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[] = {
63
void ff_h264_write_back_intra_pred_mode(H264Context *h){
64
int8_t *mode= h->intra4x4_pred_mode + h->mb2br_xy[h->mb_xy];
66
AV_COPY32(mode, h->intra4x4_pred_mode_cache + 4 + 8*4);
67
mode[4]= h->intra4x4_pred_mode_cache[7+8*3];
68
mode[5]= h->intra4x4_pred_mode_cache[7+8*2];
69
mode[6]= h->intra4x4_pred_mode_cache[7+8*1];
73
* checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
75
int ff_h264_check_intra4x4_pred_mode(H264Context *h){
76
MpegEncContext * const s = &h->s;
77
static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
78
static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
81
if(!(h->top_samples_available&0x8000)){
83
int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
85
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);
88
h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
93
if((h->left_samples_available&0x8888)!=0x8888){
94
static const int mask[4]={0x8000,0x2000,0x80,0x20};
96
if(!(h->left_samples_available&mask[i])){
97
int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
99
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);
102
h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
109
} //FIXME cleanup like ff_h264_check_intra_pred_mode
112
* checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
114
int ff_h264_check_intra_pred_mode(H264Context *h, int mode){
115
MpegEncContext * const s = &h->s;
116
static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
117
static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
120
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);
124
if(!(h->top_samples_available&0x8000)){
127
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);
132
if((h->left_samples_available&0x8080) != 0x8080){
134
if(h->left_samples_available&0x8080){ //mad cow disease mode, aka MBAFF + constrained_intra_pred
135
mode= ALZHEIMER_DC_L0T_PRED8x8 + (!(h->left_samples_available&0x8000)) + 2*(mode == DC_128_PRED8x8);
138
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);
146
const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src, int *dst_length, int *consumed, int length){
151
// src[0]&0x80; //forbidden bit
152
h->nal_ref_idc= src[0]>>5;
153
h->nal_unit_type= src[0]&0x1F;
157
#if HAVE_FAST_UNALIGNED
160
for(i=0; i+1<length; i+=9){
161
if(!((~AV_RN64A(src+i) & (AV_RN64A(src+i) - 0x0100010001000101ULL)) & 0x8000800080008080ULL))
164
for(i=0; i+1<length; i+=5){
165
if(!((~AV_RN32A(src+i) & (AV_RN32A(src+i) - 0x01000101U)) & 0x80008080U))
168
if(i>0 && !src[i]) i--;
172
for(i=0; i+1<length; i+=2){
174
if(i>0 && src[i-1]==0) i--;
176
if(i+2<length && src[i+1]==0 && src[i+2]<=3){
178
/* startcode, so we must be past the end */
186
if(i>=length-1){ //no escaped 0
188
*consumed= length+1; //+1 for the header
192
bufidx = h->nal_unit_type == NAL_DPC ? 1 : 0; // use second escape buffer for inter data
193
av_fast_malloc(&h->rbsp_buffer[bufidx], &h->rbsp_buffer_size[bufidx], length+FF_INPUT_BUFFER_PADDING_SIZE);
194
dst= h->rbsp_buffer[bufidx];
200
//printf("decoding esc\n");
204
//remove escapes (very rare 1:2^22)
206
dst[di++]= src[si++];
207
dst[di++]= src[si++];
208
}else if(src[si]==0 && src[si+1]==0){
209
if(src[si+2]==3){ //escape
214
}else //next start code
218
dst[di++]= src[si++];
221
dst[di++]= src[si++];
224
memset(dst+di, 0, FF_INPUT_BUFFER_PADDING_SIZE);
227
*consumed= si + 1;//+1 for the header
228
//FIXME store exact number of bits in the getbitcontext (it is needed for decoding)
233
* Identify the exact end of the bitstream
234
* @return the length of the trailing, or 0 if damaged
236
static int ff_h264_decode_rbsp_trailing(H264Context *h, const uint8_t *src){
240
tprintf(h->s.avctx, "rbsp trailing %X\n", v);
249
static inline int get_lowest_part_list_y(H264Context *h, Picture *pic, int n, int height,
250
int y_offset, int list){
251
int raw_my= h->mv_cache[list][ scan8[n] ][1];
252
int filter_height= (raw_my&3) ? 2 : 0;
253
int full_my= (raw_my>>2) + y_offset;
254
int top = full_my - filter_height, bottom = full_my + height + filter_height;
256
return FFMAX(abs(top), bottom);
259
static inline void get_lowest_part_y(H264Context *h, int refs[2][48], int n, int height,
260
int y_offset, int list0, int list1, int *nrefs){
261
MpegEncContext * const s = &h->s;
264
y_offset += 16*(s->mb_y >> MB_FIELD);
267
int ref_n = h->ref_cache[0][ scan8[n] ];
268
Picture *ref= &h->ref_list[0][ref_n];
270
// Error resilience puts the current picture in the ref list.
271
// Don't try to wait on these as it will cause a deadlock.
272
// Fields can wait on each other, though.
273
if(ref->thread_opaque != s->current_picture.thread_opaque ||
274
(ref->reference&3) != s->picture_structure) {
275
my = get_lowest_part_list_y(h, ref, n, height, y_offset, 0);
276
if (refs[0][ref_n] < 0) nrefs[0] += 1;
277
refs[0][ref_n] = FFMAX(refs[0][ref_n], my);
282
int ref_n = h->ref_cache[1][ scan8[n] ];
283
Picture *ref= &h->ref_list[1][ref_n];
285
if(ref->thread_opaque != s->current_picture.thread_opaque ||
286
(ref->reference&3) != s->picture_structure) {
287
my = get_lowest_part_list_y(h, ref, n, height, y_offset, 1);
288
if (refs[1][ref_n] < 0) nrefs[1] += 1;
289
refs[1][ref_n] = FFMAX(refs[1][ref_n], my);
295
* Wait until all reference frames are available for MC operations.
297
* @param h the H264 context
299
static void await_references(H264Context *h){
300
MpegEncContext * const s = &h->s;
301
const int mb_xy= h->mb_xy;
302
const int mb_type= s->current_picture.mb_type[mb_xy];
307
memset(refs, -1, sizeof(refs));
309
if(IS_16X16(mb_type)){
310
get_lowest_part_y(h, refs, 0, 16, 0,
311
IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs);
312
}else if(IS_16X8(mb_type)){
313
get_lowest_part_y(h, refs, 0, 8, 0,
314
IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs);
315
get_lowest_part_y(h, refs, 8, 8, 8,
316
IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1), nrefs);
317
}else if(IS_8X16(mb_type)){
318
get_lowest_part_y(h, refs, 0, 16, 0,
319
IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs);
320
get_lowest_part_y(h, refs, 4, 16, 0,
321
IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1), nrefs);
325
assert(IS_8X8(mb_type));
328
const int sub_mb_type= h->sub_mb_type[i];
330
int y_offset= (i&2)<<2;
332
if(IS_SUB_8X8(sub_mb_type)){
333
get_lowest_part_y(h, refs, n , 8, y_offset,
334
IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs);
335
}else if(IS_SUB_8X4(sub_mb_type)){
336
get_lowest_part_y(h, refs, n , 4, y_offset,
337
IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs);
338
get_lowest_part_y(h, refs, n+2, 4, y_offset+4,
339
IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs);
340
}else if(IS_SUB_4X8(sub_mb_type)){
341
get_lowest_part_y(h, refs, n , 8, y_offset,
342
IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs);
343
get_lowest_part_y(h, refs, n+1, 8, y_offset,
344
IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs);
347
assert(IS_SUB_4X4(sub_mb_type));
349
int sub_y_offset= y_offset + 2*(j&2);
350
get_lowest_part_y(h, refs, n+j, 4, sub_y_offset,
351
IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs);
357
for(list=h->list_count-1; list>=0; list--){
358
for(ref=0; ref<48 && nrefs[list]; ref++){
359
int row = refs[list][ref];
361
Picture *ref_pic = &h->ref_list[list][ref];
362
int ref_field = ref_pic->reference - 1;
363
int ref_field_picture = ref_pic->field_picture;
364
int pic_height = 16*s->mb_height >> ref_field_picture;
369
if(!FIELD_PICTURE && ref_field_picture){ // frame referencing two fields
370
ff_thread_await_progress((AVFrame*)ref_pic, FFMIN((row >> 1) - !(row&1), pic_height-1), 1);
371
ff_thread_await_progress((AVFrame*)ref_pic, FFMIN((row >> 1) , pic_height-1), 0);
372
}else if(FIELD_PICTURE && !ref_field_picture){ // field referencing one field of a frame
373
ff_thread_await_progress((AVFrame*)ref_pic, FFMIN(row*2 + ref_field , pic_height-1), 0);
374
}else if(FIELD_PICTURE){
375
ff_thread_await_progress((AVFrame*)ref_pic, FFMIN(row, pic_height-1), ref_field);
377
ff_thread_await_progress((AVFrame*)ref_pic, FFMIN(row, pic_height-1), 0);
386
* DCT transforms the 16 dc values.
387
* @param qp quantization parameter ??? FIXME
389
static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
390
// const int qmul= dequant_coeff[qp][0];
392
int temp[16]; //FIXME check if this is a good idea
393
static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
394
static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
397
const int offset= y_offset[i];
398
const int z0= block[offset+stride*0] + block[offset+stride*4];
399
const int z1= block[offset+stride*0] - block[offset+stride*4];
400
const int z2= block[offset+stride*1] - block[offset+stride*5];
401
const int z3= block[offset+stride*1] + block[offset+stride*5];
410
const int offset= x_offset[i];
411
const int z0= temp[4*0+i] + temp[4*2+i];
412
const int z1= temp[4*0+i] - temp[4*2+i];
413
const int z2= temp[4*1+i] - temp[4*3+i];
414
const int z3= temp[4*1+i] + temp[4*3+i];
416
block[stride*0 +offset]= (z0 + z3)>>1;
417
block[stride*2 +offset]= (z1 + z2)>>1;
418
block[stride*8 +offset]= (z1 - z2)>>1;
419
block[stride*10+offset]= (z0 - z3)>>1;
428
static void chroma_dc_dct_c(DCTELEM *block){
429
const int stride= 16*2;
430
const int xStride= 16;
433
a= block[stride*0 + xStride*0];
434
b= block[stride*0 + xStride*1];
435
c= block[stride*1 + xStride*0];
436
d= block[stride*1 + xStride*1];
443
block[stride*0 + xStride*0]= (a+c);
444
block[stride*0 + xStride*1]= (e+b);
445
block[stride*1 + xStride*0]= (a-c);
446
block[stride*1 + xStride*1]= (e-b);
450
static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
451
uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
452
int src_x_offset, int src_y_offset,
453
qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op,
454
int pixel_shift, int chroma444){
455
MpegEncContext * const s = &h->s;
456
const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
457
int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
458
const int luma_xy= (mx&3) + ((my&3)<<2);
459
int offset = ((mx>>2) << pixel_shift) + (my>>2)*h->mb_linesize;
460
uint8_t * src_y = pic->data[0] + offset;
461
uint8_t * src_cb, * src_cr;
462
int extra_width= h->emu_edge_width;
463
int extra_height= h->emu_edge_height;
465
const int full_mx= mx>>2;
466
const int full_my= my>>2;
467
const int pic_width = 16*s->mb_width;
468
const int pic_height = 16*s->mb_height >> MB_FIELD;
470
if(mx&7) extra_width -= 3;
471
if(my&7) extra_height -= 3;
473
if( full_mx < 0-extra_width
474
|| full_my < 0-extra_height
475
|| full_mx + 16/*FIXME*/ > pic_width + extra_width
476
|| full_my + 16/*FIXME*/ > pic_height + extra_height){
477
s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_y - (2 << pixel_shift) - 2*h->mb_linesize, h->mb_linesize, 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
478
src_y= s->edge_emu_buffer + (2 << pixel_shift) + 2*h->mb_linesize;
482
qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); //FIXME try variable height perhaps?
484
qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
487
if(CONFIG_GRAY && s->flags&CODEC_FLAG_GRAY) return;
490
src_cb = pic->data[1] + offset;
492
s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cb - (2 << pixel_shift) - 2*h->mb_linesize, h->mb_linesize,
493
16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
494
src_cb= s->edge_emu_buffer + (2 << pixel_shift) + 2*h->mb_linesize;
496
qpix_op[luma_xy](dest_cb, src_cb, h->mb_linesize); //FIXME try variable height perhaps?
498
qpix_op[luma_xy](dest_cb + delta, src_cb + delta, h->mb_linesize);
501
src_cr = pic->data[2] + offset;
503
s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cr - (2 << pixel_shift) - 2*h->mb_linesize, h->mb_linesize,
504
16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
505
src_cr= s->edge_emu_buffer + (2 << pixel_shift) + 2*h->mb_linesize;
507
qpix_op[luma_xy](dest_cr, src_cr, h->mb_linesize); //FIXME try variable height perhaps?
509
qpix_op[luma_xy](dest_cr + delta, src_cr + delta, h->mb_linesize);
515
// chroma offset when predicting from a field of opposite parity
516
my += 2 * ((s->mb_y & 1) - (pic->reference - 1));
517
emu |= (my>>3) < 0 || (my>>3) + 8 >= (pic_height>>1);
519
src_cb= pic->data[1] + ((mx>>3) << pixel_shift) + (my>>3)*h->mb_uvlinesize;
520
src_cr= pic->data[2] + ((mx>>3) << pixel_shift) + (my>>3)*h->mb_uvlinesize;
523
s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cb, h->mb_uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
524
src_cb= s->edge_emu_buffer;
526
chroma_op(dest_cb, src_cb, h->mb_uvlinesize, chroma_height, mx&7, my&7);
529
s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cr, h->mb_uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
530
src_cr= s->edge_emu_buffer;
532
chroma_op(dest_cr, src_cr, h->mb_uvlinesize, chroma_height, mx&7, my&7);
535
static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
536
uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
537
int x_offset, int y_offset,
538
qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
539
qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
540
int list0, int list1, int pixel_shift, int chroma444){
541
MpegEncContext * const s = &h->s;
542
qpel_mc_func *qpix_op= qpix_put;
543
h264_chroma_mc_func chroma_op= chroma_put;
545
dest_y += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
547
dest_cb += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
548
dest_cr += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
550
dest_cb += ( x_offset << pixel_shift) + y_offset*h->mb_uvlinesize;
551
dest_cr += ( x_offset << pixel_shift) + y_offset*h->mb_uvlinesize;
553
x_offset += 8*s->mb_x;
554
y_offset += 8*(s->mb_y >> MB_FIELD);
557
Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
558
mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
559
dest_y, dest_cb, dest_cr, x_offset, y_offset,
560
qpix_op, chroma_op, pixel_shift, chroma444);
563
chroma_op= chroma_avg;
567
Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
568
mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
569
dest_y, dest_cb, dest_cr, x_offset, y_offset,
570
qpix_op, chroma_op, pixel_shift, chroma444);
574
static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
575
uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
576
int x_offset, int y_offset,
577
qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
578
h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
579
h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
580
int list0, int list1, int pixel_shift, int chroma444){
581
MpegEncContext * const s = &h->s;
583
dest_y += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
585
chroma_weight_avg = luma_weight_avg;
586
chroma_weight_op = luma_weight_op;
587
dest_cb += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
588
dest_cr += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
590
dest_cb += ( x_offset << pixel_shift) + y_offset*h->mb_uvlinesize;
591
dest_cr += ( x_offset << pixel_shift) + y_offset*h->mb_uvlinesize;
593
x_offset += 8*s->mb_x;
594
y_offset += 8*(s->mb_y >> MB_FIELD);
597
/* don't optimize for luma-only case, since B-frames usually
598
* use implicit weights => chroma too. */
599
uint8_t *tmp_cb = s->obmc_scratchpad;
600
uint8_t *tmp_cr = s->obmc_scratchpad + (16 << pixel_shift);
601
uint8_t *tmp_y = s->obmc_scratchpad + 16*h->mb_uvlinesize;
602
int refn0 = h->ref_cache[0][ scan8[n] ];
603
int refn1 = h->ref_cache[1][ scan8[n] ];
605
mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
606
dest_y, dest_cb, dest_cr,
607
x_offset, y_offset, qpix_put, chroma_put, pixel_shift, chroma444);
608
mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
609
tmp_y, tmp_cb, tmp_cr,
610
x_offset, y_offset, qpix_put, chroma_put, pixel_shift, chroma444);
612
if(h->use_weight == 2){
613
int weight0 = h->implicit_weight[refn0][refn1][s->mb_y&1];
614
int weight1 = 64 - weight0;
615
luma_weight_avg( dest_y, tmp_y, h-> mb_linesize, 5, weight0, weight1, 0);
616
chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, 5, weight0, weight1, 0);
617
chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, 5, weight0, weight1, 0);
619
luma_weight_avg(dest_y, tmp_y, h->mb_linesize, h->luma_log2_weight_denom,
620
h->luma_weight[refn0][0][0] , h->luma_weight[refn1][1][0],
621
h->luma_weight[refn0][0][1] + h->luma_weight[refn1][1][1]);
622
chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
623
h->chroma_weight[refn0][0][0][0] , h->chroma_weight[refn1][1][0][0],
624
h->chroma_weight[refn0][0][0][1] + h->chroma_weight[refn1][1][0][1]);
625
chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
626
h->chroma_weight[refn0][0][1][0] , h->chroma_weight[refn1][1][1][0],
627
h->chroma_weight[refn0][0][1][1] + h->chroma_weight[refn1][1][1][1]);
630
int list = list1 ? 1 : 0;
631
int refn = h->ref_cache[list][ scan8[n] ];
632
Picture *ref= &h->ref_list[list][refn];
633
mc_dir_part(h, ref, n, square, chroma_height, delta, list,
634
dest_y, dest_cb, dest_cr, x_offset, y_offset,
635
qpix_put, chroma_put, pixel_shift, chroma444);
637
luma_weight_op(dest_y, h->mb_linesize, h->luma_log2_weight_denom,
638
h->luma_weight[refn][list][0], h->luma_weight[refn][list][1]);
639
if(h->use_weight_chroma){
640
chroma_weight_op(dest_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
641
h->chroma_weight[refn][list][0][0], h->chroma_weight[refn][list][0][1]);
642
chroma_weight_op(dest_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
643
h->chroma_weight[refn][list][1][0], h->chroma_weight[refn][list][1][1]);
648
static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
649
uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
650
int x_offset, int y_offset,
651
qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
652
qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
653
h264_weight_func *weight_op, h264_biweight_func *weight_avg,
654
int list0, int list1, int pixel_shift, int chroma444){
655
if((h->use_weight==2 && list0 && list1
656
&& (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ][h->s.mb_y&1] != 32))
658
mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
659
x_offset, y_offset, qpix_put, chroma_put,
660
weight_op[0], weight_op[3], weight_avg[0],
661
weight_avg[3], list0, list1, pixel_shift, chroma444);
663
mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
664
x_offset, y_offset, qpix_put, chroma_put, qpix_avg,
665
chroma_avg, list0, list1, pixel_shift, chroma444);
668
static inline void prefetch_motion(H264Context *h, int list, int pixel_shift, int chroma444){
669
/* fetch pixels for estimated mv 4 macroblocks ahead
670
* optimized for 64byte cache lines */
671
MpegEncContext * const s = &h->s;
672
const int refn = h->ref_cache[list][scan8[0]];
674
const int mx= (h->mv_cache[list][scan8[0]][0]>>2) + 16*s->mb_x + 8;
675
const int my= (h->mv_cache[list][scan8[0]][1]>>2) + 16*s->mb_y;
676
uint8_t **src= h->ref_list[list][refn].data;
677
int off= (mx << pixel_shift) + (my + (s->mb_x&3)*4)*h->mb_linesize + (64 << pixel_shift);
678
s->dsp.prefetch(src[0]+off, s->linesize, 4);
680
s->dsp.prefetch(src[1]+off, s->linesize, 4);
681
s->dsp.prefetch(src[2]+off, s->linesize, 4);
683
off= ((mx>>1) << pixel_shift) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + (64 << pixel_shift);
684
s->dsp.prefetch(src[1]+off, src[2]-src[1], 2);
689
static av_always_inline void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
690
qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
691
qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
692
h264_weight_func *weight_op, h264_biweight_func *weight_avg,
693
int pixel_shift, int chroma444){
694
MpegEncContext * const s = &h->s;
695
const int mb_xy= h->mb_xy;
696
const int mb_type= s->current_picture.mb_type[mb_xy];
698
assert(IS_INTER(mb_type));
700
if(HAVE_PTHREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME))
702
prefetch_motion(h, 0, pixel_shift, chroma444);
704
if(IS_16X16(mb_type)){
705
mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
706
qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
707
weight_op, weight_avg,
708
IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1),
709
pixel_shift, chroma444);
710
}else if(IS_16X8(mb_type)){
711
mc_part(h, 0, 0, 4, 8 << pixel_shift, dest_y, dest_cb, dest_cr, 0, 0,
712
qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
713
&weight_op[1], &weight_avg[1],
714
IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1),
715
pixel_shift, chroma444);
716
mc_part(h, 8, 0, 4, 8 << pixel_shift, dest_y, dest_cb, dest_cr, 0, 4,
717
qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
718
&weight_op[1], &weight_avg[1],
719
IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1),
720
pixel_shift, chroma444);
721
}else if(IS_8X16(mb_type)){
722
mc_part(h, 0, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 0, 0,
723
qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
724
&weight_op[2], &weight_avg[2],
725
IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1),
726
pixel_shift, chroma444);
727
mc_part(h, 4, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 4, 0,
728
qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
729
&weight_op[2], &weight_avg[2],
730
IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1),
731
pixel_shift, chroma444);
735
assert(IS_8X8(mb_type));
738
const int sub_mb_type= h->sub_mb_type[i];
740
int x_offset= (i&1)<<2;
741
int y_offset= (i&2)<<1;
743
if(IS_SUB_8X8(sub_mb_type)){
744
mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
745
qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
746
&weight_op[3], &weight_avg[3],
747
IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
748
pixel_shift, chroma444);
749
}else if(IS_SUB_8X4(sub_mb_type)){
750
mc_part(h, n , 0, 2, 4 << pixel_shift, dest_y, dest_cb, dest_cr, x_offset, y_offset,
751
qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
752
&weight_op[4], &weight_avg[4],
753
IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
754
pixel_shift, chroma444);
755
mc_part(h, n+2, 0, 2, 4 << pixel_shift, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
756
qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
757
&weight_op[4], &weight_avg[4],
758
IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
759
pixel_shift, chroma444);
760
}else if(IS_SUB_4X8(sub_mb_type)){
761
mc_part(h, n , 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
762
qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
763
&weight_op[5], &weight_avg[5],
764
IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
765
pixel_shift, chroma444);
766
mc_part(h, n+1, 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
767
qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
768
&weight_op[5], &weight_avg[5],
769
IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
770
pixel_shift, chroma444);
773
assert(IS_SUB_4X4(sub_mb_type));
775
int sub_x_offset= x_offset + 2*(j&1);
776
int sub_y_offset= y_offset + (j&2);
777
mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
778
qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
779
&weight_op[6], &weight_avg[6],
780
IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
781
pixel_shift, chroma444);
787
prefetch_motion(h, 1, pixel_shift, chroma444);
790
#define hl_motion_fn(sh, bits) \
791
static av_always_inline void hl_motion_ ## bits(H264Context *h, \
793
uint8_t *dest_cb, uint8_t *dest_cr, \
794
qpel_mc_func (*qpix_put)[16], \
795
h264_chroma_mc_func (*chroma_put), \
796
qpel_mc_func (*qpix_avg)[16], \
797
h264_chroma_mc_func (*chroma_avg), \
798
h264_weight_func *weight_op, \
799
h264_biweight_func *weight_avg, \
802
hl_motion(h, dest_y, dest_cb, dest_cr, qpix_put, chroma_put, \
803
qpix_avg, chroma_avg, weight_op, weight_avg, sh, chroma444); \
808
static void free_tables(H264Context *h, int free_rbsp){
812
av_freep(&h->intra4x4_pred_mode);
813
av_freep(&h->chroma_pred_mode_table);
814
av_freep(&h->cbp_table);
815
av_freep(&h->mvd_table[0]);
816
av_freep(&h->mvd_table[1]);
817
av_freep(&h->direct_table);
818
av_freep(&h->non_zero_count);
819
av_freep(&h->slice_table_base);
820
h->slice_table= NULL;
821
av_freep(&h->list_counts);
823
av_freep(&h->mb2b_xy);
824
av_freep(&h->mb2br_xy);
826
for(i = 0; i < MAX_THREADS; i++) {
827
hx = h->thread_context[i];
829
av_freep(&hx->top_borders[1]);
830
av_freep(&hx->top_borders[0]);
831
av_freep(&hx->s.obmc_scratchpad);
833
av_freep(&hx->rbsp_buffer[1]);
834
av_freep(&hx->rbsp_buffer[0]);
835
hx->rbsp_buffer_size[0] = 0;
836
hx->rbsp_buffer_size[1] = 0;
838
if (i) av_freep(&h->thread_context[i]);
842
static void init_dequant8_coeff_table(H264Context *h){
844
const int max_qp = 51 + 6*(h->sps.bit_depth_luma-8);
847
h->dequant8_coeff[i] = h->dequant8_buffer[i];
849
if(!memcmp(h->pps.scaling_matrix8[j], h->pps.scaling_matrix8[i], 64*sizeof(uint8_t))){
850
h->dequant8_coeff[i] = h->dequant8_buffer[j];
857
for(q=0; q<max_qp+1; q++){
861
h->dequant8_coeff[i][q][(x>>3)|((x&7)<<3)] =
862
((uint32_t)dequant8_coeff_init[idx][ dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] *
863
h->pps.scaling_matrix8[i][x]) << shift;
868
static void init_dequant4_coeff_table(H264Context *h){
870
const int max_qp = 51 + 6*(h->sps.bit_depth_luma-8);
872
h->dequant4_coeff[i] = h->dequant4_buffer[i];
874
if(!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i], 16*sizeof(uint8_t))){
875
h->dequant4_coeff[i] = h->dequant4_buffer[j];
882
for(q=0; q<max_qp+1; q++){
883
int shift = div6[q] + 2;
886
h->dequant4_coeff[i][q][(x>>2)|((x<<2)&0xF)] =
887
((uint32_t)dequant4_coeff_init[idx][(x&1) + ((x>>2)&1)] *
888
h->pps.scaling_matrix4[i][x]) << shift;
893
static void init_dequant_tables(H264Context *h){
895
init_dequant4_coeff_table(h);
896
if(h->pps.transform_8x8_mode)
897
init_dequant8_coeff_table(h);
898
if(h->sps.transform_bypass){
901
h->dequant4_coeff[i][0][x] = 1<<6;
902
if(h->pps.transform_8x8_mode)
905
h->dequant8_coeff[i][0][x] = 1<<6;
910
int ff_h264_alloc_tables(H264Context *h){
911
MpegEncContext * const s = &h->s;
912
const int big_mb_num= s->mb_stride * (s->mb_height+1);
913
const int row_mb_num= 2*s->mb_stride*s->avctx->thread_count;
916
FF_ALLOCZ_OR_GOTO(h->s.avctx, h->intra4x4_pred_mode, row_mb_num * 8 * sizeof(uint8_t), fail)
918
FF_ALLOCZ_OR_GOTO(h->s.avctx, h->non_zero_count , big_mb_num * 48 * sizeof(uint8_t), fail)
919
FF_ALLOCZ_OR_GOTO(h->s.avctx, h->slice_table_base , (big_mb_num+s->mb_stride) * sizeof(*h->slice_table_base), fail)
920
FF_ALLOCZ_OR_GOTO(h->s.avctx, h->cbp_table, big_mb_num * sizeof(uint16_t), fail)
922
FF_ALLOCZ_OR_GOTO(h->s.avctx, h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t), fail)
923
FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[0], 16*row_mb_num * sizeof(uint8_t), fail);
924
FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[1], 16*row_mb_num * sizeof(uint8_t), fail);
925
FF_ALLOCZ_OR_GOTO(h->s.avctx, h->direct_table, 4*big_mb_num * sizeof(uint8_t) , fail);
926
FF_ALLOCZ_OR_GOTO(h->s.avctx, h->list_counts, big_mb_num * sizeof(uint8_t), fail)
928
memset(h->slice_table_base, -1, (big_mb_num+s->mb_stride) * sizeof(*h->slice_table_base));
929
h->slice_table= h->slice_table_base + s->mb_stride*2 + 1;
931
FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2b_xy , big_mb_num * sizeof(uint32_t), fail);
932
FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2br_xy , big_mb_num * sizeof(uint32_t), fail);
933
for(y=0; y<s->mb_height; y++){
934
for(x=0; x<s->mb_width; x++){
935
const int mb_xy= x + y*s->mb_stride;
936
const int b_xy = 4*x + 4*y*h->b_stride;
938
h->mb2b_xy [mb_xy]= b_xy;
939
h->mb2br_xy[mb_xy]= 8*(FMO ? mb_xy : (mb_xy % (2*s->mb_stride)));
943
s->obmc_scratchpad = NULL;
945
if(!h->dequant4_coeff[0])
946
init_dequant_tables(h);
955
* Mimic alloc_tables(), but for every context thread.
957
static void clone_tables(H264Context *dst, H264Context *src, int i){
958
MpegEncContext * const s = &src->s;
959
dst->intra4x4_pred_mode = src->intra4x4_pred_mode + i*8*2*s->mb_stride;
960
dst->non_zero_count = src->non_zero_count;
961
dst->slice_table = src->slice_table;
962
dst->cbp_table = src->cbp_table;
963
dst->mb2b_xy = src->mb2b_xy;
964
dst->mb2br_xy = src->mb2br_xy;
965
dst->chroma_pred_mode_table = src->chroma_pred_mode_table;
966
dst->mvd_table[0] = src->mvd_table[0] + i*8*2*s->mb_stride;
967
dst->mvd_table[1] = src->mvd_table[1] + i*8*2*s->mb_stride;
968
dst->direct_table = src->direct_table;
969
dst->list_counts = src->list_counts;
971
dst->s.obmc_scratchpad = NULL;
972
ff_h264_pred_init(&dst->hpc, src->s.codec_id, src->sps.bit_depth_luma);
977
* Allocate buffers which are not shared amongst multiple threads.
979
static int context_init(H264Context *h){
980
FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[0], h->s.mb_width * 16*3 * sizeof(uint8_t)*2, fail)
981
FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[1], h->s.mb_width * 16*3 * sizeof(uint8_t)*2, fail)
983
h->ref_cache[0][scan8[5 ]+1] = h->ref_cache[0][scan8[7 ]+1] = h->ref_cache[0][scan8[13]+1] =
984
h->ref_cache[1][scan8[5 ]+1] = h->ref_cache[1][scan8[7 ]+1] = h->ref_cache[1][scan8[13]+1] = PART_NOT_AVAILABLE;
988
return -1; // free_tables will clean up for us
991
static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size);
993
static av_cold void common_init(H264Context *h){
994
MpegEncContext * const s = &h->s;
996
s->width = s->avctx->width;
997
s->height = s->avctx->height;
998
s->codec_id= s->avctx->codec->id;
1000
ff_h264dsp_init(&h->h264dsp, 8);
1001
ff_h264_pred_init(&h->hpc, s->codec_id, 8);
1003
h->dequant_coeff_pps= -1;
1004
s->unrestricted_mv=1;
1005
s->decode=1; //FIXME
1007
dsputil_init(&s->dsp, s->avctx); // needed so that idct permutation is known early
1009
memset(h->pps.scaling_matrix4, 16, 6*16*sizeof(uint8_t));
1010
memset(h->pps.scaling_matrix8, 16, 2*64*sizeof(uint8_t));
1013
int ff_h264_decode_extradata(H264Context *h)
1015
AVCodecContext *avctx = h->s.avctx;
1017
if(avctx->extradata[0] == 1){
1018
int i, cnt, nalsize;
1019
unsigned char *p = avctx->extradata;
1023
if(avctx->extradata_size < 7) {
1024
av_log(avctx, AV_LOG_ERROR, "avcC too short\n");
1027
/* sps and pps in the avcC always have length coded with 2 bytes,
1028
so put a fake nal_length_size = 2 while parsing them */
1029
h->nal_length_size = 2;
1030
// Decode sps from avcC
1031
cnt = *(p+5) & 0x1f; // Number of sps
1033
for (i = 0; i < cnt; i++) {
1034
nalsize = AV_RB16(p) + 2;
1035
if(decode_nal_units(h, p, nalsize) < 0) {
1036
av_log(avctx, AV_LOG_ERROR, "Decoding sps %d from avcC failed\n", i);
1041
// Decode pps from avcC
1042
cnt = *(p++); // Number of pps
1043
for (i = 0; i < cnt; i++) {
1044
nalsize = AV_RB16(p) + 2;
1045
if (decode_nal_units(h, p, nalsize) < 0) {
1046
av_log(avctx, AV_LOG_ERROR, "Decoding pps %d from avcC failed\n", i);
1051
// Now store right nal length size, that will be use to parse all other nals
1052
h->nal_length_size = (avctx->extradata[4] & 0x03) + 1;
1055
if(decode_nal_units(h, avctx->extradata, avctx->extradata_size) < 0)
1061
av_cold int ff_h264_decode_init(AVCodecContext *avctx){
1062
H264Context *h= avctx->priv_data;
1063
MpegEncContext * const s = &h->s;
1065
MPV_decode_defaults(s);
1070
s->out_format = FMT_H264;
1071
s->workaround_bugs= avctx->workaround_bugs;
1074
// s->decode_mb= ff_h263_decode_mb;
1075
s->quarter_sample = 1;
1076
if(!avctx->has_b_frames)
1079
avctx->chroma_sample_location = AVCHROMA_LOC_LEFT;
1081
ff_h264_decode_init_vlc();
1084
h->sps.bit_depth_luma = avctx->bits_per_raw_sample = 8;
1086
h->thread_context[0] = h;
1087
h->outputed_poc = h->next_outputed_poc = INT_MIN;
1088
h->prev_poc_msb= 1<<16;
1090
ff_h264_reset_sei(h);
1091
if(avctx->codec_id == CODEC_ID_H264){
1092
if(avctx->ticks_per_frame == 1){
1093
s->avctx->time_base.den *=2;
1095
avctx->ticks_per_frame = 2;
1098
if(avctx->extradata_size > 0 && avctx->extradata &&
1099
ff_h264_decode_extradata(h))
1102
if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames < h->sps.num_reorder_frames){
1103
s->avctx->has_b_frames = h->sps.num_reorder_frames;
1110
#define IN_RANGE(a, b, size) (((a) >= (b)) && ((a) < ((b)+(size))))
1111
static void copy_picture_range(Picture **to, Picture **from, int count, MpegEncContext *new_base, MpegEncContext *old_base)
1115
for (i=0; i<count; i++){
1116
assert((IN_RANGE(from[i], old_base, sizeof(*old_base)) ||
1117
IN_RANGE(from[i], old_base->picture, sizeof(Picture) * old_base->picture_count) ||
1119
to[i] = REBASE_PICTURE(from[i], new_base, old_base);
1123
static void copy_parameter_set(void **to, void **from, int count, int size)
1127
for (i=0; i<count; i++){
1128
if (to[i] && !from[i]) av_freep(&to[i]);
1129
else if (from[i] && !to[i]) to[i] = av_malloc(size);
1131
if (from[i]) memcpy(to[i], from[i], size);
1135
static int decode_init_thread_copy(AVCodecContext *avctx){
1136
H264Context *h= avctx->priv_data;
1138
if (!avctx->is_copy) return 0;
1139
memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
1140
memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
1145
#define copy_fields(to, from, start_field, end_field) memcpy(&to->start_field, &from->start_field, (char*)&to->end_field - (char*)&to->start_field)
1146
static int decode_update_thread_context(AVCodecContext *dst, const AVCodecContext *src){
1147
H264Context *h= dst->priv_data, *h1= src->priv_data;
1148
MpegEncContext * const s = &h->s, * const s1 = &h1->s;
1149
int inited = s->context_initialized, err;
1152
if(dst == src || !s1->context_initialized) return 0;
1154
err = ff_mpeg_update_thread_context(dst, src);
1157
//FIXME handle width/height changing
1159
for(i = 0; i < MAX_SPS_COUNT; i++)
1160
av_freep(h->sps_buffers + i);
1162
for(i = 0; i < MAX_PPS_COUNT; i++)
1163
av_freep(h->pps_buffers + i);
1165
memcpy(&h->s + 1, &h1->s + 1, sizeof(H264Context) - sizeof(MpegEncContext)); //copy all fields after MpegEnc
1166
memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
1167
memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
1168
if (ff_h264_alloc_tables(h) < 0) {
1169
av_log(dst, AV_LOG_ERROR, "Could not allocate memory for h264\n");
1170
return AVERROR(ENOMEM);
1175
h->rbsp_buffer[i] = NULL;
1176
h->rbsp_buffer_size[i] = 0;
1179
h->thread_context[0] = h;
1181
// frame_start may not be called for the next thread (if it's decoding a bottom field)
1182
// so this has to be allocated here
1183
h->s.obmc_scratchpad = av_malloc(16*6*s->linesize);
1185
s->dsp.clear_blocks(h->mb);
1186
s->dsp.clear_blocks(h->mb+(24*16<<h->pixel_shift));
1189
//extradata/NAL handling
1190
h->is_avc = h1->is_avc;
1193
copy_parameter_set((void**)h->sps_buffers, (void**)h1->sps_buffers, MAX_SPS_COUNT, sizeof(SPS));
1195
copy_parameter_set((void**)h->pps_buffers, (void**)h1->pps_buffers, MAX_PPS_COUNT, sizeof(PPS));
1198
//Dequantization matrices
1199
//FIXME these are big - can they be only copied when PPS changes?
1200
copy_fields(h, h1, dequant4_buffer, dequant4_coeff);
1203
h->dequant4_coeff[i] = h->dequant4_buffer[0] + (h1->dequant4_coeff[i] - h1->dequant4_buffer[0]);
1206
h->dequant8_coeff[i] = h->dequant8_buffer[0] + (h1->dequant8_coeff[i] - h1->dequant8_buffer[0]);
1208
h->dequant_coeff_pps = h1->dequant_coeff_pps;
1211
copy_fields(h, h1, poc_lsb, redundant_pic_count);
1214
copy_fields(h, h1, ref_count, list_count);
1215
copy_fields(h, h1, ref_list, intra_gb);
1216
copy_fields(h, h1, short_ref, cabac_init_idc);
1218
copy_picture_range(h->short_ref, h1->short_ref, 32, s, s1);
1219
copy_picture_range(h->long_ref, h1->long_ref, 32, s, s1);
1220
copy_picture_range(h->delayed_pic, h1->delayed_pic, MAX_DELAYED_PIC_COUNT+2, s, s1);
1222
h->last_slice_type = h1->last_slice_type;
1224
if(!s->current_picture_ptr) return 0;
1227
ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1228
h->prev_poc_msb = h->poc_msb;
1229
h->prev_poc_lsb = h->poc_lsb;
1231
h->prev_frame_num_offset= h->frame_num_offset;
1232
h->prev_frame_num = h->frame_num;
1233
h->outputed_poc = h->next_outputed_poc;
1238
int ff_h264_frame_start(H264Context *h){
1239
MpegEncContext * const s = &h->s;
1241
const int pixel_shift = h->pixel_shift;
1242
int thread_count = (s->avctx->active_thread_type & FF_THREAD_SLICE) ? s->avctx->thread_count : 1;
1244
if(MPV_frame_start(s, s->avctx) < 0)
1246
ff_er_frame_start(s);
1248
* MPV_frame_start uses pict_type to derive key_frame.
1249
* This is incorrect for H.264; IDR markings must be used.
1250
* Zero here; IDR markings per slice in frame or fields are ORed in later.
1251
* See decode_nal_units().
1253
s->current_picture_ptr->key_frame= 0;
1254
s->current_picture_ptr->mmco_reset= 0;
1256
assert(s->linesize && s->uvlinesize);
1258
for(i=0; i<16; i++){
1259
h->block_offset[i]= (4*((scan8[i] - scan8[0])&7) << pixel_shift) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
1260
h->block_offset[48+i]= (4*((scan8[i] - scan8[0])&7) << pixel_shift) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
1262
for(i=0; i<16; i++){
1263
h->block_offset[16+i]=
1264
h->block_offset[32+i]= (4*((scan8[i] - scan8[0])&7) << pixel_shift) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
1265
h->block_offset[48+16+i]=
1266
h->block_offset[48+32+i]= (4*((scan8[i] - scan8[0])&7) << pixel_shift) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
1269
/* can't be in alloc_tables because linesize isn't known there.
1270
* FIXME: redo bipred weight to not require extra buffer? */
1271
for(i = 0; i < thread_count; i++)
1272
if(h->thread_context[i] && !h->thread_context[i]->s.obmc_scratchpad)
1273
h->thread_context[i]->s.obmc_scratchpad = av_malloc(16*6*s->linesize);
1275
/* some macroblocks can be accessed before they're available in case of lost slices, mbaff or threading*/
1276
memset(h->slice_table, -1, (s->mb_height*s->mb_stride-1) * sizeof(*h->slice_table));
1278
// s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
1280
// We mark the current picture as non-reference after allocating it, so
1281
// that if we break out due to an error it can be released automatically
1282
// in the next MPV_frame_start().
1283
// SVQ3 as well as most other codecs have only last/next/current and thus
1284
// get released even with set reference, besides SVQ3 and others do not
1285
// mark frames as reference later "naturally".
1286
if(s->codec_id != CODEC_ID_SVQ3)
1287
s->current_picture_ptr->reference= 0;
1289
s->current_picture_ptr->field_poc[0]=
1290
s->current_picture_ptr->field_poc[1]= INT_MAX;
1292
h->next_output_pic = NULL;
1294
assert(s->current_picture_ptr->long_ref==0);
1300
* Run setup operations that must be run after slice header decoding.
1301
* This includes finding the next displayed frame.
1303
* @param h h264 master context
1304
* @param setup_finished enough NALs have been read that we can call
1305
* ff_thread_finish_setup()
1307
static void decode_postinit(H264Context *h, int setup_finished){
1308
MpegEncContext * const s = &h->s;
1309
Picture *out = s->current_picture_ptr;
1310
Picture *cur = s->current_picture_ptr;
1311
int i, pics, out_of_order, out_idx;
1313
s->current_picture_ptr->qscale_type= FF_QSCALE_TYPE_H264;
1314
s->current_picture_ptr->pict_type= s->pict_type;
1316
if (h->next_output_pic) return;
1318
if (cur->field_poc[0]==INT_MAX || cur->field_poc[1]==INT_MAX) {
1319
//FIXME: if we have two PAFF fields in one packet, we can't start the next thread here.
1320
//If we have one field per packet, we can. The check in decode_nal_units() is not good enough
1321
//to find this yet, so we assume the worst for now.
1322
//if (setup_finished)
1323
// ff_thread_finish_setup(s->avctx);
1327
cur->interlaced_frame = 0;
1328
cur->repeat_pict = 0;
1330
/* Signal interlacing information externally. */
1331
/* Prioritize picture timing SEI information over used decoding process if it exists. */
1333
if(h->sps.pic_struct_present_flag){
1334
switch (h->sei_pic_struct)
1336
case SEI_PIC_STRUCT_FRAME:
1338
case SEI_PIC_STRUCT_TOP_FIELD:
1339
case SEI_PIC_STRUCT_BOTTOM_FIELD:
1340
cur->interlaced_frame = 1;
1342
case SEI_PIC_STRUCT_TOP_BOTTOM:
1343
case SEI_PIC_STRUCT_BOTTOM_TOP:
1344
if (FIELD_OR_MBAFF_PICTURE)
1345
cur->interlaced_frame = 1;
1347
// try to flag soft telecine progressive
1348
cur->interlaced_frame = h->prev_interlaced_frame;
1350
case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
1351
case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
1352
// Signal the possibility of telecined film externally (pic_struct 5,6)
1353
// From these hints, let the applications decide if they apply deinterlacing.
1354
cur->repeat_pict = 1;
1356
case SEI_PIC_STRUCT_FRAME_DOUBLING:
1357
// Force progressive here, as doubling interlaced frame is a bad idea.
1358
cur->repeat_pict = 2;
1360
case SEI_PIC_STRUCT_FRAME_TRIPLING:
1361
cur->repeat_pict = 4;
1365
if ((h->sei_ct_type & 3) && h->sei_pic_struct <= SEI_PIC_STRUCT_BOTTOM_TOP)
1366
cur->interlaced_frame = (h->sei_ct_type & (1<<1)) != 0;
1368
/* Derive interlacing flag from used decoding process. */
1369
cur->interlaced_frame = FIELD_OR_MBAFF_PICTURE;
1371
h->prev_interlaced_frame = cur->interlaced_frame;
1373
if (cur->field_poc[0] != cur->field_poc[1]){
1374
/* Derive top_field_first from field pocs. */
1375
cur->top_field_first = cur->field_poc[0] < cur->field_poc[1];
1377
if(cur->interlaced_frame || h->sps.pic_struct_present_flag){
1378
/* Use picture timing SEI information. Even if it is a information of a past frame, better than nothing. */
1379
if(h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM
1380
|| h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM_TOP)
1381
cur->top_field_first = 1;
1383
cur->top_field_first = 0;
1385
/* Most likely progressive */
1386
cur->top_field_first = 0;
1390
//FIXME do something with unavailable reference frames
1392
/* Sort B-frames into display order */
1394
if(h->sps.bitstream_restriction_flag
1395
&& s->avctx->has_b_frames < h->sps.num_reorder_frames){
1396
s->avctx->has_b_frames = h->sps.num_reorder_frames;
1400
if( s->avctx->strict_std_compliance >= FF_COMPLIANCE_STRICT
1401
&& !h->sps.bitstream_restriction_flag){
1402
s->avctx->has_b_frames= MAX_DELAYED_PIC_COUNT;
1407
while(h->delayed_pic[pics]) pics++;
1409
assert(pics <= MAX_DELAYED_PIC_COUNT);
1411
h->delayed_pic[pics++] = cur;
1412
if(cur->reference == 0)
1413
cur->reference = DELAYED_PIC_REF;
1415
out = h->delayed_pic[0];
1417
for(i=1; h->delayed_pic[i] && !h->delayed_pic[i]->key_frame && !h->delayed_pic[i]->mmco_reset; i++)
1418
if(h->delayed_pic[i]->poc < out->poc){
1419
out = h->delayed_pic[i];
1422
if(s->avctx->has_b_frames == 0 && (h->delayed_pic[0]->key_frame || h->delayed_pic[0]->mmco_reset))
1423
h->next_outputed_poc= INT_MIN;
1424
out_of_order = out->poc < h->next_outputed_poc;
1426
if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames >= h->sps.num_reorder_frames)
1428
else if((out_of_order && pics-1 == s->avctx->has_b_frames && s->avctx->has_b_frames < MAX_DELAYED_PIC_COUNT)
1430
((h->next_outputed_poc != INT_MIN && out->poc > h->next_outputed_poc + 2)
1431
|| cur->pict_type == AV_PICTURE_TYPE_B)))
1434
s->avctx->has_b_frames++;
1437
if(out_of_order || pics > s->avctx->has_b_frames){
1438
out->reference &= ~DELAYED_PIC_REF;
1439
out->owner2 = s; // for frame threading, the owner must be the second field's thread
1440
// or else the first thread can release the picture and reuse it unsafely
1441
for(i=out_idx; h->delayed_pic[i]; i++)
1442
h->delayed_pic[i] = h->delayed_pic[i+1];
1444
if(!out_of_order && pics > s->avctx->has_b_frames){
1445
h->next_output_pic = out;
1446
if(out_idx==0 && h->delayed_pic[0] && (h->delayed_pic[0]->key_frame || h->delayed_pic[0]->mmco_reset)) {
1447
h->next_outputed_poc = INT_MIN;
1449
h->next_outputed_poc = out->poc;
1451
av_log(s->avctx, AV_LOG_DEBUG, "no picture\n");
1455
ff_thread_finish_setup(s->avctx);
1458
static inline void backup_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize, int chroma444, int simple){
1459
MpegEncContext * const s = &h->s;
1460
uint8_t *top_border;
1462
const int pixel_shift = h->pixel_shift;
1465
src_cb -= uvlinesize;
1466
src_cr -= uvlinesize;
1468
if(!simple && FRAME_MBAFF){
1471
top_border = h->top_borders[0][s->mb_x];
1472
AV_COPY128(top_border, src_y + 15*linesize);
1474
AV_COPY128(top_border+16, src_y+15*linesize+16);
1475
if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1478
AV_COPY128(top_border+32, src_cb + 15*uvlinesize);
1479
AV_COPY128(top_border+48, src_cb + 15*uvlinesize+16);
1480
AV_COPY128(top_border+64, src_cr + 15*uvlinesize);
1481
AV_COPY128(top_border+80, src_cr + 15*uvlinesize+16);
1483
AV_COPY128(top_border+16, src_cb + 15*uvlinesize);
1484
AV_COPY128(top_border+32, src_cr + 15*uvlinesize);
1488
AV_COPY128(top_border+32, src_cb+7*uvlinesize);
1489
AV_COPY128(top_border+48, src_cr+7*uvlinesize);
1491
AV_COPY64(top_border+16, src_cb+7*uvlinesize);
1492
AV_COPY64(top_border+24, src_cr+7*uvlinesize);
1503
top_border = h->top_borders[top_idx][s->mb_x];
1504
// There are two lines saved, the line above the the top macroblock of a pair,
1505
// and the line above the bottom macroblock
1506
AV_COPY128(top_border, src_y + 16*linesize);
1508
AV_COPY128(top_border+16, src_y+16*linesize+16);
1510
if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1513
AV_COPY128(top_border+32, src_cb + 16*linesize);
1514
AV_COPY128(top_border+48, src_cb + 16*linesize+16);
1515
AV_COPY128(top_border+64, src_cr + 16*linesize);
1516
AV_COPY128(top_border+80, src_cr + 16*linesize+16);
1518
AV_COPY128(top_border+16, src_cb + 16*linesize);
1519
AV_COPY128(top_border+32, src_cr + 16*linesize);
1523
AV_COPY128(top_border+32, src_cb+8*uvlinesize);
1524
AV_COPY128(top_border+48, src_cr+8*uvlinesize);
1526
AV_COPY64(top_border+16, src_cb+8*uvlinesize);
1527
AV_COPY64(top_border+24, src_cr+8*uvlinesize);
1533
static inline void xchg_mb_border(H264Context *h, uint8_t *src_y,
1534
uint8_t *src_cb, uint8_t *src_cr,
1535
int linesize, int uvlinesize,
1536
int xchg, int chroma444,
1537
int simple, int pixel_shift){
1538
MpegEncContext * const s = &h->s;
1539
int deblock_topleft;
1542
uint8_t *top_border_m1;
1543
uint8_t *top_border;
1545
if(!simple && FRAME_MBAFF){
1550
top_idx = MB_MBAFF ? 0 : 1;
1554
if(h->deblocking_filter == 2) {
1555
deblock_topleft = h->slice_table[h->mb_xy - 1 - s->mb_stride] == h->slice_num;
1556
deblock_top = h->top_type;
1558
deblock_topleft = (s->mb_x > 0);
1559
deblock_top = (s->mb_y > !!MB_FIELD);
1562
src_y -= linesize + 1 + pixel_shift;
1563
src_cb -= uvlinesize + 1 + pixel_shift;
1564
src_cr -= uvlinesize + 1 + pixel_shift;
1566
top_border_m1 = h->top_borders[top_idx][s->mb_x-1];
1567
top_border = h->top_borders[top_idx][s->mb_x];
1569
#define XCHG(a,b,xchg)\
1572
AV_SWAP64(b+0,a+0);\
1573
AV_SWAP64(b+8,a+8);\
1578
if (xchg) AV_SWAP64(b,a);\
1579
else AV_COPY64(b,a);
1582
if(deblock_topleft){
1583
XCHG(top_border_m1 + (8 << pixel_shift), src_y - (7 << pixel_shift), 1);
1585
XCHG(top_border + (0 << pixel_shift), src_y + (1 << pixel_shift), xchg);
1586
XCHG(top_border + (8 << pixel_shift), src_y + (9 << pixel_shift), 1);
1587
if(s->mb_x+1 < s->mb_width){
1588
XCHG(h->top_borders[top_idx][s->mb_x+1], src_y + (17 << pixel_shift), 1);
1591
if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1593
if(deblock_topleft){
1594
XCHG(top_border_m1 + (24 << pixel_shift), src_cb - (7 << pixel_shift), 1);
1595
XCHG(top_border_m1 + (40 << pixel_shift), src_cr - (7 << pixel_shift), 1);
1597
XCHG(top_border + (16 << pixel_shift), src_cb + (1 << pixel_shift), xchg);
1598
XCHG(top_border + (24 << pixel_shift), src_cb + (9 << pixel_shift), 1);
1599
XCHG(top_border + (32 << pixel_shift), src_cr + (1 << pixel_shift), xchg);
1600
XCHG(top_border + (40 << pixel_shift), src_cr + (9 << pixel_shift), 1);
1601
if(s->mb_x+1 < s->mb_width){
1602
XCHG(h->top_borders[top_idx][s->mb_x+1] + (16 << pixel_shift), src_cb + (17 << pixel_shift), 1);
1603
XCHG(h->top_borders[top_idx][s->mb_x+1] + (32 << pixel_shift), src_cr + (17 << pixel_shift), 1);
1607
if(deblock_topleft){
1608
XCHG(top_border_m1 + (16 << pixel_shift), src_cb - (7 << pixel_shift), 1);
1609
XCHG(top_border_m1 + (24 << pixel_shift), src_cr - (7 << pixel_shift), 1);
1611
XCHG(top_border + (16 << pixel_shift), src_cb+1+pixel_shift, 1);
1612
XCHG(top_border + (24 << pixel_shift), src_cr+1+pixel_shift, 1);
1618
static av_always_inline int dctcoef_get(DCTELEM *mb, int high_bit_depth, int index) {
1619
if (high_bit_depth) {
1620
return AV_RN32A(((int32_t*)mb) + index);
1622
return AV_RN16A(mb + index);
1625
static av_always_inline void dctcoef_set(DCTELEM *mb, int high_bit_depth, int index, int value) {
1626
if (high_bit_depth) {
1627
AV_WN32A(((int32_t*)mb) + index, value);
1629
AV_WN16A(mb + index, value);
1632
static av_always_inline void hl_decode_mb_predict_luma(H264Context *h, int mb_type, int is_h264, int simple, int transform_bypass,
1633
int pixel_shift, int *block_offset, int linesize, uint8_t *dest_y, int p)
1635
MpegEncContext * const s = &h->s;
1636
void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1637
void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
1639
int qscale = p == 0 ? s->qscale : h->chroma_qp[p-1];
1640
block_offset += 16*p;
1641
if(IS_INTRA4x4(mb_type)){
1642
if(simple || !s->encoding){
1643
if(IS_8x8DCT(mb_type)){
1644
if(transform_bypass){
1646
idct_add = s->dsp.add_pixels8;
1648
idct_dc_add = h->h264dsp.h264_idct8_dc_add;
1649
idct_add = h->h264dsp.h264_idct8_add;
1651
for(i=0; i<16; i+=4){
1652
uint8_t * const ptr= dest_y + block_offset[i];
1653
const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1654
if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1655
h->hpc.pred8x8l_add[dir](ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1657
const int nnz = h->non_zero_count_cache[ scan8[i+p*16] ];
1658
h->hpc.pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
1659
(h->topright_samples_available<<i)&0x4000, linesize);
1661
if(nnz == 1 && dctcoef_get(h->mb, pixel_shift, i*16+p*256))
1662
idct_dc_add(ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1664
idct_add (ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1669
if(transform_bypass){
1671
idct_add = s->dsp.add_pixels4;
1673
idct_dc_add = h->h264dsp.h264_idct_dc_add;
1674
idct_add = h->h264dsp.h264_idct_add;
1676
for(i=0; i<16; i++){
1677
uint8_t * const ptr= dest_y + block_offset[i];
1678
const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1680
if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1681
h->hpc.pred4x4_add[dir](ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1686
if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
1687
const int topright_avail= (h->topright_samples_available<<i)&0x8000;
1688
assert(s->mb_y || linesize <= block_offset[i]);
1689
if(!topright_avail){
1691
tr_high= ((uint16_t*)ptr)[3 - linesize/2]*0x0001000100010001ULL;
1692
topright= (uint8_t*) &tr_high;
1694
tr= ptr[3 - linesize]*0x01010101;
1695
topright= (uint8_t*) &tr;
1698
topright= ptr + (4 << pixel_shift) - linesize;
1702
h->hpc.pred4x4[ dir ](ptr, topright, linesize);
1703
nnz = h->non_zero_count_cache[ scan8[i+p*16] ];
1706
if(nnz == 1 && dctcoef_get(h->mb, pixel_shift, i*16+p*256))
1707
idct_dc_add(ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1709
idct_add (ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1711
ff_svq3_add_idct_c(ptr, h->mb + i*16+p*256, linesize, qscale, 0);
1718
h->hpc.pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
1720
if(h->non_zero_count_cache[ scan8[LUMA_DC_BLOCK_INDEX+p] ]){
1721
if(!transform_bypass)
1722
h->h264dsp.h264_luma_dc_dequant_idct(h->mb+(p*256 << pixel_shift), h->mb_luma_dc[p], h->dequant4_coeff[p][qscale][0]);
1724
static const uint8_t dc_mapping[16] = { 0*16, 1*16, 4*16, 5*16, 2*16, 3*16, 6*16, 7*16,
1725
8*16, 9*16,12*16,13*16,10*16,11*16,14*16,15*16};
1726
for(i = 0; i < 16; i++)
1727
dctcoef_set(h->mb+p*256, pixel_shift, dc_mapping[i], dctcoef_get(h->mb_luma_dc[p], pixel_shift, i));
1731
ff_svq3_luma_dc_dequant_idct_c(h->mb+p*256, h->mb_luma_dc[p], qscale);
1735
static av_always_inline void hl_decode_mb_idct_luma(H264Context *h, int mb_type, int is_h264, int simple, int transform_bypass,
1736
int pixel_shift, int *block_offset, int linesize, uint8_t *dest_y, int p)
1738
MpegEncContext * const s = &h->s;
1739
void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1741
block_offset += 16*p;
1742
if(!IS_INTRA4x4(mb_type)){
1744
if(IS_INTRA16x16(mb_type)){
1745
if(transform_bypass){
1746
if(h->sps.profile_idc==244 && (h->intra16x16_pred_mode==VERT_PRED8x8 || h->intra16x16_pred_mode==HOR_PRED8x8)){
1747
h->hpc.pred16x16_add[h->intra16x16_pred_mode](dest_y, block_offset, h->mb + (p*256 << pixel_shift), linesize);
1749
for(i=0; i<16; i++){
1750
if(h->non_zero_count_cache[ scan8[i+p*16] ] || dctcoef_get(h->mb, pixel_shift, i*16+p*256))
1751
s->dsp.add_pixels4(dest_y + block_offset[i], h->mb + (i*16+p*256 << pixel_shift), linesize);
1755
h->h264dsp.h264_idct_add16intra(dest_y, block_offset, h->mb + (p*256 << pixel_shift), linesize, h->non_zero_count_cache+p*5*8);
1757
}else if(h->cbp&15){
1758
if(transform_bypass){
1759
const int di = IS_8x8DCT(mb_type) ? 4 : 1;
1760
idct_add= IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4;
1761
for(i=0; i<16; i+=di){
1762
if(h->non_zero_count_cache[ scan8[i+p*16] ]){
1763
idct_add(dest_y + block_offset[i], h->mb + (i*16+p*256 << pixel_shift), linesize);
1767
if(IS_8x8DCT(mb_type)){
1768
h->h264dsp.h264_idct8_add4(dest_y, block_offset, h->mb + (p*256 << pixel_shift), linesize, h->non_zero_count_cache+p*5*8);
1770
h->h264dsp.h264_idct_add16(dest_y, block_offset, h->mb + (p*256 << pixel_shift), linesize, h->non_zero_count_cache+p*5*8);
1775
for(i=0; i<16; i++){
1776
if(h->non_zero_count_cache[ scan8[i+p*16] ] || h->mb[i*16+p*256]){ //FIXME benchmark weird rule, & below
1777
uint8_t * const ptr= dest_y + block_offset[i];
1778
ff_svq3_add_idct_c(ptr, h->mb + i*16 + p*256, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
1785
static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple, int pixel_shift){
1786
MpegEncContext * const s = &h->s;
1787
const int mb_x= s->mb_x;
1788
const int mb_y= s->mb_y;
1789
const int mb_xy= h->mb_xy;
1790
const int mb_type= s->current_picture.mb_type[mb_xy];
1791
uint8_t *dest_y, *dest_cb, *dest_cr;
1792
int linesize, uvlinesize /*dct_offset*/;
1794
int *block_offset = &h->block_offset[0];
1795
const int transform_bypass = !simple && (s->qscale == 0 && h->sps.transform_bypass);
1796
/* is_h264 should always be true if SVQ3 is disabled. */
1797
const int is_h264 = !CONFIG_SVQ3_DECODER || simple || s->codec_id == CODEC_ID_H264;
1798
void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1800
dest_y = s->current_picture.data[0] + ((mb_x << pixel_shift) + mb_y * s->linesize ) * 16;
1801
dest_cb = s->current_picture.data[1] + ((mb_x << pixel_shift) + mb_y * s->uvlinesize) * 8;
1802
dest_cr = s->current_picture.data[2] + ((mb_x << pixel_shift) + mb_y * s->uvlinesize) * 8;
1804
s->dsp.prefetch(dest_y + (s->mb_x&3)*4*s->linesize + (64 << pixel_shift), s->linesize, 4);
1805
s->dsp.prefetch(dest_cb + (s->mb_x&7)*s->uvlinesize + (64 << pixel_shift), dest_cr - dest_cb, 2);
1807
h->list_counts[mb_xy]= h->list_count;
1809
if (!simple && MB_FIELD) {
1810
linesize = h->mb_linesize = s->linesize * 2;
1811
uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
1812
block_offset = &h->block_offset[48];
1813
if(mb_y&1){ //FIXME move out of this function?
1814
dest_y -= s->linesize*15;
1815
dest_cb-= s->uvlinesize*7;
1816
dest_cr-= s->uvlinesize*7;
1820
for(list=0; list<h->list_count; list++){
1821
if(!USES_LIST(mb_type, list))
1823
if(IS_16X16(mb_type)){
1824
int8_t *ref = &h->ref_cache[list][scan8[0]];
1825
fill_rectangle(ref, 4, 4, 8, (16+*ref)^(s->mb_y&1), 1);
1827
for(i=0; i<16; i+=4){
1828
int ref = h->ref_cache[list][scan8[i]];
1830
fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, (16+ref)^(s->mb_y&1), 1);
1836
linesize = h->mb_linesize = s->linesize;
1837
uvlinesize = h->mb_uvlinesize = s->uvlinesize;
1838
// dct_offset = s->linesize * 16;
1841
if (!simple && IS_INTRA_PCM(mb_type)) {
1843
const int bit_depth = h->sps.bit_depth_luma;
1846
init_get_bits(&gb, (uint8_t*)h->mb, 384*bit_depth);
1848
for (i = 0; i < 16; i++) {
1849
uint16_t *tmp_y = (uint16_t*)(dest_y + i*linesize);
1850
for (j = 0; j < 16; j++)
1851
tmp_y[j] = get_bits(&gb, bit_depth);
1853
if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1854
for (i = 0; i < 8; i++) {
1855
uint16_t *tmp_cb = (uint16_t*)(dest_cb + i*uvlinesize);
1856
for (j = 0; j < 8; j++)
1857
tmp_cb[j] = get_bits(&gb, bit_depth);
1859
for (i = 0; i < 8; i++) {
1860
uint16_t *tmp_cr = (uint16_t*)(dest_cr + i*uvlinesize);
1861
for (j = 0; j < 8; j++)
1862
tmp_cr[j] = get_bits(&gb, bit_depth);
1866
for (i=0; i<16; i++) {
1867
memcpy(dest_y + i* linesize, h->mb + i*8, 16);
1869
if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1870
for (i=0; i<8; i++) {
1871
memcpy(dest_cb+ i*uvlinesize, h->mb + 128 + i*4, 8);
1872
memcpy(dest_cr+ i*uvlinesize, h->mb + 160 + i*4, 8);
1877
if(IS_INTRA(mb_type)){
1878
if(h->deblocking_filter)
1879
xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1, 0, simple, pixel_shift);
1881
if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1882
h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
1883
h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
1886
hl_decode_mb_predict_luma(h, mb_type, is_h264, simple, transform_bypass, pixel_shift, block_offset, linesize, dest_y, 0);
1888
if(h->deblocking_filter)
1889
xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0, 0, simple, pixel_shift);
1892
hl_motion_16(h, dest_y, dest_cb, dest_cr,
1893
s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
1894
s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
1895
h->h264dsp.weight_h264_pixels_tab,
1896
h->h264dsp.biweight_h264_pixels_tab, 0);
1898
hl_motion_8(h, dest_y, dest_cb, dest_cr,
1899
s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
1900
s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
1901
h->h264dsp.weight_h264_pixels_tab,
1902
h->h264dsp.biweight_h264_pixels_tab, 0);
1905
hl_decode_mb_idct_luma(h, mb_type, is_h264, simple, transform_bypass, pixel_shift, block_offset, linesize, dest_y, 0);
1907
if((simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)) && (h->cbp&0x30)){
1908
uint8_t *dest[2] = {dest_cb, dest_cr};
1909
if(transform_bypass){
1910
if(IS_INTRA(mb_type) && h->sps.profile_idc==244 && (h->chroma_pred_mode==VERT_PRED8x8 || h->chroma_pred_mode==HOR_PRED8x8)){
1911
h->hpc.pred8x8_add[h->chroma_pred_mode](dest[0], block_offset + 16, h->mb + (16*16*1 << pixel_shift), uvlinesize);
1912
h->hpc.pred8x8_add[h->chroma_pred_mode](dest[1], block_offset + 32, h->mb + (16*16*2 << pixel_shift), uvlinesize);
1914
idct_add = s->dsp.add_pixels4;
1916
for(i=j*16; i<j*16+4; i++){
1917
if(h->non_zero_count_cache[ scan8[i] ] || dctcoef_get(h->mb, pixel_shift, i*16))
1918
idct_add (dest[j-1] + block_offset[i], h->mb + (i*16 << pixel_shift), uvlinesize);
1924
if(h->non_zero_count_cache[ scan8[CHROMA_DC_BLOCK_INDEX+0] ])
1925
h->h264dsp.h264_chroma_dc_dequant_idct(h->mb + (16*16*1 << pixel_shift), h->dequant4_coeff[IS_INTRA(mb_type) ? 1:4][h->chroma_qp[0]][0]);
1926
if(h->non_zero_count_cache[ scan8[CHROMA_DC_BLOCK_INDEX+1] ])
1927
h->h264dsp.h264_chroma_dc_dequant_idct(h->mb + (16*16*2 << pixel_shift), h->dequant4_coeff[IS_INTRA(mb_type) ? 2:5][h->chroma_qp[1]][0]);
1928
h->h264dsp.h264_idct_add8(dest, block_offset,
1930
h->non_zero_count_cache);
1932
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]);
1933
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]);
1935
for(i=j*16; i<j*16+4; i++){
1936
if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
1937
uint8_t * const ptr= dest[j-1] + block_offset[i];
1938
ff_svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, ff_h264_chroma_qp[0][s->qscale + 12] - 12, 2);
1946
if(h->cbp || IS_INTRA(mb_type))
1948
s->dsp.clear_blocks(h->mb);
1949
s->dsp.clear_blocks(h->mb+(24*16<<pixel_shift));
1953
static av_always_inline void hl_decode_mb_444_internal(H264Context *h, int simple, int pixel_shift){
1954
MpegEncContext * const s = &h->s;
1955
const int mb_x= s->mb_x;
1956
const int mb_y= s->mb_y;
1957
const int mb_xy= h->mb_xy;
1958
const int mb_type= s->current_picture.mb_type[mb_xy];
1962
int *block_offset = &h->block_offset[0];
1963
const int transform_bypass = !simple && (s->qscale == 0 && h->sps.transform_bypass);
1964
const int plane_count = (simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)) ? 3 : 1;
1966
for (p = 0; p < plane_count; p++)
1968
dest[p] = s->current_picture.data[p] + ((mb_x << pixel_shift) + mb_y * s->linesize) * 16;
1969
s->dsp.prefetch(dest[p] + (s->mb_x&3)*4*s->linesize + (64 << pixel_shift), s->linesize, 4);
1972
h->list_counts[mb_xy]= h->list_count;
1974
if (!simple && MB_FIELD) {
1975
linesize = h->mb_linesize = h->mb_uvlinesize = s->linesize * 2;
1976
block_offset = &h->block_offset[48];
1977
if(mb_y&1) //FIXME move out of this function?
1978
for (p = 0; p < 3; p++)
1979
dest[p] -= s->linesize*15;
1982
for(list=0; list<h->list_count; list++){
1983
if(!USES_LIST(mb_type, list))
1985
if(IS_16X16(mb_type)){
1986
int8_t *ref = &h->ref_cache[list][scan8[0]];
1987
fill_rectangle(ref, 4, 4, 8, (16+*ref)^(s->mb_y&1), 1);
1989
for(i=0; i<16; i+=4){
1990
int ref = h->ref_cache[list][scan8[i]];
1992
fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, (16+ref)^(s->mb_y&1), 1);
1998
linesize = h->mb_linesize = h->mb_uvlinesize = s->linesize;
2001
if (!simple && IS_INTRA_PCM(mb_type)) {
2003
const int bit_depth = h->sps.bit_depth_luma;
2005
init_get_bits(&gb, (uint8_t*)h->mb, 768*bit_depth);
2007
for (p = 0; p < plane_count; p++) {
2008
for (i = 0; i < 16; i++) {
2009
uint16_t *tmp = (uint16_t*)(dest[p] + i*linesize);
2010
for (j = 0; j < 16; j++)
2011
tmp[j] = get_bits(&gb, bit_depth);
2015
for (p = 0; p < plane_count; p++) {
2016
for (i = 0; i < 16; i++) {
2017
memcpy(dest[p] + i*linesize, h->mb + p*128 + i*8, 16);
2022
if(IS_INTRA(mb_type)){
2023
if(h->deblocking_filter)
2024
xchg_mb_border(h, dest[0], dest[1], dest[2], linesize, linesize, 1, 1, simple, pixel_shift);
2026
for (p = 0; p < plane_count; p++)
2027
hl_decode_mb_predict_luma(h, mb_type, 1, simple, transform_bypass, pixel_shift, block_offset, linesize, dest[p], p);
2029
if(h->deblocking_filter)
2030
xchg_mb_border(h, dest[0], dest[1], dest[2], linesize, linesize, 0, 1, simple, pixel_shift);
2033
hl_motion_16(h, dest[0], dest[1], dest[2],
2034
s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
2035
s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
2036
h->h264dsp.weight_h264_pixels_tab,
2037
h->h264dsp.biweight_h264_pixels_tab, 1);
2039
hl_motion_8(h, dest[0], dest[1], dest[2],
2040
s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
2041
s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
2042
h->h264dsp.weight_h264_pixels_tab,
2043
h->h264dsp.biweight_h264_pixels_tab, 1);
2046
for (p = 0; p < plane_count; p++)
2047
hl_decode_mb_idct_luma(h, mb_type, 1, simple, transform_bypass, pixel_shift, block_offset, linesize, dest[p], p);
2049
if(h->cbp || IS_INTRA(mb_type))
2051
s->dsp.clear_blocks(h->mb);
2052
s->dsp.clear_blocks(h->mb+(24*16<<pixel_shift));
2057
* Process a macroblock; this case avoids checks for expensive uncommon cases.
2059
#define hl_decode_mb_simple(sh, bits) \
2060
static void hl_decode_mb_simple_ ## bits(H264Context *h){ \
2061
hl_decode_mb_internal(h, 1, sh); \
2063
hl_decode_mb_simple(0, 8);
2064
hl_decode_mb_simple(1, 16);
2067
* Process a macroblock; this handles edge cases, such as interlacing.
2069
static void av_noinline hl_decode_mb_complex(H264Context *h){
2070
hl_decode_mb_internal(h, 0, h->pixel_shift);
2073
static void av_noinline hl_decode_mb_444_complex(H264Context *h){
2074
hl_decode_mb_444_internal(h, 0, h->pixel_shift);
2077
static void av_noinline hl_decode_mb_444_simple(H264Context *h){
2078
hl_decode_mb_444_internal(h, 1, 0);
2081
void ff_h264_hl_decode_mb(H264Context *h){
2082
MpegEncContext * const s = &h->s;
2083
const int mb_xy= h->mb_xy;
2084
const int mb_type= s->current_picture.mb_type[mb_xy];
2085
int is_complex = CONFIG_SMALL || h->is_complex || IS_INTRA_PCM(mb_type) || s->qscale == 0;
2088
if(is_complex || h->pixel_shift)
2089
hl_decode_mb_444_complex(h);
2091
hl_decode_mb_444_simple(h);
2092
} else if (is_complex) {
2093
hl_decode_mb_complex(h);
2094
} else if (h->pixel_shift) {
2095
hl_decode_mb_simple_16(h);
2097
hl_decode_mb_simple_8(h);
2100
static int pred_weight_table(H264Context *h){
2101
MpegEncContext * const s = &h->s;
2103
int luma_def, chroma_def;
2106
h->use_weight_chroma= 0;
2107
h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
2108
if(h->sps.chroma_format_idc)
2109
h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
2110
luma_def = 1<<h->luma_log2_weight_denom;
2111
chroma_def = 1<<h->chroma_log2_weight_denom;
2113
for(list=0; list<2; list++){
2114
h->luma_weight_flag[list] = 0;
2115
h->chroma_weight_flag[list] = 0;
2116
for(i=0; i<h->ref_count[list]; i++){
2117
int luma_weight_flag, chroma_weight_flag;
2119
luma_weight_flag= get_bits1(&s->gb);
2120
if(luma_weight_flag){
2121
h->luma_weight[i][list][0]= get_se_golomb(&s->gb);
2122
h->luma_weight[i][list][1]= get_se_golomb(&s->gb);
2123
if( h->luma_weight[i][list][0] != luma_def
2124
|| h->luma_weight[i][list][1] != 0) {
2126
h->luma_weight_flag[list]= 1;
2129
h->luma_weight[i][list][0]= luma_def;
2130
h->luma_weight[i][list][1]= 0;
2133
if(h->sps.chroma_format_idc){
2134
chroma_weight_flag= get_bits1(&s->gb);
2135
if(chroma_weight_flag){
2138
h->chroma_weight[i][list][j][0]= get_se_golomb(&s->gb);
2139
h->chroma_weight[i][list][j][1]= get_se_golomb(&s->gb);
2140
if( h->chroma_weight[i][list][j][0] != chroma_def
2141
|| h->chroma_weight[i][list][j][1] != 0) {
2142
h->use_weight_chroma= 1;
2143
h->chroma_weight_flag[list]= 1;
2149
h->chroma_weight[i][list][j][0]= chroma_def;
2150
h->chroma_weight[i][list][j][1]= 0;
2155
if(h->slice_type_nos != AV_PICTURE_TYPE_B) break;
2157
h->use_weight= h->use_weight || h->use_weight_chroma;
2162
* Initialize implicit_weight table.
2163
* @param field 0/1 initialize the weight for interlaced MBAFF
2164
* -1 initializes the rest
2166
static void implicit_weight_table(H264Context *h, int field){
2167
MpegEncContext * const s = &h->s;
2168
int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
2170
for (i = 0; i < 2; i++) {
2171
h->luma_weight_flag[i] = 0;
2172
h->chroma_weight_flag[i] = 0;
2176
cur_poc = s->current_picture_ptr->poc;
2177
if( h->ref_count[0] == 1 && h->ref_count[1] == 1 && !FRAME_MBAFF
2178
&& h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
2180
h->use_weight_chroma= 0;
2184
ref_count0= h->ref_count[0];
2185
ref_count1= h->ref_count[1];
2187
cur_poc = s->current_picture_ptr->field_poc[field];
2189
ref_count0= 16+2*h->ref_count[0];
2190
ref_count1= 16+2*h->ref_count[1];
2194
h->use_weight_chroma= 2;
2195
h->luma_log2_weight_denom= 5;
2196
h->chroma_log2_weight_denom= 5;
2198
for(ref0=ref_start; ref0 < ref_count0; ref0++){
2199
int poc0 = h->ref_list[0][ref0].poc;
2200
for(ref1=ref_start; ref1 < ref_count1; ref1++){
2202
if (!h->ref_list[0][ref0].long_ref && !h->ref_list[1][ref1].long_ref) {
2203
int poc1 = h->ref_list[1][ref1].poc;
2204
int td = av_clip(poc1 - poc0, -128, 127);
2206
int tb = av_clip(cur_poc - poc0, -128, 127);
2207
int tx = (16384 + (FFABS(td) >> 1)) / td;
2208
int dist_scale_factor = (tb*tx + 32) >> 8;
2209
if(dist_scale_factor >= -64 && dist_scale_factor <= 128)
2210
w = 64 - dist_scale_factor;
2214
h->implicit_weight[ref0][ref1][0]=
2215
h->implicit_weight[ref0][ref1][1]= w;
2217
h->implicit_weight[ref0][ref1][field]=w;
2224
* instantaneous decoder refresh.
2226
static void idr(H264Context *h){
2227
ff_h264_remove_all_refs(h);
2228
h->prev_frame_num= 0;
2229
h->prev_frame_num_offset= 0;
2234
/* forget old pics after a seek */
2235
static void flush_dpb(AVCodecContext *avctx){
2236
H264Context *h= avctx->priv_data;
2238
for(i=0; i<MAX_DELAYED_PIC_COUNT; i++) {
2239
if(h->delayed_pic[i])
2240
h->delayed_pic[i]->reference= 0;
2241
h->delayed_pic[i]= NULL;
2243
h->outputed_poc=h->next_outputed_poc= INT_MIN;
2244
h->prev_interlaced_frame = 1;
2246
if(h->s.current_picture_ptr)
2247
h->s.current_picture_ptr->reference= 0;
2248
h->s.first_field= 0;
2249
ff_h264_reset_sei(h);
2250
ff_mpeg_flush(avctx);
2253
static int init_poc(H264Context *h){
2254
MpegEncContext * const s = &h->s;
2255
const int max_frame_num= 1<<h->sps.log2_max_frame_num;
2257
Picture *cur = s->current_picture_ptr;
2259
h->frame_num_offset= h->prev_frame_num_offset;
2260
if(h->frame_num < h->prev_frame_num)
2261
h->frame_num_offset += max_frame_num;
2263
if(h->sps.poc_type==0){
2264
const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
2266
if (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
2267
h->poc_msb = h->prev_poc_msb + max_poc_lsb;
2268
else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
2269
h->poc_msb = h->prev_poc_msb - max_poc_lsb;
2271
h->poc_msb = h->prev_poc_msb;
2272
//printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
2274
field_poc[1] = h->poc_msb + h->poc_lsb;
2275
if(s->picture_structure == PICT_FRAME)
2276
field_poc[1] += h->delta_poc_bottom;
2277
}else if(h->sps.poc_type==1){
2278
int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
2281
if(h->sps.poc_cycle_length != 0)
2282
abs_frame_num = h->frame_num_offset + h->frame_num;
2286
if(h->nal_ref_idc==0 && abs_frame_num > 0)
2289
expected_delta_per_poc_cycle = 0;
2290
for(i=0; i < h->sps.poc_cycle_length; i++)
2291
expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
2293
if(abs_frame_num > 0){
2294
int poc_cycle_cnt = (abs_frame_num - 1) / h->sps.poc_cycle_length;
2295
int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
2297
expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
2298
for(i = 0; i <= frame_num_in_poc_cycle; i++)
2299
expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
2303
if(h->nal_ref_idc == 0)
2304
expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
2306
field_poc[0] = expectedpoc + h->delta_poc[0];
2307
field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
2309
if(s->picture_structure == PICT_FRAME)
2310
field_poc[1] += h->delta_poc[1];
2312
int poc= 2*(h->frame_num_offset + h->frame_num);
2321
if(s->picture_structure != PICT_BOTTOM_FIELD)
2322
s->current_picture_ptr->field_poc[0]= field_poc[0];
2323
if(s->picture_structure != PICT_TOP_FIELD)
2324
s->current_picture_ptr->field_poc[1]= field_poc[1];
2325
cur->poc= FFMIN(cur->field_poc[0], cur->field_poc[1]);
2332
* initialize scan tables
2334
static void init_scan_tables(H264Context *h){
2336
for(i=0; i<16; i++){
2337
#define T(x) (x>>2) | ((x<<2) & 0xF)
2338
h->zigzag_scan[i] = T(zigzag_scan[i]);
2339
h-> field_scan[i] = T( field_scan[i]);
2342
for(i=0; i<64; i++){
2343
#define T(x) (x>>3) | ((x&7)<<3)
2344
h->zigzag_scan8x8[i] = T(ff_zigzag_direct[i]);
2345
h->zigzag_scan8x8_cavlc[i] = T(zigzag_scan8x8_cavlc[i]);
2346
h->field_scan8x8[i] = T(field_scan8x8[i]);
2347
h->field_scan8x8_cavlc[i] = T(field_scan8x8_cavlc[i]);
2350
if(h->sps.transform_bypass){ //FIXME same ugly
2351
h->zigzag_scan_q0 = zigzag_scan;
2352
h->zigzag_scan8x8_q0 = ff_zigzag_direct;
2353
h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc;
2354
h->field_scan_q0 = field_scan;
2355
h->field_scan8x8_q0 = field_scan8x8;
2356
h->field_scan8x8_cavlc_q0 = field_scan8x8_cavlc;
2358
h->zigzag_scan_q0 = h->zigzag_scan;
2359
h->zigzag_scan8x8_q0 = h->zigzag_scan8x8;
2360
h->zigzag_scan8x8_cavlc_q0 = h->zigzag_scan8x8_cavlc;
2361
h->field_scan_q0 = h->field_scan;
2362
h->field_scan8x8_q0 = h->field_scan8x8;
2363
h->field_scan8x8_cavlc_q0 = h->field_scan8x8_cavlc;
2367
static void field_end(H264Context *h, int in_setup){
2368
MpegEncContext * const s = &h->s;
2369
AVCodecContext * const avctx= s->avctx;
2372
if (!in_setup && !s->dropable)
2373
ff_thread_report_progress((AVFrame*)s->current_picture_ptr, (16*s->mb_height >> FIELD_PICTURE) - 1,
2374
s->picture_structure==PICT_BOTTOM_FIELD);
2376
if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
2377
ff_vdpau_h264_set_reference_frames(s);
2379
if(in_setup || !(avctx->active_thread_type&FF_THREAD_FRAME)){
2381
ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
2382
h->prev_poc_msb= h->poc_msb;
2383
h->prev_poc_lsb= h->poc_lsb;
2385
h->prev_frame_num_offset= h->frame_num_offset;
2386
h->prev_frame_num= h->frame_num;
2387
h->outputed_poc = h->next_outputed_poc;
2390
if (avctx->hwaccel) {
2391
if (avctx->hwaccel->end_frame(avctx) < 0)
2392
av_log(avctx, AV_LOG_ERROR, "hardware accelerator failed to decode picture\n");
2395
if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
2396
ff_vdpau_h264_picture_complete(s);
2399
* FIXME: Error handling code does not seem to support interlaced
2400
* when slices span multiple rows
2401
* The ff_er_add_slice calls don't work right for bottom
2402
* fields; they cause massive erroneous error concealing
2403
* Error marking covers both fields (top and bottom).
2404
* This causes a mismatched s->error_count
2405
* and a bad error table. Further, the error count goes to
2406
* INT_MAX when called for bottom field, because mb_y is
2407
* past end by one (callers fault) and resync_mb_y != 0
2408
* causes problems for the first MB line, too.
2419
* Replicate H264 "master" context to thread contexts.
2421
static void clone_slice(H264Context *dst, H264Context *src)
2423
memcpy(dst->block_offset, src->block_offset, sizeof(dst->block_offset));
2424
dst->s.current_picture_ptr = src->s.current_picture_ptr;
2425
dst->s.current_picture = src->s.current_picture;
2426
dst->s.linesize = src->s.linesize;
2427
dst->s.uvlinesize = src->s.uvlinesize;
2428
dst->s.first_field = src->s.first_field;
2430
dst->prev_poc_msb = src->prev_poc_msb;
2431
dst->prev_poc_lsb = src->prev_poc_lsb;
2432
dst->prev_frame_num_offset = src->prev_frame_num_offset;
2433
dst->prev_frame_num = src->prev_frame_num;
2434
dst->short_ref_count = src->short_ref_count;
2436
memcpy(dst->short_ref, src->short_ref, sizeof(dst->short_ref));
2437
memcpy(dst->long_ref, src->long_ref, sizeof(dst->long_ref));
2438
memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
2439
memcpy(dst->ref_list, src->ref_list, sizeof(dst->ref_list));
2441
memcpy(dst->dequant4_coeff, src->dequant4_coeff, sizeof(src->dequant4_coeff));
2442
memcpy(dst->dequant8_coeff, src->dequant8_coeff, sizeof(src->dequant8_coeff));
2446
* computes profile from profile_idc and constraint_set?_flags
2450
* @return profile as defined by FF_PROFILE_H264_*
2452
int ff_h264_get_profile(SPS *sps)
2454
int profile = sps->profile_idc;
2456
switch(sps->profile_idc) {
2457
case FF_PROFILE_H264_BASELINE:
2458
// constraint_set1_flag set to 1
2459
profile |= (sps->constraint_set_flags & 1<<1) ? FF_PROFILE_H264_CONSTRAINED : 0;
2461
case FF_PROFILE_H264_HIGH_10:
2462
case FF_PROFILE_H264_HIGH_422:
2463
case FF_PROFILE_H264_HIGH_444_PREDICTIVE:
2464
// constraint_set3_flag set to 1
2465
profile |= (sps->constraint_set_flags & 1<<3) ? FF_PROFILE_H264_INTRA : 0;
2473
* decodes a slice header.
2474
* This will also call MPV_common_init() and frame_start() as needed.
2476
* @param h h264context
2477
* @param h0 h264 master context (differs from 'h' when doing sliced based parallel decoding)
2479
* @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
2481
static int decode_slice_header(H264Context *h, H264Context *h0){
2482
MpegEncContext * const s = &h->s;
2483
MpegEncContext * const s0 = &h0->s;
2484
unsigned int first_mb_in_slice;
2485
unsigned int pps_id;
2486
int num_ref_idx_active_override_flag;
2487
unsigned int slice_type, tmp, i, j;
2488
int default_ref_list_done = 0;
2489
int last_pic_structure;
2491
s->dropable= h->nal_ref_idc == 0;
2493
/* FIXME: 2tap qpel isn't implemented for high bit depth. */
2494
if((s->avctx->flags2 & CODEC_FLAG2_FAST) && !h->nal_ref_idc && !h->pixel_shift){
2495
s->me.qpel_put= s->dsp.put_2tap_qpel_pixels_tab;
2496
s->me.qpel_avg= s->dsp.avg_2tap_qpel_pixels_tab;
2498
s->me.qpel_put= s->dsp.put_h264_qpel_pixels_tab;
2499
s->me.qpel_avg= s->dsp.avg_h264_qpel_pixels_tab;
2502
first_mb_in_slice= get_ue_golomb(&s->gb);
2504
if(first_mb_in_slice == 0){ //FIXME better field boundary detection
2505
if(h0->current_slice && FIELD_PICTURE){
2509
h0->current_slice = 0;
2510
if (!s0->first_field)
2511
s->current_picture_ptr= NULL;
2514
slice_type= get_ue_golomb_31(&s->gb);
2516
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);
2521
h->slice_type_fixed=1;
2523
h->slice_type_fixed=0;
2525
slice_type= golomb_to_pict_type[ slice_type ];
2526
if (slice_type == AV_PICTURE_TYPE_I
2527
|| (h0->current_slice != 0 && slice_type == h0->last_slice_type) ) {
2528
default_ref_list_done = 1;
2530
h->slice_type= slice_type;
2531
h->slice_type_nos= slice_type & 3;
2533
s->pict_type= h->slice_type; // to make a few old functions happy, it's wrong though
2535
pps_id= get_ue_golomb(&s->gb);
2536
if(pps_id>=MAX_PPS_COUNT){
2537
av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
2540
if(!h0->pps_buffers[pps_id]) {
2541
av_log(h->s.avctx, AV_LOG_ERROR, "non-existing PPS %u referenced\n", pps_id);
2544
h->pps= *h0->pps_buffers[pps_id];
2546
if(!h0->sps_buffers[h->pps.sps_id]) {
2547
av_log(h->s.avctx, AV_LOG_ERROR, "non-existing SPS %u referenced\n", h->pps.sps_id);
2550
h->sps = *h0->sps_buffers[h->pps.sps_id];
2552
s->avctx->profile = ff_h264_get_profile(&h->sps);
2553
s->avctx->level = h->sps.level_idc;
2554
s->avctx->refs = h->sps.ref_frame_count;
2556
if(h == h0 && h->dequant_coeff_pps != pps_id){
2557
h->dequant_coeff_pps = pps_id;
2558
init_dequant_tables(h);
2561
s->mb_width= h->sps.mb_width;
2562
s->mb_height= h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
2564
h->b_stride= s->mb_width*4;
2566
s->width = 16*s->mb_width - (2>>CHROMA444)*FFMIN(h->sps.crop_right, (8<<CHROMA444)-1);
2567
if(h->sps.frame_mbs_only_flag)
2568
s->height= 16*s->mb_height - (2>>CHROMA444)*FFMIN(h->sps.crop_bottom, (8<<CHROMA444)-1);
2570
s->height= 16*s->mb_height - (4>>CHROMA444)*FFMIN(h->sps.crop_bottom, (8<<CHROMA444)-1);
2572
if (s->context_initialized
2573
&& ( s->width != s->avctx->width || s->height != s->avctx->height
2574
|| av_cmp_q(h->sps.sar, s->avctx->sample_aspect_ratio))) {
2576
av_log_missing_feature(s->avctx, "Width/height changing with threads is", 0);
2577
return -1; // width / height changed during parallelized decoding
2580
flush_dpb(s->avctx);
2583
if (!s->context_initialized) {
2585
av_log(h->s.avctx, AV_LOG_ERROR, "Cannot (re-)initialize context during parallel decoding.\n");
2589
avcodec_set_dimensions(s->avctx, s->width, s->height);
2590
s->avctx->sample_aspect_ratio= h->sps.sar;
2591
av_assert0(s->avctx->sample_aspect_ratio.den);
2593
h->s.avctx->coded_width = 16*s->mb_width;
2594
h->s.avctx->coded_height = 16*s->mb_height;
2596
if(h->sps.video_signal_type_present_flag){
2597
s->avctx->color_range = h->sps.full_range ? AVCOL_RANGE_JPEG : AVCOL_RANGE_MPEG;
2598
if(h->sps.colour_description_present_flag){
2599
s->avctx->color_primaries = h->sps.color_primaries;
2600
s->avctx->color_trc = h->sps.color_trc;
2601
s->avctx->colorspace = h->sps.colorspace;
2605
if(h->sps.timing_info_present_flag){
2606
int64_t den= h->sps.time_scale;
2607
if(h->x264_build < 44U)
2609
av_reduce(&s->avctx->time_base.num, &s->avctx->time_base.den,
2610
h->sps.num_units_in_tick, den, 1<<30);
2613
switch (h->sps.bit_depth_luma) {
2615
s->avctx->pix_fmt = CHROMA444 ? PIX_FMT_YUV444P9 : PIX_FMT_YUV420P9;
2618
s->avctx->pix_fmt = CHROMA444 ? PIX_FMT_YUV444P10 : PIX_FMT_YUV420P10;
2622
s->avctx->pix_fmt = s->avctx->color_range == AVCOL_RANGE_JPEG ? PIX_FMT_YUVJ444P : PIX_FMT_YUV444P;
2624
s->avctx->pix_fmt = s->avctx->get_format(s->avctx,
2625
s->avctx->codec->pix_fmts ?
2626
s->avctx->codec->pix_fmts :
2627
s->avctx->color_range == AVCOL_RANGE_JPEG ?
2628
hwaccel_pixfmt_list_h264_jpeg_420 :
2629
ff_hwaccel_pixfmt_list_420);
2633
s->avctx->hwaccel = ff_find_hwaccel(s->avctx->codec->id, s->avctx->pix_fmt);
2635
if (MPV_common_init(s) < 0) {
2636
av_log(h->s.avctx, AV_LOG_ERROR, "MPV_common_init() failed.\n");
2640
h->prev_interlaced_frame = 1;
2642
init_scan_tables(h);
2643
if (ff_h264_alloc_tables(h) < 0) {
2644
av_log(h->s.avctx, AV_LOG_ERROR, "Could not allocate memory for h264\n");
2645
return AVERROR(ENOMEM);
2648
if (!HAVE_THREADS || !(s->avctx->active_thread_type&FF_THREAD_SLICE)) {
2649
if (context_init(h) < 0) {
2650
av_log(h->s.avctx, AV_LOG_ERROR, "context_init() failed.\n");
2654
for(i = 1; i < s->avctx->thread_count; i++) {
2656
c = h->thread_context[i] = av_malloc(sizeof(H264Context));
2657
memcpy(c, h->s.thread_context[i], sizeof(MpegEncContext));
2658
memset(&c->s + 1, 0, sizeof(H264Context) - sizeof(MpegEncContext));
2659
c->h264dsp = h->h264dsp;
2662
c->pixel_shift = h->pixel_shift;
2663
init_scan_tables(c);
2664
clone_tables(c, h, i);
2667
for(i = 0; i < s->avctx->thread_count; i++)
2668
if (context_init(h->thread_context[i]) < 0) {
2669
av_log(h->s.avctx, AV_LOG_ERROR, "context_init() failed.\n");
2675
h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
2678
h->mb_aff_frame = 0;
2679
last_pic_structure = s0->picture_structure;
2680
if(h->sps.frame_mbs_only_flag){
2681
s->picture_structure= PICT_FRAME;
2683
if(get_bits1(&s->gb)) { //field_pic_flag
2684
s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
2686
s->picture_structure= PICT_FRAME;
2687
h->mb_aff_frame = h->sps.mb_aff;
2690
h->mb_field_decoding_flag= s->picture_structure != PICT_FRAME;
2692
if(h0->current_slice == 0){
2693
// Shorten frame num gaps so we don't have to allocate reference frames just to throw them away
2694
if(h->frame_num != h->prev_frame_num) {
2695
int unwrap_prev_frame_num = h->prev_frame_num, max_frame_num = 1<<h->sps.log2_max_frame_num;
2697
if (unwrap_prev_frame_num > h->frame_num) unwrap_prev_frame_num -= max_frame_num;
2699
if ((h->frame_num - unwrap_prev_frame_num) > h->sps.ref_frame_count) {
2700
unwrap_prev_frame_num = (h->frame_num - h->sps.ref_frame_count) - 1;
2701
if (unwrap_prev_frame_num < 0)
2702
unwrap_prev_frame_num += max_frame_num;
2704
h->prev_frame_num = unwrap_prev_frame_num;
2708
while(h->frame_num != h->prev_frame_num &&
2709
h->frame_num != (h->prev_frame_num+1)%(1<<h->sps.log2_max_frame_num)){
2710
Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
2711
av_log(h->s.avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n", h->frame_num, h->prev_frame_num);
2712
if (ff_h264_frame_start(h) < 0)
2714
h->prev_frame_num++;
2715
h->prev_frame_num %= 1<<h->sps.log2_max_frame_num;
2716
s->current_picture_ptr->frame_num= h->prev_frame_num;
2717
ff_thread_report_progress((AVFrame*)s->current_picture_ptr, INT_MAX, 0);
2718
ff_thread_report_progress((AVFrame*)s->current_picture_ptr, INT_MAX, 1);
2719
ff_generate_sliding_window_mmcos(h);
2720
ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
2721
/* Error concealment: if a ref is missing, copy the previous ref in its place.
2722
* FIXME: avoiding a memcpy would be nice, but ref handling makes many assumptions
2723
* about there being no actual duplicates.
2724
* FIXME: this doesn't copy padding for out-of-frame motion vectors. Given we're
2725
* concealing a lost frame, this probably isn't noticable by comparison, but it should
2727
if (h->short_ref_count) {
2729
av_image_copy(h->short_ref[0]->data, h->short_ref[0]->linesize,
2730
(const uint8_t**)prev->data, prev->linesize,
2731
s->avctx->pix_fmt, s->mb_width*16, s->mb_height*16);
2732
h->short_ref[0]->poc = prev->poc+2;
2734
h->short_ref[0]->frame_num = h->prev_frame_num;
2738
/* See if we have a decoded first field looking for a pair... */
2739
if (s0->first_field) {
2740
assert(s0->current_picture_ptr);
2741
assert(s0->current_picture_ptr->data[0]);
2742
assert(s0->current_picture_ptr->reference != DELAYED_PIC_REF);
2744
/* figure out if we have a complementary field pair */
2745
if (!FIELD_PICTURE || s->picture_structure == last_pic_structure) {
2747
* Previous field is unmatched. Don't display it, but let it
2748
* remain for reference if marked as such.
2750
s0->current_picture_ptr = NULL;
2751
s0->first_field = FIELD_PICTURE;
2754
if (h->nal_ref_idc &&
2755
s0->current_picture_ptr->reference &&
2756
s0->current_picture_ptr->frame_num != h->frame_num) {
2758
* This and previous field were reference, but had
2759
* different frame_nums. Consider this field first in
2760
* pair. Throw away previous field except for reference
2763
s0->first_field = 1;
2764
s0->current_picture_ptr = NULL;
2767
/* Second field in complementary pair */
2768
s0->first_field = 0;
2773
/* Frame or first field in a potentially complementary pair */
2774
assert(!s0->current_picture_ptr);
2775
s0->first_field = FIELD_PICTURE;
2778
if(!FIELD_PICTURE || s0->first_field) {
2779
if (ff_h264_frame_start(h) < 0) {
2780
s0->first_field = 0;
2784
ff_release_unused_pictures(s, 0);
2790
s->current_picture_ptr->frame_num= h->frame_num; //FIXME frame_num cleanup
2792
assert(s->mb_num == s->mb_width * s->mb_height);
2793
if(first_mb_in_slice << FIELD_OR_MBAFF_PICTURE >= s->mb_num ||
2794
first_mb_in_slice >= s->mb_num){
2795
av_log(h->s.avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
2798
s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
2799
s->resync_mb_y = s->mb_y = (first_mb_in_slice / s->mb_width) << FIELD_OR_MBAFF_PICTURE;
2800
if (s->picture_structure == PICT_BOTTOM_FIELD)
2801
s->resync_mb_y = s->mb_y = s->mb_y + 1;
2802
assert(s->mb_y < s->mb_height);
2804
if(s->picture_structure==PICT_FRAME){
2805
h->curr_pic_num= h->frame_num;
2806
h->max_pic_num= 1<< h->sps.log2_max_frame_num;
2808
h->curr_pic_num= 2*h->frame_num + 1;
2809
h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
2812
if(h->nal_unit_type == NAL_IDR_SLICE){
2813
get_ue_golomb(&s->gb); /* idr_pic_id */
2816
if(h->sps.poc_type==0){
2817
h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
2819
if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
2820
h->delta_poc_bottom= get_se_golomb(&s->gb);
2824
if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
2825
h->delta_poc[0]= get_se_golomb(&s->gb);
2827
if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
2828
h->delta_poc[1]= get_se_golomb(&s->gb);
2833
if(h->pps.redundant_pic_cnt_present){
2834
h->redundant_pic_count= get_ue_golomb(&s->gb);
2837
//set defaults, might be overridden a few lines later
2838
h->ref_count[0]= h->pps.ref_count[0];
2839
h->ref_count[1]= h->pps.ref_count[1];
2841
if(h->slice_type_nos != AV_PICTURE_TYPE_I){
2842
if(h->slice_type_nos == AV_PICTURE_TYPE_B){
2843
h->direct_spatial_mv_pred= get_bits1(&s->gb);
2845
num_ref_idx_active_override_flag= get_bits1(&s->gb);
2847
if(num_ref_idx_active_override_flag){
2848
h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
2849
if(h->slice_type_nos==AV_PICTURE_TYPE_B)
2850
h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
2852
if(h->ref_count[0]-1 > 32-1 || h->ref_count[1]-1 > 32-1){
2853
av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
2854
h->ref_count[0]= h->ref_count[1]= 1;
2858
if(h->slice_type_nos == AV_PICTURE_TYPE_B)
2865
if(!default_ref_list_done){
2866
ff_h264_fill_default_ref_list(h);
2869
if(h->slice_type_nos!=AV_PICTURE_TYPE_I && ff_h264_decode_ref_pic_list_reordering(h) < 0)
2872
if(h->slice_type_nos!=AV_PICTURE_TYPE_I){
2873
s->last_picture_ptr= &h->ref_list[0][0];
2874
ff_copy_picture(&s->last_picture, s->last_picture_ptr);
2876
if(h->slice_type_nos==AV_PICTURE_TYPE_B){
2877
s->next_picture_ptr= &h->ref_list[1][0];
2878
ff_copy_picture(&s->next_picture, s->next_picture_ptr);
2881
if( (h->pps.weighted_pred && h->slice_type_nos == AV_PICTURE_TYPE_P )
2882
|| (h->pps.weighted_bipred_idc==1 && h->slice_type_nos== AV_PICTURE_TYPE_B ) )
2883
pred_weight_table(h);
2884
else if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== AV_PICTURE_TYPE_B){
2885
implicit_weight_table(h, -1);
2888
for (i = 0; i < 2; i++) {
2889
h->luma_weight_flag[i] = 0;
2890
h->chroma_weight_flag[i] = 0;
2895
ff_h264_decode_ref_pic_marking(h0, &s->gb);
2898
ff_h264_fill_mbaff_ref_list(h);
2900
if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== AV_PICTURE_TYPE_B){
2901
implicit_weight_table(h, 0);
2902
implicit_weight_table(h, 1);
2906
if(h->slice_type_nos==AV_PICTURE_TYPE_B && !h->direct_spatial_mv_pred)
2907
ff_h264_direct_dist_scale_factor(h);
2908
ff_h264_direct_ref_list_init(h);
2910
if( h->slice_type_nos != AV_PICTURE_TYPE_I && h->pps.cabac ){
2911
tmp = get_ue_golomb_31(&s->gb);
2913
av_log(s->avctx, AV_LOG_ERROR, "cabac_init_idc overflow\n");
2916
h->cabac_init_idc= tmp;
2919
h->last_qscale_diff = 0;
2920
tmp = h->pps.init_qp + get_se_golomb(&s->gb);
2921
if(tmp>51+6*(h->sps.bit_depth_luma-8)){
2922
av_log(s->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
2926
h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
2927
h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
2928
//FIXME qscale / qp ... stuff
2929
if(h->slice_type == AV_PICTURE_TYPE_SP){
2930
get_bits1(&s->gb); /* sp_for_switch_flag */
2932
if(h->slice_type==AV_PICTURE_TYPE_SP || h->slice_type == AV_PICTURE_TYPE_SI){
2933
get_se_golomb(&s->gb); /* slice_qs_delta */
2936
h->deblocking_filter = 1;
2937
h->slice_alpha_c0_offset = 52;
2938
h->slice_beta_offset = 52;
2939
if( h->pps.deblocking_filter_parameters_present ) {
2940
tmp= get_ue_golomb_31(&s->gb);
2942
av_log(s->avctx, AV_LOG_ERROR, "deblocking_filter_idc %u out of range\n", tmp);
2945
h->deblocking_filter= tmp;
2946
if(h->deblocking_filter < 2)
2947
h->deblocking_filter^= 1; // 1<->0
2949
if( h->deblocking_filter ) {
2950
h->slice_alpha_c0_offset += get_se_golomb(&s->gb) << 1;
2951
h->slice_beta_offset += get_se_golomb(&s->gb) << 1;
2952
if( h->slice_alpha_c0_offset > 104U
2953
|| h->slice_beta_offset > 104U){
2954
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);
2960
if( s->avctx->skip_loop_filter >= AVDISCARD_ALL
2961
||(s->avctx->skip_loop_filter >= AVDISCARD_NONKEY && h->slice_type_nos != AV_PICTURE_TYPE_I)
2962
||(s->avctx->skip_loop_filter >= AVDISCARD_BIDIR && h->slice_type_nos == AV_PICTURE_TYPE_B)
2963
||(s->avctx->skip_loop_filter >= AVDISCARD_NONREF && h->nal_ref_idc == 0))
2964
h->deblocking_filter= 0;
2966
if(h->deblocking_filter == 1 && h0->max_contexts > 1) {
2967
if(s->avctx->flags2 & CODEC_FLAG2_FAST) {
2968
/* Cheat slightly for speed:
2969
Do not bother to deblock across slices. */
2970
h->deblocking_filter = 2;
2972
h0->max_contexts = 1;
2973
if(!h0->single_decode_warning) {
2974
av_log(s->avctx, AV_LOG_INFO, "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
2975
h0->single_decode_warning = 1;
2978
av_log(h->s.avctx, AV_LOG_ERROR, "Deblocking switched inside frame.\n");
2983
h->qp_thresh= 15 + 52 - FFMIN(h->slice_alpha_c0_offset, h->slice_beta_offset) - FFMAX3(0, h->pps.chroma_qp_index_offset[0], h->pps.chroma_qp_index_offset[1]);
2986
if( h->pps.num_slice_groups > 1 && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
2987
slice_group_change_cycle= get_bits(&s->gb, ?);
2990
h0->last_slice_type = slice_type;
2991
h->slice_num = ++h0->current_slice;
2992
if(h->slice_num >= MAX_SLICES){
2993
av_log(s->avctx, AV_LOG_ERROR, "Too many slices, increase MAX_SLICES and recompile\n");
2998
int *ref2frm= h->ref2frm[h->slice_num&(MAX_SLICES-1)][j];
2999
for(i=0; i<16; i++){
3001
if(h->ref_list[j][i].data[0]){
3003
uint8_t *base= h->ref_list[j][i].base[0];
3004
for(k=0; k<h->short_ref_count; k++)
3005
if(h->short_ref[k]->base[0] == base){
3009
for(k=0; k<h->long_ref_count; k++)
3010
if(h->long_ref[k] && h->long_ref[k]->base[0] == base){
3011
id_list[i]= h->short_ref_count + k;
3020
ref2frm[i+2]= 4*id_list[i]
3021
+(h->ref_list[j][i].reference&3);
3024
for(i=16; i<48; i++)
3025
ref2frm[i+4]= 4*id_list[(i-16)>>1]
3026
+(h->ref_list[j][i].reference&3);
3029
//FIXME: fix draw_edges+PAFF+frame threads
3030
h->emu_edge_width= (s->flags&CODEC_FLAG_EMU_EDGE || (!h->sps.frame_mbs_only_flag && s->avctx->active_thread_type)) ? 0 : 16;
3031
h->emu_edge_height= (FRAME_MBAFF || FIELD_PICTURE) ? 0 : h->emu_edge_width;
3033
if(s->avctx->debug&FF_DEBUG_PICT_INFO){
3034
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",
3036
(s->picture_structure==PICT_FRAME ? "F" : s->picture_structure==PICT_TOP_FIELD ? "T" : "B"),
3038
av_get_picture_type_char(h->slice_type), h->slice_type_fixed ? " fix" : "", h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
3039
pps_id, h->frame_num,
3040
s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
3041
h->ref_count[0], h->ref_count[1],
3043
h->deblocking_filter, h->slice_alpha_c0_offset/2-26, h->slice_beta_offset/2-26,
3045
h->use_weight==1 && h->use_weight_chroma ? "c" : "",
3046
h->slice_type == AV_PICTURE_TYPE_B ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : ""
3053
int ff_h264_get_slice_type(const H264Context *h)
3055
switch (h->slice_type) {
3056
case AV_PICTURE_TYPE_P: return 0;
3057
case AV_PICTURE_TYPE_B: return 1;
3058
case AV_PICTURE_TYPE_I: return 2;
3059
case AV_PICTURE_TYPE_SP: return 3;
3060
case AV_PICTURE_TYPE_SI: return 4;
3067
* @return non zero if the loop filter can be skiped
3069
static int fill_filter_caches(H264Context *h, int mb_type){
3070
MpegEncContext * const s = &h->s;
3071
const int mb_xy= h->mb_xy;
3072
int top_xy, left_xy[2];
3073
int top_type, left_type[2];
3075
top_xy = mb_xy - (s->mb_stride << MB_FIELD);
3077
//FIXME deblocking could skip the intra and nnz parts.
3079
/* Wow, what a mess, why didn't they simplify the interlacing & intra
3080
* stuff, I can't imagine that these complex rules are worth it. */
3082
left_xy[1] = left_xy[0] = mb_xy-1;
3084
const int left_mb_field_flag = IS_INTERLACED(s->current_picture.mb_type[mb_xy-1]);
3085
const int curr_mb_field_flag = IS_INTERLACED(mb_type);
3087
if (left_mb_field_flag != curr_mb_field_flag) {
3088
left_xy[0] -= s->mb_stride;
3091
if(curr_mb_field_flag){
3092
top_xy += s->mb_stride & (((s->current_picture.mb_type[top_xy ]>>7)&1)-1);
3094
if (left_mb_field_flag != curr_mb_field_flag) {
3095
left_xy[1] += s->mb_stride;
3100
h->top_mb_xy = top_xy;
3101
h->left_mb_xy[0] = left_xy[0];
3102
h->left_mb_xy[1] = left_xy[1];
3104
//for sufficiently low qp, filtering wouldn't do anything
3105
//this is a conservative estimate: could also check beta_offset and more accurate chroma_qp
3106
int qp_thresh = h->qp_thresh; //FIXME strictly we should store qp_thresh for each mb of a slice
3107
int qp = s->current_picture.qscale_table[mb_xy];
3109
&& (left_xy[0]<0 || ((qp + s->current_picture.qscale_table[left_xy[0]] + 1)>>1) <= qp_thresh)
3110
&& (top_xy < 0 || ((qp + s->current_picture.qscale_table[top_xy ] + 1)>>1) <= qp_thresh)){
3113
if( (left_xy[0]< 0 || ((qp + s->current_picture.qscale_table[left_xy[1] ] + 1)>>1) <= qp_thresh)
3114
&& (top_xy < s->mb_stride || ((qp + s->current_picture.qscale_table[top_xy -s->mb_stride] + 1)>>1) <= qp_thresh))
3119
top_type = s->current_picture.mb_type[top_xy] ;
3120
left_type[0] = s->current_picture.mb_type[left_xy[0]];
3121
left_type[1] = s->current_picture.mb_type[left_xy[1]];
3122
if(h->deblocking_filter == 2){
3123
if(h->slice_table[top_xy ] != h->slice_num) top_type= 0;
3124
if(h->slice_table[left_xy[0] ] != h->slice_num) left_type[0]= left_type[1]= 0;
3126
if(h->slice_table[top_xy ] == 0xFFFF) top_type= 0;
3127
if(h->slice_table[left_xy[0] ] == 0xFFFF) left_type[0]= left_type[1] =0;
3129
h->top_type = top_type ;
3130
h->left_type[0]= left_type[0];
3131
h->left_type[1]= left_type[1];
3133
if(IS_INTRA(mb_type))
3136
AV_COPY32(&h->non_zero_count_cache[4+8* 1], &h->non_zero_count[mb_xy][ 0]);
3137
AV_COPY32(&h->non_zero_count_cache[4+8* 2], &h->non_zero_count[mb_xy][ 4]);
3138
AV_COPY32(&h->non_zero_count_cache[4+8* 3], &h->non_zero_count[mb_xy][ 8]);
3139
AV_COPY32(&h->non_zero_count_cache[4+8* 4], &h->non_zero_count[mb_xy][12]);
3141
h->cbp= h->cbp_table[mb_xy];
3145
for(list=0; list<h->list_count; list++){
3148
int16_t (*mv_dst)[2];
3149
int16_t (*mv_src)[2];
3151
if(!USES_LIST(mb_type, list)){
3152
fill_rectangle( h->mv_cache[list][scan8[0]], 4, 4, 8, pack16to32(0,0), 4);
3153
AV_WN32A(&h->ref_cache[list][scan8[ 0]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
3154
AV_WN32A(&h->ref_cache[list][scan8[ 2]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
3155
AV_WN32A(&h->ref_cache[list][scan8[ 8]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
3156
AV_WN32A(&h->ref_cache[list][scan8[10]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
3160
ref = &s->current_picture.ref_index[list][4*mb_xy];
3162
int (*ref2frm)[64] = h->ref2frm[ h->slice_num&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
3163
AV_WN32A(&h->ref_cache[list][scan8[ 0]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
3164
AV_WN32A(&h->ref_cache[list][scan8[ 2]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
3166
AV_WN32A(&h->ref_cache[list][scan8[ 8]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
3167
AV_WN32A(&h->ref_cache[list][scan8[10]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
3170
b_stride = h->b_stride;
3171
mv_dst = &h->mv_cache[list][scan8[0]];
3172
mv_src = &s->current_picture.motion_val[list][4*s->mb_x + 4*s->mb_y*b_stride];
3174
AV_COPY128(mv_dst + 8*y, mv_src + y*b_stride);
3189
//FIXME constraint_intra_pred & partitioning & nnz (let us hope this is just a typo in the spec)
3191
AV_COPY32(&h->non_zero_count_cache[4+8*0], &h->non_zero_count[top_xy][3*4]);
3195
h->non_zero_count_cache[3+8*1]= h->non_zero_count[left_xy[0]][3+0*4];
3196
h->non_zero_count_cache[3+8*2]= h->non_zero_count[left_xy[0]][3+1*4];
3197
h->non_zero_count_cache[3+8*3]= h->non_zero_count[left_xy[0]][3+2*4];
3198
h->non_zero_count_cache[3+8*4]= h->non_zero_count[left_xy[0]][3+3*4];
3201
// CAVLC 8x8dct requires NNZ values for residual decoding that differ from what the loop filter needs
3202
if(!CABAC && h->pps.transform_8x8_mode){
3203
if(IS_8x8DCT(top_type)){
3204
h->non_zero_count_cache[4+8*0]=
3205
h->non_zero_count_cache[5+8*0]= (h->cbp_table[top_xy] & 0x4000) >> 12;
3206
h->non_zero_count_cache[6+8*0]=
3207
h->non_zero_count_cache[7+8*0]= (h->cbp_table[top_xy] & 0x8000) >> 12;
3209
if(IS_8x8DCT(left_type[0])){
3210
h->non_zero_count_cache[3+8*1]=
3211
h->non_zero_count_cache[3+8*2]= (h->cbp_table[left_xy[0]]&0x2000) >> 12; //FIXME check MBAFF
3213
if(IS_8x8DCT(left_type[1])){
3214
h->non_zero_count_cache[3+8*3]=
3215
h->non_zero_count_cache[3+8*4]= (h->cbp_table[left_xy[1]]&0x8000) >> 12; //FIXME check MBAFF
3218
if(IS_8x8DCT(mb_type)){
3219
h->non_zero_count_cache[scan8[0 ]]= h->non_zero_count_cache[scan8[1 ]]=
3220
h->non_zero_count_cache[scan8[2 ]]= h->non_zero_count_cache[scan8[3 ]]= (h->cbp & 0x1000) >> 12;
3222
h->non_zero_count_cache[scan8[0+ 4]]= h->non_zero_count_cache[scan8[1+ 4]]=
3223
h->non_zero_count_cache[scan8[2+ 4]]= h->non_zero_count_cache[scan8[3+ 4]]= (h->cbp & 0x2000) >> 12;
3225
h->non_zero_count_cache[scan8[0+ 8]]= h->non_zero_count_cache[scan8[1+ 8]]=
3226
h->non_zero_count_cache[scan8[2+ 8]]= h->non_zero_count_cache[scan8[3+ 8]]= (h->cbp & 0x4000) >> 12;
3228
h->non_zero_count_cache[scan8[0+12]]= h->non_zero_count_cache[scan8[1+12]]=
3229
h->non_zero_count_cache[scan8[2+12]]= h->non_zero_count_cache[scan8[3+12]]= (h->cbp & 0x8000) >> 12;
3233
if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
3235
for(list=0; list<h->list_count; list++){
3236
if(USES_LIST(top_type, list)){
3237
const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
3238
const int b8_xy= 4*top_xy + 2;
3239
int (*ref2frm)[64] = h->ref2frm[ h->slice_table[top_xy]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
3240
AV_COPY128(h->mv_cache[list][scan8[0] + 0 - 1*8], s->current_picture.motion_val[list][b_xy + 0]);
3241
h->ref_cache[list][scan8[0] + 0 - 1*8]=
3242
h->ref_cache[list][scan8[0] + 1 - 1*8]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 0]];
3243
h->ref_cache[list][scan8[0] + 2 - 1*8]=
3244
h->ref_cache[list][scan8[0] + 3 - 1*8]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 1]];
3246
AV_ZERO128(h->mv_cache[list][scan8[0] + 0 - 1*8]);
3247
AV_WN32A(&h->ref_cache[list][scan8[0] + 0 - 1*8], ((LIST_NOT_USED)&0xFF)*0x01010101u);
3250
if(!IS_INTERLACED(mb_type^left_type[0])){
3251
if(USES_LIST(left_type[0], list)){
3252
const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
3253
const int b8_xy= 4*left_xy[0] + 1;
3254
int (*ref2frm)[64] = h->ref2frm[ h->slice_table[left_xy[0]]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
3255
AV_COPY32(h->mv_cache[list][scan8[0] - 1 + 0 ], s->current_picture.motion_val[list][b_xy + h->b_stride*0]);
3256
AV_COPY32(h->mv_cache[list][scan8[0] - 1 + 8 ], s->current_picture.motion_val[list][b_xy + h->b_stride*1]);
3257
AV_COPY32(h->mv_cache[list][scan8[0] - 1 +16 ], s->current_picture.motion_val[list][b_xy + h->b_stride*2]);
3258
AV_COPY32(h->mv_cache[list][scan8[0] - 1 +24 ], s->current_picture.motion_val[list][b_xy + h->b_stride*3]);
3259
h->ref_cache[list][scan8[0] - 1 + 0 ]=
3260
h->ref_cache[list][scan8[0] - 1 + 8 ]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 2*0]];
3261
h->ref_cache[list][scan8[0] - 1 +16 ]=
3262
h->ref_cache[list][scan8[0] - 1 +24 ]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 2*1]];
3264
AV_ZERO32(h->mv_cache [list][scan8[0] - 1 + 0 ]);
3265
AV_ZERO32(h->mv_cache [list][scan8[0] - 1 + 8 ]);
3266
AV_ZERO32(h->mv_cache [list][scan8[0] - 1 +16 ]);
3267
AV_ZERO32(h->mv_cache [list][scan8[0] - 1 +24 ]);
3268
h->ref_cache[list][scan8[0] - 1 + 0 ]=
3269
h->ref_cache[list][scan8[0] - 1 + 8 ]=
3270
h->ref_cache[list][scan8[0] - 1 + 16 ]=
3271
h->ref_cache[list][scan8[0] - 1 + 24 ]= LIST_NOT_USED;
3280
static void loop_filter(H264Context *h, int start_x, int end_x){
3281
MpegEncContext * const s = &h->s;
3282
uint8_t *dest_y, *dest_cb, *dest_cr;
3283
int linesize, uvlinesize, mb_x, mb_y;
3284
const int end_mb_y= s->mb_y + FRAME_MBAFF;
3285
const int old_slice_type= h->slice_type;
3286
const int pixel_shift = h->pixel_shift;
3288
if(h->deblocking_filter) {
3289
for(mb_x= start_x; mb_x<end_x; mb_x++){
3290
for(mb_y=end_mb_y - FRAME_MBAFF; mb_y<= end_mb_y; mb_y++){
3292
mb_xy = h->mb_xy = mb_x + mb_y*s->mb_stride;
3293
h->slice_num= h->slice_table[mb_xy];
3294
mb_type= s->current_picture.mb_type[mb_xy];
3295
h->list_count= h->list_counts[mb_xy];
3298
h->mb_mbaff = h->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
3302
dest_y = s->current_picture.data[0] + ((mb_x << pixel_shift) + mb_y * s->linesize ) * 16;
3303
dest_cb = s->current_picture.data[1] + ((mb_x << pixel_shift) + mb_y * s->uvlinesize) * (8 << CHROMA444);
3304
dest_cr = s->current_picture.data[2] + ((mb_x << pixel_shift) + mb_y * s->uvlinesize) * (8 << CHROMA444);
3305
//FIXME simplify above
3308
linesize = h->mb_linesize = s->linesize * 2;
3309
uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
3310
if(mb_y&1){ //FIXME move out of this function?
3311
dest_y -= s->linesize*15;
3312
dest_cb-= s->uvlinesize*((8 << CHROMA444)-1);
3313
dest_cr-= s->uvlinesize*((8 << CHROMA444)-1);
3316
linesize = h->mb_linesize = s->linesize;
3317
uvlinesize = h->mb_uvlinesize = s->uvlinesize;
3319
backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, CHROMA444, 0);
3320
if(fill_filter_caches(h, mb_type))
3322
h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.qscale_table[mb_xy]);
3323
h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.qscale_table[mb_xy]);
3326
ff_h264_filter_mb (h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
3328
ff_h264_filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
3333
h->slice_type= old_slice_type;
3335
s->mb_y= end_mb_y - FRAME_MBAFF;
3336
h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
3337
h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
3340
static void predict_field_decoding_flag(H264Context *h){
3341
MpegEncContext * const s = &h->s;
3342
const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
3343
int mb_type = (h->slice_table[mb_xy-1] == h->slice_num)
3344
? s->current_picture.mb_type[mb_xy-1]
3345
: (h->slice_table[mb_xy-s->mb_stride] == h->slice_num)
3346
? s->current_picture.mb_type[mb_xy-s->mb_stride]
3348
h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
3352
* Draw edges and report progress for the last MB row.
3354
static void decode_finish_row(H264Context *h){
3355
MpegEncContext * const s = &h->s;
3356
int top = 16*(s->mb_y >> FIELD_PICTURE);
3357
int height = 16 << FRAME_MBAFF;
3358
int deblock_border = (16 + 4) << FRAME_MBAFF;
3359
int pic_height = 16*s->mb_height >> FIELD_PICTURE;
3361
if (h->deblocking_filter) {
3362
if((top + height) >= pic_height)
3363
height += deblock_border;
3365
top -= deblock_border;
3368
if (top >= pic_height || (top + height) < h->emu_edge_height)
3371
height = FFMIN(height, pic_height - top);
3372
if (top < h->emu_edge_height) {
3373
height = top+height;
3377
ff_draw_horiz_band(s, top, height);
3379
if (s->dropable) return;
3381
ff_thread_report_progress((AVFrame*)s->current_picture_ptr, top + height - 1,
3382
s->picture_structure==PICT_BOTTOM_FIELD);
3385
static int decode_slice(struct AVCodecContext *avctx, void *arg){
3386
H264Context *h = *(void**)arg;
3387
MpegEncContext * const s = &h->s;
3388
const int part_mask= s->partitioned_frame ? (AC_END|AC_ERROR) : 0x7F;
3389
int lf_x_start = s->mb_x;
3393
h->is_complex = FRAME_MBAFF || s->picture_structure != PICT_FRAME || s->codec_id != CODEC_ID_H264 ||
3394
(CONFIG_GRAY && (s->flags&CODEC_FLAG_GRAY));
3396
if( h->pps.cabac ) {
3398
align_get_bits( &s->gb );
3401
ff_init_cabac_states( &h->cabac);
3402
ff_init_cabac_decoder( &h->cabac,
3403
s->gb.buffer + get_bits_count(&s->gb)/8,
3404
(get_bits_left(&s->gb) + 7)/8);
3406
ff_h264_init_cabac_states(h);
3410
int ret = ff_h264_decode_mb_cabac(h);
3412
//STOP_TIMER("decode_mb_cabac")
3414
if(ret>=0) ff_h264_hl_decode_mb(h);
3416
if( ret >= 0 && FRAME_MBAFF ) { //FIXME optimal? or let mb_decode decode 16x32 ?
3419
ret = ff_h264_decode_mb_cabac(h);
3421
if(ret>=0) ff_h264_hl_decode_mb(h);
3424
eos = get_cabac_terminate( &h->cabac );
3426
if((s->workaround_bugs & FF_BUG_TRUNCATED) && h->cabac.bytestream > h->cabac.bytestream_end + 2){
3427
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);
3428
if (s->mb_x >= lf_x_start) loop_filter(h, lf_x_start, s->mb_x + 1);
3431
if( ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 2) {
3432
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);
3433
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);
3437
if( ++s->mb_x >= s->mb_width ) {
3438
loop_filter(h, lf_x_start, s->mb_x);
3439
s->mb_x = lf_x_start = 0;
3440
decode_finish_row(h);
3442
if(FIELD_OR_MBAFF_PICTURE) {
3444
if(FRAME_MBAFF && s->mb_y < s->mb_height)
3445
predict_field_decoding_flag(h);
3449
if( eos || s->mb_y >= s->mb_height ) {
3450
tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
3451
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);
3452
if (s->mb_x > lf_x_start) loop_filter(h, lf_x_start, s->mb_x);
3459
int ret = ff_h264_decode_mb_cavlc(h);
3461
if(ret>=0) ff_h264_hl_decode_mb(h);
3463
if(ret>=0 && FRAME_MBAFF){ //FIXME optimal? or let mb_decode decode 16x32 ?
3465
ret = ff_h264_decode_mb_cavlc(h);
3467
if(ret>=0) ff_h264_hl_decode_mb(h);
3472
av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
3473
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);
3477
if(++s->mb_x >= s->mb_width){
3478
loop_filter(h, lf_x_start, s->mb_x);
3479
s->mb_x = lf_x_start = 0;
3480
decode_finish_row(h);
3482
if(FIELD_OR_MBAFF_PICTURE) {
3484
if(FRAME_MBAFF && s->mb_y < s->mb_height)
3485
predict_field_decoding_flag(h);
3487
if(s->mb_y >= s->mb_height){
3488
tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
3490
if(get_bits_count(&s->gb) == s->gb.size_in_bits ) {
3491
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);
3495
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);
3502
if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->mb_skip_run<=0){
3503
tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
3504
if(get_bits_count(&s->gb) == s->gb.size_in_bits ){
3505
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);
3506
if (s->mb_x > lf_x_start) loop_filter(h, lf_x_start, s->mb_x);
3510
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);
3519
for(;s->mb_y < s->mb_height; s->mb_y++){
3520
for(;s->mb_x < s->mb_width; s->mb_x++){
3521
int ret= decode_mb(h);
3523
ff_h264_hl_decode_mb(h);
3526
av_log(s->avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
3527
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);
3532
if(++s->mb_x >= s->mb_width){
3534
if(++s->mb_y >= s->mb_height){
3535
if(get_bits_count(s->gb) == s->gb.size_in_bits){
3536
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);
3540
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);
3547
if(get_bits_count(s->?gb) >= s->gb?.size_in_bits){
3548
if(get_bits_count(s->gb) == s->gb.size_in_bits){
3549
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
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);
3560
ff_draw_horiz_band(s, 16*s->mb_y, 16);
3563
return -1; //not reached
3567
* Call decode_slice() for each context.
3569
* @param h h264 master context
3570
* @param context_count number of contexts to execute
3572
static void execute_decode_slices(H264Context *h, int context_count){
3573
MpegEncContext * const s = &h->s;
3574
AVCodecContext * const avctx= s->avctx;
3578
if (s->avctx->hwaccel)
3580
if(s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
3582
if(context_count == 1) {
3583
decode_slice(avctx, &h);
3585
for(i = 1; i < context_count; i++) {
3586
hx = h->thread_context[i];
3587
hx->s.error_recognition = avctx->error_recognition;
3588
hx->s.error_count = 0;
3591
avctx->execute(avctx, (void *)decode_slice,
3592
h->thread_context, NULL, context_count, sizeof(void*));
3594
/* pull back stuff from slices to master context */
3595
hx = h->thread_context[context_count - 1];
3596
s->mb_x = hx->s.mb_x;
3597
s->mb_y = hx->s.mb_y;
3598
s->dropable = hx->s.dropable;
3599
s->picture_structure = hx->s.picture_structure;
3600
for(i = 1; i < context_count; i++)
3601
h->s.error_count += h->thread_context[i]->s.error_count;
3606
static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size){
3607
MpegEncContext * const s = &h->s;
3608
AVCodecContext * const avctx= s->avctx;
3609
H264Context *hx; ///< thread context
3613
int pass = !(avctx->active_thread_type & FF_THREAD_FRAME);
3614
int nals_needed=0; ///< number of NALs that need decoding before the next frame thread starts
3617
h->max_contexts = (HAVE_THREADS && (s->avctx->active_thread_type&FF_THREAD_SLICE)) ? avctx->thread_count : 1;
3618
if(!(s->flags2 & CODEC_FLAG2_CHUNKS)){
3619
h->current_slice = 0;
3620
if (!s->first_field)
3621
s->current_picture_ptr= NULL;
3622
ff_h264_reset_sei(h);
3625
for(;pass <= 1;pass++){
3628
next_avc = h->is_avc ? 0 : buf_size;
3638
if(buf_index >= next_avc) {
3639
if(buf_index >= buf_size) break;
3641
for(i = 0; i < h->nal_length_size; i++)
3642
nalsize = (nalsize << 8) | buf[buf_index++];
3643
if(nalsize <= 0 || nalsize > buf_size - buf_index){
3644
av_log(h->s.avctx, AV_LOG_ERROR, "AVC: nal size %d\n", nalsize);
3647
next_avc= buf_index + nalsize;
3649
// start code prefix search
3650
for(; buf_index + 3 < next_avc; buf_index++){
3651
// This should always succeed in the first iteration.
3652
if(buf[buf_index] == 0 && buf[buf_index+1] == 0 && buf[buf_index+2] == 1)
3656
if(buf_index+3 >= buf_size) break;
3659
if(buf_index >= next_avc) continue;
3662
hx = h->thread_context[context_count];
3664
ptr= ff_h264_decode_nal(hx, buf + buf_index, &dst_length, &consumed, next_avc - buf_index);
3665
if (ptr==NULL || dst_length < 0){
3668
i= buf_index + consumed;
3669
if((s->workaround_bugs & FF_BUG_AUTODETECT) && i+3<next_avc &&
3670
buf[i]==0x00 && buf[i+1]==0x00 && buf[i+2]==0x01 && buf[i+3]==0xE0)
3671
s->workaround_bugs |= FF_BUG_TRUNCATED;
3673
if(!(s->workaround_bugs & FF_BUG_TRUNCATED)){
3674
while(ptr[dst_length - 1] == 0 && dst_length > 0)
3677
bit_length= !dst_length ? 0 : (8*dst_length - ff_h264_decode_rbsp_trailing(h, ptr + dst_length - 1));
3679
if(s->avctx->debug&FF_DEBUG_STARTCODE){
3680
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);
3683
if (h->is_avc && (nalsize != consumed) && nalsize){
3684
av_log(h->s.avctx, AV_LOG_DEBUG, "AVC: Consumed only %d bytes instead of %d\n", consumed, nalsize);
3687
buf_index += consumed;
3691
// packets can sometimes contain multiple PPS/SPS
3692
// e.g. two PAFF field pictures in one packet, or a demuxer which splits NALs strangely
3693
// if so, when frame threading we can't start the next thread until we've read all of them
3694
switch (hx->nal_unit_type) {
3699
nals_needed = nal_index;
3704
//FIXME do not discard SEI id
3705
if(avctx->skip_frame >= AVDISCARD_NONREF && h->nal_ref_idc == 0)
3710
switch(hx->nal_unit_type){
3712
if (h->nal_unit_type != NAL_IDR_SLICE) {
3713
av_log(h->s.avctx, AV_LOG_ERROR, "Invalid mix of idr and non-idr slices");
3716
idr(h); //FIXME ensure we don't loose some frames if there is reordering
3718
init_get_bits(&hx->s.gb, ptr, bit_length);
3720
hx->inter_gb_ptr= &hx->s.gb;
3721
hx->s.data_partitioning = 0;
3723
if((err = decode_slice_header(hx, h)))
3726
s->current_picture_ptr->key_frame |=
3727
(hx->nal_unit_type == NAL_IDR_SLICE) ||
3728
(h->sei_recovery_frame_cnt >= 0);
3730
if (h->current_slice == 1) {
3731
if(!(s->flags2 & CODEC_FLAG2_CHUNKS)) {
3732
decode_postinit(h, nal_index >= nals_needed);
3735
if (s->avctx->hwaccel && s->avctx->hwaccel->start_frame(s->avctx, NULL, 0) < 0)
3737
if(CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
3738
ff_vdpau_h264_picture_start(s);
3741
if(hx->redundant_pic_count==0
3742
&& (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
3743
&& (avctx->skip_frame < AVDISCARD_BIDIR || hx->slice_type_nos!=AV_PICTURE_TYPE_B)
3744
&& (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==AV_PICTURE_TYPE_I)
3745
&& avctx->skip_frame < AVDISCARD_ALL){
3746
if(avctx->hwaccel) {
3747
if (avctx->hwaccel->decode_slice(avctx, &buf[buf_index - consumed], consumed) < 0)
3750
if(CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU){
3751
static const uint8_t start_code[] = {0x00, 0x00, 0x01};
3752
ff_vdpau_add_data_chunk(s, start_code, sizeof(start_code));
3753
ff_vdpau_add_data_chunk(s, &buf[buf_index - consumed], consumed );
3759
init_get_bits(&hx->s.gb, ptr, bit_length);
3761
hx->inter_gb_ptr= NULL;
3763
if ((err = decode_slice_header(hx, h)) < 0)
3766
hx->s.data_partitioning = 1;
3770
init_get_bits(&hx->intra_gb, ptr, bit_length);
3771
hx->intra_gb_ptr= &hx->intra_gb;
3774
init_get_bits(&hx->inter_gb, ptr, bit_length);
3775
hx->inter_gb_ptr= &hx->inter_gb;
3777
if(hx->redundant_pic_count==0 && hx->intra_gb_ptr && hx->s.data_partitioning
3778
&& s->context_initialized
3779
&& (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
3780
&& (avctx->skip_frame < AVDISCARD_BIDIR || hx->slice_type_nos!=AV_PICTURE_TYPE_B)
3781
&& (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==AV_PICTURE_TYPE_I)
3782
&& avctx->skip_frame < AVDISCARD_ALL)
3786
init_get_bits(&s->gb, ptr, bit_length);
3787
ff_h264_decode_sei(h);
3790
init_get_bits(&s->gb, ptr, bit_length);
3791
ff_h264_decode_seq_parameter_set(h);
3793
if (s->flags& CODEC_FLAG_LOW_DELAY ||
3794
(h->sps.bitstream_restriction_flag && !h->sps.num_reorder_frames))
3797
if(avctx->has_b_frames < 2)
3798
avctx->has_b_frames= !s->low_delay;
3800
if (avctx->bits_per_raw_sample != h->sps.bit_depth_luma) {
3801
if (h->sps.bit_depth_luma >= 8 && h->sps.bit_depth_luma <= 10) {
3802
avctx->bits_per_raw_sample = h->sps.bit_depth_luma;
3803
h->pixel_shift = h->sps.bit_depth_luma > 8;
3805
ff_h264dsp_init(&h->h264dsp, h->sps.bit_depth_luma);
3806
ff_h264_pred_init(&h->hpc, s->codec_id, h->sps.bit_depth_luma);
3807
dsputil_init(&s->dsp, s->avctx);
3809
av_log(avctx, AV_LOG_DEBUG, "Unsupported bit depth: %d\n", h->sps.bit_depth_luma);
3815
init_get_bits(&s->gb, ptr, bit_length);
3817
ff_h264_decode_picture_parameter_set(h, bit_length);
3821
case NAL_END_SEQUENCE:
3822
case NAL_END_STREAM:
3823
case NAL_FILLER_DATA:
3825
case NAL_AUXILIARY_SLICE:
3828
av_log(avctx, AV_LOG_DEBUG, "Unknown NAL code: %d (%d bits)\n", hx->nal_unit_type, bit_length);
3831
if(context_count == h->max_contexts) {
3832
execute_decode_slices(h, context_count);
3837
av_log(h->s.avctx, AV_LOG_ERROR, "decode_slice_header error\n");
3839
/* Slice could not be decoded in parallel mode, copy down
3840
* NAL unit stuff to context 0 and restart. Note that
3841
* rbsp_buffer is not transferred, but since we no longer
3842
* run in parallel mode this should not be an issue. */
3843
h->nal_unit_type = hx->nal_unit_type;
3844
h->nal_ref_idc = hx->nal_ref_idc;
3851
execute_decode_slices(h, context_count);
3856
* returns the number of bytes consumed for building the current frame
3858
static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size){
3859
if(pos==0) pos=1; //avoid infinite loops (i doubt that is needed but ...)
3860
if(pos+10>buf_size) pos=buf_size; // oops ;)
3865
static int decode_frame(AVCodecContext *avctx,
3866
void *data, int *data_size,
3869
const uint8_t *buf = avpkt->data;
3870
int buf_size = avpkt->size;
3871
H264Context *h = avctx->priv_data;
3872
MpegEncContext *s = &h->s;
3873
AVFrame *pict = data;
3876
s->flags= avctx->flags;
3877
s->flags2= avctx->flags2;
3879
/* end of stream, output what is still in the buffers */
3881
if (buf_size == 0) {
3885
s->current_picture_ptr = NULL;
3887
//FIXME factorize this with the output code below
3888
out = h->delayed_pic[0];
3890
for(i=1; h->delayed_pic[i] && !h->delayed_pic[i]->key_frame && !h->delayed_pic[i]->mmco_reset; i++)
3891
if(h->delayed_pic[i]->poc < out->poc){
3892
out = h->delayed_pic[i];
3896
for(i=out_idx; h->delayed_pic[i]; i++)
3897
h->delayed_pic[i] = h->delayed_pic[i+1];
3900
*data_size = sizeof(AVFrame);
3901
*pict= *(AVFrame*)out;
3907
buf_index=decode_nal_units(h, buf, buf_size);
3911
if (!s->current_picture_ptr && h->nal_unit_type == NAL_END_SEQUENCE) {
3916
if(!(s->flags2 & CODEC_FLAG2_CHUNKS) && !s->current_picture_ptr){
3917
if (avctx->skip_frame >= AVDISCARD_NONREF)
3919
av_log(avctx, AV_LOG_ERROR, "no frame!\n");
3923
if(!(s->flags2 & CODEC_FLAG2_CHUNKS) || (s->mb_y >= s->mb_height && s->mb_height)){
3925
if(s->flags2 & CODEC_FLAG2_CHUNKS) decode_postinit(h, 1);
3929
if (!h->next_output_pic) {
3930
/* Wait for second field. */
3934
*data_size = sizeof(AVFrame);
3935
*pict = *(AVFrame*)h->next_output_pic;
3939
assert(pict->data[0] || !*data_size);
3940
ff_print_debug_info(s, pict);
3941
//printf("out %d\n", (int)pict->data[0]);
3943
return get_consumed_bytes(s, buf_index, buf_size);
3946
static inline void fill_mb_avail(H264Context *h){
3947
MpegEncContext * const s = &h->s;
3948
const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
3951
h->mb_avail[0]= s->mb_x && h->slice_table[mb_xy - s->mb_stride - 1] == h->slice_num;
3952
h->mb_avail[1]= h->slice_table[mb_xy - s->mb_stride ] == h->slice_num;
3953
h->mb_avail[2]= s->mb_x+1 < s->mb_width && h->slice_table[mb_xy - s->mb_stride + 1] == h->slice_num;
3959
h->mb_avail[3]= s->mb_x && h->slice_table[mb_xy - 1] == h->slice_num;
3960
h->mb_avail[4]= 1; //FIXME move out
3961
h->mb_avail[5]= 0; //FIXME move out
3969
#define SIZE (COUNT*40)
3975
// int int_temp[10000];
3977
AVCodecContext avctx;
3979
dsputil_init(&dsp, &avctx);
3981
init_put_bits(&pb, temp, SIZE);
3982
printf("testing unsigned exp golomb\n");
3983
for(i=0; i<COUNT; i++){
3985
set_ue_golomb(&pb, i);
3986
STOP_TIMER("set_ue_golomb");
3988
flush_put_bits(&pb);
3990
init_get_bits(&gb, temp, 8*SIZE);
3991
for(i=0; i<COUNT; i++){
3994
s= show_bits(&gb, 24);
3997
j= get_ue_golomb(&gb);
3999
printf("mismatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
4002
STOP_TIMER("get_ue_golomb");
4006
init_put_bits(&pb, temp, SIZE);
4007
printf("testing signed exp golomb\n");
4008
for(i=0; i<COUNT; i++){
4010
set_se_golomb(&pb, i - COUNT/2);
4011
STOP_TIMER("set_se_golomb");
4013
flush_put_bits(&pb);
4015
init_get_bits(&gb, temp, 8*SIZE);
4016
for(i=0; i<COUNT; i++){
4019
s= show_bits(&gb, 24);
4022
j= get_se_golomb(&gb);
4023
if(j != i - COUNT/2){
4024
printf("mismatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
4027
STOP_TIMER("get_se_golomb");
4031
printf("testing 4x4 (I)DCT\n");
4034
uint8_t src[16], ref[16];
4035
uint64_t error= 0, max_error=0;
4037
for(i=0; i<COUNT; i++){
4039
// printf("%d %d %d\n", r1, r2, (r2-r1)*16);
4040
for(j=0; j<16; j++){
4041
ref[j]= random()%255;
4042
src[j]= random()%255;
4045
h264_diff_dct_c(block, src, ref, 4);
4048
for(j=0; j<16; j++){
4049
// printf("%d ", block[j]);
4050
block[j]= block[j]*4;
4051
if(j&1) block[j]= (block[j]*4 + 2)/5;
4052
if(j&4) block[j]= (block[j]*4 + 2)/5;
4056
h->h264dsp.h264_idct_add(ref, block, 4);
4057
/* for(j=0; j<16; j++){
4058
printf("%d ", ref[j]);
4062
for(j=0; j<16; j++){
4063
int diff= FFABS(src[j] - ref[j]);
4066
max_error= FFMAX(max_error, diff);
4069
printf("error=%f max_error=%d\n", ((float)error)/COUNT/16, (int)max_error );
4070
printf("testing quantizer\n");
4071
for(qp=0; qp<52; qp++){
4073
src1_block[i]= src2_block[i]= random()%255;
4076
printf("Testing NAL layer\n");
4078
uint8_t bitstream[COUNT];
4079
uint8_t nal[COUNT*2];
4081
memset(&h, 0, sizeof(H264Context));
4083
for(i=0; i<COUNT; i++){
4091
for(j=0; j<COUNT; j++){
4092
bitstream[j]= (random() % 255) + 1;
4095
for(j=0; j<zeros; j++){
4096
int pos= random() % COUNT;
4097
while(bitstream[pos] == 0){
4106
nal_length= encode_nal(&h, nal, bitstream, COUNT, COUNT*2);
4108
printf("encoding failed\n");
4112
out= ff_h264_decode_nal(&h, nal, &out_length, &consumed, nal_length);
4116
if(out_length != COUNT){
4117
printf("incorrect length %d %d\n", out_length, COUNT);
4121
if(consumed != nal_length){
4122
printf("incorrect consumed length %d %d\n", nal_length, consumed);
4126
if(memcmp(bitstream, out, COUNT)){
4127
printf("mismatch\n");
4133
printf("Testing RBSP\n");
4141
av_cold void ff_h264_free_context(H264Context *h)
4145
free_tables(h, 1); //FIXME cleanup init stuff perhaps
4147
for(i = 0; i < MAX_SPS_COUNT; i++)
4148
av_freep(h->sps_buffers + i);
4150
for(i = 0; i < MAX_PPS_COUNT; i++)
4151
av_freep(h->pps_buffers + i);
4154
av_cold int ff_h264_decode_end(AVCodecContext *avctx)
4156
H264Context *h = avctx->priv_data;
4157
MpegEncContext *s = &h->s;
4159
ff_h264_free_context(h);
4163
// memset(h, 0, sizeof(H264Context));
4168
static const AVProfile profiles[] = {
4169
{ FF_PROFILE_H264_BASELINE, "Baseline" },
4170
{ FF_PROFILE_H264_CONSTRAINED_BASELINE, "Constrained Baseline" },
4171
{ FF_PROFILE_H264_MAIN, "Main" },
4172
{ FF_PROFILE_H264_EXTENDED, "Extended" },
4173
{ FF_PROFILE_H264_HIGH, "High" },
4174
{ FF_PROFILE_H264_HIGH_10, "High 10" },
4175
{ FF_PROFILE_H264_HIGH_10_INTRA, "High 10 Intra" },
4176
{ FF_PROFILE_H264_HIGH_422, "High 4:2:2" },
4177
{ FF_PROFILE_H264_HIGH_422_INTRA, "High 4:2:2 Intra" },
4178
{ FF_PROFILE_H264_HIGH_444, "High 4:4:4" },
4179
{ FF_PROFILE_H264_HIGH_444_PREDICTIVE, "High 4:4:4 Predictive" },
4180
{ FF_PROFILE_H264_HIGH_444_INTRA, "High 4:4:4 Intra" },
4181
{ FF_PROFILE_H264_CAVLC_444, "CAVLC 4:4:4" },
4182
{ FF_PROFILE_UNKNOWN },
4185
AVCodec ff_h264_decoder = {
4189
sizeof(H264Context),
4190
ff_h264_decode_init,
4194
/*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 | CODEC_CAP_DELAY |
4195
CODEC_CAP_SLICE_THREADS | CODEC_CAP_FRAME_THREADS,
4197
.long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10"),
4198
.init_thread_copy = ONLY_IF_THREADS_ENABLED(decode_init_thread_copy),
4199
.update_thread_context = ONLY_IF_THREADS_ENABLED(decode_update_thread_context),
4200
.profiles = NULL_IF_CONFIG_SMALL(profiles),
4203
#if CONFIG_H264_VDPAU_DECODER
4204
AVCodec ff_h264_vdpau_decoder = {
4208
sizeof(H264Context),
4209
ff_h264_decode_init,
4213
CODEC_CAP_DR1 | CODEC_CAP_DELAY | CODEC_CAP_HWACCEL_VDPAU,
4215
.long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10 (VDPAU acceleration)"),
4216
.pix_fmts = (const enum PixelFormat[]){PIX_FMT_VDPAU_H264, PIX_FMT_NONE},
4217
.profiles = NULL_IF_CONFIG_SMALL(profiles),