100l, I broke H.264 again, forgot one hunk.
[libav.git] / libavcodec / h264.c
1 /*
2 * H.26L/H.264/AVC/JVT/14496-10/... encoder/decoder
3 * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
4 *
5 * This file is part of FFmpeg.
6 *
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
11 *
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
16 *
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20 */
21
22 /**
23 * @file h264.c
24 * H.264 / AVC / MPEG4 part10 codec.
25 * @author Michael Niedermayer <michaelni@gmx.at>
26 */
27
28 #include "dsputil.h"
29 #include "avcodec.h"
30 #include "mpegvideo.h"
31 #include "h264.h"
32 #include "h264data.h"
33 #include "h264_parser.h"
34 #include "golomb.h"
35 #include "rectangle.h"
36
37 #include "cabac.h"
38 #ifdef ARCH_X86
39 #include "x86/h264_i386.h"
40 #endif
41
42 //#undef NDEBUG
43 #include <assert.h>
44
45 /**
46 * Value of Picture.reference when Picture is not a reference picture, but
47 * is held for delayed output.
48 */
49 #define DELAYED_PIC_REF 4
50
51 static VLC coeff_token_vlc[4];
52 static VLC_TYPE coeff_token_vlc_tables[520+332+280+256][2];
53 static const int coeff_token_vlc_tables_size[4]={520,332,280,256};
54
55 static VLC chroma_dc_coeff_token_vlc;
56 static VLC_TYPE chroma_dc_coeff_token_vlc_table[256][2];
57 static const int chroma_dc_coeff_token_vlc_table_size = 256;
58
59 static VLC total_zeros_vlc[15];
60 static VLC_TYPE total_zeros_vlc_tables[15][512][2];
61 static const int total_zeros_vlc_tables_size = 512;
62
63 static VLC chroma_dc_total_zeros_vlc[3];
64 static VLC_TYPE chroma_dc_total_zeros_vlc_tables[3][8][2];
65 static const int chroma_dc_total_zeros_vlc_tables_size = 8;
66
67 static VLC run_vlc[6];
68 static VLC_TYPE run_vlc_tables[6][8][2];
69 static const int run_vlc_tables_size = 8;
70
71 static VLC run7_vlc;
72 static VLC_TYPE run7_vlc_table[96][2];
73 static const int run7_vlc_table_size = 96;
74
75 static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
76 static void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
77 static void filter_mb( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr, unsigned int linesize, unsigned int uvlinesize);
78 static void filter_mb_fast( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr, unsigned int linesize, unsigned int uvlinesize);
79 static Picture * remove_long(H264Context *h, int i, int ref_mask);
80
81 static av_always_inline uint32_t pack16to32(int a, int b){
82 #ifdef WORDS_BIGENDIAN
83 return (b&0xFFFF) + (a<<16);
84 #else
85 return (a&0xFFFF) + (b<<16);
86 #endif
87 }
88
89 static const uint8_t rem6[52]={
90 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,
91 };
92
93 static const uint8_t div6[52]={
94 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 4, 4, 4, 4, 4, 4, 5, 5, 5, 5, 5, 5, 6, 6, 6, 6, 6, 6, 7, 7, 7, 7, 7, 7, 8, 8, 8, 8,
95 };
96
97 static const int left_block_options[4][8]={
98 {0,1,2,3,7,10,8,11},
99 {2,2,3,3,8,11,8,11},
100 {0,0,1,1,7,10,7,10},
101 {0,2,0,2,7,10,7,10}
102 };
103
104 static void fill_caches(H264Context *h, int mb_type, int for_deblock){
105 MpegEncContext * const s = &h->s;
106 const int mb_xy= h->mb_xy;
107 int topleft_xy, top_xy, topright_xy, left_xy[2];
108 int topleft_type, top_type, topright_type, left_type[2];
109 const int * left_block;
110 int topleft_partition= -1;
111 int i;
112
113 top_xy = mb_xy - (s->mb_stride << FIELD_PICTURE);
114
115 //FIXME deblocking could skip the intra and nnz parts.
116 if(for_deblock && (h->slice_num == 1 || h->slice_table[mb_xy] == h->slice_table[top_xy]) && !FRAME_MBAFF)
117 return;
118
119 /* Wow, what a mess, why didn't they simplify the interlacing & intra
120 * stuff, I can't imagine that these complex rules are worth it. */
121
122 topleft_xy = top_xy - 1;
123 topright_xy= top_xy + 1;
124 left_xy[1] = left_xy[0] = mb_xy-1;
125 left_block = left_block_options[0];
126 if(FRAME_MBAFF){
127 const int pair_xy = s->mb_x + (s->mb_y & ~1)*s->mb_stride;
128 const int top_pair_xy = pair_xy - s->mb_stride;
129 const int topleft_pair_xy = top_pair_xy - 1;
130 const int topright_pair_xy = top_pair_xy + 1;
131 const int topleft_mb_field_flag = IS_INTERLACED(s->current_picture.mb_type[topleft_pair_xy]);
132 const int top_mb_field_flag = IS_INTERLACED(s->current_picture.mb_type[top_pair_xy]);
133 const int topright_mb_field_flag = IS_INTERLACED(s->current_picture.mb_type[topright_pair_xy]);
134 const int left_mb_field_flag = IS_INTERLACED(s->current_picture.mb_type[pair_xy-1]);
135 const int curr_mb_field_flag = IS_INTERLACED(mb_type);
136 const int bottom = (s->mb_y & 1);
137 tprintf(s->avctx, "fill_caches: curr_mb_field_flag:%d, left_mb_field_flag:%d, topleft_mb_field_flag:%d, top_mb_field_flag:%d, topright_mb_field_flag:%d\n", curr_mb_field_flag, left_mb_field_flag, topleft_mb_field_flag, top_mb_field_flag, topright_mb_field_flag);
138
139 if (curr_mb_field_flag && (bottom || top_mb_field_flag)){
140 top_xy -= s->mb_stride;
141 }
142 if (curr_mb_field_flag && (bottom || topleft_mb_field_flag)){
143 topleft_xy -= s->mb_stride;
144 } else if(bottom && !curr_mb_field_flag && left_mb_field_flag) {
145 topleft_xy += s->mb_stride;
146 // take top left mv from the middle of the mb, as opposed to all other modes which use the bottom right partition
147 topleft_partition = 0;
148 }
149 if (curr_mb_field_flag && (bottom || topright_mb_field_flag)){
150 topright_xy -= s->mb_stride;
151 }
152 if (left_mb_field_flag != curr_mb_field_flag) {
153 left_xy[1] = left_xy[0] = pair_xy - 1;
154 if (curr_mb_field_flag) {
155 left_xy[1] += s->mb_stride;
156 left_block = left_block_options[3];
157 } else {
158 left_block= left_block_options[2 - bottom];
159 }
160 }
161 }
162
163 h->top_mb_xy = top_xy;
164 h->left_mb_xy[0] = left_xy[0];
165 h->left_mb_xy[1] = left_xy[1];
166 if(for_deblock){
167 topleft_type = 0;
168 topright_type = 0;
169 top_type = h->slice_table[top_xy ] < 0xFFFF ? s->current_picture.mb_type[top_xy] : 0;
170 left_type[0] = h->slice_table[left_xy[0] ] < 0xFFFF ? s->current_picture.mb_type[left_xy[0]] : 0;
171 left_type[1] = h->slice_table[left_xy[1] ] < 0xFFFF ? s->current_picture.mb_type[left_xy[1]] : 0;
172
173 if(MB_MBAFF && !IS_INTRA(mb_type)){
174 int list;
175 for(list=0; list<h->list_count; list++){
176 //These values where changed for ease of performing MC, we need to change them back
177 //FIXME maybe we can make MC and loop filter use the same values or prevent
178 //the MC code from changing ref_cache and rather use a temporary array.
179 if(USES_LIST(mb_type,list)){
180 int8_t *ref = &s->current_picture.ref_index[list][h->mb2b8_xy[mb_xy]];
181 *(uint32_t*)&h->ref_cache[list][scan8[ 0]] =
182 *(uint32_t*)&h->ref_cache[list][scan8[ 2]] = (pack16to32(ref[0],ref[1])&0x00FF00FF)*0x0101;
183 ref += h->b8_stride;
184 *(uint32_t*)&h->ref_cache[list][scan8[ 8]] =
185 *(uint32_t*)&h->ref_cache[list][scan8[10]] = (pack16to32(ref[0],ref[1])&0x00FF00FF)*0x0101;
186 }
187 }
188 }
189 }else{
190 topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
191 top_type = h->slice_table[top_xy ] == h->slice_num ? s->current_picture.mb_type[top_xy] : 0;
192 topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
193 left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
194 left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
195
196 if(IS_INTRA(mb_type)){
197 int type_mask= h->pps.constrained_intra_pred ? IS_INTRA(-1) : -1;
198 h->topleft_samples_available=
199 h->top_samples_available=
200 h->left_samples_available= 0xFFFF;
201 h->topright_samples_available= 0xEEEA;
202
203 if(!(top_type & type_mask)){
204 h->topleft_samples_available= 0xB3FF;
205 h->top_samples_available= 0x33FF;
206 h->topright_samples_available= 0x26EA;
207 }
208 if(IS_INTERLACED(mb_type) != IS_INTERLACED(left_type[0])){
209 if(IS_INTERLACED(mb_type)){
210 if(!(left_type[0] & type_mask)){
211 h->topleft_samples_available&= 0xDFFF;
212 h->left_samples_available&= 0x5FFF;
213 }
214 if(!(left_type[1] & type_mask)){
215 h->topleft_samples_available&= 0xFF5F;
216 h->left_samples_available&= 0xFF5F;
217 }
218 }else{
219 int left_typei = h->slice_table[left_xy[0] + s->mb_stride ] == h->slice_num
220 ? s->current_picture.mb_type[left_xy[0] + s->mb_stride] : 0;
221 assert(left_xy[0] == left_xy[1]);
222 if(!((left_typei & type_mask) && (left_type[0] & type_mask))){
223 h->topleft_samples_available&= 0xDF5F;
224 h->left_samples_available&= 0x5F5F;
225 }
226 }
227 }else{
228 if(!(left_type[0] & type_mask)){
229 h->topleft_samples_available&= 0xDF5F;
230 h->left_samples_available&= 0x5F5F;
231 }
232 }
233
234 if(!(topleft_type & type_mask))
235 h->topleft_samples_available&= 0x7FFF;
236
237 if(!(topright_type & type_mask))
238 h->topright_samples_available&= 0xFBFF;
239
240 if(IS_INTRA4x4(mb_type)){
241 if(IS_INTRA4x4(top_type)){
242 h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
243 h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
244 h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
245 h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
246 }else{
247 int pred;
248 if(!(top_type & type_mask))
249 pred= -1;
250 else{
251 pred= 2;
252 }
253 h->intra4x4_pred_mode_cache[4+8*0]=
254 h->intra4x4_pred_mode_cache[5+8*0]=
255 h->intra4x4_pred_mode_cache[6+8*0]=
256 h->intra4x4_pred_mode_cache[7+8*0]= pred;
257 }
258 for(i=0; i<2; i++){
259 if(IS_INTRA4x4(left_type[i])){
260 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
261 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
262 }else{
263 int pred;
264 if(!(left_type[i] & type_mask))
265 pred= -1;
266 else{
267 pred= 2;
268 }
269 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
270 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
271 }
272 }
273 }
274 }
275 }
276
277
278 /*
279 0 . T T. T T T T
280 1 L . .L . . . .
281 2 L . .L . . . .
282 3 . T TL . . . .
283 4 L . .L . . . .
284 5 L . .. . . . .
285 */
286 //FIXME constraint_intra_pred & partitioning & nnz (let us hope this is just a typo in the spec)
287 if(top_type){
288 h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][4];
289 h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][5];
290 h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][6];
291 h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3];
292
293 h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][9];
294 h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8];
295
296 h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][12];
297 h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11];
298
299 }else{
300 h->non_zero_count_cache[4+8*0]=
301 h->non_zero_count_cache[5+8*0]=
302 h->non_zero_count_cache[6+8*0]=
303 h->non_zero_count_cache[7+8*0]=
304
305 h->non_zero_count_cache[1+8*0]=
306 h->non_zero_count_cache[2+8*0]=
307
308 h->non_zero_count_cache[1+8*3]=
309 h->non_zero_count_cache[2+8*3]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
310
311 }
312
313 for (i=0; i<2; i++) {
314 if(left_type[i]){
315 h->non_zero_count_cache[3+8*1 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[0+2*i]];
316 h->non_zero_count_cache[3+8*2 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[1+2*i]];
317 h->non_zero_count_cache[0+8*1 + 8*i]= h->non_zero_count[left_xy[i]][left_block[4+2*i]];
318 h->non_zero_count_cache[0+8*4 + 8*i]= h->non_zero_count[left_xy[i]][left_block[5+2*i]];
319 }else{
320 h->non_zero_count_cache[3+8*1 + 2*8*i]=
321 h->non_zero_count_cache[3+8*2 + 2*8*i]=
322 h->non_zero_count_cache[0+8*1 + 8*i]=
323 h->non_zero_count_cache[0+8*4 + 8*i]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
324 }
325 }
326
327 if( h->pps.cabac ) {
328 // top_cbp
329 if(top_type) {
330 h->top_cbp = h->cbp_table[top_xy];
331 } else if(IS_INTRA(mb_type)) {
332 h->top_cbp = 0x1C0;
333 } else {
334 h->top_cbp = 0;
335 }
336 // left_cbp
337 if (left_type[0]) {
338 h->left_cbp = h->cbp_table[left_xy[0]] & 0x1f0;
339 } else if(IS_INTRA(mb_type)) {
340 h->left_cbp = 0x1C0;
341 } else {
342 h->left_cbp = 0;
343 }
344 if (left_type[0]) {
345 h->left_cbp |= ((h->cbp_table[left_xy[0]]>>((left_block[0]&(~1))+1))&0x1) << 1;
346 }
347 if (left_type[1]) {
348 h->left_cbp |= ((h->cbp_table[left_xy[1]]>>((left_block[2]&(~1))+1))&0x1) << 3;
349 }
350 }
351
352 #if 1
353 if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
354 int list;
355 for(list=0; list<h->list_count; list++){
356 if(!USES_LIST(mb_type, list) && !IS_DIRECT(mb_type) && !h->deblocking_filter){
357 /*if(!h->mv_cache_clean[list]){
358 memset(h->mv_cache [list], 0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
359 memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
360 h->mv_cache_clean[list]= 1;
361 }*/
362 continue;
363 }
364 h->mv_cache_clean[list]= 0;
365
366 if(USES_LIST(top_type, list)){
367 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
368 const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
369 *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0];
370 *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];
371 *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];
372 *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];
373 h->ref_cache[list][scan8[0] + 0 - 1*8]=
374 h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
375 h->ref_cache[list][scan8[0] + 2 - 1*8]=
376 h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
377 }else{
378 *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]=
379 *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]=
380 *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]=
381 *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;
382 *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
383 }
384
385 for(i=0; i<2; i++){
386 int cache_idx = scan8[0] - 1 + i*2*8;
387 if(USES_LIST(left_type[i], list)){
388 const int b_xy= h->mb2b_xy[left_xy[i]] + 3;
389 const int b8_xy= h->mb2b8_xy[left_xy[i]] + 1;
390 *(uint32_t*)h->mv_cache[list][cache_idx ]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[0+i*2]];
391 *(uint32_t*)h->mv_cache[list][cache_idx+8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[1+i*2]];
392 h->ref_cache[list][cache_idx ]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0+i*2]>>1)];
393 h->ref_cache[list][cache_idx+8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[1+i*2]>>1)];
394 }else{
395 *(uint32_t*)h->mv_cache [list][cache_idx ]=
396 *(uint32_t*)h->mv_cache [list][cache_idx+8]= 0;
397 h->ref_cache[list][cache_idx ]=
398 h->ref_cache[list][cache_idx+8]= left_type[i] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
399 }
400 }
401
402 if(for_deblock || ((IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred) && !FRAME_MBAFF))
403 continue;
404
405 if(USES_LIST(topleft_type, list)){
406 const int b_xy = h->mb2b_xy[topleft_xy] + 3 + h->b_stride + (topleft_partition & 2*h->b_stride);
407 const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + (topleft_partition & h->b8_stride);
408 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
409 h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
410 }else{
411 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
412 h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
413 }
414
415 if(USES_LIST(topright_type, list)){
416 const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
417 const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
418 *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
419 h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
420 }else{
421 *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
422 h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
423 }
424
425 if((IS_SKIP(mb_type) || IS_DIRECT(mb_type)) && !FRAME_MBAFF)
426 continue;
427
428 h->ref_cache[list][scan8[5 ]+1] =
429 h->ref_cache[list][scan8[7 ]+1] =
430 h->ref_cache[list][scan8[13]+1] = //FIXME remove past 3 (init somewhere else)
431 h->ref_cache[list][scan8[4 ]] =
432 h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
433 *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
434 *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
435 *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
436 *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
437 *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
438
439 if( h->pps.cabac ) {
440 /* XXX beurk, Load mvd */
441 if(USES_LIST(top_type, list)){
442 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
443 *(uint32_t*)h->mvd_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 0];
444 *(uint32_t*)h->mvd_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 1];
445 *(uint32_t*)h->mvd_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 2];
446 *(uint32_t*)h->mvd_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 3];
447 }else{
448 *(uint32_t*)h->mvd_cache [list][scan8[0] + 0 - 1*8]=
449 *(uint32_t*)h->mvd_cache [list][scan8[0] + 1 - 1*8]=
450 *(uint32_t*)h->mvd_cache [list][scan8[0] + 2 - 1*8]=
451 *(uint32_t*)h->mvd_cache [list][scan8[0] + 3 - 1*8]= 0;
452 }
453 if(USES_LIST(left_type[0], list)){
454 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
455 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 0*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[0]];
456 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[1]];
457 }else{
458 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 0*8]=
459 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 1*8]= 0;
460 }
461 if(USES_LIST(left_type[1], list)){
462 const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
463 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 2*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[2]];
464 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 3*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[3]];
465 }else{
466 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 2*8]=
467 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 3*8]= 0;
468 }
469 *(uint32_t*)h->mvd_cache [list][scan8[5 ]+1]=
470 *(uint32_t*)h->mvd_cache [list][scan8[7 ]+1]=
471 *(uint32_t*)h->mvd_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
472 *(uint32_t*)h->mvd_cache [list][scan8[4 ]]=
473 *(uint32_t*)h->mvd_cache [list][scan8[12]]= 0;
474
475 if(h->slice_type_nos == FF_B_TYPE){
476 fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, 0, 1);
477
478 if(IS_DIRECT(top_type)){
479 *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0x01010101;
480 }else if(IS_8X8(top_type)){
481 int b8_xy = h->mb2b8_xy[top_xy] + h->b8_stride;
482 h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy];
483 h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 1];
484 }else{
485 *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0;
486 }
487
488 if(IS_DIRECT(left_type[0]))
489 h->direct_cache[scan8[0] - 1 + 0*8]= 1;
490 else if(IS_8X8(left_type[0]))
491 h->direct_cache[scan8[0] - 1 + 0*8]= h->direct_table[h->mb2b8_xy[left_xy[0]] + 1 + h->b8_stride*(left_block[0]>>1)];
492 else
493 h->direct_cache[scan8[0] - 1 + 0*8]= 0;
494
495 if(IS_DIRECT(left_type[1]))
496 h->direct_cache[scan8[0] - 1 + 2*8]= 1;
497 else if(IS_8X8(left_type[1]))
498 h->direct_cache[scan8[0] - 1 + 2*8]= h->direct_table[h->mb2b8_xy[left_xy[1]] + 1 + h->b8_stride*(left_block[2]>>1)];
499 else
500 h->direct_cache[scan8[0] - 1 + 2*8]= 0;
501 }
502 }
503
504 if(FRAME_MBAFF){
505 #define MAP_MVS\
506 MAP_F2F(scan8[0] - 1 - 1*8, topleft_type)\
507 MAP_F2F(scan8[0] + 0 - 1*8, top_type)\
508 MAP_F2F(scan8[0] + 1 - 1*8, top_type)\
509 MAP_F2F(scan8[0] + 2 - 1*8, top_type)\
510 MAP_F2F(scan8[0] + 3 - 1*8, top_type)\
511 MAP_F2F(scan8[0] + 4 - 1*8, topright_type)\
512 MAP_F2F(scan8[0] - 1 + 0*8, left_type[0])\
513 MAP_F2F(scan8[0] - 1 + 1*8, left_type[0])\
514 MAP_F2F(scan8[0] - 1 + 2*8, left_type[1])\
515 MAP_F2F(scan8[0] - 1 + 3*8, left_type[1])
516 if(MB_FIELD){
517 #define MAP_F2F(idx, mb_type)\
518 if(!IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
519 h->ref_cache[list][idx] <<= 1;\
520 h->mv_cache[list][idx][1] /= 2;\
521 h->mvd_cache[list][idx][1] /= 2;\
522 }
523 MAP_MVS
524 #undef MAP_F2F
525 }else{
526 #define MAP_F2F(idx, mb_type)\
527 if(IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
528 h->ref_cache[list][idx] >>= 1;\
529 h->mv_cache[list][idx][1] <<= 1;\
530 h->mvd_cache[list][idx][1] <<= 1;\
531 }
532 MAP_MVS
533 #undef MAP_F2F
534 }
535 }
536 }
537 }
538 #endif
539
540 h->neighbor_transform_size= !!IS_8x8DCT(top_type) + !!IS_8x8DCT(left_type[0]);
541 }
542
543 static inline void write_back_intra_pred_mode(H264Context *h){
544 const int mb_xy= h->mb_xy;
545
546 h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
547 h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
548 h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
549 h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
550 h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
551 h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
552 h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
553 }
554
555 /**
556 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
557 */
558 static inline int check_intra4x4_pred_mode(H264Context *h){
559 MpegEncContext * const s = &h->s;
560 static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
561 static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
562 int i;
563
564 if(!(h->top_samples_available&0x8000)){
565 for(i=0; i<4; i++){
566 int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
567 if(status<0){
568 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);
569 return -1;
570 } else if(status){
571 h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
572 }
573 }
574 }
575
576 if((h->left_samples_available&0x8888)!=0x8888){
577 static const int mask[4]={0x8000,0x2000,0x80,0x20};
578 for(i=0; i<4; i++){
579 if(!(h->left_samples_available&mask[i])){
580 int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
581 if(status<0){
582 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);
583 return -1;
584 } else if(status){
585 h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
586 }
587 }
588 }
589 }
590
591 return 0;
592 } //FIXME cleanup like next
593
594 /**
595 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
596 */
597 static inline int check_intra_pred_mode(H264Context *h, int mode){
598 MpegEncContext * const s = &h->s;
599 static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
600 static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
601
602 if(mode > 6U) {
603 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);
604 return -1;
605 }
606
607 if(!(h->top_samples_available&0x8000)){
608 mode= top[ mode ];
609 if(mode<0){
610 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);
611 return -1;
612 }
613 }
614
615 if((h->left_samples_available&0x8080) != 0x8080){
616 mode= left[ mode ];
617 if(h->left_samples_available&0x8080){ //mad cow disease mode, aka MBAFF + constrained_intra_pred
618 mode= ALZHEIMER_DC_L0T_PRED8x8 + (!(h->left_samples_available&0x8000)) + 2*(mode == DC_128_PRED8x8);
619 }
620 if(mode<0){
621 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);
622 return -1;
623 }
624 }
625
626 return mode;
627 }
628
629 /**
630 * gets the predicted intra4x4 prediction mode.
631 */
632 static inline int pred_intra_mode(H264Context *h, int n){
633 const int index8= scan8[n];
634 const int left= h->intra4x4_pred_mode_cache[index8 - 1];
635 const int top = h->intra4x4_pred_mode_cache[index8 - 8];
636 const int min= FFMIN(left, top);
637
638 tprintf(h->s.avctx, "mode:%d %d min:%d\n", left ,top, min);
639
640 if(min<0) return DC_PRED;
641 else return min;
642 }
643
644 static inline void write_back_non_zero_count(H264Context *h){
645 const int mb_xy= h->mb_xy;
646
647 h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[7+8*1];
648 h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[7+8*2];
649 h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[7+8*3];
650 h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];
651 h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[4+8*4];
652 h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[5+8*4];
653 h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[6+8*4];
654
655 h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[1+8*2];
656 h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
657 h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[2+8*1];
658
659 h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[1+8*5];
660 h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
661 h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[2+8*4];
662 }
663
664 /**
665 * gets the predicted number of non-zero coefficients.
666 * @param n block index
667 */
668 static inline int pred_non_zero_count(H264Context *h, int n){
669 const int index8= scan8[n];
670 const int left= h->non_zero_count_cache[index8 - 1];
671 const int top = h->non_zero_count_cache[index8 - 8];
672 int i= left + top;
673
674 if(i<64) i= (i+1)>>1;
675
676 tprintf(h->s.avctx, "pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
677
678 return i&31;
679 }
680
681 static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
682 const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];
683 MpegEncContext *s = &h->s;
684
685 /* there is no consistent mapping of mvs to neighboring locations that will
686 * make mbaff happy, so we can't move all this logic to fill_caches */
687 if(FRAME_MBAFF){
688 const uint32_t *mb_types = s->current_picture_ptr->mb_type;
689 const int16_t *mv;
690 *(uint32_t*)h->mv_cache[list][scan8[0]-2] = 0;
691 *C = h->mv_cache[list][scan8[0]-2];
692
693 if(!MB_FIELD
694 && (s->mb_y&1) && i < scan8[0]+8 && topright_ref != PART_NOT_AVAILABLE){
695 int topright_xy = s->mb_x + (s->mb_y-1)*s->mb_stride + (i == scan8[0]+3);
696 if(IS_INTERLACED(mb_types[topright_xy])){
697 #define SET_DIAG_MV(MV_OP, REF_OP, X4, Y4)\
698 const int x4 = X4, y4 = Y4;\
699 const int mb_type = mb_types[(x4>>2)+(y4>>2)*s->mb_stride];\
700 if(!USES_LIST(mb_type,list))\
701 return LIST_NOT_USED;\
702 mv = s->current_picture_ptr->motion_val[list][x4 + y4*h->b_stride];\
703 h->mv_cache[list][scan8[0]-2][0] = mv[0];\
704 h->mv_cache[list][scan8[0]-2][1] = mv[1] MV_OP;\
705 return s->current_picture_ptr->ref_index[list][(x4>>1) + (y4>>1)*h->b8_stride] REF_OP;
706
707 SET_DIAG_MV(*2, >>1, s->mb_x*4+(i&7)-4+part_width, s->mb_y*4-1);
708 }
709 }
710 if(topright_ref == PART_NOT_AVAILABLE
711 && ((s->mb_y&1) || i >= scan8[0]+8) && (i&7)==4
712 && h->ref_cache[list][scan8[0]-1] != PART_NOT_AVAILABLE){
713 if(!MB_FIELD
714 && IS_INTERLACED(mb_types[h->left_mb_xy[0]])){
715 SET_DIAG_MV(*2, >>1, s->mb_x*4-1, (s->mb_y|1)*4+(s->mb_y&1)*2+(i>>4)-1);
716 }
717 if(MB_FIELD
718 && !IS_INTERLACED(mb_types[h->left_mb_xy[0]])
719 && i >= scan8[0]+8){
720 // left shift will turn LIST_NOT_USED into PART_NOT_AVAILABLE, but that's OK.
721 SET_DIAG_MV(/2, <<1, s->mb_x*4-1, (s->mb_y&~1)*4 - 1 + ((i-scan8[0])>>3)*2);
722 }
723 }
724 #undef SET_DIAG_MV
725 }
726
727 if(topright_ref != PART_NOT_AVAILABLE){
728 *C= h->mv_cache[list][ i - 8 + part_width ];
729 return topright_ref;
730 }else{
731 tprintf(s->avctx, "topright MV not available\n");
732
733 *C= h->mv_cache[list][ i - 8 - 1 ];
734 return h->ref_cache[list][ i - 8 - 1 ];
735 }
736 }
737
738 /**
739 * gets the predicted MV.
740 * @param n the block index
741 * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)
742 * @param mx the x component of the predicted motion vector
743 * @param my the y component of the predicted motion vector
744 */
745 static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
746 const int index8= scan8[n];
747 const int top_ref= h->ref_cache[list][ index8 - 8 ];
748 const int left_ref= h->ref_cache[list][ index8 - 1 ];
749 const int16_t * const A= h->mv_cache[list][ index8 - 1 ];
750 const int16_t * const B= h->mv_cache[list][ index8 - 8 ];
751 const int16_t * C;
752 int diagonal_ref, match_count;
753
754 assert(part_width==1 || part_width==2 || part_width==4);
755
756 /* mv_cache
757 B . . A T T T T
758 U . . L . . , .
759 U . . L . . . .
760 U . . L . . , .
761 . . . L . . . .
762 */
763
764 diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);
765 match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);
766 tprintf(h->s.avctx, "pred_motion match_count=%d\n", match_count);
767 if(match_count > 1){ //most common
768 *mx= mid_pred(A[0], B[0], C[0]);
769 *my= mid_pred(A[1], B[1], C[1]);
770 }else if(match_count==1){
771 if(left_ref==ref){
772 *mx= A[0];
773 *my= A[1];
774 }else if(top_ref==ref){
775 *mx= B[0];
776 *my= B[1];
777 }else{
778 *mx= C[0];
779 *my= C[1];
780 }
781 }else{
782 if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
783 *mx= A[0];
784 *my= A[1];
785 }else{
786 *mx= mid_pred(A[0], B[0], C[0]);
787 *my= mid_pred(A[1], B[1], C[1]);
788 }
789 }
790
791 tprintf(h->s.avctx, "pred_motion (%2d %2d %2d) (%2d %2d %2d) (%2d %2d %2d) -> (%2d %2d %2d) at %2d %2d %d list %d\n", top_ref, B[0], B[1], diagonal_ref, C[0], C[1], left_ref, A[0], A[1], ref, *mx, *my, h->s.mb_x, h->s.mb_y, n, list);
792 }
793
794 /**
795 * gets the directionally predicted 16x8 MV.
796 * @param n the block index
797 * @param mx the x component of the predicted motion vector
798 * @param my the y component of the predicted motion vector
799 */
800 static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
801 if(n==0){
802 const int top_ref= h->ref_cache[list][ scan8[0] - 8 ];
803 const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
804
805 tprintf(h->s.avctx, "pred_16x8: (%2d %2d %2d) at %2d %2d %d list %d\n", top_ref, B[0], B[1], h->s.mb_x, h->s.mb_y, n, list);
806
807 if(top_ref == ref){
808 *mx= B[0];
809 *my= B[1];
810 return;
811 }
812 }else{
813 const int left_ref= h->ref_cache[list][ scan8[8] - 1 ];
814 const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];
815
816 tprintf(h->s.avctx, "pred_16x8: (%2d %2d %2d) at %2d %2d %d list %d\n", left_ref, A[0], A[1], h->s.mb_x, h->s.mb_y, n, list);
817
818 if(left_ref == ref){
819 *mx= A[0];
820 *my= A[1];
821 return;
822 }
823 }
824
825 //RARE
826 pred_motion(h, n, 4, list, ref, mx, my);
827 }
828
829 /**
830 * gets the directionally predicted 8x16 MV.
831 * @param n the block index
832 * @param mx the x component of the predicted motion vector
833 * @param my the y component of the predicted motion vector
834 */
835 static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
836 if(n==0){
837 const int left_ref= h->ref_cache[list][ scan8[0] - 1 ];
838 const int16_t * const A= h->mv_cache[list][ scan8[0] - 1 ];
839
840 tprintf(h->s.avctx, "pred_8x16: (%2d %2d %2d) at %2d %2d %d list %d\n", left_ref, A[0], A[1], h->s.mb_x, h->s.mb_y, n, list);
841
842 if(left_ref == ref){
843 *mx= A[0];
844 *my= A[1];
845 return;
846 }
847 }else{
848 const int16_t * C;
849 int diagonal_ref;
850
851 diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);
852
853 tprintf(h->s.avctx, "pred_8x16: (%2d %2d %2d) at %2d %2d %d list %d\n", diagonal_ref, C[0], C[1], h->s.mb_x, h->s.mb_y, n, list);
854
855 if(diagonal_ref == ref){
856 *mx= C[0];
857 *my= C[1];
858 return;
859 }
860 }
861
862 //RARE
863 pred_motion(h, n, 2, list, ref, mx, my);
864 }
865
866 static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
867 const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
868 const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];
869
870 tprintf(h->s.avctx, "pred_pskip: (%d) (%d) at %2d %2d\n", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
871
872 if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
873 || (top_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0)
874 || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){
875
876 *mx = *my = 0;
877 return;
878 }
879
880 pred_motion(h, 0, 4, 0, 0, mx, my);
881
882 return;
883 }
884
885 static int get_scale_factor(H264Context * const h, int poc, int poc1, int i){
886 int poc0 = h->ref_list[0][i].poc;
887 int td = av_clip(poc1 - poc0, -128, 127);
888 if(td == 0 || h->ref_list[0][i].long_ref){
889 return 256;
890 }else{
891 int tb = av_clip(poc - poc0, -128, 127);
892 int tx = (16384 + (FFABS(td) >> 1)) / td;
893 return av_clip((tb*tx + 32) >> 6, -1024, 1023);
894 }
895 }
896
897 static inline void direct_dist_scale_factor(H264Context * const h){
898 MpegEncContext * const s = &h->s;
899 const int poc = h->s.current_picture_ptr->field_poc[ s->picture_structure == PICT_BOTTOM_FIELD ];
900 const int poc1 = h->ref_list[1][0].poc;
901 int i, field;
902 for(field=0; field<2; field++){
903 const int poc = h->s.current_picture_ptr->field_poc[field];
904 const int poc1 = h->ref_list[1][0].field_poc[field];
905 for(i=0; i < 2*h->ref_count[0]; i++)
906 h->dist_scale_factor_field[field][i^field] = get_scale_factor(h, poc, poc1, i+16);
907 }
908
909 for(i=0; i<h->ref_count[0]; i++){
910 h->dist_scale_factor[i] = get_scale_factor(h, poc, poc1, i);
911 }
912 }
913
914 static void fill_colmap(H264Context *h, int map[2][16+32], int list, int field, int colfield, int mbafi){
915 MpegEncContext * const s = &h->s;
916 Picture * const ref1 = &h->ref_list[1][0];
917 int j, old_ref, rfield;
918 int start= mbafi ? 16 : 0;
919 int end = mbafi ? 16+2*h->ref_count[list] : h->ref_count[list];
920 int interl= mbafi || s->picture_structure != PICT_FRAME;
921
922 /* bogus; fills in for missing frames */
923 memset(map[list], 0, sizeof(map[list]));
924
925 for(rfield=0; rfield<2; rfield++){
926 for(old_ref=0; old_ref<ref1->ref_count[colfield][list]; old_ref++){
927 int poc = ref1->ref_poc[colfield][list][old_ref];
928
929 if (!interl)
930 poc |= 3;
931 else if( interl && (poc&3) == 3) //FIXME store all MBAFF references so this isnt needed
932 poc= (poc&~3) + rfield + 1;
933
934 for(j=start; j<end; j++){
935 if(4*h->ref_list[list][j].frame_num + (h->ref_list[list][j].reference&3) == poc){
936 int cur_ref= mbafi ? (j-16)^field : j;
937 map[list][2*old_ref + (rfield^field) + 16] = cur_ref;
938 if(rfield == field)
939 map[list][old_ref] = cur_ref;
940 break;
941 }
942 }
943 }
944 }
945 }
946
947 static inline void direct_ref_list_init(H264Context * const h){
948 MpegEncContext * const s = &h->s;
949 Picture * const ref1 = &h->ref_list[1][0];
950 Picture * const cur = s->current_picture_ptr;
951 int list, j, field;
952 int sidx= (s->picture_structure&1)^1;
953 int ref1sidx= (ref1->reference&1)^1;
954
955 for(list=0; list<2; list++){
956 cur->ref_count[sidx][list] = h->ref_count[list];
957 for(j=0; j<h->ref_count[list]; j++)
958 cur->ref_poc[sidx][list][j] = 4*h->ref_list[list][j].frame_num + (h->ref_list[list][j].reference&3);
959 }
960
961 if(s->picture_structure == PICT_FRAME){
962 memcpy(cur->ref_count[1], cur->ref_count[0], sizeof(cur->ref_count[0]));
963 memcpy(cur->ref_poc [1], cur->ref_poc [0], sizeof(cur->ref_poc [0]));
964 }
965
966 cur->mbaff= FRAME_MBAFF;
967
968 if(cur->pict_type != FF_B_TYPE || h->direct_spatial_mv_pred)
969 return;
970
971 for(list=0; list<2; list++){
972 fill_colmap(h, h->map_col_to_list0, list, sidx, ref1sidx, 0);
973 for(field=0; field<2; field++)
974 fill_colmap(h, h->map_col_to_list0_field[field], list, field, field, 1);
975 }
976 }
977
978 static inline void pred_direct_motion(H264Context * const h, int *mb_type){
979 MpegEncContext * const s = &h->s;
980 int b8_stride = h->b8_stride;
981 int b4_stride = h->b_stride;
982 int mb_xy = h->mb_xy;
983 int mb_type_col[2];
984 const int16_t (*l1mv0)[2], (*l1mv1)[2];
985 const int8_t *l1ref0, *l1ref1;
986 const int is_b8x8 = IS_8X8(*mb_type);
987 unsigned int sub_mb_type;
988 int i8, i4;
989
990 #define MB_TYPE_16x16_OR_INTRA (MB_TYPE_16x16|MB_TYPE_INTRA4x4|MB_TYPE_INTRA16x16|MB_TYPE_INTRA_PCM)
991
992 if(IS_INTERLACED(h->ref_list[1][0].mb_type[mb_xy])){ // AFL/AFR/FR/FL -> AFL/FL
993 if(!IS_INTERLACED(*mb_type)){ // AFR/FR -> AFL/FL
994 int cur_poc = s->current_picture_ptr->poc;
995 int *col_poc = h->ref_list[1]->field_poc;
996 int col_parity = FFABS(col_poc[0] - cur_poc) >= FFABS(col_poc[1] - cur_poc);
997 mb_xy= s->mb_x + ((s->mb_y&~1) + col_parity)*s->mb_stride;
998 b8_stride = 0;
999 }else if(!(s->picture_structure & h->ref_list[1][0].reference) && !h->ref_list[1][0].mbaff){// FL -> FL & differ parity
1000 int fieldoff= 2*(h->ref_list[1][0].reference)-3;
1001 mb_xy += s->mb_stride*fieldoff;
1002 }
1003 goto single_col;
1004 }else{ // AFL/AFR/FR/FL -> AFR/FR
1005 if(IS_INTERLACED(*mb_type)){ // AFL /FL -> AFR/FR
1006 mb_xy= s->mb_x + (s->mb_y&~1)*s->mb_stride;
1007 mb_type_col[0] = h->ref_list[1][0].mb_type[mb_xy];
1008 mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy + s->mb_stride];
1009 b8_stride *= 3;
1010 b4_stride *= 6;
1011 //FIXME IS_8X8(mb_type_col[0]) && !h->sps.direct_8x8_inference_flag
1012 if( (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)
1013 && (mb_type_col[1] & MB_TYPE_16x16_OR_INTRA)
1014 && !is_b8x8){
1015 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1016 *mb_type |= MB_TYPE_16x8 |MB_TYPE_L0L1|MB_TYPE_DIRECT2; /* B_16x8 */
1017 }else{
1018 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1019 *mb_type |= MB_TYPE_8x8|MB_TYPE_L0L1;
1020 }
1021 }else{ // AFR/FR -> AFR/FR
1022 single_col:
1023 mb_type_col[0] =
1024 mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy];
1025 if(IS_8X8(mb_type_col[0]) && !h->sps.direct_8x8_inference_flag){
1026 /* FIXME save sub mb types from previous frames (or derive from MVs)
1027 * so we know exactly what block size to use */
1028 sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */
1029 *mb_type |= MB_TYPE_8x8|MB_TYPE_L0L1;
1030 }else if(!is_b8x8 && (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)){
1031 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1032 *mb_type |= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */
1033 }else{
1034 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1035 *mb_type |= MB_TYPE_8x8|MB_TYPE_L0L1;
1036 }
1037 }
1038 }
1039
1040 l1mv0 = &h->ref_list[1][0].motion_val[0][h->mb2b_xy [mb_xy]];
1041 l1mv1 = &h->ref_list[1][0].motion_val[1][h->mb2b_xy [mb_xy]];
1042 l1ref0 = &h->ref_list[1][0].ref_index [0][h->mb2b8_xy[mb_xy]];
1043 l1ref1 = &h->ref_list[1][0].ref_index [1][h->mb2b8_xy[mb_xy]];
1044 if(!b8_stride){
1045 if(s->mb_y&1){
1046 l1ref0 += h->b8_stride;
1047 l1ref1 += h->b8_stride;
1048 l1mv0 += 2*b4_stride;
1049 l1mv1 += 2*b4_stride;
1050 }
1051 }
1052
1053 if(h->direct_spatial_mv_pred){
1054 int ref[2];
1055 int mv[2][2];
1056 int list;
1057
1058 /* FIXME interlacing + spatial direct uses wrong colocated block positions */
1059
1060 /* ref = min(neighbors) */
1061 for(list=0; list<2; list++){
1062 int refa = h->ref_cache[list][scan8[0] - 1];
1063 int refb = h->ref_cache[list][scan8[0] - 8];
1064 int refc = h->ref_cache[list][scan8[0] - 8 + 4];
1065 if(refc == PART_NOT_AVAILABLE)
1066 refc = h->ref_cache[list][scan8[0] - 8 - 1];
1067 ref[list] = FFMIN3((unsigned)refa, (unsigned)refb, (unsigned)refc);
1068 if(ref[list] < 0)
1069 ref[list] = -1;
1070 }
1071
1072 if(ref[0] < 0 && ref[1] < 0){
1073 ref[0] = ref[1] = 0;
1074 mv[0][0] = mv[0][1] =
1075 mv[1][0] = mv[1][1] = 0;
1076 }else{
1077 for(list=0; list<2; list++){
1078 if(ref[list] >= 0)
1079 pred_motion(h, 0, 4, list, ref[list], &mv[list][0], &mv[list][1]);
1080 else
1081 mv[list][0] = mv[list][1] = 0;
1082 }
1083 }
1084
1085 if(ref[1] < 0){
1086 if(!is_b8x8)
1087 *mb_type &= ~MB_TYPE_L1;
1088 sub_mb_type &= ~MB_TYPE_L1;
1089 }else if(ref[0] < 0){
1090 if(!is_b8x8)
1091 *mb_type &= ~MB_TYPE_L0;
1092 sub_mb_type &= ~MB_TYPE_L0;
1093 }
1094
1095 if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col[0])){
1096 for(i8=0; i8<4; i8++){
1097 int x8 = i8&1;
1098 int y8 = i8>>1;
1099 int xy8 = x8+y8*b8_stride;
1100 int xy4 = 3*x8+y8*b4_stride;
1101 int a=0, b=0;
1102
1103 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1104 continue;
1105 h->sub_mb_type[i8] = sub_mb_type;
1106
1107 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);
1108 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);
1109 if(!IS_INTRA(mb_type_col[y8])
1110 && ( (l1ref0[xy8] == 0 && FFABS(l1mv0[xy4][0]) <= 1 && FFABS(l1mv0[xy4][1]) <= 1)
1111 || (l1ref0[xy8] < 0 && l1ref1[xy8] == 0 && FFABS(l1mv1[xy4][0]) <= 1 && FFABS(l1mv1[xy4][1]) <= 1))){
1112 if(ref[0] > 0)
1113 a= pack16to32(mv[0][0],mv[0][1]);
1114 if(ref[1] > 0)
1115 b= pack16to32(mv[1][0],mv[1][1]);
1116 }else{
1117 a= pack16to32(mv[0][0],mv[0][1]);
1118 b= pack16to32(mv[1][0],mv[1][1]);
1119 }
1120 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, a, 4);
1121 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, b, 4);
1122 }
1123 }else if(IS_16X16(*mb_type)){
1124 int a=0, b=0;
1125
1126 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, (uint8_t)ref[0], 1);
1127 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, (uint8_t)ref[1], 1);
1128 if(!IS_INTRA(mb_type_col[0])
1129 && ( (l1ref0[0] == 0 && FFABS(l1mv0[0][0]) <= 1 && FFABS(l1mv0[0][1]) <= 1)
1130 || (l1ref0[0] < 0 && l1ref1[0] == 0 && FFABS(l1mv1[0][0]) <= 1 && FFABS(l1mv1[0][1]) <= 1
1131 && (h->x264_build>33 || !h->x264_build)))){
1132 if(ref[0] > 0)
1133 a= pack16to32(mv[0][0],mv[0][1]);
1134 if(ref[1] > 0)
1135 b= pack16to32(mv[1][0],mv[1][1]);
1136 }else{
1137 a= pack16to32(mv[0][0],mv[0][1]);
1138 b= pack16to32(mv[1][0],mv[1][1]);
1139 }
1140 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, a, 4);
1141 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, b, 4);
1142 }else{
1143 for(i8=0; i8<4; i8++){
1144 const int x8 = i8&1;
1145 const int y8 = i8>>1;
1146
1147 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1148 continue;
1149 h->sub_mb_type[i8] = sub_mb_type;
1150
1151 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1152 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1153 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);
1154 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);
1155
1156 /* col_zero_flag */
1157 if(!IS_INTRA(mb_type_col[0]) && ( l1ref0[x8 + y8*b8_stride] == 0
1158 || (l1ref0[x8 + y8*b8_stride] < 0 && l1ref1[x8 + y8*b8_stride] == 0
1159 && (h->x264_build>33 || !h->x264_build)))){
1160 const int16_t (*l1mv)[2]= l1ref0[x8 + y8*b8_stride] == 0 ? l1mv0 : l1mv1;
1161 if(IS_SUB_8X8(sub_mb_type)){
1162 const int16_t *mv_col = l1mv[x8*3 + y8*3*b4_stride];
1163 if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
1164 if(ref[0] == 0)
1165 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1166 if(ref[1] == 0)
1167 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1168 }
1169 }else
1170 for(i4=0; i4<4; i4++){
1171 const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*b4_stride];
1172 if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
1173 if(ref[0] == 0)
1174 *(uint32_t*)h->mv_cache[0][scan8[i8*4+i4]] = 0;
1175 if(ref[1] == 0)
1176 *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = 0;
1177 }
1178 }
1179 }
1180 }
1181 }
1182 }else{ /* direct temporal mv pred */
1183 const int *map_col_to_list0[2] = {h->map_col_to_list0[0], h->map_col_to_list0[1]};
1184 const int *dist_scale_factor = h->dist_scale_factor;
1185 int ref_offset= 0;
1186
1187 if(FRAME_MBAFF && IS_INTERLACED(*mb_type)){
1188 map_col_to_list0[0] = h->map_col_to_list0_field[s->mb_y&1][0];
1189 map_col_to_list0[1] = h->map_col_to_list0_field[s->mb_y&1][1];
1190 dist_scale_factor =h->dist_scale_factor_field[s->mb_y&1];
1191 }
1192 if(h->ref_list[1][0].mbaff && IS_INTERLACED(mb_type_col[0]))
1193 ref_offset += 16;
1194
1195 if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col[0])){
1196 /* FIXME assumes direct_8x8_inference == 1 */
1197 int y_shift = 2*!IS_INTERLACED(*mb_type);
1198
1199 for(i8=0; i8<4; i8++){
1200 const int x8 = i8&1;
1201 const int y8 = i8>>1;
1202 int ref0, scale;
1203 const int16_t (*l1mv)[2]= l1mv0;
1204
1205 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1206 continue;
1207 h->sub_mb_type[i8] = sub_mb_type;
1208
1209 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1210 if(IS_INTRA(mb_type_col[y8])){
1211 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1212 fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1213 fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1214 continue;
1215 }
1216
1217 ref0 = l1ref0[x8 + y8*b8_stride];
1218 if(ref0 >= 0)
1219 ref0 = map_col_to_list0[0][ref0 + ref_offset];
1220 else{
1221 ref0 = map_col_to_list0[1][l1ref1[x8 + y8*b8_stride] + ref_offset];
1222 l1mv= l1mv1;
1223 }
1224 scale = dist_scale_factor[ref0];
1225 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1226
1227 {
1228 const int16_t *mv_col = l1mv[x8*3 + y8*b4_stride];
1229 int my_col = (mv_col[1]<<y_shift)/2;
1230 int mx = (scale * mv_col[0] + 128) >> 8;
1231 int my = (scale * my_col + 128) >> 8;
1232 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
1233 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-my_col), 4);
1234 }
1235 }
1236 return;
1237 }
1238
1239 /* one-to-one mv scaling */
1240
1241 if(IS_16X16(*mb_type)){
1242 int ref, mv0, mv1;
1243
1244 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1);
1245 if(IS_INTRA(mb_type_col[0])){
1246 ref=mv0=mv1=0;
1247 }else{
1248 const int ref0 = l1ref0[0] >= 0 ? map_col_to_list0[0][l1ref0[0] + ref_offset]
1249 : map_col_to_list0[1][l1ref1[0] + ref_offset];
1250 const int scale = dist_scale_factor[ref0];
1251 const int16_t *mv_col = l1ref0[0] >= 0 ? l1mv0[0] : l1mv1[0];
1252 int mv_l0[2];
1253 mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
1254 mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
1255 ref= ref0;
1256 mv0= pack16to32(mv_l0[0],mv_l0[1]);
1257 mv1= pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
1258 }
1259 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref, 1);
1260 fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, mv0, 4);
1261 fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, mv1, 4);
1262 }else{
1263 for(i8=0; i8<4; i8++){
1264 const int x8 = i8&1;
1265 const int y8 = i8>>1;
1266 int ref0, scale;
1267 const int16_t (*l1mv)[2]= l1mv0;
1268
1269 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1270 continue;
1271 h->sub_mb_type[i8] = sub_mb_type;
1272 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1273 if(IS_INTRA(mb_type_col[0])){
1274 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1275 fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1276 fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1277 continue;
1278 }
1279
1280 ref0 = l1ref0[x8 + y8*b8_stride] + ref_offset;
1281 if(ref0 >= 0)
1282 ref0 = map_col_to_list0[0][ref0];
1283 else{
1284 ref0 = map_col_to_list0[1][l1ref1[x8 + y8*b8_stride] + ref_offset];
1285 l1mv= l1mv1;
1286 }
1287 scale = dist_scale_factor[ref0];
1288
1289 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1290 if(IS_SUB_8X8(sub_mb_type)){
1291 const int16_t *mv_col = l1mv[x8*3 + y8*3*b4_stride];
1292 int mx = (scale * mv_col[0] + 128) >> 8;
1293 int my = (scale * mv_col[1] + 128) >> 8;
1294 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
1295 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-mv_col[1]), 4);
1296 }else
1297 for(i4=0; i4<4; i4++){
1298 const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*b4_stride];
1299 int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]];
1300 mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
1301 mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
1302 *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] =
1303 pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
1304 }
1305 }
1306 }
1307 }
1308 }
1309
1310 static inline void write_back_motion(H264Context *h, int mb_type){
1311 MpegEncContext * const s = &h->s;
1312 const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1313 const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1314 int list;
1315
1316 if(!USES_LIST(mb_type, 0))
1317 fill_rectangle(&s->current_picture.ref_index[0][b8_xy], 2, 2, h->b8_stride, (uint8_t)LIST_NOT_USED, 1);
1318
1319 for(list=0; list<h->list_count; list++){
1320 int y;
1321 if(!USES_LIST(mb_type, list))
1322 continue;
1323
1324 for(y=0; y<4; y++){
1325 *(uint64_t*)s->current_picture.motion_val[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mv_cache[list][scan8[0]+0 + 8*y];
1326 *(uint64_t*)s->current_picture.motion_val[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mv_cache[list][scan8[0]+2 + 8*y];
1327 }
1328 if( h->pps.cabac ) {
1329 if(IS_SKIP(mb_type))
1330 fill_rectangle(h->mvd_table[list][b_xy], 4, 4, h->b_stride, 0, 4);
1331 else
1332 for(y=0; y<4; y++){
1333 *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+0 + 8*y];
1334 *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+2 + 8*y];
1335 }
1336 }
1337
1338 {
1339 int8_t *ref_index = &s->current_picture.ref_index[list][b8_xy];
1340 ref_index[0+0*h->b8_stride]= h->ref_cache[list][scan8[0]];
1341 ref_index[1+0*h->b8_stride]= h->ref_cache[list][scan8[4]];
1342 ref_index[0+1*h->b8_stride]= h->ref_cache[list][scan8[8]];
1343 ref_index[1+1*h->b8_stride]= h->ref_cache[list][scan8[12]];
1344 }
1345 }
1346
1347 if(h->slice_type_nos == FF_B_TYPE && h->pps.cabac){
1348 if(IS_8X8(mb_type)){
1349 uint8_t *direct_table = &h->direct_table[b8_xy];
1350 direct_table[1+0*h->b8_stride] = IS_DIRECT(h->sub_mb_type[1]) ? 1 : 0;
1351 direct_table[0+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[2]) ? 1 : 0;
1352 direct_table[1+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[3]) ? 1 : 0;
1353 }
1354 }
1355 }
1356
1357 /**
1358 * Decodes a network abstraction layer unit.
1359 * @param consumed is the number of bytes used as input
1360 * @param length is the length of the array
1361 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp tailing?
1362 * @returns decoded bytes, might be src+1 if no escapes
1363 */
1364 static const uint8_t *decode_nal(H264Context *h, const uint8_t *src, int *dst_length, int *consumed, int length){
1365 int i, si, di;
1366 uint8_t *dst;
1367 int bufidx;
1368
1369 // src[0]&0x80; //forbidden bit
1370 h->nal_ref_idc= src[0]>>5;
1371 h->nal_unit_type= src[0]&0x1F;
1372
1373 src++; length--;
1374 #if 0
1375 for(i=0; i<length; i++)
1376 printf("%2X ", src[i]);
1377 #endif
1378
1379 #ifdef HAVE_FAST_UNALIGNED
1380 # ifdef HAVE_FAST_64BIT
1381 # define RS 7
1382 for(i=0; i+1<length; i+=9){
1383 if(!((~*(uint64_t*)(src+i) & (*(uint64_t*)(src+i) - 0x0100010001000101ULL)) & 0x8000800080008080ULL))
1384 # else
1385 # define RS 3
1386 for(i=0; i+1<length; i+=5){
1387 if(!((~*(uint32_t*)(src+i) & (*(uint32_t*)(src+i) - 0x01000101U)) & 0x80008080U))
1388 # endif
1389 continue;
1390 if(i>0 && !src[i]) i--;
1391 while(src[i]) i++;
1392 #else
1393 # define RS 0
1394 for(i=0; i+1<length; i+=2){
1395 if(src[i]) continue;
1396 if(i>0 && src[i-1]==0) i--;
1397 #endif
1398 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1399 if(src[i+2]!=3){
1400 /* startcode, so we must be past the end */
1401 length=i;
1402 }
1403 break;
1404 }
1405 i-= RS;
1406 }
1407
1408 if(i>=length-1){ //no escaped 0
1409 *dst_length= length;
1410 *consumed= length+1; //+1 for the header
1411 return src;
1412 }
1413
1414 bufidx = h->nal_unit_type == NAL_DPC ? 1 : 0; // use second escape buffer for inter data
1415 h->rbsp_buffer[bufidx]= av_fast_realloc(h->rbsp_buffer[bufidx], &h->rbsp_buffer_size[bufidx], length+FF_INPUT_BUFFER_PADDING_SIZE);
1416 dst= h->rbsp_buffer[bufidx];
1417
1418 if (dst == NULL){
1419 return NULL;
1420 }
1421
1422 //printf("decoding esc\n");
1423 si=di=0;
1424 while(si<length){
1425 //remove escapes (very rare 1:2^22)
1426 if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1427 if(src[si+2]==3){ //escape
1428 dst[di++]= 0;
1429 dst[di++]= 0;
1430 si+=3;
1431 continue;
1432 }else //next start code
1433 break;
1434 }
1435
1436 dst[di++]= src[si++];
1437 }
1438
1439 memset(dst+di, 0, FF_INPUT_BUFFER_PADDING_SIZE);
1440
1441 *dst_length= di;
1442 *consumed= si + 1;//+1 for the header
1443 //FIXME store exact number of bits in the getbitcontext (it is needed for decoding)
1444 return dst;
1445 }
1446
1447 /**
1448 * identifies the exact end of the bitstream
1449 * @return the length of the trailing, or 0 if damaged
1450 */
1451 static int decode_rbsp_trailing(H264Context *h, const uint8_t *src){
1452 int v= *src;
1453 int r;
1454
1455 tprintf(h->s.avctx, "rbsp trailing %X\n", v);
1456
1457 for(r=1; r<9; r++){
1458 if(v&1) return r;
1459 v>>=1;
1460 }
1461 return 0;
1462 }
1463
1464 /**
1465 * IDCT transforms the 16 dc values and dequantizes them.
1466 * @param qp quantization parameter
1467 */
1468 static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
1469 #define stride 16
1470 int i;
1471 int temp[16]; //FIXME check if this is a good idea
1472 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
1473 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1474
1475 //memset(block, 64, 2*256);
1476 //return;
1477 for(i=0; i<4; i++){
1478 const int offset= y_offset[i];
1479 const int z0= block[offset+stride*0] + block[offset+stride*4];
1480 const int z1= block[offset+stride*0] - block[offset+stride*4];
1481 const int z2= block[offset+stride*1] - block[offset+stride*5];
1482 const int z3= block[offset+stride*1] + block[offset+stride*5];
1483
1484 temp[4*i+0]= z0+z3;
1485 temp[4*i+1]= z1+z2;
1486 temp[4*i+2]= z1-z2;
1487 temp[4*i+3]= z0-z3;
1488 }
1489
1490 for(i=0; i<4; i++){
1491 const int offset= x_offset[i];
1492 const int z0= temp[4*0+i] + temp[4*2+i];
1493 const int z1= temp[4*0+i] - temp[4*2+i];
1494 const int z2= temp[4*1+i] - temp[4*3+i];
1495 const int z3= temp[4*1+i] + temp[4*3+i];
1496
1497 block[stride*0 +offset]= ((((z0 + z3)*qmul + 128 ) >> 8)); //FIXME think about merging this into decode_residual
1498 block[stride*2 +offset]= ((((z1 + z2)*qmul + 128 ) >> 8));
1499 block[stride*8 +offset]= ((((z1 - z2)*qmul + 128 ) >> 8));
1500 block[stride*10+offset]= ((((z0 - z3)*qmul + 128 ) >> 8));
1501 }
1502 }
1503
1504 #if 0
1505 /**
1506 * DCT transforms the 16 dc values.
1507 * @param qp quantization parameter ??? FIXME
1508 */
1509 static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
1510 // const int qmul= dequant_coeff[qp][0];
1511 int i;
1512 int temp[16]; //FIXME check if this is a good idea
1513 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
1514 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1515
1516 for(i=0; i<4; i++){
1517 const int offset= y_offset[i];
1518 const int z0= block[offset+stride*0] + block[offset+stride*4];
1519 const int z1= block[offset+stride*0] - block[offset+stride*4];
1520 const int z2= block[offset+stride*1] - block[offset+stride*5];
1521 const int z3= block[offset+stride*1] + block[offset+stride*5];
1522
1523 temp[4*i+0]= z0+z3;
1524 temp[4*i+1]= z1+z2;
1525 temp[4*i+2]= z1-z2;
1526 temp[4*i+3]= z0-z3;
1527 }
1528
1529 for(i=0; i<4; i++){
1530 const int offset= x_offset[i];
1531 const int z0= temp[4*0+i] + temp[4*2+i];
1532 const int z1= temp[4*0+i] - temp[4*2+i];
1533 const int z2= temp[4*1+i] - temp[4*3+i];
1534 const int z3= temp[4*1+i] + temp[4*3+i];
1535
1536 block[stride*0 +offset]= (z0 + z3)>>1;
1537 block[stride*2 +offset]= (z1 + z2)>>1;
1538 block[stride*8 +offset]= (z1 - z2)>>1;
1539 block[stride*10+offset]= (z0 - z3)>>1;
1540 }
1541 }
1542 #endif
1543
1544 #undef xStride
1545 #undef stride
1546
1547 static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
1548 const int stride= 16*2;
1549 const int xStride= 16;
1550 int a,b,c,d,e;
1551
1552 a= block[stride*0 + xStride*0];
1553 b= block[stride*0 + xStride*1];
1554 c= block[stride*1 + xStride*0];
1555 d= block[stride*1 + xStride*1];
1556
1557 e= a-b;
1558 a= a+b;
1559 b= c-d;
1560 c= c+d;
1561
1562 block[stride*0 + xStride*0]= ((a+c)*qmul) >> 7;
1563 block[stride*0 + xStride*1]= ((e+b)*qmul) >> 7;
1564 block[stride*1 + xStride*0]= ((a-c)*qmul) >> 7;
1565 block[stride*1 + xStride*1]= ((e-b)*qmul) >> 7;
1566 }
1567
1568 #if 0
1569 static void chroma_dc_dct_c(DCTELEM *block){
1570 const int stride= 16*2;
1571 const int xStride= 16;
1572 int a,b,c,d,e;
1573
1574 a= block[stride*0 + xStride*0];
1575 b= block[stride*0 + xStride*1];
1576 c= block[stride*1 + xStride*0];
1577 d= block[stride*1 + xStride*1];
1578
1579 e= a-b;
1580 a= a+b;
1581 b= c-d;
1582 c= c+d;
1583
1584 block[stride*0 + xStride*0]= (a+c);
1585 block[stride*0 + xStride*1]= (e+b);
1586 block[stride*1 + xStride*0]= (a-c);
1587 block[stride*1 + xStride*1]= (e-b);
1588 }
1589 #endif
1590
1591 /**
1592 * gets the chroma qp.
1593 */
1594 static inline int get_chroma_qp(H264Context *h, int t, int qscale){
1595 return h->pps.chroma_qp_table[t][qscale];
1596 }
1597
1598 static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
1599 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1600 int src_x_offset, int src_y_offset,
1601 qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
1602 MpegEncContext * const s = &h->s;
1603 const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
1604 int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
1605 const int luma_xy= (mx&3) + ((my&3)<<2);
1606 uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*h->mb_linesize;
1607 uint8_t * src_cb, * src_cr;
1608 int extra_width= h->emu_edge_width;
1609 int extra_height= h->emu_edge_height;
1610 int emu=0;
1611 const int full_mx= mx>>2;
1612 const int full_my= my>>2;
1613 const int pic_width = 16*s->mb_width;
1614 const int pic_height = 16*s->mb_height >> MB_FIELD;
1615
1616 if(mx&7) extra_width -= 3;
1617 if(my&7) extra_height -= 3;
1618
1619 if( full_mx < 0-extra_width
1620 || full_my < 0-extra_height
1621 || full_mx + 16/*FIXME*/ > pic_width + extra_width
1622 || full_my + 16/*FIXME*/ > pic_height + extra_height){
1623 ff_emulated_edge_mc(s->edge_emu_buffer, src_y - 2 - 2*h->mb_linesize, h->mb_linesize, 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
1624 src_y= s->edge_emu_buffer + 2 + 2*h->mb_linesize;
1625 emu=1;
1626 }
1627
1628 qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); //FIXME try variable height perhaps?
1629 if(!square){
1630 qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
1631 }
1632
1633 if(ENABLE_GRAY && s->flags&CODEC_FLAG_GRAY) return;
1634
1635 if(MB_FIELD){
1636 // chroma offset when predicting from a field of opposite parity
1637 my += 2 * ((s->mb_y & 1) - (pic->reference - 1));
1638 emu |= (my>>3) < 0 || (my>>3) + 8 >= (pic_height>>1);
1639 }
1640 src_cb= pic->data[1] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
1641 src_cr= pic->data[2] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
1642
1643 if(emu){
1644 ff_emulated_edge_mc(s->edge_emu_buffer, src_cb, h->mb_uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
1645 src_cb= s->edge_emu_buffer;
1646 }
1647 chroma_op(dest_cb, src_cb, h->mb_uvlinesize, chroma_height, mx&7, my&7);
1648
1649 if(emu){
1650 ff_emulated_edge_mc(s->edge_emu_buffer, src_cr, h->mb_uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
1651 src_cr= s->edge_emu_buffer;
1652 }
1653 chroma_op(dest_cr, src_cr, h->mb_uvlinesize, chroma_height, mx&7, my&7);
1654 }
1655
1656 static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
1657 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1658 int x_offset, int y_offset,
1659 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
1660 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
1661 int list0, int list1){
1662 MpegEncContext * const s = &h->s;
1663 qpel_mc_func *qpix_op= qpix_put;
1664 h264_chroma_mc_func chroma_op= chroma_put;
1665
1666 dest_y += 2*x_offset + 2*y_offset*h-> mb_linesize;
1667 dest_cb += x_offset + y_offset*h->mb_uvlinesize;
1668 dest_cr += x_offset + y_offset*h->mb_uvlinesize;
1669 x_offset += 8*s->mb_x;
1670 y_offset += 8*(s->mb_y >> MB_FIELD);
1671
1672 if(list0){
1673 Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
1674 mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
1675 dest_y, dest_cb, dest_cr, x_offset, y_offset,
1676 qpix_op, chroma_op);
1677
1678 qpix_op= qpix_avg;
1679 chroma_op= chroma_avg;
1680 }
1681
1682 if(list1){
1683 Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
1684 mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
1685 dest_y, dest_cb, dest_cr, x_offset, y_offset,
1686 qpix_op, chroma_op);
1687 }
1688 }
1689
1690 static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
1691 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1692 int x_offset, int y_offset,
1693 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
1694 h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
1695 h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
1696 int list0, int list1){
1697 MpegEncContext * const s = &h->s;
1698
1699 dest_y += 2*x_offset + 2*y_offset*h-> mb_linesize;
1700 dest_cb += x_offset + y_offset*h->mb_uvlinesize;
1701 dest_cr += x_offset + y_offset*h->mb_uvlinesize;
1702 x_offset += 8*s->mb_x;
1703 y_offset += 8*(s->mb_y >> MB_FIELD);
1704
1705 if(list0 && list1){
1706 /* don't optimize for luma-only case, since B-frames usually
1707 * use implicit weights => chroma too. */
1708 uint8_t *tmp_cb = s->obmc_scratchpad;
1709 uint8_t *tmp_cr = s->obmc_scratchpad + 8;
1710 uint8_t *tmp_y = s->obmc_scratchpad + 8*h->mb_uvlinesize;
1711 int refn0 = h->ref_cache[0][ scan8[n] ];
1712 int refn1 = h->ref_cache[1][ scan8[n] ];
1713
1714 mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
1715 dest_y, dest_cb, dest_cr,
1716 x_offset, y_offset, qpix_put, chroma_put);
1717 mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
1718 tmp_y, tmp_cb, tmp_cr,
1719 x_offset, y_offset, qpix_put, chroma_put);
1720
1721 if(h->use_weight == 2){
1722 int weight0 = h->implicit_weight[refn0][refn1];
1723 int weight1 = 64 - weight0;
1724 luma_weight_avg( dest_y, tmp_y, h-> mb_linesize, 5, weight0, weight1, 0);
1725 chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, 5, weight0, weight1, 0);
1726 chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, 5, weight0, weight1, 0);
1727 }else{
1728 luma_weight_avg(dest_y, tmp_y, h->mb_linesize, h->luma_log2_weight_denom,
1729 h->luma_weight[0][refn0], h->luma_weight[1][refn1],
1730 h->luma_offset[0][refn0] + h->luma_offset[1][refn1]);
1731 chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
1732 h->chroma_weight[0][refn0][0], h->chroma_weight[1][refn1][0],
1733 h->chroma_offset[0][refn0][0] + h->chroma_offset[1][refn1][0]);
1734 chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
1735 h->chroma_weight[0][refn0][1], h->chroma_weight[1][refn1][1],
1736 h->chroma_offset[0][refn0][1] + h->chroma_offset[1][refn1][1]);
1737 }
1738 }else{
1739 int list = list1 ? 1 : 0;
1740 int refn = h->ref_cache[list][ scan8[n] ];
1741 Picture *ref= &h->ref_list[list][refn];
1742 mc_dir_part(h, ref, n, square, chroma_height, delta, list,
1743 dest_y, dest_cb, dest_cr, x_offset, y_offset,
1744 qpix_put, chroma_put);
1745
1746 luma_weight_op(dest_y, h->mb_linesize, h->luma_log2_weight_denom,
1747 h->luma_weight[list][refn], h->luma_offset[list][refn]);
1748 if(h->use_weight_chroma){
1749 chroma_weight_op(dest_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
1750 h->chroma_weight[list][refn][0], h->chroma_offset[list][refn][0]);
1751 chroma_weight_op(dest_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
1752 h->chroma_weight[list][refn][1], h->chroma_offset[list][refn][1]);
1753 }
1754 }
1755 }
1756
1757 static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
1758 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1759 int x_offset, int y_offset,
1760 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
1761 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
1762 h264_weight_func *weight_op, h264_biweight_func *weight_avg,
1763 int list0, int list1){
1764 if((h->use_weight==2 && list0 && list1
1765 && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ] != 32))
1766 || h->use_weight==1)
1767 mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
1768 x_offset, y_offset, qpix_put, chroma_put,
1769 weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
1770 else
1771 mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
1772 x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
1773 }
1774
1775 static inline void prefetch_motion(H264Context *h, int list){
1776 /* fetch pixels for estimated mv 4 macroblocks ahead
1777 * optimized for 64byte cache lines */
1778 MpegEncContext * const s = &h->s;
1779 const int refn = h->ref_cache[list][scan8[0]];
1780 if(refn >= 0){
1781 const int mx= (h->mv_cache[list][scan8[0]][0]>>2) + 16*s->mb_x + 8;
1782 const int my= (h->mv_cache[list][scan8[0]][1]>>2) + 16*s->mb_y;
1783 uint8_t **src= h->ref_list[list][refn].data;
1784 int off= mx + (my + (s->mb_x&3)*4)*h->mb_linesize + 64;
1785 s->dsp.prefetch(src[0]+off, s->linesize, 4);
1786 off= (mx>>1) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + 64;
1787 s->dsp.prefetch(src[1]+off, src[2]-src[1], 2);
1788 }
1789 }
1790
1791 static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1792 qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
1793 qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
1794 h264_weight_func *weight_op, h264_biweight_func *weight_avg){
1795 MpegEncContext * const s = &h->s;
1796 const int mb_xy= h->mb_xy;
1797 const int mb_type= s->current_picture.mb_type[mb_xy];
1798
1799 assert(IS_INTER(mb_type));
1800
1801 prefetch_motion(h, 0);
1802
1803 if(IS_16X16(mb_type)){
1804 mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
1805 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
1806 &weight_op[0], &weight_avg[0],
1807 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1808 }else if(IS_16X8(mb_type)){
1809 mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
1810 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
1811 &weight_op[1], &weight_avg[1],
1812 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1813 mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
1814 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
1815 &weight_op[1], &weight_avg[1],
1816 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
1817 }else if(IS_8X16(mb_type)){
1818 mc_part(h, 0, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 0, 0,
1819 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
1820 &weight_op[2], &weight_avg[2],
1821 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1822 mc_part(h, 4, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 4, 0,
1823 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
1824 &weight_op[2], &weight_avg[2],
1825 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
1826 }else{
1827 int i;
1828
1829 assert(IS_8X8(mb_type));
1830
1831 for(i=0; i<4; i++){
1832 const int sub_mb_type= h->sub_mb_type[i];
1833 const int n= 4*i;
1834 int x_offset= (i&1)<<2;
1835 int y_offset= (i&2)<<1;
1836
1837 if(IS_SUB_8X8(sub_mb_type)){
1838 mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
1839 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
1840 &weight_op[3], &weight_avg[3],
1841 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1842 }else if(IS_SUB_8X4(sub_mb_type)){
1843 mc_part(h, n , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
1844 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
1845 &weight_op[4], &weight_avg[4],
1846 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1847 mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
1848 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
1849 &weight_op[4], &weight_avg[4],
1850 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1851 }else if(IS_SUB_4X8(sub_mb_type)){
1852 mc_part(h, n , 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
1853 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
1854 &weight_op[5], &weight_avg[5],
1855 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1856 mc_part(h, n+1, 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
1857 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
1858 &weight_op[5], &weight_avg[5],
1859 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1860 }else{
1861 int j;
1862 assert(IS_SUB_4X4(sub_mb_type));
1863 for(j=0; j<4; j++){
1864 int sub_x_offset= x_offset + 2*(j&1);
1865 int sub_y_offset= y_offset + (j&2);
1866 mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
1867 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
1868 &weight_op[6], &weight_avg[6],
1869 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1870 }
1871 }
1872 }
1873 }
1874
1875 prefetch_motion(h, 1);
1876 }
1877
1878 static av_cold void decode_init_vlc(void){
1879 static int done = 0;
1880
1881 if (!done) {
1882 int i;
1883 int offset;
1884 done = 1;
1885
1886 chroma_dc_coeff_token_vlc.table = chroma_dc_coeff_token_vlc_table;
1887 chroma_dc_coeff_token_vlc.table_allocated = chroma_dc_coeff_token_vlc_table_size;
1888 init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5,
1889 &chroma_dc_coeff_token_len [0], 1, 1,
1890 &chroma_dc_coeff_token_bits[0], 1, 1,
1891 INIT_VLC_USE_NEW_STATIC);
1892
1893 offset = 0;
1894 for(i=0; i<4; i++){
1895 coeff_token_vlc[i].table = coeff_token_vlc_tables+offset;
1896 coeff_token_vlc[i].table_allocated = coeff_token_vlc_tables_size[i];
1897 init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17,
1898 &coeff_token_len [i][0], 1, 1,
1899 &coeff_token_bits[i][0], 1, 1,
1900 INIT_VLC_USE_NEW_STATIC);
1901 offset += coeff_token_vlc_tables_size[i];
1902 }
1903 /*
1904 * This is a one time safety check to make sure that
1905 * the packed static coeff_token_vlc table sizes
1906 * were initialized correctly.
1907 */
1908 assert(offset == FF_ARRAY_ELEMS(coeff_token_vlc_tables));
1909
1910 for(i=0; i<3; i++){
1911 chroma_dc_total_zeros_vlc[i].table = chroma_dc_total_zeros_vlc_tables[i];
1912 chroma_dc_total_zeros_vlc[i].table_allocated = chroma_dc_total_zeros_vlc_tables_size;
1913 init_vlc(&chroma_dc_total_zeros_vlc[i],
1914 CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
1915 &chroma_dc_total_zeros_len [i][0], 1, 1,
1916 &chroma_dc_total_zeros_bits[i][0], 1, 1,
1917 INIT_VLC_USE_NEW_STATIC);
1918 }
1919 for(i=0; i<15; i++){
1920 total_zeros_vlc[i].table = total_zeros_vlc_tables[i];
1921 total_zeros_vlc[i].table_allocated = total_zeros_vlc_tables_size;
1922 init_vlc(&total_zeros_vlc[i],
1923 TOTAL_ZEROS_VLC_BITS, 16,
1924 &total_zeros_len [i][0], 1, 1,
1925 &total_zeros_bits[i][0], 1, 1,
1926 INIT_VLC_USE_NEW_STATIC);
1927 }
1928
1929 for(i=0; i<6; i++){
1930 run_vlc[i].table = run_vlc_tables[i];
1931 run_vlc[i].table_allocated = run_vlc_tables_size;
1932 init_vlc(&run_vlc[i],
1933 RUN_VLC_BITS, 7,
1934 &run_len [i][0], 1, 1,
1935 &run_bits[i][0], 1, 1,
1936 INIT_VLC_USE_NEW_STATIC);
1937 }
1938 run7_vlc.table = run7_vlc_table,
1939 run7_vlc.table_allocated = run7_vlc_table_size;
1940 init_vlc(&run7_vlc, RUN7_VLC_BITS, 16,
1941 &run_len [6][0], 1, 1,
1942 &run_bits[6][0], 1, 1,
1943 INIT_VLC_USE_NEW_STATIC);
1944 }
1945 }
1946
1947 static void free_tables(H264Context *h){
1948 int i;
1949 H264Context *hx;
1950 av_freep(&h->intra4x4_pred_mode);
1951 av_freep(&h->chroma_pred_mode_table);
1952 av_freep(&h->cbp_table);
1953 av_freep(&h->mvd_table[0]);
1954 av_freep(&h->mvd_table[1]);
1955 av_freep(&h->direct_table);
1956 av_freep(&h->non_zero_count);
1957 av_freep(&h->slice_table_base);
1958 h->slice_table= NULL;
1959
1960 av_freep(&h->mb2b_xy);
1961 av_freep(&h->mb2b8_xy);
1962
1963 for(i = 0; i < h->s.avctx->thread_count; i++) {
1964 hx = h->thread_context[i];
1965 if(!hx) continue;
1966 av_freep(&hx->top_borders[1]);
1967 av_freep(&hx->top_borders[0]);
1968 av_freep(&hx->s.obmc_scratchpad);
1969 }
1970 }
1971
1972 static void init_dequant8_coeff_table(H264Context *h){
1973 int i,q,x;
1974 const int transpose = (h->s.dsp.h264_idct8_add != ff_h264_idct8_add_c); //FIXME ugly
1975 h->dequant8_coeff[0] = h->dequant8_buffer[0];
1976 h->dequant8_coeff[1] = h->dequant8_buffer[1];
1977
1978 for(i=0; i<2; i++ ){
1979 if(i && !memcmp(h->pps.scaling_matrix8[0], h->pps.scaling_matrix8[1], 64*sizeof(uint8_t))){
1980 h->dequant8_coeff[1] = h->dequant8_buffer[0];
1981 break;
1982 }
1983
1984 for(q=0; q<52; q++){
1985 int shift = div6[q];
1986 int idx = rem6[q];
1987 for(x=0; x<64; x++)
1988 h->dequant8_coeff[i][q][transpose ? (x>>3)|((x&7)<<3) : x] =
1989 ((uint32_t)dequant8_coeff_init[idx][ dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] *
1990 h->pps.scaling_matrix8[i][x]) << shift;
1991 }
1992 }
1993 }
1994
1995 static void init_dequant4_coeff_table(H264Context *h){
1996 int i,j,q,x;
1997 const int transpose = (h->s.dsp.h264_idct_add != ff_h264_idct_add_c); //FIXME ugly
1998 for(i=0; i<6; i++ ){
1999 h->dequant4_coeff[i] = h->dequant4_buffer[i];
2000 for(j=0; j<i; j++){
2001 if(!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i], 16*sizeof(uint8_t))){
2002 h->dequant4_coeff[i] = h->dequant4_buffer[j];
2003 break;
2004 }
2005 }
2006 if(j<i)
2007 continue;
2008
2009 for(q=0; q<52; q++){
2010 int shift = div6[q] + 2;
2011 int idx = rem6[q];
2012 for(x=0; x<16; x++)
2013 h->dequant4_coeff[i][q][transpose ? (x>>2)|((x<<2)&0xF) : x] =
2014 ((uint32_t)dequant4_coeff_init[idx][(x&1) + ((x>>2)&1)] *
2015 h->pps.scaling_matrix4[i][x]) << shift;
2016 }
2017 }
2018 }
2019
2020 static void init_dequant_tables(H264Context *h){
2021 int i,x;
2022 init_dequant4_coeff_table(h);
2023 if(h->pps.transform_8x8_mode)
2024 init_dequant8_coeff_table(h);
2025 if(h->sps.transform_bypass){
2026 for(i=0; i<6; i++)
2027 for(x=0; x<16; x++)
2028 h->dequant4_coeff[i][0][x] = 1<<6;
2029 if(h->pps.transform_8x8_mode)
2030 for(i=0; i<2; i++)
2031 for(x=0; x<64; x++)
2032 h->dequant8_coeff[i][0][x] = 1<<6;
2033 }
2034 }
2035
2036
2037 /**
2038 * allocates tables.
2039 * needs width/height
2040 */
2041 static int alloc_tables(H264Context *h){
2042 MpegEncContext * const s = &h->s;
2043 const int big_mb_num= s->mb_stride * (s->mb_height+1);
2044 int x,y;
2045
2046 CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8 * sizeof(uint8_t))
2047
2048 CHECKED_ALLOCZ(h->non_zero_count , big_mb_num * 16 * sizeof(uint8_t))
2049 CHECKED_ALLOCZ(h->slice_table_base , (big_mb_num+s->mb_stride) * sizeof(*h->slice_table_base))
2050 CHECKED_ALLOCZ(h->cbp_table, big_mb_num * sizeof(uint16_t))
2051
2052 CHECKED_ALLOCZ(h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t))
2053 CHECKED_ALLOCZ(h->mvd_table[0], 32*big_mb_num * sizeof(uint16_t));
2054 CHECKED_ALLOCZ(h->mvd_table[1], 32*big_mb_num * sizeof(uint16_t));
2055 CHECKED_ALLOCZ(h->direct_table, 32*big_mb_num * sizeof(uint8_t));
2056
2057 memset(h->slice_table_base, -1, (big_mb_num+s->mb_stride) * sizeof(*h->slice_table_base));
2058 h->slice_table= h->slice_table_base + s->mb_stride*2 + 1;
2059
2060 CHECKED_ALLOCZ(h->mb2b_xy , big_mb_num * sizeof(uint32_t));
2061 CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint32_t));
2062 for(y=0; y<s->mb_height; y++){
2063 for(x=0; x<s->mb_width; x++){
2064 const int mb_xy= x + y*s->mb_stride;
2065 const int b_xy = 4*x + 4*y*h->b_stride;
2066 const int b8_xy= 2*x + 2*y*h->b8_stride;
2067
2068 h->mb2b_xy [mb_xy]= b_xy;
2069 h->mb2b8_xy[mb_xy]= b8_xy;
2070 }
2071 }
2072
2073 s->obmc_scratchpad = NULL;
2074
2075 if(!h->dequant4_coeff[0])
2076 init_dequant_tables(h);
2077
2078 return 0;
2079 fail:
2080 free_tables(h);
2081 return -1;
2082 }
2083
2084 /**
2085 * Mimic alloc_tables(), but for every context thread.
2086 */
2087 static void clone_tables(H264Context *dst, H264Context *src){
2088 dst->intra4x4_pred_mode = src->intra4x4_pred_mode;
2089 dst->non_zero_count = src->non_zero_count;
2090 dst->slice_table = src->slice_table;
2091 dst->cbp_table = src->cbp_table;
2092 dst->mb2b_xy = src->mb2b_xy;
2093 dst->mb2b8_xy = src->mb2b8_xy;
2094 dst->chroma_pred_mode_table = src->chroma_pred_mode_table;
2095 dst->mvd_table[0] = src->mvd_table[0];
2096 dst->mvd_table[1] = src->mvd_table[1];
2097 dst->direct_table = src->direct_table;
2098
2099 dst->s.obmc_scratchpad = NULL;
2100 ff_h264_pred_init(&dst->hpc, src->s.codec_id);
2101 }
2102
2103 /**
2104 * Init context
2105 * Allocate buffers which are not shared amongst multiple threads.
2106 */
2107 static int context_init(H264Context *h){
2108 CHECKED_ALLOCZ(h->top_borders[0], h->s.mb_width * (16+8+8) * sizeof(uint8_t))
2109 CHECKED_ALLOCZ(h->top_borders[1], h->s.mb_width * (16+8+8) * sizeof(uint8_t))
2110
2111 return 0;
2112 fail:
2113 return -1; // free_tables will clean up for us
2114 }
2115
2116 static av_cold void common_init(H264Context *h){
2117 MpegEncContext * const s = &h->s;
2118
2119 s->width = s->avctx->width;
2120 s->height = s->avctx->height;
2121 s->codec_id= s->avctx->codec->id;
2122
2123 ff_h264_pred_init(&h->hpc, s->codec_id);
2124
2125 h->dequant_coeff_pps= -1;
2126 s->unrestricted_mv=1;
2127 s->decode=1; //FIXME
2128
2129 dsputil_init(&s->dsp, s->avctx); // needed so that idct permutation is known early
2130
2131 memset(h->pps.scaling_matrix4, 16, 6*16*sizeof(uint8_t));
2132 memset(h->pps.scaling_matrix8, 16, 2*64*sizeof(uint8_t));
2133 }
2134
2135 static av_cold int decode_init(AVCodecContext *avctx){
2136 H264Context *h= avctx->priv_data;
2137 MpegEncContext * const s = &h->s;
2138
2139 MPV_decode_defaults(s);
2140
2141 s->avctx = avctx;
2142 common_init(h);
2143
2144 s->out_format = FMT_H264;
2145 s->workaround_bugs= avctx->workaround_bugs;
2146
2147 // set defaults
2148 // s->decode_mb= ff_h263_decode_mb;
2149 s->quarter_sample = 1;
2150 s->low_delay= 1;
2151
2152 if(avctx->codec_id == CODEC_ID_SVQ3)
2153 avctx->pix_fmt= PIX_FMT_YUVJ420P;
2154 else
2155 avctx->pix_fmt= PIX_FMT_YUV420P;
2156
2157 decode_init_vlc();
2158
2159 if(avctx->extradata_size > 0 && avctx->extradata &&
2160 *(char *)avctx->extradata == 1){
2161 h->is_avc = 1;
2162 h->got_avcC = 0;
2163 } else {
2164 h->is_avc = 0;
2165 }
2166
2167 h->thread_context[0] = h;
2168 h->outputed_poc = INT_MIN;
2169 h->prev_poc_msb= 1<<16;
2170 return 0;
2171 }
2172
2173 static int frame_start(H264Context *h){
2174 MpegEncContext * const s = &h->s;
2175 int i;
2176
2177 if(MPV_frame_start(s, s->avctx) < 0)
2178 return -1;
2179 ff_er_frame_start(s);
2180 /*
2181 * MPV_frame_start uses pict_type to derive key_frame.
2182 * This is incorrect for H.264; IDR markings must be used.
2183 * Zero here; IDR markings per slice in frame or fields are ORed in later.
2184 * See decode_nal_units().
2185 */
2186 s->current_picture_ptr->key_frame= 0;
2187
2188 assert(s->linesize && s->uvlinesize);
2189
2190 for(i=0; i<16; i++){
2191 h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
2192 h->block_offset[24+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
2193 }
2194 for(i=0; i<4; i++){
2195 h->block_offset[16+i]=
2196 h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2197 h->block_offset[24+16+i]=
2198 h->block_offset[24+20+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2199 }
2200
2201 /* can't be in alloc_tables because linesize isn't known there.
2202 * FIXME: redo bipred weight to not require extra buffer? */
2203 for(i = 0; i < s->avctx->thread_count; i++)
2204 if(!h->thread_context[i]->s.obmc_scratchpad)
2205 h->thread_context[i]->s.obmc_scratchpad = av_malloc(16*2*s->linesize + 8*2*s->uvlinesize);
2206
2207 /* some macroblocks will be accessed before they're available */
2208 if(FRAME_MBAFF || s->avctx->thread_count > 1)
2209 memset(h->slice_table, -1, (s->mb_height*s->mb_stride-1) * sizeof(*h->slice_table));
2210
2211 // s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
2212
2213 // We mark the current picture as non-reference after allocating it, so
2214 // that if we break out due to an error it can be released automatically
2215 // in the next MPV_frame_start().
2216 // SVQ3 as well as most other codecs have only last/next/current and thus
2217 // get released even with set reference, besides SVQ3 and others do not
2218 // mark frames as reference later "naturally".
2219 if(s->codec_id != CODEC_ID_SVQ3)
2220 s->current_picture_ptr->reference= 0;
2221
2222 s->current_picture_ptr->field_poc[0]=
2223 s->current_picture_ptr->field_poc[1]= INT_MAX;
2224 assert(s->current_picture_ptr->long_ref==0);
2225
2226 return 0;
2227 }
2228
2229 static inline void backup_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize, int simple){
2230 MpegEncContext * const s = &h->s;
2231 int i;
2232 int step = 1;
2233 int offset = 1;
2234 int uvoffset= 1;
2235 int top_idx = 1;
2236 int skiplast= 0;
2237
2238 src_y -= linesize;
2239 src_cb -= uvlinesize;
2240 src_cr -= uvlinesize;
2241
2242 if(!simple && FRAME_MBAFF){
2243 if(s->mb_y&1){
2244 offset = MB_MBAFF ? 1 : 17;
2245 uvoffset= MB_MBAFF ? 1 : 9;
2246 if(!MB_MBAFF){
2247 *(uint64_t*)(h->top_borders[0][s->mb_x]+ 0)= *(uint64_t*)(src_y + 15*linesize);
2248 *(uint64_t*)(h->top_borders[0][s->mb_x]+ 8)= *(uint64_t*)(src_y +8+15*linesize);
2249 if(simple || !ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2250 *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+7*uvlinesize);
2251 *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+7*uvlinesize);
2252 }
2253 }
2254 }else{
2255 if(!MB_MBAFF){
2256 h->left_border[0]= h->top_borders[0][s->mb_x][15];
2257 if(simple || !ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2258 h->left_border[34 ]= h->top_borders[0][s->mb_x][16+7 ];
2259 h->left_border[34+18]= h->top_borders[0][s->mb_x][16+8+7];
2260 }
2261 skiplast= 1;
2262 }
2263 offset =
2264 uvoffset=
2265 top_idx = MB_MBAFF ? 0 : 1;
2266 }
2267 step= MB_MBAFF ? 2 : 1;
2268 }
2269
2270 // There are two lines saved, the line above the the top macroblock of a pair,
2271 // and the line above the bottom macroblock
2272 h->left_border[offset]= h->top_borders[top_idx][s->mb_x][15];
2273 for(i=1; i<17 - skiplast; i++){
2274 h->left_border[offset+i*step]= src_y[15+i* linesize];
2275 }
2276
2277 *(uint64_t*)(h->top_borders[top_idx][s->mb_x]+0)= *(uint64_t*)(src_y + 16*linesize);
2278 *(uint64_t*)(h->top_borders[top_idx][s->mb_x]+8)= *(uint64_t*)(src_y +8+16*linesize);
2279
2280 if(simple || !ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2281 h->left_border[uvoffset+34 ]= h->top_borders[top_idx][s->mb_x][16+7];
2282 h->left_border[uvoffset+34+18]= h->top_borders[top_idx][s->mb_x][24+7];
2283 for(i=1; i<9 - skiplast; i++){
2284 h->left_border[uvoffset+34 +i*step]= src_cb[7+i*uvlinesize];
2285 h->left_border[uvoffset+34+18+i*step]= src_cr[7+i*uvlinesize];
2286 }
2287 *(uint64_t*)(h->top_borders[top_idx][s->mb_x]+16)= *(uint64_t*)(src_cb+8*uvlinesize);
2288 *(uint64_t*)(h->top_borders[top_idx][s->mb_x]+24)= *(uint64_t*)(src_cr+8*uvlinesize);
2289 }
2290 }
2291
2292 static inline void xchg_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize, int xchg, int simple){
2293 MpegEncContext * const s = &h->s;
2294 int temp8, i;
2295 uint64_t temp64;
2296 int deblock_left;
2297 int deblock_top;
2298 int mb_xy;
2299 int step = 1;
2300 int offset = 1;
2301 int uvoffset= 1;
2302 int top_idx = 1;
2303
2304 if(!simple && FRAME_MBAFF){
2305 if(s->mb_y&1){
2306 offset = MB_MBAFF ? 1 : 17;
2307 uvoffset= MB_MBAFF ? 1 : 9;
2308 }else{
2309 offset =
2310 uvoffset=
2311 top_idx = MB_MBAFF ? 0 : 1;
2312 }
2313 step= MB_MBAFF ? 2 : 1;
2314 }
2315
2316 if(h->deblocking_filter == 2) {
2317 mb_xy = h->mb_xy;
2318 deblock_left = h->slice_table[mb_xy] == h->slice_table[mb_xy - 1];
2319 deblock_top = h->slice_table[mb_xy] == h->slice_table[h->top_mb_xy];
2320 } else {
2321 deblock_left = (s->mb_x > 0);
2322 deblock_top = (s->mb_y > !!MB_FIELD);
2323 }
2324
2325 src_y -= linesize + 1;
2326 src_cb -= uvlinesize + 1;
2327 src_cr -= uvlinesize + 1;
2328
2329 #define XCHG(a,b,t,xchg)\
2330 t= a;\
2331 if(xchg)\
2332 a= b;\
2333 b= t;
2334
2335 if(deblock_left){
2336 for(i = !deblock_top; i<16; i++){
2337 XCHG(h->left_border[offset+i*step], src_y [i* linesize], temp8, xchg);
2338 }
2339 XCHG(h->left_border[offset+i*step], src_y [i* linesize], temp8, 1);
2340 }
2341
2342 if(deblock_top){
2343 XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
2344 XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
2345 if(s->mb_x+1 < s->mb_width){
2346 XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x+1]), *(uint64_t*)(src_y +17), temp64, 1);
2347 }
2348 }
2349
2350 if(simple || !ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2351 if(deblock_left){
2352 for(i = !deblock_top; i<8; i++){
2353 XCHG(h->left_border[uvoffset+34 +i*step], src_cb[i*uvlinesize], temp8, xchg);
2354 XCHG(h->left_border[uvoffset+34+18+i*step], src_cr[i*uvlinesize], temp8, xchg);
2355 }
2356 XCHG(h->left_border[uvoffset+34 +i*step], src_cb[i*uvlinesize], temp8, 1);
2357 XCHG(h->left_border[uvoffset+34+18+i*step], src_cr[i*uvlinesize], temp8, 1);
2358 }
2359 if(deblock_top){
2360 XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
2361 XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
2362 }
2363 }
2364 }
2365
2366 static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
2367 MpegEncContext * const s = &h->s;
2368 const int mb_x= s->mb_x;
2369 const int mb_y= s->mb_y;
2370 const int mb_xy= h->mb_xy;
2371 const int mb_type= s->current_picture.mb_type[mb_xy];
2372 uint8_t *dest_y, *dest_cb, *dest_cr;
2373 int linesize, uvlinesize /*dct_offset*/;
2374 int i;
2375 int *block_offset = &h->block_offset[0];
2376 const int transform_bypass = !simple && (s->qscale == 0 && h->sps.transform_bypass);
2377 const int is_h264 = simple || s->codec_id == CODEC_ID_H264;
2378 void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
2379 void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
2380
2381 dest_y = s->current_picture.data[0] + (mb_x + mb_y * s->linesize ) * 16;
2382 dest_cb = s->current_picture.data[1] + (mb_x + mb_y * s->uvlinesize) * 8;
2383 dest_cr = s->current_picture.data[2] + (mb_x + mb_y * s->uvlinesize) * 8;
2384
2385 s->dsp.prefetch(dest_y + (s->mb_x&3)*4*s->linesize + 64, s->linesize, 4);
2386 s->dsp.prefetch(dest_cb + (s->mb_x&7)*s->uvlinesize + 64, dest_cr - dest_cb, 2);
2387
2388 if (!simple && MB_FIELD) {
2389 linesize = h->mb_linesize = s->linesize * 2;
2390 uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
2391 block_offset = &h->block_offset[24];
2392 if(mb_y&1){ //FIXME move out of this function?
2393 dest_y -= s->linesize*15;
2394 dest_cb-= s->uvlinesize*7;
2395 dest_cr-= s->uvlinesize*7;
2396 }
2397 if(FRAME_MBAFF) {
2398 int list;
2399 for(list=0; list<h->list_count; list++){
2400 if(!USES_LIST(mb_type, list))
2401 continue;
2402 if(IS_16X16(mb_type)){
2403 int8_t *ref = &h->ref_cache[list][scan8[0]];
2404 fill_rectangle(ref, 4, 4, 8, (16+*ref)^(s->mb_y&1), 1);
2405 }else{
2406 for(i=0; i<16; i+=4){
2407 int ref = h->ref_cache[list][scan8[i]];
2408 if(ref >= 0)
2409 fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, (16+ref)^(s->mb_y&1), 1);
2410 }
2411 }
2412 }
2413 }
2414 } else {
2415 linesize = h->mb_linesize = s->linesize;
2416 uvlinesize = h->mb_uvlinesize = s->uvlinesize;
2417 // dct_offset = s->linesize * 16;
2418 }
2419
2420 if (!simple && IS_INTRA_PCM(mb_type)) {
2421 for (i=0; i<16; i++) {
2422 memcpy(dest_y + i* linesize, h->mb + i*8, 16);
2423 }
2424 for (i=0; i<8; i++) {
2425 memcpy(dest_cb+ i*uvlinesize, h->mb + 128 + i*4, 8);
2426 memcpy(dest_cr+ i*uvlinesize, h->mb + 160 + i*4, 8);
2427 }
2428 } else {
2429 if(IS_INTRA(mb_type)){
2430 if(h->deblocking_filter)
2431 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1, simple);
2432
2433 if(simple || !ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2434 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
2435 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
2436 }
2437
2438 if(IS_INTRA4x4(mb_type)){
2439 if(simple || !s->encoding){
2440 if(IS_8x8DCT(mb_type)){
2441 if(transform_bypass){
2442 idct_dc_add =
2443 idct_add = s->dsp.add_pixels8;
2444 }else{
2445 idct_dc_add = s->dsp.h264_idct8_dc_add;
2446 idct_add = s->dsp.h264_idct8_add;
2447 }
2448 for(i=0; i<16; i+=4){
2449 uint8_t * const ptr= dest_y + block_offset[i];
2450 const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
2451 if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
2452 h->hpc.pred8x8l_add[dir](ptr, h->mb + i*16, linesize);
2453 }else{
2454 const int nnz = h->non_zero_count_cache[ scan8[i] ];
2455 h->hpc.pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
2456 (h->topright_samples_available<<i)&0x4000, linesize);
2457 if(nnz){
2458 if(nnz == 1 && h->mb[i*16])
2459 idct_dc_add(ptr, h->mb + i*16, linesize);
2460 else
2461 idct_add (ptr, h->mb + i*16, linesize);
2462 }
2463 }
2464 }
2465 }else{
2466 if(transform_bypass){
2467 idct_dc_add =
2468 idct_add = s->dsp.add_pixels4;
2469 }else{
2470 idct_dc_add = s->dsp.h264_idct_dc_add;
2471 idct_add = s->dsp.h264_idct_add;
2472 }
2473 for(i=0; i<16; i++){
2474 uint8_t * const ptr= dest_y + block_offset[i];
2475 const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
2476
2477 if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
2478 h->hpc.pred4x4_add[dir](ptr, h->mb + i*16, linesize);
2479 }else{
2480 uint8_t *topright;
2481 int nnz, tr;
2482 if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
2483 const int topright_avail= (h->topright_samples_available<<i)&0x8000;
2484 assert(mb_y || linesize <= block_offset[i]);
2485 if(!topright_avail){
2486 tr= ptr[3 - linesize]*0x01010101;
2487 topright= (uint8_t*) &tr;
2488 }else
2489 topright= ptr + 4 - linesize;
2490 }else
2491 topright= NULL;
2492
2493 h->hpc.pred4x4[ dir ](ptr, topright, linesize);
2494 nnz = h->non_zero_count_cache[ scan8[i] ];
2495 if(nnz){
2496 if(is_h264){
2497 if(nnz == 1 && h->mb[i*16])
2498 idct_dc_add(ptr, h->mb + i*16, linesize);
2499 else
2500 idct_add (ptr, h->mb + i*16, linesize);
2501 }else
2502 svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
2503 }
2504 }
2505 }
2506 }
2507 }
2508 }else{
2509 h->hpc.pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
2510 if(is_h264){
2511 if(!transform_bypass)
2512 h264_luma_dc_dequant_idct_c(h->mb, s->qscale, h->dequant4_coeff[0][s->qscale][0]);
2513 }else
2514 svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
2515 }
2516 if(h->deblocking_filter)
2517 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0, simple);
2518 }else if(is_h264){
2519 hl_motion(h, dest_y, dest_cb, dest_cr,
2520 s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
2521 s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
2522 s->dsp.weight_h264_pixels_tab, s->dsp.biweight_h264_pixels_tab);
2523 }
2524
2525
2526 if(!IS_INTRA4x4(mb_type)){
2527 if(is_h264){
2528 if(IS_INTRA16x16(mb_type)){
2529 if(transform_bypass){
2530 if(h->sps.profile_idc==244 && (h->intra16x16_pred_mode==VERT_PRED8x8 || h->intra16x16_pred_mode==HOR_PRED8x8)){
2531 h->hpc.pred16x16_add[h->intra16x16_pred_mode](dest_y, block_offset, h->mb, linesize);
2532 }else{
2533 for(i=0; i<16; i++){
2534 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16])
2535 s->dsp.add_pixels4(dest_y + block_offset[i], h->mb + i*16, linesize);
2536 }
2537 }
2538 }else{
2539 s->dsp.h264_idct_add16intra(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
2540 }
2541 }else if(h->cbp&