Cosmetic changes to h264_idct_10bit.asm.
[libav.git] / libavcodec / h264.c
CommitLineData
0da71265 1/*
ff3d4310 2 * H.26L/H.264/AVC/JVT/14496-10/... decoder
0da71265
MN
3 * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
4 *
2912e87a 5 * This file is part of Libav.
b78e7197 6 *
2912e87a 7 * Libav is free software; you can redistribute it and/or
0da71265
MN
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
b78e7197 10 * version 2.1 of the License, or (at your option) any later version.
0da71265 11 *
2912e87a 12 * Libav is distributed in the hope that it will be useful,
0da71265
MN
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.
16 *
17 * You should have received a copy of the GNU Lesser General Public
2912e87a 18 * License along with Libav; if not, write to the Free Software
5509bffa 19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
0da71265 20 */
115329f1 21
0da71265 22/**
ba87f080 23 * @file
0da71265
MN
24 * H.264 / AVC / MPEG4 part10 codec.
25 * @author Michael Niedermayer <michaelni@gmx.at>
26 */
27
737eb597 28#include "libavutil/imgutils.h"
40e5d31b 29#include "internal.h"
0da71265
MN
30#include "dsputil.h"
31#include "avcodec.h"
32#include "mpegvideo.h"
26b4fe82 33#include "h264.h"
0da71265 34#include "h264data.h"
188d3c51 35#include "h264_mvpred.h"
0da71265 36#include "golomb.h"
199436b9 37#include "mathops.h"
626464fb 38#include "rectangle.h"
369122dd 39#include "vdpau_internal.h"
cfa5a81e 40#include "libavutil/avassert.h"
0da71265 41
e5017ab8
LA
42#include "cabac.h"
43
2848ce84 44//#undef NDEBUG
0da71265
MN
45#include <assert.h>
46
fcc0224e
OA
47static const uint8_t rem6[QP_MAX_NUM+1]={
480, 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,
acd8d10f
PI
49};
50
fcc0224e
OA
51static const uint8_t div6[QP_MAX_NUM+1]={
520, 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,
acd8d10f
PI
53};
54
0435fb16
BC
55static const enum PixelFormat hwaccel_pixfmt_list_h264_jpeg_420[] = {
56 PIX_FMT_DXVA2_VLD,
57 PIX_FMT_VAAPI_VLD,
58 PIX_FMT_YUVJ420P,
59 PIX_FMT_NONE
60};
61
903d58f6 62void ff_h264_write_back_intra_pred_mode(H264Context *h){
5b0fb524 63 int8_t *mode= h->intra4x4_pred_mode + h->mb2br_xy[h->mb_xy];
0da71265 64
662a5b23
MN
65 AV_COPY32(mode, h->intra4x4_pred_mode_cache + 4 + 8*4);
66 mode[4]= h->intra4x4_pred_mode_cache[7+8*3];
67 mode[5]= h->intra4x4_pred_mode_cache[7+8*2];
68 mode[6]= h->intra4x4_pred_mode_cache[7+8*1];
0da71265
MN
69}
70
71/**
72 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
73 */
2bedc0e8
MN
74int ff_h264_check_intra4x4_pred_mode(H264Context *h){
75 MpegEncContext * const s = &h->s;
76 static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
77 static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
78 int i;
79
80 if(!(h->top_samples_available&0x8000)){
81 for(i=0; i<4; i++){
82 int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
83 if(status<0){
84 av_log(h->s.avctx, AV_LOG_ERROR, "top block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y);
85 return -1;
86 } else if(status){
87 h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
88 }
89 }
90 }
91
92 if((h->left_samples_available&0x8888)!=0x8888){
93 static const int mask[4]={0x8000,0x2000,0x80,0x20};
94 for(i=0; i<4; i++){
95 if(!(h->left_samples_available&mask[i])){
96 int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
97 if(status<0){
98 av_log(h->s.avctx, AV_LOG_ERROR, "left block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y);
99 return -1;
100 } else if(status){
101 h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
102 }
103 }
104 }
105 }
106
107 return 0;
108} //FIXME cleanup like ff_h264_check_intra_pred_mode
109
110/**
111 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
112 */
903d58f6 113int ff_h264_check_intra_pred_mode(H264Context *h, int mode){
0da71265
MN
114 MpegEncContext * const s = &h->s;
115 static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
116 static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
115329f1 117
43ff0714 118 if(mode > 6U) {
5175b937 119 av_log(h->s.avctx, AV_LOG_ERROR, "out of range intra chroma pred mode at %d %d\n", s->mb_x, s->mb_y);
7440fe83 120 return -1;
5175b937 121 }
115329f1 122
0da71265
MN
123 if(!(h->top_samples_available&0x8000)){
124 mode= top[ mode ];
125 if(mode<0){
9b879566 126 av_log(h->s.avctx, AV_LOG_ERROR, "top block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);
0da71265
MN
127 return -1;
128 }
129 }
115329f1 130
d1d10e91 131 if((h->left_samples_available&0x8080) != 0x8080){
0da71265 132 mode= left[ mode ];
d1d10e91
MN
133 if(h->left_samples_available&0x8080){ //mad cow disease mode, aka MBAFF + constrained_intra_pred
134 mode= ALZHEIMER_DC_L0T_PRED8x8 + (!(h->left_samples_available&0x8000)) + 2*(mode == DC_128_PRED8x8);
135 }
0da71265 136 if(mode<0){
9b879566 137 av_log(h->s.avctx, AV_LOG_ERROR, "left block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);
0da71265 138 return -1;
115329f1 139 }
0da71265
MN
140 }
141
142 return mode;
143}
144
1790a5e9 145const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src, int *dst_length, int *consumed, int length){
0da71265
MN
146 int i, si, di;
147 uint8_t *dst;
24456882 148 int bufidx;
0da71265 149
bb270c08 150// src[0]&0x80; //forbidden bit
0da71265
MN
151 h->nal_ref_idc= src[0]>>5;
152 h->nal_unit_type= src[0]&0x1F;
153
154 src++; length--;
e08715d3 155
b250f9c6
AJ
156#if HAVE_FAST_UNALIGNED
157# if HAVE_FAST_64BIT
e08715d3
MN
158# define RS 7
159 for(i=0; i+1<length; i+=9){
19769ece 160 if(!((~AV_RN64A(src+i) & (AV_RN64A(src+i) - 0x0100010001000101ULL)) & 0x8000800080008080ULL))
e08715d3
MN
161# else
162# define RS 3
163 for(i=0; i+1<length; i+=5){
19769ece 164 if(!((~AV_RN32A(src+i) & (AV_RN32A(src+i) - 0x01000101U)) & 0x80008080U))
e08715d3
MN
165# endif
166 continue;
167 if(i>0 && !src[i]) i--;
168 while(src[i]) i++;
169#else
170# define RS 0
0da71265
MN
171 for(i=0; i+1<length; i+=2){
172 if(src[i]) continue;
173 if(i>0 && src[i-1]==0) i--;
e08715d3 174#endif
0da71265
MN
175 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
176 if(src[i+2]!=3){
177 /* startcode, so we must be past the end */
178 length=i;
179 }
180 break;
181 }
abb27cfb 182 i-= RS;
0da71265
MN
183 }
184
185 if(i>=length-1){ //no escaped 0
186 *dst_length= length;
187 *consumed= length+1; //+1 for the header
115329f1 188 return src;
0da71265
MN
189 }
190
24456882 191 bufidx = h->nal_unit_type == NAL_DPC ? 1 : 0; // use second escape buffer for inter data
238ef6da 192 av_fast_malloc(&h->rbsp_buffer[bufidx], &h->rbsp_buffer_size[bufidx], length+FF_INPUT_BUFFER_PADDING_SIZE);
24456882 193 dst= h->rbsp_buffer[bufidx];
0da71265 194
ac658be5
FOL
195 if (dst == NULL){
196 return NULL;
197 }
198
3b66c4c5 199//printf("decoding esc\n");
593af7cd
MN
200 memcpy(dst, src, i);
201 si=di=i;
202 while(si+2<length){
0da71265 203 //remove escapes (very rare 1:2^22)
593af7cd
MN
204 if(src[si+2]>3){
205 dst[di++]= src[si++];
206 dst[di++]= src[si++];
207 }else if(src[si]==0 && src[si+1]==0){
0da71265
MN
208 if(src[si+2]==3){ //escape
209 dst[di++]= 0;
210 dst[di++]= 0;
211 si+=3;
c8470cc1 212 continue;
0da71265 213 }else //next start code
593af7cd 214 goto nsc;
0da71265
MN
215 }
216
217 dst[di++]= src[si++];
218 }
593af7cd
MN
219 while(si<length)
220 dst[di++]= src[si++];
221nsc:
0da71265 222
d4369630
AS
223 memset(dst+di, 0, FF_INPUT_BUFFER_PADDING_SIZE);
224
0da71265
MN
225 *dst_length= di;
226 *consumed= si + 1;//+1 for the header
90b5b51e 227//FIXME store exact number of bits in the getbitcontext (it is needed for decoding)
0da71265
MN
228 return dst;
229}
230
85297319
DEP
231/**
232 * Identify the exact end of the bitstream
233 * @return the length of the trailing, or 0 if damaged
234 */
235static int ff_h264_decode_rbsp_trailing(H264Context *h, const uint8_t *src){
0da71265
MN
236 int v= *src;
237 int r;
238
a9c9a240 239 tprintf(h->s.avctx, "rbsp trailing %X\n", v);
0da71265
MN
240
241 for(r=1; r<9; r++){
242 if(v&1) return r;
243 v>>=1;
244 }
245 return 0;
246}
247
66c6b5e2
RB
248#if 0
249/**
250 * DCT transforms the 16 dc values.
251 * @param qp quantization parameter ??? FIXME
252 */
253static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
254// const int qmul= dequant_coeff[qp][0];
255 int i;
256 int temp[16]; //FIXME check if this is a good idea
257 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
258 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
259
260 for(i=0; i<4; i++){
261 const int offset= y_offset[i];
262 const int z0= block[offset+stride*0] + block[offset+stride*4];
263 const int z1= block[offset+stride*0] - block[offset+stride*4];
264 const int z2= block[offset+stride*1] - block[offset+stride*5];
265 const int z3= block[offset+stride*1] + block[offset+stride*5];
266
267 temp[4*i+0]= z0+z3;
268 temp[4*i+1]= z1+z2;
269 temp[4*i+2]= z1-z2;
270 temp[4*i+3]= z0-z3;
271 }
272
273 for(i=0; i<4; i++){
274 const int offset= x_offset[i];
275 const int z0= temp[4*0+i] + temp[4*2+i];
276 const int z1= temp[4*0+i] - temp[4*2+i];
277 const int z2= temp[4*1+i] - temp[4*3+i];
278 const int z3= temp[4*1+i] + temp[4*3+i];
279
280 block[stride*0 +offset]= (z0 + z3)>>1;
281 block[stride*2 +offset]= (z1 + z2)>>1;
282 block[stride*8 +offset]= (z1 - z2)>>1;
283 block[stride*10+offset]= (z0 - z3)>>1;
284 }
285}
286#endif
287
288#undef xStride
289#undef stride
290
66c6b5e2
RB
291#if 0
292static void chroma_dc_dct_c(DCTELEM *block){
293 const int stride= 16*2;
294 const int xStride= 16;
295 int a,b,c,d,e;
296
297 a= block[stride*0 + xStride*0];
298 b= block[stride*0 + xStride*1];
299 c= block[stride*1 + xStride*0];
300 d= block[stride*1 + xStride*1];
301
302 e= a-b;
303 a= a+b;
304 b= c-d;
305 c= c+d;
306
307 block[stride*0 + xStride*0]= (a+c);
308 block[stride*0 + xStride*1]= (e+b);
309 block[stride*1 + xStride*0]= (a-c);
310 block[stride*1 + xStride*1]= (e-b);
311}
312#endif
313
0da71265
MN
314static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
315 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
316 int src_x_offset, int src_y_offset,
6e3ef511
OA
317 qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op,
318 int pixel_shift){
0da71265
MN
319 MpegEncContext * const s = &h->s;
320 const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
5d18eaad 321 int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
0da71265 322 const int luma_xy= (mx&3) + ((my&3)<<2);
6e3ef511 323 uint8_t * src_y = pic->data[0] + ((mx>>2) << pixel_shift) + (my>>2)*h->mb_linesize;
5d18eaad
LM
324 uint8_t * src_cb, * src_cr;
325 int extra_width= h->emu_edge_width;
326 int extra_height= h->emu_edge_height;
0da71265
MN
327 int emu=0;
328 const int full_mx= mx>>2;
329 const int full_my= my>>2;
fbd312fd 330 const int pic_width = 16*s->mb_width;
0d43dd8c 331 const int pic_height = 16*s->mb_height >> MB_FIELD;
115329f1 332
0da71265
MN
333 if(mx&7) extra_width -= 3;
334 if(my&7) extra_height -= 3;
115329f1
DB
335
336 if( full_mx < 0-extra_width
337 || full_my < 0-extra_height
338 || full_mx + 16/*FIXME*/ > pic_width + extra_width
fbd312fd 339 || full_my + 16/*FIXME*/ > pic_height + extra_height){
6e3ef511
OA
340 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);
341 src_y= s->edge_emu_buffer + (2 << pixel_shift) + 2*h->mb_linesize;
0da71265
MN
342 emu=1;
343 }
115329f1 344
5d18eaad 345 qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); //FIXME try variable height perhaps?
0da71265 346 if(!square){
5d18eaad 347 qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
0da71265 348 }
115329f1 349
49fb20cb 350 if(CONFIG_GRAY && s->flags&CODEC_FLAG_GRAY) return;
115329f1 351
0d43dd8c 352 if(MB_FIELD){
5d18eaad 353 // chroma offset when predicting from a field of opposite parity
2143b118 354 my += 2 * ((s->mb_y & 1) - (pic->reference - 1));
5d18eaad
LM
355 emu |= (my>>3) < 0 || (my>>3) + 8 >= (pic_height>>1);
356 }
6e3ef511
OA
357 src_cb= pic->data[1] + ((mx>>3) << pixel_shift) + (my>>3)*h->mb_uvlinesize;
358 src_cr= pic->data[2] + ((mx>>3) << pixel_shift) + (my>>3)*h->mb_uvlinesize;
5d18eaad 359
0da71265 360 if(emu){
2e279598 361 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);
0da71265
MN
362 src_cb= s->edge_emu_buffer;
363 }
5d18eaad 364 chroma_op(dest_cb, src_cb, h->mb_uvlinesize, chroma_height, mx&7, my&7);
0da71265
MN
365
366 if(emu){
2e279598 367 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);
0da71265
MN
368 src_cr= s->edge_emu_buffer;
369 }
5d18eaad 370 chroma_op(dest_cr, src_cr, h->mb_uvlinesize, chroma_height, mx&7, my&7);
0da71265
MN
371}
372
9f2d1b4f 373static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
0da71265
MN
374 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
375 int x_offset, int y_offset,
376 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
377 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
6e3ef511 378 int list0, int list1, int pixel_shift){
0da71265
MN
379 MpegEncContext * const s = &h->s;
380 qpel_mc_func *qpix_op= qpix_put;
381 h264_chroma_mc_func chroma_op= chroma_put;
115329f1 382
6e3ef511
OA
383 dest_y += (2*x_offset << pixel_shift) + 2*y_offset*h-> mb_linesize;
384 dest_cb += ( x_offset << pixel_shift) + y_offset*h->mb_uvlinesize;
385 dest_cr += ( x_offset << pixel_shift) + y_offset*h->mb_uvlinesize;
0da71265 386 x_offset += 8*s->mb_x;
0d43dd8c 387 y_offset += 8*(s->mb_y >> MB_FIELD);
115329f1 388
0da71265 389 if(list0){
1924f3ce 390 Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
0da71265
MN
391 mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
392 dest_y, dest_cb, dest_cr, x_offset, y_offset,
6e3ef511 393 qpix_op, chroma_op, pixel_shift);
0da71265
MN
394
395 qpix_op= qpix_avg;
396 chroma_op= chroma_avg;
397 }
398
399 if(list1){
1924f3ce 400 Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
0da71265
MN
401 mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
402 dest_y, dest_cb, dest_cr, x_offset, y_offset,
6e3ef511 403 qpix_op, chroma_op, pixel_shift);
0da71265
MN
404 }
405}
406
9f2d1b4f
LM
407static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
408 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
409 int x_offset, int y_offset,
410 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
411 h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
412 h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
6e3ef511 413 int list0, int list1, int pixel_shift){
9f2d1b4f
LM
414 MpegEncContext * const s = &h->s;
415
6e3ef511
OA
416 dest_y += (2*x_offset << pixel_shift) + 2*y_offset*h-> mb_linesize;
417 dest_cb += ( x_offset << pixel_shift) + y_offset*h->mb_uvlinesize;
418 dest_cr += ( x_offset << pixel_shift) + y_offset*h->mb_uvlinesize;
9f2d1b4f 419 x_offset += 8*s->mb_x;
0d43dd8c 420 y_offset += 8*(s->mb_y >> MB_FIELD);
115329f1 421
9f2d1b4f
LM
422 if(list0 && list1){
423 /* don't optimize for luma-only case, since B-frames usually
424 * use implicit weights => chroma too. */
425 uint8_t *tmp_cb = s->obmc_scratchpad;
6e3ef511 426 uint8_t *tmp_cr = s->obmc_scratchpad + (8 << pixel_shift);
5d18eaad 427 uint8_t *tmp_y = s->obmc_scratchpad + 8*h->mb_uvlinesize;
9f2d1b4f
LM
428 int refn0 = h->ref_cache[0][ scan8[n] ];
429 int refn1 = h->ref_cache[1][ scan8[n] ];
430
431 mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
432 dest_y, dest_cb, dest_cr,
6e3ef511 433 x_offset, y_offset, qpix_put, chroma_put, pixel_shift);
9f2d1b4f
LM
434 mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
435 tmp_y, tmp_cb, tmp_cr,
6e3ef511 436 x_offset, y_offset, qpix_put, chroma_put, pixel_shift);
9f2d1b4f
LM
437
438 if(h->use_weight == 2){
1052b76f 439 int weight0 = h->implicit_weight[refn0][refn1][s->mb_y&1];
9f2d1b4f 440 int weight1 = 64 - weight0;
5d18eaad
LM
441 luma_weight_avg( dest_y, tmp_y, h-> mb_linesize, 5, weight0, weight1, 0);
442 chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, 5, weight0, weight1, 0);
443 chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, 5, weight0, weight1, 0);
9f2d1b4f 444 }else{
5d18eaad 445 luma_weight_avg(dest_y, tmp_y, h->mb_linesize, h->luma_log2_weight_denom,
3d9137c8
MN
446 h->luma_weight[refn0][0][0] , h->luma_weight[refn1][1][0],
447 h->luma_weight[refn0][0][1] + h->luma_weight[refn1][1][1]);
5d18eaad 448 chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
3d9137c8
MN
449 h->chroma_weight[refn0][0][0][0] , h->chroma_weight[refn1][1][0][0],
450 h->chroma_weight[refn0][0][0][1] + h->chroma_weight[refn1][1][0][1]);
5d18eaad 451 chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
3d9137c8
MN
452 h->chroma_weight[refn0][0][1][0] , h->chroma_weight[refn1][1][1][0],
453 h->chroma_weight[refn0][0][1][1] + h->chroma_weight[refn1][1][1][1]);
9f2d1b4f
LM
454 }
455 }else{
456 int list = list1 ? 1 : 0;
457 int refn = h->ref_cache[list][ scan8[n] ];
458 Picture *ref= &h->ref_list[list][refn];
459 mc_dir_part(h, ref, n, square, chroma_height, delta, list,
460 dest_y, dest_cb, dest_cr, x_offset, y_offset,
6e3ef511 461 qpix_put, chroma_put, pixel_shift);
9f2d1b4f 462
5d18eaad 463 luma_weight_op(dest_y, h->mb_linesize, h->luma_log2_weight_denom,
3d9137c8 464 h->luma_weight[refn][list][0], h->luma_weight[refn][list][1]);
9f2d1b4f 465 if(h->use_weight_chroma){
5d18eaad 466 chroma_weight_op(dest_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
3d9137c8 467 h->chroma_weight[refn][list][0][0], h->chroma_weight[refn][list][0][1]);
5d18eaad 468 chroma_weight_op(dest_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
3d9137c8 469 h->chroma_weight[refn][list][1][0], h->chroma_weight[refn][list][1][1]);
9f2d1b4f
LM
470 }
471 }
472}
473
474static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
475 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
476 int x_offset, int y_offset,
477 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
478 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
115329f1 479 h264_weight_func *weight_op, h264_biweight_func *weight_avg,
6e3ef511 480 int list0, int list1, int pixel_shift){
9f2d1b4f 481 if((h->use_weight==2 && list0 && list1
1052b76f 482 && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ][h->s.mb_y&1] != 32))
9f2d1b4f
LM
483 || h->use_weight==1)
484 mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
485 x_offset, y_offset, qpix_put, chroma_put,
6e3ef511
OA
486 weight_op[0], weight_op[3], weight_avg[0],
487 weight_avg[3], list0, list1, pixel_shift);
9f2d1b4f
LM
488 else
489 mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
6e3ef511
OA
490 x_offset, y_offset, qpix_put, chroma_put, qpix_avg,
491 chroma_avg, list0, list1, pixel_shift);
9f2d1b4f
LM
492}
493
6e3ef511 494static inline void prefetch_motion(H264Context *h, int list, int pixel_shift){
513fbd8e
LM
495 /* fetch pixels for estimated mv 4 macroblocks ahead
496 * optimized for 64byte cache lines */
497 MpegEncContext * const s = &h->s;
498 const int refn = h->ref_cache[list][scan8[0]];
499 if(refn >= 0){
500 const int mx= (h->mv_cache[list][scan8[0]][0]>>2) + 16*s->mb_x + 8;
501 const int my= (h->mv_cache[list][scan8[0]][1]>>2) + 16*s->mb_y;
502 uint8_t **src= h->ref_list[list][refn].data;
6e3ef511 503 int off= (mx << pixel_shift) + (my + (s->mb_x&3)*4)*h->mb_linesize + (64 << pixel_shift);
513fbd8e 504 s->dsp.prefetch(src[0]+off, s->linesize, 4);
6e3ef511 505 off= ((mx>>1) << pixel_shift) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + (64 << pixel_shift);
513fbd8e
LM
506 s->dsp.prefetch(src[1]+off, src[2]-src[1], 2);
507 }
508}
509
6e3ef511 510static av_always_inline void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
0da71265 511 qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
9f2d1b4f 512 qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
6e3ef511
OA
513 h264_weight_func *weight_op, h264_biweight_func *weight_avg,
514 int pixel_shift){
0da71265 515 MpegEncContext * const s = &h->s;
64514ee8 516 const int mb_xy= h->mb_xy;
0da71265 517 const int mb_type= s->current_picture.mb_type[mb_xy];
115329f1 518
0da71265 519 assert(IS_INTER(mb_type));
115329f1 520
6e3ef511 521 prefetch_motion(h, 0, pixel_shift);
513fbd8e 522
0da71265
MN
523 if(IS_16X16(mb_type)){
524 mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
525 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
7231ccf4 526 weight_op, weight_avg,
6e3ef511
OA
527 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1),
528 pixel_shift);
0da71265 529 }else if(IS_16X8(mb_type)){
6e3ef511 530 mc_part(h, 0, 0, 4, 8 << pixel_shift, dest_y, dest_cb, dest_cr, 0, 0,
0da71265 531 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
9f2d1b4f 532 &weight_op[1], &weight_avg[1],
6e3ef511
OA
533 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1),
534 pixel_shift);
535 mc_part(h, 8, 0, 4, 8 << pixel_shift, dest_y, dest_cb, dest_cr, 0, 4,
0da71265 536 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
9f2d1b4f 537 &weight_op[1], &weight_avg[1],
6e3ef511
OA
538 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1),
539 pixel_shift);
0da71265 540 }else if(IS_8X16(mb_type)){
5d18eaad 541 mc_part(h, 0, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 0, 0,
0da71265 542 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
9f2d1b4f 543 &weight_op[2], &weight_avg[2],
6e3ef511
OA
544 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1),
545 pixel_shift);
5d18eaad 546 mc_part(h, 4, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 4, 0,
0da71265 547 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
9f2d1b4f 548 &weight_op[2], &weight_avg[2],
6e3ef511
OA
549 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1),
550 pixel_shift);
0da71265
MN
551 }else{
552 int i;
115329f1 553
0da71265
MN
554 assert(IS_8X8(mb_type));
555
556 for(i=0; i<4; i++){
557 const int sub_mb_type= h->sub_mb_type[i];
558 const int n= 4*i;
559 int x_offset= (i&1)<<2;
560 int y_offset= (i&2)<<1;
561
562 if(IS_SUB_8X8(sub_mb_type)){
563 mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
564 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
9f2d1b4f 565 &weight_op[3], &weight_avg[3],
6e3ef511
OA
566 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
567 pixel_shift);
0da71265 568 }else if(IS_SUB_8X4(sub_mb_type)){
6e3ef511 569 mc_part(h, n , 0, 2, 4 << pixel_shift, dest_y, dest_cb, dest_cr, x_offset, y_offset,
0da71265 570 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
9f2d1b4f 571 &weight_op[4], &weight_avg[4],
6e3ef511
OA
572 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
573 pixel_shift);
574 mc_part(h, n+2, 0, 2, 4 << pixel_shift, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
0da71265 575 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
9f2d1b4f 576 &weight_op[4], &weight_avg[4],
6e3ef511
OA
577 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
578 pixel_shift);
0da71265 579 }else if(IS_SUB_4X8(sub_mb_type)){
5d18eaad 580 mc_part(h, n , 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
0da71265 581 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
9f2d1b4f 582 &weight_op[5], &weight_avg[5],
6e3ef511
OA
583 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
584 pixel_shift);
5d18eaad 585 mc_part(h, n+1, 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
0da71265 586 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
9f2d1b4f 587 &weight_op[5], &weight_avg[5],
6e3ef511
OA
588 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
589 pixel_shift);
0da71265
MN
590 }else{
591 int j;
592 assert(IS_SUB_4X4(sub_mb_type));
593 for(j=0; j<4; j++){
594 int sub_x_offset= x_offset + 2*(j&1);
595 int sub_y_offset= y_offset + (j&2);
596 mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
597 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
9f2d1b4f 598 &weight_op[6], &weight_avg[6],
6e3ef511
OA
599 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
600 pixel_shift);
0da71265
MN
601 }
602 }
603 }
604 }
513fbd8e 605
6e3ef511 606 prefetch_motion(h, 1, pixel_shift);
0da71265
MN
607}
608
6e3ef511
OA
609#define hl_motion_fn(sh, bits) \
610static av_always_inline void hl_motion_ ## bits(H264Context *h, \
611 uint8_t *dest_y, \
612 uint8_t *dest_cb, uint8_t *dest_cr, \
613 qpel_mc_func (*qpix_put)[16], \
614 h264_chroma_mc_func (*chroma_put), \
615 qpel_mc_func (*qpix_avg)[16], \
616 h264_chroma_mc_func (*chroma_avg), \
617 h264_weight_func *weight_op, \
618 h264_biweight_func *weight_avg) \
619{ \
620 hl_motion(h, dest_y, dest_cb, dest_cr, qpix_put, chroma_put, \
621 qpix_avg, chroma_avg, weight_op, weight_avg, sh); \
622}
623hl_motion_fn(0, 8);
624hl_motion_fn(1, 16);
0da71265 625
91078926 626static void free_tables(H264Context *h, int free_rbsp){
7978debd 627 int i;
afebe2f7 628 H264Context *hx;
0da71265 629 av_freep(&h->intra4x4_pred_mode);
e5017ab8
LA
630 av_freep(&h->chroma_pred_mode_table);
631 av_freep(&h->cbp_table);
9e528114
LA
632 av_freep(&h->mvd_table[0]);
633 av_freep(&h->mvd_table[1]);
5ad984c9 634 av_freep(&h->direct_table);
0da71265
MN
635 av_freep(&h->non_zero_count);
636 av_freep(&h->slice_table_base);
637 h->slice_table= NULL;
c988f975 638 av_freep(&h->list_counts);
e5017ab8 639
0da71265 640 av_freep(&h->mb2b_xy);
d43c1922 641 av_freep(&h->mb2br_xy);
9f2d1b4f 642
6752dd5a 643 for(i = 0; i < MAX_THREADS; i++) {
afebe2f7
644 hx = h->thread_context[i];
645 if(!hx) continue;
646 av_freep(&hx->top_borders[1]);
647 av_freep(&hx->top_borders[0]);
648 av_freep(&hx->s.obmc_scratchpad);
91078926 649 if (free_rbsp){
fcb7e535
RB
650 av_freep(&hx->rbsp_buffer[1]);
651 av_freep(&hx->rbsp_buffer[0]);
652 hx->rbsp_buffer_size[0] = 0;
653 hx->rbsp_buffer_size[1] = 0;
91078926 654 }
d2d5e067 655 if (i) av_freep(&h->thread_context[i]);
afebe2f7 656 }
0da71265
MN
657}
658
239ea04c
LM
659static void init_dequant8_coeff_table(H264Context *h){
660 int i,q,x;
fcc0224e 661 const int max_qp = 51 + 6*(h->sps.bit_depth_luma-8);
239ea04c
LM
662 h->dequant8_coeff[0] = h->dequant8_buffer[0];
663 h->dequant8_coeff[1] = h->dequant8_buffer[1];
664
665 for(i=0; i<2; i++ ){
666 if(i && !memcmp(h->pps.scaling_matrix8[0], h->pps.scaling_matrix8[1], 64*sizeof(uint8_t))){
667 h->dequant8_coeff[1] = h->dequant8_buffer[0];
668 break;
669 }
670
fcc0224e 671 for(q=0; q<max_qp+1; q++){
d9ec210b
DP
672 int shift = div6[q];
673 int idx = rem6[q];
239ea04c 674 for(x=0; x<64; x++)
ca32f7f2 675 h->dequant8_coeff[i][q][(x>>3)|((x&7)<<3)] =
548a1c8a
LM
676 ((uint32_t)dequant8_coeff_init[idx][ dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] *
677 h->pps.scaling_matrix8[i][x]) << shift;
239ea04c
LM
678 }
679 }
680}
681
682static void init_dequant4_coeff_table(H264Context *h){
683 int i,j,q,x;
fcc0224e 684 const int max_qp = 51 + 6*(h->sps.bit_depth_luma-8);
239ea04c
LM
685 for(i=0; i<6; i++ ){
686 h->dequant4_coeff[i] = h->dequant4_buffer[i];
687 for(j=0; j<i; j++){
688 if(!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i], 16*sizeof(uint8_t))){
689 h->dequant4_coeff[i] = h->dequant4_buffer[j];
690 break;
691 }
692 }
693 if(j<i)
694 continue;
695
fcc0224e 696 for(q=0; q<max_qp+1; q++){
d9ec210b
DP
697 int shift = div6[q] + 2;
698 int idx = rem6[q];
239ea04c 699 for(x=0; x<16; x++)
ca32f7f2 700 h->dequant4_coeff[i][q][(x>>2)|((x<<2)&0xF)] =
ab2e3e2c 701 ((uint32_t)dequant4_coeff_init[idx][(x&1) + ((x>>2)&1)] *
239ea04c
LM
702 h->pps.scaling_matrix4[i][x]) << shift;
703 }
704 }
705}
706
707static void init_dequant_tables(H264Context *h){
708 int i,x;
709 init_dequant4_coeff_table(h);
710 if(h->pps.transform_8x8_mode)
711 init_dequant8_coeff_table(h);
712 if(h->sps.transform_bypass){
713 for(i=0; i<6; i++)
714 for(x=0; x<16; x++)
715 h->dequant4_coeff[i][0][x] = 1<<6;
716 if(h->pps.transform_8x8_mode)
717 for(i=0; i<2; i++)
718 for(x=0; x<64; x++)
719 h->dequant8_coeff[i][0][x] = 1<<6;
720 }
721}
722
723
903d58f6 724int ff_h264_alloc_tables(H264Context *h){
0da71265 725 MpegEncContext * const s = &h->s;
7bc9090a 726 const int big_mb_num= s->mb_stride * (s->mb_height+1);
145061a1 727 const int row_mb_num= 2*s->mb_stride*s->avctx->thread_count;
239ea04c 728 int x,y;
0da71265 729
145061a1 730 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->intra4x4_pred_mode, row_mb_num * 8 * sizeof(uint8_t), fail)
e5017ab8 731
c988f975 732 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->non_zero_count , big_mb_num * 32 * sizeof(uint8_t), fail)
d31dbec3
RP
733 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->slice_table_base , (big_mb_num+s->mb_stride) * sizeof(*h->slice_table_base), fail)
734 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->cbp_table, big_mb_num * sizeof(uint16_t), fail)
0da71265 735
d31dbec3 736 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t), fail)
145061a1
MN
737 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[0], 16*row_mb_num * sizeof(uint8_t), fail);
738 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[1], 16*row_mb_num * sizeof(uint8_t), fail);
36b54927 739 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->direct_table, 4*big_mb_num * sizeof(uint8_t) , fail);
c988f975 740 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->list_counts, big_mb_num * sizeof(uint8_t), fail)
e5017ab8 741
b735aeea 742 memset(h->slice_table_base, -1, (big_mb_num+s->mb_stride) * sizeof(*h->slice_table_base));
5d18eaad 743 h->slice_table= h->slice_table_base + s->mb_stride*2 + 1;
0da71265 744
d31dbec3 745 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2b_xy , big_mb_num * sizeof(uint32_t), fail);
d43c1922 746 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2br_xy , big_mb_num * sizeof(uint32_t), fail);
0da71265
MN
747 for(y=0; y<s->mb_height; y++){
748 for(x=0; x<s->mb_width; x++){
7bc9090a 749 const int mb_xy= x + y*s->mb_stride;
0da71265 750 const int b_xy = 4*x + 4*y*h->b_stride;
115329f1 751
0da71265 752 h->mb2b_xy [mb_xy]= b_xy;
e1c88a21 753 h->mb2br_xy[mb_xy]= 8*(FMO ? mb_xy : (mb_xy % (2*s->mb_stride)));
0da71265
MN
754 }
755 }
9f2d1b4f 756
9c6221ae
GV
757 s->obmc_scratchpad = NULL;
758
56edbd81
LM
759 if(!h->dequant4_coeff[0])
760 init_dequant_tables(h);
761
0da71265
MN
762 return 0;
763fail:
91078926 764 free_tables(h, 1);
0da71265
MN
765 return -1;
766}
767
afebe2f7
768/**
769 * Mimic alloc_tables(), but for every context thread.
770 */
145061a1
MN
771static void clone_tables(H264Context *dst, H264Context *src, int i){
772 MpegEncContext * const s = &src->s;
773 dst->intra4x4_pred_mode = src->intra4x4_pred_mode + i*8*2*s->mb_stride;
afebe2f7
774 dst->non_zero_count = src->non_zero_count;
775 dst->slice_table = src->slice_table;
776 dst->cbp_table = src->cbp_table;
777 dst->mb2b_xy = src->mb2b_xy;
d43c1922 778 dst->mb2br_xy = src->mb2br_xy;
afebe2f7 779 dst->chroma_pred_mode_table = src->chroma_pred_mode_table;
145061a1
MN
780 dst->mvd_table[0] = src->mvd_table[0] + i*8*2*s->mb_stride;
781 dst->mvd_table[1] = src->mvd_table[1] + i*8*2*s->mb_stride;
afebe2f7 782 dst->direct_table = src->direct_table;
fb823b77 783 dst->list_counts = src->list_counts;
afebe2f7 784
afebe2f7 785 dst->s.obmc_scratchpad = NULL;
19a0729b 786 ff_h264_pred_init(&dst->hpc, src->s.codec_id, src->sps.bit_depth_luma);
afebe2f7
787}
788
789/**
790 * Init context
791 * Allocate buffers which are not shared amongst multiple threads.
792 */
793static int context_init(H264Context *h){
6e3ef511
OA
794 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[0], h->s.mb_width * (16+8+8) * sizeof(uint8_t)*2, fail)
795 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[1], h->s.mb_width * (16+8+8) * sizeof(uint8_t)*2, fail)
afebe2f7 796
145061a1
MN
797 h->ref_cache[0][scan8[5 ]+1] = h->ref_cache[0][scan8[7 ]+1] = h->ref_cache[0][scan8[13]+1] =
798 h->ref_cache[1][scan8[5 ]+1] = h->ref_cache[1][scan8[7 ]+1] = h->ref_cache[1][scan8[13]+1] = PART_NOT_AVAILABLE;
799
afebe2f7
800 return 0;
801fail:
802 return -1; // free_tables will clean up for us
803}
804
9855b2e3
MN
805static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size);
806
98a6fff9 807static av_cold void common_init(H264Context *h){
0da71265 808 MpegEncContext * const s = &h->s;
0da71265
MN
809
810 s->width = s->avctx->width;
811 s->height = s->avctx->height;
812 s->codec_id= s->avctx->codec->id;
115329f1 813
19a0729b
OA
814 ff_h264dsp_init(&h->h264dsp, 8);
815 ff_h264_pred_init(&h->hpc, s->codec_id, 8);
0da71265 816
239ea04c 817 h->dequant_coeff_pps= -1;
9a41c2c7 818 s->unrestricted_mv=1;
0da71265 819 s->decode=1; //FIXME
56edbd81 820
a5805aa9
MN
821 dsputil_init(&s->dsp, s->avctx); // needed so that idct permutation is known early
822
56edbd81
LM
823 memset(h->pps.scaling_matrix4, 16, 6*16*sizeof(uint8_t));
824 memset(h->pps.scaling_matrix8, 16, 2*64*sizeof(uint8_t));
0da71265
MN
825}
826
05e95319
HC
827int ff_h264_decode_extradata(H264Context *h)
828{
829 AVCodecContext *avctx = h->s.avctx;
9855b2e3 830
05e95319 831 if(*(char *)avctx->extradata == 1){
9855b2e3
MN
832 int i, cnt, nalsize;
833 unsigned char *p = avctx->extradata;
834
835 h->is_avc = 1;
836
837 if(avctx->extradata_size < 7) {
838 av_log(avctx, AV_LOG_ERROR, "avcC too short\n");
839 return -1;
840 }
841 /* sps and pps in the avcC always have length coded with 2 bytes,
842 so put a fake nal_length_size = 2 while parsing them */
843 h->nal_length_size = 2;
844 // Decode sps from avcC
845 cnt = *(p+5) & 0x1f; // Number of sps
846 p += 6;
847 for (i = 0; i < cnt; i++) {
848 nalsize = AV_RB16(p) + 2;
849 if(decode_nal_units(h, p, nalsize) < 0) {
850 av_log(avctx, AV_LOG_ERROR, "Decoding sps %d from avcC failed\n", i);
851 return -1;
852 }
853 p += nalsize;
854 }
855 // Decode pps from avcC
856 cnt = *(p++); // Number of pps
857 for (i = 0; i < cnt; i++) {
858 nalsize = AV_RB16(p) + 2;
859 if(decode_nal_units(h, p, nalsize) != nalsize) {
860 av_log(avctx, AV_LOG_ERROR, "Decoding pps %d from avcC failed\n", i);
861 return -1;
862 }
863 p += nalsize;
864 }
865 // Now store right nal length size, that will be use to parse all other nals
866 h->nal_length_size = ((*(((char*)(avctx->extradata))+4))&0x03)+1;
867 } else {
868 h->is_avc = 0;
05e95319 869 if(decode_nal_units(h, avctx->extradata, avctx->extradata_size) < 0)
9855b2e3
MN
870 return -1;
871 }
05e95319
HC
872 return 0;
873}
874
875av_cold int ff_h264_decode_init(AVCodecContext *avctx){
876 H264Context *h= avctx->priv_data;
877 MpegEncContext * const s = &h->s;
878
879 MPV_decode_defaults(s);
880
881 s->avctx = avctx;
882 common_init(h);
883
884 s->out_format = FMT_H264;
885 s->workaround_bugs= avctx->workaround_bugs;
886
887 // set defaults
888// s->decode_mb= ff_h263_decode_mb;
889 s->quarter_sample = 1;
890 if(!avctx->has_b_frames)
891 s->low_delay= 1;
892
893 avctx->chroma_sample_location = AVCHROMA_LOC_LEFT;
894
895 ff_h264_decode_init_vlc();
896
6e3ef511 897 h->pixel_shift = 0;
19a0729b 898 h->sps.bit_depth_luma = avctx->bits_per_raw_sample = 8;
6e3ef511 899
05e95319
HC
900 h->thread_context[0] = h;
901 h->outputed_poc = INT_MIN;
902 h->prev_poc_msb= 1<<16;
903 h->x264_build = -1;
904 ff_h264_reset_sei(h);
905 if(avctx->codec_id == CODEC_ID_H264){
906 if(avctx->ticks_per_frame == 1){
907 s->avctx->time_base.den *=2;
908 }
909 avctx->ticks_per_frame = 2;
910 }
911
912 if(avctx->extradata_size > 0 && avctx->extradata &&
913 ff_h264_decode_extradata(h))
914 return -1;
915
db8cb47d
MN
916 if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames < h->sps.num_reorder_frames){
917 s->avctx->has_b_frames = h->sps.num_reorder_frames;
918 s->low_delay = 0;
919 }
9855b2e3 920
0da71265
MN
921 return 0;
922}
923
903d58f6 924int ff_h264_frame_start(H264Context *h){
0da71265
MN
925 MpegEncContext * const s = &h->s;
926 int i;
6e3ef511 927 const int pixel_shift = h->pixel_shift;
0da71265 928
af8aa846
MN
929 if(MPV_frame_start(s, s->avctx) < 0)
930 return -1;
0da71265 931 ff_er_frame_start(s);
3a22d7fa
JD
932 /*
933 * MPV_frame_start uses pict_type to derive key_frame.
934 * This is incorrect for H.264; IDR markings must be used.
1412060e 935 * Zero here; IDR markings per slice in frame or fields are ORed in later.
3a22d7fa
JD
936 * See decode_nal_units().
937 */
938 s->current_picture_ptr->key_frame= 0;
c173a088 939 s->current_picture_ptr->mmco_reset= 0;
0da71265
MN
940
941 assert(s->linesize && s->uvlinesize);
942
943 for(i=0; i<16; i++){
6e3ef511
OA
944 h->block_offset[i]= (4*((scan8[i] - scan8[0])&7) << pixel_shift) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
945 h->block_offset[24+i]= (4*((scan8[i] - scan8[0])&7) << pixel_shift) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
0da71265
MN
946 }
947 for(i=0; i<4; i++){
948 h->block_offset[16+i]=
6e3ef511 949 h->block_offset[20+i]= (4*((scan8[i] - scan8[0])&7) << pixel_shift) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
6867a90b 950 h->block_offset[24+16+i]=
6e3ef511 951 h->block_offset[24+20+i]= (4*((scan8[i] - scan8[0])&7) << pixel_shift) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
0da71265
MN
952 }
953
934b0821
LM
954 /* can't be in alloc_tables because linesize isn't known there.
955 * FIXME: redo bipred weight to not require extra buffer? */
afebe2f7 956 for(i = 0; i < s->avctx->thread_count; i++)
0f016023 957 if(h->thread_context[i] && !h->thread_context[i]->s.obmc_scratchpad)
afebe2f7 958 h->thread_context[i]->s.obmc_scratchpad = av_malloc(16*2*s->linesize + 8*2*s->uvlinesize);
5d18eaad 959
2ce1c2e0 960 /* some macroblocks can be accessed before they're available in case of lost slices, mbaff or threading*/
5820b90d 961 memset(h->slice_table, -1, (s->mb_height*s->mb_stride-1) * sizeof(*h->slice_table));
934b0821 962
0da71265 963// s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
28bb9eb2 964
1412060e 965 // We mark the current picture as non-reference after allocating it, so
28bb9eb2
MN
966 // that if we break out due to an error it can be released automatically
967 // in the next MPV_frame_start().
968 // SVQ3 as well as most other codecs have only last/next/current and thus
969 // get released even with set reference, besides SVQ3 and others do not
970 // mark frames as reference later "naturally".
971 if(s->codec_id != CODEC_ID_SVQ3)
972 s->current_picture_ptr->reference= 0;
357282c6
MN
973
974 s->current_picture_ptr->field_poc[0]=
975 s->current_picture_ptr->field_poc[1]= INT_MAX;
5118c6c7 976 assert(s->current_picture_ptr->long_ref==0);
357282c6 977
af8aa846 978 return 0;
0da71265
MN
979}
980
93cc10fa 981static inline void backup_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize, int simple){
53c05b1e 982 MpegEncContext * const s = &h->s;
0b69d625 983 uint8_t *top_border;
5f7f9719 984 int top_idx = 1;
6e3ef511 985 const int pixel_shift = h->pixel_shift;
115329f1 986
53c05b1e
MN
987 src_y -= linesize;
988 src_cb -= uvlinesize;
989 src_cr -= uvlinesize;
990
5f7f9719
MN
991 if(!simple && FRAME_MBAFF){
992 if(s->mb_y&1){
5f7f9719 993 if(!MB_MBAFF){
0b69d625
AS
994 top_border = h->top_borders[0][s->mb_x];
995 AV_COPY128(top_border, src_y + 15*linesize);
6e3ef511
OA
996 if (pixel_shift)
997 AV_COPY128(top_border+16, src_y+15*linesize+16);
49fb20cb 998 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
6e3ef511
OA
999 if (pixel_shift) {
1000 AV_COPY128(top_border+32, src_cb+7*uvlinesize);
1001 AV_COPY128(top_border+48, src_cr+7*uvlinesize);
1002 } else {
0b69d625
AS
1003 AV_COPY64(top_border+16, src_cb+7*uvlinesize);
1004 AV_COPY64(top_border+24, src_cr+7*uvlinesize);
6e3ef511 1005 }
5f7f9719
MN
1006 }
1007 }
c988f975
MN
1008 }else if(MB_MBAFF){
1009 top_idx = 0;
1010 }else
1011 return;
5f7f9719
MN
1012 }
1013
0b69d625 1014 top_border = h->top_borders[top_idx][s->mb_x];
3b66c4c5 1015 // There are two lines saved, the line above the the top macroblock of a pair,
6867a90b 1016 // and the line above the bottom macroblock
0b69d625 1017 AV_COPY128(top_border, src_y + 16*linesize);
6e3ef511
OA
1018 if (pixel_shift)
1019 AV_COPY128(top_border+16, src_y+16*linesize+16);
53c05b1e 1020
49fb20cb 1021 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
6e3ef511
OA
1022 if (pixel_shift) {
1023 AV_COPY128(top_border+32, src_cb+8*uvlinesize);
1024 AV_COPY128(top_border+48, src_cr+8*uvlinesize);
1025 } else {
0b69d625
AS
1026 AV_COPY64(top_border+16, src_cb+8*uvlinesize);
1027 AV_COPY64(top_border+24, src_cr+8*uvlinesize);
6e3ef511 1028 }
53c05b1e
MN
1029 }
1030}
1031
6e3ef511
OA
1032static inline void xchg_mb_border(H264Context *h, uint8_t *src_y,
1033 uint8_t *src_cb, uint8_t *src_cr,
1034 int linesize, int uvlinesize,
1035 int xchg, int simple, int pixel_shift){
53c05b1e 1036 MpegEncContext * const s = &h->s;
4e987f82 1037 int deblock_topleft;
b69378e2 1038 int deblock_top;
5f7f9719 1039 int top_idx = 1;
1e4f1c56
AS
1040 uint8_t *top_border_m1;
1041 uint8_t *top_border;
5f7f9719
MN
1042
1043 if(!simple && FRAME_MBAFF){
1044 if(s->mb_y&1){
c988f975
MN
1045 if(!MB_MBAFF)
1046 return;
5f7f9719 1047 }else{
5f7f9719
MN
1048 top_idx = MB_MBAFF ? 0 : 1;
1049 }
5f7f9719 1050 }
b69378e2
1051
1052 if(h->deblocking_filter == 2) {
4e987f82
RB
1053 deblock_topleft = h->slice_table[h->mb_xy - 1 - s->mb_stride] == h->slice_num;
1054 deblock_top = h->top_type;
b69378e2 1055 } else {
4e987f82
RB
1056 deblock_topleft = (s->mb_x > 0);
1057 deblock_top = (s->mb_y > !!MB_FIELD);
b69378e2 1058 }
53c05b1e 1059
6e3ef511
OA
1060 src_y -= linesize + 1 + pixel_shift;
1061 src_cb -= uvlinesize + 1 + pixel_shift;
1062 src_cr -= uvlinesize + 1 + pixel_shift;
53c05b1e 1063
1e4f1c56
AS
1064 top_border_m1 = h->top_borders[top_idx][s->mb_x-1];
1065 top_border = h->top_borders[top_idx][s->mb_x];
1066
0b69d625 1067#define XCHG(a,b,xchg)\
6e3ef511
OA
1068 if (pixel_shift) {\
1069 if (xchg) {\
1070 AV_SWAP64(b+0,a+0);\
1071 AV_SWAP64(b+8,a+8);\
1072 } else {\
1073 AV_COPY128(b,a); \
1074 }\
1075 } else \
0b69d625
AS
1076if (xchg) AV_SWAP64(b,a);\
1077else AV_COPY64(b,a);
d89dc06a 1078
d89dc06a 1079 if(deblock_top){
4e987f82 1080 if(deblock_topleft){
6e3ef511 1081 XCHG(top_border_m1 + (8 << pixel_shift), src_y - (7 << pixel_shift), 1);
c988f975 1082 }
6e3ef511
OA
1083 XCHG(top_border + (0 << pixel_shift), src_y + (1 << pixel_shift), xchg);
1084 XCHG(top_border + (8 << pixel_shift), src_y + (9 << pixel_shift), 1);
cad4368a 1085 if(s->mb_x+1 < s->mb_width){
6e3ef511 1086 XCHG(h->top_borders[top_idx][s->mb_x+1], src_y + (17 << pixel_shift), 1);
43efd19a 1087 }
53c05b1e 1088 }
49fb20cb 1089 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
d89dc06a 1090 if(deblock_top){
4e987f82 1091 if(deblock_topleft){
6e3ef511
OA
1092 XCHG(top_border_m1 + (16 << pixel_shift), src_cb - (7 << pixel_shift), 1);
1093 XCHG(top_border_m1 + (24 << pixel_shift), src_cr - (7 << pixel_shift), 1);
c988f975 1094 }
6e3ef511
OA
1095 XCHG(top_border + (16 << pixel_shift), src_cb+1+pixel_shift, 1);
1096 XCHG(top_border + (24 << pixel_shift), src_cr+1+pixel_shift, 1);
53c05b1e 1097 }
53c05b1e
MN
1098 }
1099}
1100
6e3ef511
OA
1101static av_always_inline int dctcoef_get(DCTELEM *mb, int high_bit_depth, int index) {
1102 if (high_bit_depth) {
1103 return AV_RN32A(((int32_t*)mb) + index);
1104 } else
1105 return AV_RN16A(mb + index);
1106}
1107
1108static av_always_inline void dctcoef_set(DCTELEM *mb, int high_bit_depth, int index, int value) {
1109 if (high_bit_depth) {
1110 AV_WN32A(((int32_t*)mb) + index, value);
1111 } else
1112 AV_WN16A(mb + index, value);
1113}
1114
1115static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple, int pixel_shift){
0da71265
MN
1116 MpegEncContext * const s = &h->s;
1117 const int mb_x= s->mb_x;
1118 const int mb_y= s->mb_y;
64514ee8 1119 const int mb_xy= h->mb_xy;
0da71265
MN
1120 const int mb_type= s->current_picture.mb_type[mb_xy];
1121 uint8_t *dest_y, *dest_cb, *dest_cr;
1122 int linesize, uvlinesize /*dct_offset*/;
1123 int i;
6867a90b 1124 int *block_offset = &h->block_offset[0];
41e4055b 1125 const int transform_bypass = !simple && (s->qscale == 0 && h->sps.transform_bypass);
8b6871ed 1126 /* is_h264 should always be true if SVQ3 is disabled. */
49fb20cb 1127 const int is_h264 = !CONFIG_SVQ3_DECODER || simple || s->codec_id == CODEC_ID_H264;
36940eca 1128 void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
ef9d1d15 1129 void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
0da71265 1130
6e3ef511
OA
1131 dest_y = s->current_picture.data[0] + ((mb_x << pixel_shift) + mb_y * s->linesize ) * 16;
1132 dest_cb = s->current_picture.data[1] + ((mb_x << pixel_shift) + mb_y * s->uvlinesize) * 8;
1133 dest_cr = s->current_picture.data[2] + ((mb_x << pixel_shift) + mb_y * s->uvlinesize) * 8;
0da71265 1134
6e3ef511
OA
1135 s->dsp.prefetch(dest_y + (s->mb_x&3)*4*s->linesize + (64 << pixel_shift), s->linesize, 4);
1136 s->dsp.prefetch(dest_cb + (s->mb_x&7)*s->uvlinesize + (64 << pixel_shift), dest_cr - dest_cb, 2);
a957c27b 1137
c988f975
MN
1138 h->list_counts[mb_xy]= h->list_count;
1139
bd91fee3 1140 if (!simple && MB_FIELD) {
5d18eaad
LM
1141 linesize = h->mb_linesize = s->linesize * 2;
1142 uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
6867a90b 1143 block_offset = &h->block_offset[24];
1412060e 1144 if(mb_y&1){ //FIXME move out of this function?
0da71265 1145 dest_y -= s->linesize*15;
6867a90b
LLL
1146 dest_cb-= s->uvlinesize*7;
1147 dest_cr-= s->uvlinesize*7;
0da71265 1148 }
5d18eaad
LM
1149 if(FRAME_MBAFF) {
1150 int list;
3425501d 1151 for(list=0; list<h->list_count; list++){
5d18eaad
LM
1152 if(!USES_LIST(mb_type, list))
1153 continue;
1154 if(IS_16X16(mb_type)){
1155 int8_t *ref = &h->ref_cache[list][scan8[0]];
1710856c 1156 fill_rectangle(ref, 4, 4, 8, (16+*ref)^(s->mb_y&1), 1);
5d18eaad
LM
1157 }else{
1158 for(i=0; i<16; i+=4){
5d18eaad
LM
1159 int ref = h->ref_cache[list][scan8[i]];
1160 if(ref >= 0)
1710856c 1161 fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, (16+ref)^(s->mb_y&1), 1);
5d18eaad
LM
1162 }
1163 }
1164 }
1165 }
0da71265 1166 } else {
5d18eaad
LM
1167 linesize = h->mb_linesize = s->linesize;
1168 uvlinesize = h->mb_uvlinesize = s->uvlinesize;
0da71265
MN
1169// dct_offset = s->linesize * 16;
1170 }
115329f1 1171
bd91fee3 1172 if (!simple && IS_INTRA_PCM(mb_type)) {
6e3ef511
OA
1173 if (pixel_shift) {
1174 const int bit_depth = h->sps.bit_depth_luma;
1175 int j;
1176 GetBitContext gb;
1177 init_get_bits(&gb, (uint8_t*)h->mb, 384*bit_depth);
1178
1179 for (i = 0; i < 16; i++) {
1180 uint16_t *tmp_y = (uint16_t*)(dest_y + i*linesize);
1181 for (j = 0; j < 16; j++)
1182 tmp_y[j] = get_bits(&gb, bit_depth);
1183 }
1184 for (i = 0; i < 8; i++) {
1185 uint16_t *tmp_cb = (uint16_t*)(dest_cb + i*uvlinesize);
1186 for (j = 0; j < 8; j++)
1187 tmp_cb[j] = get_bits(&gb, bit_depth);
1188 }
1189 for (i = 0; i < 8; i++) {
1190 uint16_t *tmp_cr = (uint16_t*)(dest_cr + i*uvlinesize);
1191 for (j = 0; j < 8; j++)
1192 tmp_cr[j] = get_bits(&gb, bit_depth);
1193 }
1194 } else {
c1708e8d
MN
1195 for (i=0; i<16; i++) {
1196 memcpy(dest_y + i* linesize, h->mb + i*8, 16);
6fbcaaa0 1197 }
c1708e8d
MN
1198 for (i=0; i<8; i++) {
1199 memcpy(dest_cb+ i*uvlinesize, h->mb + 128 + i*4, 8);
1200 memcpy(dest_cr+ i*uvlinesize, h->mb + 160 + i*4, 8);
6fbcaaa0 1201 }
6e3ef511 1202 }
e7e09b49
LLL
1203 } else {
1204 if(IS_INTRA(mb_type)){
5f7f9719 1205 if(h->deblocking_filter)
6e3ef511 1206 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1, simple, pixel_shift);
53c05b1e 1207
49fb20cb 1208 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
c92a30bb
KS
1209 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
1210 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
e7e09b49 1211 }
0da71265 1212
e7e09b49 1213 if(IS_INTRA4x4(mb_type)){
bd91fee3 1214 if(simple || !s->encoding){
43efd19a 1215 if(IS_8x8DCT(mb_type)){
1eb96035
MN
1216 if(transform_bypass){
1217 idct_dc_add =
1218 idct_add = s->dsp.add_pixels8;
dae006d7 1219 }else{
4693b031
MR
1220 idct_dc_add = h->h264dsp.h264_idct8_dc_add;
1221 idct_add = h->h264dsp.h264_idct8_add;
1eb96035 1222 }
43efd19a
LM
1223 for(i=0; i<16; i+=4){
1224 uint8_t * const ptr= dest_y + block_offset[i];
1225 const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
41e4055b 1226 if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
6e3ef511 1227 h->hpc.pred8x8l_add[dir](ptr, h->mb + (i*16 << pixel_shift), linesize);
41e4055b 1228 }else{
ac0623b2
MN
1229 const int nnz = h->non_zero_count_cache[ scan8[i] ];
1230 h->hpc.pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
1231 (h->topright_samples_available<<i)&0x4000, linesize);
1232 if(nnz){
6e3ef511
OA
1233 if(nnz == 1 && dctcoef_get(h->mb, pixel_shift, i*16))
1234 idct_dc_add(ptr, h->mb + (i*16 << pixel_shift), linesize);
ac0623b2 1235 else
6e3ef511 1236 idct_add (ptr, h->mb + (i*16 << pixel_shift), linesize);
ac0623b2 1237 }
41e4055b 1238 }
43efd19a 1239 }
1eb96035
MN
1240 }else{
1241 if(transform_bypass){
1242 idct_dc_add =
1243 idct_add = s->dsp.add_pixels4;
1244 }else{
4693b031
MR
1245 idct_dc_add = h->h264dsp.h264_idct_dc_add;
1246 idct_add = h->h264dsp.h264_idct_add;
1eb96035 1247 }
aebb5d6d
MN
1248 for(i=0; i<16; i++){
1249 uint8_t * const ptr= dest_y + block_offset[i];
1250 const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
e7e09b49 1251
aebb5d6d 1252 if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
6e3ef511 1253 h->hpc.pred4x4_add[dir](ptr, h->mb + (i*16 << pixel_shift), linesize);
aebb5d6d
MN
1254 }else{
1255 uint8_t *topright;
1256 int nnz, tr;
6e3ef511 1257 uint64_t tr_high;
aebb5d6d
MN
1258 if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
1259 const int topright_avail= (h->topright_samples_available<<i)&0x8000;
1260 assert(mb_y || linesize <= block_offset[i]);
1261 if(!topright_avail){
6e3ef511
OA
1262 if (pixel_shift) {
1263 tr_high= ((uint16_t*)ptr)[3 - linesize/2]*0x0001000100010001ULL;
1264 topright= (uint8_t*) &tr_high;
1265 } else {
aebb5d6d
MN
1266 tr= ptr[3 - linesize]*0x01010101;
1267 topright= (uint8_t*) &tr;
6e3ef511 1268 }
aebb5d6d 1269 }else
6e3ef511 1270 topright= ptr + (4 << pixel_shift) - linesize;
ac0623b2 1271 }else
aebb5d6d
MN
1272 topright= NULL;
1273
1274 h->hpc.pred4x4[ dir ](ptr, topright, linesize);
1275 nnz = h->non_zero_count_cache[ scan8[i] ];
1276 if(nnz){
1277 if(is_h264){
6e3ef511
OA
1278 if(nnz == 1 && dctcoef_get(h->mb, pixel_shift, i*16))
1279 idct_dc_add(ptr, h->mb + (i*16 << pixel_shift), linesize);
aebb5d6d 1280 else
6e3ef511 1281 idct_add (ptr, h->mb + (i*16 << pixel_shift), linesize);
aebb5d6d 1282 }else
881b5b80 1283 ff_svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
aebb5d6d 1284 }
ac0623b2 1285 }
41e4055b 1286 }
8b82a956 1287 }
0da71265 1288 }
e7e09b49 1289 }else{
c92a30bb 1290 h->hpc.pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
2e186601
JGG
1291 if(is_h264){
1292 if(h->non_zero_count_cache[ scan8[LUMA_DC_BLOCK_INDEX] ]){
0d1d01cf
JGG
1293 if(!transform_bypass)
1294 h->h264dsp.h264_luma_dc_dequant_idct(h->mb, h->mb_luma_dc, h->dequant4_coeff[0][s->qscale][0]);
1295 else{
1296 static const uint8_t dc_mapping[16] = { 0*16, 1*16, 4*16, 5*16, 2*16, 3*16, 6*16, 7*16,
1297 8*16, 9*16,12*16,13*16,10*16,11*16,14*16,15*16};
1298 for(i = 0; i < 16; i++)
6e3ef511 1299 dctcoef_set(h->mb, pixel_shift, dc_mapping[i], dctcoef_get(h->mb_luma_dc, pixel_shift, i));
0d1d01cf 1300 }
2e186601
JGG
1301 }
1302 }else
290fabc6 1303 ff_svq3_luma_dc_dequant_idct_c(h->mb, h->mb_luma_dc, s->qscale);
0da71265 1304 }
5f7f9719 1305 if(h->deblocking_filter)
6e3ef511 1306 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0, simple, pixel_shift);
bd91fee3 1307 }else if(is_h264){
6e3ef511
OA
1308 if (pixel_shift) {
1309 hl_motion_16(h, dest_y, dest_cb, dest_cr,
1310 s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
1311 s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
1312 h->h264dsp.weight_h264_pixels_tab,
1313 h->h264dsp.biweight_h264_pixels_tab);
1314 } else
1315 hl_motion_8(h, dest_y, dest_cb, dest_cr,
1316 s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
1317 s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
1318 h->h264dsp.weight_h264_pixels_tab,
1319 h->h264dsp.biweight_h264_pixels_tab);
0da71265 1320 }
e7e09b49
LLL
1321
1322
1323 if(!IS_INTRA4x4(mb_type)){
bd91fee3 1324 if(is_h264){
ef9d1d15 1325 if(IS_INTRA16x16(mb_type)){
2fd1f0e0
MN
1326 if(transform_bypass){
1327 if(h->sps.profile_idc==244 && (h->intra16x16_pred_mode==VERT_PRED8x8 || h->intra16x16_pred_mode==HOR_PRED8x8)){
0a8ca22f
MN
1328 h->hpc.pred16x16_add[h->intra16x16_pred_mode](dest_y, block_offset, h->mb, linesize);
1329 }else{
1330 for(i=0; i<16; i++){
6e3ef511
OA
1331 if(h->non_zero_count_cache[ scan8[i] ] || dctcoef_get(h->mb, pixel_shift, i*16))
1332 s->dsp.add_pixels4(dest_y + block_offset[i], h->mb + (i*16 << pixel_shift), linesize);
0a8ca22f 1333 }
2fd1f0e0
MN
1334 }
1335 }else{
4693b031 1336 h->h264dsp.h264_idct_add16intra(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
41e4055b 1337 }
49c084a7 1338 }else if(h->cbp&15){
2fd1f0e0 1339 if(transform_bypass){
0a8ca22f 1340 const int di = IS_8x8DCT(mb_type) ? 4 : 1;
1eb96035 1341 idct_add= IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4;
0a8ca22f 1342 for(i=0; i<16; i+=di){
62bc966f 1343 if(h->non_zero_count_cache[ scan8[i] ]){
6e3ef511 1344 idct_add(dest_y + block_offset[i], h->mb + (i*16 << pixel_shift), linesize);
0a8ca22f 1345 }
ef9d1d15 1346 }
2fd1f0e0
MN
1347 }else{
1348 if(IS_8x8DCT(mb_type)){
4693b031 1349 h->h264dsp.h264_idct8_add4(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
2fd1f0e0 1350 }else{
4693b031 1351 h->h264dsp.h264_idct_add16(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
2fd1f0e0
MN
1352 }
1353 }
4704097a 1354 }
e7e09b49
LLL
1355 }else{
1356 for(i=0; i<16; i++){
1357 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
6867a90b 1358 uint8_t * const ptr= dest_y + block_offset[i];
881b5b80 1359 ff_svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
e7e09b49 1360 }
4704097a 1361 }
0da71265
MN
1362 }
1363 }
0da71265 1364
49fb20cb 1365 if((simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)) && (h->cbp&0x30)){
ef9d1d15
LM
1366 uint8_t *dest[2] = {dest_cb, dest_cr};
1367 if(transform_bypass){
96465b90 1368 if(IS_INTRA(mb_type) && h->sps.profile_idc==244 && (h->chroma_pred_mode==VERT_PRED8x8 || h->chroma_pred_mode==HOR_PRED8x8)){
6e3ef511
OA
1369 h->hpc.pred8x8_add[h->chroma_pred_mode](dest[0], block_offset + 16, h->mb + (16*16 << pixel_shift), uvlinesize);
1370 h->hpc.pred8x8_add[h->chroma_pred_mode](dest[1], block_offset + 20, h->mb + (20*16 << pixel_shift), uvlinesize);
96465b90 1371 }else{
c25ac15a 1372 idct_add = s->dsp.add_pixels4;
96465b90 1373 for(i=16; i<16+8; i++){
6e3ef511
OA
1374 if(h->non_zero_count_cache[ scan8[i] ] || dctcoef_get(h->mb, pixel_shift, i*16))
1375 idct_add (dest[(i&4)>>2] + block_offset[i], h->mb + (i*16 << pixel_shift), uvlinesize);
96465b90
MN
1376 }
1377 }
ef9d1d15 1378 }else{
aebb5d6d 1379 if(is_h264){
2e186601 1380 if(h->non_zero_count_cache[ scan8[CHROMA_DC_BLOCK_INDEX+0] ])
6e3ef511 1381 h->h264dsp.h264_chroma_dc_dequant_idct(h->mb + (16*16 << pixel_shift) , h->dequant4_coeff[IS_INTRA(mb_type) ? 1:4][h->chroma_qp[0]][0]);
2e186601 1382 if(h->non_zero_count_cache[ scan8[CHROMA_DC_BLOCK_INDEX+1] ])
6e3ef511 1383 h->h264dsp.h264_chroma_dc_dequant_idct(h->mb + ((16*16+4*16) << pixel_shift), h->dequant4_coeff[IS_INTRA(mb_type) ? 2:5][h->chroma_qp[1]][0]);
1d16a1cf
RB
1384 h->h264dsp.h264_idct_add8(dest, block_offset,
1385 h->mb, uvlinesize,
1386 h->non_zero_count_cache);
aebb5d6d 1387 }else{
e39e3aba
OA
1388 h->h264dsp.h264_chroma_dc_dequant_idct(h->mb + 16*16 , h->dequant4_coeff[IS_INTRA(mb_type) ? 1:4][h->chroma_qp[0]][0]);
1389 h->h264dsp.h264_chroma_dc_dequant_idct(h->mb + 16*16+4*16, h->dequant4_coeff[IS_INTRA(mb_type) ? 2:5][h->chroma_qp[1]][0]);
aebb5d6d
MN
1390 for(i=16; i<16+8; i++){
1391 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
1392 uint8_t * const ptr= dest[(i&4)>>2] + block_offset[i];
fcc0224e 1393 ff_svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, ff_h264_chroma_qp[0][s->qscale + 12] - 12, 2);
aebb5d6d 1394 }
e7e09b49 1395 }
4704097a 1396 }
0da71265
MN
1397 }
1398 }
1399 }
c212fb0c
MN
1400 if(h->cbp || IS_INTRA(mb_type))
1401 s->dsp.clear_blocks(h->mb);
0da71265
MN
1402}
1403
0da71265 1404/**
bd91fee3
AS
1405 * Process a macroblock; this case avoids checks for expensive uncommon cases.
1406 */
6e3ef511
OA
1407#define hl_decode_mb_simple(sh, bits) \
1408static void hl_decode_mb_simple_ ## bits(H264Context *h){ \
1409 hl_decode_mb_internal(h, 1, sh); \
bd91fee3 1410}
6e3ef511
OA
1411hl_decode_mb_simple(0, 8);
1412hl_decode_mb_simple(1, 16);
bd91fee3
AS
1413
1414/**
1415 * Process a macroblock; this handles edge cases, such as interlacing.
1416 */
1417static void av_noinline hl_decode_mb_complex(H264Context *h){
6e3ef511 1418 hl_decode_mb_internal(h, 0, h->pixel_shift);
bd91fee3
AS
1419}
1420
903d58f6 1421void ff_h264_hl_decode_mb(H264Context *h){
bd91fee3 1422 MpegEncContext * const s = &h->s;
64514ee8 1423 const int mb_xy= h->mb_xy;
bd91fee3 1424 const int mb_type= s->current_picture.mb_type[mb_xy];
49fb20cb 1425 int is_complex = CONFIG_SMALL || h->is_complex || IS_INTRA_PCM(mb_type) || s->qscale == 0;
bd91fee3 1426
6e3ef511 1427 if (is_complex) {
bd91fee3 1428 hl_decode_mb_complex(h);
6e3ef511
OA
1429 } else if (h->pixel_shift) {
1430 hl_decode_mb_simple_16(h);
1431 } else
1432 hl_decode_mb_simple_8(h);
bd91fee3
AS
1433}
1434
0da71265
MN
1435static int pred_weight_table(H264Context *h){
1436 MpegEncContext * const s = &h->s;
1437 int list, i;
9f2d1b4f 1438 int luma_def, chroma_def;
115329f1 1439
9f2d1b4f
LM
1440 h->use_weight= 0;
1441 h->use_weight_chroma= 0;
0da71265 1442 h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
09fffe9b
JGG
1443 if(CHROMA)
1444 h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
9f2d1b4f
LM
1445 luma_def = 1<<h->luma_log2_weight_denom;
1446 chroma_def = 1<<h->chroma_log2_weight_denom;
0da71265
MN
1447
1448 for(list=0; list<2; list++){
cb99c652
GB
1449 h->luma_weight_flag[list] = 0;
1450 h->chroma_weight_flag[list] = 0;
0da71265
MN
1451 for(i=0; i<h->ref_count[list]; i++){
1452 int luma_weight_flag, chroma_weight_flag;
115329f1 1453
0da71265
MN
1454 luma_weight_flag= get_bits1(&s->gb);
1455 if(luma_weight_flag){
3d9137c8
MN
1456 h->luma_weight[i][list][0]= get_se_golomb(&s->gb);
1457 h->luma_weight[i][list][1]= get_se_golomb(&s->gb);
1458 if( h->luma_weight[i][list][0] != luma_def
1459 || h->luma_weight[i][list][1] != 0) {
9f2d1b4f 1460 h->use_weight= 1;
cb99c652
GB
1461 h->luma_weight_flag[list]= 1;
1462 }
9f2d1b4f 1463 }else{
3d9137c8
MN
1464 h->luma_weight[i][list][0]= luma_def;
1465 h->luma_weight[i][list][1]= 0;
0da71265
MN
1466 }
1467
0af6967e 1468 if(CHROMA){
fef744d4
MN
1469 chroma_weight_flag= get_bits1(&s->gb);
1470 if(chroma_weight_flag){
1471 int j;
1472 for(j=0; j<2; j++){
3d9137c8
MN
1473 h->chroma_weight[i][list][j][0]= get_se_golomb(&s->gb);
1474 h->chroma_weight[i][list][j][1]= get_se_golomb(&s->gb);
1475 if( h->chroma_weight[i][list][j][0] != chroma_def
1476 || h->chroma_weight[i][list][j][1] != 0) {
fef744d4 1477 h->use_weight_chroma= 1;
cb99c652
GB
1478 h->chroma_weight_flag[list]= 1;
1479 }
fef744d4
MN
1480 }
1481 }else{
1482 int j;
1483 for(j=0; j<2; j++){
3d9137c8
MN
1484 h->chroma_weight[i][list][j][0]= chroma_def;
1485 h->chroma_weight[i][list][j][1]= 0;
fef744d4 1486 }
0da71265
MN
1487 }
1488 }
1489 }
975a1447 1490 if(h->slice_type_nos != AV_PICTURE_TYPE_B) break;
0da71265 1491 }
9f2d1b4f 1492 h->use_weight= h->use_weight || h->use_weight_chroma;
0da71265
MN
1493 return 0;
1494}
1495
1052b76f
MN
1496/**
1497 * Initialize implicit_weight table.
6da88bd3 1498 * @param field 0/1 initialize the weight for interlaced MBAFF
1052b76f
MN
1499 * -1 initializes the rest
1500 */
1501static void implicit_weight_table(H264Context *h, int field){
9f2d1b4f 1502 MpegEncContext * const s = &h->s;
1052b76f 1503 int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
9f2d1b4f 1504
ce09f927
GB
1505 for (i = 0; i < 2; i++) {
1506 h->luma_weight_flag[i] = 0;
1507 h->chroma_weight_flag[i] = 0;
1508 }
1509
1052b76f
MN
1510 if(field < 0){
1511 cur_poc = s->current_picture_ptr->poc;
1512 if( h->ref_count[0] == 1 && h->ref_count[1] == 1 && !FRAME_MBAFF
9f2d1b4f
LM
1513 && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
1514 h->use_weight= 0;
1515 h->use_weight_chroma= 0;
1516 return;
1517 }
1052b76f
MN
1518 ref_start= 0;
1519 ref_count0= h->ref_count[0];
1520 ref_count1= h->ref_count[1];
1521 }else{
1522 cur_poc = s->current_picture_ptr->field_poc[field];
1523 ref_start= 16;
1524 ref_count0= 16+2*h->ref_count[0];
1525 ref_count1= 16+2*h->ref_count[1];
1526 }
9f2d1b4f
LM
1527
1528 h->use_weight= 2;
1529 h->use_weight_chroma= 2;
1530 h->luma_log2_weight_denom= 5;
1531 h->chroma_log2_weight_denom= 5;
1532
1052b76f 1533 for(ref0=ref_start; ref0 < ref_count0; ref0++){
9f2d1b4f 1534 int poc0 = h->ref_list[0][ref0].poc;
1052b76f 1535 for(ref1=ref_start; ref1 < ref_count1; ref1++){
738386a5 1536 int poc1 = h->ref_list[1][ref1].poc;
f66e4f5f 1537 int td = av_clip(poc1 - poc0, -128, 127);
1052b76f 1538 int w= 32;
9f2d1b4f 1539 if(td){
f66e4f5f 1540 int tb = av_clip(cur_poc - poc0, -128, 127);
c26abfa5 1541 int tx = (16384 + (FFABS(td) >> 1)) / td;
72f86ec0
MN
1542 int dist_scale_factor = (tb*tx + 32) >> 8;
1543 if(dist_scale_factor >= -64 && dist_scale_factor <= 128)
1052b76f
MN
1544 w = 64 - dist_scale_factor;
1545 }
1546 if(field<0){
1547 h->implicit_weight[ref0][ref1][0]=
1548 h->implicit_weight[ref0][ref1][1]= w;
1549 }else{
1550 h->implicit_weight[ref0][ref1][field]=w;
72f86ec0 1551 }
9f2d1b4f
LM
1552 }
1553 }
1554}
1555
8fd57a66 1556/**
5175b937 1557 * instantaneous decoder refresh.
0da71265
MN
1558 */
1559static void idr(H264Context *h){
ea6f00c4 1560 ff_h264_remove_all_refs(h);
a149c1a5 1561 h->prev_frame_num= 0;
80f8e035
MN
1562 h->prev_frame_num_offset= 0;
1563 h->prev_poc_msb=
1564 h->prev_poc_lsb= 0;
0da71265
MN
1565}
1566
7c33ad19
LM
1567/* forget old pics after a seek */
1568static void flush_dpb(AVCodecContext *avctx){
1569 H264Context *h= avctx->priv_data;
1570 int i;
64b9d48f 1571 for(i=0; i<MAX_DELAYED_PIC_COUNT; i++) {
285b570f
LM
1572 if(h->delayed_pic[i])
1573 h->delayed_pic[i]->reference= 0;
7c33ad19 1574 h->delayed_pic[i]= NULL;
285b570f 1575 }
df8a7dff 1576 h->outputed_poc= INT_MIN;
b19d493f 1577 h->prev_interlaced_frame = 1;
7c33ad19 1578 idr(h);
ca159196
MR
1579 if(h->s.current_picture_ptr)
1580 h->s.current_picture_ptr->reference= 0;
12d96de3 1581 h->s.first_field= 0;
9c095463 1582 ff_h264_reset_sei(h);
e240f898 1583 ff_mpeg_flush(avctx);
7c33ad19
LM
1584}
1585
0da71265
MN
1586static int init_poc(H264Context *h){
1587 MpegEncContext * const s = &h->s;
1588 const int max_frame_num= 1<<h->sps.log2_max_frame_num;
1589 int field_poc[2];
357282c6 1590 Picture *cur = s->current_picture_ptr;
0da71265 1591
b78a6baa 1592 h->frame_num_offset= h->prev_frame_num_offset;
5710b371 1593 if(h->frame_num < h->prev_frame_num)
b78a6baa 1594 h->frame_num_offset += max_frame_num;
0da71265
MN
1595
1596 if(h->sps.poc_type==0){
1597 const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
1598
1599 if (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
1600 h->poc_msb = h->prev_poc_msb + max_poc_lsb;
1601 else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
1602 h->poc_msb = h->prev_poc_msb - max_poc_lsb;
1603 else
1604 h->poc_msb = h->prev_poc_msb;
1605//printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
115329f1 1606 field_poc[0] =
0da71265 1607 field_poc[1] = h->poc_msb + h->poc_lsb;
115329f1 1608 if(s->picture_structure == PICT_FRAME)
0da71265
MN
1609 field_poc[1] += h->delta_poc_bottom;
1610 }else if(h->sps.poc_type==1){
1611 int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
1612 int i;
1613
1614 if(h->sps.poc_cycle_length != 0)
1615 abs_frame_num = h->frame_num_offset + h->frame_num;
1616 else
1617 abs_frame_num = 0;
1618
1619 if(h->nal_ref_idc==0 && abs_frame_num > 0)
1620 abs_frame_num--;
115329f1 1621
0da71265
MN
1622 expected_delta_per_poc_cycle = 0;
1623 for(i=0; i < h->sps.poc_cycle_length; i++)
1624 expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
1625
1626 if(abs_frame_num > 0){
1627 int poc_cycle_cnt = (abs_frame_num - 1) / h->sps.poc_cycle_length;
1628 int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
1629
1630 expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
1631 for(i = 0; i <= frame_num_in_poc_cycle; i++)
1632 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
1633 } else
1634 expectedpoc = 0;
1635
115329f1 1636 if(h->nal_ref_idc == 0)
0da71265 1637 expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
115329f1 1638
0da71265
MN
1639 field_poc[0] = expectedpoc + h->delta_poc[0];
1640 field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
1641
1642 if(s->picture_structure == PICT_FRAME)
1643 field_poc[1] += h->delta_poc[1];
1644 }else{
b78a6baa 1645 int poc= 2*(h->frame_num_offset + h->frame_num);
5710b371 1646
b78a6baa
MN
1647 if(!h->nal_ref_idc)
1648 poc--;
5710b371 1649
0da71265
MN
1650 field_poc[0]= poc;
1651 field_poc[1]= poc;
1652 }
115329f1 1653
357282c6 1654 if(s->picture_structure != PICT_BOTTOM_FIELD)
0da71265 1655 s->current_picture_ptr->field_poc[0]= field_poc[0];
357282c6 1656 if(s->picture_structure != PICT_TOP_FIELD)
0da71265 1657 s->current_picture_ptr->field_poc[1]= field_poc[1];
357282c6 1658 cur->poc= FFMIN(cur->field_poc[0], cur->field_poc[1]);
0da71265
MN
1659
1660 return 0;
1661}
1662
b41c1db3
1663
1664/**
1665 * initialize scan tables
1666 */
1667static void init_scan_tables(H264Context *h){
b41c1db3 1668 int i;
ca32f7f2 1669 for(i=0; i<16; i++){
b41c1db3 1670#define T(x) (x>>2) | ((x<<2) & 0xF)
ca32f7f2
JGG
1671 h->zigzag_scan[i] = T(zigzag_scan[i]);
1672 h-> field_scan[i] = T( field_scan[i]);
b41c1db3 1673#undef T
b41c1db3 1674 }
ca32f7f2 1675 for(i=0; i<64; i++){
b41c1db3 1676#define T(x) (x>>3) | ((x&7)<<3)
ca32f7f2
JGG
1677 h->zigzag_scan8x8[i] = T(ff_zigzag_direct[i]);
1678 h->zigzag_scan8x8_cavlc[i] = T(zigzag_scan8x8_cavlc[i]);
1679 h->field_scan8x8[i] = T(field_scan8x8[i]);
1680 h->field_scan8x8_cavlc[i] = T(field_scan8x8_cavlc[i]);
b41c1db3 1681#undef T
b41c1db3
1682 }
1683 if(h->sps.transform_bypass){ //FIXME same ugly
1684 h->zigzag_scan_q0 = zigzag_scan;
45beb850 1685 h->zigzag_scan8x8_q0 = ff_zigzag_direct;
b41c1db3
1686 h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc;
1687 h->field_scan_q0 = field_scan;
1688 h->field_scan8x8_q0 = field_scan8x8;
1689 h->field_scan8x8_cavlc_q0 = field_scan8x8_cavlc;
1690 }else{
1691 h->zigzag_scan_q0 = h->zigzag_scan;
1692 h->zigzag_scan8x8_q0 = h->zigzag_scan8x8;
1693 h->zigzag_scan8x8_cavlc_q0 = h->zigzag_scan8x8_cavlc;
1694 h->field_scan_q0 = h->field_scan;
1695 h->field_scan8x8_q0 = h->field_scan8x8;
1696 h->field_scan8x8_cavlc_q0 = h->field_scan8x8_cavlc;
1697 }
1698}
afebe2f7 1699
256299d3
MN
1700static void field_end(H264Context *h){
1701 MpegEncContext * const s = &h->s;
1702 AVCodecContext * const avctx= s->avctx;
1703 s->mb_y= 0;
1704
1705 s->current_picture_ptr->qscale_type= FF_QSCALE_TYPE_H264;
1706 s->current_picture_ptr->pict_type= s->pict_type;
1707
1708 if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
1709 ff_vdpau_h264_set_reference_frames(s);
1710
1711 if(!s->dropable) {
ea6f00c4 1712 ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
256299d3
MN
1713 h->prev_poc_msb= h->poc_msb;
1714 h->prev_poc_lsb= h->poc_lsb;
1715 }
1716 h->prev_frame_num_offset= h->frame_num_offset;
1717 h->prev_frame_num= h->frame_num;
1718
1719 if (avctx->hwaccel) {
1720 if (avctx->hwaccel->end_frame(avctx) < 0)
1721 av_log(avctx, AV_LOG_ERROR, "hardware accelerator failed to decode picture\n");
1722 }
1723
1724 if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
1725 ff_vdpau_h264_picture_complete(s);
1726
1727 /*
1728 * FIXME: Error handling code does not seem to support interlaced
1729 * when slices span multiple rows
1730 * The ff_er_add_slice calls don't work right for bottom
1731 * fields; they cause massive erroneous error concealing
1732 * Error marking covers both fields (top and bottom).
1733 * This causes a mismatched s->error_count
1734 * and a bad error table. Further, the error count goes to
1735 * INT_MAX when called for bottom field, because mb_y is
1736 * past end by one (callers fault) and resync_mb_y != 0
1737 * causes problems for the first MB line, too.
1738 */
1739 if (!FIELD_PICTURE)
1740 ff_er_frame_end(s);
1741
1742 MPV_frame_end(s);
d225a1e2
MN
1743
1744 h->current_slice=0;
256299d3
MN
1745}
1746
afebe2f7 1747/**
49bd8e4b 1748 * Replicate H264 "master" context to thread contexts.
afebe2f7
1749 */
1750static void clone_slice(H264Context *dst, H264Context *src)
1751{
1752 memcpy(dst->block_offset, src->block_offset, sizeof(dst->block_offset));
1753 dst->s.current_picture_ptr = src->s.current_picture_ptr;
1754 dst->s.current_picture = src->s.current_picture;
1755 dst->s.linesize = src->s.linesize;
1756 dst->s.uvlinesize = src->s.uvlinesize;
12d96de3 1757 dst->s.first_field = src->s.first_field;
afebe2f7
1758
1759 dst->prev_poc_msb = src->prev_poc_msb;
1760 dst->prev_poc_lsb = src->prev_poc_lsb;
1761 dst->prev_frame_num_offset = src->prev_frame_num_offset;
1762 dst->prev_frame_num = src->prev_frame_num;
1763 dst->short_ref_count = src->short_ref_count;
1764
1765 memcpy(dst->short_ref, src->short_ref, sizeof(dst->short_ref));
1766 memcpy(dst->long_ref, src->long_ref, sizeof(dst->long_ref));
1767 memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
1768 memcpy(dst->ref_list, src->ref_list, sizeof(dst->ref_list));
50c21814
1769
1770 memcpy(dst->dequant4_coeff, src->dequant4_coeff, sizeof(src->dequant4_coeff));
1771 memcpy(dst->dequant8_coeff, src->dequant8_coeff, sizeof(src->dequant8_coeff));
afebe2f7
1772}
1773
0da71265 1774/**
fe9a3fbe
JG
1775 * computes profile from profile_idc and constraint_set?_flags
1776 *
1777 * @param sps SPS
1778 *
1779 * @return profile as defined by FF_PROFILE_H264_*
1780 */
1781int ff_h264_get_profile(SPS *sps)
1782{
1783 int profile = sps->profile_idc;
1784
1785 switch(sps->profile_idc) {
1786 case FF_PROFILE_H264_BASELINE:
1787 // constraint_set1_flag set to 1
1788 profile |= (sps->constraint_set_flags & 1<<1) ? FF_PROFILE_H264_CONSTRAINED : 0;
1789 break;
1790 case FF_PROFILE_H264_HIGH_10:
1791 case FF_PROFILE_H264_HIGH_422:
1792 case FF_PROFILE_H264_HIGH_444_PREDICTIVE:
1793 // constraint_set3_flag set to 1
1794 profile |= (sps->constraint_set_flags & 1<<3) ? FF_PROFILE_H264_INTRA : 0;
1795 break;
1796 }
1797
1798 return profile;
1799}
1800
1801/**
0da71265 1802 * decodes a slice header.
9c852bcf 1803 * This will also call MPV_common_init() and frame_start() as needed.
afebe2f7
1804 *
1805 * @param h h264context
1806 * @param h0 h264 master context (differs from 'h' when doing sliced based parallel decoding)
1807 *
d9526386 1808 * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
0da71265 1809 */
afebe2f7 1810static int decode_slice_header(H264Context *h, H264Context *h0){
0da71265 1811 MpegEncContext * const s = &h->s;
12d96de3 1812 MpegEncContext * const s0 = &h0->s;
88e7a4d1 1813 unsigned int first_mb_in_slice;
ac658be5 1814 unsigned int pps_id;
0da71265 1815 int num_ref_idx_active_override_flag;
41f5c62f 1816 unsigned int slice_type, tmp, i, j;
0bf79634 1817 int default_ref_list_done = 0;
12d96de3 1818 int last_pic_structure;
0da71265 1819
2f944356 1820 s->dropable= h->nal_ref_idc == 0;
0da71265 1821
cf653d08
JD
1822 if((s->avctx->flags2 & CODEC_FLAG2_FAST) && !h->nal_ref_idc){
1823 s->me.qpel_put= s->dsp.put_2tap_qpel_pixels_tab;
1824 s->me.qpel_avg= s->dsp.avg_2tap_qpel_pixels_tab;
1825 }else{
1826 s->me.qpel_put= s->dsp.put_h264_qpel_pixels_tab;
1827 s->me.qpel_avg= s->dsp.avg_h264_qpel_pixels_tab;
1828 }
1829
0da71265
MN
1830 first_mb_in_slice= get_ue_golomb(&s->gb);
1831
d225a1e2
MN
1832 if(first_mb_in_slice == 0){ //FIXME better field boundary detection
1833 if(h0->current_slice && FIELD_PICTURE){
1834 field_end(h);
1835 }
1836
afebe2f7 1837 h0->current_slice = 0;
12d96de3 1838 if (!s0->first_field)
f6e3c460 1839 s->current_picture_ptr= NULL;
66a4b2c1
MN
1840 }
1841
9963b332 1842 slice_type= get_ue_golomb_31(&s->gb);
0bf79634 1843 if(slice_type > 9){
9b879566 1844 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);
5175b937 1845 return -1;
0da71265 1846 }
0bf79634
LLL
1847 if(slice_type > 4){
1848 slice_type -= 5;
0da71265
MN
1849 h->slice_type_fixed=1;
1850 }else
1851 h->slice_type_fixed=0;
115329f1 1852
ee2a957f 1853 slice_type= golomb_to_pict_type[ slice_type ];
975a1447 1854 if (slice_type == AV_PICTURE_TYPE_I
afebe2f7 1855 || (h0->current_slice != 0 && slice_type == h0->last_slice_type) ) {
0bf79634
LLL
1856 default_ref_list_done = 1;
1857 }
1858 h->slice_type= slice_type;
e3e6f18f 1859 h->slice_type_nos= slice_type & 3;
0bf79634 1860
1412060e 1861 s->pict_type= h->slice_type; // to make a few old functions happy, it's wrong though
115329f1 1862
0da71265 1863 pps_id= get_ue_golomb(&s->gb);
ac658be5 1864 if(pps_id>=MAX_PPS_COUNT){
9b879566 1865 av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
0da71265
MN
1866 return -1;
1867 }
afebe2f7 1868 if(!h0->pps_buffers[pps_id]) {
a0f80050 1869 av_log(h->s.avctx, AV_LOG_ERROR, "non-existing PPS %u referenced\n", pps_id);
8b92b792
MN
1870 return -1;
1871 }
afebe2f7 1872 h->pps= *h0->pps_buffers[pps_id];
8b92b792 1873
afebe2f7 1874 if(!h0->sps_buffers[h->pps.sps_id]) {
a0f80050 1875 av_log(h->s.avctx, AV_LOG_ERROR, "non-existing SPS %u referenced\n", h->pps.sps_id);
8b92b792
MN
1876 return -1;
1877 }
afebe2f7 1878 h->sps = *h0->sps_buffers[h->pps.sps_id];
239ea04c 1879
fe9a3fbe 1880 s->avctx->profile = ff_h264_get_profile(&h->sps);
fa37cf0d 1881 s->avctx->level = h->sps.level_idc;
6752a3cc 1882 s->avctx->refs = h->sps.ref_frame_count;
b08e38e8 1883
50c21814 1884 if(h == h0 && h->dequant_coeff_pps != pps_id){
50eaa857 1885 h->dequant_coeff_pps = pps_id;
239ea04c
LM
1886 init_dequant_tables(h);
1887 }
115329f1 1888
0da71265 1889 s->mb_width= h->sps.mb_width;
6867a90b 1890 s->mb_height= h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
115329f1 1891
bf4665ee 1892 h->b_stride= s->mb_width*4;
0da71265 1893
faf3dfb9 1894 s->width = 16*s->mb_width - 2*FFMIN(h->sps.crop_right, 7);
0da71265 1895 if(h->sps.frame_mbs_only_flag)
faf3dfb9 1896 s->height= 16*s->mb_height - 2*FFMIN(h->sps.crop_bottom, 7);
0da71265 1897 else
00d1e96b 1898 s->height= 16*s->mb_height - 4*FFMIN(h->sps.crop_bottom, 7);
115329f1
DB
1899
1900 if (s->context_initialized
5388f0b4
JK
1901 && ( s->width != s->avctx->width || s->height != s->avctx->height
1902 || av_cmp_q(h->sps.sar, s->avctx->sample_aspect_ratio))) {
afebe2f7
1903 if(h != h0)
1904 return -1; // width / height changed during parallelized decoding
91078926 1905 free_tables(h, 0);
ff7f75e1 1906 flush_dpb(s->avctx);
0da71265
MN
1907 MPV_common_end(s);
1908 }
1909 if (!s->context_initialized) {
afebe2f7
1910 if(h != h0)
1911 return -1; // we cant (re-)initialize context during parallel decoding
f3bdc3da
RD
1912
1913 avcodec_set_dimensions(s->avctx, s->width, s->height);
1914 s->avctx->sample_aspect_ratio= h->sps.sar;
cfa5a81e 1915 av_assert0(s->avctx->sample_aspect_ratio.den);
f3bdc3da 1916
b47904d1
FC
1917 h->s.avctx->coded_width = 16*s->mb_width;
1918 h->s.avctx->coded_height = 16*s->mb_height;
1919
c4dffe7e
DC
1920 if(h->sps.video_signal_type_present_flag){
1921 s->avctx->color_range = h->sps.full_range ? AVCOL_RANGE_JPEG : AVCOL_RANGE_MPEG;
1922 if(h->sps.colour_description_present_flag){
1923 s->avctx->color_primaries = h->sps.color_primaries;
1924 s->avctx->color_trc = h->sps.color_trc;
1925 s->avctx->colorspace = h->sps.colorspace;
1926 }
1927 }
1928
f3bdc3da 1929 if(h->sps.timing_info_present_flag){
3102d180 1930 int64_t den= h->sps.time_scale;
055a6aa7 1931 if(h->x264_build < 44U)
3102d180 1932 den *= 2;
f3bdc3da 1933 av_reduce(&s->avctx->time_base.num, &s->avctx->time_base.den,
3102d180 1934 h->sps.num_units_in_tick, den, 1<<30);
f3bdc3da 1935 }
d545cf80
OA
1936
1937 switch (h->sps.bit_depth_luma) {
1938 case 9 :
1939 s->avctx->pix_fmt = PIX_FMT_YUV420P9;
1940 break;
1941 case 10 :
1942 s->avctx->pix_fmt = PIX_FMT_YUV420P10;
1943 break;
1944 default:
0435fb16 1945 s->avctx->pix_fmt = s->avctx->get_format(s->avctx,
9ad7dfc1
BC
1946 s->avctx->codec->pix_fmts ?
1947 s->avctx->codec->pix_fmts :
0435fb16
BC
1948 s->avctx->color_range == AVCOL_RANGE_JPEG ?
1949 hwaccel_pixfmt_list_h264_jpeg_420 :
1950 ff_hwaccel_pixfmt_list_420);
d545cf80
OA
1951 }
1952
f3bdc3da
RD
1953 s->avctx->hwaccel = ff_find_hwaccel(s->avctx->codec->id, s->avctx->pix_fmt);
1954
0da71265
MN
1955 if (MPV_common_init(s) < 0)
1956 return -1;
12d96de3 1957 s->first_field = 0;
b19d493f 1958 h->prev_interlaced_frame = 1;
115329f1 1959
b41c1db3 1960 init_scan_tables(h);
903d58f6 1961 ff_h264_alloc_tables(h);
0da71265 1962
afebe2f7
1963 for(i = 1; i < s->avctx->thread_count; i++) {
1964 H264Context *c;
1965 c = h->thread_context[i] = av_malloc(sizeof(H264Context));
79db7ac6 1966 memcpy(c, h->s.thread_context[i], sizeof(MpegEncContext));
afebe2f7 1967 memset(&c->s + 1, 0, sizeof(H264Context) - sizeof(MpegEncContext));
7a5c850b 1968 c->h264dsp = h->h264dsp;
afebe2f7
1969 c->sps = h->sps;
1970 c->pps = h->pps;
8d44cd2c 1971 c->pixel_shift = h->pixel_shift;
afebe2f7 1972 init_scan_tables(c);
145061a1 1973 clone_tables(c, h, i);
afebe2f7
1974 }
1975
1976 for(i = 0; i < s->avctx->thread_count; i++)
1977 if(context_init(h->thread_context[i]) < 0)
1978 return -1;
0da71265
MN
1979 }
1980
0da71265
MN
1981 h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
1982
5d18eaad 1983 h->mb_mbaff = 0;
6ba71fc4 1984 h->mb_aff_frame = 0;
12d96de3 1985 last_pic_structure = s0->picture_structure;
0da71265
MN
1986 if(h->sps.frame_mbs_only_flag){
1987 s->picture_structure= PICT_FRAME;
1988 }else{
6ba71fc4 1989 if(get_bits1(&s->gb)) { //field_pic_flag
0da71265 1990 s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
6ba71fc4 1991 } else {
0da71265 1992 s->picture_structure= PICT_FRAME;
6ba71fc4 1993 h->mb_aff_frame = h->sps.mb_aff;
6867a90b 1994 }
0da71265 1995 }
44e9dcf1 1996 h->mb_field_decoding_flag= s->picture_structure != PICT_FRAME;
2ddcf84b
JD
1997
1998 if(h0->current_slice == 0){
26b86e47
MN
1999 while(h->frame_num != h->prev_frame_num &&
2000 h->frame_num != (h->prev_frame_num+1)%(1<<h->sps.log2_max_frame_num)){
4dece8c7 2001 Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
bb943bb8 2002 av_log(h->s.avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n", h->frame_num, h->prev_frame_num);
903d58f6 2003 if (ff_h264_frame_start(h) < 0)
66e6038c 2004 return -1;
26b86e47
MN
2005 h->prev_frame_num++;
2006 h->prev_frame_num %= 1<<h->sps.log2_max_frame_num;
2007 s->current_picture_ptr->frame_num= h->prev_frame_num;
3d542120
MN
2008 ff_generate_sliding_window_mmcos(h);
2009 ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
e2983d6e
JGG
2010 /* Error concealment: if a ref is missing, copy the previous ref in its place.
2011 * FIXME: avoiding a memcpy would be nice, but ref handling makes many assumptions
2012 * about there being no actual duplicates.
2013 * FIXME: this doesn't copy padding for out-of-frame motion vectors. Given we're
2014 * concealing a lost frame, this probably isn't noticable by comparison, but it should
2015 * be fixed. */
4dece8c7
JGG
2016 if (h->short_ref_count) {
2017 if (prev) {
2018 av_image_copy(h->short_ref[0]->data, h->short_ref[0]->linesize,
2019 (const uint8_t**)prev->data, prev->linesize,
5dd7f994 2020 s->avctx->pix_fmt, s->mb_width*16, s->mb_height*16);
4dece8c7
JGG
2021 h->short_ref[0]->poc = prev->poc+2;
2022 }
2023 h->short_ref[0]->frame_num = h->prev_frame_num;
2024 }
26b86e47
MN
2025 }
2026
12d96de3
JD
2027 /* See if we have a decoded first field looking for a pair... */
2028 if (s0->first_field) {
2029 assert(s0->current_picture_ptr);
2030 assert(s0->current_picture_ptr->data[0]);
2031 assert(s0->current_picture_ptr->reference != DELAYED_PIC_REF);
2032
2033 /* figure out if we have a complementary field pair */
2034 if (!FIELD_PICTURE || s->picture_structure == last_pic_structure) {
2035 /*
2036 * Previous field is unmatched. Don't display it, but let it
2037 * remain for reference if marked as such.
2038 */
2039 s0->current_picture_ptr = NULL;
2040 s0->first_field = FIELD_PICTURE;
2041
2042 } else {
2043 if (h->nal_ref_idc &&
2044 s0->current_picture_ptr->reference &&
2045 s0->current_picture_ptr->frame_num != h->frame_num) {
2046 /*
2047 * This and previous field were reference, but had
2048 * different frame_nums. Consider this field first in
2049 * pair. Throw away previous field except for reference
2050 * purposes.
2051 */
2052 s0->first_field = 1;
2053 s0->current_picture_ptr = NULL;
2054
2055 } else {
2056 /* Second field in complementary pair */
2057 s0->first_field = 0;
2058 }
2059 }
2060
2061 } else {
2062 /* Frame or first field in a potentially complementary pair */
2063 assert(!s0->current_picture_ptr);
2064 s0->first_field = FIELD_PICTURE;
2065 }
2066
903d58f6 2067 if((!FIELD_PICTURE || s0->first_field) && ff_h264_frame_start(h) < 0) {
12d96de3 2068 s0->first_field = 0;
2ddcf84b 2069 return -1;
12d96de3 2070 }
2ddcf84b
JD
2071 }
2072 if(h != h0)
2073 clone_slice(h, h0);
2074
2075 s->current_picture_ptr->frame_num= h->frame_num; //FIXME frame_num cleanup
2076
88e7a4d1 2077 assert(s->mb_num == s->mb_width * s->mb_height);
f3e53d9f 2078 if(first_mb_in_slice << FIELD_OR_MBAFF_PICTURE >= s->mb_num ||
88e7a4d1
MN
2079 first_mb_in_slice >= s->mb_num){
2080 av_log(h->s.avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
6b53b87e
MN
2081 return -1;
2082 }
88e7a4d1 2083 s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
f3e53d9f
JD
2084 s->resync_mb_y = s->mb_y = (first_mb_in_slice / s->mb_width) << FIELD_OR_MBAFF_PICTURE;
2085 if (s->picture_structure == PICT_BOTTOM_FIELD)
2086 s->resync_mb_y = s->mb_y = s->mb_y + 1;
88e7a4d1 2087 assert(s->mb_y < s->mb_height);
115329f1 2088
0da71265
MN
2089 if(s->picture_structure==PICT_FRAME){
2090 h->curr_pic_num= h->frame_num;
2091 h->max_pic_num= 1<< h->sps.log2_max_frame_num;
2092 }else{
f57e2af6 2093 h->curr_pic_num= 2*h->frame_num + 1;
0da71265
MN
2094 h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
2095 }
115329f1 2096
0da71265 2097 if(h->nal_unit_type == NAL_IDR_SLICE){
1df1df0b 2098 get_ue_golomb(&s->gb); /* idr_pic_id */
0da71265 2099 }
115329f1 2100
0da71265
MN
2101 if(h->sps.poc_type==0){
2102 h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
115329f1 2103
0da71265
MN
2104 if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
2105 h->delta_poc_bottom= get_se_golomb(&s->gb);
2106 }
2107 }
115329f1 2108
0da71265
MN
2109 if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
2110 h->delta_poc[0]= get_se_golomb(&s->gb);
115329f1 2111
0da71265
MN
2112 if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
2113 h->delta_poc[1]= get_se_golomb(&s->gb);
2114 }
115329f1 2115
0da71265 2116 init_poc(h);
115329f1 2117
0da71265
MN
2118 if(h->pps.redundant_pic_cnt_present){
2119 h->redundant_pic_count= get_ue_golomb(&s->gb);
2120 }
2121
1412060e 2122 //set defaults, might be overridden a few lines later
0da71265
MN
2123 h->ref_count[0]= h->pps.ref_count[0];
2124 h->ref_count[1]= h->pps.ref_count[1];
2125
975a1447
SS
2126 if(h->slice_type_nos != AV_PICTURE_TYPE_I){
2127 if(h->slice_type_nos == AV_PICTURE_TYPE_B){
0da71265
MN
2128 h->direct_spatial_mv_pred= get_bits1(&s->gb);
2129 }
2130 num_ref_idx_active_override_flag= get_bits1(&s->gb);
115329f1 2131
0da71265
MN
2132 if(num_ref_idx_active_override_flag){
2133 h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
975a1447 2134 if(h->slice_type_nos==AV_PICTURE_TYPE_B)
0da71265
MN
2135 h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
2136
187696fa 2137 if(h->ref_count[0]-1 > 32-1 || h->ref_count[1]-1 > 32-1){
9b879566 2138 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
88e7a4d1 2139 h->ref_count[0]= h->ref_count[1]= 1;
0da71265
MN
2140 return -1;
2141 }
2142 }
975a1447 2143 if(h->slice_type_nos == AV_PICTURE_TYPE_B)
187696fa
MN
2144 h->list_count= 2;
2145 else
2146 h->list_count= 1;
2147 }else
2148 h->list_count= 0;
0da71265 2149
0bf79634 2150 if(!default_ref_list_done){
ea6f00c4 2151 ff_h264_fill_default_ref_list(h);
0da71265
MN
2152 }
2153
975a1447 2154 if(h->slice_type_nos!=AV_PICTURE_TYPE_I && ff_h264_decode_ref_pic_list_reordering(h) < 0)
806bb93f 2155 return -1;
0da71265 2156
975a1447 2157 if(h->slice_type_nos!=AV_PICTURE_TYPE_I){
07dff5c7 2158 s->last_picture_ptr= &h->ref_list[0][0];
8d2fc163 2159 ff_copy_picture(&s->last_picture, s->last_picture_ptr);
07dff5c7 2160 }
975a1447 2161 if(h->slice_type_nos==AV_PICTURE_TYPE_B){
07dff5c7 2162 s->next_picture_ptr= &h->ref_list[1][0];
8d2fc163 2163 ff_copy_picture(&s->next_picture, s->next_picture_ptr);
07dff5c7
MN
2164 }
2165
975a1447
SS
2166 if( (h->pps.weighted_pred && h->slice_type_nos == AV_PICTURE_TYPE_P )
2167 || (h->pps.weighted_bipred_idc==1 && h->slice_type_nos== AV_PICTURE_TYPE_B ) )
0da71265 2168 pred_weight_table(h);
975a1447 2169 else if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== AV_PICTURE_TYPE_B){
1052b76f 2170 implicit_weight_table(h, -1);
1a29c6a0 2171 }else {
9f2d1b4f 2172 h->use_weight = 0;
cb99c652
GB
2173 for (i = 0; i < 2; i++) {
2174 h->luma_weight_flag[i] = 0;
2175 h->chroma_weight_flag[i] = 0;
2176 }
2177 }
115329f1 2178
2ddcf84b 2179 if(h->nal_ref_idc)
ea6f00c4 2180 ff_h264_decode_ref_pic_marking(h0, &s->gb);
0da71265 2181
1052b76f 2182 if(FRAME_MBAFF){
ea6f00c4 2183 ff_h264_fill_mbaff_ref_list(h);
5d18eaad 2184
975a1447 2185 if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== AV_PICTURE_TYPE_B){
1052b76f
MN
2186 implicit_weight_table(h, 0);
2187 implicit_weight_table(h, 1);
2188 }
2189 }
2190
975a1447 2191 if(h->slice_type_nos==AV_PICTURE_TYPE_B && !h->direct_spatial_mv_pred)
943f69a6
MN
2192 ff_h264_direct_dist_scale_factor(h);
2193 ff_h264_direct_ref_list_init(h);
8f56e219 2194
975a1447 2195 if( h->slice_type_nos != AV_PICTURE_TYPE_I && h->pps.cabac ){
9963b332 2196 tmp = get_ue_golomb_31(&s->gb);
88e7a4d1
MN
2197 if(tmp > 2){
2198 av_log(s->avctx, AV_LOG_ERROR, "cabac_init_idc overflow\n");
2199 return -1;
2200 }
2201 h->cabac_init_idc= tmp;
2202 }
e5017ab8
LA
2203
2204 h->last_qscale_diff = 0;
88e7a4d1 2205 tmp = h->pps.init_qp + get_se_golomb(&s->gb);
fcc0224e 2206 if(tmp>51+6*(h->sps.bit_depth_luma-8)){
88e7a4d1 2207 av_log(s->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
3ebc7e04
MN
2208 return -1;
2209 }
88e7a4d1 2210 s->qscale= tmp;
4691a77d
2211 h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
2212 h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
0da71265 2213 //FIXME qscale / qp ... stuff
975a1447 2214 if(h->slice_type == AV_PICTURE_TYPE_SP){
1df1df0b 2215 get_bits1(&s->gb); /* sp_for_switch_flag */
0da71265 2216 }
975a1447 2217 if(h->slice_type==AV_PICTURE_TYPE_SP || h->slice_type == AV_PICTURE_TYPE_SI){
1df1df0b 2218 get_se_golomb(&s->gb); /* slice_qs_delta */
0da71265
MN
2219 }
2220
53c05b1e 2221 h->deblocking_filter = 1;
0c32e19d
MN
2222 h->slice_alpha_c0_offset = 52;
2223 h->slice_beta_offset = 52;
0da71265 2224 if( h->pps.deblocking_filter_parameters_present ) {
9963b332 2225 tmp= get_ue_golomb_31(&s->gb);
88e7a4d1
MN
2226 if(tmp > 2){
2227 av_log(s->avctx, AV_LOG_ERROR, "deblocking_filter_idc %u out of range\n", tmp);
2228 return -1;
2229 }
2230 h->deblocking_filter= tmp;
115329f1 2231 if(h->deblocking_filter < 2)
53c05b1e
MN
2232 h->deblocking_filter^= 1; // 1<->0
2233
2234 if( h->deblocking_filter ) {
0c32e19d
MN
2235 h->slice_alpha_c0_offset += get_se_golomb(&s->gb) << 1;
2236 h->slice_beta_offset += get_se_golomb(&s->gb) << 1;
2237 if( h->slice_alpha_c0_offset > 104U
2238 || h->slice_beta_offset > 104U){
2239 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);
2240 return -1;
2241 }
0da71265 2242 }
980a82b7 2243 }
afebe2f7 2244
61858a76 2245 if( s->avctx->skip_loop_filter >= AVDISCARD_ALL
975a1447
SS
2246 ||(s->avctx->skip_loop_filter >= AVDISCARD_NONKEY && h->slice_type_nos != AV_PICTURE_TYPE_I)
2247 ||(s->avctx->skip_loop_filter >= AVDISCARD_BIDIR && h->slice_type_nos == AV_PICTURE_TYPE_B)
61858a76
RD
2248 ||(s->avctx->skip_loop_filter >= AVDISCARD_NONREF && h->nal_ref_idc == 0))
2249 h->deblocking_filter= 0;
2250
afebe2f7 2251 if(h->deblocking_filter == 1 && h0->max_contexts > 1) {
ec970c21
2252 if(s->avctx->flags2 & CODEC_FLAG2_FAST) {
2253 /* Cheat slightly for speed:
5d81d641 2254 Do not bother to deblock across slices. */
ec970c21
2255 h->deblocking_filter = 2;
2256 } else {
7ae94d52
2257 h0->max_contexts = 1;
2258 if(!h0->single_decode_warning) {
2259 av_log(s->avctx, AV_LOG_INFO, "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
2260 h0->single_decode_warning = 1;
2261 }
2262 if(h != h0)
2263 return 1; // deblocking switched inside frame
ec970c21 2264 }
afebe2f7 2265 }
0c32e19d 2266 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]);
afebe2f7 2267
0da71265
MN
2268#if 0 //FMO
2269 if( h->pps.num_slice_groups > 1 && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
2270 slice_group_change_cycle= get_bits(&s->gb, ?);
2271#endif
2272
afebe2f7
2273 h0->last_slice_type = slice_type;
2274 h->slice_num = ++h0->current_slice;
b735aeea
MN
2275 if(h->slice_num >= MAX_SLICES){
2276 av_log(s->avctx, AV_LOG_ERROR, "Too many slices, increase MAX_SLICES and recompile\n");
2277 }
5175b937 2278
c32867b5 2279 for(j=0; j<2; j++){
6d7e6b26 2280 int id_list[16];
b735aeea 2281 int *ref2frm= h->ref2frm[h->slice_num&(MAX_SLICES-1)][j];
6d7e6b26
MN
2282 for(i=0; i<16; i++){
2283 id_list[i]= 60;
2284 if(h->ref_list[j][i].data[0]){
2285 int k;
2286 uint8_t *base= h->ref_list[j][i].base[0];
2287 for(k=0; k<h->short_ref_count; k++)
2288 if(h->short_ref[k]->base[0] == base){
2289 id_list[i]= k;
2290 break;
2291 }
2292 for(k=0; k<h->long_ref_count; k++)
2293 if(h->long_ref[k] && h->long_ref[k]->base[0] == base){
2294 id_list[i]= h->short_ref_count + k;
2295 break;
2296 }
2297 }
2298 }
2299
c32867b5
MN
2300 ref2frm[0]=
2301 ref2frm[1]= -1;
d50cdd82 2302 for(i=0; i<16; i++)
6d7e6b26 2303 ref2frm[i+2]= 4*id_list[i]
c32867b5 2304 +(h->ref_list[j][i].reference&3);
d50cdd82
MN
2305 ref2frm[18+0]=
2306 ref2frm[18+1]= -1;
2307 for(i=16; i<48; i++)
6d7e6b26 2308 ref2frm[i+4]= 4*id_list[(i-16)>>1]
d50cdd82 2309 +(h->ref_list[j][i].reference&3);
c32867b5
MN
2310 }
2311
5d18eaad 2312 h->emu_edge_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16;
8a11a969 2313 h->emu_edge_height= (FRAME_MBAFF || FIELD_PICTURE) ? 0 : h->emu_edge_width;
5d18eaad 2314
0da71265 2315 if(s->avctx->debug&FF_DEBUG_PICT_INFO){
49573a87 2316 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",
6867a90b
LLL
2317 h->slice_num,
2318 (s->picture_structure==PICT_FRAME ? "F" : s->picture_structure==PICT_TOP_FIELD ? "T" : "B"),
115329f1 2319 first_mb_in_slice,
6209669d 2320 av_get_picture_type_char(h->slice_type), h->slice_type_fixed ? " fix" : "", h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
0da71265
MN
2321 pps_id, h->frame_num,
2322 s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
2323 h->ref_count[0], h->ref_count[1],
2324 s->qscale,
0c32e19d 2325 h->deblocking_filter, h->slice_alpha_c0_offset/2-26, h->slice_beta_offset/2-26,
9f2d1b4f 2326 h->use_weight,
4806b922 2327 h->use_weight==1 && h->use_weight_chroma ? "c" : "",
975a1447 2328 h->slice_type == AV_PICTURE_TYPE_B ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : ""
0da71265
MN
2329 );
2330 }
2331
2332 return 0;
2333}
2334
0dc343d4 2335int ff_h264_get_slice_type(const H264Context *h)
75dd6938
LA
2336{
2337 switch (h->slice_type) {
975a1447
SS
2338 case AV_PICTURE_TYPE_P: return 0;
2339 case AV_PICTURE_TYPE_B: return 1;
2340 case AV_PICTURE_TYPE_I: return 2;
2341 case AV_PICTURE_TYPE_SP: return 3;
2342 case AV_PICTURE_TYPE_SI: return 4;
75dd6938
LA
2343 default: return -1;
2344 }
2345}
2346
d02bb3ec
DB
2347/**
2348 *
2349 * @return non zero if the loop filter can be skiped
2350 */
2351static int fill_filter_caches(H264Context *h, int mb_type){
2352 MpegEncContext * const s = &h->s;
2353 const int mb_xy= h->mb_xy;
2354 int top_xy, left_xy[2];
2355 int top_type, left_type[2];
2356
2357 top_xy = mb_xy - (s->mb_stride << MB_FIELD);
2358
2359 //FIXME deblocking could skip the intra and nnz parts.
2360
2361 /* Wow, what a mess, why didn't they simplify the interlacing & intra
2362 * stuff, I can't imagine that these complex rules are worth it. */
2363
2364 left_xy[1] = left_xy[0] = mb_xy-1;
2365 if(FRAME_MBAFF){
2366 const int left_mb_field_flag = IS_INTERLACED(s->current_picture.mb_type[mb_xy-1]);
2367 const int curr_mb_field_flag = IS_INTERLACED(mb_type);
2368 if(s->mb_y&1){
2369 if (left_mb_field_flag != curr_mb_field_flag) {
2370 left_xy[0] -= s->mb_stride;
2371 }
2372 }else{
2373 if(curr_mb_field_flag){
2374 top_xy += s->mb_stride & (((s->current_picture.mb_type[top_xy ]>>7)&1)-1);
2375 }
2376 if (left_mb_field_flag != curr_mb_field_flag) {
2377 left_xy[1] += s->mb_stride;
2378 }
2379 }
2380 }
2381
2382 h->top_mb_xy = top_xy;
2383 h->left_mb_xy[0] = left_xy[0];
2384 h->left_mb_xy[1] = left_xy[1];
2385 {
2386 //for sufficiently low qp, filtering wouldn't do anything
2387 //this is a conservative estimate: could also check beta_offset and more accurate chroma_qp
2388 int qp_thresh = h->qp_thresh; //FIXME strictly we should store qp_thresh for each mb of a slice
2389 int qp = s->current_picture.qscale_table[mb_xy];
2390 if(qp <= qp_thresh
2391 && (left_xy[0]<0 || ((qp + s->current_picture.qscale_table[left_xy[0]] + 1)>>1) <= qp_thresh)
2392 && (top_xy < 0 || ((qp + s->current_picture.qscale_table[top_xy ] + 1)>>1) <= qp_thresh)){
2393 if(!FRAME_MBAFF)
2394 return 1;
2395 if( (left_xy[0]< 0 || ((qp + s->current_picture.qscale_table[left_xy[1] ] + 1)>>1) <= qp_thresh)
2396 && (top_xy < s->mb_stride || ((qp + s->current_picture.qscale_table[top_xy -s->mb_stride] + 1)>>1) <= qp_thresh))
2397 return 1;
2398 }
2399 }
2400
2401 top_type = s->current_picture.mb_type[top_xy] ;
2402 left_type[0] = s->current_picture.mb_type[left_xy[0]];
2403 left_type[1] = s->current_picture.mb_type[left_xy[1]];
2404 if(h->deblocking_filter == 2){
2405 if(h->slice_table[top_xy ] != h->slice_num) top_type= 0;
2406 if(h->slice_table[left_xy[0] ] != h->slice_num) left_type[0]= left_type[1]= 0;
2407 }else{
2408 if(h->slice_table[top_xy ] == 0xFFFF) top_type= 0;
2409 if(h->slice_table[left_xy[0] ] == 0xFFFF) left_type[0]= left_type[1] =0;
2410 }
2411 h->top_type = top_type ;
2412 h->left_type[0]= left_type[0];
2413 h->left_type[1]= left_type[1];
2414
2415 if(IS_INTRA(mb_type))
2416 return 0;
2417
2418 AV_COPY64(&h->non_zero_count_cache[0+8*1], &h->non_zero_count[mb_xy][ 0]);
2419 AV_COPY64(&h->non_zero_count_cache[0+8*2], &h->non_zero_count[mb_xy][ 8]);
2420 AV_COPY32(&h->non_zero_count_cache[0+8*5], &h->non_zero_count[mb_xy][16]);
2421 AV_COPY32(&h->non_zero_count_cache[4+8*3], &h->non_zero_count[mb_xy][20]);
2422 AV_COPY64(&h->non_zero_count_cache[0+8*4], &h->non_zero_count[mb_xy][24]);
2423
2424 h->cbp= h->cbp_table[mb_xy];
2425
2426 {
2427 int list;
2428 for(list=0; list<h->list_count; list++){
2429 int8_t *ref;
2430 int y, b_stride;
2431 int16_t (*mv_dst)[2];
2432 int16_t (*mv_src)[2];
2433
2434 if(!USES_LIST(mb_type, list)){
2435 fill_rectangle( h->mv_cache[list][scan8[0]], 4, 4, 8, pack16to32(0,0), 4);
2436 AV_WN32A(&h->ref_cache[list][scan8[ 0]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
2437 AV_WN32A(&h->ref_cache[list][scan8[ 2]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
2438 AV_WN32A(&h->ref_cache[list][scan8[ 8]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
2439 AV_WN32A(&h->ref_cache[list][scan8[10]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
2440 continue;
2441 }
2442
2443 ref = &s->current_picture.ref_index[list][4*mb_xy];
2444 {
2445 int (*ref2frm)[64] = h->ref2frm[ h->slice_num&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
2446 AV_WN32A(&h->ref_cache[list][scan8[ 0]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
2447 AV_WN32A(&h->ref_cache[list][scan8[ 2]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
2448 ref += 2;
2449 AV_WN32A(&h->ref_cache[list][scan8[ 8]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
2450 AV_WN32A(&h->ref_cache[list][scan8[10]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
2451 }
2452
2453 b_stride = h->b_stride;
2454 mv_dst = &h->mv_cache[list][scan8[0]];
2455 mv_src = &s->current_picture.motion_val[list][4*s->mb_x + 4*s->mb_y*b_stride];
2456 for(y=0; y<4; y++){
2457 AV_COPY128(mv_dst + 8*y, mv_src + y*b_stride);
2458 }
2459
2460 }
2461 }
2462
2463
2464/*
24650 . T T. T T T T
24661 L . .L . . . .
24672 L . .L . . . .
24683 . T TL . . . .
24694 L . .L . . . .
24705 L . .. . . . .
2471*/
2472//FIXME constraint_intra_pred & partitioning & nnz (let us hope this is just a typo in the spec)
2473 if(top_type){
2474 AV_COPY32(&h->non_zero_count_cache[4+8*0], &h->non_zero_count[top_xy][4+3*8]);
2475 }
2476
2477 if(left_type[0]){
2478 h->non_zero_count_cache[3+8*1]= h->non_zero_count[left_xy[0]][7+0*8];
2479 h->non_zero_count_cache[3+8*2]= h->non_zero_count[left_xy[0]][7+1*8];
2480 h->non_zero_count_cache[3+8*3]= h->non_zero_count[left_xy[0]][7+2*8];
2481 h->non_zero_count_cache[3+8*4]= h->non_zero_count[left_xy[0]][7+3*8];
2482 }
2483
2484 // CAVLC 8x8dct requires NNZ values for residual decoding that differ from what the loop filter needs
2485 if(!CABAC && h->pps.transform_8x8_mode){
2486 if(IS_8x8DCT(top_type)){
2487 h->non_zero_count_cache[4+8*0]=
2488 h->non_zero_count_cache[5+8*0]= h->cbp_table[top_xy] & 4;
2489 h->non_zero_count_cache[6+8*0]=
2490 h->non_zero_count_cache[7+8*0]= h->cbp_table[top_xy] & 8;
2491 }
2492 if(IS_8x8DCT(left_type[0])){
2493 h->non_zero_count_cache[3+8*1]=
2494 h->non_zero_count_cache[3+8*2]= h->cbp_table[left_xy[0]]&2; //FIXME check MBAFF
2495 }
2496 if(IS_8x8DCT(left_type[1])){
2497 h->non_zero_count_cache[3+8*3]=
2498 h->non_zero_count_cache[3+8*4]= h->cbp_table[left_xy[1]]&8; //FIXME check MBAFF
2499 }
2500
2501 if(IS_8x8DCT(mb_type)){
2502 h->non_zero_count_cache[scan8[0 ]]= h->non_zero_count_cache[scan8[1 ]]=
2503 h->non_zero_count_cache[scan8[2 ]]= h->non_zero_count_cache[scan8[3 ]]= h->cbp & 1;
2504
2505 h->non_zero_count_cache[scan8[0+ 4]]= h->non_zero_count_cache[scan8[1+ 4]]=
2506 h->non_zero_count_cache[scan8[2+ 4]]= h->non_zero_count_cache[scan8[3+ 4]]= h->cbp & 2;
2507
2508 h->non_zero_count_cache[scan8[0+ 8]]= h->non_zero_count_cache[scan8[1+ 8]]=
2509 h->non_zero_count_cache[scan8[2+ 8]]= h->non_zero_count_cache[scan8[3+ 8]]= h->cbp & 4;
2510
2511 h->non_zero_count_cache[scan8[0+12]]= h->non_zero_count_cache[scan8[1+12]]=
2512 h->non_zero_count_cache[scan8[2+12]]= h->non_zero_count_cache[scan8[3+12]]= h->cbp & 8;
2513 }
2514 }
2515
2516 if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
2517 int list;
2518 for(list=0; list<h->list_count; list++){
2519 if(USES_LIST(top_type, list)){
2520 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
2521 const int b8_xy= 4*top_xy + 2;
2522 int (*ref2frm)[64] = h->ref2frm[ h->slice_table[top_xy]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
2523 AV_COPY128(h->mv_cache[list][scan8[0] + 0 - 1*8], s->current_picture.motion_val[list][b_xy + 0]);
2524 h->ref_cache[list][scan8[0] + 0 - 1*8]=
2525 h->ref_cache[list][scan8[0] + 1 - 1*8]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 0]];
2526 h->ref_cache[list][scan8[0] + 2 - 1*8]=
2527 h->ref_cache[list][scan8[0] + 3 - 1*8]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 1]];
2528 }else{
2529 AV_ZERO128(h->mv_cache[list][scan8[0] + 0 - 1*8]);
2530 AV_WN32A(&h->ref_cache[list][scan8[0] + 0 - 1*8], ((LIST_NOT_USED)&0xFF)*0x01010101u);
2531 }
2532
2533 if(!IS_INTERLACED(mb_type^left_type[0])){
2534 if(USES_LIST(left_type[0], list)){
2535 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
2536 const int b8_xy= 4*left_xy[0] + 1;
2537 int (*ref2frm)[64] = h->ref2frm[ h->slice_table[left_xy[0]]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
2538 AV_COPY32(h->mv_cache[list][scan8[0] - 1 + 0 ], s->current_picture.motion_val[list][b_xy + h->b_stride*0]);
2539 AV_COPY32(h->mv_cache[list][scan8[0] - 1 + 8 ], s->current_picture.motion_val[list][b_xy + h->b_stride*1]);
2540 AV_COPY32(h->mv_cache[list][scan8[0] - 1 +16 ], s->current_picture.motion_val[list][b_xy + h->b_stride*2]);
2541 AV_COPY32(h->mv_cache[list][scan8[0] - 1 +24 ], s->current_picture.motion_val[list][b_xy + h->b_stride*3]);
2542 h->ref_cache[list][scan8[0] - 1 + 0 ]=
2543 h->ref_cache[list][scan8[0] - 1 + 8 ]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 2*0]];
2544 h->ref_cache[list][scan8[0] - 1 +16 ]=
2545 h->ref_cache[list][scan8[0] - 1 +24 ]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 2*1]];
2546 }else{
2547 AV_ZERO32(h->mv_cache [list][scan8[0] - 1 + 0 ]);
2548 AV_ZERO32(h->mv_cache [list][scan8[0] - 1 + 8 ]);
2549 AV_ZERO32(h->mv_cache [list][scan8[0] - 1 +16 ]);
2550 AV_ZERO32(h->mv_cache [list][scan8[0] - 1 +24 ]);
2551 h->ref_cache[list][scan8[0] - 1 + 0 ]=
2552 h->ref_cache[list][scan8[0] - 1 + 8 ]=
2553 h->ref_cache[list][scan8[0] - 1 + 16 ]=
2554 h->ref_cache[list][scan8[0] - 1 + 24 ]= LIST_NOT_USED;
2555 }
2556 }
2557 }
2558 }
2559
2560 return 0;
2561}
2562
0ffc8415 2563static void loop_filter(H264Context *h, int start_x, int end_x){
c988f975
MN
2564 MpegEncContext * const s = &h->s;
2565 uint8_t *dest_y, *dest_cb, *dest_cr;
2566 int linesize, uvlinesize, mb_x, mb_y;
2567 const int end_mb_y= s->mb_y + FRAME_MBAFF;
2568 const int old_slice_type= h->slice_type;
6e3ef511 2569 const int pixel_shift = h->pixel_shift;
c988f975
MN
2570
2571 if(h->deblocking_filter) {
0ffc8415 2572 for(mb_x= start_x; mb_x<end_x; mb_x++){
c988f975 2573 for(mb_y=end_mb_y - FRAME_MBAFF; mb_y<= end_mb_y; mb_y++){
78998bf2 2574 int mb_xy, mb_type;
c988f975
MN
2575 mb_xy = h->mb_xy = mb_x + mb_y*s->mb_stride;
2576 h->slice_num= h->slice_table[mb_xy];
2577 mb_type= s->current_picture.mb_type[mb_xy];
2578 h->list_count= h->list_counts[mb_xy];
c988f975
MN
2579
2580 if(FRAME_MBAFF)
2581 h->mb_mbaff = h->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
2582
c988f975
MN
2583 s->mb_x= mb_x;
2584 s->mb_y= mb_y;
6e3ef511
OA
2585 dest_y = s->current_picture.data[0] + ((mb_x << pixel_shift) + mb_y * s->linesize ) * 16;
2586 dest_cb = s->current_picture.data[1] + ((mb_x << pixel_shift) + mb_y * s->uvlinesize) * 8;
2587 dest_cr = s->current_picture.data[2] + ((mb_x << pixel_shift) + mb_y * s->uvlinesize) * 8;
c988f975
MN
2588 //FIXME simplify above
2589
2590 if (MB_FIELD) {
2591 linesize = h->mb_linesize = s->linesize * 2;
2592 uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
2593 if(mb_y&1){ //FIXME move out of this function?
2594 dest_y -= s->linesize*15;
2595 dest_cb-= s->uvlinesize*7;
2596 dest_cr-= s->uvlinesize*7;
2597 }
2598 } else {
2599 linesize = h->mb_linesize = s->linesize;
2600 uvlinesize = h->mb_uvlinesize = s->uvlinesize;
2601 }
77d40dce 2602 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);
aaa995d7 2603 if(fill_filter_caches(h, mb_type))
44a5e7b6 2604 continue;
c988f975
MN
2605 h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.qscale_table[mb_xy]);
2606 h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.qscale_table[mb_xy]);
2607
77d40dce 2608 if (FRAME_MBAFF) {
c988f975
MN
2609 ff_h264_filter_mb (h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2610 } else {
2611 ff_h264_filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2612 }
2613 }
2614 }
2615 }
2616 h->slice_type= old_slice_type;
4e987f82 2617 s->mb_x= end_x;
c988f975 2618 s->mb_y= end_mb_y - FRAME_MBAFF;
f4b8b825
MN
2619 h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
2620 h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
c988f975
MN
2621}
2622
69a28f3e
MN
2623static void predict_field_decoding_flag(H264Context *h){
2624 MpegEncContext * const s = &h->s;
2625 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
2626 int mb_type = (h->slice_table[mb_xy-1] == h->slice_num)
2627 ? s->current_picture.mb_type[mb_xy-1]
2628 : (h->slice_table[mb_xy-s->mb_stride] == h->slice_num)
2629 ? s->current_picture.mb_type[mb_xy-s->mb_stride]
2630 : 0;
2631 h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
2632}
2633
3a84713a
RS
2634static int decode_slice(struct AVCodecContext *avctx, void *arg){
2635 H264Context *h = *(void**)arg;
0da71265
MN
2636 MpegEncContext * const s = &h->s;
2637 const int part_mask= s->partitioned_frame ? (AC_END|AC_ERROR) : 0x7F;
0ffc8415 2638 int lf_x_start = s->mb_x;
0da71265
MN
2639
2640 s->mb_skip_run= -1;
0da71265 2641
89db0bae 2642 h->is_complex = FRAME_MBAFF || s->picture_structure != PICT_FRAME || s->codec_id != CODEC_ID_H264 ||
5317c95b 2643 (CONFIG_GRAY && (s->flags&CODEC_FLAG_GRAY));
89db0bae 2644
e5017ab8 2645 if( h->pps.cabac ) {
e5017ab8
LA
2646 /* realign */
2647 align_get_bits( &s->gb );
2648
2649 /* init cabac */
d61c4e73 2650 ff_init_cabac_states( &h->cabac);
e5017ab8
LA
2651 ff_init_cabac_decoder( &h->cabac,
2652 s->gb.buffer + get_bits_count(&s->gb)/8,
6e44ba15 2653 (get_bits_left(&s->gb) + 7)/8);
cc51b282
MN
2654
2655 ff_h264_init_cabac_states(h);
95c26348 2656
e5017ab8 2657 for(;;){
851ded89 2658//START_TIMER
cc51b282 2659 int ret = ff_h264_decode_mb_cabac(h);
6867a90b 2660 int eos;
851ded89 2661//STOP_TIMER("decode_mb_cabac")
0da71265 2662
903d58f6 2663 if(ret>=0) ff_h264_hl_decode_mb(h);
0da71265 2664
5d18eaad 2665 if( ret >= 0 && FRAME_MBAFF ) { //FIXME optimal? or let mb_decode decode 16x32 ?
e5017ab8
LA
2666 s->mb_y++;
2667
cc51b282 2668 ret = ff_h264_decode_mb_cabac(h);
e5017ab8 2669
903d58f6 2670 if(ret>=0) ff_h264_hl_decode_mb(h);
e5017ab8
LA
2671 s->mb_y--;
2672 }
6867a90b 2673 eos = get_cabac_terminate( &h->cabac );
e5017ab8 2674
3566042a
MN
2675 if((s->workaround_bugs & FF_BUG_TRUNCATED) && h->cabac.bytestream > h->cabac.bytestream_end + 2){
2676 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);
0ffc8415 2677 if (s->mb_x >= lf_x_start) loop_filter(h, lf_x_start, s->mb_x + 1);
3566042a
MN
2678 return 0;
2679 }
5659b509 2680 if( ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 2) {
706da4af 2681 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);
e5017ab8
LA
2682 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);
2683 return -1;
2684 }
2685
2686 if( ++s->mb_x >= s->mb_width ) {
0ffc8415
RB
2687 loop_filter(h, lf_x_start, s->mb_x);
2688 s->mb_x = lf_x_start = 0;
e5017ab8 2689 ff_draw_horiz_band(s, 16*s->mb_y, 16);
5175b937 2690 ++s->mb_y;
f3e53d9f 2691 if(FIELD_OR_MBAFF_PICTURE) {
6867a90b 2692 ++s->mb_y;
69cc3183
MN
2693 if(FRAME_MBAFF && s->mb_y < s->mb_height)
2694 predict_field_decoding_flag(h);
6867a90b 2695 }
0da71265 2696 }
0da71265 2697
e5017ab8 2698 if( eos || s->mb_y >= s->mb_height ) {
a9c9a240 2699 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
e5017ab8 2700 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);
0ffc8415 2701 if (s->mb_x > lf_x_start) loop_filter(h, lf_x_start, s->mb_x);
0da71265 2702 return 0;
e5017ab8 2703 }
e5017ab8
LA
2704 }
2705
2706 } else {
2707 for(;;){
e1e94902 2708 int ret = ff_h264_decode_mb_cavlc(h);
e5017ab8 2709
903d58f6 2710 if(ret>=0) ff_h264_hl_decode_mb(h);
e5017ab8 2711
5d18eaad 2712 if(ret>=0 && FRAME_MBAFF){ //FIXME optimal? or let mb_decode decode 16x32 ?
e5017ab8 2713 s->mb_y++;
e1e94902 2714 ret = ff_h264_decode_mb_cavlc(h);
e5017ab8 2715
903d58f6 2716 if(ret>=0) ff_h264_hl_decode_mb(h);
e5017ab8
LA
2717 s->mb_y--;
2718 }
2719
2720 if(ret<0){
2721 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
0da71265 2722 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);
0da71265
MN
2723 return -1;
2724 }
e5017ab8
LA
2725
2726 if(++s->mb_x >= s->mb_width){
0ffc8415
RB
2727 loop_filter(h, lf_x_start, s->mb_x);
2728 s->mb_x = lf_x_start = 0;
e5017ab8 2729 ff_draw_horiz_band(s, 16*s->mb_y, 16);
6867a90b 2730 ++s->mb_y;
f3e53d9f 2731 if(FIELD_OR_MBAFF_PICTURE) {
6867a90b 2732 ++s->mb_y;
69cc3183
MN
2733 if(FRAME_MBAFF && s->mb_y < s->mb_height)
2734 predict_field_decoding_flag(h);
6867a90b
LLL
2735 }
2736 if(s->mb_y >= s->mb_height){
a9c9a240 2737 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
e5017ab8
LA
2738
2739 if(get_bits_count(&s->gb) == s->gb.size_in_bits ) {
2740 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);
2741
2742 return 0;
2743 }else{
2744 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);
2745
2746 return -1;
2747 }
2748 }
2749 }
2750
2751 if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->mb_skip_run<=0){
a9c9a240 2752 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
e5017ab8
LA
2753 if(get_bits_count(&s->gb) == s->gb.size_in_bits ){
2754 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);
0ffc8415 2755 if (s->mb_x > lf_x_start) loop_filter(h, lf_x_start, s->mb_x);
e5017ab8
LA
2756
2757 return 0;
2758 }else{
2759 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);
2760
2761 return -1;
2762 }
2763 }
0da71265
MN
2764 }
2765 }
e5017ab8 2766
0da71265
MN
2767#if 0
2768 for(;s->mb_y < s->mb_height; s->mb_y++){
2769 for(;s->mb_x < s->mb_width; s->mb_x++){
2770 int ret= decode_mb(h);
115329f1 2771
903d58f6 2772 ff_h264_hl_decode_mb(h);
0da71265
MN
2773
2774 if(ret<0){
267f7edc 2775 av_log(s->avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
0da71265
MN
2776 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);
2777
2778 return -1;
2779 }
115329f1 2780
0da71265
MN
2781 if(++s->mb_x >= s->mb_width){
2782 s->mb_x=0;
2783 if(++s->mb_y >= s->mb_height){
2784 if(get_bits_count(s->gb) == s->gb.size_in_bits){
2785 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);
2786
2787 return 0;
2788 }else{
2789 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);
2790
2791 return -1;
2792 }
2793 }
2794 }
115329f1 2795
0da71265
MN
2796 if(get_bits_count(s->?gb) >= s->gb?.size_in_bits){
2797 if(get_bits_count(s->gb) == s->gb.size_in_bits){
2798 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);
2799
2800 return 0;
2801 }else{
2802 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);
2803
2804 return -1;
2805 }
2806 }
2807 }
2808 s->mb_x=0;
2809 ff_draw_horiz_band(s, 16*s->mb_y, 16);
2810 }
2811#endif
2812 return -1; //not reached
2813}
2814
afebe2f7
2815/**
2816 * Call decode_slice() for each context.
2817 *
2818 * @param h h264 master context
2819 * @param context_count number of contexts to execute
2820 */
2821static void execute_decode_slices(H264Context *h, int context_count){
2822 MpegEncContext * const s = &h->s;
2823 AVCodecContext * const avctx= s->avctx;
2824 H264Context *hx;
2825 int i;
2826
40e5d31b
GB
2827 if (s->avctx->hwaccel)
2828 return;
0d3d172f 2829 if(s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
369122dd 2830 return;
afebe2f7 2831 if(context_count == 1) {
74e8b78b 2832 decode_slice(avctx, &h);
afebe2f7
2833 } else {
2834 for(i = 1; i < context_count; i++) {
2835 hx = h->thread_context[i];
047599a4 2836 hx->s.error_recognition = avctx->error_recognition;
afebe2f7
2837 hx->s.error_count = 0;
2838 }
2839
2840 avctx->execute(avctx, (void *)decode_slice,
01418506 2841 h->thread_context, NULL, context_count, sizeof(void*));
afebe2f7
2842
2843 /* pull back stuff from slices to master context */
2844 hx = h->thread_context[context_count - 1];
2845 s->mb_x = hx->s.mb_x;
2846 s->mb_y = hx->s.mb_y;
12d96de3
JD
2847 s->dropable = hx->s.dropable;
2848 s->picture_structure = hx->s.picture_structure;
afebe2f7
2849 for(i = 1; i < context_count; i++)
2850 h->s.error_count += h->thread_context[i]->s.error_count;
2851 }
2852}
2853
2854
30317501 2855static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size){
0da71265
MN
2856 MpegEncContext * const s = &h->s;
2857 AVCodecContext * const avctx= s->avctx;
2858 int buf_index=0;
afebe2f7
2859 H264Context *hx; ///< thread context
2860 int context_count = 0;
74b14aac 2861 int next_avc= h->is_avc ? 0 : buf_size;
afebe2f7
2862
2863 h->max_contexts = avctx->thread_count;
66a4b2c1 2864 if(!(s->flags2 & CODEC_FLAG2_CHUNKS)){
afebe2f7 2865 h->current_slice = 0;
12d96de3 2866 if (!s->first_field)
f6e3c460 2867 s->current_picture_ptr= NULL;
9c095463 2868 ff_h264_reset_sei(h);
66a4b2c1
MN
2869 }
2870
0da71265
MN
2871 for(;;){
2872 int consumed;
2873 int dst_length;
2874 int bit_length;
30317501 2875 const uint8_t *ptr;
4770b1b4 2876 int i, nalsize = 0;
afebe2f7 2877 int err;
115329f1 2878
74b14aac 2879 if(buf_index >= next_avc) {
1c48415b
2880 if(buf_index >= buf_size) break;
2881 nalsize = 0;
2882 for(i = 0; i < h->nal_length_size; i++)
2883 nalsize = (nalsize << 8) | buf[buf_index++];
9d252137
BC
2884 if(nalsize <= 0 || nalsize > buf_size - buf_index){
2885 av_log(h->s.avctx, AV_LOG_ERROR, "AVC: nal size %d\n", nalsize);
2886 break;
1c48415b 2887 }
74b14aac 2888 next_avc= buf_index + nalsize;
1c48415b
2889 } else {
2890 // start code prefix search
52255d17 2891 for(; buf_index + 3 < next_avc; buf_index++){
1c48415b
2892 // This should always succeed in the first iteration.
2893 if(buf[buf_index] == 0 && buf[buf_index+1] == 0 && buf[buf_index+2] == 1)
2894 break;
8b031359 2895 }
115329f1 2896
1c48415b 2897 if(buf_index+3 >= buf_size) break;
115329f1 2898
1c48415b 2899 buf_index+=3;
52255d17 2900 if(buf_index >= next_avc) continue;
1c48415b 2901 }
115329f1 2902
afebe2f7
2903 hx = h->thread_context[context_count];
2904
74b14aac 2905 ptr= ff_h264_decode_nal(hx, buf + buf_index, &dst_length, &consumed, next_avc - buf_index);
ff82e429 2906 if (ptr==NULL || dst_length < 0){
ac658be5
FOL
2907 return -1;
2908 }
3566042a
MN
2909 i= buf_index + consumed;
2910 if((s->workaround_bugs & FF_BUG_AUTODETECT) && i+3<next_avc &&
2911 buf[i]==0x00 && buf[i+1]==0x00 && buf[i+2]==0x01 && buf[i+3]==0xE0)
2912 s->workaround_bugs |= FF_BUG_TRUNCATED;
2913
2914 if(!(s->workaround_bugs & FF_BUG_TRUNCATED)){
6ac9696e 2915 while(ptr[dst_length - 1] == 0 && dst_length > 0)
c4da83fb 2916 dst_length--;
3566042a 2917 }
1790a5e9 2918 bit_length= !dst_length ? 0 : (8*dst_length - ff_h264_decode_rbsp_trailing(h, ptr + dst_length - 1));
0da71265
MN
2919
2920 if(s->avctx->debug&FF_DEBUG_STARTCODE){
afebe2f7 2921 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);
0da71265 2922 }
115329f1 2923
74b14aac 2924 if (h->is_avc && (nalsize != consumed) && nalsize){
e262365d 2925 av_log(h->s.avctx, AV_LOG_DEBUG, "AVC: Consumed only %d bytes instead of %d\n", consumed, nalsize);
9d2cc8c1 2926 }
4770b1b4 2927
0da71265
MN
2928 buf_index += consumed;
2929
8ed2ae09 2930 //FIXME do not discard SEI id
e7021c0e 2931 if(avctx->skip_frame >= AVDISCARD_NONREF && h->nal_ref_idc == 0)
0da71265 2932 continue;
115329f1 2933
afebe2f7
2934 again:
2935 err = 0;
2936 switch(hx->nal_unit_type){
0da71265 2937 case NAL_IDR_SLICE:
afebe2f7
2938 if (h->nal_unit_type != NAL_IDR_SLICE) {
2939 av_log(h->s.avctx, AV_LOG_ERROR, "Invalid mix of idr and non-idr slices");
2940 return -1;
2941 }
3b66c4c5 2942 idr(h); //FIXME ensure we don't loose some frames if there is reordering
0da71265 2943 case NAL_SLICE:
afebe2f7
2944 init_get_bits(&hx->s.gb, ptr, bit_length);
2945 hx->intra_gb_ptr=
2946 hx->inter_gb_ptr= &hx->s.gb;
2947 hx->s.data_partitioning = 0;
2948
2949 if((err = decode_slice_header(hx, h)))
2950 break;
2951
3bccd93a
SW
2952 if (h->current_slice == 1) {
2953 if (s->avctx->hwaccel && s->avctx->hwaccel->start_frame(s->avctx, NULL, 0) < 0)
6026a096 2954 return -1;
3bccd93a
SW
2955 if(CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
2956 ff_vdpau_h264_picture_start(s);
6026a096
GB
2957 }
2958
37a558fe
IS
2959 s->current_picture_ptr->key_frame |=
2960 (hx->nal_unit_type == NAL_IDR_SLICE) ||
2961 (h->sei_recovery_frame_cnt >= 0);
8ed2ae09 2962 if(hx->redundant_pic_count==0
afebe2f7 2963 && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
975a1447
SS
2964 && (avctx->skip_frame < AVDISCARD_BIDIR || hx->slice_type_nos!=AV_PICTURE_TYPE_B)
2965 && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==AV_PICTURE_TYPE_I)
369122dd 2966 && avctx->skip_frame < AVDISCARD_ALL){
d404b3ed
MN
2967 if(avctx->hwaccel) {
2968 if (avctx->hwaccel->decode_slice(avctx, &buf[buf_index - consumed], consumed) < 0)
2969 return -1;
2970 }else
0d3d172f 2971 if(CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU){
369122dd 2972 static const uint8_t start_code[] = {0x00, 0x00, 0x01};
c639fc72
CEH
2973 ff_vdpau_add_data_chunk(s, start_code, sizeof(start_code));
2974 ff_vdpau_add_data_chunk(s, &buf[buf_index - consumed], consumed );
369122dd 2975 }else
f2c214a1 2976 context_count++;
369122dd 2977 }
0da71265
MN
2978 break;
2979 case NAL_DPA:
afebe2f7
2980 init_get_bits(&hx->s.gb, ptr, bit_length);
2981 hx->intra_gb_ptr=
2982 hx->inter_gb_ptr= NULL;
0410ee8f
AS
2983
2984 if ((err = decode_slice_header(hx, h)) < 0)
2985 break;
2986
afebe2f7 2987 hx->s.data_partitioning = 1;
115329f1 2988
0da71265
MN
2989 break;
2990 case NAL_DPB:
afebe2f7
2991 init_get_bits(&hx->intra_gb, ptr, bit_length);
2992 hx->intra_gb_ptr= &hx->intra_gb;
0da71265
MN
2993 break;
2994 case NAL_DPC:
afebe2f7
2995 init_get_bits(&hx->inter_gb, ptr, bit_length);
2996 hx->inter_gb_ptr= &hx->inter_gb;
8b92b792 2997
afebe2f7 2998 if(hx->redundant_pic_count==0 && hx->intra_gb_ptr && hx->s.data_partitioning
b18e5c03 2999 && s->context_initialized
afebe2f7 3000 && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
975a1447
SS
3001 && (avctx->skip_frame < AVDISCARD_BIDIR || hx->slice_type_nos!=AV_PICTURE_TYPE_B)
3002 && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==AV_PICTURE_TYPE_I)
e0111b32 3003 && avctx->skip_frame < AVDISCARD_ALL)
afebe2f7 3004 context_count++;
0da71265
MN
3005 break;
3006 case NAL_SEI:
cdd10689 3007 init_get_bits(&s->gb, ptr, bit_length);
1790a5e9 3008 ff_h264_decode_sei(h);
0da71265
MN
3009 break;
3010 case NAL_SPS:
3011 init_get_bits(&s->gb, ptr, bit_length);
1790a5e9 3012 ff_h264_decode_seq_parameter_set(h);
115329f1 3013
0da71265
MN
3014 if(s->flags& CODEC_FLAG_LOW_DELAY)
3015 s->low_delay=1;
115329f1 3016
a18030bb
LM
3017 if(avctx->has_b_frames < 2)
3018 avctx->has_b_frames= !s->low_delay;
19a0729b
OA
3019
3020 if (avctx->bits_per_raw_sample != h->sps.bit_depth_luma) {
3021 if (h->sps.bit_depth_luma >= 8 && h->sps.bit_depth_luma <= 10) {
3022 avctx->bits_per_raw_sample = h->sps.bit_depth_luma;
3023 h->pixel_shift = h->sps.bit_depth_luma > 8;
3024
3025 ff_h264dsp_init(&h->h264dsp, h->sps.bit_depth_luma);
3026 ff_h264_pred_init(&h->hpc, s->codec_id, h->sps.bit_depth_luma);
3027 dsputil_init(&s->dsp, s->avctx);
3028 } else {
3029 av_log(avctx, AV_LOG_DEBUG, "Unsupported bit depth: %d\n", h->sps.bit_depth_luma);
3030 return -1;
3031 }
3032 }
0da71265
MN
3033 break;
3034 case NAL_PPS:
3035 init_get_bits(&s->gb, ptr, bit_length);
115329f1 3036
1790a5e9 3037 ff_h264_decode_picture_parameter_set(h, bit_length);
0da71265
MN
3038
3039 break;
ab470fa7
LM
3040 case NAL_AUD:
3041 case NAL_END_SEQUENCE:
3042 case NAL_END_STREAM:
3043 case NAL_FILLER_DATA:
3044 case NAL_SPS_EXT:
3045 case NAL_AUXILIARY_SLICE:
0da71265 3046 break;
bb270c08 3047 default:
4ad04da2 3048 av_log(avctx, AV_LOG_DEBUG, "Unknown NAL code: %d (%d bits)\n", hx->nal_unit_type, bit_length);
115329f1 3049 }
115329f1 3050
afebe2f7
3051 if(context_count == h->max_contexts) {
3052 execute_decode_slices(h, context_count);
3053 context_count = 0;
3054 }
3055
3056 if (err < 0)
3057 av_log(h->s.avctx, AV_LOG_ERROR, "decode_slice_header error\n");
3058 else if(err == 1) {
3059 /* Slice could not be decoded in parallel mode, copy down
3060 * NAL unit stuff to context 0 and restart. Note that
1412060e 3061 * rbsp_buffer is not transferred, but since we no longer
afebe2f7
3062 * run in parallel mode this should not be an issue. */
3063 h->nal_unit_type = hx->nal_unit_type;
3064 h->nal_ref_idc = hx->nal_ref_idc;
3065 hx = h;
3066 goto again;
3067 }
3068 }
3069 if(context_count)
3070 execute_decode_slices(h, context_count);
0da71265
MN
3071 return buf_index;
3072}
3073
3074/**
3b66c4c5 3075 * returns the number of bytes consumed for building the current frame
0da71265
MN
3076 */
3077static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size){
755bfeab 3078 if(pos==0) pos=1; //avoid infinite loops (i doubt that is needed but ...)
0da71265
MN
3079 if(pos+10>buf_size) pos=buf_size; // oops ;)
3080
3081 return pos;
0da71265
MN
3082}
3083
115329f1 3084static int decode_frame(AVCodecContext *avctx,
0da71265 3085 void *data, int *data_size,
7a00bbad 3086 AVPacket *avpkt)
0da71265 3087{
7a00bbad
TB
3088 const uint8_t *buf = avpkt->data;
3089 int buf_size = avpkt->size;
0da71265
MN
3090 H264Context *h = avctx->priv_data;
3091 MpegEncContext *s = &h->s;
115329f1 3092 AVFrame *pict = data;
0da71265 3093 int buf_index;
115329f1 3094
0da71265 3095 s->flags= avctx->flags;
303e50e6 3096 s->flags2= avctx->flags2;
0da71265 3097
1412060e 3098 /* end of stream, output what is still in the buffers */
9d252137 3099 out:
0da71265 3100 if (buf_size == 0) {
97bbb885
MN
3101 Picture *out;
3102 int i, out_idx;
3103
3104//FIXME factorize this with the output code below
3105 out = h->delayed_pic[0];
3106 out_idx = 0;
c173a088 3107 for(i=1; h->delayed_pic[i] && !h->delayed_pic[i]->key_frame && !h->delayed_pic[i]->mmco_reset; i++)
97bbb885
MN
3108 if(h->delayed_pic[i]->poc < out->poc){
3109 out = h->delayed_pic[i];
3110 out_idx = i;
3111 }
3112
3113 for(i=out_idx; h->delayed_pic[i]; i++)
3114 h->delayed_pic[i] = h->delayed_pic[i+1];
3115
3116 if(out){
3117 *data_size = sizeof(AVFrame);
3118 *pict= *(AVFrame*)out;
3119 }
3120
0da71265
MN
3121 return 0;
3122 }
115329f1 3123
0da71265 3124 buf_index=decode_nal_units(h, buf, buf_size);
115329f1 3125 if(buf_index < 0)