cosmetics: comment typo fixes
[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
39 //#undef NDEBUG
40 #include <assert.h>
41
42 /**
43 * Value of Picture.reference when Picture is not a reference picture, but
44 * is held for delayed output.
45 */
46 #define DELAYED_PIC_REF 4
47
48 static VLC coeff_token_vlc[4];
49 static VLC chroma_dc_coeff_token_vlc;
50
51 static VLC total_zeros_vlc[15];
52 static VLC chroma_dc_total_zeros_vlc[3];
53
54 static VLC run_vlc[6];
55 static VLC run7_vlc;
56
57 static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
58 static void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
59 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);
60 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);
61
62 static av_always_inline uint32_t pack16to32(int a, int b){
63 #ifdef WORDS_BIGENDIAN
64 return (b&0xFFFF) + (a<<16);
65 #else
66 return (a&0xFFFF) + (b<<16);
67 #endif
68 }
69
70 const uint8_t ff_rem6[52]={
71 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,
72 };
73
74 const uint8_t ff_div6[52]={
75 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,
76 };
77
78
79 static void fill_caches(H264Context *h, int mb_type, int for_deblock){
80 MpegEncContext * const s = &h->s;
81 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
82 int topleft_xy, top_xy, topright_xy, left_xy[2];
83 int topleft_type, top_type, topright_type, left_type[2];
84 int left_block[8];
85 int i;
86
87 top_xy = mb_xy - (s->mb_stride << FIELD_PICTURE);
88
89 //FIXME deblocking could skip the intra and nnz parts.
90 if(for_deblock && (h->slice_num == 1 || h->slice_table[mb_xy] == h->slice_table[top_xy]) && !FRAME_MBAFF)
91 return;
92
93 //wow what a mess, why didn't they simplify the interlacing&intra stuff, i can't imagine that these complex rules are worth it
94
95 topleft_xy = top_xy - 1;
96 topright_xy= top_xy + 1;
97 left_xy[1] = left_xy[0] = mb_xy-1;
98 left_block[0]= 0;
99 left_block[1]= 1;
100 left_block[2]= 2;
101 left_block[3]= 3;
102 left_block[4]= 7;
103 left_block[5]= 10;
104 left_block[6]= 8;
105 left_block[7]= 11;
106 if(FRAME_MBAFF){
107 const int pair_xy = s->mb_x + (s->mb_y & ~1)*s->mb_stride;
108 const int top_pair_xy = pair_xy - s->mb_stride;
109 const int topleft_pair_xy = top_pair_xy - 1;
110 const int topright_pair_xy = top_pair_xy + 1;
111 const int topleft_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[topleft_pair_xy]);
112 const int top_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[top_pair_xy]);
113 const int topright_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[topright_pair_xy]);
114 const int left_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[pair_xy-1]);
115 const int curr_mb_frame_flag = !IS_INTERLACED(mb_type);
116 const int bottom = (s->mb_y & 1);
117 tprintf(s->avctx, "fill_caches: curr_mb_frame_flag:%d, left_mb_frame_flag:%d, topleft_mb_frame_flag:%d, top_mb_frame_flag:%d, topright_mb_frame_flag:%d\n", curr_mb_frame_flag, left_mb_frame_flag, topleft_mb_frame_flag, top_mb_frame_flag, topright_mb_frame_flag);
118 if (bottom
119 ? !curr_mb_frame_flag // bottom macroblock
120 : (!curr_mb_frame_flag && !top_mb_frame_flag) // top macroblock
121 ) {
122 top_xy -= s->mb_stride;
123 }
124 if (bottom
125 ? !curr_mb_frame_flag // bottom macroblock
126 : (!curr_mb_frame_flag && !topleft_mb_frame_flag) // top macroblock
127 ) {
128 topleft_xy -= s->mb_stride;
129 }
130 if (bottom
131 ? !curr_mb_frame_flag // bottom macroblock
132 : (!curr_mb_frame_flag && !topright_mb_frame_flag) // top macroblock
133 ) {
134 topright_xy -= s->mb_stride;
135 }
136 if (left_mb_frame_flag != curr_mb_frame_flag) {
137 left_xy[1] = left_xy[0] = pair_xy - 1;
138 if (curr_mb_frame_flag) {
139 if (bottom) {
140 left_block[0]= 2;
141 left_block[1]= 2;
142 left_block[2]= 3;
143 left_block[3]= 3;
144 left_block[4]= 8;
145 left_block[5]= 11;
146 left_block[6]= 8;
147 left_block[7]= 11;
148 } else {
149 left_block[0]= 0;
150 left_block[1]= 0;
151 left_block[2]= 1;
152 left_block[3]= 1;
153 left_block[4]= 7;
154 left_block[5]= 10;
155 left_block[6]= 7;
156 left_block[7]= 10;
157 }
158 } else {
159 left_xy[1] += s->mb_stride;
160 //left_block[0]= 0;
161 left_block[1]= 2;
162 left_block[2]= 0;
163 left_block[3]= 2;
164 //left_block[4]= 7;
165 left_block[5]= 10;
166 left_block[6]= 7;
167 left_block[7]= 10;
168 }
169 }
170 }
171
172 h->top_mb_xy = top_xy;
173 h->left_mb_xy[0] = left_xy[0];
174 h->left_mb_xy[1] = left_xy[1];
175 if(for_deblock){
176 topleft_type = 0;
177 topright_type = 0;
178 top_type = h->slice_table[top_xy ] < 255 ? s->current_picture.mb_type[top_xy] : 0;
179 left_type[0] = h->slice_table[left_xy[0] ] < 255 ? s->current_picture.mb_type[left_xy[0]] : 0;
180 left_type[1] = h->slice_table[left_xy[1] ] < 255 ? s->current_picture.mb_type[left_xy[1]] : 0;
181
182 if(FRAME_MBAFF && !IS_INTRA(mb_type)){
183 int list;
184 int v = *(uint16_t*)&h->non_zero_count[mb_xy][14];
185 for(i=0; i<16; i++)
186 h->non_zero_count_cache[scan8[i]] = (v>>i)&1;
187 for(list=0; list<h->list_count; list++){
188 if(USES_LIST(mb_type,list)){
189 uint32_t *src = (uint32_t*)s->current_picture.motion_val[list][h->mb2b_xy[mb_xy]];
190 uint32_t *dst = (uint32_t*)h->mv_cache[list][scan8[0]];
191 int8_t *ref = &s->current_picture.ref_index[list][h->mb2b8_xy[mb_xy]];
192 for(i=0; i<4; i++, dst+=8, src+=h->b_stride){
193 dst[0] = src[0];
194 dst[1] = src[1];
195 dst[2] = src[2];
196 dst[3] = src[3];
197 }
198 *(uint32_t*)&h->ref_cache[list][scan8[ 0]] =
199 *(uint32_t*)&h->ref_cache[list][scan8[ 2]] = pack16to32(ref[0],ref[1])*0x0101;
200 ref += h->b8_stride;
201 *(uint32_t*)&h->ref_cache[list][scan8[ 8]] =
202 *(uint32_t*)&h->ref_cache[list][scan8[10]] = pack16to32(ref[0],ref[1])*0x0101;
203 }else{
204 fill_rectangle(&h-> mv_cache[list][scan8[ 0]], 4, 4, 8, 0, 4);
205 fill_rectangle(&h->ref_cache[list][scan8[ 0]], 4, 4, 8, (uint8_t)LIST_NOT_USED, 1);
206 }
207 }
208 }
209 }else{
210 topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
211 top_type = h->slice_table[top_xy ] == h->slice_num ? s->current_picture.mb_type[top_xy] : 0;
212 topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
213 left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
214 left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
215 }
216
217 if(IS_INTRA(mb_type)){
218 h->topleft_samples_available=
219 h->top_samples_available=
220 h->left_samples_available= 0xFFFF;
221 h->topright_samples_available= 0xEEEA;
222
223 if(!IS_INTRA(top_type) && (top_type==0 || h->pps.constrained_intra_pred)){
224 h->topleft_samples_available= 0xB3FF;
225 h->top_samples_available= 0x33FF;
226 h->topright_samples_available= 0x26EA;
227 }
228 for(i=0; i<2; i++){
229 if(!IS_INTRA(left_type[i]) && (left_type[i]==0 || h->pps.constrained_intra_pred)){
230 h->topleft_samples_available&= 0xDF5F;
231 h->left_samples_available&= 0x5F5F;
232 }
233 }
234
235 if(!IS_INTRA(topleft_type) && (topleft_type==0 || h->pps.constrained_intra_pred))
236 h->topleft_samples_available&= 0x7FFF;
237
238 if(!IS_INTRA(topright_type) && (topright_type==0 || h->pps.constrained_intra_pred))
239 h->topright_samples_available&= 0xFBFF;
240
241 if(IS_INTRA4x4(mb_type)){
242 if(IS_INTRA4x4(top_type)){
243 h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
244 h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
245 h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
246 h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
247 }else{
248 int pred;
249 if(!top_type || (IS_INTER(top_type) && h->pps.constrained_intra_pred))
250 pred= -1;
251 else{
252 pred= 2;
253 }
254 h->intra4x4_pred_mode_cache[4+8*0]=
255 h->intra4x4_pred_mode_cache[5+8*0]=
256 h->intra4x4_pred_mode_cache[6+8*0]=
257 h->intra4x4_pred_mode_cache[7+8*0]= pred;
258 }
259 for(i=0; i<2; i++){
260 if(IS_INTRA4x4(left_type[i])){
261 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
262 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
263 }else{
264 int pred;
265 if(!left_type[i] || (IS_INTER(left_type[i]) && h->pps.constrained_intra_pred))
266 pred= -1;
267 else{
268 pred= 2;
269 }
270 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
271 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
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 (lets 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 + 3*h->b_stride;
407 const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + 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 == 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 MpegEncContext * const s = &h->s;
545 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
546
547 h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
548 h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
549 h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
550 h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
551 h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
552 h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
553 h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
554 }
555
556 /**
557 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
558 */
559 static inline int check_intra4x4_pred_mode(H264Context *h){
560 MpegEncContext * const s = &h->s;
561 static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
562 static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
563 int i;
564
565 if(!(h->top_samples_available&0x8000)){
566 for(i=0; i<4; i++){
567 int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
568 if(status<0){
569 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);
570 return -1;
571 } else if(status){
572 h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
573 }
574 }
575 }
576
577 if(!(h->left_samples_available&0x8000)){
578 for(i=0; i<4; i++){
579 int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
580 if(status<0){
581 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);
582 return -1;
583 } else if(status){
584 h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
585 }
586 }
587 }
588
589 return 0;
590 } //FIXME cleanup like next
591
592 /**
593 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
594 */
595 static inline int check_intra_pred_mode(H264Context *h, int mode){
596 MpegEncContext * const s = &h->s;
597 static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
598 static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
599
600 if(mode > 6U) {
601 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);
602 return -1;
603 }
604
605 if(!(h->top_samples_available&0x8000)){
606 mode= top[ mode ];
607 if(mode<0){
608 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);
609 return -1;
610 }
611 }
612
613 if(!(h->left_samples_available&0x8000)){
614 mode= left[ mode ];
615 if(mode<0){
616 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);
617 return -1;
618 }
619 }
620
621 return mode;
622 }
623
624 /**
625 * gets the predicted intra4x4 prediction mode.
626 */
627 static inline int pred_intra_mode(H264Context *h, int n){
628 const int index8= scan8[n];
629 const int left= h->intra4x4_pred_mode_cache[index8 - 1];
630 const int top = h->intra4x4_pred_mode_cache[index8 - 8];
631 const int min= FFMIN(left, top);
632
633 tprintf(h->s.avctx, "mode:%d %d min:%d\n", left ,top, min);
634
635 if(min<0) return DC_PRED;
636 else return min;
637 }
638
639 static inline void write_back_non_zero_count(H264Context *h){
640 MpegEncContext * const s = &h->s;
641 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
642
643 h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[7+8*1];
644 h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[7+8*2];
645 h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[7+8*3];
646 h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];
647 h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[4+8*4];
648 h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[5+8*4];
649 h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[6+8*4];
650
651 h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[1+8*2];
652 h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
653 h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[2+8*1];
654
655 h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[1+8*5];
656 h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
657 h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[2+8*4];
658
659 if(FRAME_MBAFF){
660 // store all luma nnzs, for deblocking
661 int v = 0, i;
662 for(i=0; i<16; i++)
663 v += (!!h->non_zero_count_cache[scan8[i]]) << i;
664 *(uint16_t*)&h->non_zero_count[mb_xy][14] = v;
665 }
666 }
667
668 /**
669 * gets the predicted number of non zero coefficients.
670 * @param n block index
671 */
672 static inline int pred_non_zero_count(H264Context *h, int n){
673 const int index8= scan8[n];
674 const int left= h->non_zero_count_cache[index8 - 1];
675 const int top = h->non_zero_count_cache[index8 - 8];
676 int i= left + top;
677
678 if(i<64) i= (i+1)>>1;
679
680 tprintf(h->s.avctx, "pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
681
682 return i&31;
683 }
684
685 static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
686 const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];
687 MpegEncContext *s = &h->s;
688
689 /* there is no consistent mapping of mvs to neighboring locations that will
690 * make mbaff happy, so we can't move all this logic to fill_caches */
691 if(FRAME_MBAFF){
692 const uint32_t *mb_types = s->current_picture_ptr->mb_type;
693 const int16_t *mv;
694 *(uint32_t*)h->mv_cache[list][scan8[0]-2] = 0;
695 *C = h->mv_cache[list][scan8[0]-2];
696
697 if(!MB_FIELD
698 && (s->mb_y&1) && i < scan8[0]+8 && topright_ref != PART_NOT_AVAILABLE){
699 int topright_xy = s->mb_x + (s->mb_y-1)*s->mb_stride + (i == scan8[0]+3);
700 if(IS_INTERLACED(mb_types[topright_xy])){
701 #define SET_DIAG_MV(MV_OP, REF_OP, X4, Y4)\
702 const int x4 = X4, y4 = Y4;\
703 const int mb_type = mb_types[(x4>>2)+(y4>>2)*s->mb_stride];\
704 if(!USES_LIST(mb_type,list) && !IS_8X8(mb_type))\
705 return LIST_NOT_USED;\
706 mv = s->current_picture_ptr->motion_val[list][x4 + y4*h->b_stride];\
707 h->mv_cache[list][scan8[0]-2][0] = mv[0];\
708 h->mv_cache[list][scan8[0]-2][1] = mv[1] MV_OP;\
709 return s->current_picture_ptr->ref_index[list][(x4>>1) + (y4>>1)*h->b8_stride] REF_OP;
710
711 SET_DIAG_MV(*2, >>1, s->mb_x*4+(i&7)-4+part_width, s->mb_y*4-1);
712 }
713 }
714 if(topright_ref == PART_NOT_AVAILABLE
715 && ((s->mb_y&1) || i >= scan8[0]+8) && (i&7)==4
716 && h->ref_cache[list][scan8[0]-1] != PART_NOT_AVAILABLE){
717 if(!MB_FIELD
718 && IS_INTERLACED(mb_types[h->left_mb_xy[0]])){
719 SET_DIAG_MV(*2, >>1, s->mb_x*4-1, (s->mb_y|1)*4+(s->mb_y&1)*2+(i>>4)-1);
720 }
721 if(MB_FIELD
722 && !IS_INTERLACED(mb_types[h->left_mb_xy[0]])
723 && i >= scan8[0]+8){
724 // leftshift will turn LIST_NOT_USED into PART_NOT_AVAILABLE, but that's ok.
725 SET_DIAG_MV(>>1, <<1, s->mb_x*4-1, (s->mb_y&~1)*4 - 1 + ((i-scan8[0])>>3)*2);
726 }
727 }
728 #undef SET_DIAG_MV
729 }
730
731 if(topright_ref != PART_NOT_AVAILABLE){
732 *C= h->mv_cache[list][ i - 8 + part_width ];
733 return topright_ref;
734 }else{
735 tprintf(s->avctx, "topright MV not available\n");
736
737 *C= h->mv_cache[list][ i - 8 - 1 ];
738 return h->ref_cache[list][ i - 8 - 1 ];
739 }
740 }
741
742 /**
743 * gets the predicted MV.
744 * @param n the block index
745 * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)
746 * @param mx the x component of the predicted motion vector
747 * @param my the y component of the predicted motion vector
748 */
749 static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
750 const int index8= scan8[n];
751 const int top_ref= h->ref_cache[list][ index8 - 8 ];
752 const int left_ref= h->ref_cache[list][ index8 - 1 ];
753 const int16_t * const A= h->mv_cache[list][ index8 - 1 ];
754 const int16_t * const B= h->mv_cache[list][ index8 - 8 ];
755 const int16_t * C;
756 int diagonal_ref, match_count;
757
758 assert(part_width==1 || part_width==2 || part_width==4);
759
760 /* mv_cache
761 B . . A T T T T
762 U . . L . . , .
763 U . . L . . . .
764 U . . L . . , .
765 . . . L . . . .
766 */
767
768 diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);
769 match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);
770 tprintf(h->s.avctx, "pred_motion match_count=%d\n", match_count);
771 if(match_count > 1){ //most common
772 *mx= mid_pred(A[0], B[0], C[0]);
773 *my= mid_pred(A[1], B[1], C[1]);
774 }else if(match_count==1){
775 if(left_ref==ref){
776 *mx= A[0];
777 *my= A[1];
778 }else if(top_ref==ref){
779 *mx= B[0];
780 *my= B[1];
781 }else{
782 *mx= C[0];
783 *my= C[1];
784 }
785 }else{
786 if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
787 *mx= A[0];
788 *my= A[1];
789 }else{
790 *mx= mid_pred(A[0], B[0], C[0]);
791 *my= mid_pred(A[1], B[1], C[1]);
792 }
793 }
794
795 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);
796 }
797
798 /**
799 * gets the directionally predicted 16x8 MV.
800 * @param n the block index
801 * @param mx the x component of the predicted motion vector
802 * @param my the y component of the predicted motion vector
803 */
804 static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
805 if(n==0){
806 const int top_ref= h->ref_cache[list][ scan8[0] - 8 ];
807 const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
808
809 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);
810
811 if(top_ref == ref){
812 *mx= B[0];
813 *my= B[1];
814 return;
815 }
816 }else{
817 const int left_ref= h->ref_cache[list][ scan8[8] - 1 ];
818 const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];
819
820 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);
821
822 if(left_ref == ref){
823 *mx= A[0];
824 *my= A[1];
825 return;
826 }
827 }
828
829 //RARE
830 pred_motion(h, n, 4, list, ref, mx, my);
831 }
832
833 /**
834 * gets the directionally predicted 8x16 MV.
835 * @param n the block index
836 * @param mx the x component of the predicted motion vector
837 * @param my the y component of the predicted motion vector
838 */
839 static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
840 if(n==0){
841 const int left_ref= h->ref_cache[list][ scan8[0] - 1 ];
842 const int16_t * const A= h->mv_cache[list][ scan8[0] - 1 ];
843
844 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);
845
846 if(left_ref == ref){
847 *mx= A[0];
848 *my= A[1];
849 return;
850 }
851 }else{
852 const int16_t * C;
853 int diagonal_ref;
854
855 diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);
856
857 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);
858
859 if(diagonal_ref == ref){
860 *mx= C[0];
861 *my= C[1];
862 return;
863 }
864 }
865
866 //RARE
867 pred_motion(h, n, 2, list, ref, mx, my);
868 }
869
870 static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
871 const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
872 const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];
873
874 tprintf(h->s.avctx, "pred_pskip: (%d) (%d) at %2d %2d\n", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
875
876 if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
877 || (top_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0)
878 || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){
879
880 *mx = *my = 0;
881 return;
882 }
883
884 pred_motion(h, 0, 4, 0, 0, mx, my);
885
886 return;
887 }
888
889 static inline void direct_dist_scale_factor(H264Context * const h){
890 const int poc = h->s.current_picture_ptr->poc;
891 const int poc1 = h->ref_list[1][0].poc;
892 int i;
893 for(i=0; i<h->ref_count[0]; i++){
894 int poc0 = h->ref_list[0][i].poc;
895 int td = av_clip(poc1 - poc0, -128, 127);
896 if(td == 0 /* FIXME || pic0 is a long-term ref */){
897 h->dist_scale_factor[i] = 256;
898 }else{
899 int tb = av_clip(poc - poc0, -128, 127);
900 int tx = (16384 + (FFABS(td) >> 1)) / td;
901 h->dist_scale_factor[i] = av_clip((tb*tx + 32) >> 6, -1024, 1023);
902 }
903 }
904 if(FRAME_MBAFF){
905 for(i=0; i<h->ref_count[0]; i++){
906 h->dist_scale_factor_field[2*i] =
907 h->dist_scale_factor_field[2*i+1] = h->dist_scale_factor[i];
908 }
909 }
910 }
911 static inline void direct_ref_list_init(H264Context * const h){
912 MpegEncContext * const s = &h->s;
913 Picture * const ref1 = &h->ref_list[1][0];
914 Picture * const cur = s->current_picture_ptr;
915 int list, i, j;
916 if(cur->pict_type == I_TYPE)
917 cur->ref_count[0] = 0;
918 if(cur->pict_type != B_TYPE)
919 cur->ref_count[1] = 0;
920 for(list=0; list<2; list++){
921 cur->ref_count[list] = h->ref_count[list];
922 for(j=0; j<h->ref_count[list]; j++)
923 cur->ref_poc[list][j] = h->ref_list[list][j].poc;
924 }
925 if(cur->pict_type != B_TYPE || h->direct_spatial_mv_pred)
926 return;
927 for(list=0; list<2; list++){
928 for(i=0; i<ref1->ref_count[list]; i++){
929 const int poc = ref1->ref_poc[list][i];
930 h->map_col_to_list0[list][i] = 0; /* bogus; fills in for missing frames */
931 for(j=0; j<h->ref_count[list]; j++)
932 if(h->ref_list[list][j].poc == poc){
933 h->map_col_to_list0[list][i] = j;
934 break;
935 }
936 }
937 }
938 if(FRAME_MBAFF){
939 for(list=0; list<2; list++){
940 for(i=0; i<ref1->ref_count[list]; i++){
941 j = h->map_col_to_list0[list][i];
942 h->map_col_to_list0_field[list][2*i] = 2*j;
943 h->map_col_to_list0_field[list][2*i+1] = 2*j+1;
944 }
945 }
946 }
947 }
948
949 static inline void pred_direct_motion(H264Context * const h, int *mb_type){
950 MpegEncContext * const s = &h->s;
951 const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
952 const int b8_xy = 2*s->mb_x + 2*s->mb_y*h->b8_stride;
953 const int b4_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
954 const int mb_type_col = h->ref_list[1][0].mb_type[mb_xy];
955 const int16_t (*l1mv0)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[0][b4_xy];
956 const int16_t (*l1mv1)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[1][b4_xy];
957 const int8_t *l1ref0 = &h->ref_list[1][0].ref_index[0][b8_xy];
958 const int8_t *l1ref1 = &h->ref_list[1][0].ref_index[1][b8_xy];
959 const int is_b8x8 = IS_8X8(*mb_type);
960 unsigned int sub_mb_type;
961 int i8, i4;
962
963 #define MB_TYPE_16x16_OR_INTRA (MB_TYPE_16x16|MB_TYPE_INTRA4x4|MB_TYPE_INTRA16x16|MB_TYPE_INTRA_PCM)
964 if(IS_8X8(mb_type_col) && !h->sps.direct_8x8_inference_flag){
965 /* FIXME save sub mb types from previous frames (or derive from MVs)
966 * so we know exactly what block size to use */
967 sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */
968 *mb_type = MB_TYPE_8x8|MB_TYPE_L0L1;
969 }else if(!is_b8x8 && (mb_type_col & MB_TYPE_16x16_OR_INTRA)){
970 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
971 *mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */
972 }else{
973 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
974 *mb_type = MB_TYPE_8x8|MB_TYPE_L0L1;
975 }
976 if(!is_b8x8)
977 *mb_type |= MB_TYPE_DIRECT2;
978 if(MB_FIELD)
979 *mb_type |= MB_TYPE_INTERLACED;
980
981 tprintf(s->avctx, "mb_type = %08x, sub_mb_type = %08x, is_b8x8 = %d, mb_type_col = %08x\n", *mb_type, sub_mb_type, is_b8x8, mb_type_col);
982
983 if(h->direct_spatial_mv_pred){
984 int ref[2];
985 int mv[2][2];
986 int list;
987
988 /* FIXME interlacing + spatial direct uses wrong colocated block positions */
989
990 /* ref = min(neighbors) */
991 for(list=0; list<2; list++){
992 int refa = h->ref_cache[list][scan8[0] - 1];
993 int refb = h->ref_cache[list][scan8[0] - 8];
994 int refc = h->ref_cache[list][scan8[0] - 8 + 4];
995 if(refc == -2)
996 refc = h->ref_cache[list][scan8[0] - 8 - 1];
997 ref[list] = refa;
998 if(ref[list] < 0 || (refb < ref[list] && refb >= 0))
999 ref[list] = refb;
1000 if(ref[list] < 0 || (refc < ref[list] && refc >= 0))
1001 ref[list] = refc;
1002 if(ref[list] < 0)
1003 ref[list] = -1;
1004 }
1005
1006 if(ref[0] < 0 && ref[1] < 0){
1007 ref[0] = ref[1] = 0;
1008 mv[0][0] = mv[0][1] =
1009 mv[1][0] = mv[1][1] = 0;
1010 }else{
1011 for(list=0; list<2; list++){
1012 if(ref[list] >= 0)
1013 pred_motion(h, 0, 4, list, ref[list], &mv[list][0], &mv[list][1]);
1014 else
1015 mv[list][0] = mv[list][1] = 0;
1016 }
1017 }
1018
1019 if(ref[1] < 0){
1020 *mb_type &= ~MB_TYPE_P0L1;
1021 sub_mb_type &= ~MB_TYPE_P0L1;
1022 }else if(ref[0] < 0){
1023 *mb_type &= ~MB_TYPE_P0L0;
1024 sub_mb_type &= ~MB_TYPE_P0L0;
1025 }
1026
1027 if(IS_16X16(*mb_type)){
1028 int a=0, b=0;
1029
1030 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, (uint8_t)ref[0], 1);
1031 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, (uint8_t)ref[1], 1);
1032 if(!IS_INTRA(mb_type_col)
1033 && ( (l1ref0[0] == 0 && FFABS(l1mv0[0][0]) <= 1 && FFABS(l1mv0[0][1]) <= 1)
1034 || (l1ref0[0] < 0 && l1ref1[0] == 0 && FFABS(l1mv1[0][0]) <= 1 && FFABS(l1mv1[0][1]) <= 1
1035 && (h->x264_build>33 || !h->x264_build)))){
1036 if(ref[0] > 0)
1037 a= pack16to32(mv[0][0],mv[0][1]);
1038 if(ref[1] > 0)
1039 b= pack16to32(mv[1][0],mv[1][1]);
1040 }else{
1041 a= pack16to32(mv[0][0],mv[0][1]);
1042 b= pack16to32(mv[1][0],mv[1][1]);
1043 }
1044 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, a, 4);
1045 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, b, 4);
1046 }else{
1047 for(i8=0; i8<4; i8++){
1048 const int x8 = i8&1;
1049 const int y8 = i8>>1;
1050
1051 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1052 continue;
1053 h->sub_mb_type[i8] = sub_mb_type;
1054
1055 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1056 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1057 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);
1058 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);
1059
1060 /* col_zero_flag */
1061 if(!IS_INTRA(mb_type_col) && ( l1ref0[x8 + y8*h->b8_stride] == 0
1062 || (l1ref0[x8 + y8*h->b8_stride] < 0 && l1ref1[x8 + y8*h->b8_stride] == 0
1063 && (h->x264_build>33 || !h->x264_build)))){
1064 const int16_t (*l1mv)[2]= l1ref0[x8 + y8*h->b8_stride] == 0 ? l1mv0 : l1mv1;
1065 if(IS_SUB_8X8(sub_mb_type)){
1066 const int16_t *mv_col = l1mv[x8*3 + y8*3*h->b_stride];
1067 if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
1068 if(ref[0] == 0)
1069 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1070 if(ref[1] == 0)
1071 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1072 }
1073 }else
1074 for(i4=0; i4<4; i4++){
1075 const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1076 if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
1077 if(ref[0] == 0)
1078 *(uint32_t*)h->mv_cache[0][scan8[i8*4+i4]] = 0;
1079 if(ref[1] == 0)
1080 *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = 0;
1081 }
1082 }
1083 }
1084 }
1085 }
1086 }else{ /* direct temporal mv pred */
1087 const int *map_col_to_list0[2] = {h->map_col_to_list0[0], h->map_col_to_list0[1]};
1088 const int *dist_scale_factor = h->dist_scale_factor;
1089
1090 if(FRAME_MBAFF){
1091 if(IS_INTERLACED(*mb_type)){
1092 map_col_to_list0[0] = h->map_col_to_list0_field[0];
1093 map_col_to_list0[1] = h->map_col_to_list0_field[1];
1094 dist_scale_factor = h->dist_scale_factor_field;
1095 }
1096 if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col)){
1097 /* FIXME assumes direct_8x8_inference == 1 */
1098 const int pair_xy = s->mb_x + (s->mb_y&~1)*s->mb_stride;
1099 int mb_types_col[2];
1100 int y_shift;
1101
1102 *mb_type = MB_TYPE_8x8|MB_TYPE_L0L1
1103 | (is_b8x8 ? 0 : MB_TYPE_DIRECT2)
1104 | (*mb_type & MB_TYPE_INTERLACED);
1105 sub_mb_type = MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2|MB_TYPE_16x16;
1106
1107 if(IS_INTERLACED(*mb_type)){
1108 /* frame to field scaling */
1109 mb_types_col[0] = h->ref_list[1][0].mb_type[pair_xy];
1110 mb_types_col[1] = h->ref_list[1][0].mb_type[pair_xy+s->mb_stride];
1111 if(s->mb_y&1){
1112 l1ref0 -= 2*h->b8_stride;
1113 l1ref1 -= 2*h->b8_stride;
1114 l1mv0 -= 4*h->b_stride;
1115 l1mv1 -= 4*h->b_stride;
1116 }
1117 y_shift = 0;
1118
1119 if( (mb_types_col[0] & MB_TYPE_16x16_OR_INTRA)
1120 && (mb_types_col[1] & MB_TYPE_16x16_OR_INTRA)
1121 && !is_b8x8)
1122 *mb_type |= MB_TYPE_16x8;
1123 else
1124 *mb_type |= MB_TYPE_8x8;
1125 }else{
1126 /* field to frame scaling */
1127 /* col_mb_y = (mb_y&~1) + (topAbsDiffPOC < bottomAbsDiffPOC ? 0 : 1)
1128 * but in MBAFF, top and bottom POC are equal */
1129 int dy = (s->mb_y&1) ? 1 : 2;
1130 mb_types_col[0] =
1131 mb_types_col[1] = h->ref_list[1][0].mb_type[pair_xy+s->mb_stride];
1132 l1ref0 += dy*h->b8_stride;
1133 l1ref1 += dy*h->b8_stride;
1134 l1mv0 += 2*dy*h->b_stride;
1135 l1mv1 += 2*dy*h->b_stride;
1136 y_shift = 2;
1137
1138 if((mb_types_col[0] & (MB_TYPE_16x16_OR_INTRA|MB_TYPE_16x8))
1139 && !is_b8x8)
1140 *mb_type |= MB_TYPE_16x16;
1141 else
1142 *mb_type |= MB_TYPE_8x8;
1143 }
1144
1145 for(i8=0; i8<4; i8++){
1146 const int x8 = i8&1;
1147 const int y8 = i8>>1;
1148 int ref0, scale;
1149 const int16_t (*l1mv)[2]= l1mv0;
1150
1151 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1152 continue;
1153 h->sub_mb_type[i8] = sub_mb_type;
1154
1155 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1156 if(IS_INTRA(mb_types_col[y8])){
1157 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1158 fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1159 fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1160 continue;
1161 }
1162
1163 ref0 = l1ref0[x8 + (y8*2>>y_shift)*h->b8_stride];
1164 if(ref0 >= 0)
1165 ref0 = map_col_to_list0[0][ref0*2>>y_shift];
1166 else{
1167 ref0 = map_col_to_list0[1][l1ref1[x8 + (y8*2>>y_shift)*h->b8_stride]*2>>y_shift];
1168 l1mv= l1mv1;
1169 }
1170 scale = dist_scale_factor[ref0];
1171 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1172
1173 {
1174 const int16_t *mv_col = l1mv[x8*3 + (y8*6>>y_shift)*h->b_stride];
1175 int my_col = (mv_col[1]<<y_shift)/2;
1176 int mx = (scale * mv_col[0] + 128) >> 8;
1177 int my = (scale * my_col + 128) >> 8;
1178 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
1179 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-my_col), 4);
1180 }
1181 }
1182 return;
1183 }
1184 }
1185
1186 /* one-to-one mv scaling */
1187
1188 if(IS_16X16(*mb_type)){
1189 int ref, mv0, mv1;
1190
1191 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1);
1192 if(IS_INTRA(mb_type_col)){
1193 ref=mv0=mv1=0;
1194 }else{
1195 const int ref0 = l1ref0[0] >= 0 ? map_col_to_list0[0][l1ref0[0]]
1196 : map_col_to_list0[1][l1ref1[0]];
1197 const int scale = dist_scale_factor[ref0];
1198 const int16_t *mv_col = l1ref0[0] >= 0 ? l1mv0[0] : l1mv1[0];
1199 int mv_l0[2];
1200 mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
1201 mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
1202 ref= ref0;
1203 mv0= pack16to32(mv_l0[0],mv_l0[1]);
1204 mv1= pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
1205 }
1206 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref, 1);
1207 fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, mv0, 4);
1208 fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, mv1, 4);
1209 }else{
1210 for(i8=0; i8<4; i8++){
1211 const int x8 = i8&1;
1212 const int y8 = i8>>1;
1213 int ref0, scale;
1214 const int16_t (*l1mv)[2]= l1mv0;
1215
1216 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1217 continue;
1218 h->sub_mb_type[i8] = sub_mb_type;
1219 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1220 if(IS_INTRA(mb_type_col)){
1221 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1222 fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1223 fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1224 continue;
1225 }
1226
1227 ref0 = l1ref0[x8 + y8*h->b8_stride];
1228 if(ref0 >= 0)
1229 ref0 = map_col_to_list0[0][ref0];
1230 else{
1231 ref0 = map_col_to_list0[1][l1ref1[x8 + y8*h->b8_stride]];
1232 l1mv= l1mv1;
1233 }
1234 scale = dist_scale_factor[ref0];
1235
1236 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1237 if(IS_SUB_8X8(sub_mb_type)){
1238 const int16_t *mv_col = l1mv[x8*3 + y8*3*h->b_stride];
1239 int mx = (scale * mv_col[0] + 128) >> 8;
1240 int my = (scale * mv_col[1] + 128) >> 8;
1241 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
1242 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-mv_col[1]), 4);
1243 }else
1244 for(i4=0; i4<4; i4++){
1245 const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1246 int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]];
1247 mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
1248 mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
1249 *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] =
1250 pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
1251 }
1252 }
1253 }
1254 }
1255 }
1256
1257 static inline void write_back_motion(H264Context *h, int mb_type){
1258 MpegEncContext * const s = &h->s;
1259 const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1260 const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1261 int list;
1262
1263 if(!USES_LIST(mb_type, 0))
1264 fill_rectangle(&s->current_picture.ref_index[0][b8_xy], 2, 2, h->b8_stride, (uint8_t)LIST_NOT_USED, 1);
1265
1266 for(list=0; list<h->list_count; list++){
1267 int y;
1268 if(!USES_LIST(mb_type, list))
1269 continue;
1270
1271 for(y=0; y<4; y++){
1272 *(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];
1273 *(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];
1274 }
1275 if( h->pps.cabac ) {
1276 if(IS_SKIP(mb_type))
1277 fill_rectangle(h->mvd_table[list][b_xy], 4, 4, h->b_stride, 0, 4);
1278 else
1279 for(y=0; y<4; y++){
1280 *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+0 + 8*y];
1281 *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+2 + 8*y];
1282 }
1283 }
1284
1285 {
1286 int8_t *ref_index = &s->current_picture.ref_index[list][b8_xy];
1287 ref_index[0+0*h->b8_stride]= h->ref_cache[list][scan8[0]];
1288 ref_index[1+0*h->b8_stride]= h->ref_cache[list][scan8[4]];
1289 ref_index[0+1*h->b8_stride]= h->ref_cache[list][scan8[8]];
1290 ref_index[1+1*h->b8_stride]= h->ref_cache[list][scan8[12]];
1291 }
1292 }
1293
1294 if(h->slice_type == B_TYPE && h->pps.cabac){
1295 if(IS_8X8(mb_type)){
1296 uint8_t *direct_table = &h->direct_table[b8_xy];
1297 direct_table[1+0*h->b8_stride] = IS_DIRECT(h->sub_mb_type[1]) ? 1 : 0;
1298 direct_table[0+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[2]) ? 1 : 0;
1299 direct_table[1+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[3]) ? 1 : 0;
1300 }
1301 }
1302 }
1303
1304 /**
1305 * Decodes a network abstraction layer unit.
1306 * @param consumed is the number of bytes used as input
1307 * @param length is the length of the array
1308 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp tailing?
1309 * @returns decoded bytes, might be src+1 if no escapes
1310 */
1311 static uint8_t *decode_nal(H264Context *h, uint8_t *src, int *dst_length, int *consumed, int length){
1312 int i, si, di;
1313 uint8_t *dst;
1314 int bufidx;
1315
1316 // src[0]&0x80; //forbidden bit
1317 h->nal_ref_idc= src[0]>>5;
1318 h->nal_unit_type= src[0]&0x1F;
1319
1320 src++; length--;
1321 #if 0
1322 for(i=0; i<length; i++)
1323 printf("%2X ", src[i]);
1324 #endif
1325 for(i=0; i+1<length; i+=2){
1326 if(src[i]) continue;
1327 if(i>0 && src[i-1]==0) i--;
1328 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1329 if(src[i+2]!=3){
1330 /* startcode, so we must be past the end */
1331 length=i;
1332 }
1333 break;
1334 }
1335 }
1336
1337 if(i>=length-1){ //no escaped 0
1338 *dst_length= length;
1339 *consumed= length+1; //+1 for the header
1340 return src;
1341 }
1342
1343 bufidx = h->nal_unit_type == NAL_DPC ? 1 : 0; // use second escape buffer for inter data
1344 h->rbsp_buffer[bufidx]= av_fast_realloc(h->rbsp_buffer[bufidx], &h->rbsp_buffer_size[bufidx], length);
1345 dst= h->rbsp_buffer[bufidx];
1346
1347 if (dst == NULL){
1348 return NULL;
1349 }
1350
1351 //printf("decoding esc\n");
1352 si=di=0;
1353 while(si<length){
1354 //remove escapes (very rare 1:2^22)
1355 if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1356 if(src[si+2]==3){ //escape
1357 dst[di++]= 0;
1358 dst[di++]= 0;
1359 si+=3;
1360 continue;
1361 }else //next start code
1362 break;
1363 }
1364
1365 dst[di++]= src[si++];
1366 }
1367
1368 *dst_length= di;
1369 *consumed= si + 1;//+1 for the header
1370 //FIXME store exact number of bits in the getbitcontext (it is needed for decoding)
1371 return dst;
1372 }
1373
1374 /**
1375 * identifies the exact end of the bitstream
1376 * @return the length of the trailing, or 0 if damaged
1377 */
1378 static int decode_rbsp_trailing(H264Context *h, uint8_t *src){
1379 int v= *src;
1380 int r;
1381
1382 tprintf(h->s.avctx, "rbsp trailing %X\n", v);
1383
1384 for(r=1; r<9; r++){
1385 if(v&1) return r;
1386 v>>=1;
1387 }
1388 return 0;
1389 }
1390
1391 /**
1392 * idct tranforms the 16 dc values and dequantize them.
1393 * @param qp quantization parameter
1394 */
1395 static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
1396 #define stride 16
1397 int i;
1398 int temp[16]; //FIXME check if this is a good idea
1399 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
1400 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1401
1402 //memset(block, 64, 2*256);
1403 //return;
1404 for(i=0; i<4; i++){
1405 const int offset= y_offset[i];
1406 const int z0= block[offset+stride*0] + block[offset+stride*4];
1407 const int z1= block[offset+stride*0] - block[offset+stride*4];
1408 const int z2= block[offset+stride*1] - block[offset+stride*5];
1409 const int z3= block[offset+stride*1] + block[offset+stride*5];
1410
1411 temp[4*i+0]= z0+z3;
1412 temp[4*i+1]= z1+z2;
1413 temp[4*i+2]= z1-z2;
1414 temp[4*i+3]= z0-z3;
1415 }
1416
1417 for(i=0; i<4; i++){
1418 const int offset= x_offset[i];
1419 const int z0= temp[4*0+i] + temp[4*2+i];
1420 const int z1= temp[4*0+i] - temp[4*2+i];
1421 const int z2= temp[4*1+i] - temp[4*3+i];
1422 const int z3= temp[4*1+i] + temp[4*3+i];
1423
1424 block[stride*0 +offset]= ((((z0 + z3)*qmul + 128 ) >> 8)); //FIXME think about merging this into decode_resdual
1425 block[stride*2 +offset]= ((((z1 + z2)*qmul + 128 ) >> 8));
1426 block[stride*8 +offset]= ((((z1 - z2)*qmul + 128 ) >> 8));
1427 block[stride*10+offset]= ((((z0 - z3)*qmul + 128 ) >> 8));
1428 }
1429 }
1430
1431 #if 0
1432 /**
1433 * dct tranforms the 16 dc values.
1434 * @param qp quantization parameter ??? FIXME
1435 */
1436 static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
1437 // const int qmul= dequant_coeff[qp][0];
1438 int i;
1439 int temp[16]; //FIXME check if this is a good idea
1440 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
1441 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1442
1443 for(i=0; i<4; i++){
1444 const int offset= y_offset[i];
1445 const int z0= block[offset+stride*0] + block[offset+stride*4];
1446 const int z1= block[offset+stride*0] - block[offset+stride*4];
1447 const int z2= block[offset+stride*1] - block[offset+stride*5];
1448 const int z3= block[offset+stride*1] + block[offset+stride*5];
1449
1450 temp[4*i+0]= z0+z3;
1451 temp[4*i+1]= z1+z2;
1452 temp[4*i+2]= z1-z2;
1453 temp[4*i+3]= z0-z3;
1454 }
1455
1456 for(i=0; i<4; i++){
1457 const int offset= x_offset[i];
1458 const int z0= temp[4*0+i] + temp[4*2+i];
1459 const int z1= temp[4*0+i] - temp[4*2+i];
1460 const int z2= temp[4*1+i] - temp[4*3+i];
1461 const int z3= temp[4*1+i] + temp[4*3+i];
1462
1463 block[stride*0 +offset]= (z0 + z3)>>1;
1464 block[stride*2 +offset]= (z1 + z2)>>1;
1465 block[stride*8 +offset]= (z1 - z2)>>1;
1466 block[stride*10+offset]= (z0 - z3)>>1;
1467 }
1468 }
1469 #endif
1470
1471 #undef xStride
1472 #undef stride
1473
1474 static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
1475 const int stride= 16*2;
1476 const int xStride= 16;
1477 int a,b,c,d,e;
1478
1479 a= block[stride*0 + xStride*0];
1480 b= block[stride*0 + xStride*1];
1481 c= block[stride*1 + xStride*0];
1482 d= block[stride*1 + xStride*1];
1483
1484 e= a-b;
1485 a= a+b;
1486 b= c-d;
1487 c= c+d;
1488
1489 block[stride*0 + xStride*0]= ((a+c)*qmul) >> 7;
1490 block[stride*0 + xStride*1]= ((e+b)*qmul) >> 7;
1491 block[stride*1 + xStride*0]= ((a-c)*qmul) >> 7;
1492 block[stride*1 + xStride*1]= ((e-b)*qmul) >> 7;
1493 }
1494
1495 #if 0
1496 static void chroma_dc_dct_c(DCTELEM *block){
1497 const int stride= 16*2;
1498 const int xStride= 16;
1499 int a,b,c,d,e;
1500
1501 a= block[stride*0 + xStride*0];
1502 b= block[stride*0 + xStride*1];
1503 c= block[stride*1 + xStride*0];
1504 d= block[stride*1 + xStride*1];
1505
1506 e= a-b;
1507 a= a+b;
1508 b= c-d;
1509 c= c+d;
1510
1511 block[stride*0 + xStride*0]= (a+c);
1512 block[stride*0 + xStride*1]= (e+b);
1513 block[stride*1 + xStride*0]= (a-c);
1514 block[stride*1 + xStride*1]= (e-b);
1515 }
1516 #endif
1517
1518 /**
1519 * gets the chroma qp.
1520 */
1521 static inline int get_chroma_qp(H264Context *h, int t, int qscale){
1522 return h->pps.chroma_qp_table[t][qscale & 0xff];
1523 }
1524
1525 //FIXME need to check that this does not overflow signed 32 bit for low qp, i am not sure, it's very close
1526 //FIXME check that gcc inlines this (and optimizes intra & separate_dc stuff away)
1527 static inline int quantize_c(DCTELEM *block, uint8_t *scantable, int qscale, int intra, int separate_dc){
1528 int i;
1529 const int * const quant_table= quant_coeff[qscale];
1530 const int bias= intra ? (1<<QUANT_SHIFT)/3 : (1<<QUANT_SHIFT)/6;
1531 const unsigned int threshold1= (1<<QUANT_SHIFT) - bias - 1;
1532 const unsigned int threshold2= (threshold1<<1);
1533 int last_non_zero;
1534
1535 if(separate_dc){
1536 if(qscale<=18){
1537 //avoid overflows
1538 const int dc_bias= intra ? (1<<(QUANT_SHIFT-2))/3 : (1<<(QUANT_SHIFT-2))/6;
1539 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT-2)) - dc_bias - 1;
1540 const unsigned int dc_threshold2= (dc_threshold1<<1);
1541
1542 int level= block[0]*quant_coeff[qscale+18][0];
1543 if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1544 if(level>0){
1545 level= (dc_bias + level)>>(QUANT_SHIFT-2);
1546 block[0]= level;
1547 }else{
1548 level= (dc_bias - level)>>(QUANT_SHIFT-2);
1549 block[0]= -level;
1550 }
1551 // last_non_zero = i;
1552 }else{
1553 block[0]=0;
1554 }
1555 }else{
1556 const int dc_bias= intra ? (1<<(QUANT_SHIFT+1))/3 : (1<<(QUANT_SHIFT+1))/6;
1557 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT+1)) - dc_bias - 1;
1558 const unsigned int dc_threshold2= (dc_threshold1<<1);
1559
1560 int level= block[0]*quant_table[0];
1561 if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1562 if(level>0){
1563 level= (dc_bias + level)>>(QUANT_SHIFT+1);
1564 block[0]= level;
1565 }else{
1566 level= (dc_bias - level)>>(QUANT_SHIFT+1);
1567 block[0]= -level;
1568 }
1569 // last_non_zero = i;
1570 }else{
1571 block[0]=0;
1572 }
1573 }
1574 last_non_zero= 0;
1575 i=1;
1576 }else{
1577 last_non_zero= -1;
1578 i=0;
1579 }
1580
1581 for(; i<16; i++){
1582 const int j= scantable[i];
1583 int level= block[j]*quant_table[j];
1584
1585 // if( bias+level >= (1<<(QMAT_SHIFT - 3))
1586 // || bias-level >= (1<<(QMAT_SHIFT - 3))){
1587 if(((unsigned)(level+threshold1))>threshold2){
1588 if(level>0){
1589 level= (bias + level)>>QUANT_SHIFT;
1590 block[j]= level;
1591 }else{
1592 level= (bias - level)>>QUANT_SHIFT;
1593 block[j]= -level;
1594 }
1595 last_non_zero = i;
1596 }else{
1597 block[j]=0;
1598 }
1599 }
1600
1601 return last_non_zero;
1602 }
1603
1604 static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
1605 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1606 int src_x_offset, int src_y_offset,
1607 qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
1608 MpegEncContext * const s = &h->s;
1609 const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
1610 int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
1611 const int luma_xy= (mx&3) + ((my&3)<<2);
1612 uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*h->mb_linesize;
1613 uint8_t * src_cb, * src_cr;
1614 int extra_width= h->emu_edge_width;
1615 int extra_height= h->emu_edge_height;
1616 int emu=0;
1617 const int full_mx= mx>>2;
1618 const int full_my= my>>2;
1619 const int pic_width = 16*s->mb_width;
1620 const int pic_height = 16*s->mb_height >> MB_FIELD;
1621
1622 if(!pic->data[0]) //FIXME this is unacceptable, some senseable error concealment must be done for missing reference frames
1623 return;
1624
1625 if(mx&7) extra_width -= 3;
1626 if(my&7) extra_height -= 3;
1627
1628 if( full_mx < 0-extra_width
1629 || full_my < 0-extra_height
1630 || full_mx + 16/*FIXME*/ > pic_width + extra_width
1631 || full_my + 16/*FIXME*/ > pic_height + extra_height){
1632 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);
1633 src_y= s->edge_emu_buffer + 2 + 2*h->mb_linesize;
1634 emu=1;
1635 }
1636
1637 qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); //FIXME try variable height perhaps?
1638 if(!square){
1639 qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
1640 }
1641
1642 if(ENABLE_GRAY && s->flags&CODEC_FLAG_GRAY) return;
1643
1644 if(MB_FIELD){
1645 // chroma offset when predicting from a field of opposite parity
1646 my += 2 * ((s->mb_y & 1) - (pic->reference - 1));
1647 emu |= (my>>3) < 0 || (my>>3) + 8 >= (pic_height>>1);
1648 }
1649 src_cb= pic->data[1] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
1650 src_cr= pic->data[2] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
1651
1652 if(emu){
1653 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);
1654 src_cb= s->edge_emu_buffer;
1655 }
1656 chroma_op(dest_cb, src_cb, h->mb_uvlinesize, chroma_height, mx&7, my&7);
1657
1658 if(emu){
1659 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);
1660 src_cr= s->edge_emu_buffer;
1661 }
1662 chroma_op(dest_cr, src_cr, h->mb_uvlinesize, chroma_height, mx&7, my&7);
1663 }
1664
1665 static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
1666 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1667 int x_offset, int y_offset,
1668 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
1669 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
1670 int list0, int list1){
1671 MpegEncContext * const s = &h->s;
1672 qpel_mc_func *qpix_op= qpix_put;
1673 h264_chroma_mc_func chroma_op= chroma_put;
1674
1675 dest_y += 2*x_offset + 2*y_offset*h-> mb_linesize;
1676 dest_cb += x_offset + y_offset*h->mb_uvlinesize;
1677 dest_cr += x_offset + y_offset*h->mb_uvlinesize;
1678 x_offset += 8*s->mb_x;
1679 y_offset += 8*(s->mb_y >> MB_FIELD);
1680
1681 if(list0){
1682 Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
1683 mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
1684 dest_y, dest_cb, dest_cr, x_offset, y_offset,
1685 qpix_op, chroma_op);
1686
1687 qpix_op= qpix_avg;
1688 chroma_op= chroma_avg;
1689 }
1690
1691 if(list1){
1692 Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
1693 mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
1694 dest_y, dest_cb, dest_cr, x_offset, y_offset,
1695 qpix_op, chroma_op);
1696 }
1697 }
1698
1699 static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
1700 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1701 int x_offset, int y_offset,
1702 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
1703 h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
1704 h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
1705 int list0, int list1){
1706 MpegEncContext * const s = &h->s;
1707
1708 dest_y += 2*x_offset + 2*y_offset*h-> mb_linesize;
1709 dest_cb += x_offset + y_offset*h->mb_uvlinesize;
1710 dest_cr += x_offset + y_offset*h->mb_uvlinesize;
1711 x_offset += 8*s->mb_x;
1712 y_offset += 8*(s->mb_y >> MB_FIELD);
1713
1714 if(list0 && list1){
1715 /* don't optimize for luma-only case, since B-frames usually
1716 * use implicit weights => chroma too. */
1717 uint8_t *tmp_cb = s->obmc_scratchpad;
1718 uint8_t *tmp_cr = s->obmc_scratchpad + 8;
1719 uint8_t *tmp_y = s->obmc_scratchpad + 8*h->mb_uvlinesize;
1720 int refn0 = h->ref_cache[0][ scan8[n] ];
1721 int refn1 = h->ref_cache[1][ scan8[n] ];
1722
1723 mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
1724 dest_y, dest_cb, dest_cr,
1725 x_offset, y_offset, qpix_put, chroma_put);
1726 mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
1727 tmp_y, tmp_cb, tmp_cr,
1728 x_offset, y_offset, qpix_put, chroma_put);
1729
1730 if(h->use_weight == 2){
1731 int weight0 = h->implicit_weight[refn0][refn1];
1732 int weight1 = 64 - weight0;
1733 luma_weight_avg( dest_y, tmp_y, h-> mb_linesize, 5, weight0, weight1, 0);
1734 chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, 5, weight0, weight1, 0);
1735 chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, 5, weight0, weight1, 0);
1736 }else{
1737 luma_weight_avg(dest_y, tmp_y, h->mb_linesize, h->luma_log2_weight_denom,
1738 h->luma_weight[0][refn0], h->luma_weight[1][refn1],
1739 h->luma_offset[0][refn0] + h->luma_offset[1][refn1]);
1740 chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
1741 h->chroma_weight[0][refn0][0], h->chroma_weight[1][refn1][0],
1742 h->chroma_offset[0][refn0][0] + h->chroma_offset[1][refn1][0]);
1743 chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
1744 h->chroma_weight[0][refn0][1], h->chroma_weight[1][refn1][1],
1745 h->chroma_offset[0][refn0][1] + h->chroma_offset[1][refn1][1]);
1746 }
1747 }else{
1748 int list = list1 ? 1 : 0;
1749 int refn = h->ref_cache[list][ scan8[n] ];
1750 Picture *ref= &h->ref_list[list][refn];
1751 mc_dir_part(h, ref, n, square, chroma_height, delta, list,
1752 dest_y, dest_cb, dest_cr, x_offset, y_offset,
1753 qpix_put, chroma_put);
1754
1755 luma_weight_op(dest_y, h->mb_linesize, h->luma_log2_weight_denom,
1756 h->luma_weight[list][refn], h->luma_offset[list][refn]);
1757 if(h->use_weight_chroma){
1758 chroma_weight_op(dest_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
1759 h->chroma_weight[list][refn][0], h->chroma_offset[list][refn][0]);
1760 chroma_weight_op(dest_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
1761 h->chroma_weight[list][refn][1], h->chroma_offset[list][refn][1]);
1762 }
1763 }
1764 }
1765
1766 static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
1767 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1768 int x_offset, int y_offset,
1769 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
1770 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
1771 h264_weight_func *weight_op, h264_biweight_func *weight_avg,
1772 int list0, int list1){
1773 if((h->use_weight==2 && list0 && list1
1774 && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ] != 32))
1775 || h->use_weight==1)
1776 mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
1777 x_offset, y_offset, qpix_put, chroma_put,
1778 weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
1779 else
1780 mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
1781 x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
1782 }
1783
1784 static inline void prefetch_motion(H264Context *h, int list){
1785 /* fetch pixels for estimated mv 4 macroblocks ahead
1786 * optimized for 64byte cache lines */
1787 MpegEncContext * const s = &h->s;
1788 const int refn = h->ref_cache[list][scan8[0]];
1789 if(refn >= 0){
1790 const int mx= (h->mv_cache[list][scan8[0]][0]>>2) + 16*s->mb_x + 8;
1791 const int my= (h->mv_cache[list][scan8[0]][1]>>2) + 16*s->mb_y;
1792 uint8_t **src= h->ref_list[list][refn].data;
1793 int off= mx + (my + (s->mb_x&3)*4)*h->mb_linesize + 64;
1794 s->dsp.prefetch(src[0]+off, s->linesize, 4);
1795 off= (mx>>1) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + 64;
1796 s->dsp.prefetch(src[1]+off, src[2]-src[1], 2);
1797 }
1798 }
1799
1800 static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1801 qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
1802 qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
1803 h264_weight_func *weight_op, h264_biweight_func *weight_avg){
1804 MpegEncContext * const s = &h->s;
1805 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
1806 const int mb_type= s->current_picture.mb_type[mb_xy];
1807
1808 assert(IS_INTER(mb_type));
1809
1810 prefetch_motion(h, 0);
1811
1812 if(IS_16X16(mb_type)){
1813 mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
1814 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
1815 &weight_op[0], &weight_avg[0],
1816 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1817 }else if(IS_16X8(mb_type)){
1818 mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
1819 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
1820 &weight_op[1], &weight_avg[1],
1821 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1822 mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
1823 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
1824 &weight_op[1], &weight_avg[1],
1825 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
1826 }else if(IS_8X16(mb_type)){
1827 mc_part(h, 0, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 0, 0,
1828 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
1829 &weight_op[2], &weight_avg[2],
1830 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1831 mc_part(h, 4, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 4, 0,
1832 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
1833 &weight_op[2], &weight_avg[2],
1834 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
1835 }else{
1836 int i;
1837
1838 assert(IS_8X8(mb_type));
1839
1840 for(i=0; i<4; i++){
1841 const int sub_mb_type= h->sub_mb_type[i];
1842 const int n= 4*i;
1843 int x_offset= (i&1)<<2;
1844 int y_offset= (i&2)<<1;
1845
1846 if(IS_SUB_8X8(sub_mb_type)){
1847 mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
1848 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
1849 &weight_op[3], &weight_avg[3],
1850 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1851 }else if(IS_SUB_8X4(sub_mb_type)){
1852 mc_part(h, n , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
1853 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
1854 &weight_op[4], &weight_avg[4],
1855 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1856 mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
1857 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
1858 &weight_op[4], &weight_avg[4],
1859 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1860 }else if(IS_SUB_4X8(sub_mb_type)){
1861 mc_part(h, n , 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
1862 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
1863 &weight_op[5], &weight_avg[5],
1864 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1865 mc_part(h, n+1, 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
1866 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
1867 &weight_op[5], &weight_avg[5],
1868 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1869 }else{
1870 int j;
1871 assert(IS_SUB_4X4(sub_mb_type));
1872 for(j=0; j<4; j++){
1873 int sub_x_offset= x_offset + 2*(j&1);
1874 int sub_y_offset= y_offset + (j&2);
1875 mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
1876 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
1877 &weight_op[6], &weight_avg[6],
1878 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1879 }
1880 }
1881 }
1882 }
1883
1884 prefetch_motion(h, 1);
1885 }
1886
1887 static void decode_init_vlc(void){
1888 static int done = 0;
1889
1890 if (!done) {
1891 int i;
1892 done = 1;
1893
1894 init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5,
1895 &chroma_dc_coeff_token_len [0], 1, 1,
1896 &chroma_dc_coeff_token_bits[0], 1, 1, 1);
1897
1898 for(i=0; i<4; i++){
1899 init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17,
1900 &coeff_token_len [i][0], 1, 1,
1901 &coeff_token_bits[i][0], 1, 1, 1);
1902 }
1903
1904 for(i=0; i<3; i++){
1905 init_vlc(&chroma_dc_total_zeros_vlc[i], CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
1906 &chroma_dc_total_zeros_len [i][0], 1, 1,
1907 &chroma_dc_total_zeros_bits[i][0], 1, 1, 1);
1908 }
1909 for(i=0; i<15; i++){
1910 init_vlc(&total_zeros_vlc[i], TOTAL_ZEROS_VLC_BITS, 16,
1911 &total_zeros_len [i][0], 1, 1,
1912 &total_zeros_bits[i][0], 1, 1, 1);
1913 }
1914
1915 for(i=0; i<6; i++){
1916 init_vlc(&run_vlc[i], RUN_VLC_BITS, 7,
1917 &run_len [i][0], 1, 1,
1918 &run_bits[i][0], 1, 1, 1);
1919 }
1920 init_vlc(&run7_vlc, RUN7_VLC_BITS, 16,
1921 &run_len [6][0], 1, 1,
1922 &run_bits[6][0], 1, 1, 1);
1923 }
1924 }
1925
1926 static void free_tables(H264Context *h){
1927 int i;
1928 H264Context *hx;
1929 av_freep(&h->intra4x4_pred_mode);
1930 av_freep(&h->chroma_pred_mode_table);
1931 av_freep(&h->cbp_table);
1932 av_freep(&h->mvd_table[0]);
1933 av_freep(&h->mvd_table[1]);
1934 av_freep(&h->direct_table);
1935 av_freep(&h->non_zero_count);
1936 av_freep(&h->slice_table_base);
1937 h->slice_table= NULL;
1938
1939 av_freep(&h->mb2b_xy);
1940 av_freep(&h->mb2b8_xy);
1941
1942 for(i = 0; i < MAX_SPS_COUNT; i++)
1943 av_freep(h->sps_buffers + i);
1944
1945 for(i = 0; i < MAX_PPS_COUNT; i++)
1946 av_freep(h->pps_buffers + i);
1947
1948 for(i = 0; i < h->s.avctx->thread_count; i++) {
1949 hx = h->thread_context[i];
1950 if(!hx) continue;
1951 av_freep(&hx->top_borders[1]);
1952 av_freep(&hx->top_borders[0]);
1953 av_freep(&hx->s.obmc_scratchpad);
1954 av_freep(&hx->s.allocated_edge_emu_buffer);
1955 }
1956 }
1957
1958 static void init_dequant8_coeff_table(H264Context *h){
1959 int i,q,x;
1960 const int transpose = (h->s.dsp.h264_idct8_add != ff_h264_idct8_add_c); //FIXME ugly
1961 h->dequant8_coeff[0] = h->dequant8_buffer[0];
1962 h->dequant8_coeff[1] = h->dequant8_buffer[1];
1963
1964 for(i=0; i<2; i++ ){
1965 if(i && !memcmp(h->pps.scaling_matrix8[0], h->pps.scaling_matrix8[1], 64*sizeof(uint8_t))){
1966 h->dequant8_coeff[1] = h->dequant8_buffer[0];
1967 break;
1968 }
1969
1970 for(q=0; q<52; q++){
1971 int shift = ff_div6[q];
1972 int idx = ff_rem6[q];
1973 for(x=0; x<64; x++)
1974 h->dequant8_coeff[i][q][transpose ? (x>>3)|((x&7)<<3) : x] =
1975 ((uint32_t)dequant8_coeff_init[idx][ dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] *
1976 h->pps.scaling_matrix8[i][x]) << shift;
1977 }
1978 }
1979 }
1980
1981 static void init_dequant4_coeff_table(H264Context *h){
1982 int i,j,q,x;
1983 const int transpose = (h->s.dsp.h264_idct_add != ff_h264_idct_add_c); //FIXME ugly
1984 for(i=0; i<6; i++ ){
1985 h->dequant4_coeff[i] = h->dequant4_buffer[i];
1986 for(j=0; j<i; j++){
1987 if(!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i], 16*sizeof(uint8_t))){
1988 h->dequant4_coeff[i] = h->dequant4_buffer[j];
1989 break;
1990 }
1991 }
1992 if(j<i)
1993 continue;
1994
1995 for(q=0; q<52; q++){
1996 int shift = ff_div6[q] + 2;
1997 int idx = ff_rem6[q];
1998 for(x=0; x<16; x++)
1999 h->dequant4_coeff[i][q][transpose ? (x>>2)|((x<<2)&0xF) : x] =
2000 ((uint32_t)dequant4_coeff_init[idx][(x&1) + ((x>>2)&1)] *
2001 h->pps.scaling_matrix4[i][x]) << shift;
2002 }
2003 }
2004 }
2005
2006 static void init_dequant_tables(H264Context *h){
2007 int i,x;
2008 init_dequant4_coeff_table(h);
2009 if(h->pps.transform_8x8_mode)
2010 init_dequant8_coeff_table(h);
2011 if(h->sps.transform_bypass){
2012 for(i=0; i<6; i++)
2013 for(x=0; x<16; x++)
2014 h->dequant4_coeff[i][0][x] = 1<<6;
2015 if(h->pps.transform_8x8_mode)
2016 for(i=0; i<2; i++)
2017 for(x=0; x<64; x++)
2018 h->dequant8_coeff[i][0][x] = 1<<6;
2019 }
2020 }
2021
2022
2023 /**
2024 * allocates tables.
2025 * needs width/height
2026 */
2027 static int alloc_tables(H264Context *h){
2028 MpegEncContext * const s = &h->s;
2029 const int big_mb_num= s->mb_stride * (s->mb_height+1);
2030 int x,y;
2031
2032 CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8 * sizeof(uint8_t))
2033
2034 CHECKED_ALLOCZ(h->non_zero_count , big_mb_num * 16 * sizeof(uint8_t))
2035 CHECKED_ALLOCZ(h->slice_table_base , (big_mb_num+s->mb_stride) * sizeof(uint8_t))
2036 CHECKED_ALLOCZ(h->cbp_table, big_mb_num * sizeof(uint16_t))
2037
2038 if( h->pps.cabac ) {
2039 CHECKED_ALLOCZ(h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t))
2040 CHECKED_ALLOCZ(h->mvd_table[0], 32*big_mb_num * sizeof(uint16_t));
2041 CHECKED_ALLOCZ(h->mvd_table[1], 32*big_mb_num * sizeof(uint16_t));
2042 CHECKED_ALLOCZ(h->direct_table, 32*big_mb_num * sizeof(uint8_t));
2043 }
2044
2045 memset(h->slice_table_base, -1, (big_mb_num+s->mb_stride) * sizeof(uint8_t));
2046 h->slice_table= h->slice_table_base + s->mb_stride*2 + 1;
2047
2048 CHECKED_ALLOCZ(h->mb2b_xy , big_mb_num * sizeof(uint32_t));
2049 CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint32_t));
2050 for(y=0; y<s->mb_height; y++){
2051 for(x=0; x<s->mb_width; x++){
2052 const int mb_xy= x + y*s->mb_stride;
2053 const int b_xy = 4*x + 4*y*h->b_stride;
2054 const int b8_xy= 2*x + 2*y*h->b8_stride;
2055
2056 h->mb2b_xy [mb_xy]= b_xy;
2057 h->mb2b8_xy[mb_xy]= b8_xy;
2058 }
2059 }
2060
2061 s->obmc_scratchpad = NULL;
2062
2063 if(!h->dequant4_coeff[0])
2064 init_dequant_tables(h);
2065
2066 return 0;
2067 fail:
2068 free_tables(h);
2069 return -1;
2070 }
2071
2072 /**
2073 * Mimic alloc_tables(), but for every context thread.
2074 */
2075 static void clone_tables(H264Context *dst, H264Context *src){
2076 dst->intra4x4_pred_mode = src->intra4x4_pred_mode;
2077 dst->non_zero_count = src->non_zero_count;
2078 dst->slice_table = src->slice_table;
2079 dst->cbp_table = src->cbp_table;
2080 dst->mb2b_xy = src->mb2b_xy;
2081 dst->mb2b8_xy = src->mb2b8_xy;
2082 dst->chroma_pred_mode_table = src->chroma_pred_mode_table;
2083 dst->mvd_table[0] = src->mvd_table[0];
2084 dst->mvd_table[1] = src->mvd_table[1];
2085 dst->direct_table = src->direct_table;
2086
2087 dst->s.obmc_scratchpad = NULL;
2088 ff_h264_pred_init(&dst->hpc, src->s.codec_id);
2089 }
2090
2091 /**
2092 * Init context
2093 * Allocate buffers which are not shared amongst multiple threads.
2094 */
2095 static int context_init(H264Context *h){
2096 MpegEncContext * const s = &h->s;
2097
2098 CHECKED_ALLOCZ(h->top_borders[0], h->s.mb_width * (16+8+8) * sizeof(uint8_t))
2099 CHECKED_ALLOCZ(h->top_borders[1], h->s.mb_width * (16+8+8) * sizeof(uint8_t))
2100
2101 // edge emu needs blocksize + filter length - 1 (=17x17 for halfpel / 21x21 for h264)
2102 CHECKED_ALLOCZ(s->allocated_edge_emu_buffer,
2103 (s->width+64)*2*21*2); //(width + edge + align)*interlaced*MBsize*tolerance
2104 s->edge_emu_buffer= s->allocated_edge_emu_buffer + (s->width+64)*2*21;
2105 return 0;
2106 fail:
2107 return -1; // free_tables will clean up for us
2108 }
2109
2110 static void common_init(H264Context *h){
2111 MpegEncContext * const s = &h->s;
2112
2113 s->width = s->avctx->width;
2114 s->height = s->avctx->height;
2115 s->codec_id= s->avctx->codec->id;
2116
2117 ff_h264_pred_init(&h->hpc, s->codec_id);
2118
2119 h->dequant_coeff_pps= -1;
2120 s->unrestricted_mv=1;
2121 s->decode=1; //FIXME
2122
2123 memset(h->pps.scaling_matrix4, 16, 6*16*sizeof(uint8_t));
2124 memset(h->pps.scaling_matrix8, 16, 2*64*sizeof(uint8_t));
2125 }
2126
2127 static int decode_init(AVCodecContext *avctx){
2128 H264Context *h= avctx->priv_data;
2129 MpegEncContext * const s = &h->s;
2130
2131 MPV_decode_defaults(s);
2132
2133 s->avctx = avctx;
2134 common_init(h);
2135
2136 s->out_format = FMT_H264;
2137 s->workaround_bugs= avctx->workaround_bugs;
2138
2139 // set defaults
2140 // s->decode_mb= ff_h263_decode_mb;
2141 s->quarter_sample = 1;
2142 s->low_delay= 1;
2143 avctx->pix_fmt= PIX_FMT_YUV420P;
2144
2145 decode_init_vlc();
2146
2147 if(avctx->extradata_size > 0 && avctx->extradata &&
2148 *(char *)avctx->extradata == 1){
2149 h->is_avc = 1;
2150 h->got_avcC = 0;
2151 } else {
2152 h->is_avc = 0;
2153 }
2154
2155 h->thread_context[0] = h;
2156 return 0;
2157 }
2158
2159 static int frame_start(H264Context *h){
2160 MpegEncContext * const s = &h->s;
2161 int i;
2162
2163 if(MPV_frame_start(s, s->avctx) < 0)
2164 return -1;
2165 ff_er_frame_start(s);
2166 /*
2167 * MPV_frame_start uses pict_type to derive key_frame.
2168 * This is incorrect for H.264; IDR markings must be used.
2169 * Zero here; IDR markings per slice in frame or fields are OR'd in later.
2170 * See decode_nal_units().
2171 */
2172 s->current_picture_ptr->key_frame= 0;
2173
2174 assert(s->linesize && s->uvlinesize);
2175
2176 for(i=0; i<16; i++){
2177 h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
2178 h->block_offset[24+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
2179 }
2180 for(i=0; i<4; i++){
2181 h->block_offset[16+i]=
2182 h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2183 h->block_offset[24+16+i]=
2184 h->block_offset[24+20+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2185 }
2186
2187 /* can't be in alloc_tables because linesize isn't known there.
2188 * FIXME: redo bipred weight to not require extra buffer? */
2189 for(i = 0; i < s->avctx->thread_count; i++)
2190 if(!h->thread_context[i]->s.obmc_scratchpad)
2191 h->thread_context[i]->s.obmc_scratchpad = av_malloc(16*2*s->linesize + 8*2*s->uvlinesize);
2192
2193 /* some macroblocks will be accessed before they're available */
2194 if(FRAME_MBAFF || s->avctx->thread_count > 1)
2195 memset(h->slice_table, -1, (s->mb_height*s->mb_stride-1) * sizeof(uint8_t));
2196
2197 // s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
2198 return 0;
2199 }
2200
2201 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){
2202 MpegEncContext * const s = &h->s;
2203 int i;
2204
2205 src_y -= linesize;
2206 src_cb -= uvlinesize;
2207 src_cr -= uvlinesize;
2208
2209 // There are two lines saved, the line above the the top macroblock of a pair,
2210 // and the line above the bottom macroblock
2211 h->left_border[0]= h->top_borders[0][s->mb_x][15];
2212 for(i=1; i<17; i++){
2213 h->left_border[i]= src_y[15+i* linesize];
2214 }
2215
2216 *(uint64_t*)(h->top_borders[0][s->mb_x]+0)= *(uint64_t*)(src_y + 16*linesize);
2217 *(uint64_t*)(h->top_borders[0][s->mb_x]+8)= *(uint64_t*)(src_y +8+16*linesize);
2218
2219 if(simple || !ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2220 h->left_border[17 ]= h->top_borders[0][s->mb_x][16+7];
2221 h->left_border[17+9]= h->top_borders[0][s->mb_x][24+7];
2222 for(i=1; i<9; i++){
2223 h->left_border[i+17 ]= src_cb[7+i*uvlinesize];
2224 h->left_border[i+17+9]= src_cr[7+i*uvlinesize];
2225 }
2226 *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+8*uvlinesize);
2227 *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+8*uvlinesize);
2228 }
2229 }
2230
2231 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){
2232 MpegEncContext * const s = &h->s;
2233 int temp8, i;
2234 uint64_t temp64;
2235 int deblock_left;
2236 int deblock_top;
2237 int mb_xy;
2238
2239 if(h->deblocking_filter == 2) {
2240 mb_xy = s->mb_x + s->mb_y*s->mb_stride;
2241 deblock_left = h->slice_table[mb_xy] == h->slice_table[mb_xy - 1];
2242 deblock_top = h->slice_table[mb_xy] == h->slice_table[h->top_mb_xy];
2243 } else {
2244 deblock_left = (s->mb_x > 0);
2245 deblock_top = (s->mb_y > 0);
2246 }
2247
2248 src_y -= linesize + 1;
2249 src_cb -= uvlinesize + 1;
2250 src_cr -= uvlinesize + 1;
2251
2252 #define XCHG(a,b,t,xchg)\
2253 t= a;\
2254 if(xchg)\
2255 a= b;\
2256 b= t;
2257
2258 if(deblock_left){
2259 for(i = !deblock_top; i<17; i++){
2260 XCHG(h->left_border[i ], src_y [i* linesize], temp8, xchg);
2261 }
2262 }
2263
2264 if(deblock_top){
2265 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
2266 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
2267 if(s->mb_x+1 < s->mb_width){
2268 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x+1]), *(uint64_t*)(src_y +17), temp64, 1);
2269 }
2270 }
2271
2272 if(simple || !ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2273 if(deblock_left){
2274 for(i = !deblock_top; i<9; i++){
2275 XCHG(h->left_border[i+17 ], src_cb[i*uvlinesize], temp8, xchg);
2276 XCHG(h->left_border[i+17+9], src_cr[i*uvlinesize], temp8, xchg);
2277 }
2278 }
2279 if(deblock_top){
2280 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
2281 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
2282 }
2283 }
2284 }
2285
2286 static inline void backup_pair_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){
2287 MpegEncContext * const s = &h->s;
2288 int i;
2289
2290 src_y -= 2 * linesize;
2291 src_cb -= 2 * uvlinesize;
2292 src_cr -= 2 * uvlinesize;
2293
2294 // There are two lines saved, the line above the the top macroblock of a pair,
2295 // and the line above the bottom macroblock
2296 h->left_border[0]= h->top_borders[0][s->mb_x][15];
2297 h->left_border[1]= h->top_borders[1][s->mb_x][15];
2298 for(i=2; i<34; i++){
2299 h->left_border[i]= src_y[15+i* linesize];
2300 }
2301
2302 *(uint64_t*)(h->top_borders[0][s->mb_x]+0)= *(uint64_t*)(src_y + 32*linesize);
2303 *(uint64_t*)(h->top_borders[0][s->mb_x]+8)= *(uint64_t*)(src_y +8+32*linesize);
2304 *(uint64_t*)(h->top_borders[1][s->mb_x]+0)= *(uint64_t*)(src_y + 33*linesize);
2305 *(uint64_t*)(h->top_borders[1][s->mb_x]+8)= *(uint64_t*)(src_y +8+33*linesize);
2306
2307 if(!ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2308 h->left_border[34 ]= h->top_borders[0][s->mb_x][16+7];
2309 h->left_border[34+ 1]= h->top_borders[1][s->mb_x][16+7];
2310 h->left_border[34+18 ]= h->top_borders[0][s->mb_x][24+7];
2311 h->left_border[34+18+1]= h->top_borders[1][s->mb_x][24+7];
2312 for(i=2; i<18; i++){
2313 h->left_border[i+34 ]= src_cb[7+i*uvlinesize];
2314 h->left_border[i+34+18]= src_cr[7+i*uvlinesize];
2315 }
2316 *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+16*uvlinesize);
2317 *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+16*uvlinesize);
2318 *(uint64_t*)(h->top_borders[1][s->mb_x]+16)= *(uint64_t*)(src_cb+17*uvlinesize);
2319 *(uint64_t*)(h->top_borders[1][s->mb_x]+24)= *(uint64_t*)(src_cr+17*uvlinesize);
2320 }
2321 }
2322
2323 static inline void xchg_pair_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize, int xchg){
2324 MpegEncContext * const s = &h->s;
2325 int temp8, i;
2326 uint64_t temp64;
2327 int deblock_left = (s->mb_x > 0);
2328 int deblock_top = (s->mb_y > 1);
2329
2330 tprintf(s->avctx, "xchg_pair_border: src_y:%p src_cb:%p src_cr:%p ls:%d uvls:%d\n", src_y, src_cb, src_cr, linesize, uvlinesize);
2331
2332 src_y -= 2 * linesize + 1;
2333 src_cb -= 2 * uvlinesize + 1;
2334 src_cr -= 2 * uvlinesize + 1;
2335
2336 #define XCHG(a,b,t,xchg)\
2337 t= a;\
2338 if(xchg)\
2339 a= b;\
2340 b= t;
2341
2342 if(deblock_left){
2343 for(i = (!deblock_top)<<1; i<34; i++){
2344 XCHG(h->left_border[i ], src_y [i* linesize], temp8, xchg);
2345 }
2346 }
2347
2348 if(deblock_top){
2349 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
2350 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
2351 XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+0), *(uint64_t*)(src_y +1 +linesize), temp64, xchg);
2352 XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+8), *(uint64_t*)(src_y +9 +linesize), temp64, 1);
2353 if(s->mb_x+1 < s->mb_width){
2354 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x+1]), *(uint64_t*)(src_y +17), temp64, 1);
2355 XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x+1]), *(uint64_t*)(src_y +17 +linesize), temp64, 1);
2356 }
2357 }
2358
2359 if(!ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2360 if(deblock_left){
2361 for(i = (!deblock_top) << 1; i<18; i++){
2362 XCHG(h->left_border[i+34 ], src_cb[i*uvlinesize], temp8, xchg);
2363 XCHG(h->left_border[i+34+18], src_cr[i*uvlinesize], temp8, xchg);
2364 }
2365 }
2366 if(deblock_top){
2367 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
2368 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
2369 XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+16), *(uint64_t*)(src_cb+1 +uvlinesize), temp64, 1);
2370 XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+24), *(uint64_t*)(src_cr+1 +uvlinesize), temp64, 1);
2371 }
2372 }
2373 }
2374
2375 static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
2376 MpegEncContext * const s = &h->s;
2377 const int mb_x= s->mb_x;
2378 const int mb_y= s->mb_y;
2379 const int mb_xy= mb_x + mb_y*s->mb_stride;
2380 const int mb_type= s->current_picture.mb_type[mb_xy];
2381 uint8_t *dest_y, *dest_cb, *dest_cr;
2382 int linesize, uvlinesize /*dct_offset*/;
2383 int i;
2384 int *block_offset = &h->block_offset[0];
2385 const unsigned int bottom = mb_y & 1;
2386 const int transform_bypass = (s->qscale == 0 && h->sps.transform_bypass), is_h264 = (simple || s->codec_id == CODEC_ID_H264);
2387 void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
2388 void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
2389
2390 dest_y = s->current_picture.data[0] + (mb_y * 16* s->linesize ) + mb_x * 16;
2391 dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2392 dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2393
2394 s->dsp.prefetch(dest_y + (s->mb_x&3)*4*s->linesize + 64, s->linesize, 4);
2395 s->dsp.prefetch(dest_cb + (s->mb_x&7)*s->uvlinesize + 64, dest_cr - dest_cb, 2);
2396
2397 if (!simple && MB_FIELD) {
2398 linesize = h->mb_linesize = s->linesize * 2;
2399 uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
2400 block_offset = &h->block_offset[24];
2401 if(mb_y&1){ //FIXME move out of this func?
2402 dest_y -= s->linesize*15;
2403 dest_cb-= s->uvlinesize*7;
2404 dest_cr-= s->uvlinesize*7;
2405 }
2406 if(FRAME_MBAFF) {
2407 int list;
2408 for(list=0; list<h->list_count; list++){
2409 if(!USES_LIST(mb_type, list))
2410 continue;
2411 if(IS_16X16(mb_type)){
2412 int8_t *ref = &h->ref_cache[list][scan8[0]];
2413 fill_rectangle(ref, 4, 4, 8, 16+*ref^(s->mb_y&1), 1);
2414 }else{
2415 for(i=0; i<16; i+=4){
2416 //FIXME can refs be smaller than 8x8 when !direct_8x8_inference ?
2417 int ref = h->ref_cache[list][scan8[i]];
2418 if(ref >= 0)
2419 fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, 16+ref^(s->mb_y&1), 1);
2420 }
2421 }
2422 }
2423 }
2424 } else {
2425 linesize = h->mb_linesize = s->linesize;
2426 uvlinesize = h->mb_uvlinesize = s->uvlinesize;
2427 // dct_offset = s->linesize * 16;
2428 }
2429
2430 if(transform_bypass){
2431 idct_dc_add =
2432 idct_add = IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4;
2433 }else if(IS_8x8DCT(mb_type)){
2434 idct_dc_add = s->dsp.h264_idct8_dc_add;
2435 idct_add = s->dsp.h264_idct8_add;
2436 }else{
2437 idct_dc_add = s->dsp.h264_idct_dc_add;
2438 idct_add = s->dsp.h264_idct_add;
2439 }
2440
2441 if(!simple && FRAME_MBAFF && h->deblocking_filter && IS_INTRA(mb_type)
2442 && (!bottom || !IS_INTRA(s->current_picture.mb_type[mb_xy-s->mb_stride]))){
2443 int mbt_y = mb_y&~1;
2444 uint8_t *top_y = s->current_picture.data[0] + (mbt_y * 16* s->linesize ) + mb_x * 16;
2445 uint8_t *top_cb = s->current_picture.data[1] + (mbt_y * 8 * s->uvlinesize) + mb_x * 8;
2446 uint8_t *top_cr = s->current_picture.data[2] + (mbt_y * 8 * s->uvlinesize) + mb_x * 8;
2447 xchg_pair_border(h, top_y, top_cb, top_cr, s->linesize, s->uvlinesize, 1);
2448 }
2449
2450 if (!simple && IS_INTRA_PCM(mb_type)) {
2451 unsigned int x, y;
2452
2453 // The pixels are stored in h->mb array in the same order as levels,
2454 // copy them in output in the correct order.
2455 for(i=0; i<16; i++) {
2456 for (y=0; y<4; y++) {
2457 for (x=0; x<4; x++) {
2458 *(dest_y + block_offset[i] + y*linesize + x) = h->mb[i*16+y*4+x];
2459 }
2460 }
2461 }
2462 for(i=16; i<16+4; i++) {
2463 for (y=0; y<4; y++) {
2464 for (x=0; x<4; x++) {
2465 *(dest_cb + block_offset[i] + y*uvlinesize + x) = h->mb[i*16+y*4+x];
2466 }
2467 }
2468 }
2469 for(i=20; i<20+4; i++) {
2470 for (y=0; y<4; y++) {
2471 for (x=0; x<4; x++) {
2472 *(dest_cr + block_offset[i] + y*uvlinesize + x) = h->mb[i*16+y*4+x];
2473 }
2474 }
2475 }
2476 } else {
2477 if(IS_INTRA(mb_type)){
2478 if(h->deblocking_filter && (simple || !FRAME_MBAFF))
2479 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1, simple);
2480
2481 if(simple || !ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2482 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
2483 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
2484 }
2485
2486 if(IS_INTRA4x4(mb_type)){
2487 if(simple || !s->encoding){
2488 if(IS_8x8DCT(mb_type)){
2489 for(i=0; i<16; i+=4){
2490 uint8_t * const ptr= dest_y + block_offset[i];
2491 const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
2492 const int nnz = h->non_zero_count_cache[ scan8[i] ];
2493 h->hpc.pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
2494 (h->topright_samples_available<<i)&0x4000, linesize);
2495 if(nnz){
2496 if(nnz == 1 && h->mb[i*16])
2497 idct_dc_add(ptr, h->mb + i*16, linesize);
2498 else
2499 idct_add(ptr, h->mb + i*16, linesize);
2500 }
2501 }
2502 }else
2503 for(i=0; i<16; i++){
2504 uint8_t * const ptr= dest_y + block_offset[i];
2505 uint8_t *topright;
2506 const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
2507 int nnz, tr;
2508
2509 if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
2510 const int topright_avail= (h->topright_samples_available<<i)&0x8000;
2511 assert(mb_y || linesize <= block_offset[i]);
2512 if(!topright_avail){
2513 tr= ptr[3 - linesize]*0x01010101;
2514 topright= (uint8_t*) &tr;
2515 }else
2516 topright= ptr + 4 - linesize;
2517 }else
2518 topright= NULL;
2519
2520 h->hpc.pred4x4[ dir ](ptr, topright, linesize);
2521 nnz = h->non_zero_count_cache[ scan8[i] ];
2522 if(nnz){
2523 if(is_h264){
2524 if(nnz == 1 && h->mb[i*16])
2525 idct_dc_add(ptr, h->mb + i*16, linesize);
2526 else
2527 idct_add(ptr, h->mb + i*16, linesize);
2528 }else
2529 svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
2530 }
2531 }
2532 }
2533 }else{
2534 h