f70ad50ce4da812f14dd038d0d8baad7ddf49b45
[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 /**
24 * @file h264.c
25 * H.264 / AVC / MPEG4 part10 codec.
26 * @author Michael Niedermayer <michaelni@gmx.at>
27 */
28
29 #include "dsputil.h"
30 #include "avcodec.h"
31 #include "mpegvideo.h"
32 #include "h264.h"
33 #include "h264data.h"
34 #include "h264_parser.h"
35 #include "golomb.h"
36
37 #include "cabac.h"
38
39 //#undef NDEBUG
40 #include <assert.h>
41
42 static VLC coeff_token_vlc[4];
43 static VLC chroma_dc_coeff_token_vlc;
44
45 static VLC total_zeros_vlc[15];
46 static VLC chroma_dc_total_zeros_vlc[3];
47
48 static VLC run_vlc[6];
49 static VLC run7_vlc;
50
51 static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
52 static void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
53 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);
54 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);
55
56 static av_always_inline uint32_t pack16to32(int a, int b){
57 #ifdef WORDS_BIGENDIAN
58 return (b&0xFFFF) + (a<<16);
59 #else
60 return (a&0xFFFF) + (b<<16);
61 #endif
62 }
63
64 const uint8_t ff_rem6[52]={
65 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,
66 };
67
68 const uint8_t ff_div6[52]={
69 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,
70 };
71
72
73 /**
74 * fill a rectangle.
75 * @param h height of the rectangle, should be a constant
76 * @param w width of the rectangle, should be a constant
77 * @param size the size of val (1 or 4), should be a constant
78 */
79 static av_always_inline void fill_rectangle(void *vp, int w, int h, int stride, uint32_t val, int size){
80 uint8_t *p= (uint8_t*)vp;
81 assert(size==1 || size==4);
82 assert(w<=4);
83
84 w *= size;
85 stride *= size;
86
87 assert((((long)vp)&(FFMIN(w, STRIDE_ALIGN)-1)) == 0);
88 assert((stride&(w-1))==0);
89 if(w==2){
90 const uint16_t v= size==4 ? val : val*0x0101;
91 *(uint16_t*)(p + 0*stride)= v;
92 if(h==1) return;
93 *(uint16_t*)(p + 1*stride)= v;
94 if(h==2) return;
95 *(uint16_t*)(p + 2*stride)=
96 *(uint16_t*)(p + 3*stride)= v;
97 }else if(w==4){
98 const uint32_t v= size==4 ? val : val*0x01010101;
99 *(uint32_t*)(p + 0*stride)= v;
100 if(h==1) return;
101 *(uint32_t*)(p + 1*stride)= v;
102 if(h==2) return;
103 *(uint32_t*)(p + 2*stride)=
104 *(uint32_t*)(p + 3*stride)= v;
105 }else if(w==8){
106 //gcc can't optimize 64bit math on x86_32
107 #if defined(ARCH_X86_64) || (defined(MP_WORDSIZE) && MP_WORDSIZE >= 64)
108 const uint64_t v= val*0x0100000001ULL;
109 *(uint64_t*)(p + 0*stride)= v;
110 if(h==1) return;
111 *(uint64_t*)(p + 1*stride)= v;
112 if(h==2) return;
113 *(uint64_t*)(p + 2*stride)=
114 *(uint64_t*)(p + 3*stride)= v;
115 }else if(w==16){
116 const uint64_t v= val*0x0100000001ULL;
117 *(uint64_t*)(p + 0+0*stride)=
118 *(uint64_t*)(p + 8+0*stride)=
119 *(uint64_t*)(p + 0+1*stride)=
120 *(uint64_t*)(p + 8+1*stride)= v;
121 if(h==2) return;
122 *(uint64_t*)(p + 0+2*stride)=
123 *(uint64_t*)(p + 8+2*stride)=
124 *(uint64_t*)(p + 0+3*stride)=
125 *(uint64_t*)(p + 8+3*stride)= v;
126 #else
127 *(uint32_t*)(p + 0+0*stride)=
128 *(uint32_t*)(p + 4+0*stride)= val;
129 if(h==1) return;
130 *(uint32_t*)(p + 0+1*stride)=
131 *(uint32_t*)(p + 4+1*stride)= val;
132 if(h==2) return;
133 *(uint32_t*)(p + 0+2*stride)=
134 *(uint32_t*)(p + 4+2*stride)=
135 *(uint32_t*)(p + 0+3*stride)=
136 *(uint32_t*)(p + 4+3*stride)= val;
137 }else if(w==16){
138 *(uint32_t*)(p + 0+0*stride)=
139 *(uint32_t*)(p + 4+0*stride)=
140 *(uint32_t*)(p + 8+0*stride)=
141 *(uint32_t*)(p +12+0*stride)=
142 *(uint32_t*)(p + 0+1*stride)=
143 *(uint32_t*)(p + 4+1*stride)=
144 *(uint32_t*)(p + 8+1*stride)=
145 *(uint32_t*)(p +12+1*stride)= val;
146 if(h==2) return;
147 *(uint32_t*)(p + 0+2*stride)=
148 *(uint32_t*)(p + 4+2*stride)=
149 *(uint32_t*)(p + 8+2*stride)=
150 *(uint32_t*)(p +12+2*stride)=
151 *(uint32_t*)(p + 0+3*stride)=
152 *(uint32_t*)(p + 4+3*stride)=
153 *(uint32_t*)(p + 8+3*stride)=
154 *(uint32_t*)(p +12+3*stride)= val;
155 #endif
156 }else
157 assert(0);
158 assert(h==4);
159 }
160
161 static void fill_caches(H264Context *h, int mb_type, int for_deblock){
162 MpegEncContext * const s = &h->s;
163 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
164 int topleft_xy, top_xy, topright_xy, left_xy[2];
165 int topleft_type, top_type, topright_type, left_type[2];
166 int left_block[8];
167 int i;
168
169 //FIXME deblocking could skip the intra and nnz parts.
170 if(for_deblock && (h->slice_num == 1 || h->slice_table[mb_xy] == h->slice_table[mb_xy-s->mb_stride]) && !FRAME_MBAFF)
171 return;
172
173 //wow what a mess, why didn't they simplify the interlacing&intra stuff, i can't imagine that these complex rules are worth it
174
175 top_xy = mb_xy - s->mb_stride;
176 topleft_xy = top_xy - 1;
177 topright_xy= top_xy + 1;
178 left_xy[1] = left_xy[0] = mb_xy-1;
179 left_block[0]= 0;
180 left_block[1]= 1;
181 left_block[2]= 2;
182 left_block[3]= 3;
183 left_block[4]= 7;
184 left_block[5]= 10;
185 left_block[6]= 8;
186 left_block[7]= 11;
187 if(FRAME_MBAFF){
188 const int pair_xy = s->mb_x + (s->mb_y & ~1)*s->mb_stride;
189 const int top_pair_xy = pair_xy - s->mb_stride;
190 const int topleft_pair_xy = top_pair_xy - 1;
191 const int topright_pair_xy = top_pair_xy + 1;
192 const int topleft_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[topleft_pair_xy]);
193 const int top_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[top_pair_xy]);
194 const int topright_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[topright_pair_xy]);
195 const int left_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[pair_xy-1]);
196 const int curr_mb_frame_flag = !IS_INTERLACED(mb_type);
197 const int bottom = (s->mb_y & 1);
198 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);
199 if (bottom
200 ? !curr_mb_frame_flag // bottom macroblock
201 : (!curr_mb_frame_flag && !top_mb_frame_flag) // top macroblock
202 ) {
203 top_xy -= s->mb_stride;
204 }
205 if (bottom
206 ? !curr_mb_frame_flag // bottom macroblock
207 : (!curr_mb_frame_flag && !topleft_mb_frame_flag) // top macroblock
208 ) {
209 topleft_xy -= s->mb_stride;
210 }
211 if (bottom
212 ? !curr_mb_frame_flag // bottom macroblock
213 : (!curr_mb_frame_flag && !topright_mb_frame_flag) // top macroblock
214 ) {
215 topright_xy -= s->mb_stride;
216 }
217 if (left_mb_frame_flag != curr_mb_frame_flag) {
218 left_xy[1] = left_xy[0] = pair_xy - 1;
219 if (curr_mb_frame_flag) {
220 if (bottom) {
221 left_block[0]= 2;
222 left_block[1]= 2;
223 left_block[2]= 3;
224 left_block[3]= 3;
225 left_block[4]= 8;
226 left_block[5]= 11;
227 left_block[6]= 8;
228 left_block[7]= 11;
229 } else {
230 left_block[0]= 0;
231 left_block[1]= 0;
232 left_block[2]= 1;
233 left_block[3]= 1;
234 left_block[4]= 7;
235 left_block[5]= 10;
236 left_block[6]= 7;
237 left_block[7]= 10;
238 }
239 } else {
240 left_xy[1] += s->mb_stride;
241 //left_block[0]= 0;
242 left_block[1]= 2;
243 left_block[2]= 0;
244 left_block[3]= 2;
245 //left_block[4]= 7;
246 left_block[5]= 10;
247 left_block[6]= 7;
248 left_block[7]= 10;
249 }
250 }
251 }
252
253 h->top_mb_xy = top_xy;
254 h->left_mb_xy[0] = left_xy[0];
255 h->left_mb_xy[1] = left_xy[1];
256 if(for_deblock){
257 topleft_type = 0;
258 topright_type = 0;
259 top_type = h->slice_table[top_xy ] < 255 ? s->current_picture.mb_type[top_xy] : 0;
260 left_type[0] = h->slice_table[left_xy[0] ] < 255 ? s->current_picture.mb_type[left_xy[0]] : 0;
261 left_type[1] = h->slice_table[left_xy[1] ] < 255 ? s->current_picture.mb_type[left_xy[1]] : 0;
262
263 if(FRAME_MBAFF && !IS_INTRA(mb_type)){
264 int list;
265 int v = *(uint16_t*)&h->non_zero_count[mb_xy][14];
266 for(i=0; i<16; i++)
267 h->non_zero_count_cache[scan8[i]] = (v>>i)&1;
268 for(list=0; list<h->list_count; list++){
269 if(USES_LIST(mb_type,list)){
270 uint32_t *src = (uint32_t*)s->current_picture.motion_val[list][h->mb2b_xy[mb_xy]];
271 uint32_t *dst = (uint32_t*)h->mv_cache[list][scan8[0]];
272 int8_t *ref = &s->current_picture.ref_index[list][h->mb2b8_xy[mb_xy]];
273 for(i=0; i<4; i++, dst+=8, src+=h->b_stride){
274 dst[0] = src[0];
275 dst[1] = src[1];
276 dst[2] = src[2];
277 dst[3] = src[3];
278 }
279 *(uint32_t*)&h->ref_cache[list][scan8[ 0]] =
280 *(uint32_t*)&h->ref_cache[list][scan8[ 2]] = pack16to32(ref[0],ref[1])*0x0101;
281 ref += h->b8_stride;
282 *(uint32_t*)&h->ref_cache[list][scan8[ 8]] =
283 *(uint32_t*)&h->ref_cache[list][scan8[10]] = pack16to32(ref[0],ref[1])*0x0101;
284 }else{
285 fill_rectangle(&h-> mv_cache[list][scan8[ 0]], 4, 4, 8, 0, 4);
286 fill_rectangle(&h->ref_cache[list][scan8[ 0]], 4, 4, 8, (uint8_t)LIST_NOT_USED, 1);
287 }
288 }
289 }
290 }else{
291 topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
292 top_type = h->slice_table[top_xy ] == h->slice_num ? s->current_picture.mb_type[top_xy] : 0;
293 topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
294 left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
295 left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
296 }
297
298 if(IS_INTRA(mb_type)){
299 h->topleft_samples_available=
300 h->top_samples_available=
301 h->left_samples_available= 0xFFFF;
302 h->topright_samples_available= 0xEEEA;
303
304 if(!IS_INTRA(top_type) && (top_type==0 || h->pps.constrained_intra_pred)){
305 h->topleft_samples_available= 0xB3FF;
306 h->top_samples_available= 0x33FF;
307 h->topright_samples_available= 0x26EA;
308 }
309 for(i=0; i<2; i++){
310 if(!IS_INTRA(left_type[i]) && (left_type[i]==0 || h->pps.constrained_intra_pred)){
311 h->topleft_samples_available&= 0xDF5F;
312 h->left_samples_available&= 0x5F5F;
313 }
314 }
315
316 if(!IS_INTRA(topleft_type) && (topleft_type==0 || h->pps.constrained_intra_pred))
317 h->topleft_samples_available&= 0x7FFF;
318
319 if(!IS_INTRA(topright_type) && (topright_type==0 || h->pps.constrained_intra_pred))
320 h->topright_samples_available&= 0xFBFF;
321
322 if(IS_INTRA4x4(mb_type)){
323 if(IS_INTRA4x4(top_type)){
324 h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
325 h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
326 h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
327 h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
328 }else{
329 int pred;
330 if(!top_type || (IS_INTER(top_type) && h->pps.constrained_intra_pred))
331 pred= -1;
332 else{
333 pred= 2;
334 }
335 h->intra4x4_pred_mode_cache[4+8*0]=
336 h->intra4x4_pred_mode_cache[5+8*0]=
337 h->intra4x4_pred_mode_cache[6+8*0]=
338 h->intra4x4_pred_mode_cache[7+8*0]= pred;
339 }
340 for(i=0; i<2; i++){
341 if(IS_INTRA4x4(left_type[i])){
342 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
343 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
344 }else{
345 int pred;
346 if(!left_type[i] || (IS_INTER(left_type[i]) && h->pps.constrained_intra_pred))
347 pred= -1;
348 else{
349 pred= 2;
350 }
351 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
352 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
353 }
354 }
355 }
356 }
357
358
359 /*
360 0 . T T. T T T T
361 1 L . .L . . . .
362 2 L . .L . . . .
363 3 . T TL . . . .
364 4 L . .L . . . .
365 5 L . .. . . . .
366 */
367 //FIXME constraint_intra_pred & partitioning & nnz (lets hope this is just a typo in the spec)
368 if(top_type){
369 h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][4];
370 h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][5];
371 h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][6];
372 h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3];
373
374 h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][9];
375 h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8];
376
377 h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][12];
378 h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11];
379
380 }else{
381 h->non_zero_count_cache[4+8*0]=
382 h->non_zero_count_cache[5+8*0]=
383 h->non_zero_count_cache[6+8*0]=
384 h->non_zero_count_cache[7+8*0]=
385
386 h->non_zero_count_cache[1+8*0]=
387 h->non_zero_count_cache[2+8*0]=
388
389 h->non_zero_count_cache[1+8*3]=
390 h->non_zero_count_cache[2+8*3]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
391
392 }
393
394 for (i=0; i<2; i++) {
395 if(left_type[i]){
396 h->non_zero_count_cache[3+8*1 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[0+2*i]];
397 h->non_zero_count_cache[3+8*2 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[1+2*i]];
398 h->non_zero_count_cache[0+8*1 + 8*i]= h->non_zero_count[left_xy[i]][left_block[4+2*i]];
399 h->non_zero_count_cache[0+8*4 + 8*i]= h->non_zero_count[left_xy[i]][left_block[5+2*i]];
400 }else{
401 h->non_zero_count_cache[3+8*1 + 2*8*i]=
402 h->non_zero_count_cache[3+8*2 + 2*8*i]=
403 h->non_zero_count_cache[0+8*1 + 8*i]=
404 h->non_zero_count_cache[0+8*4 + 8*i]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
405 }
406 }
407
408 if( h->pps.cabac ) {
409 // top_cbp
410 if(top_type) {
411 h->top_cbp = h->cbp_table[top_xy];
412 } else if(IS_INTRA(mb_type)) {
413 h->top_cbp = 0x1C0;
414 } else {
415 h->top_cbp = 0;
416 }
417 // left_cbp
418 if (left_type[0]) {
419 h->left_cbp = h->cbp_table[left_xy[0]] & 0x1f0;
420 } else if(IS_INTRA(mb_type)) {
421 h->left_cbp = 0x1C0;
422 } else {
423 h->left_cbp = 0;
424 }
425 if (left_type[0]) {
426 h->left_cbp |= ((h->cbp_table[left_xy[0]]>>((left_block[0]&(~1))+1))&0x1) << 1;
427 }
428 if (left_type[1]) {
429 h->left_cbp |= ((h->cbp_table[left_xy[1]]>>((left_block[2]&(~1))+1))&0x1) << 3;
430 }
431 }
432
433 #if 1
434 if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
435 int list;
436 for(list=0; list<h->list_count; list++){
437 if(!USES_LIST(mb_type, list) && !IS_DIRECT(mb_type) && !h->deblocking_filter){
438 /*if(!h->mv_cache_clean[list]){
439 memset(h->mv_cache [list], 0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
440 memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
441 h->mv_cache_clean[list]= 1;
442 }*/
443 continue;
444 }
445 h->mv_cache_clean[list]= 0;
446
447 if(USES_LIST(top_type, list)){
448 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
449 const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
450 *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0];
451 *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];
452 *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];
453 *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];
454 h->ref_cache[list][scan8[0] + 0 - 1*8]=
455 h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
456 h->ref_cache[list][scan8[0] + 2 - 1*8]=
457 h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
458 }else{
459 *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]=
460 *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]=
461 *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]=
462 *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;
463 *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
464 }
465
466 for(i=0; i<2; i++){
467 int cache_idx = scan8[0] - 1 + i*2*8;
468 if(USES_LIST(left_type[i], list)){
469 const int b_xy= h->mb2b_xy[left_xy[i]] + 3;
470 const int b8_xy= h->mb2b8_xy[left_xy[i]] + 1;
471 *(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]];
472 *(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]];
473 h->ref_cache[list][cache_idx ]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0+i*2]>>1)];
474 h->ref_cache[list][cache_idx+8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[1+i*2]>>1)];
475 }else{
476 *(uint32_t*)h->mv_cache [list][cache_idx ]=
477 *(uint32_t*)h->mv_cache [list][cache_idx+8]= 0;
478 h->ref_cache[list][cache_idx ]=
479 h->ref_cache[list][cache_idx+8]= left_type[i] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
480 }
481 }
482
483 if((for_deblock || (IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred)) && !FRAME_MBAFF)
484 continue;
485
486 if(USES_LIST(topleft_type, list)){
487 const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
488 const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + h->b8_stride;
489 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
490 h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
491 }else{
492 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
493 h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
494 }
495
496 if(USES_LIST(topright_type, list)){
497 const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
498 const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
499 *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
500 h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
501 }else{
502 *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
503 h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
504 }
505
506 if((IS_SKIP(mb_type) || IS_DIRECT(mb_type)) && !FRAME_MBAFF)
507 continue;
508
509 h->ref_cache[list][scan8[5 ]+1] =
510 h->ref_cache[list][scan8[7 ]+1] =
511 h->ref_cache[list][scan8[13]+1] = //FIXME remove past 3 (init somewhere else)
512 h->ref_cache[list][scan8[4 ]] =
513 h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
514 *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
515 *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
516 *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
517 *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
518 *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
519
520 if( h->pps.cabac ) {
521 /* XXX beurk, Load mvd */
522 if(USES_LIST(top_type, list)){
523 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
524 *(uint32_t*)h->mvd_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 0];
525 *(uint32_t*)h->mvd_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 1];
526 *(uint32_t*)h->mvd_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 2];
527 *(uint32_t*)h->mvd_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 3];
528 }else{
529 *(uint32_t*)h->mvd_cache [list][scan8[0] + 0 - 1*8]=
530 *(uint32_t*)h->mvd_cache [list][scan8[0] + 1 - 1*8]=
531 *(uint32_t*)h->mvd_cache [list][scan8[0] + 2 - 1*8]=
532 *(uint32_t*)h->mvd_cache [list][scan8[0] + 3 - 1*8]= 0;
533 }
534 if(USES_LIST(left_type[0], list)){
535 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
536 *(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]];
537 *(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]];
538 }else{
539 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 0*8]=
540 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 1*8]= 0;
541 }
542 if(USES_LIST(left_type[1], list)){
543 const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
544 *(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]];
545 *(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]];
546 }else{
547 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 2*8]=
548 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 3*8]= 0;
549 }
550 *(uint32_t*)h->mvd_cache [list][scan8[5 ]+1]=
551 *(uint32_t*)h->mvd_cache [list][scan8[7 ]+1]=
552 *(uint32_t*)h->mvd_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
553 *(uint32_t*)h->mvd_cache [list][scan8[4 ]]=
554 *(uint32_t*)h->mvd_cache [list][scan8[12]]= 0;
555
556 if(h->slice_type == B_TYPE){
557 fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, 0, 1);
558
559 if(IS_DIRECT(top_type)){
560 *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0x01010101;
561 }else if(IS_8X8(top_type)){
562 int b8_xy = h->mb2b8_xy[top_xy] + h->b8_stride;
563 h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy];
564 h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 1];
565 }else{
566 *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0;
567 }
568
569 if(IS_DIRECT(left_type[0]))
570 h->direct_cache[scan8[0] - 1 + 0*8]= 1;
571 else if(IS_8X8(left_type[0]))
572 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)];
573 else
574 h->direct_cache[scan8[0] - 1 + 0*8]= 0;
575
576 if(IS_DIRECT(left_type[1]))
577 h->direct_cache[scan8[0] - 1 + 2*8]= 1;
578 else if(IS_8X8(left_type[1]))
579 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)];
580 else
581 h->direct_cache[scan8[0] - 1 + 2*8]= 0;
582 }
583 }
584
585 if(FRAME_MBAFF){
586 #define MAP_MVS\
587 MAP_F2F(scan8[0] - 1 - 1*8, topleft_type)\
588 MAP_F2F(scan8[0] + 0 - 1*8, top_type)\
589 MAP_F2F(scan8[0] + 1 - 1*8, top_type)\
590 MAP_F2F(scan8[0] + 2 - 1*8, top_type)\
591 MAP_F2F(scan8[0] + 3 - 1*8, top_type)\
592 MAP_F2F(scan8[0] + 4 - 1*8, topright_type)\
593 MAP_F2F(scan8[0] - 1 + 0*8, left_type[0])\
594 MAP_F2F(scan8[0] - 1 + 1*8, left_type[0])\
595 MAP_F2F(scan8[0] - 1 + 2*8, left_type[1])\
596 MAP_F2F(scan8[0] - 1 + 3*8, left_type[1])
597 if(MB_FIELD){
598 #define MAP_F2F(idx, mb_type)\
599 if(!IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
600 h->ref_cache[list][idx] <<= 1;\
601 h->mv_cache[list][idx][1] /= 2;\
602 h->mvd_cache[list][idx][1] /= 2;\
603 }
604 MAP_MVS
605 #undef MAP_F2F
606 }else{
607 #define MAP_F2F(idx, mb_type)\
608 if(IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
609 h->ref_cache[list][idx] >>= 1;\
610 h->mv_cache[list][idx][1] <<= 1;\
611 h->mvd_cache[list][idx][1] <<= 1;\
612 }
613 MAP_MVS
614 #undef MAP_F2F
615 }
616 }
617 }
618 }
619 #endif
620
621 h->neighbor_transform_size= !!IS_8x8DCT(top_type) + !!IS_8x8DCT(left_type[0]);
622 }
623
624 static inline void write_back_intra_pred_mode(H264Context *h){
625 MpegEncContext * const s = &h->s;
626 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
627
628 h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
629 h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
630 h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
631 h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
632 h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
633 h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
634 h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
635 }
636
637 /**
638 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
639 */
640 static inline int check_intra4x4_pred_mode(H264Context *h){
641 MpegEncContext * const s = &h->s;
642 static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
643 static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
644 int i;
645
646 if(!(h->top_samples_available&0x8000)){
647 for(i=0; i<4; i++){
648 int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
649 if(status<0){
650 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);
651 return -1;
652 } else if(status){
653 h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
654 }
655 }
656 }
657
658 if(!(h->left_samples_available&0x8000)){
659 for(i=0; i<4; i++){
660 int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
661 if(status<0){
662 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);
663 return -1;
664 } else if(status){
665 h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
666 }
667 }
668 }
669
670 return 0;
671 } //FIXME cleanup like next
672
673 /**
674 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
675 */
676 static inline int check_intra_pred_mode(H264Context *h, int mode){
677 MpegEncContext * const s = &h->s;
678 static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
679 static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
680
681 if(mode > 6U) {
682 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);
683 return -1;
684 }
685
686 if(!(h->top_samples_available&0x8000)){
687 mode= top[ mode ];
688 if(mode<0){
689 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);
690 return -1;
691 }
692 }
693
694 if(!(h->left_samples_available&0x8000)){
695 mode= left[ mode ];
696 if(mode<0){
697 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);
698 return -1;
699 }
700 }
701
702 return mode;
703 }
704
705 /**
706 * gets the predicted intra4x4 prediction mode.
707 */
708 static inline int pred_intra_mode(H264Context *h, int n){
709 const int index8= scan8[n];
710 const int left= h->intra4x4_pred_mode_cache[index8 - 1];
711 const int top = h->intra4x4_pred_mode_cache[index8 - 8];
712 const int min= FFMIN(left, top);
713
714 tprintf(h->s.avctx, "mode:%d %d min:%d\n", left ,top, min);
715
716 if(min<0) return DC_PRED;
717 else return min;
718 }
719
720 static inline void write_back_non_zero_count(H264Context *h){
721 MpegEncContext * const s = &h->s;
722 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
723
724 h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[7+8*1];
725 h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[7+8*2];
726 h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[7+8*3];
727 h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];
728 h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[4+8*4];
729 h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[5+8*4];
730 h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[6+8*4];
731
732 h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[1+8*2];
733 h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
734 h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[2+8*1];
735
736 h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[1+8*5];
737 h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
738 h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[2+8*4];
739
740 if(FRAME_MBAFF){
741 // store all luma nnzs, for deblocking
742 int v = 0, i;
743 for(i=0; i<16; i++)
744 v += (!!h->non_zero_count_cache[scan8[i]]) << i;
745 *(uint16_t*)&h->non_zero_count[mb_xy][14] = v;
746 }
747 }
748
749 /**
750 * gets the predicted number of non zero coefficients.
751 * @param n block index
752 */
753 static inline int pred_non_zero_count(H264Context *h, int n){
754 const int index8= scan8[n];
755 const int left= h->non_zero_count_cache[index8 - 1];
756 const int top = h->non_zero_count_cache[index8 - 8];
757 int i= left + top;
758
759 if(i<64) i= (i+1)>>1;
760
761 tprintf(h->s.avctx, "pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
762
763 return i&31;
764 }
765
766 static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
767 const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];
768 MpegEncContext *s = &h->s;
769
770 /* there is no consistent mapping of mvs to neighboring locations that will
771 * make mbaff happy, so we can't move all this logic to fill_caches */
772 if(FRAME_MBAFF){
773 const uint32_t *mb_types = s->current_picture_ptr->mb_type;
774 const int16_t *mv;
775 *(uint32_t*)h->mv_cache[list][scan8[0]-2] = 0;
776 *C = h->mv_cache[list][scan8[0]-2];
777
778 if(!MB_FIELD
779 && (s->mb_y&1) && i < scan8[0]+8 && topright_ref != PART_NOT_AVAILABLE){
780 int topright_xy = s->mb_x + (s->mb_y-1)*s->mb_stride + (i == scan8[0]+3);
781 if(IS_INTERLACED(mb_types[topright_xy])){
782 #define SET_DIAG_MV(MV_OP, REF_OP, X4, Y4)\
783 const int x4 = X4, y4 = Y4;\
784 const int mb_type = mb_types[(x4>>2)+(y4>>2)*s->mb_stride];\
785 if(!USES_LIST(mb_type,list) && !IS_8X8(mb_type))\
786 return LIST_NOT_USED;\
787 mv = s->current_picture_ptr->motion_val[list][x4 + y4*h->b_stride];\
788 h->mv_cache[list][scan8[0]-2][0] = mv[0];\
789 h->mv_cache[list][scan8[0]-2][1] = mv[1] MV_OP;\
790 return s->current_picture_ptr->ref_index[list][(x4>>1) + (y4>>1)*h->b8_stride] REF_OP;
791
792 SET_DIAG_MV(*2, >>1, s->mb_x*4+(i&7)-4+part_width, s->mb_y*4-1);
793 }
794 }
795 if(topright_ref == PART_NOT_AVAILABLE
796 && ((s->mb_y&1) || i >= scan8[0]+8) && (i&7)==4
797 && h->ref_cache[list][scan8[0]-1] != PART_NOT_AVAILABLE){
798 if(!MB_FIELD
799 && IS_INTERLACED(mb_types[h->left_mb_xy[0]])){
800 SET_DIAG_MV(*2, >>1, s->mb_x*4-1, (s->mb_y|1)*4+(s->mb_y&1)*2+(i>>4)-1);
801 }
802 if(MB_FIELD
803 && !IS_INTERLACED(mb_types[h->left_mb_xy[0]])
804 && i >= scan8[0]+8){
805 // leftshift will turn LIST_NOT_USED into PART_NOT_AVAILABLE, but that's ok.
806 SET_DIAG_MV(>>1, <<1, s->mb_x*4-1, (s->mb_y&~1)*4 - 1 + ((i-scan8[0])>>3)*2);
807 }
808 }
809 #undef SET_DIAG_MV
810 }
811
812 if(topright_ref != PART_NOT_AVAILABLE){
813 *C= h->mv_cache[list][ i - 8 + part_width ];
814 return topright_ref;
815 }else{
816 tprintf(s->avctx, "topright MV not available\n");
817
818 *C= h->mv_cache[list][ i - 8 - 1 ];
819 return h->ref_cache[list][ i - 8 - 1 ];
820 }
821 }
822
823 /**
824 * gets the predicted MV.
825 * @param n the block index
826 * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)
827 * @param mx the x component of the predicted motion vector
828 * @param my the y component of the predicted motion vector
829 */
830 static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
831 const int index8= scan8[n];
832 const int top_ref= h->ref_cache[list][ index8 - 8 ];
833 const int left_ref= h->ref_cache[list][ index8 - 1 ];
834 const int16_t * const A= h->mv_cache[list][ index8 - 1 ];
835 const int16_t * const B= h->mv_cache[list][ index8 - 8 ];
836 const int16_t * C;
837 int diagonal_ref, match_count;
838
839 assert(part_width==1 || part_width==2 || part_width==4);
840
841 /* mv_cache
842 B . . A T T T T
843 U . . L . . , .
844 U . . L . . . .
845 U . . L . . , .
846 . . . L . . . .
847 */
848
849 diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);
850 match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);
851 tprintf(h->s.avctx, "pred_motion match_count=%d\n", match_count);
852 if(match_count > 1){ //most common
853 *mx= mid_pred(A[0], B[0], C[0]);
854 *my= mid_pred(A[1], B[1], C[1]);
855 }else if(match_count==1){
856 if(left_ref==ref){
857 *mx= A[0];
858 *my= A[1];
859 }else if(top_ref==ref){
860 *mx= B[0];
861 *my= B[1];
862 }else{
863 *mx= C[0];
864 *my= C[1];
865 }
866 }else{
867 if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
868 *mx= A[0];
869 *my= A[1];
870 }else{
871 *mx= mid_pred(A[0], B[0], C[0]);
872 *my= mid_pred(A[1], B[1], C[1]);
873 }
874 }
875
876 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);
877 }
878
879 /**
880 * gets the directionally predicted 16x8 MV.
881 * @param n the block index
882 * @param mx the x component of the predicted motion vector
883 * @param my the y component of the predicted motion vector
884 */
885 static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
886 if(n==0){
887 const int top_ref= h->ref_cache[list][ scan8[0] - 8 ];
888 const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
889
890 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);
891
892 if(top_ref == ref){
893 *mx= B[0];
894 *my= B[1];
895 return;
896 }
897 }else{
898 const int left_ref= h->ref_cache[list][ scan8[8] - 1 ];
899 const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];
900
901 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);
902
903 if(left_ref == ref){
904 *mx= A[0];
905 *my= A[1];
906 return;
907 }
908 }
909
910 //RARE
911 pred_motion(h, n, 4, list, ref, mx, my);
912 }
913
914 /**
915 * gets the directionally predicted 8x16 MV.
916 * @param n the block index
917 * @param mx the x component of the predicted motion vector
918 * @param my the y component of the predicted motion vector
919 */
920 static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
921 if(n==0){
922 const int left_ref= h->ref_cache[list][ scan8[0] - 1 ];
923 const int16_t * const A= h->mv_cache[list][ scan8[0] - 1 ];
924
925 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);
926
927 if(left_ref == ref){
928 *mx= A[0];
929 *my= A[1];
930 return;
931 }
932 }else{
933 const int16_t * C;
934 int diagonal_ref;
935
936 diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);
937
938 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);
939
940 if(diagonal_ref == ref){
941 *mx= C[0];
942 *my= C[1];
943 return;
944 }
945 }
946
947 //RARE
948 pred_motion(h, n, 2, list, ref, mx, my);
949 }
950
951 static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
952 const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
953 const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];
954
955 tprintf(h->s.avctx, "pred_pskip: (%d) (%d) at %2d %2d\n", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
956
957 if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
958 || (top_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0)
959 || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){
960
961 *mx = *my = 0;
962 return;
963 }
964
965 pred_motion(h, 0, 4, 0, 0, mx, my);
966
967 return;
968 }
969
970 static inline void direct_dist_scale_factor(H264Context * const h){
971 const int poc = h->s.current_picture_ptr->poc;
972 const int poc1 = h->ref_list[1][0].poc;
973 int i;
974 for(i=0; i<h->ref_count[0]; i++){
975 int poc0 = h->ref_list[0][i].poc;
976 int td = av_clip(poc1 - poc0, -128, 127);
977 if(td == 0 /* FIXME || pic0 is a long-term ref */){
978 h->dist_scale_factor[i] = 256;
979 }else{
980 int tb = av_clip(poc - poc0, -128, 127);
981 int tx = (16384 + (FFABS(td) >> 1)) / td;
982 h->dist_scale_factor[i] = av_clip((tb*tx + 32) >> 6, -1024, 1023);
983 }
984 }
985 if(FRAME_MBAFF){
986 for(i=0; i<h->ref_count[0]; i++){
987 h->dist_scale_factor_field[2*i] =
988 h->dist_scale_factor_field[2*i+1] = h->dist_scale_factor[i];
989 }
990 }
991 }
992 static inline void direct_ref_list_init(H264Context * const h){
993 MpegEncContext * const s = &h->s;
994 Picture * const ref1 = &h->ref_list[1][0];
995 Picture * const cur = s->current_picture_ptr;
996 int list, i, j;
997 if(cur->pict_type == I_TYPE)
998 cur->ref_count[0] = 0;
999 if(cur->pict_type != B_TYPE)
1000 cur->ref_count[1] = 0;
1001 for(list=0; list<2; list++){
1002 cur->ref_count[list] = h->ref_count[list];
1003 for(j=0; j<h->ref_count[list]; j++)
1004 cur->ref_poc[list][j] = h->ref_list[list][j].poc;
1005 }
1006 if(cur->pict_type != B_TYPE || h->direct_spatial_mv_pred)
1007 return;
1008 for(list=0; list<2; list++){
1009 for(i=0; i<ref1->ref_count[list]; i++){
1010 const int poc = ref1->ref_poc[list][i];
1011 h->map_col_to_list0[list][i] = 0; /* bogus; fills in for missing frames */
1012 for(j=0; j<h->ref_count[list]; j++)
1013 if(h->ref_list[list][j].poc == poc){
1014 h->map_col_to_list0[list][i] = j;
1015 break;
1016 }
1017 }
1018 }
1019 if(FRAME_MBAFF){
1020 for(list=0; list<2; list++){
1021 for(i=0; i<ref1->ref_count[list]; i++){
1022 j = h->map_col_to_list0[list][i];
1023 h->map_col_to_list0_field[list][2*i] = 2*j;
1024 h->map_col_to_list0_field[list][2*i+1] = 2*j+1;
1025 }
1026 }
1027 }
1028 }
1029
1030 static inline void pred_direct_motion(H264Context * const h, int *mb_type){
1031 MpegEncContext * const s = &h->s;
1032 const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
1033 const int b8_xy = 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1034 const int b4_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1035 const int mb_type_col = h->ref_list[1][0].mb_type[mb_xy];
1036 const int16_t (*l1mv0)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[0][b4_xy];
1037 const int16_t (*l1mv1)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[1][b4_xy];
1038 const int8_t *l1ref0 = &h->ref_list[1][0].ref_index[0][b8_xy];
1039 const int8_t *l1ref1 = &h->ref_list[1][0].ref_index[1][b8_xy];
1040 const int is_b8x8 = IS_8X8(*mb_type);
1041 unsigned int sub_mb_type;
1042 int i8, i4;
1043
1044 #define MB_TYPE_16x16_OR_INTRA (MB_TYPE_16x16|MB_TYPE_INTRA4x4|MB_TYPE_INTRA16x16|MB_TYPE_INTRA_PCM)
1045 if(IS_8X8(mb_type_col) && !h->sps.direct_8x8_inference_flag){
1046 /* FIXME save sub mb types from previous frames (or derive from MVs)
1047 * so we know exactly what block size to use */
1048 sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */
1049 *mb_type = MB_TYPE_8x8|MB_TYPE_L0L1;
1050 }else if(!is_b8x8 && (mb_type_col & MB_TYPE_16x16_OR_INTRA)){
1051 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1052 *mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */
1053 }else{
1054 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1055 *mb_type = MB_TYPE_8x8|MB_TYPE_L0L1;
1056 }
1057 if(!is_b8x8)
1058 *mb_type |= MB_TYPE_DIRECT2;
1059 if(MB_FIELD)
1060 *mb_type |= MB_TYPE_INTERLACED;
1061
1062 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);
1063
1064 if(h->direct_spatial_mv_pred){
1065 int ref[2];
1066 int mv[2][2];
1067 int list;
1068
1069 /* FIXME interlacing + spatial direct uses wrong colocated block positions */
1070
1071 /* ref = min(neighbors) */
1072 for(list=0; list<2; list++){
1073 int refa = h->ref_cache[list][scan8[0] - 1];
1074 int refb = h->ref_cache[list][scan8[0] - 8];
1075 int refc = h->ref_cache[list][scan8[0] - 8 + 4];
1076 if(refc == -2)
1077 refc = h->ref_cache[list][scan8[0] - 8 - 1];
1078 ref[list] = refa;
1079 if(ref[list] < 0 || (refb < ref[list] && refb >= 0))
1080 ref[list] = refb;
1081 if(ref[list] < 0 || (refc < ref[list] && refc >= 0))
1082 ref[list] = refc;
1083 if(ref[list] < 0)
1084 ref[list] = -1;
1085 }
1086
1087 if(ref[0] < 0 && ref[1] < 0){
1088 ref[0] = ref[1] = 0;
1089 mv[0][0] = mv[0][1] =
1090 mv[1][0] = mv[1][1] = 0;
1091 }else{
1092 for(list=0; list<2; list++){
1093 if(ref[list] >= 0)
1094 pred_motion(h, 0, 4, list, ref[list], &mv[list][0], &mv[list][1]);
1095 else
1096 mv[list][0] = mv[list][1] = 0;
1097 }
1098 }
1099
1100 if(ref[1] < 0){
1101 *mb_type &= ~MB_TYPE_P0L1;
1102 sub_mb_type &= ~MB_TYPE_P0L1;
1103 }else if(ref[0] < 0){
1104 *mb_type &= ~MB_TYPE_P0L0;
1105 sub_mb_type &= ~MB_TYPE_P0L0;
1106 }
1107
1108 if(IS_16X16(*mb_type)){
1109 int a=0, b=0;
1110
1111 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, (uint8_t)ref[0], 1);
1112 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, (uint8_t)ref[1], 1);
1113 if(!IS_INTRA(mb_type_col)
1114 && ( (l1ref0[0] == 0 && FFABS(l1mv0[0][0]) <= 1 && FFABS(l1mv0[0][1]) <= 1)
1115 || (l1ref0[0] < 0 && l1ref1[0] == 0 && FFABS(l1mv1[0][0]) <= 1 && FFABS(l1mv1[0][1]) <= 1
1116 && (h->x264_build>33 || !h->x264_build)))){
1117 if(ref[0] > 0)
1118 a= pack16to32(mv[0][0],mv[0][1]);
1119 if(ref[1] > 0)
1120 b= pack16to32(mv[1][0],mv[1][1]);
1121 }else{
1122 a= pack16to32(mv[0][0],mv[0][1]);
1123 b= pack16to32(mv[1][0],mv[1][1]);
1124 }
1125 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, a, 4);
1126 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, b, 4);
1127 }else{
1128 for(i8=0; i8<4; i8++){
1129 const int x8 = i8&1;
1130 const int y8 = i8>>1;
1131
1132 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1133 continue;
1134 h->sub_mb_type[i8] = sub_mb_type;
1135
1136 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1137 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1138 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);
1139 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);
1140
1141 /* col_zero_flag */
1142 if(!IS_INTRA(mb_type_col) && ( l1ref0[x8 + y8*h->b8_stride] == 0
1143 || (l1ref0[x8 + y8*h->b8_stride] < 0 && l1ref1[x8 + y8*h->b8_stride] == 0
1144 && (h->x264_build>33 || !h->x264_build)))){
1145 const int16_t (*l1mv)[2]= l1ref0[x8 + y8*h->b8_stride] == 0 ? l1mv0 : l1mv1;
1146 if(IS_SUB_8X8(sub_mb_type)){
1147 const int16_t *mv_col = l1mv[x8*3 + y8*3*h->b_stride];
1148 if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
1149 if(ref[0] == 0)
1150 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1151 if(ref[1] == 0)
1152 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1153 }
1154 }else
1155 for(i4=0; i4<4; i4++){
1156 const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1157 if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
1158 if(ref[0] == 0)
1159 *(uint32_t*)h->mv_cache[0][scan8[i8*4+i4]] = 0;
1160 if(ref[1] == 0)
1161 *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = 0;
1162 }
1163 }
1164 }
1165 }
1166 }
1167 }else{ /* direct temporal mv pred */
1168 const int *map_col_to_list0[2] = {h->map_col_to_list0[0], h->map_col_to_list0[1]};
1169 const int *dist_scale_factor = h->dist_scale_factor;
1170
1171 if(FRAME_MBAFF){
1172 if(IS_INTERLACED(*mb_type)){
1173 map_col_to_list0[0] = h->map_col_to_list0_field[0];
1174 map_col_to_list0[1] = h->map_col_to_list0_field[1];
1175 dist_scale_factor = h->dist_scale_factor_field;
1176 }
1177 if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col)){
1178 /* FIXME assumes direct_8x8_inference == 1 */
1179 const int pair_xy = s->mb_x + (s->mb_y&~1)*s->mb_stride;
1180 int mb_types_col[2];
1181 int y_shift;
1182
1183 *mb_type = MB_TYPE_8x8|MB_TYPE_L0L1
1184 | (is_b8x8 ? 0 : MB_TYPE_DIRECT2)
1185 | (*mb_type & MB_TYPE_INTERLACED);
1186 sub_mb_type = MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2|MB_TYPE_16x16;
1187
1188 if(IS_INTERLACED(*mb_type)){
1189 /* frame to field scaling */
1190 mb_types_col[0] = h->ref_list[1][0].mb_type[pair_xy];
1191 mb_types_col[1] = h->ref_list[1][0].mb_type[pair_xy+s->mb_stride];
1192 if(s->mb_y&1){
1193 l1ref0 -= 2*h->b8_stride;
1194 l1ref1 -= 2*h->b8_stride;
1195 l1mv0 -= 4*h->b_stride;
1196 l1mv1 -= 4*h->b_stride;
1197 }
1198 y_shift = 0;
1199
1200 if( (mb_types_col[0] & MB_TYPE_16x16_OR_INTRA)
1201 && (mb_types_col[1] & MB_TYPE_16x16_OR_INTRA)
1202 && !is_b8x8)
1203 *mb_type |= MB_TYPE_16x8;
1204 else
1205 *mb_type |= MB_TYPE_8x8;
1206 }else{
1207 /* field to frame scaling */
1208 /* col_mb_y = (mb_y&~1) + (topAbsDiffPOC < bottomAbsDiffPOC ? 0 : 1)
1209 * but in MBAFF, top and bottom POC are equal */
1210 int dy = (s->mb_y&1) ? 1 : 2;
1211 mb_types_col[0] =
1212 mb_types_col[1] = h->ref_list[1][0].mb_type[pair_xy+s->mb_stride];
1213 l1ref0 += dy*h->b8_stride;
1214 l1ref1 += dy*h->b8_stride;
1215 l1mv0 += 2*dy*h->b_stride;
1216 l1mv1 += 2*dy*h->b_stride;
1217 y_shift = 2;
1218
1219 if((mb_types_col[0] & (MB_TYPE_16x16_OR_INTRA|MB_TYPE_16x8))
1220 && !is_b8x8)
1221 *mb_type |= MB_TYPE_16x16;
1222 else
1223 *mb_type |= MB_TYPE_8x8;
1224 }
1225
1226 for(i8=0; i8<4; i8++){
1227 const int x8 = i8&1;
1228 const int y8 = i8>>1;
1229 int ref0, scale;
1230 const int16_t (*l1mv)[2]= l1mv0;
1231
1232 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1233 continue;
1234 h->sub_mb_type[i8] = sub_mb_type;
1235
1236 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1237 if(IS_INTRA(mb_types_col[y8])){
1238 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1239 fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1240 fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1241 continue;
1242 }
1243
1244 ref0 = l1ref0[x8 + (y8*2>>y_shift)*h->b8_stride];
1245 if(ref0 >= 0)
1246 ref0 = map_col_to_list0[0][ref0*2>>y_shift];
1247 else{
1248 ref0 = map_col_to_list0[1][l1ref1[x8 + (y8*2>>y_shift)*h->b8_stride]*2>>y_shift];
1249 l1mv= l1mv1;
1250 }
1251 scale = dist_scale_factor[ref0];
1252 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1253
1254 {
1255 const int16_t *mv_col = l1mv[x8*3 + (y8*6>>y_shift)*h->b_stride];
1256 int my_col = (mv_col[1]<<y_shift)/2;
1257 int mx = (scale * mv_col[0] + 128) >> 8;
1258 int my = (scale * my_col + 128) >> 8;
1259 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
1260 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-my_col), 4);
1261 }
1262 }
1263 return;
1264 }
1265 }
1266
1267 /* one-to-one mv scaling */
1268
1269 if(IS_16X16(*mb_type)){
1270 int ref, mv0, mv1;
1271
1272 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1);
1273 if(IS_INTRA(mb_type_col)){
1274 ref=mv0=mv1=0;
1275 }else{
1276 const int ref0 = l1ref0[0] >= 0 ? map_col_to_list0[0][l1ref0[0]]
1277 : map_col_to_list0[1][l1ref1[0]];
1278 const int scale = dist_scale_factor[ref0];
1279 const int16_t *mv_col = l1ref0[0] >= 0 ? l1mv0[0] : l1mv1[0];
1280 int mv_l0[2];
1281 mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
1282 mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
1283 ref= ref0;
1284 mv0= pack16to32(mv_l0[0],mv_l0[1]);
1285 mv1= pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
1286 }
1287 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref, 1);
1288 fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, mv0, 4);
1289 fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, mv1, 4);
1290 }else{
1291 for(i8=0; i8<4; i8++){
1292 const int x8 = i8&1;
1293 const int y8 = i8>>1;
1294 int ref0, scale;
1295 const int16_t (*l1mv)[2]= l1mv0;
1296
1297 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1298 continue;
1299 h->sub_mb_type[i8] = sub_mb_type;
1300 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1301 if(IS_INTRA(mb_type_col)){
1302 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1303 fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1304 fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1305 continue;
1306 }
1307
1308 ref0 = l1ref0[x8 + y8*h->b8_stride];
1309 if(ref0 >= 0)
1310 ref0 = map_col_to_list0[0][ref0];
1311 else{
1312 ref0 = map_col_to_list0[1][l1ref1[x8 + y8*h->b8_stride]];
1313 l1mv= l1mv1;
1314 }
1315 scale = dist_scale_factor[ref0];
1316
1317 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1318 if(IS_SUB_8X8(sub_mb_type)){
1319 const int16_t *mv_col = l1mv[x8*3 + y8*3*h->b_stride];
1320 int mx = (scale * mv_col[0] + 128) >> 8;
1321 int my = (scale * mv_col[1] + 128) >> 8;
1322 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
1323 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-mv_col[1]), 4);
1324 }else
1325 for(i4=0; i4<4; i4++){
1326 const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1327 int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]];
1328 mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
1329 mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
1330 *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] =
1331 pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
1332 }
1333 }
1334 }
1335 }
1336 }
1337
1338 static inline void write_back_motion(H264Context *h, int mb_type){
1339 MpegEncContext * const s = &h->s;
1340 const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1341 const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1342 int list;
1343
1344 if(!USES_LIST(mb_type, 0))
1345 fill_rectangle(&s->current_picture.ref_index[0][b8_xy], 2, 2, h->b8_stride, (uint8_t)LIST_NOT_USED, 1);
1346
1347 for(list=0; list<h->list_count; list++){
1348 int y;
1349 if(!USES_LIST(mb_type, list))
1350 continue;
1351
1352 for(y=0; y<4; y++){
1353 *(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];
1354 *(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];
1355 }
1356 if( h->pps.cabac ) {
1357 if(IS_SKIP(mb_type))
1358 fill_rectangle(h->mvd_table[list][b_xy], 4, 4, h->b_stride, 0, 4);
1359 else
1360 for(y=0; y<4; y++){
1361 *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+0 + 8*y];
1362 *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+2 + 8*y];
1363 }
1364 }
1365
1366 {
1367 int8_t *ref_index = &s->current_picture.ref_index[list][b8_xy];
1368 ref_index[0+0*h->b8_stride]= h->ref_cache[list][scan8[0]];
1369 ref_index[1+0*h->b8_stride]= h->ref_cache[list][scan8[4]];
1370 ref_index[0+1*h->b8_stride]= h->ref_cache[list][scan8[8]];
1371 ref_index[1+1*h->b8_stride]= h->ref_cache[list][scan8[12]];
1372 }
1373 }
1374
1375 if(h->slice_type == B_TYPE && h->pps.cabac){
1376 if(IS_8X8(mb_type)){
1377 uint8_t *direct_table = &h->direct_table[b8_xy];
1378 direct_table[1+0*h->b8_stride] = IS_DIRECT(h->sub_mb_type[1]) ? 1 : 0;
1379 direct_table[0+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[2]) ? 1 : 0;
1380 direct_table[1+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[3]) ? 1 : 0;
1381 }
1382 }
1383 }
1384
1385 /**
1386 * Decodes a network abstraction layer unit.
1387 * @param consumed is the number of bytes used as input
1388 * @param length is the length of the array
1389 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp tailing?
1390 * @returns decoded bytes, might be src+1 if no escapes
1391 */
1392 static uint8_t *decode_nal(H264Context *h, uint8_t *src, int *dst_length, int *consumed, int length){
1393 int i, si, di;
1394 uint8_t *dst;
1395
1396 // src[0]&0x80; //forbidden bit
1397 h->nal_ref_idc= src[0]>>5;
1398 h->nal_unit_type= src[0]&0x1F;
1399
1400 src++; length--;
1401 #if 0
1402 for(i=0; i<length; i++)
1403 printf("%2X ", src[i]);
1404 #endif
1405 for(i=0; i+1<length; i+=2){
1406 if(src[i]) continue;
1407 if(i>0 && src[i-1]==0) i--;
1408 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1409 if(src[i+2]!=3){
1410 /* startcode, so we must be past the end */
1411 length=i;
1412 }
1413 break;
1414 }
1415 }
1416
1417 if(i>=length-1){ //no escaped 0
1418 *dst_length= length;
1419 *consumed= length+1; //+1 for the header
1420 return src;
1421 }
1422
1423 h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length);
1424 dst= h->rbsp_buffer;
1425
1426 if (dst == NULL){
1427 return NULL;
1428 }
1429
1430 //printf("decoding esc\n");
1431 si=di=0;
1432 while(si<length){
1433 //remove escapes (very rare 1:2^22)
1434 if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1435 if(src[si+2]==3){ //escape
1436 dst[di++]= 0;
1437 dst[di++]= 0;
1438 si+=3;
1439 continue;
1440 }else //next start code
1441 break;
1442 }
1443
1444 dst[di++]= src[si++];
1445 }
1446
1447 *dst_length= di;
1448 *consumed= si + 1;//+1 for the header
1449 //FIXME store exact number of bits in the getbitcontext (its needed for decoding)
1450 return dst;
1451 }
1452
1453 /**
1454 * identifies the exact end of the bitstream
1455 * @return the length of the trailing, or 0 if damaged
1456 */
1457 static int decode_rbsp_trailing(H264Context *h, uint8_t *src){
1458 int v= *src;
1459 int r;
1460
1461 tprintf(h->s.avctx, "rbsp trailing %X\n", v);
1462
1463 for(r=1; r<9; r++){
1464 if(v&1) return r;
1465 v>>=1;
1466 }
1467 return 0;
1468 }
1469
1470 /**
1471 * idct tranforms the 16 dc values and dequantize them.
1472 * @param qp quantization parameter
1473 */
1474 static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
1475 #define stride 16
1476 int i;
1477 int temp[16]; //FIXME check if this is a good idea
1478 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
1479 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1480
1481 //memset(block, 64, 2*256);
1482 //return;
1483 for(i=0; i<4; i++){
1484 const int offset= y_offset[i];
1485 const int z0= block[offset+stride*0] + block[offset+stride*4];
1486 const int z1= block[offset+stride*0] - block[offset+stride*4];
1487 const int z2= block[offset+stride*1] - block[offset+stride*5];
1488 const int z3= block[offset+stride*1] + block[offset+stride*5];
1489
1490 temp[4*i+0]= z0+z3;
1491 temp[4*i+1]= z1+z2;
1492 temp[4*i+2]= z1-z2;
1493 temp[4*i+3]= z0-z3;
1494 }
1495
1496 for(i=0; i<4; i++){
1497 const int offset= x_offset[i];
1498 const int z0= temp[4*0+i] + temp[4*2+i];
1499 const int z1= temp[4*0+i] - temp[4*2+i];
1500 const int z2= temp[4*1+i] - temp[4*3+i];
1501 const int z3= temp[4*1+i] + temp[4*3+i];
1502
1503 block[stride*0 +offset]= ((((z0 + z3)*qmul + 128 ) >> 8)); //FIXME think about merging this into decode_resdual
1504 block[stride*2 +offset]= ((((z1 + z2)*qmul + 128 ) >> 8));
1505 block[stride*8 +offset]= ((((z1 - z2)*qmul + 128 ) >> 8));
1506 block[stride*10+offset]= ((((z0 - z3)*qmul + 128 ) >> 8));
1507 }
1508 }
1509
1510 #if 0
1511 /**
1512 * dct tranforms the 16 dc values.
1513 * @param qp quantization parameter ??? FIXME
1514 */
1515 static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
1516 // const int qmul= dequant_coeff[qp][0];
1517 int i;
1518 int temp[16]; //FIXME check if this is a good idea
1519 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
1520 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1521
1522 for(i=0; i<4; i++){
1523 const int offset= y_offset[i];
1524 const int z0= block[offset+stride*0] + block[offset+stride*4];
1525 const int z1= block[offset+stride*0] - block[offset+stride*4];
1526 const int z2= block[offset+stride*1] - block[offset+stride*5];
1527 const int z3= block[offset+stride*1] + block[offset+stride*5];
1528
1529 temp[4*i+0]= z0+z3;
1530 temp[4*i+1]= z1+z2;
1531 temp[4*i+2]= z1-z2;
1532 temp[4*i+3]= z0-z3;
1533 }
1534
1535 for(i=0; i<4; i++){
1536 const int offset= x_offset[i];
1537 const int z0= temp[4*0+i] + temp[4*2+i];
1538 const int z1= temp[4*0+i] - temp[4*2+i];
1539 const int z2= temp[4*1+i] - temp[4*3+i];
1540 const int z3= temp[4*1+i] + temp[4*3+i];
1541
1542 block[stride*0 +offset]= (z0 + z3)>>1;
1543 block[stride*2 +offset]= (z1 + z2)>>1;
1544 block[stride*8 +offset]= (z1 - z2)>>1;
1545 block[stride*10+offset]= (z0 - z3)>>1;
1546 }
1547 }
1548 #endif
1549
1550 #undef xStride
1551 #undef stride
1552
1553 static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
1554 const int stride= 16*2;
1555 const int xStride= 16;
1556 int a,b,c,d,e;
1557
1558 a= block[stride*0 + xStride*0];
1559 b= block[stride*0 + xStride*1];
1560 c= block[stride*1 + xStride*0];
1561 d= block[stride*1 + xStride*1];
1562
1563 e= a-b;
1564 a= a+b;
1565 b= c-d;
1566 c= c+d;
1567
1568 block[stride*0 + xStride*0]= ((a+c)*qmul) >> 7;
1569 block[stride*0 + xStride*1]= ((e+b)*qmul) >> 7;
1570 block[stride*1 + xStride*0]= ((a-c)*qmul) >> 7;
1571 block[stride*1 + xStride*1]= ((e-b)*qmul) >> 7;
1572 }
1573
1574 #if 0
1575 static void chroma_dc_dct_c(DCTELEM *block){
1576 const int stride= 16*2;
1577 const int xStride= 16;
1578 int a,b,c,d,e;
1579
1580 a= block[stride*0 + xStride*0];
1581 b= block[stride*0 + xStride*1];
1582 c= block[stride*1 + xStride*0];
1583 d= block[stride*1 + xStride*1];
1584
1585 e= a-b;
1586 a= a+b;
1587 b= c-d;
1588 c= c+d;
1589
1590 block[stride*0 + xStride*0]= (a+c);
1591 block[stride*0 + xStride*1]= (e+b);
1592 block[stride*1 + xStride*0]= (a-c);
1593 block[stride*1 + xStride*1]= (e-b);
1594 }
1595 #endif
1596
1597 /**
1598 * gets the chroma qp.
1599 */
1600 static inline int get_chroma_qp(int chroma_qp_index_offset, int qscale){
1601
1602 return chroma_qp[av_clip(qscale + chroma_qp_index_offset, 0, 51)];
1603 }
1604
1605 //FIXME need to check that this doesnt overflow signed 32 bit for low qp, i am not sure, it's very close
1606 //FIXME check that gcc inlines this (and optimizes intra & separate_dc stuff away)
1607 static inline int quantize_c(DCTELEM *block, uint8_t *scantable, int qscale, int intra, int separate_dc){
1608 int i;
1609 const int * const quant_table= quant_coeff[qscale];
1610 const int bias= intra ? (1<<QUANT_SHIFT)/3 : (1<<QUANT_SHIFT)/6;
1611 const unsigned int threshold1= (1<<QUANT_SHIFT) - bias - 1;
1612 const unsigned int threshold2= (threshold1<<1);
1613 int last_non_zero;
1614
1615 if(separate_dc){
1616 if(qscale<=18){
1617 //avoid overflows
1618 const int dc_bias= intra ? (1<<(QUANT_SHIFT-2))/3 : (1<<(QUANT_SHIFT-2))/6;
1619 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT-2)) - dc_bias - 1;
1620 const unsigned int dc_threshold2= (dc_threshold1<<1);
1621
1622 int level= block[0]*quant_coeff[qscale+18][0];
1623 if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1624 if(level>0){
1625 level= (dc_bias + level)>>(QUANT_SHIFT-2);
1626 block[0]= level;
1627 }else{
1628 level= (dc_bias - level)>>(QUANT_SHIFT-2);
1629 block[0]= -level;
1630 }
1631 // last_non_zero = i;
1632 }else{
1633 block[0]=0;
1634 }
1635 }else{
1636 const int dc_bias= intra ? (1<<(QUANT_SHIFT+1))/3 : (1<<(QUANT_SHIFT+1))/6;
1637 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT+1)) - dc_bias - 1;
1638 const unsigned int dc_threshold2= (dc_threshold1<<1);
1639
1640 int level= block[0]*quant_table[0];
1641 if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1642 if(level>0){
1643 level= (dc_bias + level)>>(QUANT_SHIFT+1);
1644 block[0]= level;
1645 }else{
1646 level= (dc_bias - level)>>(QUANT_SHIFT+1);
1647 block[0]= -level;
1648 }
1649 // last_non_zero = i;
1650 }else{
1651 block[0]=0;
1652 }
1653 }
1654 last_non_zero= 0;
1655 i=1;
1656 }else{
1657 last_non_zero= -1;
1658 i=0;
1659 }
1660
1661 for(; i<16; i++){
1662 const int j= scantable[i];
1663 int level= block[j]*quant_table[j];
1664
1665 // if( bias+level >= (1<<(QMAT_SHIFT - 3))
1666 // || bias-level >= (1<<(QMAT_SHIFT - 3))){
1667 if(((unsigned)(level+threshold1))>threshold2){
1668 if(level>0){
1669 level= (bias + level)>>QUANT_SHIFT;
1670 block[j]= level;
1671 }else{
1672 level= (bias - level)>>QUANT_SHIFT;
1673 block[j]= -level;
1674 }
1675 last_non_zero = i;
1676 }else{
1677 block[j]=0;
1678 }
1679 }
1680
1681 return last_non_zero;
1682 }
1683
1684 static void pred4x4_vertical_c(uint8_t *src, uint8_t *topright, int stride){
1685 const uint32_t a= ((uint32_t*)(src-stride))[0];
1686 ((uint32_t*)(src+0*stride))[0]= a;
1687 ((uint32_t*)(src+1*stride))[0]= a;
1688 ((uint32_t*)(src+2*stride))[0]= a;
1689 ((uint32_t*)(src+3*stride))[0]= a;
1690 }
1691
1692 static void pred4x4_horizontal_c(uint8_t *src, uint8_t *topright, int stride){
1693 ((uint32_t*)(src+0*stride))[0]= src[-1+0*stride]*0x01010101;
1694 ((uint32_t*)(src+1*stride))[0]= src[-1+1*stride]*0x01010101;
1695 ((uint32_t*)(src+2*stride))[0]= src[-1+2*stride]*0x01010101;
1696 ((uint32_t*)(src+3*stride))[0]= src[-1+3*stride]*0x01010101;
1697 }
1698
1699 static void pred4x4_dc_c(uint8_t *src, uint8_t *topright, int stride){
1700 const int dc= ( src[-stride] + src[1-stride] + src[2-stride] + src[3-stride]
1701 + src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 4) >>3;
1702
1703 ((uint32_t*)(src+0*stride))[0]=
1704 ((uint32_t*)(src+1*stride))[0]=
1705 ((uint32_t*)(src+2*stride))[0]=
1706 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1707 }
1708
1709 static void pred4x4_left_dc_c(uint8_t *src, uint8_t *topright, int stride){
1710 const int dc= ( src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 2) >>2;
1711
1712 ((uint32_t*)(src+0*stride))[0]=
1713 ((uint32_t*)(src+1*stride))[0]=
1714 ((uint32_t*)(src+2*stride))[0]=
1715 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1716 }
1717
1718 static void pred4x4_top_dc_c(uint8_t *src, uint8_t *topright, int stride){
1719 const int dc= ( src[-stride] + src[1-stride] + src[2-stride] + src[3-stride] + 2) >>2;
1720
1721 ((uint32_t*)(src+0*stride))[0]=
1722 ((uint32_t*)(src+1*stride))[0]=
1723 ((uint32_t*)(src+2*stride))[0]=
1724 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1725 }
1726
1727 static void pred4x4_128_dc_c(uint8_t *src, uint8_t *topright, int stride){
1728 ((uint32_t*)(src+0*stride))[0]=
1729 ((uint32_t*)(src+1*stride))[0]=
1730 ((uint32_t*)(src+2*stride))[0]=
1731 ((uint32_t*)(src+3*stride))[0]= 128U*0x01010101U;
1732 }
1733
1734
1735 #define LOAD_TOP_RIGHT_EDGE\
1736 const int av_unused t4= topright[0];\
1737 const int av_unused t5= topright[1];\
1738 const int av_unused t6= topright[2];\
1739 const int av_unused t7= topright[3];\
1740
1741 #define LOAD_LEFT_EDGE\
1742 const int av_unused l0= src[-1+0*stride];\
1743 const int av_unused l1= src[-1+1*stride];\
1744 const int av_unused l2= src[-1+2*stride];\
1745 const int av_unused l3= src[-1+3*stride];\
1746
1747 #define LOAD_TOP_EDGE\
1748 const int av_unused t0= src[ 0-1*stride];\
1749 const int av_unused t1= src[ 1-1*stride];\
1750 const int av_unused t2= src[ 2-1*stride];\
1751 const int av_unused t3= src[ 3-1*stride];\
1752
1753 static void pred4x4_down_right_c(uint8_t *src, uint8_t *topright, int stride){
1754 const int lt= src[-1-1*stride];
1755 LOAD_TOP_EDGE
1756 LOAD_LEFT_EDGE
1757
1758 src[0+3*stride]=(l3 + 2*l2 + l1 + 2)>>2;
1759 src[0+2*stride]=
1760 src[1+3*stride]=(l2 + 2*l1 + l0 + 2)>>2;
1761 src[0+1*stride]=
1762 src[1+2*stride]=
1763 src[2+3*stride]=(l1 + 2*l0 + lt + 2)>>2;
1764 src[0+0*stride]=
1765 src[1+1*stride]=
1766 src[2+2*stride]=
1767 src[3+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1768 src[1+0*stride]=
1769 src[2+1*stride]=
1770 src[3+2*stride]=(lt + 2*t0 + t1 + 2)>>2;
1771 src[2+0*stride]=
1772 src[3+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1773 src[3+0*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1774 }
1775
1776 static void pred4x4_down_left_c(uint8_t *src, uint8_t *topright, int stride){
1777 LOAD_TOP_EDGE
1778 LOAD_TOP_RIGHT_EDGE
1779 // LOAD_LEFT_EDGE
1780
1781 src[0+0*stride]=(t0 + t2 + 2*t1 + 2)>>2;
1782 src[1+0*stride]=
1783 src[0+1*stride]=(t1 + t3 + 2*t2 + 2)>>2;
1784 src[2+0*stride]=
1785 src[1+1*stride]=
1786 src[0+2*stride]=(t2 + t4 + 2*t3 + 2)>>2;
1787 src[3+0*stride]=
1788 src[2+1*stride]=
1789 src[1+2*stride]=
1790 src[0+3*stride]=(t3 + t5 + 2*t4 + 2)>>2;
1791 src[3+1*stride]=
1792 src[2+2*stride]=
1793 src[1+3*stride]=(t4 + t6 + 2*t5 + 2)>>2;
1794 src[3+2*stride]=
1795 src[2+3*stride]=(t5 + t7 + 2*t6 + 2)>>2;
1796 src[3+3*stride]=(t6 + 3*t7 + 2)>>2;
1797 }
1798
1799 static void pred4x4_vertical_right_c(uint8_t *src, uint8_t *topright, int stride){
1800 const int lt= src[-1-1*stride];
1801 LOAD_TOP_EDGE
1802 LOAD_LEFT_EDGE
1803
1804 src[0+0*stride]=
1805 src[1+2*stride]=(lt + t0 + 1)>>1;
1806 src[1+0*stride]=
1807 src[2+2*stride]=(t0 + t1 + 1)>>1;
1808 src[2+0*stride]=
1809 src[3+2*stride]=(t1 + t2 + 1)>>1;
1810 src[3+0*stride]=(t2 + t3 + 1)>>1;
1811 src[0+1*stride]=
1812 src[1+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1813 src[1+1*stride]=
1814 src[2+3*stride]=(lt + 2*t0 + t1 + 2)>>2;
1815 src[2+1*stride]=
1816 src[3+3*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1817 src[3+1*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1818 src[0+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1819 src[0+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1820 }
1821
1822 static void pred4x4_vertical_left_c(uint8_t *src, uint8_t *topright, int stride){
1823 LOAD_TOP_EDGE
1824 LOAD_TOP_RIGHT_EDGE
1825
1826 src[0+0*stride]=(t0 + t1 + 1)>>1;
1827 src[1+0*stride]=
1828 src[0+2*stride]=(t1 + t2 + 1)>>1;
1829 src[2+0*stride]=
1830 src[1+2*stride]=(t2 + t3 + 1)>>1;
1831 src[3+0*stride]=
1832 src[2+2*stride]=(t3 + t4+ 1)>>1;
1833 src[3+2*stride]=(t4 + t5+ 1)>>1;
1834 src[0+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1835 src[1+1*stride]=
1836 src[0+3*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1837 src[2+1*stride]=
1838 src[1+3*stride]=(t2 + 2*t3 + t4 + 2)>>2;
1839 src[3+1*stride]=
1840 src[2+3*stride]=(t3 + 2*t4 + t5 + 2)>>2;
1841 src[3+3*stride]=(t4 + 2*t5 + t6 + 2)>>2;
1842 }
1843
1844 static void pred4x4_horizontal_up_c(uint8_t *src, uint8_t *topright, int stride){
1845 LOAD_LEFT_EDGE
1846
1847 src[0+0*stride]=(l0 + l1 + 1)>>1;
1848 src[1+0*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1849 src[2+0*stride]=
1850 src[0+1*stride]=(l1 + l2 + 1)>>1;
1851 src[3+0*stride]=
1852 src[1+1*stride]=(l1 + 2*l2 + l3 + 2)>>2;
1853 src[2+1*stride]=
1854 src[0+2*stride]=(l2 + l3 + 1)>>1;
1855 src[3+1*stride]=
1856 src[1+2*stride]=(l2 + 2*l3 + l3 + 2)>>2;
1857 src[3+2*stride]=
1858 src[1+3*stride]=
1859 src[0+3*stride]=
1860 src[2+2*stride]=
1861 src[2+3*stride]=
1862 src[3+3*stride]=l3;
1863 }
1864
1865 static void pred4x4_horizontal_down_c(uint8_t *src, uint8_t *topright, int stride){
1866 const int lt= src[-1-1*stride];
1867 LOAD_TOP_EDGE
1868 LOAD_LEFT_EDGE
1869
1870 src[0+0*stride]=
1871 src[2+1*stride]=(lt + l0 + 1)>>1;
1872 src[1+0*stride]=
1873 src[3+1*stride]=(l0 + 2*lt + t0 + 2)>>2;
1874 src[2+0*stride]=(lt + 2*t0 + t1 + 2)>>2;
1875 src[3+0*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1876 src[0+1*stride]=
1877 src[2+2*stride]=(l0 + l1 + 1)>>1;
1878 src[1+1*stride]=
1879 src[3+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1880 src[0+2*stride]=
1881 src[2+3*stride]=(l1 + l2+ 1)>>1;
1882 src[1+2*stride]=
1883 src[3+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1884 src[0+3*stride]=(l2 + l3 + 1)>>1;
1885 src[1+3*stride]=(l1 + 2*l2 + l3 + 2)>>2;
1886 }
1887
1888 void ff_pred16x16_vertical_c(uint8_t *src, int stride){
1889 int i;
1890 const uint32_t a= ((uint32_t*)(src-stride))[0];
1891 const uint32_t b= ((uint32_t*)(src-stride))[1];
1892 const uint32_t c= ((uint32_t*)(src-stride))[2];
1893 const uint32_t d= ((uint32_t*)(src-stride))[3];
1894
1895 for(i=0; i<16; i++){
1896 ((uint32_t*)(src+i*stride))[0]= a;
1897 ((uint32_t*)(src+i*stride))[1]= b;
1898 ((uint32_t*)(src+i*stride))[2]= c;
1899 ((uint32_t*)(src+i*stride))[3]= d;
1900 }
1901 }
1902
1903 void ff_pred16x16_horizontal_c(uint8_t *src, int stride){
1904 int i;
1905
1906 for(i=0; i<16; i++){
1907 ((uint32_t*)(src+i*stride))[0]=
1908 ((uint32_t*)(src+i*stride))[1]=
1909 ((uint32_t*)(src+i*stride))[2]=
1910 ((uint32_t*)(src+i*stride))[3]= src[-1+i*stride]*0x01010101;
1911 }
1912 }
1913
1914 void ff_pred16x16_dc_c(uint8_t *src, int stride){
1915 int i, dc=0;
1916
1917 for(i=0;i<16; i++){
1918 dc+= src[-1+i*stride];
1919 }
1920
1921 for(i=0;i<16; i++){
1922 dc+= src[i-stride];
1923 }
1924
1925 dc= 0x01010101*((dc + 16)>>5);
1926
1927 for(i=0; i<16; i++){
1928 ((uint32_t*)(src+i*stride))[0]=
1929 ((uint32_t*)(src+i*stride))[1]=
1930 ((uint32_t*)(src+i*stride))[2]=
1931 ((uint32_t*)(src+i*stride))[3]= dc;
1932 }
1933 }
1934
1935 void ff_pred16x16_left_dc_c(uint8_t *src, int stride){
1936 int i, dc=0;
1937
1938 for(i=0;i<16; i++){
1939 dc+= src[-1+i*stride];
1940 }
1941
1942 dc= 0x01010101*((dc + 8)>>4);
1943
1944 for(i=0; i<16; i++){
1945 ((uint32_t*)(src+i*stride))[0]=
1946 ((uint32_t*)(src+i*stride))[1]=
1947 ((uint32_t*)(src+i*stride))[2]=
1948 ((uint32_t*)(src+i*stride))[3]= dc;
1949 }
1950 }
1951
1952 void ff_pred16x16_top_dc_c(uint8_t *src, int stride){
1953 int i, dc=0;
1954
1955 for(i=0;i<16; i++){
1956 dc+= src[i-stride];
1957 }
1958 dc= 0x01010101*((dc + 8)>>4);
1959
1960 for(i=0; i<16; i++){
1961 ((uint32_t*)(src+i*stride))[0]=
1962 ((uint32_t*)(src+i*stride))[1]=
1963 ((uint32_t*)(src+i*stride))[2]=
1964 ((uint32_t*)(src+i*stride))[3]= dc;
1965 }
1966 }
1967
1968 void ff_pred16x16_128_dc_c(uint8_t *src, int stride){
1969 int i;
1970
1971 for(i=0; i<16; i++){
1972 ((uint32_t*)(src+i*stride))[0]=
1973 ((uint32_t*)(src+i*stride))[1]=
1974 ((uint32_t*)(src+i*stride))[2]=
1975 ((uint32_t*)(src+i*stride))[3]= 0x01010101U*128U;
1976 }
1977 }
1978
1979 static inline void pred16x16_plane_compat_c(uint8_t *src, int stride, const int svq3){
1980 int i, j, k;
1981 int a;
1982 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
1983 const uint8_t * const src0 = src+7-stride;
1984 const uint8_t *src1 = src+8*stride-1;
1985 const uint8_t *src2 = src1-2*stride; // == src+6*stride-1;
1986 int H = src0[1] - src0[-1];
1987 int V = src1[0] - src2[ 0];
1988 for(k=2; k<=8; ++k) {
1989 src1 += stride; src2 -= stride;
1990 H += k*(src0[k] - src0[-k]);
1991 V += k*(src1[0] - src2[ 0]);
1992 }
1993 if(svq3){
1994 H = ( 5*(H/4) ) / 16;
1995 V = ( 5*(V/4) ) / 16;
1996
1997 /* required for 100% accuracy */
1998 i = H; H = V; V = i;
1999 }else{
2000 H = ( 5*H+32 ) >> 6;
2001 V = ( 5*V+32 ) >> 6;
2002 }
2003
2004 a = 16*(src1[0] + src2[16] + 1) - 7*(V+H);
2005 for(j=16; j>0; --j) {
2006 int b = a;
2007 a += V;
2008 for(i=-16; i<0; i+=4) {
2009 src[16+i] = cm[ (b ) >> 5 ];
2010 src[17+i] = cm[ (b+ H) >> 5 ];
2011 src[18+i] = cm[ (b+2*H) >> 5 ];
2012 src[19+i] = cm[ (b+3*H) >> 5 ];
2013 b += 4*H;
2014 }
2015 src += stride;
2016 }
2017 }
2018
2019 void ff_pred16x16_plane_c(uint8_t *src, int stride){
2020 pred16x16_plane_compat_c(src, stride, 0);
2021 }
2022
2023 void ff_pred8x8_vertical_c(uint8_t *src, int stride){
2024 int i;
2025 const uint32_t a= ((uint32_t*)(src-stride))[0];
2026 const uint32_t b= ((uint32_t*)(src-stride))[1];
2027
2028 for(i=0; i<8; i++){
2029 ((uint32_t*)(src+i*stride))[0]= a;
2030 ((uint32_t*)(src+i*stride))[1]= b;
2031 }
2032 }
2033
2034 void ff_pred8x8_horizontal_c(uint8_t *src, int stride){
2035 int i;
2036
2037 for(i=0; i<8; i++){
2038 ((uint32_t*)(src+i*stride))[0]=
2039 ((uint32_t*)(src+i*stride))[1]= src[-1+i*stride]*0x01010101;
2040 }
2041 }
2042
2043 void ff_pred8x8_128_dc_c(uint8_t *src, int stride){
2044 int i;
2045
2046 for(i=0; i<8; i++){
2047 ((uint32_t*)(src+i*stride))[0]=
2048 ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
2049 }
2050 }
2051
2052 void ff_pred8x8_left_dc_c(uint8_t *src, int stride){
2053 int i;
2054 int dc0, dc2;
2055
2056 dc0=dc2=0;
2057 for(i=0;i<4; i++){
2058 dc0+= src[-1+i*stride];
2059 dc2+= src[-1+(i+4)*stride];
2060 }
2061 dc0= 0x01010101*((dc0 + 2)>>2);
2062 dc2= 0x01010101*((dc2 + 2)>>2);
2063
2064 for(i=0; i<4; i++){
2065 ((uint32_t*)(src+i*stride))[0]=
2066 ((uint32_t*)(src+i*stride))[1]= dc0;
2067 }
2068 for(i=4; i<8; i++){
2069 ((uint32_t*)(src+i*stride))[0]=
2070 ((uint32_t*)(src+i*stride))[1]= dc2;
2071 }
2072 }
2073
2074 void ff_pred8x8_top_dc_c(uint8_t *src, int stride){
2075 int i;
2076 int dc0, dc1;
2077
2078 dc0=dc1=0;
2079 for(i=0;i<4; i++){
2080 dc0+= src[i-stride];
2081 dc1+= src[4+i-stride];
2082 }
2083 dc0= 0x01010101*((dc0 + 2)>>2);
2084 dc1= 0x01010101*((dc1 + 2)>>2);
2085
2086 for(i=0; i<4; i++){
2087 ((uint32_t*)(src+i*stride))[0]= dc0;
2088 ((uint32_t*)(src+i*stride))[1]= dc1;
2089 }
2090 for(i=4; i<8; i++){
2091 ((uint32_t*)(src+i*stride))[0]= dc0;
2092 ((uint32_t*)(src+i*stride))[1]= dc1;
2093 }
2094 }
2095
2096
2097 void ff_pred8x8_dc_c(uint8_t *src, int stride){
2098 int i;
2099 int dc0, dc1, dc2, dc3;
2100
2101 dc0=dc1=dc2=0;
2102 for(i=0;i<4; i++){
2103 dc0+= src[-1+i*stride] + src[i-stride];
2104 dc1+= src[4+i-stride];
2105 dc2+= src[-1+(i+4)*stride];
2106 }
2107 dc3= 0x01010101*((dc1 + dc2 + 4)>>3);
2108 dc0= 0x01010101*((dc0 + 4)>>3);
2109 dc1= 0x01010101*((dc1 + 2)>>2);
2110 dc2= 0x01010101*((dc2 + 2)>>2);
2111
2112 for(i=0; i<4; i++){
2113 ((uint32_t*)(src+i*stride))[0]= dc0;
2114 ((uint32_t*)(src+i*stride))[1]= dc1;
2115 }
2116 for(i=4; i<8; i++){
2117 ((uint32_t*)(src+i*stride))[0]= dc2;
2118 ((uint32_t*)(src+i*stride))[1]= dc3;
2119 }
2120 }
2121
2122 void ff_pred8x8_plane_c(uint8_t *src, int stride){
2123 int j, k;
2124 int a;
2125 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
2126 const uint8_t * const src0 = src+3-stride;
2127 const uint8_t *src1 = src+4*stride-1;
2128 const uint8_t *src2 = src1-2*stride; // == src+2*stride-1;
2129 int H = src0[1] - src0[-1];
2130 int V = src1[0] - src2[ 0];
2131 for(k=2; k<=4; ++k) {
2132 src1 += stride; src2 -= stride;
2133 H += k*(src0[k] - src0[-k]);
2134 V += k*(src1[0] - src2[ 0]);
2135 }
2136 H = ( 17*H+16 ) >> 5;
2137 V = ( 17*V+16 ) >> 5;
2138
2139 a = 16*(src1[0] + src2[8]+1) - 3*(V+H);
2140 for(j=8; j>0; --j) {
2141 int b = a;
2142 a += V;
2143 src[0] = cm[ (b ) >> 5 ];
2144 src[1] = cm[ (b+ H) >> 5 ];
2145 src[2] = cm[ (b+2*H) >> 5 ];
2146 src[3] = cm[ (b+3*H) >> 5 ];
2147 src[4] = cm[ (b+4*H) >> 5 ];
2148 src[5] = cm[ (b+5*H) >> 5 ];
2149 src[6] = cm[ (b+6*H) >> 5 ];
2150 src[7] = cm[ (b+7*H) >> 5 ];
2151 src += stride;
2152 }
2153 }
2154
2155 #define SRC(x,y) src[(x)+(y)*stride]
2156 #define PL(y) \
2157 const int l##y = (SRC(-1,y-1) + 2*SRC(-1,y) + SRC(-1,y+1) + 2) >> 2;
2158 #define PREDICT_8x8_LOAD_LEFT \
2159 const int l0 = ((has_topleft ? SRC(-1,-1) : SRC(-1,0)) \
2160 + 2*SRC(-1,0) + SRC(-1,1) + 2) >> 2; \
2161 PL(1) PL(2) PL(3) PL(4) PL(5) PL(6) \
2162 const int l7 av_unused = (SRC(-1,6) + 3*SRC(-1,7) + 2) >> 2
2163
2164 #define PT(x) \
2165 const int t##x = (SRC(x-1,-1) + 2*SRC(x,-1) + SRC(x+1,-1) + 2) >> 2;
2166 #define PREDICT_8x8_LOAD_TOP \
2167 const int t0 = ((has_topleft ? SRC(-1,-1) : SRC(0,-1)) \
2168 + 2*SRC(0,-1) + SRC(1,-1) + 2) >> 2; \
2169 PT(1) PT(2) PT(3) PT(4) PT(5) PT(6) \
2170 const int t7 av_unused = ((has_topright ? SRC(8,-1) : SRC(7,-1)) \
2171 + 2*SRC(7,-1) + SRC(6,-1) + 2) >> 2
2172
2173 #define PTR(x) \
2174 t##x = (SRC(x-1,-1) + 2*SRC(x,-1) + SRC(x+1,-1) + 2) >> 2;
2175 #define PREDICT_8x8_LOAD_TOPRIGHT \
2176 int t8, t9, t10, t11, t12, t13, t14, t15; \
2177 if(has_topright) { \
2178 PTR(8) PTR(9) PTR(10) PTR(11) PTR(12) PTR(13) PTR(14) \
2179 t15 = (SRC(14,-1) + 3*SRC(15,-1) + 2) >> 2; \
2180 } else t8=t9=t10=t11=t12=t13=t14=t15= SRC(7,-1);
2181
2182 #define PREDICT_8x8_LOAD_TOPLEFT \
2183 const int lt = (SRC(-1,0) + 2*SRC(-1,-1) + SRC(0,-1) + 2) >> 2
2184
2185 #define PREDICT_8x8_DC(v) \
2186 int y; \
2187 for( y = 0; y < 8; y++ ) { \
2188 ((uint32_t*)src)[0] = \
2189 ((uint32_t*)src)[1] = v; \
2190 src += stride; \
2191 }
2192
2193 static void pred8x8l_128_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2194 {
2195 PREDICT_8x8_DC(0x80808080);
2196 }
2197 static void pred8x8l_left_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2198 {
2199 PREDICT_8x8_LOAD_LEFT;
2200 const uint32_t dc = ((l0+l1+l2+l3+l4+l5+l6+l7+4) >> 3) * 0x01010101;
2201 PREDICT_8x8_DC(dc);
2202 }
2203 static void pred8x8l_top_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2204 {
2205 PREDICT_8x8_LOAD_TOP;
2206 const uint32_t dc = ((t0+t1+t2+t3+t4+t5+t6+t7+4) >> 3) * 0x01010101;
2207 PREDICT_8x8_DC(dc);
2208 }
2209 static void pred8x8l_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2210 {
2211 PREDICT_8x8_LOAD_LEFT;
2212 PREDICT_8x8_LOAD_TOP;
2213 const uint32_t dc = ((l0+l1+l2+l3+l4+l5+l6+l7
2214 +t0+t1+t2+t3+t4+t5+t6+t7+8) >> 4) * 0x01010101;
2215 PREDICT_8x8_DC(dc);
2216 }
2217 static void pred8x8l_horizontal_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2218 {
2219 PREDICT_8x8_LOAD_LEFT;
2220 #define ROW(y) ((uint32_t*)(src+y*stride))[0] =\
2221 ((uint32_t*)(src+y*stride))[1] = 0x01010101 * l##y
2222 ROW(0); ROW(1); ROW(2); ROW(3); ROW(4); ROW(5); ROW(6); ROW(7);
2223 #undef ROW
2224 }
2225 static void pred8x8l_vertical_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2226 {
2227 int y;
2228 PREDICT_8x8_LOAD_TOP;
2229 src[0] = t0;
2230 src[1] = t1;
2231 src[2] = t2;
2232 src[3] = t3;
2233 src[4] = t4;
2234 src[5] = t5;
2235 src[6] = t6;
2236 src[7] = t7;
2237 for( y = 1; y < 8; y++ )
2238 *(uint64_t*)(src+y*stride) = *(uint64_t*)src;
2239 }
2240 static void pred8x8l_down_left_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2241 {
2242 PREDICT_8x8_LOAD_TOP;
2243 PREDICT_8x8_LOAD_TOPRIGHT;
2244 SRC(0,0)= (t0 + 2*t1 + t2 + 2) >> 2;
2245 SRC(0,1)=SRC(1,0)= (t1 + 2*t2 + t3 + 2) >> 2;
2246 SRC(0,2)=SRC(1,1)=SRC(2,0)= (t2 + 2*t3 + t4 + 2) >> 2;
2247 SRC(0,3)=SRC(1,2)=SRC(2,1)=SRC(3,0)= (t3 + 2*t4 + t5 + 2) >> 2;
2248 SRC(0,4)=SRC(1,3)=SRC(2,2)=SRC(3,1)=SRC(4,0)= (t4 + 2*t5 + t6 + 2) >> 2;
2249 SRC(0,5)=SRC(1,4)=SRC(2,3)=SRC(3,2)=SRC(4,1)=SRC(5,0)= (t5 + 2*t6 + t7 + 2) >> 2;
2250 SRC(0,6)=SRC(1,5)=SRC(2,4)=SRC(3,3)=SRC(4,2)=SRC(5,1)=SRC(6,0)= (t6 + 2*t7 + t8 + 2) >> 2;
2251 SRC(0,7)=SRC(1,6)=SRC(2,5)=SRC(3,4)=SRC(4,3)=SRC(5,2)=SRC(6,1)=SRC(7,0)= (t7 + 2*t8 + t9 + 2) >> 2;
2252 SRC(1,7)=SRC(2,6)=SRC(3,5)=SRC(4,4)=SRC(5,3)=SRC(6,2)=SRC(7,1)= (t8 + 2*t9 + t10 + 2) >> 2;
2253 SRC(2,7)=SRC(3,6)=SRC(4,5)=SRC(5,4)=SRC(6,3)=SRC(7,2)= (t9 + 2*t10 + t11 + 2) >> 2;
2254 SRC(3,7)=SRC(4,6)=SRC(5,5)=SRC(6,4)=SRC(7,3)= (t10 + 2*t11 + t12 + 2) >> 2;
2255 SRC(4,7)=SRC(5,6)=SRC(6,5)=SRC(7,4)= (t11 + 2*t12 + t13 + 2) >> 2;
2256 SRC(5,7)=SRC(6,6)=SRC(7,5)= (t12 + 2*t13 + t14 + 2) >> 2;
2257 SRC(6,7)=SRC(7,6)= (t13 + 2*t14 + t15 + 2) >> 2;
2258 SRC(7,7)= (t14 + 3*t15 + 2) >> 2;
2259 }
2260 static void pred8x8l_down_right_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2261 {
2262 PREDICT_8x8_LOAD_TOP;
2263 PREDICT_8x8_LOAD_LEFT;
2264 PREDICT_8x8_LOAD_TOPLEFT;
2265 SRC(0,7)= (l7 + 2*l6 + l5 + 2) >> 2;
2266 SRC(0,6)=SRC(1,7)= (l6 + 2*l5 + l4 + 2) >> 2;
2267 SRC(0,5)=SRC(1,6)=SRC(2,7)= (l5 + 2*l4 + l3 + 2) >> 2;
2268 SRC(0,4)=SRC(1,5)=SRC(2,6)=SRC(3,7)= (l4 + 2*l3 + l2 + 2) >> 2;
2269 SRC(0,3)=SRC(1,4)=SRC(2,5)=SRC(3,6)=SRC(4,7)= (l3 + 2*l2 + l1 + 2) >> 2;
2270 SRC(0,2)=SRC(1,3)=SRC(2,4)=SRC(3,5)=SRC(4,6)=SRC(5,7)= (l2 + 2*l1 + l0 + 2) >> 2;
2271 SRC(0,1)=SRC(1,2)=SRC(2,3)=SRC(3,4)=SRC(4,5)=SRC(5,6)=SRC(6,7)= (l1 + 2*l0 + lt + 2) >> 2;
2272 SRC(0,0)=SRC(1,1)=SRC(2,2)=SRC(3,3)=SRC(4,4)=SRC(5,5)=SRC(6,6)=SRC(7,7)= (l0 + 2*lt + t0 + 2) >> 2;
2273 SRC(1,0)=SRC(2,1)=SRC(3,2)=SRC(4,3)=SRC(5,4)=SRC(6,5)=SRC(7,6)= (lt + 2*t0 + t1 + 2) >> 2;
2274 SRC(2,0)=SRC(3,1)=SRC(4,2)=SRC(5,3)=SRC(6,4)=SRC(7,5)= (t0 + 2*t1 + t2 + 2) >> 2;
2275 SRC(3,0)=SRC(4,1)=SRC(5,2)=SRC(6,3)=SRC(7,4)= (t1 + 2*t2 + t3 + 2) >> 2;
2276 SRC(4,0)=SRC(5,1)=SRC(6,2)=SRC(7,3)= (t2 + 2*t3 + t4 + 2) >> 2;
2277 SRC(5,0)=SRC(6,1)=SRC(7,2)= (t3 + 2*t4 + t5 + 2) >> 2;
2278 SRC(6,0)=SRC(7,1)= (t4 + 2*t5 + t6 + 2) >> 2;
2279 SRC(7,0)= (t5 + 2*t6 + t7 + 2) >> 2;
2280
2281 }
2282 static void pred8x8l_vertical_right_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2283 {
2284 PREDICT_8x8_LOAD_TOP;
2285 PREDICT_8x8_LOAD_LEFT;
2286 PREDICT_8x8_LOAD_TOPLEFT;
2287 SRC(0,6)= (l5 + 2*l4 + l3 + 2) >> 2;
2288 SRC(0,7)= (l6 + 2*l5 + l4 + 2) >> 2;
2289 SRC(0,4)=SRC(1,6)= (l3 + 2*l2 + l1 + 2) >> 2;
2290 SRC(0,5)=SRC(1,7)= (l4 + 2*l3 + l2 + 2) >> 2;
2291 SRC(0,2)=SRC(1,4)=SRC(2,6)= (l1 + 2*l0 + lt + 2) >> 2;
2292 SRC(0,3)=SRC(1,5)=SRC(2,7)= (l2 + 2*l1 + l0 + 2) >> 2;
2293 SRC(0,1)=SRC(1,3)=SRC(2,5)=SRC(3,7)= (l0 + 2*lt + t0 + 2) >> 2;
2294 SRC(0,0)=SRC(1,2)=SRC(2,4)=SRC(3,6)= (lt + t0 + 1) >> 1;
2295 SRC(1,1)=SRC(2,3)=SRC(3,5)=SRC(4,7)= (lt + 2*t0 + t1 + 2) >> 2;
2296 SRC(1,0)=SRC(2,2)=SRC(3,4)=SRC(4,6)= (t0 + t1 + 1) >> 1;
2297 SRC(2,1)=SRC(3,3)=SRC(4,5)=SRC(5,7)= (t0 + 2*t1 + t2 + 2) >> 2;
2298 SRC(2,0)=SRC(3,2)=SRC(4,4)=SRC(5,6)= (t1 + t2 + 1) >> 1;
2299 SRC(3,1)=SRC(4,3)=SRC(5,5)=SRC(6,7)= (t1 + 2*t2 + t3 + 2) >> 2;
2300 SRC(3,0)=SRC(4,2)=SRC(5,4)=SRC(6,6)= (t2 + t3 + 1) >> 1;
2301 SRC(4,1)=SRC(5,3)=SRC(6,5)=SRC(7,7)= (t2 + 2*t3 + t4 + 2) >> 2;
2302 SRC(4,0)=SRC(5,2)=SRC(6,4)=SRC(7,6)= (t3 + t4 + 1) >> 1;
2303 SRC(5,1)=SRC(6,3)=SRC(7,5)= (t3 + 2*t4 + t5 + 2) >> 2;
2304 SRC(5,0)=SRC(6,2)=SRC(7,4)= (t4 + t5 + 1) >> 1;
2305 SRC(6,1)=SRC(7,3)= (t4 + 2*t5 + t6 + 2) >> 2;
2306 SRC(6,0)=SRC(7,2)= (t5 + t6 + 1) >> 1;
2307 SRC(7,1)= (t5 + 2*t6 + t7 + 2) >> 2;
2308 SRC(7,0)= (t6 + t7 + 1) >> 1;
2309 }
2310 static void pred8x8l_horizontal_down_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2311 {
2312 PREDICT_8x8_LOAD_TOP;
2313 PREDICT_8x8_LOAD_LEFT;
2314 PREDICT_8x8_LOAD_TOPLEFT;
2315 SRC(0,7)= (l6 + l7 + 1) >> 1;
2316 SRC(1,7)= (l5 + 2*l6 + l7 + 2) >> 2;
2317 SRC(0,6)=SRC(2,7)= (l5 + l6 + 1) >> 1;
2318 SRC(1,6)=SRC(3,7)= (l4 + 2*l5 + l6 + 2) >> 2;
2319 SRC(0,5)=SRC(2,6)=SRC(4,7)= (l4 + l5 + 1) >> 1;
2320 SRC(1,5)=SRC(3,6)=SRC(5,7)= (l3 + 2*l4 + l5 + 2) >> 2;
2321 SRC(0,4)=SRC(2,5)=SRC(4,6)=SRC(6,7)= (l3 + l4 + 1) >> 1;
2322 SRC(1,4)=SRC(3,5)=SRC(5,6)=SRC(7,7)= (l2 + 2*l3 + l4 + 2) >> 2;
2323 SRC(0,3)=SRC(2,4)=SRC(4,5)=SRC(6,6)= (l2 + l3 + 1) >> 1;
2324 SRC(1,3)=SRC(3,4)=SRC(5,5)=SRC(7,6)= (l1 + 2*l2 + l3 + 2) >> 2;
2325 SRC(0,2)=SRC(2,3)=SRC(4,4)=SRC(6,5)= (l1 + l2 + 1) >> 1;
2326 SRC(1,2)=SRC(3,3)=SRC(5,4)=SRC(7,5)= (l0 + 2*l1 + l2 + 2) >> 2;
2327 SRC(0,1)=SRC(2,2)=SRC(4,3)=SRC(6,4)= (l0 + l1 + 1) >> 1;
2328 SRC(1,1)=SRC(3,2)=SRC(5,3)=SRC(7,4)= (lt + 2*l0 + l1 + 2) >> 2;
2329 SRC(0,0)=SRC(2,1)=SRC(4,2)=SRC(6,3)= (lt + l0 + 1) >> 1;
2330 SRC(1,0)=SRC(3,1)=SRC(5,2)=SRC(7,3)= (l0 + 2*lt + t0 + 2) >> 2;
2331 SRC(2,0)=SRC(4,1)=SRC(6,2)= (t1 + 2*t0 + lt + 2) >> 2;
2332 SRC(3,0)=SRC(5,1)=SRC(7,2)= (t2 + 2*t1 + t0 + 2) >> 2;
2333 SRC(4,0)=SRC(6,1)= (t3 + 2*t2 + t1 + 2) >> 2;
2334 SRC(5,0)=SRC(7,1)= (t4 + 2*t3 + t2 + 2) >> 2;
2335 SRC(6,0)= (t5 + 2*t4 + t3 + 2) >> 2;
2336 SRC(7,0)= (t6 + 2*t5 + t4 + 2) >> 2;
2337 }
2338 static void pred8x8l_vertical_left_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2339 {
2340 PREDICT_8x8_LOAD_TOP;
2341 PREDICT_8x8_LOAD_TOPRIGHT;
2342 SRC(0,0)= (t0 + t1 + 1) >> 1;
2343 SRC(0,1)= (t0 + 2*t1 + t2 + 2) >> 2;
2344 SRC(0,2)=SRC(1,0)= (t1 + t2 + 1) >> 1;
2345 SRC(0,3)=SRC(1,1)= (t1 + 2*t2 + t3 + 2) >> 2;
2346 SRC(0,4)=SRC(1,2)=SRC(2,0)= (t2 + t3 + 1) >> 1;
2347 SRC(0,5)=SRC(1,3)=SRC(2,1)= (t2 + 2*t3 + t4 + 2) >> 2;
2348 SRC(0,6)=SRC(1,4)=SRC(2,2)=SRC(3,0)= (t3 + t4 + 1) >> 1;
2349 SRC(0,7)=SRC(1,5)=SRC(2,3)=SRC(3,1)= (t3 + 2*t4 + t5 + 2) >> 2;
2350 SRC(1,6)=SRC(2,4)=SRC(3,2)=SRC(4,0)= (t4 + t5 + 1) >> 1;
2351 SRC(1,7)=SRC(2,5)=SRC(3,3)=SRC(4,1)= (t4 + 2*t5 + t6 + 2) >> 2;
2352 SRC(2,6)=SRC(3,4)=SRC(4,2)=SRC(5,0)= (t5 + t6 + 1) >> 1;
2353 SRC(2,7)=SRC(3,5)=SRC(4,3)=SRC(5,1)= (t5 + 2*t6 + t7 + 2) >> 2;
2354 SRC(3,6)=SRC(4,4)=SRC(5,2)=SRC(6,0)= (t6 + t7 + 1) >> 1;
2355 SRC(3,7)=SRC(4,5)=SRC(5,3)=SRC(6,1)= (t6 + 2*t7 + t8 + 2) >> 2;
2356 SRC(4,6)=SRC(5,4)=SRC(6,2)=SRC(7,0)= (t7 + t8 + 1) >> 1;
2357 SRC(4,7)=SRC(5,5)=SRC(6,3)=SRC(7,1)= (t7 + 2*t8 + t9 + 2) >> 2;
2358 SRC(5,6)=SRC(6,4)=SRC(7,2)= (t8 + t9 + 1) >> 1;
2359 SRC(5,7)=SRC(6,5)=SRC(7,3)= (t8 + 2*t9 + t10 + 2) >> 2;
2360 SRC(6,6)=SRC(7,4)= (t9 + t10 + 1) >> 1;
2361 SRC(6,7)=SRC(7,5)= (t9 + 2*t10 + t11 + 2) >> 2;
2362 SRC(7,6)= (t10 + t11 + 1) >> 1;
2363 SRC(7,7)= (t10 + 2*t11 + t12 + 2) >> 2;
2364 }
2365 static void pred8x8l_horizontal_up_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2366 {
2367 PREDICT_8x8_LOAD_LEFT;
2368 SRC(0,0)= (l0 + l1 + 1) >> 1;
2369 SRC(1,0)= (l0 + 2*l1 + l2 + 2) >> 2;
2370 SRC(0,1)=SRC(2,0)= (l1 + l2 + 1) >> 1;
2371 SRC(1,1)=SRC(3,0)= (l1 + 2*l2 + l3 + 2) >> 2;
2372 SRC(0,2)=SRC(2,1)=SRC(4,0)= (l2 + l3 + 1) >> 1;
2373 SRC(1,2)=SRC(3,1)=SRC(5,0)= (l2 + 2*l3 + l4 + 2) >> 2;
2374 SRC(0,3)=SRC(2,2)=SRC(4,1)=SRC(6,0)= (l3 + l4 + 1) >> 1;
2375 SRC(1,3)=SRC(3,2)=SRC(5,1)=SRC(7,0)= (l3 + 2*l4 + l5 + 2) >> 2;
2376 SRC(0,4)=SRC(2,3)=SRC(4,2)=SRC(6,1)= (l4 + l5 + 1) >> 1;
2377 SRC(1,4)=SRC(3,3)=SRC(5,2)=SRC(7,1)= (l4 + 2*l5 + l6 + 2) >> 2;
2378 SRC(0,5)=SRC(2,4)=SRC(4,3)=SRC(6,2)= (l5 + l6 + 1) >> 1;
2379 SRC(1,5)=SRC(3,4)=SRC(5,3)=SRC(7,2)= (l5 + 2*l6 + l7 + 2) >> 2;
2380 SRC(0,6)=SRC(2,5)=SRC(4,4)=SRC(6,3)= (l6 + l7 + 1) >> 1;
2381 SRC(1,6)=SRC(3,5)=SRC(5,4)=SRC(7,3)= (l6 + 3*l7 + 2) >> 2;
2382 SRC(0,7)=SRC(1,7)=SRC(2,6)=SRC(2,7)=SRC(3,6)=
2383 SRC(3,7)=SRC(4,5)=SRC(4,6)=SRC(4,7)=SRC(5,5)=
2384 SRC(5,6)=SRC(5,7)=SRC(6,4)=SRC(6,5)=SRC(6,6)=
2385 SRC(6,7)=SRC(7,4)=SRC(7,5)=SRC(7,6)=SRC(7,7)= l7;
2386 }
2387 #undef PREDICT_8x8_LOAD_LEFT
2388 #undef PREDICT_8x8_LOAD_TOP
2389 #undef PREDICT_8x8_LOAD_TOPLEFT
2390 #undef PREDICT_8x8_LOAD_TOPRIGHT
2391 #undef PREDICT_8x8_DC
2392 #undef PTR
2393 #undef PT
2394 #undef PL
2395 #undef SRC
2396
2397 static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
2398 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2399 int src_x_offset, int src_y_offset,
2400 qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
2401 MpegEncContext * const s = &h->s;
2402 const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
2403 int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
2404 const int luma_xy= (mx&3) + ((my&3)<<2);
2405 uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*h->mb_linesize;
2406 uint8_t * src_cb, * src_cr;
2407 int extra_width= h->emu_edge_width;
2408 int extra_height= h->emu_edge_height;
2409 int emu=0;
2410 const int full_mx= mx>>2;
2411 const int full_my= my>>2;
2412 const int pic_width = 16*s->mb_width;
2413 const int pic_height = 16*s->mb_height >> MB_MBAFF;
2414
2415 if(!pic->data[0]) //FIXME this is unacceptable, some senseable error concealment must be done for missing reference frames
2416 return;
2417
2418 if(mx&7) extra_width -= 3;
2419 if(my&7) extra_height -= 3;
2420
2421 if( full_mx < 0-extra_width
2422 || full_my < 0-extra_height
2423 || full_mx + 16/*FIXME*/ > pic_width + extra_width
2424 || full_my + 16/*FIXME*/ > pic_height + extra_height){
2425 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);
2426 src_y= s->edge_emu_buffer + 2 + 2*h->mb_linesize;
2427 emu=1;
2428 }
2429
2430 qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); //FIXME try variable height perhaps?
2431 if(!square){
2432 qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
2433 }
2434
2435 if(s->flags&CODEC_FLAG_GRAY) return;
2436
2437 if(MB_MBAFF){
2438 // chroma offset when predicting from a field of opposite parity
2439 my += 2 * ((s->mb_y & 1) - (h->ref_cache[list][scan8[n]] & 1));
2440 emu |= (my>>3) < 0 || (my>>3) + 8 >= (pic_height>>1);
2441 }
2442 src_cb= pic->data[1] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
2443 src_cr= pic->data[2] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
2444
2445 if(emu){
2446 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);
2447 src_cb= s->edge_emu_buffer;
2448 }
2449 chroma_op(dest_cb, src_cb, h->mb_uvlinesize, chroma_height, mx&7, my&7);
2450
2451 if(emu){
2452 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);
2453 src_cr= s->edge_emu_buffer;
2454 }
2455 chroma_op(dest_cr, src_cr, h->mb_uvlinesize, chroma_height, mx&7, my&7);
2456 }
2457
2458 static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
2459 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2460 int x_offset, int y_offset,
2461 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2462 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
2463 int list0, int list1){
2464 MpegEncContext * const s = &h->s;
2465 qpel_mc_func *qpix_op= qpix_put;
2466 h264_chroma_mc_func chroma_op= chroma_put;
2467
2468 dest_y += 2*x_offset + 2*y_offset*h-> mb_linesize;
2469 dest_cb += x_offset + y_offset*h->mb_uvlinesize;
2470 dest_cr += x_offset + y_offset*h->mb_uvlinesize;
2471 x_offset += 8*s->mb_x;
2472 y_offset += 8*(s->mb_y >> MB_MBAFF);
2473
2474 if(list0){
2475 Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
2476 mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
2477 dest_y, dest_cb, dest_cr, x_offset, y_offset,
2478 qpix_op, chroma_op);
2479
2480 qpix_op= qpix_avg;
2481 chroma_op= chroma_avg;
2482 }
2483
2484 if(list1){
2485 Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
2486 mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
2487 dest_y, dest_cb, dest_cr, x_offset, y_offset,
2488 qpix_op, chroma_op);
2489 }
2490 }
2491
2492 static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
2493 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2494 int x_offset, int y_offset,
2495 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2496 h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
2497 h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
2498 int list0, int list1){
2499 MpegEncContext * const s = &h->s;
2500
2501 dest_y += 2*x_offset + 2*y_offset*h-> mb_linesize;
2502 dest_cb += x_offset + y_offset*h->mb_uvlinesize;
2503 dest_cr += x_offset + y_offset*h->mb_uvlinesize;
2504 x_offset += 8*s->mb_x;
2505 y_offset += 8*(s->mb_y >> MB_MBAFF);
2506
2507 if(list0 && list1){
2508 /* don't optimize for luma-only case, since B-frames usually
2509 * use implicit weights => chroma too. */
2510 uint8_t *tmp_cb = s->obmc_scratchpad;
2511 uint8_t *tmp_cr = s->obmc_scratchpad + 8;
2512 uint8_t *tmp_y = s->obmc_scratchpad + 8*h->mb_uvlinesize;
2513 int refn0 = h->ref_cache[0][ scan8[n] ];
2514 int refn1 = h->ref_cache[1][ scan8[n] ];
2515
2516 mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
2517 dest_y, dest_cb, dest_cr,
2518 x_offset, y_offset, qpix_put, chroma_put);
2519 mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
2520 tmp_y, tmp_cb, tmp_cr,
2521 x_offset, y_offset, qpix_put, chroma_put);
2522
2523 if(h->use_weight == 2){
2524 int weight0 = h->implicit_weight[refn0][refn1];
2525 int weight1 = 64 - weight0;
2526 luma_weight_avg( dest_y, tmp_y, h-> mb_linesize, 5, weight0, weight1, 0);
2527 chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, 5, weight0, weight1, 0);
2528 chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, 5, weight0, weight1, 0);
2529 }else{
2530 luma_weight_avg(dest_y, tmp_y, h->mb_linesize, h->luma_log2_weight_denom,
2531 h->luma_weight[0][refn0], h->luma_weight[1][refn1],
2532 h->luma_offset[0][refn0] + h->luma_offset[1][refn1]);
2533 chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
2534 h->chroma_weight[0][refn0][0], h->chroma_weight[1][refn1][0],
2535 h->chroma_offset[0][refn0][0] + h->chroma_offset[1][refn1][0]);
2536 chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
2537 h->chroma_weight[0][refn0][1], h->chroma_weight[1][refn1][1],
2538 h->chroma_offset[0][refn0][1] + h->chroma_offset[1][refn1][1]);
2539 }
2540 }else{
2541 int list = list1 ? 1 : 0;
2542 int refn = h->ref_cache[list][ scan8[n] ];
2543 Picture *ref= &h->ref_list[list][refn];
2544 mc_dir_part(h, ref, n, square, chroma_height, delta, list,
2545 dest_y, dest_cb, dest_cr, x_offset, y_offset,
2546 qpix_put, chroma_put);
2547
2548 luma_weight_op(dest_y, h->mb_linesize, h->luma_log2_weight_denom,
2549 h->luma_weight[list][refn], h->luma_offset[list][refn]);
2550 if(h->use_weight_chroma){
2551 chroma_weight_op(dest_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
2552 h->chroma_weight[list][refn][0], h->chroma_offset[list][refn][0]);
2553 chroma_weight_op(dest_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
2554 h->chroma_weight[list][refn][1], h->chroma_offset[list][refn][1]);
2555 }
2556 }
2557 }
2558
2559 static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
2560 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2561 int x_offset, int y_offset,
2562 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2563 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
2564 h264_weight_func *weight_op, h264_biweight_func *weight_avg,
2565 int list0, int list1){
2566 if((h->use_weight==2 && list0 && list1
2567 && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ] != 32))
2568 || h->use_weight==1)
2569 mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
2570 x_offset, y_offset, qpix_put, chroma_put,
2571 weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
2572 else
2573 mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
2574 x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
2575 }
2576
2577 static inline void prefetch_motion(H264Context *h, int list){
2578 /* fetch pixels for estimated mv 4 macroblocks ahead
2579 * optimized for 64byte cache lines */
2580 MpegEncContext * const s = &h->s;
2581 const int refn = h->ref_cache[list][scan8[0]];
2582 if(refn >= 0){
2583 const int mx= (h->mv_cache[list][scan8[0]][0]>>2) + 16*s->mb_x + 8;
2584 const int my= (h->mv_cache[list][scan8[0]][1]>>2) + 16*s->mb_y;
2585 uint8_t **src= h->ref_list[list][refn].data;
2586 int off= mx + (my + (s->mb_x&3)*4)*h->mb_linesize + 64;
2587 s->dsp.prefetch(src[0]+off, s->linesize, 4);
2588 off= (mx>>1) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + 64;
2589 s->dsp.prefetch(src[1]+off, src[2]-src[1], 2);
2590 }
2591 }
2592
2593 static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2594 qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
2595 qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
2596 h264_weight_func *weight_op, h264_biweight_func *weight_avg){
2597 MpegEncContext * const s = &h->s;
2598 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
2599 const int mb_type= s->current_picture.mb_type[mb_xy];
2600
2601 assert(IS_INTER(mb_type));
2602
2603 prefetch_motion(h, 0);
2604
2605 if(IS_16X16(mb_type)){
2606 mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
2607 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
2608 &weight_op[0], &weight_avg[0],
2609 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2610 }else if(IS_16X8(mb_type)){
2611 mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
2612 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
2613 &weight_op[1], &weight_avg[1],
2614 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2615 mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
2616 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
2617 &weight_op[1], &weight_avg[1],
2618 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2619 }else if(IS_8X16(mb_type)){
2620 mc_part(h, 0, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 0, 0,
2621 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2622 &weight_op[2], &weight_avg[2],
2623 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2624 mc_part(h, 4, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 4, 0,
2625 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2626 &weight_op[2], &weight_avg[2],
2627 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2628 }else{
2629 int i;
2630
2631 assert(IS_8X8(mb_type));
2632
2633 for(i=0; i<4; i++){
2634 const int sub_mb_type= h->sub_mb_type[i];
2635 const int n= 4*i;
2636 int x_offset= (i&1)<<2;
2637 int y_offset= (i&2)<<1;
2638
2639 if(IS_SUB_8X8(sub_mb_type)){
2640 mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2641 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2642 &weight_op[3], &weight_avg[3],
2643 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2644 }else if(IS_SUB_8X4(sub_mb_type)){
2645 mc_part(h, n , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2646 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2647 &weight_op[4], &weight_avg[4],
2648 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2649 mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
2650 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2651 &weight_op[4], &weight_avg[4],
2652 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2653 }else if(IS_SUB_4X8(sub_mb_type)){
2654 mc_part(h, n , 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2655 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2656 &weight_op[5], &weight_avg[5]