE(from[i], old_base->picture, sizeof(Picture) * old_base->picture_count) ||
1156 to[i] = REBASE_PICTURE(from[i], new_base, old_base);
1160 static void copy_parameter_set(void **to, void **from, int count, int size)
1164 for (i=0; i<count; i++){
1165 if (to[i] && !from[i]) av_freep(&to[i]);
1166 else if (from[i] && !to[i]) to[i] = av_malloc(size);
1168 if (from[i]) memcpy(to[i], from[i], size);
1172 static int decode_init_thread_copy(AVCodecContext *avctx){
1173 H264Context *h= avctx->priv_data;
1175 if (!avctx->is_copy) return 0;
1176 memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
1177 memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
1182 #define copy_fields(to, from, start_field, end_field) memcpy(&to->start_field, &from->start_field, (char*)&to->end_field - (char*)&to->start_field)
1183 static int decode_update_thread_context(AVCodecContext *dst, const AVCodecContext *src){
1184 H264Context *h= dst->priv_data, *h1= src->priv_data;
1185 MpegEncContext * const s = &h->s, * const s1 = &h1->s;
1186 int inited = s->context_initialized, err;
1189 if(dst == src || !s1->context_initialized) return 0;
1191 err = ff_mpeg_update_thread_context(dst, src);
1194 //FIXME handle width/height changing
1196 for(i = 0; i < MAX_SPS_COUNT; i++)
1197 av_freep(h->sps_buffers + i);
1199 for(i = 0; i < MAX_PPS_COUNT; i++)
1200 av_freep(h->pps_buffers + i);
1202 memcpy(&h->s + 1, &h1->s + 1, sizeof(H264Context) - sizeof(MpegEncContext)); //copy all fields after MpegEnc
1203 memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
1204 memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
1205 if (ff_h264_alloc_tables(h) < 0) {
1206 av_log(dst, AV_LOG_ERROR, "Could not allocate memory for h264\n");
1207 return AVERROR(ENOMEM);
1212 h->rbsp_buffer[i] = NULL;
1213 h->rbsp_buffer_size[i] = 0;
1216 h->thread_context[0] = h;
1218 // frame_start may not be called for the next thread (if it's decoding a bottom field)
1219 // so this has to be allocated here
1220 h->s.obmc_scratchpad = av_malloc(16*6*s->linesize);
1222 s->dsp.clear_blocks(h->mb);
1223 s->dsp.clear_blocks(h->mb+(24*16<<h->pixel_shift));
1226 //extradata/NAL handling
1227 h->is_avc = h1->is_avc;
1230 copy_parameter_set((void**)h->sps_buffers, (void**)h1->sps_buffers, MAX_SPS_COUNT, sizeof(SPS));
1232 copy_parameter_set((void**)h->pps_buffers, (void**)h1->pps_buffers, MAX_PPS_COUNT, sizeof(PPS));
1235 //Dequantization matrices
1236 //FIXME these are big - can they be only copied when PPS changes?
1237 copy_fields(h, h1, dequant4_buffer, dequant4_coeff);
1240 h->dequant4_coeff[i] = h->dequant4_buffer[0] + (h1->dequant4_coeff[i] - h1->dequant4_buffer[0]);
1243 h->dequant8_coeff[i] = h->dequant8_buffer[0] + (h1->dequant8_coeff[i] - h1->dequant8_buffer[0]);
1245 h->dequant_coeff_pps = h1->dequant_coeff_pps;
1248 copy_fields(h, h1, poc_lsb, redundant_pic_count);
1251 copy_fields(h, h1, ref_count, list_count);
1252 copy_fields(h, h1, ref_list, intra_gb);
1253 copy_fields(h, h1, short_ref, cabac_init_idc);
1255 copy_picture_range(h->short_ref, h1->short_ref, 32, s, s1);
1256 copy_picture_range(h->long_ref, h1->long_ref, 32, s, s1);
1257 copy_picture_range(h->delayed_pic, h1->delayed_pic, MAX_DELAYED_PIC_COUNT+2, s, s1);
1259 h->last_slice_type = h1->last_slice_type;
1261 if(!s->current_picture_ptr) return 0;
1264 err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1265 h->prev_poc_msb = h->poc_msb;
1266 h->prev_poc_lsb = h->poc_lsb;
1268 h->prev_frame_num_offset= h->frame_num_offset;
1269 h->prev_frame_num = h->frame_num;
1270 h->outputed_poc = h->next_outputed_poc;
1275 int ff_h264_frame_start(H264Context *h){
1276 MpegEncContext * const s = &h->s;
1278 const int pixel_shift = h->pixel_shift;
1279 int thread_count = (s->avctx->active_thread_type & FF_THREAD_SLICE) ? s->avctx->thread_count : 1;
1281 if(MPV_frame_start(s, s->avctx) < 0)
1283 ff_er_frame_start(s);
1285 * MPV_frame_start uses pict_type to derive key_frame.
1286 * This is incorrect for H.264; IDR markings must be used.
1287 * Zero here; IDR markings per slice in frame or fields are ORed in later.
1288 * See decode_nal_units().
1290 s->current_picture_ptr->f.key_frame = 0;
1291 s->current_picture_ptr->mmco_reset= 0;
1293 assert(s->linesize && s->uvlinesize);
1295 for(i=0; i<16; i++){
1296 h->block_offset[i]= (4*((scan8[i] - scan8[0])&7) << pixel_shift) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
1297 h->block_offset[48+i]= (4*((scan8[i] - scan8[0])&7) << pixel_shift) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
1299 for(i=0; i<16; i++){
1300 h->block_offset[16+i]=
1301 h->block_offset[32+i]= (4*((scan8[i] - scan8[0])&7) << pixel_shift) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
1302 h->block_offset[48+16+i]=
1303 h->block_offset[48+32+i]= (4*((scan8[i] - scan8[0])&7) << pixel_shift) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
1306 /* can't be in alloc_tables because linesize isn't known there.
1307 * FIXME: redo bipred weight to not require extra buffer? */
1308 for(i = 0; i < thread_count; i++)
1309 if(h->thread_context[i] && !h->thread_context[i]->s.obmc_scratchpad)
1310 h->thread_context[i]->s.obmc_scratchpad = av_malloc(16*6*s->linesize);
1312 /* some macroblocks can be accessed before they're available in case of lost slices, mbaff or threading*/
1313 memset(h->slice_table, -1, (s->mb_height*s->mb_stride-1) * sizeof(*h->slice_table));
1315 // s->decode = (s->flags & CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.f.reference /*|| h->contains_intra*/ || 1;
1317 // We mark the current picture as non-reference after allocating it, so
1318 // that if we break out due to an error it can be released automatically
1319 // in the next MPV_frame_start().
1320 // SVQ3 as well as most other codecs have only last/next/current and thus
1321 // get released even with set reference, besides SVQ3 and others do not
1322 // mark frames as reference later "naturally".
1323 if(s->codec_id != CODEC_ID_SVQ3)
1324 s->current_picture_ptr->f.reference = 0;
1326 s->current_picture_ptr->field_poc[0]=
1327 s->current_picture_ptr->field_poc[1]= INT_MAX;
1329 h->next_output_pic = NULL;
1331 assert(s->current_picture_ptr->long_ref==0);
1337 * Run setup operations that must be run after slice header decoding.
1338 * This includes finding the next displayed frame.
1340 * @param h h264 master context
1341 * @param setup_finished enough NALs have been read that we can call
1342 * ff_thread_finish_setup()
1344 static void decode_postinit(H264Context *h, int setup_finished){
1345 MpegEncContext * const s = &h->s;
1346 Picture *out = s->current_picture_ptr;
1347 Picture *cur = s->current_picture_ptr;
1348 int i, pics, out_of_order, out_idx;
1350 s->current_picture_ptr->f.qscale_type = FF_QSCALE_TYPE_H264;
1351 s->current_picture_ptr->f.pict_type = s->pict_type;
1353 if (h->next_output_pic) return;
1355 if (cur->field_poc[0]==INT_MAX || cur->field_poc[1]==INT_MAX) {
1356 //FIXME: if we have two PAFF fields in one packet, we can't start the next thread here.
1357 //If we have one field per packet, we can. The check in decode_nal_units() is not good enough
1358 //to find this yet, so we assume the worst for now.
1359 //if (setup_finished)
1360 // ff_thread_finish_setup(s->avctx);
1364 cur->f.interlaced_frame = 0;
1365 cur->f.repeat_pict = 0;
1367 /* Signal interlacing information externally. */
1368 /* Prioritize picture timing SEI information over used decoding process if it exists. */
1370 if(h->sps.pic_struct_present_flag){
1371 switch (h->sei_pic_struct)
1373 case SEI_PIC_STRUCT_FRAME:
1375 case SEI_PIC_STRUCT_TOP_FIELD:
1376 case SEI_PIC_STRUCT_BOTTOM_FIELD:
1377 cur->f.interlaced_frame = 1;
1379 case SEI_PIC_STRUCT_TOP_BOTTOM:
1380 case SEI_PIC_STRUCT_BOTTOM_TOP:
1381 if (FIELD_OR_MBAFF_PICTURE)
1382 cur->f.interlaced_frame = 1;
1384 // try to flag soft telecine progressive
1385 cur->f.interlaced_frame = h->prev_interlaced_frame;
1387 case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
1388 case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
1389 // Signal the possibility of telecined film externally (pic_struct 5,6)
1390 // From these hints, let the applications decide if they apply deinterlacing.
1391 cur->f.repeat_pict = 1;
1393 case SEI_PIC_STRUCT_FRAME_DOUBLING:
1394 // Force progressive here, as doubling interlaced frame is a bad idea.
1395 cur->f.repeat_pict = 2;
1397 case SEI_PIC_STRUCT_FRAME_TRIPLING:
1398 cur->f.repeat_pict = 4;
1402 if ((h->sei_ct_type & 3) && h->sei_pic_struct <= SEI_PIC_STRUCT_BOTTOM_TOP)
1403 cur->f.interlaced_frame = (h->sei_ct_type & (1 << 1)) != 0;
1405 /* Derive interlacing flag from used decoding process. */
1406 cur->f.interlaced_frame = FIELD_OR_MBAFF_PICTURE;
1408 h->prev_interlaced_frame = cur->f.interlaced_frame;
1410 if (cur->field_poc[0] != cur->field_poc[1]){
1411 /* Derive top_field_first from field pocs. */
1412 cur->f.top_field_first = cur->field_poc[0] < cur->field_poc[1];
1414 if (cur->f.interlaced_frame || h->sps.pic_struct_present_flag) {
1415 /* Use picture timing SEI information. Even if it is a information of a past frame, better than nothing. */
1416 if(h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM
1417 || h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM_TOP)
1418 cur->f.top_field_first = 1;
1420 cur->f.top_field_first = 0;
1422 /* Most likely progressive */
1423 cur->f.top_field_first = 0;
1427 //FIXME do something with unavailable reference frames
1429 /* Sort B-frames into display order */
1431 if(h->sps.bitstream_restriction_flag
1432 && s->avctx->has_b_frames < h->sps.num_reorder_frames){
1433 s->avctx->has_b_frames = h->sps.num_reorder_frames;
1437 if( s->avctx->strict_std_compliance >= FF_COMPLIANCE_STRICT
1438 && !h->sps.bitstream_restriction_flag){
1439 s->avctx->has_b_frames= MAX_DELAYED_PIC_COUNT;
1444 while(h->delayed_pic[pics]) pics++;
1446 assert(pics <= MAX_DELAYED_PIC_COUNT);
1448 h->delayed_pic[pics++] = cur;
1449 if (cur->f.reference == 0)
1450 cur->f.reference = DELAYED_PIC_REF;
1452 out = h->delayed_pic[0];
1454 for (i = 1; h->delayed_pic[i] && !h->delayed_pic[i]->f.key_frame && !h->delayed_pic[i]->mmco_reset; i++)
1455 if(h->delayed_pic[i]->poc < out->poc){
1456 out = h->delayed_pic[i];
1459 if (s->avctx->has_b_frames == 0 && (h->delayed_pic[0]->f.key_frame || h->delayed_pic[0]->mmco_reset))
1460 h->next_outputed_poc= INT_MIN;
1461 out_of_order = out->poc < h->next_outputed_poc;
1463 if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames >= h->sps.num_reorder_frames)
1465 else if (out_of_order && pics-1 == s->avctx->has_b_frames &&
1466 s->avctx->has_b_frames < MAX_DELAYED_PIC_COUNT) {
1467 int cnt = 0, invalid = 0;
1468 for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++) {
1469 cnt += out->poc < h->last_pocs[i];
1470 invalid += h->last_pocs[i] == INT_MIN;
1472 if (invalid + cnt < MAX_DELAYED_PIC_COUNT) {
1473 s->avctx->has_b_frames = FFMAX(s->avctx->has_b_frames, cnt);
1475 for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
1476 h->last_pocs[i] = INT_MIN;
1479 } else if (s->low_delay &&
1480 ((h->next_outputed_poc != INT_MIN && out->poc > h->next_outputed_poc + 2) ||
1481 cur->f.pict_type == AV_PICTURE_TYPE_B)) {
1483 s->avctx->has_b_frames++;
1486 if(out_of_order || pics > s->avctx->has_b_frames){
1487 out->f.reference &= ~DELAYED_PIC_REF;
1488 out->owner2 = s; // for frame threading, the owner must be the second field's thread
1489 // or else the first thread can release the picture and reuse it unsafely
1490 for(i=out_idx; h->delayed_pic[i]; i++)
1491 h->delayed_pic[i] = h->delayed_pic[i+1];
1493 memmove(h->last_pocs, &h->last_pocs[1], sizeof(*h->last_pocs) * (MAX_DELAYED_PIC_COUNT - 1));
1494 h->last_pocs[MAX_DELAYED_PIC_COUNT - 1] = out->poc;
1495 if(!out_of_order && pics > s->avctx->has_b_frames){
1496 h->next_output_pic = out;
1497 if (out_idx == 0 && h->delayed_pic[0] && (h->delayed_pic[0]->f.key_frame || h->delayed_pic[0]->mmco_reset)) {
1498 h->next_outputed_poc = INT_MIN;
1500 h->next_outputed_poc = out->poc;
1502 av_log(s->avctx, AV_LOG_DEBUG, "no picture\n");
1506 ff_thread_finish_setup(s->avctx);
1509 static av_always_inline void backup_mb_border(H264Context *h, uint8_t *src_y,
1510 uint8_t *src_cb, uint8_t *src_cr,
1511 int linesize, int uvlinesize, int simple)
1513 MpegEncContext * const s = &h->s;
1514 uint8_t *top_border;
1516 const int pixel_shift = h->pixel_shift;
1517 int chroma444 = CHROMA444;
1518 int chroma422 = CHROMA422;
1521 src_cb -= uvlinesize;
1522 src_cr -= uvlinesize;
1524 if(!simple && FRAME_MBAFF){
1527 top_border = h->top_borders[0][s->mb_x];
1528 AV_COPY128(top_border, src_y + 15*linesize);
1530 AV_COPY128(top_border+16, src_y+15*linesize+16);
1531 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1534 AV_COPY128(top_border+32, src_cb + 15*uvlinesize);
1535 AV_COPY128(top_border+48, src_cb + 15*uvlinesize+16);
1536 AV_COPY128(top_border+64, src_cr + 15*uvlinesize);
1537 AV_COPY128(top_border+80, src_cr + 15*uvlinesize+16);
1539 AV_COPY128(top_border+16, src_cb + 15*uvlinesize);
1540 AV_COPY128(top_border+32, src_cr + 15*uvlinesize);
1542 } else if(chroma422) {
1544 AV_COPY128(top_border+32, src_cb + 15*uvlinesize);
1545 AV_COPY128(top_border+48, src_cr + 15*uvlinesize);
1547 AV_COPY64(top_border+16, src_cb + 15*uvlinesize);
1548 AV_COPY64(top_border+24, src_cr + 15*uvlinesize);
1552 AV_COPY128(top_border+32, src_cb+7*uvlinesize);
1553 AV_COPY128(top_border+48, src_cr+7*uvlinesize);
1555 AV_COPY64(top_border+16, src_cb+7*uvlinesize);
1556 AV_COPY64(top_border+24, src_cr+7*uvlinesize);
1567 top_border = h->top_borders[top_idx][s->mb_x];
1568 // There are two lines saved, the line above the the top macroblock of a pair,
1569 // and the line above the bottom macroblock
1570 AV_COPY128(top_border, src_y + 16*linesize);
1572 AV_COPY128(top_border+16, src_y+16*linesize+16);
1574 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1577 AV_COPY128(top_border+32, src_cb + 16*linesize);
1578 AV_COPY128(top_border+48, src_cb + 16*linesize+16);
1579 AV_COPY128(top_border+64, src_cr + 16*linesize);
1580 AV_COPY128(top_border+80, src_cr + 16*linesize+16);
1582 AV_COPY128(top_border+16, src_cb + 16*linesize);
1583 AV_COPY128(top_border+32, src_cr + 16*linesize);
1585 } else if(chroma422) {
1587 AV_COPY128(top_border+32, src_cb+16*uvlinesize);
1588 AV_COPY128(top_border+48, src_cr+16*uvlinesize);
1590 AV_COPY64(top_border+16, src_cb+16*uvlinesize);
1591 AV_COPY64(top_border+24, src_cr+16*uvlinesize);
1595 AV_COPY128(top_border+32, src_cb+8*uvlinesize);
1596 AV_COPY128(top_border+48, src_cr+8*uvlinesize);
1598 AV_COPY64(top_border+16, src_cb+8*uvlinesize);
1599 AV_COPY64(top_border+24, src_cr+8*uvlinesize);
1605 static av_always_inline void xchg_mb_border(H264Context *h, uint8_t *src_y,
1606 uint8_t *src_cb, uint8_t *src_cr,
1607 int linesize, int uvlinesize,
1608 int xchg, int chroma444,
1609 int simple, int pixel_shift){
1610 MpegEncContext * const s = &h->s;
1611 int deblock_topleft;
1614 uint8_t *top_border_m1;
1615 uint8_t *top_border;
1617 if(!simple && FRAME_MBAFF){
1622 top_idx = MB_MBAFF ? 0 : 1;
1626 if(h->deblocking_filter == 2) {
1627 deblock_topleft = h->slice_table[h->mb_xy - 1 - s->mb_stride] == h->slice_num;
1628 deblock_top = h->top_type;
1630 deblock_topleft = (s->mb_x > 0);
1631 deblock_top = (s->mb_y > !!MB_FIELD);
1634 src_y -= linesize + 1 + pixel_shift;
1635 src_cb -= uvlinesize + 1 + pixel_shift;
1636 src_cr -= uvlinesize + 1 + pixel_shift;
1638 top_border_m1 = h->top_borders[top_idx][s->mb_x-1];
1639 top_border = h->top_borders[top_idx][s->mb_x];
1641 #define XCHG(a,b,xchg)\
1644 AV_SWAP64(b+0,a+0);\
1645 AV_SWAP64(b+8,a+8);\
1650 if (xchg) AV_SWAP64(b,a);\
1651 else AV_COPY64(b,a);
1654 if(deblock_topleft){
1655 XCHG(top_border_m1 + (8 << pixel_shift), src_y - (7 << pixel_shift), 1);
1657 XCHG(top_border + (0 << pixel_shift), src_y + (1 << pixel_shift), xchg);
1658 XCHG(top_border + (8 << pixel_shift), src_y + (9 << pixel_shift), 1);
1659 if(s->mb_x+1 < s->mb_width){
1660 XCHG(h->top_borders[top_idx][s->mb_x+1], src_y + (17 << pixel_shift), 1);
1663 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1665 if(deblock_topleft){
1666 XCHG(top_border_m1 + (24 << pixel_shift), src_cb - (7 << pixel_shift), 1);
1667 XCHG(top_border_m1 + (40 << pixel_shift), src_cr - (7 << pixel_shift), 1);
1669 XCHG(top_border + (16 << pixel_shift), src_cb + (1 << pixel_shift), xchg);
1670 XCHG(top_border + (24 << pixel_shift), src_cb + (9 << pixel_shift), 1);
1671 XCHG(top_border + (32 << pixel_shift), src_cr + (1 << pixel_shift), xchg);
1672 XCHG(top_border + (40 << pixel_shift), src_cr + (9 << pixel_shift), 1);
1673 if(s->mb_x+1 < s->mb_width){
1674 XCHG(h->top_borders[top_idx][s->mb_x+1] + (16 << pixel_shift), src_cb + (17 << pixel_shift), 1);
1675 XCHG(h->top_borders[top_idx][s->mb_x+1] + (32 << pixel_shift), src_cr + (17 << pixel_shift), 1);
1679 if(deblock_topleft){
1680 XCHG(top_border_m1 + (16 << pixel_shift), src_cb - (7 << pixel_shift), 1);
1681 XCHG(top_border_m1 + (24 << pixel_shift), src_cr - (7 << pixel_shift), 1);
1683 XCHG(top_border + (16 << pixel_shift), src_cb+1+pixel_shift, 1);
1684 XCHG(top_border + (24 << pixel_shift), src_cr+1+pixel_shift, 1);
1690 static av_always_inline int dctcoef_get(DCTELEM *mb, int high_bit_depth, int index) {
1691 if (high_bit_depth) {
1692 return AV_RN32A(((int32_t*)mb) + index);
1694 return AV_RN16A(mb + index);
1697 static av_always_inline void dctcoef_set(DCTELEM *mb, int high_bit_depth, int index, int value) {
1698 if (high_bit_depth) {
1699 AV_WN32A(((int32_t*)mb) + index, value);
1701 AV_WN16A(mb + index, value);
1704 static av_always_inline void hl_decode_mb_predict_luma(H264Context *h, int mb_type, int is_h264, int simple, int transform_bypass,
1705 int pixel_shift, int *block_offset, int linesize, uint8_t *dest_y, int p)
1707 MpegEncContext * const s = &h->s;
1708 void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1709 void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
1711 int qscale = p == 0 ? s->qscale : h->chroma_qp[p-1];
1712 block_offset += 16*p;
1713 if(IS_INTRA4x4(mb_type)){
1714 if(simple || !s->encoding){
1715 if(IS_8x8DCT(mb_type)){
1716 if(transform_bypass){
1718 idct_add = s->dsp.add_pixels8;
1720 idct_dc_add = h->h264dsp.h264_idct8_dc_add;
1721 idct_add = h->h264dsp.h264_idct8_add;
1723 for(i=0; i<16; i+=4){
1724 uint8_t * const ptr= dest_y + block_offset[i];
1725 const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1726 if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1727 h->hpc.pred8x8l_add[dir](ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1729 const int nnz = h->non_zero_count_cache[ scan8[i+p*16] ];
1730 h->hpc.pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
1731 (h->topright_samples_available<<i)&0x4000, linesize);
1733 if(nnz == 1 && dctcoef_get(h->mb, pixel_shift, i*16+p*256))
1734 idct_dc_add(ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1736 idct_add (ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1741 if(transform_bypass){
1743 idct_add = s->dsp.add_pixels4;
1745 idct_dc_add = h->h264dsp.h264_idct_dc_add;
1746 idct_add = h->h264dsp.h264_idct_add;
1748 for(i=0; i<16; i++){
1749 uint8_t * const ptr= dest_y + block_offset[i];
1750 const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1752 if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1753 h->hpc.pred4x4_add[dir](ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1758 if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
1759 const int topright_avail= (h->topright_samples_available<<i)&0x8000;
1760 assert(s->mb_y || linesize <= block_offset[i]);
1761 if(!topright_avail){
1763 tr_high= ((uint16_t*)ptr)[3 - linesize/2]*0x0001000100010001ULL;
1764 topright= (uint8_t*) &tr_high;
1766 tr= ptr[3 - linesize]*0x01010101u;
1767 topright= (uint8_t*) &tr;
1770 topright= ptr + (4 << pixel_shift) - linesize;
1774 h->hpc.pred4x4[ dir ](ptr, topright, linesize);
1775 nnz = h->non_zero_count_cache[ scan8[i+p*16] ];
1778 if(nnz == 1 && dctcoef_get(h->mb, pixel_shift, i*16+p*256))
1779 idct_dc_add(ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1781 idct_add (ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1783 ff_svq3_add_idct_c(ptr, h->mb + i*16+p*256, linesize, qscale, 0);
1790 h->hpc.pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
1792 if(h->non_zero_count_cache[ scan8[LUMA_DC_BLOCK_INDEX+p] ]){
1793 if(!transform_bypass)
1794 h->h264dsp.h264_luma_dc_dequant_idct(h->mb+(p*256 << pixel_shift), h->mb_luma_dc[p], h->dequant4_coeff[p][qscale][0]);
1796 static const uint8_t dc_mapping[16] = { 0*16, 1*16, 4*16, 5*16, 2*16, 3*16, 6*16, 7*16,
1797 8*16, 9*16,12*16,13*16,10*16,11*16,14*16,15*16};
1798 for(i = 0; i < 16; i++)
1799 dctcoef_set(h->mb+(p*256 << pixel_shift), pixel_shift, dc_mapping[i], dctcoef_get(h->mb_luma_dc[p], pixel_shift, i));
1803 ff_svq3_luma_dc_dequant_idct_c(h->mb+p*256, h->mb_luma_dc[p], qscale);
1807 static av_always_inline void hl_decode_mb_idct_luma(H264Context *h, int mb_type, int is_h264, int simple, int transform_bypass,
1808 int pixel_shift, int *block_offset, int linesize, uint8_t *dest_y, int p)
1810 MpegEncContext * const s = &h->s;
1811 void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1813 block_offset += 16*p;
1814 if(!IS_INTRA4x4(mb_type)){
1816 if(IS_INTRA16x16(mb_type)){
1817 if(transform_bypass){
1818 if(h->sps.profile_idc==244 && (h->intra16x16_pred_mode==VERT_PRED8x8 || h->intra16x16_pred_mode==HOR_PRED8x8)){
1819 h->hpc.pred16x16_add[h->intra16x16_pred_mode](dest_y, block_offset, h->mb + (p*256 << pixel_shift), linesize);
1821 for(i=0; i<16; i++){
1822 if(h->non_zero_count_cache[ scan8[i+p*16] ] || dctcoef_get(h->mb, pixel_shift, i*16+p*256))
1823 s->dsp.add_pixels4(dest_y + block_offset[i], h->mb + (i*16+p*256 << pixel_shift), linesize);
1827 h->h264dsp.h264_idct_add16intra(dest_y, block_offset, h->mb + (p*256 << pixel_shift), linesize, h->non_zero_count_cache+p*5*8);
1829 }else if(h->cbp&15){
1830 if(transform_bypass){
1831 const int di = IS_8x8DCT(mb_type) ? 4 : 1;
1832 idct_add= IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4;
1833 for(i=0; i<16; i+=di){
1834 if(h->non_zero_count_cache[ scan8[i+p*16] ]){
1835 idct_add(dest_y + block_offset[i], h->mb + (i*16+p*256 << pixel_shift), linesize);
1839 if(IS_8x8DCT(mb_type)){
1840 h->h264dsp.h264_idct8_add4(dest_y, block_offset, h->mb + (p*256 << pixel_shift), linesize, h->non_zero_count_cache+p*5*8);
1842 h->h264dsp.h264_idct_add16(dest_y, block_offset, h->mb + (p*256 << pixel_shift), linesize, h->non_zero_count_cache+p*5*8);
1847 for(i=0; i<16; i++){
1848 if(h->non_zero_count_cache[ scan8[i+p*16] ] || h->mb[i*16+p*256]){ //FIXME benchmark weird rule, & below
1849 uint8_t * const ptr= dest_y + block_offset[i];
1850 ff_svq3_add_idct_c(ptr, h->mb + i*16 + p*256, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
1857 static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple, int pixel_shift)
1859 MpegEncContext * const s = &h->s;
1860 const int mb_x= s->mb_x;
1861 const int mb_y= s->mb_y;
1862 const int mb_xy= h->mb_xy;
1863 const int mb_type = s->current_picture.f.mb_type[mb_xy];
1864 uint8_t *dest_y, *dest_cb, *dest_cr;
1865 int linesize, uvlinesize /*dct_offset*/;
1867 int *block_offset = &h->block_offset[0];
1868 const int transform_bypass = !simple && (s->qscale == 0 && h->sps.transform_bypass);
1869 /* is_h264 should always be true if SVQ3 is disabled. */
1870 const int is_h264 = !CONFIG_SVQ3_DECODER || simple || s->codec_id == CODEC_ID_H264;
1871 void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1872 const int block_h = 16 >> s->chroma_y_shift;
1873 const int chroma422 = CHROMA422;
1875 dest_y = s->current_picture.f.data[0] + ((mb_x << pixel_shift) + mb_y * s->linesize ) * 16;
1876 dest_cb = s->current_picture.f.data[1] + (mb_x << pixel_shift)*8 + mb_y * s->uvlinesize * block_h;
1877 dest_cr = s->current_picture.f.data[2] + (mb_x << pixel_shift)*8 + mb_y * s->uvlinesize * block_h;
1879 s->dsp.prefetch(dest_y + (s->mb_x&3)*4*s->linesize + (64 << pixel_shift), s->linesize, 4);
1880 s->dsp.prefetch(dest_cb + (s->mb_x&7)*s->uvlinesize + (64 << pixel_shift), dest_cr - dest_cb, 2);
1882 h->list_counts[mb_xy]= h->list_count;
1884 if (!simple && MB_FIELD) {
1885 linesize = h->mb_linesize = s->linesize * 2;
1886 uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
1887 block_offset = &h->block_offset[48];
1888 if(mb_y&1){ //FIXME move out of this function?
1889 dest_y -= s->linesize*15;
1890 dest_cb-= s->uvlinesize * (block_h - 1);
1891 dest_cr-= s->uvlinesize * (block_h - 1);
1895 for(list=0; list<h->list_count; list++){
1896 if(!USES_LIST(mb_type, list))
1898 if(IS_16X16(mb_type)){
1899 int8_t *ref = &h->ref_cache[list][scan8[0]];
1900 fill_rectangle(ref, 4, 4, 8, (16+*ref)^(s->mb_y&1), 1);
1902 for(i=0; i<16; i+=4){
1903 int ref = h->ref_cache[list][scan8[i]];
1905 fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, (16+ref)^(s->mb_y&1), 1);
1911 linesize = h->mb_linesize = s->linesize;
1912 uvlinesize = h->mb_uvlinesize = s->uvlinesize;
1913 // dct_offset = s->linesize * 16;
1916 if (!simple && IS_INTRA_PCM(mb_type)) {
1918 const int bit_depth = h->sps.bit_depth_luma;
1921 init_get_bits(&gb, (uint8_t*)h->mb, 384*bit_depth);
1923 for (i = 0; i < 16; i++) {
1924 uint16_t *tmp_y = (uint16_t*)(dest_y + i*linesize);
1925 for (j = 0; j < 16; j++)
1926 tmp_y[j] = get_bits(&gb, bit_depth);
1928 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1929 if (!h->sps.chroma_format_idc) {
1930 for (i = 0; i < block_h; i++) {
1931 uint16_t *tmp_cb = (uint16_t*)(dest_cb + i*uvlinesize);
1932 for (j = 0; j < 8; j++) {
1933 tmp_cb[j] = 1 << (bit_depth - 1);
1936 for (i = 0; i < block_h; i++) {
1937 uint16_t *tmp_cr = (uint16_t*)(dest_cr + i*uvlinesize);
1938 for (j = 0; j < 8; j++) {
1939 tmp_cr[j] = 1 << (bit_depth - 1);
1943 for (i = 0; i < block_h; i++) {
1944 uint16_t *tmp_cb = (uint16_t*)(dest_cb + i*uvlinesize);
1945 for (j = 0; j < 8; j++)
1946 tmp_cb[j] = get_bits(&gb, bit_depth);
1948 for (i = 0; i < block_h; i++) {
1949 uint16_t *tmp_cr = (uint16_t*)(dest_cr + i*uvlinesize);
1950 for (j = 0; j < 8; j++)
1951 tmp_cr[j] = get_bits(&gb, bit_depth);
1956 for (i=0; i<16; i++) {
1957 memcpy(dest_y + i* linesize, h->mb + i*8, 16);
1959 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1960 if (!h->sps.chroma_format_idc) {
1961 for (i = 0; i < block_h; i++) {
1962 memset(dest_cb + i*uvlinesize, 128, 8);
1963 memset(dest_cr + i*uvlinesize, 128, 8);
1966 for (i = 0; i < block_h; i++) {
1967 memcpy(dest_cb + i*uvlinesize, h->mb + 128 + i*4, 8);
1968 memcpy(dest_cr + i*uvlinesize, h->mb + 160 + i*4, 8);
1974 if(IS_INTRA(mb_type)){
1975 if(h->deblocking_filter)
1976 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1, 0, simple, pixel_shift);
1978 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1979 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
1980 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
1983 hl_decode_mb_predict_luma(h, mb_type, is_h264, simple, transform_bypass, pixel_shift, block_offset, linesize, dest_y, 0);
1985 if(h->deblocking_filter)
1986 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0, 0, simple, pixel_shift);
1989 hl_motion_422(h, dest_y, dest_cb, dest_cr,
1990 s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
1991 s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
1992 h->h264dsp.weight_h264_pixels_tab,
1993 h->h264dsp.biweight_h264_pixels_tab,
1996 hl_motion_420(h, dest_y, dest_cb, dest_cr,
1997 s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
1998 s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
1999 h->h264dsp.weight_h264_pixels_tab,
2000 h->h264dsp.biweight_h264_pixels_tab,
2005 hl_decode_mb_idct_luma(h, mb_type, is_h264, simple, transform_bypass, pixel_shift, block_offset, linesize, dest_y, 0);
2007 if((simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)) && (h->cbp&0x30)){
2008 uint8_t *dest[2] = {dest_cb, dest_cr};
2009 if(transform_bypass){
2010 if(IS_INTRA(mb_type) && h->sps.profile_idc==244 && (h->chroma_pred_mode==VERT_PRED8x8 || h->chroma_pred_mode==HOR_PRED8x8)){
2011 h->hpc.pred8x8_add[h->chroma_pred_mode](dest[0], block_offset + 16, h->mb + (16*16*1 << pixel_shift), uvlinesize);
2012 h->hpc.pred8x8_add[h->chroma_pred_mode](dest[1], block_offset + 32, h->mb + (16*16*2 << pixel_shift), uvlinesize);
2014 idct_add = s->dsp.add_pixels4;
2016 for(i=j*16; i<j*16+4; i++){
2017 if(h->non_zero_count_cache[ scan8[i] ] || dctcoef_get(h->mb, pixel_shift, i*16))
2018 idct_add (dest[j-1] + block_offset[i], h->mb + (i*16 << pixel_shift), uvlinesize);
2021 for(i=j*16+4; i<j*16+8; i++){
2022 if(h->non_zero_count_cache[ scan8[i+4] ] || dctcoef_get(h->mb, pixel_shift, i*16))
2023 idct_add (dest[j-1] + block_offset[i+4], h->mb + (i*16 << pixel_shift), uvlinesize);
2032 qp[0] = h->chroma_qp[0] + 3;
2033 qp[1] = h->chroma_qp[1] + 3;
2035 qp[0] = h->chroma_qp[0];
2036 qp[1] = h->chroma_qp[1];
2038 if(h->non_zero_count_cache[ scan8[CHROMA_DC_BLOCK_INDEX+0] ])
2039 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]);
2040 if(h->non_zero_count_cache[ scan8[CHROMA_DC_BLOCK_INDEX+1] ])
2041 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]);
2042 h->h264dsp.h264_idct_add8(dest, block_offset,
2044 h->non_zero_count_cache);
2046 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]);
2047 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]);
2049 for(i=j*16; i<j*16+4; i++){
2050 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2051 uint8_t * const ptr= dest[j-1] + block_offset[i];
2052 ff_svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, ff_h264_chroma_qp[0][s->qscale + 12] - 12, 2);
2060 if(h->cbp || IS_INTRA(mb_type))
2062 s->dsp.clear_blocks(h->mb);
2063 s->dsp.clear_blocks(h->mb+(24*16<<pixel_shift));
2067 static av_always_inline void hl_decode_mb_444_internal(H264Context *h, int simple, int pixel_shift){
2068 MpegEncContext * const s = &h->s;
2069 const int mb_x= s->mb_x;
2070 const int mb_y= s->mb_y;
2071 const int mb_xy= h->mb_xy;
2072 const int mb_type = s->current_picture.f.mb_type[mb_xy];
2076 int *block_offset = &h->block_offset[0];
2077 const int transform_bypass = !simple && (s->qscale == 0 && h->sps.transform_bypass);
2078 const int plane_count = (simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)) ? 3 : 1;
2080 for (p = 0; p < plane_count; p++)
2082 dest[p] = s->current_picture.f.data[p] + ((mb_x << pixel_shift) + mb_y * s->linesize) * 16;
2083 s->dsp.prefetch(dest[p] + (s->mb_x&3)*4*s->linesize + (64 << pixel_shift), s->linesize, 4);
2086 h->list_counts[mb_xy]= h->list_count;
2088 if (!simple && MB_FIELD) {
2089 linesize = h->mb_linesize = h->mb_uvlinesize = s->linesize * 2;
2090 block_offset = &h->block_offset[48];
2091 if(mb_y&1) //FIXME move out of this function?
2092 for (p = 0; p < 3; p++)
2093 dest[p] -= s->linesize*15;
2096 for(list=0; list<h->list_count; list++){
2097 if(!USES_LIST(mb_type, list))
2099 if(IS_16X16(mb_type)){
2100 int8_t *ref = &h->ref_cache[list][scan8[0]];
2101 fill_rectangle(ref, 4, 4, 8, (16+*ref)^(s->mb_y&1), 1);
2103 for(i=0; i<16; i+=4){
2104 int ref = h->ref_cache[list][scan8[i]];
2106 fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, (16+ref)^(s->mb_y&1), 1);
2112 linesize = h->mb_linesize = h->mb_uvlinesize = s->linesize;
2115 if (!simple && IS_INTRA_PCM(mb_type)) {
2117 const int bit_depth = h->sps.bit_depth_luma;
2119 init_get_bits(&gb, (uint8_t*)h->mb, 768*bit_depth);
2121 for (p = 0; p < plane_count; p++) {
2122 for (i = 0; i < 16; i++) {
2123 uint16_t *tmp = (uint16_t*)(dest[p] + i*linesize);
2124 for (j = 0; j < 16; j++)
2125 tmp[j] = get_bits(&gb, bit_depth);
2129 for (p = 0; p < plane_count; p++) {
2130 for (i = 0; i < 16; i++) {
2131 memcpy(dest[p] + i*linesize, h->mb + p*128 + i*8, 16);
2136 if(IS_INTRA(mb_type)){
2137 if(h->deblocking_filter)
2138 xchg_mb_border(h, dest[0], dest[1], dest[2], linesize, linesize, 1, 1, simple, pixel_shift);
2140 for (p = 0; p < plane_count; p++)
2141 hl_decode_mb_predict_luma(h, mb_type, 1, simple, transform_bypass, pixel_shift, block_offset, linesize, dest[p], p);
2143 if(h->deblocking_filter)
2144 xchg_mb_border(h, dest[0], dest[1], dest[2], linesize, linesize, 0, 1, simple, pixel_shift);
2146 hl_motion(h, dest[0], dest[1], dest[2],
2147 s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
2148 s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
2149 h->h264dsp.weight_h264_pixels_tab,
2150 h->h264dsp.biweight_h264_pixels_tab, pixel_shift, 3);
2153 for (p = 0; p < plane_count; p++)
2154 hl_decode_mb_idct_luma(h, mb_type, 1, simple, transform_bypass, pixel_shift, block_offset, linesize, dest[p], p);
2156 if(h->cbp || IS_INTRA(mb_type))
2158 s->dsp.clear_blocks(h->mb);
2159 s->dsp.clear_blocks(h->mb+(24*16<<pixel_shift));
2164 * Process a macroblock; this case avoids checks for expensive uncommon cases.
2166 #define hl_decode_mb_simple(sh, bits) \
2167 static void hl_decode_mb_simple_ ## bits(H264Context *h){ \
2168 hl_decode_mb_internal(h, 1, sh); \
2170 hl_decode_mb_simple(0, 8);
2171 hl_decode_mb_simple(1, 16);
2174 * Process a macroblock; this handles edge cases, such as interlacing.
2176 static void av_noinline hl_decode_mb_complex(H264Context *h){
2177 hl_decode_mb_internal(h, 0, h->pixel_shift);
2180 static void av_noinline hl_decode_mb_444_complex(H264Context *h){
2181 hl_decode_mb_444_internal(h, 0, h->pixel_shift);
2184 static void av_noinline hl_decode_mb_444_simple(H264Context *h){
2185 hl_decode_mb_444_internal(h, 1, 0);
2188 void ff_h264_hl_decode_mb(H264Context *h){
2189 MpegEncContext * const s = &h->s;
2190 const int mb_xy= h->mb_xy;
2191 const int mb_type = s->current_picture.f.mb_type[mb_xy];
2192 int is_complex = CONFIG_SMALL || h->is_complex || IS_INTRA_PCM(mb_type) || s->qscale == 0;
2195 if(is_complex || h->pixel_shift)
2196 hl_decode_mb_444_complex(h);
2198 hl_decode_mb_444_simple(h);
2199 } else if (is_complex) {
2200 hl_decode_mb_complex(h);
2201 } else if (h->pixel_shift) {
2202 hl_decode_mb_simple_16(h);
2204 hl_decode_mb_simple_8(h);
2207 static int pred_weight_table(H264Context *h){
2208 MpegEncContext * const s = &h->s;
2210 int luma_def, chroma_def;
2213 h->use_weight_chroma= 0;
2214 h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
2215 if(h->sps.chroma_format_idc)
2216 h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
2217 luma_def = 1<<h->luma_log2_weight_denom;
2218 chroma_def = 1<<h->chroma_log2_weight_denom;
2220 for(list=0; list<2; list++){
2221 h->luma_weight_flag[list] = 0;
2222 h->chroma_weight_flag[list] = 0;
2223 for(i=0; i<h->ref_count[list]; i++){
2224 int luma_weight_flag, chroma_weight_flag;
2226 luma_weight_flag= get_bits1(&s->gb);
2227 if(luma_weight_flag){
2228 h->luma_weight[i][list][0]= get_se_golomb(&s->gb);
2229 h->luma_weight[i][list][1]= get_se_golomb(&s->gb);
2230 if( h->luma_weight[i][list][0] != luma_def
2231 || h->luma_weight[i][list][1] != 0) {
2233 h->luma_weight_flag[list]= 1;
2236 h->luma_weight[i][list][0]= luma_def;
2237 h->luma_weight[i][list][1]= 0;
2240 if(h->sps.chroma_format_idc){
2241 chroma_weight_flag= get_bits1(&s->gb);
2242 if(chroma_weight_flag){
2245 h->chroma_weight[i][list][j][0]= get_se_golomb(&s->gb);
2246 h->chroma_weight[i][list][j][1]= get_se_golomb(&s->gb);
2247 if( h->chroma_weight[i][list][j][0] != chroma_def
2248 || h->chroma_weight[i][list][j][1] != 0) {
2249 h->use_weight_chroma= 1;
2250 h->chroma_weight_flag[list]= 1;
2256 h->chroma_weight[i][list][j][0]= chroma_def;
2257 h->chroma_weight[i][list][j][1]= 0;
2262 if(h->slice_type_nos != AV_PICTURE_TYPE_B) break;
2264 h->use_weight= h->use_weight || h->use_weight_chroma;
2269 * Initialize implicit_weight table.
2270 * @param field 0/1 initialize the weight for interlaced MBAFF
2271 * -1 initializes the rest
2273 static void implicit_weight_table(H264Context *h, int field){
2274 MpegEncContext * const s = &h->s;
2275 int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
2277 for (i = 0; i < 2; i++) {
2278 h->luma_weight_flag[i] = 0;
2279 h->chroma_weight_flag[i] = 0;
2283 if (s->picture_structure == PICT_FRAME) {
2284 cur_poc = s->current_picture_ptr->poc;
2286 cur_poc = s->current_picture_ptr->field_poc[s->picture_structure - 1];
2288 if( h->ref_count[0] == 1 && h->ref_count[1] == 1 && !FRAME_MBAFF
2289 && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
2291 h->use_weight_chroma= 0;
2295 ref_count0= h->ref_count[0];
2296 ref_count1= h->ref_count[1];
2298 cur_poc = s->current_picture_ptr->field_poc[field];
2300 ref_count0= 16+2*h->ref_count[0];
2301 ref_count1= 16+2*h->ref_count[1];
2305 h->use_weight_chroma= 2;
2306 h->luma_log2_weight_denom= 5;
2307 h->chroma_log2_weight_denom= 5;
2309 for(ref0=ref_start; ref0 < ref_count0; ref0++){
2310 int poc0 = h->ref_list[0][ref0].poc;
2311 for(ref1=ref_start; ref1 < ref_count1; ref1++){
2313 if (!h->ref_list[0][ref0].long_ref && !h->ref_list[1][ref1].long_ref) {
2314 int poc1 = h->ref_list[1][ref1].poc;
2315 int td = av_clip(poc1 - poc0, -128, 127);
2317 int tb = av_clip(cur_poc - poc0, -128, 127);
2318 int tx = (16384 + (FFABS(td) >> 1)) / td;
2319 int dist_scale_factor = (tb*tx + 32) >> 8;
2320 if(dist_scale_factor >= -64 && dist_scale_factor <= 128)
2321 w = 64 - dist_scale_factor;
2325 h->implicit_weight[ref0][ref1][0]=
2326 h->implicit_weight[ref0][ref1][1]= w;
2328 h->implicit_weight[ref0][ref1][field]=w;
2335 * instantaneous decoder refresh.
2337 static void idr(H264Context *h){
2338 ff_h264_remove_all_refs(h);
2339 h->prev_frame_num= 0;
2340 h->prev_frame_num_offset= 0;
2345 /* forget old pics after a seek */
2346 static void flush_dpb(AVCodecContext *avctx){
2347 H264Context *h= avctx->priv_data;
2349 for(i=0; i<MAX_DELAYED_PIC_COUNT; i++) {
2350 if(h->delayed_pic[i])
2351 h->delayed_pic[i]->f.reference = 0;
2352 h->delayed_pic[i]= NULL;
2354 h->outputed_poc=h->next_outputed_poc= INT_MIN;
2355 h->prev_interlaced_frame = 1;
2357 if(h->s.current_picture_ptr)
2358 h->s.current_picture_ptr->f.reference = 0;
2359 h->s.first_field= 0;
2360 ff_h264_reset_sei(h);
2361 ff_mpeg_flush(avctx);
2364 static int init_poc(H264Context *h){
2365 MpegEncContext * const s = &h->s;
2366 const int max_frame_num= 1<<h->sps.log2_max_frame_num;
2368 Picture *cur = s->current_picture_ptr;
2370 h->frame_num_offset= h->prev_frame_num_offset;
2371 if(h->frame_num < h->prev_frame_num)
2372 h->frame_num_offset += max_frame_num;
2374 if(h->sps.poc_type==0){
2375 const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
2377 if (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
2378 h->poc_msb = h->prev_poc_msb + max_poc_lsb;
2379 else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
2380 h->poc_msb = h->prev_poc_msb - max_poc_lsb;
2382 h->poc_msb = h->prev_poc_msb;
2383 //printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
2385 field_poc[1] = h->poc_msb + h->poc_lsb;
2386 if(s->picture_structure == PICT_FRAME)
2387 field_poc[1] += h->delta_poc_bottom;
2388 }else if(h->sps.poc_type==1){
2389 int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
2392 if(h->sps.poc_cycle_length != 0)
2393 abs_frame_num = h->frame_num_offset + h->frame_num;
2397 if(h->nal_ref_idc==0 && abs_frame_num > 0)
2400 expected_delta_per_poc_cycle = 0;
2401 for(i=0; i < h->sps.poc_cycle_length; i++)
2402 expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
2404 if(abs_frame_num > 0){
2405 int poc_cycle_cnt = (abs_frame_num - 1) / h->sps.poc_cycle_length;
2406 int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
2408 expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
2409 for(i = 0; i <= frame_num_in_poc_cycle; i++)
2410 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
2414 if(h->nal_ref_idc == 0)
2415 expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
2417 field_poc[0] = expectedpoc + h->delta_poc[0];
2418 field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
2420 if(s->picture_structure == PICT_FRAME)
2421 field_poc[1] += h->delta_poc[1];
2423 int poc= 2*(h->frame_num_offset + h->frame_num);
2432 if(s->picture_structure != PICT_BOTTOM_FIELD)
2433 s->current_picture_ptr->field_poc[0]= field_poc[0];
2434 if(s->picture_structure != PICT_TOP_FIELD)
2435 s->current_picture_ptr->field_poc[1]= field_poc[1];
2436 cur->poc= FFMIN(cur->field_poc[0], cur->field_poc[1]);
2443 * initialize scan tables
2445 static void init_scan_tables(H264Context *h){
2447 for(i=0; i<16; i++){
2448 #define T(x) (x>>2) | ((x<<2) & 0xF)
2449 h->zigzag_scan[i] = T(zigzag_scan[i]);
2450 h-> field_scan[i] = T( field_scan[i]);
2453 for(i=0; i<64; i++){
2454 #define T(x) (x>>3) | ((x&7)<<3)
2455 h->zigzag_scan8x8[i] = T(ff_zigzag_direct[i]);
2456 h->zigzag_scan8x8_cavlc[i] = T(zigzag_scan8x8_cavlc[i]);
2457 h->field_scan8x8[i] = T(field_scan8x8[i]);
2458 h->field_scan8x8_cavlc[i] = T(field_scan8x8_cavlc[i]);
2461 if(h->sps.transform_bypass){ //FIXME same ugly
2462 h->zigzag_scan_q0 = zigzag_scan;
2463 h->zigzag_scan8x8_q0 = ff_zigzag_direct;
2464 h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc;
2465 h->field_scan_q0 = field_scan;
2466 h->field_scan8x8_q0 = field_scan8x8;
2467 h->field_scan8x8_cavlc_q0 = field_scan8x8_cavlc;
2469 h->zigzag_scan_q0 = h->zigzag_scan;
2470 h->zigzag_scan8x8_q0 = h->zigzag_scan8x8;
2471 h->zigzag_scan8x8_cavlc_q0 = h->zigzag_scan8x8_cavlc;
2472 h->field_scan_q0 = h->field_scan;
2473 h->field_scan8x8_q0 = h->field_scan8x8;
2474 h->field_scan8x8_cavlc_q0 = h->field_scan8x8_cavlc;
2478 static int field_end(H264Context *h, int in_setup){
2479 MpegEncContext * const s = &h->s;
2480 AVCodecContext * const avctx= s->avctx;
2484 if (!in_setup && !s->dropable)
2485 ff_thread_report_progress((AVFrame*)s->current_picture_ptr, (16*s->mb_height >> FIELD_PICTURE) - 1,
2486 s->picture_structure==PICT_BOTTOM_FIELD);
2488 if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
2489 ff_vdpau_h264_set_reference_frames(s);
2491 if(in_setup || !(avctx->active_thread_type&FF_THREAD_FRAME)){
2493 err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
2494 h->prev_poc_msb= h->poc_msb;
2495 h->prev_poc_lsb= h->poc_lsb;
2497 h->prev_frame_num_offset= h->frame_num_offset;
2498 h->prev_frame_num= h->frame_num;
2499 h->outputed_poc = h->next_outputed_poc;
2502 if (avctx->hwaccel) {
2503 if (avctx->hwaccel->end_frame(avctx) < 0)
2504 av_log(avctx, AV_LOG_ERROR, "hardware accelerator failed to decode picture\n");
2507 if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
2508 ff_vdpau_h264_picture_complete(s);
2511 * FIXME: Error handling code does not seem to support interlaced
2512 * when slices span multiple rows
2513 * The ff_er_add_slice calls don't work right for bottom
2514 * fields; they cause massive erroneous error concealing
2515 * Error marking covers both fields (top and bottom).
2516 * This causes a mismatched s->error_count
2517 * and a bad error table. Further, the error count goes to
2518 * INT_MAX when called for bottom field, because mb_y is
2519 * past end by one (callers fault) and resync_mb_y != 0
2520 * causes problems for the first MB line, too.
2533 * Replicate H264 "master" context to thread contexts.
2535 static void clone_slice(H264Context *dst, H264Context *src)
2537 memcpy(dst->block_offset, src->block_offset, sizeof(dst->block_offset));
2538 dst->s.current_picture_ptr = src->s.current_picture_ptr;
2539 dst->s.current_picture = src->s.current_picture;
2540 dst->s.linesize = src->s.linesize;
2541 dst->s.uvlinesize = src->s.uvlinesize;
2542 dst->s.first_field = src->s.first_field;
2544 dst->prev_poc_msb = src->prev_poc_msb;
2545 dst->prev_poc_lsb = src->prev_poc_lsb;
2546 dst->prev_frame_num_offset = src->prev_frame_num_offset;
2547 dst->prev_frame_num = src->prev_frame_num;
2548 dst->short_ref_count = src->short_ref_count;
2550 memcpy(dst->short_ref, src->short_ref, sizeof(dst->short_ref));
2551 memcpy(dst->long_ref, src->long_ref, sizeof(dst->long_ref));
2552 memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
2553 memcpy(dst->ref_list, src->ref_list, sizeof(dst->ref_list));
2555 memcpy(dst->dequant4_coeff, src->dequant4_coeff, sizeof(src->dequant4_coeff));
2556 memcpy(dst->dequant8_coeff, src->dequant8_coeff, sizeof(src->dequant8_coeff));
2560 * computes profile from profile_idc and constraint_set?_flags
2564 * @return profile as defined by FF_PROFILE_H264_*
2566 int ff_h264_get_profile(SPS *sps)
2568 int profile = sps->profile_idc;
2570 switch(sps->profile_idc) {
2571 case FF_PROFILE_H264_BASELINE:
2572 // constraint_set1_flag set to 1
2573 profile |= (sps->constraint_set_flags & 1<<1) ? FF_PROFILE_H264_CONSTRAINED : 0;
2575 case FF_PROFILE_H264_HIGH_10:
2576 case FF_PROFILE_H264_HIGH_422:
2577 case FF_PROFILE_H264_HIGH_444_PREDICTIVE:
2578 // constraint_set3_flag set to 1
2579 profile |= (sps->constraint_set_flags & 1<<3) ? FF_PROFILE_H264_INTRA : 0;
2587 * decodes a slice header.
2588 * This will also call MPV_common_init() and frame_start() as needed.
2590 * @param h h264context
2591 * @param h0 h264 master context (differs from 'h' when doing sliced based parallel decoding)
2593 * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
2595 static int decode_slice_header(H264Context *h, H264Context *h0){
2596 MpegEncContext * const s = &h->s;
2597 MpegEncContext * const s0 = &h0->s;
2598 unsigned int first_mb_in_slice;
2599 unsigned int pps_id;
2600 int num_ref_idx_active_override_flag;
2601 unsigned int slice_type, tmp, i, j;
2602 int default_ref_list_done = 0;
2603 int last_pic_structure;
2605 s->dropable= h->nal_ref_idc == 0;
2607 /* FIXME: 2tap qpel isn't implemented for high bit depth. */
2608 if((s->avctx->flags2 & CODEC_FLAG2_FAST) && !h->nal_ref_idc && !h->pixel_shift){
2609 s->me.qpel_put= s->dsp.put_2tap_qpel_pixels_tab;
2610 s->me.qpel_avg= s->dsp.avg_2tap_qpel_pixels_tab;
2612 s->me.qpel_put= s->dsp.put_h264_qpel_pixels_tab;
2613 s->me.qpel_avg= s->dsp.avg_h264_qpel_pixels_tab;
2616 first_mb_in_slice= get_ue_golomb(&s->gb);
2618 if(first_mb_in_slice == 0){ //FIXME better field boundary detection
2619 if(h0->current_slice && FIELD_PICTURE){
2623 h0->current_slice = 0;
2624 if (!s0->first_field)
2625 s->current_picture_ptr= NULL;
2628 slice_type= get_ue_golomb_31(&s->gb);
2630 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);
2635 h->slice_type_fixed=1;
2637 h->slice_type_fixed=0;
2639 slice_type= golomb_to_pict_type[ slice_type ];
2640 if (slice_type == AV_PICTURE_TYPE_I
2641 || (h0->current_slice != 0 && slice_type == h0->last_slice_type) ) {
2642 default_ref_list_done = 1;
2644 h->slice_type= slice_type;
2645 h->slice_type_nos= slice_type & 3;
2647 s->pict_type= h->slice_type; // to make a few old functions happy, it's wrong though
2649 pps_id= get_ue_golomb(&s->gb);
2650 if(pps_id>=MAX_PPS_COUNT){
2651 av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
2654 if(!h0->pps_buffers[pps_id]) {
2655 av_log(h->s.avctx, AV_LOG_ERROR, "non-existing PPS %u referenced\n", pps_id);
2658 h->pps= *h0->pps_buffers[pps_id];
2660 if(!h0->sps_buffers[h->pps.sps_id]) {
2661 av_log(h->s.avctx, AV_LOG_ERROR, "non-existing SPS %u referenced\n", h->pps.sps_id);
2664 h->sps = *h0->sps_buffers[h->pps.sps_id];
2666 s->avctx->profile = ff_h264_get_profile(&h->sps);
2667 s->avctx->level = h->sps.level_idc;
2668 s->avctx->refs = h->sps.ref_frame_count;
2670 if(h == h0 && h->dequant_coeff_pps != pps_id){
2671 h->dequant_coeff_pps = pps_id;
2672 init_dequant_tables(h);
2675 s->mb_width= h->sps.mb_width;
2676 s->mb_height= h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
2678 h->b_stride= s->mb_width*4;
2680 s->chroma_y_shift = h->sps.chroma_format_idc <= 1; // 400 uses yuv420p
2682 s->width = 16*s->mb_width - (2>>CHROMA444)*FFMIN(h->sps.crop_right, (8<<CHROMA444)-1);
2683 if(h->sps.frame_mbs_only_flag)
2684 s->height= 16*s->mb_height - (1<<s->chroma_y_shift)*FFMIN(h->sps.crop_bottom, (16>>s->chroma_y_shift)-1);
2686 s->height= 16*s->mb_height - (2<<s->chroma_y_shift)*FFMIN(h->sps.crop_bottom, (16>>s->chroma_y_shift)-1);
2688 if (s->context_initialized
2689 && ( s->width != s->avctx->width || s->height != s->avctx->height
2690 || av_cmp_q(h->sps.sar, s->avctx->sample_aspect_ratio))) {
2692 av_log_missing_feature(s->avctx, "Width/height changing with threads is", 0);
2693 return -1; // width / height changed during parallelized decoding
2696 flush_dpb(s->avctx);
2699 if (!s->context_initialized) {
2701 av_log(h->s.avctx, AV_LOG_ERROR, "Cannot (re-)initialize context during parallel decoding.\n");
2705 avcodec_set_dimensions(s->avctx, s->width, s->height);
2706 s->avctx->sample_aspect_ratio= h->sps.sar;
2707 av_assert0(s->avctx->sample_aspect_ratio.den);
2709 if(h->sps.video_signal_type_present_flag){
2710 s->avctx->color_range = h->sps.full_range ? AVCOL_RANGE_JPEG : AVCOL_RANGE_MPEG;
2711 if(h->sps.colour_description_present_flag){
2712 s->avctx->color_primaries = h->sps.color_primaries;
2713 s->avctx->color_trc = h->sps.color_trc;
2714 s->avctx->colorspace = h->sps.colorspace;
2718 if(h->sps.timing_info_present_flag){
2719 int64_t den= h->sps.time_scale;
2720 if(h->x264_build < 44U)
2722 av_reduce(&s->avctx->time_base.num, &s->avctx->time_base.den,
2723 h->sps.num_units_in_tick, den, 1<<30);
2726 switch (h->sps.bit_depth_luma) {
2729 s->avctx->pix_fmt = PIX_FMT_YUV444P9;
2731 s->avctx->pix_fmt = PIX_FMT_YUV422P9;
2733 s->avctx->pix_fmt = PIX_FMT_YUV420P9;
2737 s->avctx->pix_fmt = PIX_FMT_YUV444P10;
2739 s->avctx->pix_fmt = PIX_FMT_YUV422P10;
2741 s->avctx->pix_fmt = PIX_FMT_YUV420P10;
2745 s->avctx->pix_fmt = s->avctx->color_range == AVCOL_RANGE_JPEG ? PIX_FMT_YUVJ444P : PIX_FMT_YUV444P;
2746 } else if (CHROMA422) {
2747 s->avctx->pix_fmt = s->avctx->color_range == AVCOL_RANGE_JPEG ? PIX_FMT_YUVJ422P : PIX_FMT_YUV422P;
2749 s->avctx->pix_fmt = s->avctx->get_format(s->avctx,
2750 s->avctx->codec->pix_fmts ?
2751 s->avctx->codec->pix_fmts :
2752 s->avctx->color_range == AVCOL_RANGE_JPEG ?
2753 hwaccel_pixfmt_list_h264_jpeg_420 :
2754 ff_hwaccel_pixfmt_list_420);
2758 s->avctx->hwaccel = ff_find_hwaccel(s->avctx->codec->id, s->avctx->pix_fmt);
2760 if (MPV_common_init(s) < 0) {
2761 av_log(h->s.avctx, AV_LOG_ERROR, "MPV_common_init() failed.\n");
2765 h->prev_interlaced_frame = 1;
2767 init_scan_tables(h);
2768 if (ff_h264_alloc_tables(h) < 0) {
2769 av_log(h->s.avctx, AV_LOG_ERROR, "Could not allocate memory for h264\n");
2770 return AVERROR(ENOMEM);
2773 if (!HAVE_THREADS || !(s->avctx->active_thread_type&FF_THREAD_SLICE)) {
2774 if (context_init(h) < 0) {
2775 av_log(h->s.avctx, AV_LOG_ERROR, "context_init() failed.\n");
2779 for(i = 1; i < s->avctx->thread_count; i++) {
2781 c = h->thread_context[i] = av_malloc(sizeof(H264Context));
2782 memcpy(c, h->s.thread_context[i], sizeof(MpegEncContext));
2783 memset(&c->s + 1, 0, sizeof(H264Context) - sizeof(MpegEncContext));
2784 c->h264dsp = h->h264dsp;
2787 c->pixel_shift = h->pixel_shift;
2788 init_scan_tables(c);
2789 clone_tables(c, h, i);
2792 for(i = 0; i < s->avctx->thread_count; i++)
2793 if (context_init(h->thread_context[i]) < 0) {
2794 av_log(h->s.avctx, AV_LOG_ERROR, "context_init() failed.\n");
2800 h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
2803 h->mb_aff_frame = 0;
2804 last_pic_structure = s0->picture_structure;
2805 if(h->sps.frame_mbs_only_flag){
2806 s->picture_structure= PICT_FRAME;
2808 if(get_bits1(&s->gb)) { //field_pic_flag
2809 s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
2811 s->picture_structure= PICT_FRAME;
2812 h->mb_aff_frame = h->sps.mb_aff;
2815 h->mb_field_decoding_flag= s->picture_structure != PICT_FRAME;
2817 if(h0->current_slice == 0){
2818 // Shorten frame num gaps so we don't have to allocate reference frames just to throw them away
2819 if(h->frame_num != h->prev_frame_num) {
2820 int unwrap_prev_frame_num = h->prev_frame_num, max_frame_num = 1<<h->sps.log2_max_frame_num;
2822 if (unwrap_prev_frame_num > h->frame_num) unwrap_prev_frame_num -= max_frame_num;
2824 if ((h->frame_num - unwrap_prev_frame_num) > h->sps.ref_frame_count) {
2825 unwrap_prev_frame_num = (h->frame_num - h->sps.ref_frame_count) - 1;
2826 if (unwrap_prev_frame_num < 0)
2827 unwrap_prev_frame_num += max_frame_num;
2829 h->prev_frame_num = unwrap_prev_frame_num;
2833 while(h->frame_num != h->prev_frame_num &&
2834 h->frame_num != (h->prev_frame_num+1)%(1<<h->sps.log2_max_frame_num)){
2835 Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
2836 av_log(h->s.avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n", h->frame_num, h->prev_frame_num);
2837 if (ff_h264_frame_start(h) < 0)
2839 h->prev_frame_num++;
2840 h->prev_frame_num %= 1<<h->sps.log2_max_frame_num;
2841 s->current_picture_ptr->frame_num= h->prev_frame_num;
2842 ff_thread_report_progress((AVFrame*)s->current_picture_ptr, INT_MAX, 0);
2843 ff_thread_report_progress((AVFrame*)s->current_picture_ptr, INT_MAX, 1);
2844 ff_generate_sliding_window_mmcos(h);
2845 if (ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index) < 0 &&
2846 (s->avctx->err_recognition & AV_EF_EXPLODE))
2847 return AVERROR_INVALIDDATA;
2848 /* Error concealment: if a ref is missing, copy the previous ref in its place.
2849 * FIXME: avoiding a memcpy would be nice, but ref handling makes many assumptions
2850 * about there being no actual duplicates.
2851 * FIXME: this doesn't copy padding for out-of-frame motion vectors. Given we're
2852 * concealing a lost frame, this probably isn't noticable by comparison, but it should
2854 if (h->short_ref_count) {
2856 av_image_copy(h->short_ref[0]->f.data, h->short_ref[0]->f.linesize,
2857 (const uint8_t**)prev->f.data, prev->f.linesize,
2858 s->avctx->pix_fmt, s->mb_width*16, s->mb_height*16);
2859 h->short_ref[0]->poc = prev->poc+2;
2861 h->short_ref[0]->frame_num = h->prev_frame_num;
2865 /* See if we have a decoded first field looking for a pair... */
2866 if (s0->first_field) {
2867 assert(s0->current_picture_ptr);
2868 assert(s0->current_picture_ptr->f.data[0]);
2869 assert(s0->current_picture_ptr->f.reference != DELAYED_PIC_REF);
2871 /* figure out if we have a complementary field pair */
2872 if (!FIELD_PICTURE || s->picture_structure == last_pic_structure) {
2874 * Previous field is unmatched. Don't display it, but let it
2875 * remain for reference if marked as such.
2877 s0->current_picture_ptr = NULL;
2878 s0->first_field = FIELD_PICTURE;
2881 if (h->nal_ref_idc &&
2882 s0->current_picture_ptr->f.reference &&
2883 s0->current_picture_ptr->frame_num != h->frame_num) {
2885 * This and previous field were reference, but had
2886 * different frame_nums. Consider this field first in
2887 * pair. Throw away previous field except for reference
2890 s0->first_field = 1;
2891 s0->current_picture_ptr = NULL;
2894 /* Second field in complementary pair */
2895 s0->first_field = 0;
2900 /* Frame or first field in a potentially complementary pair */
2901 assert(!s0->current_picture_ptr);
2902 s0->first_field = FIELD_PICTURE;
2905 if(!FIELD_PICTURE || s0->first_field) {
2906 if (ff_h264_frame_start(h) < 0) {
2907 s0->first_field = 0;
2911 ff_release_unused_pictures(s, 0);
2917 s->current_picture_ptr->frame_num= h->frame_num; //FIXME frame_num cleanup
2919 assert(s->mb_num == s->mb_width * s->mb_height);
2920 if(first_mb_in_slice << FIELD_OR_MBAFF_PICTURE >= s->mb_num ||
2921 first_mb_in_slice >= s->mb_num){
2922 av_log(h->s.avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
2925 s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
2926 s->resync_mb_y = s->mb_y = (first_mb_in_slice / s->mb_width) << FIELD_OR_MBAFF_PICTURE;
2927 if (s->picture_structure == PICT_BOTTOM_FIELD)
2928 s->resync_mb_y = s->mb_y = s->mb_y + 1;
2929 assert(s->mb_y < s->mb_height);
2931 if(s->picture_structure==PICT_FRAME){
2932 h->curr_pic_num= h->frame_num;
2933 h->max_pic_num= 1<< h->sps.log2_max_frame_num;
2935 h->curr_pic_num= 2*h->frame_num + 1;
2936 h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
2939 if(h->nal_unit_type == NAL_IDR_SLICE){
2940 get_ue_golomb(&s->gb); /* idr_pic_id */
2943 if(h->sps.poc_type==0){
2944 h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
2946 if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
2947 h->delta_poc_bottom= get_se_golomb(&s->gb);
2951 if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
2952 h->delta_poc[0]= get_se_golomb(&s->gb);
2954 if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
2955 h->delta_poc[1]= get_se_golomb(&s->gb);
2960 if(h->pps.redundant_pic_cnt_present){
2961 h->redundant_pic_count= get_ue_golomb(&s->gb);
2964 //set defaults, might be overridden a few lines later
2965 h->ref_count[0]= h->pps.ref_count[0];
2966 h->ref_count[1]= h->pps.ref_count[1];
2968 if(h->slice_type_nos != AV_PICTURE_TYPE_I){
2969 if(h->slice_type_nos == AV_PICTURE_TYPE_B){
2970 h->direct_spatial_mv_pred= get_bits1(&s->gb);
2972 num_ref_idx_active_override_flag= get_bits1(&s->gb);
2974 if(num_ref_idx_active_override_flag){
2975 h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
2976 if(h->slice_type_nos==AV_PICTURE_TYPE_B)
2977 h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
2979 if(h->ref_count[0]-1 > 32-1 || h->ref_count[1]-1 > 32-1){
2980 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
2981 h->ref_count[0]= h->ref_count[1]= 1;
2985 if(h->slice_type_nos == AV_PICTURE_TYPE_B)
2992 if(!default_ref_list_done){
2993 ff_h264_fill_default_ref_list(h);
2996 if(h->slice_type_nos!=AV_PICTURE_TYPE_I && ff_h264_decode_ref_pic_list_reordering(h) < 0) {
2997 h->ref_count[1]= h->ref_count[0]= 0;
3001 if(h->slice_type_nos!=AV_PICTURE_TYPE_I){
3002 s->last_picture_ptr= &h->ref_list[0][0];
3003 ff_copy_picture(&s->last_picture, s->last_picture_ptr);
3005 if(h->slice_type_nos==AV_PICTURE_TYPE_B){
3006 s->next_picture_ptr= &h->ref_list[1][0];
3007 ff_copy_picture(&s->next_picture, s->next_picture_ptr);
3010 if( (h->pps.weighted_pred && h->slice_type_nos == AV_PICTURE_TYPE_P )
3011 || (h->pps.weighted_bipred_idc==1 && h->slice_type_nos== AV_PICTURE_TYPE_B ) )
3012 pred_weight_table(h);
3013 else if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== AV_PICTURE_TYPE_B){
3014 implicit_weight_table(h, -1);
3017 for (i = 0; i < 2; i++) {
3018 h->luma_weight_flag[i] = 0;
3019 h->chroma_weight_flag[i] = 0;
3023 if(h->nal_ref_idc && ff_h264_decode_ref_pic_marking(h0, &s->gb) < 0 &&
3024 (s->avctx->err_recognition & AV_EF_EXPLODE))
3025 return AVERROR_INVALIDDATA;
3028 ff_h264_fill_mbaff_ref_list(h);
3030 if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== AV_PICTURE_TYPE_B){
3031 implicit_weight_table(h, 0);
3032 implicit_weight_table(h, 1);
3036 if(h->slice_type_nos==AV_PICTURE_TYPE_B && !h->direct_spatial_mv_pred)
3037 ff_h264_direct_dist_scale_factor(h);
3038 ff_h264_direct_ref_list_init(h);
3040 if( h->slice_type_nos != AV_PICTURE_TYPE_I && h->pps.cabac ){
3041 tmp = get_ue_golomb_31(&s->gb);
3043 av_log(s->avctx, AV_LOG_ERROR, "cabac_init_idc overflow\n");
3046 h->cabac_init_idc= tmp;
3049 h->last_qscale_diff = 0;
3050 tmp = h->pps.init_qp + get_se_golomb(&s->gb);
3051 if(tmp>51+6*(h->sps.bit_depth_luma-8)){
3052 av_log(s->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
3056 h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
3057 h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
3058 //FIXME qscale / qp ... stuff
3059 if(h->slice_type == AV_PICTURE_TYPE_SP){
3060 get_bits1(&s->gb); /* sp_for_switch_flag */
3062 if(h->slice_type==AV_PICTURE_TYPE_SP || h->slice_type == AV_PICTURE_TYPE_SI){
3063 get_se_golomb(&s->gb); /* slice_qs_delta */
3066 h->deblocking_filter = 1;
3067 h->slice_alpha_c0_offset = 52;
3068 h->slice_beta_offset = 52;
3069 if( h->pps.deblocking_filter_parameters_present ) {
3070 tmp= get_ue_golomb_31(&s->gb);
3072 av_log(s->avctx, AV_LOG_ERROR, "deblocking_filter_idc %u out of range\n", tmp);
3075 h->deblocking_filter= tmp;
3076 if(h->deblocking_filter < 2)
3077 h->deblocking_filter^= 1; // 1<->0
3079 if( h->deblocking_filter ) {
3080 h->slice_alpha_c0_offset += get_se_golomb(&s->gb) << 1;
3081 h->slice_beta_offset += get_se_golomb(&s->gb) << 1;
3082 if( h->slice_alpha_c0_offset > 104U
3083 || h->slice_beta_offset > 104U){
3084 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);
3090 if( s->avctx->skip_loop_filter >= AVDISCARD_ALL
3091 ||(s->avctx->skip_loop_filter >= AVDISCARD_NONKEY && h->slice_type_nos != AV_PICTURE_TYPE_I)
3092 ||(s->avctx->skip_loop_filter >= AVDISCARD_BIDIR && h->slice_type_nos == AV_PICTURE_TYPE_B)
3093 ||(s->avctx->skip_loop_filter >= AVDISCARD_NONREF && h->nal_ref_idc == 0))
3094 h->deblocking_filter= 0;
3096 if(h->deblocking_filter == 1 && h0->max_contexts > 1) {
3097 if(s->avctx->flags2 & CODEC_FLAG2_FAST) {
3098 /* Cheat slightly for speed:
3099 Do not bother to deblock across slices. */
3100 h->deblocking_filter = 2;
3102 h0->max_contexts = 1;
3103 if(!h0->single_decode_warning) {
3104 av_log(s->avctx, AV_LOG_INFO, "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
3105 h0->single_decode_warning = 1;
3108 av_log(h->s.avctx, AV_LOG_ERROR, "Deblocking switched inside frame.\n");
3113 h->qp_thresh = 15 + 52 - FFMIN(h->slice_alpha_c0_offset, h->slice_beta_offset)
3114 - FFMAX3(0, h->pps.chroma_qp_index_offset[0], h->pps.chroma_qp_index_offset[1])
3115 + 6 * (h->sps.bit_depth_luma - 8);
3118 if( h->pps.num_slice_groups > 1 && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
3119 slice_group_change_cycle= get_bits(&s->gb, ?);
3122 h0->last_slice_type = slice_type;
3123 h->slice_num = ++h0->current_slice;
3124 if(h->slice_num >= MAX_SLICES){
3125 av_log(s->avctx, AV_LOG_ERROR, "Too many slices, increase MAX_SLICES and recompile\n");
3130 int *ref2frm= h->ref2frm[h->slice_num&(MAX_SLICES-1)][j];
3131 for(i=0; i<16; i++){
3133 if (h->ref_list[j][i].f.data[0]) {
3135 uint8_t *base = h->ref_list[j][i].f.base[0];
3136 for(k=0; k<h->short_ref_count; k++)
3137 if (h->short_ref[k]->f.base[0] == base) {
3141 for(k=0; k<h->long_ref_count; k++)
3142 if (h->long_ref[k] && h->long_ref[k]->f.base[0] == base) {
3143 id_list[i]= h->short_ref_count + k;
3152 ref2frm[i+2]= 4*id_list[i]
3153 + (h->ref_list[j][i].f.reference & 3);
3156 for(i=16; i<48; i++)
3157 ref2frm[i+4]= 4*id_list[(i-16)>>1]
3158 + (h->ref_list[j][i].f.reference & 3);
3161 //FIXME: fix draw_edges+PAFF+frame threads
3162 h->emu_edge_width= (s->flags&CODEC_FLAG_EMU_EDGE || (!h->sps.frame_mbs_only_flag && s->avctx->active_thread_type)) ? 0 : 16;
3163 h->emu_edge_height= (FRAME_MBAFF || FIELD_PICTURE) ? 0 : h->emu_edge_width;
3165 if(s->avctx->debug&FF_DEBUG_PICT_INFO){
3166 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",
3168 (s->picture_structure==PICT_FRAME ? "F" : s->picture_structure==PICT_TOP_FIELD ? "T" : "B"),
3170 av_get_picture_type_char(h->slice_type), h->slice_type_fixed ? " fix" : "", h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
3171 pps_id, h->frame_num,
3172 s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
3173 h->ref_count[0], h->ref_count[1],
3175 h->deblocking_filter, h->slice_alpha_c0_offset/2-26, h->slice_beta_offset/2-26,
3177 h->use_weight==1 && h->use_weight_chroma ? "c" : "",
3178 h->slice_type == AV_PICTURE_TYPE_B ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : ""
3185 int ff_h264_get_slice_type(const H264Context *h)
3187 switch (h->slice_type) {
3188 case AV_PICTURE_TYPE_P: return 0;
3189 case AV_PICTURE_TYPE_B: return 1;
3190 case AV_PICTURE_TYPE_I: return 2;
3191 case AV_PICTURE_TYPE_SP: return 3;
3192 case AV_PICTURE_TYPE_SI: return 4;
3197 static av_always_inline void fill_filter_caches_inter(H264Context *h, MpegEncContext * const s, int mb_type, int top_xy,
3198 int left_xy[LEFT_MBS], int top_type, int left_type[LEFT_MBS], int mb_xy, int list)
3200 int b_stride = h->b_stride;
3201 int16_t (*mv_dst)[2] = &h->mv_cache[list][scan8[0]];
3202 int8_t *ref_cache = &h->ref_cache[list][scan8[0]];
3203 if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
3204 if(USES_LIST(top_type, list)){
3205 const int b_xy= h->mb2b_xy[top_xy] + 3*b_stride;
3206 const int b8_xy= 4*top_xy + 2;
3207 int (*ref2frm)[64] = h->ref2frm[ h->slice_table[top_xy]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
3208 AV_COPY128(mv_dst - 1*8, s->current_picture.f.motion_val[list][b_xy + 0]);
3210 ref_cache[1 - 1*8]= ref2frm[list][s->current_picture.f.ref_index[list][b8_xy + 0]];
3212 ref_cache[3 - 1*8]= ref2frm[list][s->current_picture.f.ref_index[list][b8_xy + 1]];
3214 AV_ZERO128(mv_dst - 1*8);
3215 AV_WN32A(&ref_cache[0 - 1*8], ((LIST_NOT_USED)&0xFF)*0x01010101u);
3218 if(!IS_INTERLACED(mb_type^left_type[LTOP])){
3219 if(USES_LIST(left_type[LTOP], list)){
3220 const int b_xy= h->mb2b_xy[left_xy[LTOP]] + 3;
3221 const int b8_xy= 4*left_xy[LTOP] + 1;
3222 int (*ref2frm)[64] = h->ref2frm[ h->slice_table[left_xy[LTOP]]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
3223 AV_COPY32(mv_dst - 1 + 0, s->current_picture.f.motion_val[list][b_xy + b_stride*0]);
3224 AV_COPY32(mv_dst - 1 + 8, s->current_picture.f.motion_val[list][b_xy + b_stride*1]);
3225 AV_COPY32(mv_dst - 1 + 16, s->current_picture.f.motion_val[list][b_xy + b_stride*2]);
3226 AV_COPY32(mv_dst - 1 + 24, s->current_picture.f.motion_val[list][b_xy + b_stride*3]);
3228 ref_cache[-1 + 8]= ref2frm[list][s->current_picture.f.ref_index[list][b8_xy + 2*0]];
3230 ref_cache[-1 + 24]= ref2frm[list][s->current_picture.f.ref_index[list][b8_xy + 2*1]];
3232 AV_ZERO32(mv_dst - 1 + 0);
3233 AV_ZERO32(mv_dst - 1 + 8);
3234 AV_ZERO32(mv_dst - 1 +16);
3235 AV_ZERO32(mv_dst - 1 +24);
3239 ref_cache[-1 + 24]= LIST_NOT_USED;
3244 if(!USES_LIST(mb_type, list)){
3245 fill_rectangle(mv_dst, 4, 4, 8, pack16to32(0,0), 4);
3246 AV_WN32A(&ref_cache[0*8], ((LIST_NOT_USED)&0xFF)*0x01010101u);
3247 AV_WN32A(&ref_cache[1*8], ((LIST_NOT_USED)&0xFF)*0x01010101u);
3248 AV_WN32A(&ref_cache[2*8], ((LIST_NOT_USED)&0xFF)*0x01010101u);
3249 AV_WN32A(&ref_cache[3*8], ((LIST_NOT_USED)&0xFF)*0x01010101u);
3254 int8_t *ref = &s->current_picture.f.ref_index[list][4*mb_xy];
3255 int (*ref2frm)[64] = h->ref2frm[ h->slice_num&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
3256 uint32_t ref01 = (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101;
3257 uint32_t ref23 = (pack16to32(ref2frm[list][ref[2]],ref2frm[list][ref[3]])&0x00FF00FF)*0x0101;
3258 AV_WN32A(&ref_cache[0*8], ref01);
3259 AV_WN32A(&ref_cache[1*8], ref01);
3260 AV_WN32A(&ref_cache[2*8], ref23);
3261 AV_WN32A(&ref_cache[3*8], ref23);
3265 int16_t (*mv_src)[2] = &s->current_picture.f.motion_val[list][4*s->mb_x + 4*s->mb_y*b_stride];
3266 AV_COPY128(mv_dst + 8*0, mv_src + 0*b_stride);
3267 AV_COPY128(mv_dst + 8*1, mv_src + 1*b_stride);
3268 AV_COPY128(mv_dst + 8*2, mv_src + 2*b_stride);
3269 AV_COPY128(mv_dst + 8*3, mv_src + 3*b_stride);
3275 * @return non zero if the loop filter can be skiped
3277 static int fill_filter_caches(H264Context *h, int mb_type){
3278 MpegEncContext * const s = &h->s;
3279 const int mb_xy= h->mb_xy;
3280 int top_xy, left_xy[LEFT_MBS];
3281 int top_type, left_type[LEFT_MBS];
3285 top_xy = mb_xy - (s->mb_stride << MB_FIELD);
3287 /* Wow, what a mess, why didn't they simplify the interlacing & intra
3288 * stuff, I can't imagine that these complex rules are worth it. */
3290 left_xy[LBOT] = left_xy[LTOP] = mb_xy-1;
3292 const int left_mb_field_flag = IS_INTERLACED(s->current_picture.f.mb_type[mb_xy - 1]);
3293 const int curr_mb_field_flag = IS_INTERLACED(mb_type);
3295 if (left_mb_field_flag != curr_mb_field_flag) {
3296 left_xy[LTOP] -= s->mb_stride;
3299 if(curr_mb_field_flag){
3300 top_xy += s->mb_stride & (((s->current_picture.f.mb_type[top_xy] >> 7) & 1) - 1);
3302 if (left_mb_field_flag != curr_mb_field_flag) {
3303 left_xy[LBOT] += s->mb_stride;
3308 h->top_mb_xy = top_xy;
3309 h->left_mb_xy[LTOP] = left_xy[LTOP];
3310 h->left_mb_xy[LBOT] = left_xy[LBOT];
3312 //for sufficiently low qp, filtering wouldn't do anything
3313 //this is a conservative estimate: could also check beta_offset and more accurate chroma_qp
3314 int qp_thresh = h->qp_thresh; //FIXME strictly we should store qp_thresh for each mb of a slice
3315 int qp = s->current_picture.f.qscale_table[mb_xy];
3317 && (left_xy[LTOP] < 0 || ((qp + s->current_picture.f.qscale_table[left_xy[LTOP]] + 1) >> 1) <= qp_thresh)
3318 && (top_xy < 0 || ((qp + s->current_picture.f.qscale_table[top_xy ] + 1) >> 1) <= qp_thresh)) {
3321 if ((left_xy[LTOP] < 0 || ((qp + s->current_picture.f.qscale_table[left_xy[LBOT] ] + 1) >> 1) <= qp_thresh) &&
3322 (top_xy < s->mb_stride || ((qp + s->current_picture.f.qscale_table[top_xy - s->mb_stride] + 1) >> 1) <= qp_thresh))
3327 top_type = s->current_picture.f.mb_type[top_xy];
3328 left_type[LTOP] = s->current_picture.f.mb_type[left_xy[LTOP]];
3329 left_type[LBOT] = s->current_picture.f.mb_type[left_xy[LBOT]];
3330 if(h->deblocking_filter == 2){
3331 if(h->slice_table[top_xy ] != h->slice_num) top_type= 0;
3332 if(h->slice_table[left_xy[LBOT]] != h->slice_num) left_type[LTOP]= left_type[LBOT]= 0;
3334 if(h->slice_table[top_xy ] == 0xFFFF) top_type= 0;
3335 if(h->slice_table[left_xy[LBOT]] == 0xFFFF) left_type[LTOP]= left_type[LBOT] =0;
3337 h->top_type = top_type;
3338 h->left_type[LTOP]= left_type[LTOP];
3339 h->left_type[LBOT]= left_type[LBOT];
3341 if(IS_INTRA(mb_type))
3344 fill_filter_caches_inter(h, s, mb_type, top_xy, left_xy, top_type, left_type, mb_xy, 0);
3345 if(h->list_count == 2)
3346 fill_filter_caches_inter(h, s, mb_type, top_xy, left_xy, top_type, left_type, mb_xy, 1);
3348 nnz = h->non_zero_count[mb_xy];
3349 nnz_cache = h->non_zero_count_cache;
3350 AV_COPY32(&nnz_cache[4+8*1], &nnz[ 0]);
3351 AV_COPY32(&nnz_cache[4+8*2], &nnz[ 4]);
3352 AV_COPY32(&nnz_cache[4+8*3], &nnz[ 8]);
3353 AV_COPY32(&nnz_cache[4+8*4], &nnz[12]);
3354 h->cbp= h->cbp_table[mb_xy];
3357 nnz = h->non_zero_count[top_xy];
3358 AV_COPY32(&nnz_cache[4+8*0], &nnz[3*4]);
3361 if(left_type[LTOP]){
3362 nnz = h->non_zero_count[left_xy[LTOP]];
3363 nnz_cache[3+8*1]= nnz[3+0*4];
3364 nnz_cache[3+8*2]= nnz[3+1*4];
3365 nnz_cache[3+8*3]= nnz[3+2*4];
3366 nnz_cache[3+8*4]= nnz[3+3*4];
3369 // CAVLC 8x8dct requires NNZ values for residual decoding that differ from what the loop filter needs
3370 if(!CABAC && h->pps.transform_8x8_mode){
3371 if(IS_8x8DCT(top_type)){
3373 nnz_cache[5+8*0]= (h->cbp_table[top_xy] & 0x4000) >> 12;
3375 nnz_cache[7+8*0]= (h->cbp_table[top_xy] & 0x8000) >> 12;
3377 if(IS_8x8DCT(left_type[LTOP])){
3379 nnz_cache[3+8*2]= (h->cbp_table[left_xy[LTOP]]&0x2000) >> 12; //FIXME check MBAFF
3381 if(IS_8x8DCT(left_type[LBOT])){
3383 nnz_cache[3+8*4]= (h->cbp_table[left_xy[LBOT]]&0x8000) >> 12; //FIXME check MBAFF
3386 if(IS_8x8DCT(mb_type)){
3387 nnz_cache[scan8[0 ]]= nnz_cache[scan8[1 ]]=
3388 nnz_cache[scan8[2 ]]= nnz_cache[scan8[3 ]]= (h->cbp & 0x1000) >> 12;
3390 nnz_cache[scan8[0+ 4]]= nnz_cache[scan8[1+ 4]]=
3391 nnz_cache[scan8[2+ 4]]= nnz_cache[scan8[3+ 4]]= (h->cbp & 0x2000) >> 12;
3393 nnz_cache[scan8[0+ 8]]= nnz_cache[scan8[1+ 8]]=
3394 nnz_cache[scan8[2+ 8]]= nnz_cache[scan8[3+ 8]]= (h->cbp & 0x4000) >> 12;
3396 nnz_cache[scan8[0+12]]= nnz_cache[scan8[1+12]]=
3397 nnz_cache[scan8[2+12]]= nnz_cache[scan8[3+12]]= (h->cbp & 0x8000) >> 12;
3404 static void loop_filter(H264Context *h, int start_x, int end_x){
3405 MpegEncContext * const s = &h->s;
3406 uint8_t *dest_y, *dest_cb, *dest_cr;
3407 int linesize, uvlinesize, mb_x, mb_y;
3408 const int end_mb_y= s->mb_y + FRAME_MBAFF;
3409 const int old_slice_type= h->slice_type;
3410 const int pixel_shift = h->pixel_shift;
3411 const int block_h = 16 >> s->chroma_y_shift;
3413 if(h->deblocking_filter) {
3414 for(mb_x= start_x; mb_x<end_x; mb_x++){
3415 for(mb_y=end_mb_y - FRAME_MBAFF; mb_y<= end_mb_y; mb_y++){
3417 mb_xy = h->mb_xy = mb_x + mb_y*s->mb_stride;
3418 h->slice_num= h->slice_table[mb_xy];
3419 mb_type = s->current_picture.f.mb_type[mb_xy];
3420 h->list_count= h->list_counts[mb_xy];
3423 h->mb_mbaff = h->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
3427 dest_y = s->current_picture.f.data[0] + ((mb_x << pixel_shift) + mb_y * s->linesize ) * 16;
3428 dest_cb = s->current_picture.f.data[1] + (mb_x << pixel_shift) * (8 << CHROMA444) + mb_y * s->uvlinesize * block_h;
3429 dest_cr = s->current_picture.f.data[2] + (mb_x << pixel_shift) * (8 << CHROMA444) + mb_y * s->uvlinesize * block_h;
3430 //FIXME simplify above
3433 linesize = h->mb_linesize = s->linesize * 2;
3434 uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
3435 if(mb_y&1){ //FIXME move out of this function?
3436 dest_y -= s->linesize*15;
3437 dest_cb-= s->uvlinesize * (block_h - 1);
3438 dest_cr-= s->uvlinesize * (block_h - 1);
3441 linesize = h->mb_linesize = s->linesize;
3442 uvlinesize = h->mb_uvlinesize = s->uvlinesize;
3444 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);
3445 if(fill_filter_caches(h, mb_type))
3447 h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.f.qscale_table[mb_xy]);
3448 h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.f.qscale_table[mb_xy]);
3451 ff_h264_filter_mb (h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
3453 ff_h264_filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
3458 h->slice_type= old_slice_type;
3460 s->mb_y= end_mb_y - FRAME_MBAFF;
3461 h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
3462 h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
3465 static void predict_field_decoding_flag(H264Context *h){
3466 MpegEncContext * const s = &h->s;
3467 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
3468 int mb_type = (h->slice_table[mb_xy-1] == h->slice_num)
3469 ? s->current_picture.f.mb_type[mb_xy - 1]
3470 : (h->slice_table[mb_xy-s->mb_stride] == h->slice_num)
3471 ? s->current_picture.f.mb_type[mb_xy - s->mb_stride]
3473 h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
3477 * Draw edges and report progress for the last MB row.
3479 static void decode_finish_row(H264Context *h){
3480 MpegEncContext * const s = &h->s;
3481 int top = 16*(s->mb_y >> FIELD_PICTURE);
3482 int height = 16 << FRAME_MBAFF;
3483 int deblock_border = (16 + 4) << FRAME_MBAFF;
3484 int pic_height = 16*s->mb_height >> FIELD_PICTURE;
3486 if (h->deblocking_filter) {
3487 if((top + height) >= pic_height)
3488 height += deblock_border;
3490 top -= deblock_border;
3493 if (top >= pic_height || (top + height) < h->emu_edge_height)
3496 height = FFMIN(height, pic_height - top);
3497 if (top < h->emu_edge_height) {
3498 height = top+height;
3502 ff_draw_horiz_band(s, top, height);
3504 if (s->dropable) return;
3506 ff_thread_report_progress((AVFrame*)s->current_picture_ptr, top + height - 1,
3507 s->picture_structure==PICT_BOTTOM_FIELD);
3510 static int decode_slice(struct AVCodecContext *avctx, void *arg){
3511 H264Context *h = *(void**)arg;
3512 MpegEncContext * const s = &h->s;
3513 const int part_mask= s->partitioned_frame ? (AC_END|AC_ERROR) : 0x7F;
3514 int lf_x_start = s->mb_x;
3518 h->is_complex = FRAME_MBAFF || s->picture_structure != PICT_FRAME || s->codec_id != CODEC_ID_H264 ||
3519 (CONFIG_GRAY && (s->flags&CODEC_FLAG_GRAY));
3521 if( h->pps.cabac ) {
3523 align_get_bits( &s->gb );
3526 ff_init_cabac_states( &h->cabac);
3527 ff_init_cabac_decoder( &h->cabac,
3528 s->gb.buffer + get_bits_count(&s->gb)/8,
3529 (get_bits_left(&s->gb) + 7)/8);
3531 ff_h264_init_cabac_states(h);
3535 int ret = ff_h264_decode_mb_cabac(h);
3537 //STOP_TIMER("decode_mb_cabac")
3539 if(ret>=0) ff_h264_hl_decode_mb(h);
3541 if( ret >= 0 && FRAME_MBAFF ) { //FIXME optimal? or let mb_decode decode 16x32 ?
3544 ret = ff_h264_decode_mb_cabac(h);
3546 if(ret>=0) ff_h264_hl_decode_mb(h);
3549 eos = get_cabac_terminate( &h->cabac );
3551 if((s->workaround_bugs & FF_BUG_TRUNCATED) && h->cabac.bytestream > h->cabac.bytestream_end + 2){
3552 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);
3553 if (s->mb_x >= lf_x_start) loop_filter(h, lf_x_start, s->mb_x + 1);
3556 if( ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 2) {
3557 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);
3558 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);
3562 if( ++s->mb_x >= s->mb_width ) {
3563 loop_filter(h, lf_x_start, s->mb_x);
3564 s->mb_x = lf_x_start = 0;
3565 decode_finish_row(h);
3567 if(FIELD_OR_MBAFF_PICTURE) {
3569 if(FRAME_MBAFF && s->mb_y < s->mb_height)
3570 predict_field_decoding_flag(h);
3574 if( eos || s->mb_y >= s->mb_height ) {
3575 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
3576 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);
3577 if (s->mb_x > lf_x_start) loop_filter(h, lf_x_start, s->mb_x);
3584 int ret = ff_h264_decode_mb_cavlc(h);
3586 if(ret>=0) ff_h264_hl_decode_mb(h);
3588 if(ret>=0 && FRAME_MBAFF){ //FIXME optimal? or let mb_decode decode 16x32 ?
3590 ret = ff_h264_decode_mb_cavlc(h);
3592 if(ret>=0) ff_h264_hl_decode_mb(h);
3597 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
3598 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);
3602 if(++s->mb_x >= s->mb_width){
3603 loop_filter(h, lf_x_start, s->mb_x);
3604 s->mb_x = lf_x_start = 0;
3605 decode_finish_row(h);
3607 if(FIELD_OR_MBAFF_PICTURE) {
3609 if(FRAME_MBAFF && s->mb_y < s->mb_height)
3610 predict_field_decoding_flag(h);
3612 if(s->mb_y >= s->mb_height){
3613 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
3615 if(get_bits_count(&s->gb) == s->gb.size_in_bits ) {
3616 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);
3620 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);
3627 if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->mb_skip_run<=0){
3628 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
3629 if(get_bits_count(&s->gb) == s->gb.size_in_bits ){
3630 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);
3631 if (s->mb_x > lf_x_start) loop_filter(h, lf_x_start, s->mb_x);
3635 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);
3645 * Call decode_slice() for each context.
3647 * @param h h264 master context
3648 * @param context_count number of contexts to execute
3650 static int execute_decode_slices(H264Context *h, int context_count){
3651 MpegEncContext * const s = &h->s;
3652 AVCodecContext * const avctx= s->avctx;
3656 if (s->avctx->hwaccel || s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
3658 if(context_count == 1) {
3659 return decode_slice(avctx, &h);
3661 for(i = 1; i < context_count; i++) {
3662 hx = h->thread_context[i];
3663 hx->s.error_recognition = avctx->error_recognition;
3664 hx->s.error_count = 0;
3667 avctx->execute(avctx, (void *)decode_slice,
3668 h->thread_context, NULL, context_count, sizeof(void*));
3670 /* pull back stuff from slices to master context */
3671 hx = h->thread_context[context_count - 1];
3672 s->mb_x = hx->s.mb_x;
3673 s->mb_y = hx->s.mb_y;
3674 s->dropable = hx->s.dropable;
3675 s->picture_structure = hx->s.picture_structure;
3676 for(i = 1; i < context_count; i++)
3677 h->s.error_count += h->thread_context[i]->s.error_count;
3684 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size){
3685 MpegEncContext * const s = &h->s;
3686 AVCodecContext * const avctx= s->avctx;
3687 H264Context *hx; ///< thread context
3691 int pass = !(avctx->active_thread_type & FF_THREAD_FRAME);
3692 int nals_needed=0; ///< number of NALs that need decoding before the next frame thread starts
3695 h->max_contexts = (HAVE_THREADS && (s->avctx->active_thread_type&FF_THREAD_SLICE)) ? avctx->thread_count : 1;
3696 if(!(s->flags2 & CODEC_FLAG2_CHUNKS)){
3697 h->current_slice = 0;
3698 if (!s->first_field)
3699 s->current_picture_ptr= NULL;
3700 ff_h264_reset_sei(h);
3703 for(;pass <= 1;pass++){
3706 next_avc = h->is_avc ? 0 : buf_size;
3716 if(buf_index >= next_avc) {
3717 if(buf_index >= buf_size) break;
3719 for(i = 0; i < h->nal_length_size; i++)
3720 nalsize = (nalsize << 8) | buf[buf_index++];
3721 if(nalsize <= 0 || nalsize > buf_size - buf_index){
3722 av_log(h->s.avctx, AV_LOG_ERROR, "AVC: nal size %d\n", nalsize);
3725 next_avc= buf_index + nalsize;
3727 // start code prefix search
3728 for(; buf_index + 3 < next_avc; buf_index++){
3729 // This should always succeed in the first iteration.
3730 if(buf[buf_index] == 0 && buf[buf_index+1] == 0 && buf[buf_index+2] == 1)
3734 if(buf_index+3 >= buf_size) break;
3737 if(buf_index >= next_avc) continue;
3740 hx = h->thread_context[context_count];
3742 ptr= ff_h264_decode_nal(hx, buf + buf_index, &dst_length, &consumed, next_avc - buf_index);
3743 if (ptr==NULL || dst_length < 0){
3746 i= buf_index + consumed;
3747 if((s->workaround_bugs & FF_BUG_AUTODETECT) && i+3<next_avc &&
3748 buf[i]==0x00 && buf[i+1]==0x00 && buf[i+2]==0x01 && buf[i+3]==0xE0)
3749 s->workaround_bugs |= FF_BUG_TRUNCATED;
3751 if(!(s->workaround_bugs & FF_BUG_TRUNCATED)){
3752 while(ptr[dst_length - 1] == 0 && dst_length > 0)
3755 bit_length= !dst_length ? 0 : (8*dst_length - ff_h264_decode_rbsp_trailing(h, ptr + dst_length - 1));
3757 if(s->avctx->debug&FF_DEBUG_STARTCODE){
3758 av_log(h->s.avctx, AV_LOG_DEBUG, "NAL %d at %d/%d length %d\n", hx->nal_unit_type, buf_index, buf_size, dst_length);
3761 if (h->is_avc && (nalsize != consumed) && nalsize){
3762 av_log(h->s.avctx, AV_LOG_DEBUG, "AVC: Consumed only %d bytes instead of %d\n", consumed, nalsize);
3765 buf_index += consumed;
3769 // packets can sometimes contain multiple PPS/SPS
3770 // e.g. two PAFF field pictures in one packet, or a demuxer which splits NALs strangely
3771 // if so, when frame threading we can't start the next thread until we've read all of them
3772 switch (hx->nal_unit_type) {
3775 nals_needed = nal_index;
3779 init_get_bits(&hx->s.gb, ptr, bit_length);
3780 if (!get_ue_golomb(&hx->s.gb))
3781 nals_needed = nal_index;
3786 //FIXME do not discard SEI id
3787 if(avctx->skip_frame >= AVDISCARD_NONREF && h->nal_ref_idc == 0)
3792 switch(hx->nal_unit_type){
3794 if (h->nal_unit_type != NAL_IDR_SLICE) {
3795 av_log(h->s.avctx, AV_LOG_ERROR, "Invalid mix of idr and non-idr slices");
3798 idr(h); //FIXME ensure we don't loose some frames if there is reordering
3800 init_get_bits(&hx->s.gb, ptr, bit_length);
3802 hx->inter_gb_ptr= &hx->s.gb;
3803 hx->s.data_partitioning = 0;
3805 if((err = decode_slice_header(hx, h)))
3808 s->current_picture_ptr->f.key_frame |=
3809 (hx->nal_unit_type == NAL_IDR_SLICE) ||
3810 (h->sei_recovery_frame_cnt >= 0);
3812 if (h->current_slice == 1) {
3813 if(!(s->flags2 & CODEC_FLAG2_CHUNKS)) {
3814 decode_postinit(h, nal_index >= nals_needed);
3817 if (s->avctx->hwaccel && s->avctx->hwaccel->start_frame(s->avctx, NULL, 0) < 0)
3819 if(CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
3820 ff_vdpau_h264_picture_start(s);
3823 if(hx->redundant_pic_count==0
3824 && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
3825 && (avctx->skip_frame < AVDISCARD_BIDIR || hx->slice_type_nos!=AV_PICTURE_TYPE_B)
3826 && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==AV_PICTURE_TYPE_I)
3827 && avctx->skip_frame < AVDISCARD_ALL){
3828 if(avctx->hwaccel) {
3829 if (avctx->hwaccel->decode_slice(avctx, &buf[buf_index - consumed], consumed) < 0)
3832 if(CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU){
3833 static const uint8_t start_code[] = {0x00, 0x00, 0x01};
3834 ff_vdpau_add_data_chunk(s, start_code, sizeof(start_code));
3835 ff_vdpau_add_data_chunk(s, &buf[buf_index - consumed], consumed );
3841 init_get_bits(&hx->s.gb, ptr, bit_length);
3843 hx->inter_gb_ptr= NULL;
3845 if ((err = decode_slice_header(hx, h)) < 0)
3848 hx->s.data_partitioning = 1;
3852 init_get_bits(&hx->intra_gb, ptr, bit_length);
3853 hx->intra_gb_ptr= &hx->intra_gb;
3856 init_get_bits(&hx->inter_gb, ptr, bit_length);
3857 hx->inter_gb_ptr= &hx->inter_gb;
3859 if(hx->redundant_pic_count==0 && hx->intra_gb_ptr && hx->s.data_partitioning
3860 && s->context_initialized
3861 && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
3862 && (avctx->skip_frame < AVDISCARD_BIDIR || hx->slice_type_nos!=AV_PICTURE_TYPE_B)
3863 && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==AV_PICTURE_TYPE_I)
3864 && avctx->skip_frame < AVDISCARD_ALL)
3868 init_get_bits(&s->gb, ptr, bit_length);
3869 ff_h264_decode_sei(h);
3872 init_get_bits(&s->gb, ptr, bit_length);
3873 ff_h264_decode_seq_parameter_set(h);
3875 if (s->flags& CODEC_FLAG_LOW_DELAY ||
3876 (h->sps.bitstream_restriction_flag && !h->sps.num_reorder_frames))
3879 if(avctx->has_b_frames < 2)
3880 avctx->has_b_frames= !s->low_delay;
3882 if (avctx->bits_per_raw_sample != h->sps.bit_depth_luma ||
3883 h->cur_chroma_format_idc != h->sps.chroma_format_idc) {
3884 if (h->sps.bit_depth_luma >= 8 && h->sps.bit_depth_luma <= 10) {
3885 avctx->bits_per_raw_sample = h->sps.bit_depth_luma;
3886 h->cur_chroma_format_idc = h->sps.chroma_format_idc;
3887 h->pixel_shift = h->sps.bit_depth_luma > 8;
3889 ff_h264dsp_init(&h->h264dsp, h->sps.bit_depth_luma, h->sps.chroma_format_idc);
3890 ff_h264_pred_init(&h->hpc, s->codec_id, h->sps.bit_depth_luma, h->sps.chroma_format_idc);
3891 s->dsp.dct_bits = h->sps.bit_depth_luma > 8 ? 32 : 16;
3892 dsputil_init(&s->dsp, s->avctx);
3894 av_log(avctx, AV_LOG_ERROR, "Unsupported bit depth: %d\n", h->sps.bit_depth_luma);
3900 init_get_bits(&s->gb, ptr, bit_length);
3902 ff_h264_decode_picture_parameter_set(h, bit_length);
3906 case NAL_END_SEQUENCE:
3907 case NAL_END_STREAM:
3908 case NAL_FILLER_DATA:
3910 case NAL_AUXILIARY_SLICE:
3913 av_log(avctx, AV_LOG_DEBUG, "Unknown NAL code: %d (%d bits)\n", hx->nal_unit_type, bit_length);
3916 if(context_count == h->max_contexts) {
3917 execute_decode_slices(h, context_count);
3922 av_log(h->s.avctx, AV_LOG_ERROR, "decode_slice_header error\n");
3924 /* Slice could not be decoded in parallel mode, copy down
3925 * NAL unit stuff to context 0 and restart. Note that
3926 * rbsp_buffer is not transferred, but since we no longer
3927 * run in parallel mode this should not be an issue. */
3928 h->nal_unit_type = hx->nal_unit_type;
3929 h->nal_ref_idc = hx->nal_ref_idc;
3936 execute_decode_slices(h, context_count);
3941 * returns the number of bytes consumed for building the current frame
3943 static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size){
3944 if(pos==0) pos=1; //avoid infinite loops (i doubt that is needed but ...)
3945 if(pos+10>buf_size) pos=buf_size; // oops ;)
3950 static int decode_frame(AVCodecContext *avctx,
3951 void *data, int *data_size,
3954 const uint8_t *buf = avpkt->data;
3955 int buf_size = avpkt->size;
3956 H264Context *h = avctx->priv_data;
3957 MpegEncContext *s = &h->s;
3958 AVFrame *pict = data;
3961 s->flags= avctx->flags;
3962 s->flags2= avctx->flags2;
3964 /* end of stream, output what is still in the buffers */
3966 if (buf_size == 0) {
3970 s->current_picture_ptr = NULL;
3972 //FIXME factorize this with the output code below
3973 out = h->delayed_pic[0];
3975 for (i = 1; h->delayed_pic[i] && !h->delayed_pic[i]->f.key_frame && !h->delayed_pic[i]->mmco_reset; i++)
3976 if(h->delayed_pic[i]->poc < out->poc){
3977 out = h->delayed_pic[i];
3981 for(i=out_idx; h->delayed_pic[i]; i++)
3982 h->delayed_pic[i] = h->delayed_pic[i+1];
3985 *data_size = sizeof(AVFrame);
3986 *pict= *(AVFrame*)out;
3992 buf_index=decode_nal_units(h, buf, buf_size);
3996 if (!s->current_picture_ptr && h->nal_unit_type == NAL_END_SEQUENCE) {
4001 if(!(s->flags2 & CODEC_FLAG2_CHUNKS) && !s->current_picture_ptr){
4002 if (avctx->skip_frame >= AVDISCARD_NONREF)
4004 av_log(avctx, AV_LOG_ERROR, "no frame!\n");
4008 if(!(s->flags2 & CODEC_FLAG2_CHUNKS) || (s->mb_y >= s->mb_height && s->mb_height)){
4010 if(s->flags2 & CODEC_FLAG2_CHUNKS) decode_postinit(h, 1);
4014 if (!h->next_output_pic) {
4015 /* Wait for second field. */
4019 *data_size = sizeof(AVFrame);
4020 *pict = *(AVFrame*)h->next_output_pic;
4024 assert(pict->data[0] || !*data_size);
4025 ff_print_debug_info(s, pict);
4026 //printf("out %d\n", (int)pict->data[0]);
4028 return get_consumed_bytes(s, buf_index, buf_size);
4031 static inline void fill_mb_avail(H264Context *h){
4032 MpegEncContext * const s = &h->s;
4033 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
4036 h->mb_avail[0]= s->mb_x && h->slice_table[mb_xy - s->mb_stride - 1] == h->slice_num;
4037 h->mb_avail[1]= h->slice_table[mb_xy - s->mb_stride ] == h->slice_num;
4038 h->mb_avail[2]= s->mb_x+1 < s->mb_width && h->slice_table[mb_xy - s->mb_stride + 1] == h->slice_num;
4044 h->mb_avail[3]= s->mb_x && h->slice_table[mb_xy - 1] == h->slice_num;
4045 h->mb_avail[4]= 1; //FIXME move out
4046 h->mb_avail[5]= 0; //FIXME move out
4054 #define SIZE (COUNT*40)
4060 // int int_temp[10000];
4062 AVCodecContext avctx;
4064 dsputil_init(&dsp, &avctx);
4066 init_put_bits(&pb, temp, SIZE);
4067 printf("testing unsigned exp golomb\n");
4068 for(i=0; i<COUNT; i++){
4070 set_ue_golomb(&pb, i);
4071 STOP_TIMER("set_ue_golomb");
4073 flush_put_bits(&pb);
4075 init_get_bits(&gb, temp, 8*SIZE);
4076 for(i=0; i<COUNT; i++){
4079 s= show_bits(&gb, 24);
4082 j= get_ue_golomb(&gb);
4084 printf("mismatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
4087 STOP_TIMER("get_ue_golomb");
4091 init_put_bits(&pb, temp, SIZE);
4092 printf("testing signed exp golomb\n");
4093 for(i=0; i<COUNT; i++){
4095 set_se_golomb(&pb, i - COUNT/2);
4096 STOP_TIMER("set_se_golomb");
4098 flush_put_bits(&pb);
4100 init_get_bits(&gb, temp, 8*SIZE);
4101 for(i=0; i<COUNT; i++){
4104 s= show_bits(&gb, 24);
4107 j= get_se_golomb(&gb);
4108 if(j != i - COUNT/2){
4109 printf("mismatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
4112 STOP_TIMER("get_se_golomb");
4115 printf("Testing RBSP\n");
4123 av_cold void ff_h264_free_context(H264Context *h)
4127 free_tables(h, 1); //FIXME cleanup init stuff perhaps
4129 for(i = 0; i < MAX_SPS_COUNT; i++)
4130 av_freep(h->sps_buffers + i);
4132 for(i = 0; i < MAX_PPS_COUNT; i++)
4133 av_freep(h->pps_buffers + i);
4136 av_cold int ff_h264_decode_end(AVCodecContext *avctx)
4138 H264Context *h = avctx->priv_data;
4139 MpegEncContext *s = &h->s;
4141 ff_h264_free_context(h);
4145 // memset(h, 0, sizeof(H264Context));
4150 static const AVProfile profiles[] = {
4151 { FF_PROFILE_H264_BASELINE, "Baseline" },
4152 { FF_PROFILE_H264_CONSTRAINED_BASELINE, "Constrained Baseline" },
4153 { FF_PROFILE_H264_MAIN, "Main" },
4154 { FF_PROFILE_H264_EXTENDED, "Extended" },
4155 { FF_PROFILE_H264_HIGH, "High" },
4156 { FF_PROFILE_H264_HIGH_10, "High 10" },
4157 { FF_PROFILE_H264_HIGH_10_INTRA, "High 10 Intra" },
4158 { FF_PROFILE_H264_HIGH_422, "High 4:2:2" },
4159 { FF_PROFILE_H264_HIGH_422_INTRA, "High 4:2:2 Intra" },
4160 { FF_PROFILE_H264_HIGH_444, "High 4:4:4" },
4161 { FF_PROFILE_H264_HIGH_444_PREDICTIVE, "High 4:4:4 Predictive" },
4162 { FF_PROFILE_H264_HIGH_444_INTRA, "High 4:4:4 Intra" },
4163 { FF_PROFILE_H264_CAVLC_444, "CAVLC 4:4:4" },
4164 { FF_PROFILE_UNKNOWN },
4167 AVCodec ff_h264_decoder = {
4169 .type = AVMEDIA_TYPE_VIDEO,
4170 .id = CODEC_ID_H264,
4171 .priv_data_size = sizeof(H264Context),
4172 .init = ff_h264_decode_init,
4173 .close = ff_h264_decode_end,
4174 .decode = decode_frame,
4175 .capabilities = /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 | CODEC_CAP_DELAY |
4176 CODEC_CAP_SLICE_THREADS | CODEC_CAP_FRAME_THREADS,
4178 .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10"),
4179 .init_thread_copy = ONLY_IF_THREADS_ENABLED(decode_init_thread_copy),
4180 .update_thread_context = ONLY_IF_THREADS_ENABLED(decode_update_thread_context),
4181 .profiles = NULL_IF_CONFIG_SMALL(profiles),
4184 #if CONFIG_H264_VDPAU_DECODER
4185 AVCodec ff_h264_vdpau_decoder = {
4186 .name = "h264_vdpau",
4187 .type = AVMEDIA_TYPE_VIDEO,
4188 .id = CODEC_ID_H264,
4189 .priv_data_size = sizeof(H264Context),
4190 .init = ff_h264_decode_init,
4191 .close = ff_h264_decode_end,
4192 .decode = decode_frame,
4193 .capabilities = CODEC_CAP_DR1 | CODEC_CAP_DELAY | CODEC_CAP_HWACCEL_VDPAU,
4195 .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10 (VDPAU acceleration)"),
4196 .pix_fmts = (const enum PixelFormat[]){PIX_FMT_VDPAU_H264, PIX_FMT_NONE},
4197 .profiles = NULL_IF_CONFIG_SMALL(profiles),