2 * H.26L/H.264/AVC/JVT/14496-10/... encoder/decoder
3 * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
5 * This file is part of FFmpeg.
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.
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.
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
25 * H.264 / AVC / MPEG4 part10 codec.
26 * @author Michael Niedermayer <michaelni@gmx.at>
31 #include "mpegvideo.h"
34 #include "h264_parser.h"
42 static VLC coeff_token_vlc[4];
43 static VLC chroma_dc_coeff_token_vlc;
45 static VLC total_zeros_vlc[15];
46 static VLC chroma_dc_total_zeros_vlc[3];
48 static VLC run_vlc[6];
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);
56 static av_always_inline uint32_t pack16to32(int a, int b){
57 #ifdef WORDS_BIGENDIAN
58 return (b&0xFFFF) + (a<<16);
60 return (a&0xFFFF) + (b<<16);
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,
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,
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
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);
87 assert((((long)vp)&(FFMIN(w, STRIDE_ALIGN)-1)) == 0);
88 assert((stride&(w-1))==0);
90 const uint16_t v= size==4 ? val : val*0x0101;
91 *(uint16_t*)(p + 0*stride)= v;
93 *(uint16_t*)(p + 1*stride)= v;
95 *(uint16_t*)(p + 2*stride)=
96 *(uint16_t*)(p + 3*stride)= v;
98 const uint32_t v= size==4 ? val : val*0x01010101;
99 *(uint32_t*)(p + 0*stride)= v;
101 *(uint32_t*)(p + 1*stride)= v;
103 *(uint32_t*)(p + 2*stride)=
104 *(uint32_t*)(p + 3*stride)= v;
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;
111 *(uint64_t*)(p + 1*stride)= v;
113 *(uint64_t*)(p + 2*stride)=
114 *(uint64_t*)(p + 3*stride)= v;
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;
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;
127 *(uint32_t*)(p + 0+0*stride)=
128 *(uint32_t*)(p + 4+0*stride)= val;
130 *(uint32_t*)(p + 0+1*stride)=
131 *(uint32_t*)(p + 4+1*stride)= val;
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;
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;
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;
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];
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)
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
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;
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);
200 ? !curr_mb_frame_flag // bottom macroblock
201 : (!curr_mb_frame_flag && !top_mb_frame_flag) // top macroblock
203 top_xy -= s->mb_stride;
206 ? !curr_mb_frame_flag // bottom macroblock
207 : (!curr_mb_frame_flag && !topleft_mb_frame_flag) // top macroblock
209 topleft_xy -= s->mb_stride;
212 ? !curr_mb_frame_flag // bottom macroblock
213 : (!curr_mb_frame_flag && !topright_mb_frame_flag) // top macroblock
215 topright_xy -= s->mb_stride;
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) {
240 left_xy[1] += s->mb_stride;
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];
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;
263 if(FRAME_MBAFF && !IS_INTRA(mb_type)){
265 int v = *(uint16_t*)&h->non_zero_count[mb_xy][14];
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){
279 *(uint32_t*)&h->ref_cache[list][scan8[ 0]] =
280 *(uint32_t*)&h->ref_cache[list][scan8[ 2]] = pack16to32(ref[0],ref[1])*0x0101;
282 *(uint32_t*)&h->ref_cache[list][scan8[ 8]] =
283 *(uint32_t*)&h->ref_cache[list][scan8[10]] = pack16to32(ref[0],ref[1])*0x0101;
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);
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;
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;
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;
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;
316 if(!IS_INTRA(topleft_type) && (topleft_type==0 || h->pps.constrained_intra_pred))
317 h->topleft_samples_available&= 0x7FFF;
319 if(!IS_INTRA(topright_type) && (topright_type==0 || h->pps.constrained_intra_pred))
320 h->topright_samples_available&= 0xFBFF;
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];
330 if(!top_type || (IS_INTER(top_type) && h->pps.constrained_intra_pred))
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;
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]];
346 if(!left_type[i] || (IS_INTER(left_type[i]) && h->pps.constrained_intra_pred))
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;
367 //FIXME constraint_intra_pred & partitioning & nnz (lets hope this is just a typo in the spec)
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];
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];
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];
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]=
386 h->non_zero_count_cache[1+8*0]=
387 h->non_zero_count_cache[2+8*0]=
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;
394 for (i=0; i<2; 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]];
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;
411 h->top_cbp = h->cbp_table[top_xy];
412 } else if(IS_INTRA(mb_type)) {
419 h->left_cbp = h->cbp_table[left_xy[0]] & 0x1f0;
420 } else if(IS_INTRA(mb_type)) {
426 h->left_cbp |= ((h->cbp_table[left_xy[0]]>>((left_block[0]&(~1))+1))&0x1) << 1;
429 h->left_cbp |= ((h->cbp_table[left_xy[1]]>>((left_block[2]&(~1))+1))&0x1) << 3;
434 if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
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;
445 h->mv_cache_clean[list]= 0;
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];
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;
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)];
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;
483 if((for_deblock || (IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred)) && !FRAME_MBAFF)
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];
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;
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];
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;
506 if((IS_SKIP(mb_type) || IS_DIRECT(mb_type)) && !FRAME_MBAFF)
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;
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];
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;
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]];
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;
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]];
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;
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;
556 if(h->slice_type == B_TYPE){
557 fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, 0, 1);
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];
566 *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0;
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)];
574 h->direct_cache[scan8[0] - 1 + 0*8]= 0;
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)];
581 h->direct_cache[scan8[0] - 1 + 2*8]= 0;
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])
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;\
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;\
621 h->neighbor_transform_size= !!IS_8x8DCT(top_type) + !!IS_8x8DCT(left_type[0]);
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;
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];
638 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
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};
646 if(!(h->top_samples_available&0x8000)){
648 int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
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);
653 h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
658 if(!(h->left_samples_available&0x8000)){
660 int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
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);
665 h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
671 } //FIXME cleanup like next
674 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
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};
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);
686 if(!(h->top_samples_available&0x8000)){
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);
694 if(!(h->left_samples_available&0x8000)){
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);
706 * gets the predicted intra4x4 prediction mode.
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);
714 tprintf(h->s.avctx, "mode:%d %d min:%d\n", left ,top, min);
716 if(min<0) return DC_PRED;
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;
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];
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];
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];
741 // store all luma nnzs, for deblocking
744 v += (!!h->non_zero_count_cache[scan8[i]]) << i;
745 *(uint16_t*)&h->non_zero_count[mb_xy][14] = v;
750 * gets the predicted number of non zero coefficients.
751 * @param n block index
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];
759 if(i<64) i= (i+1)>>1;
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);
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;
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 */
773 const uint32_t *mb_types = s->current_picture_ptr->mb_type;
775 *(uint32_t*)h->mv_cache[list][scan8[0]-2] = 0;
776 *C = h->mv_cache[list][scan8[0]-2];
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;
792 SET_DIAG_MV(*2, >>1, s->mb_x*4+(i&7)-4+part_width, s->mb_y*4-1);
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){
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);
803 && !IS_INTERLACED(mb_types[h->left_mb_xy[0]])
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);
812 if(topright_ref != PART_NOT_AVAILABLE){
813 *C= h->mv_cache[list][ i - 8 + part_width ];
816 tprintf(s->avctx, "topright MV not available\n");
818 *C= h->mv_cache[list][ i - 8 - 1 ];
819 return h->ref_cache[list][ i - 8 - 1 ];
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
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 ];
837 int diagonal_ref, match_count;
839 assert(part_width==1 || part_width==2 || part_width==4);
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){
859 }else if(top_ref==ref){
867 if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
871 *mx= mid_pred(A[0], B[0], C[0]);
872 *my= mid_pred(A[1], B[1], C[1]);
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);
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
885 static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
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 ];
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);
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 ];
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);
911 pred_motion(h, n, 4, list, ref, mx, my);
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
920 static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
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 ];
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);
936 diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);
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);
940 if(diagonal_ref == ref){
948 pred_motion(h, n, 2, list, ref, mx, my);
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 ];
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);
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)){
965 pred_motion(h, 0, 4, 0, 0, mx, my);
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;
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;
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);
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];
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;
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;
1006 if(cur->pict_type != B_TYPE || h->direct_spatial_mv_pred)
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;
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;
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;
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 */
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;
1058 *mb_type |= MB_TYPE_DIRECT2;
1060 *mb_type |= MB_TYPE_INTERLACED;
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);
1064 if(h->direct_spatial_mv_pred){
1069 /* FIXME interlacing + spatial direct uses wrong colocated block positions */
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];
1077 refc = h->ref_cache[list][scan8[0] - 8 - 1];
1079 if(ref[list] < 0 || (refb < ref[list] && refb >= 0))
1081 if(ref[list] < 0 || (refc < ref[list] && refc >= 0))
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;
1092 for(list=0; list<2; list++){
1094 pred_motion(h, 0, 4, list, ref[list], &mv[list][0], &mv[list][1]);
1096 mv[list][0] = mv[list][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;
1108 if(IS_16X16(*mb_type)){
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)))){
1118 a= pack16to32(mv[0][0],mv[0][1]);
1120 b= pack16to32(mv[1][0],mv[1][1]);
1122 a= pack16to32(mv[0][0],mv[0][1]);
1123 b= pack16to32(mv[1][0],mv[1][1]);
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);
1128 for(i8=0; i8<4; i8++){
1129 const int x8 = i8&1;
1130 const int y8 = i8>>1;
1132 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1134 h->sub_mb_type[i8] = sub_mb_type;
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);
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){
1150 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1152 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
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){
1159 *(uint32_t*)h->mv_cache[0][scan8[i8*4+i4]] = 0;
1161 *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = 0;
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;
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;
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];
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;
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];
1193 l1ref0 -= 2*h->b8_stride;
1194 l1ref1 -= 2*h->b8_stride;
1195 l1mv0 -= 4*h->b_stride;
1196 l1mv1 -= 4*h->b_stride;
1200 if( (mb_types_col[0] & MB_TYPE_16x16_OR_INTRA)
1201 && (mb_types_col[1] & MB_TYPE_16x16_OR_INTRA)
1203 *mb_type |= MB_TYPE_16x8;
1205 *mb_type |= MB_TYPE_8x8;
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;
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;
1219 if((mb_types_col[0] & (MB_TYPE_16x16_OR_INTRA|MB_TYPE_16x8))
1221 *mb_type |= MB_TYPE_16x16;
1223 *mb_type |= MB_TYPE_8x8;
1226 for(i8=0; i8<4; i8++){
1227 const int x8 = i8&1;
1228 const int y8 = i8>>1;
1230 const int16_t (*l1mv)[2]= l1mv0;
1232 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1234 h->sub_mb_type[i8] = sub_mb_type;
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);
1244 ref0 = l1ref0[x8 + (y8*2>>y_shift)*h->b8_stride];
1246 ref0 = map_col_to_list0[0][ref0*2>>y_shift];
1248 ref0 = map_col_to_list0[1][l1ref1[x8 + (y8*2>>y_shift)*h->b8_stride]*2>>y_shift];
1251 scale = dist_scale_factor[ref0];
1252 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
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);
1267 /* one-to-one mv scaling */
1269 if(IS_16X16(*mb_type)){
1272 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1);
1273 if(IS_INTRA(mb_type_col)){
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];
1281 mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
1282 mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
1284 mv0= pack16to32(mv_l0[0],mv_l0[1]);
1285 mv1= pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
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);
1291 for(i8=0; i8<4; i8++){
1292 const int x8 = i8&1;
1293 const int y8 = i8>>1;
1295 const int16_t (*l1mv)[2]= l1mv0;
1297 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
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);
1308 ref0 = l1ref0[x8 + y8*h->b8_stride];
1310 ref0 = map_col_to_list0[0][ref0];
1312 ref0 = map_col_to_list0[1][l1ref1[x8 + y8*h->b8_stride]];
1315 scale = dist_scale_factor[ref0];
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);
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]);
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;
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);
1347 for(list=0; list<h->list_count; list++){
1349 if(!USES_LIST(mb_type, list))
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];
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);
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];
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]];
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;
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
1392 static uint8_t *decode_nal(H264Context *h, uint8_t *src, int *dst_length, int *consumed, int length){
1397 // src[0]&0x80; //forbidden bit
1398 h->nal_ref_idc= src[0]>>5;
1399 h->nal_unit_type= src[0]&0x1F;
1403 for(i=0; i<length; i++)
1404 printf("%2X ", src[i]);
1406 for(i=0; i+1<length; i+=2){
1407 if(src[i]) continue;
1408 if(i>0 && src[i-1]==0) i--;
1409 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1411 /* startcode, so we must be past the end */
1418 if(i>=length-1){ //no escaped 0
1419 *dst_length= length;
1420 *consumed= length+1; //+1 for the header
1424 bufidx = h->nal_unit_type == NAL_DPC ? 1 : 0; // use second escape buffer for inter data
1425 h->rbsp_buffer[bufidx]= av_fast_realloc(h->rbsp_buffer[bufidx], &h->rbsp_buffer_size[bufidx], length);
1426 dst= h->rbsp_buffer[bufidx];
1432 //printf("decoding esc\n");
1435 //remove escapes (very rare 1:2^22)
1436 if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1437 if(src[si+2]==3){ //escape
1442 }else //next start code
1446 dst[di++]= src[si++];
1450 *consumed= si + 1;//+1 for the header
1451 //FIXME store exact number of bits in the getbitcontext (it is needed for decoding)
1456 * identifies the exact end of the bitstream
1457 * @return the length of the trailing, or 0 if damaged
1459 static int decode_rbsp_trailing(H264Context *h, uint8_t *src){
1463 tprintf(h->s.avctx, "rbsp trailing %X\n", v);
1473 * idct tranforms the 16 dc values and dequantize them.
1474 * @param qp quantization parameter
1476 static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
1479 int temp[16]; //FIXME check if this is a good idea
1480 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
1481 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1483 //memset(block, 64, 2*256);
1486 const int offset= y_offset[i];
1487 const int z0= block[offset+stride*0] + block[offset+stride*4];
1488 const int z1= block[offset+stride*0] - block[offset+stride*4];
1489 const int z2= block[offset+stride*1] - block[offset+stride*5];
1490 const int z3= block[offset+stride*1] + block[offset+stride*5];
1499 const int offset= x_offset[i];
1500 const int z0= temp[4*0+i] + temp[4*2+i];
1501 const int z1= temp[4*0+i] - temp[4*2+i];
1502 const int z2= temp[4*1+i] - temp[4*3+i];
1503 const int z3= temp[4*1+i] + temp[4*3+i];
1505 block[stride*0 +offset]= ((((z0 + z3)*qmul + 128 ) >> 8)); //FIXME think about merging this into decode_resdual
1506 block[stride*2 +offset]= ((((z1 + z2)*qmul + 128 ) >> 8));
1507 block[stride*8 +offset]= ((((z1 - z2)*qmul + 128 ) >> 8));
1508 block[stride*10+offset]= ((((z0 - z3)*qmul + 128 ) >> 8));
1514 * dct tranforms the 16 dc values.
1515 * @param qp quantization parameter ??? FIXME
1517 static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
1518 // const int qmul= dequant_coeff[qp][0];
1520 int temp[16]; //FIXME check if this is a good idea
1521 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
1522 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1525 const int offset= y_offset[i];
1526 const int z0= block[offset+stride*0] + block[offset+stride*4];
1527 const int z1= block[offset+stride*0] - block[offset+stride*4];
1528 const int z2= block[offset+stride*1] - block[offset+stride*5];
1529 const int z3= block[offset+stride*1] + block[offset+stride*5];
1538 const int offset= x_offset[i];
1539 const int z0= temp[4*0+i] + temp[4*2+i];
1540 const int z1= temp[4*0+i] - temp[4*2+i];
1541 const int z2= temp[4*1+i] - temp[4*3+i];
1542 const int z3= temp[4*1+i] + temp[4*3+i];
1544 block[stride*0 +offset]= (z0 + z3)>>1;
1545 block[stride*2 +offset]= (z1 + z2)>>1;
1546 block[stride*8 +offset]= (z1 - z2)>>1;
1547 block[stride*10+offset]= (z0 - z3)>>1;
1555 static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
1556 const int stride= 16*2;
1557 const int xStride= 16;
1560 a= block[stride*0 + xStride*0];
1561 b= block[stride*0 + xStride*1];
1562 c= block[stride*1 + xStride*0];
1563 d= block[stride*1 + xStride*1];
1570 block[stride*0 + xStride*0]= ((a+c)*qmul) >> 7;
1571 block[stride*0 + xStride*1]= ((e+b)*qmul) >> 7;
1572 block[stride*1 + xStride*0]= ((a-c)*qmul) >> 7;
1573 block[stride*1 + xStride*1]= ((e-b)*qmul) >> 7;
1577 static void chroma_dc_dct_c(DCTELEM *block){
1578 const int stride= 16*2;
1579 const int xStride= 16;
1582 a= block[stride*0 + xStride*0];
1583 b= block[stride*0 + xStride*1];
1584 c= block[stride*1 + xStride*0];
1585 d= block[stride*1 + xStride*1];
1592 block[stride*0 + xStride*0]= (a+c);
1593 block[stride*0 + xStride*1]= (e+b);
1594 block[stride*1 + xStride*0]= (a-c);
1595 block[stride*1 + xStride*1]= (e-b);
1600 * gets the chroma qp.
1602 static inline int get_chroma_qp(int chroma_qp_index_offset, int qscale){
1604 return chroma_qp[av_clip(qscale + chroma_qp_index_offset, 0, 51)];
1607 //FIXME need to check that this does not overflow signed 32 bit for low qp, i am not sure, it's very close
1608 //FIXME check that gcc inlines this (and optimizes intra & separate_dc stuff away)
1609 static inline int quantize_c(DCTELEM *block, uint8_t *scantable, int qscale, int intra, int separate_dc){
1611 const int * const quant_table= quant_coeff[qscale];
1612 const int bias= intra ? (1<<QUANT_SHIFT)/3 : (1<<QUANT_SHIFT)/6;
1613 const unsigned int threshold1= (1<<QUANT_SHIFT) - bias - 1;
1614 const unsigned int threshold2= (threshold1<<1);
1620 const int dc_bias= intra ? (1<<(QUANT_SHIFT-2))/3 : (1<<(QUANT_SHIFT-2))/6;
1621 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT-2)) - dc_bias - 1;
1622 const unsigned int dc_threshold2= (dc_threshold1<<1);
1624 int level= block[0]*quant_coeff[qscale+18][0];
1625 if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1627 level= (dc_bias + level)>>(QUANT_SHIFT-2);
1630 level= (dc_bias - level)>>(QUANT_SHIFT-2);
1633 // last_non_zero = i;
1638 const int dc_bias= intra ? (1<<(QUANT_SHIFT+1))/3 : (1<<(QUANT_SHIFT+1))/6;
1639 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT+1)) - dc_bias - 1;
1640 const unsigned int dc_threshold2= (dc_threshold1<<1);
1642 int level= block[0]*quant_table[0];
1643 if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1645 level= (dc_bias + level)>>(QUANT_SHIFT+1);
1648 level= (dc_bias - level)>>(QUANT_SHIFT+1);
1651 // last_non_zero = i;
1664 const int j= scantable[i];
1665 int level= block[j]*quant_table[j];
1667 // if( bias+level >= (1<<(QMAT_SHIFT - 3))
1668 // || bias-level >= (1<<(QMAT_SHIFT - 3))){
1669 if(((unsigned)(level+threshold1))>threshold2){
1671 level= (bias + level)>>QUANT_SHIFT;
1674 level= (bias - level)>>QUANT_SHIFT;
1683 return last_non_zero;
1686 static void pred4x4_vertical_c(uint8_t *src, uint8_t *topright, int stride){
1687 const uint32_t a= ((uint32_t*)(src-stride))[0];
1688 ((uint32_t*)(src+0*stride))[0]= a;
1689 ((uint32_t*)(src+1*stride))[0]= a;
1690 ((uint32_t*)(src+2*stride))[0]= a;
1691 ((uint32_t*)(src+3*stride))[0]= a;
1694 static void pred4x4_horizontal_c(uint8_t *src, uint8_t *topright, int stride){
1695 ((uint32_t*)(src+0*stride))[0]= src[-1+0*stride]*0x01010101;
1696 ((uint32_t*)(src+1*stride))[0]= src[-1+1*stride]*0x01010101;
1697 ((uint32_t*)(src+2*stride))[0]= src[-1+2*stride]*0x01010101;
1698 ((uint32_t*)(src+3*stride))[0]= src[-1+3*stride]*0x01010101;
1701 static void pred4x4_dc_c(uint8_t *src, uint8_t *topright, int stride){
1702 const int dc= ( src[-stride] + src[1-stride] + src[2-stride] + src[3-stride]
1703 + src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 4) >>3;
1705 ((uint32_t*)(src+0*stride))[0]=
1706 ((uint32_t*)(src+1*stride))[0]=
1707 ((uint32_t*)(src+2*stride))[0]=
1708 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1711 static void pred4x4_left_dc_c(uint8_t *src, uint8_t *topright, int stride){
1712 const int dc= ( src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 2) >>2;
1714 ((uint32_t*)(src+0*stride))[0]=
1715 ((uint32_t*)(src+1*stride))[0]=
1716 ((uint32_t*)(src+2*stride))[0]=
1717 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1720 static void pred4x4_top_dc_c(uint8_t *src, uint8_t *topright, int stride){
1721 const int dc= ( src[-stride] + src[1-stride] + src[2-stride] + src[3-stride] + 2) >>2;
1723 ((uint32_t*)(src+0*stride))[0]=
1724 ((uint32_t*)(src+1*stride))[0]=
1725 ((uint32_t*)(src+2*stride))[0]=
1726 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1729 static void pred4x4_128_dc_c(uint8_t *src, uint8_t *topright, int stride){
1730 ((uint32_t*)(src+0*stride))[0]=
1731 ((uint32_t*)(src+1*stride))[0]=
1732 ((uint32_t*)(src+2*stride))[0]=
1733 ((uint32_t*)(src+3*stride))[0]= 128U*0x01010101U;
1737 #define LOAD_TOP_RIGHT_EDGE\
1738 const int av_unused t4= topright[0];\
1739 const int av_unused t5= topright[1];\
1740 const int av_unused t6= topright[2];\
1741 const int av_unused t7= topright[3];\
1743 #define LOAD_LEFT_EDGE\
1744 const int av_unused l0= src[-1+0*stride];\
1745 const int av_unused l1= src[-1+1*stride];\
1746 const int av_unused l2= src[-1+2*stride];\
1747 const int av_unused l3= src[-1+3*stride];\
1749 #define LOAD_TOP_EDGE\
1750 const int av_unused t0= src[ 0-1*stride];\
1751 const int av_unused t1= src[ 1-1*stride];\
1752 const int av_unused t2= src[ 2-1*stride];\
1753 const int av_unused t3= src[ 3-1*stride];\
1755 static void pred4x4_down_right_c(uint8_t *src, uint8_t *topright, int stride){
1756 const int lt= src[-1-1*stride];
1760 src[0+3*stride]=(l3 + 2*l2 + l1 + 2)>>2;
1762 src[1+3*stride]=(l2 + 2*l1 + l0 + 2)>>2;
1765 src[2+3*stride]=(l1 + 2*l0 + lt + 2)>>2;
1769 src[3+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1772 src[3+2*stride]=(lt + 2*t0 + t1 + 2)>>2;
1774 src[3+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1775 src[3+0*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1778 static void pred4x4_down_left_c(uint8_t *src, uint8_t *topright, int stride){
1783 src[0+0*stride]=(t0 + t2 + 2*t1 + 2)>>2;
1785 src[0+1*stride]=(t1 + t3 + 2*t2 + 2)>>2;
1788 src[0+2*stride]=(t2 + t4 + 2*t3 + 2)>>2;
1792 src[0+3*stride]=(t3 + t5 + 2*t4 + 2)>>2;
1795 src[1+3*stride]=(t4 + t6 + 2*t5 + 2)>>2;
1797 src[2+3*stride]=(t5 + t7 + 2*t6 + 2)>>2;
1798 src[3+3*stride]=(t6 + 3*t7 + 2)>>2;
1801 static void pred4x4_vertical_right_c(uint8_t *src, uint8_t *topright, int stride){
1802 const int lt= src[-1-1*stride];
1807 src[1+2*stride]=(lt + t0 + 1)>>1;
1809 src[2+2*stride]=(t0 + t1 + 1)>>1;
1811 src[3+2*stride]=(t1 + t2 + 1)>>1;
1812 src[3+0*stride]=(t2 + t3 + 1)>>1;
1814 src[1+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1816 src[2+3*stride]=(lt + 2*t0 + t1 + 2)>>2;
1818 src[3+3*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1819 src[3+1*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1820 src[0+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1821 src[0+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1824 static void pred4x4_vertical_left_c(uint8_t *src, uint8_t *topright, int stride){
1828 src[0+0*stride]=(t0 + t1 + 1)>>1;
1830 src[0+2*stride]=(t1 + t2 + 1)>>1;
1832 src[1+2*stride]=(t2 + t3 + 1)>>1;
1834 src[2+2*stride]=(t3 + t4+ 1)>>1;
1835 src[3+2*stride]=(t4 + t5+ 1)>>1;
1836 src[0+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1838 src[0+3*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1840 src[1+3*stride]=(t2 + 2*t3 + t4 + 2)>>2;
1842 src[2+3*stride]=(t3 + 2*t4 + t5 + 2)>>2;
1843 src[3+3*stride]=(t4 + 2*t5 + t6 + 2)>>2;
1846 static void pred4x4_horizontal_up_c(uint8_t *src, uint8_t *topright, int stride){
1849 src[0+0*stride]=(l0 + l1 + 1)>>1;
1850 src[1+0*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1852 src[0+1*stride]=(l1 + l2 + 1)>>1;
1854 src[1+1*stride]=(l1 + 2*l2 + l3 + 2)>>2;
1856 src[0+2*stride]=(l2 + l3 + 1)>>1;
1858 src[1+2*stride]=(l2 + 2*l3 + l3 + 2)>>2;
1867 static void pred4x4_horizontal_down_c(uint8_t *src, uint8_t *topright, int stride){
1868 const int lt= src[-1-1*stride];
1873 src[2+1*stride]=(lt + l0 + 1)>>1;
1875 src[3+1*stride]=(l0 + 2*lt + t0 + 2)>>2;
1876 src[2+0*stride]=(lt + 2*t0 + t1 + 2)>>2;
1877 src[3+0*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1879 src[2+2*stride]=(l0 + l1 + 1)>>1;
1881 src[3+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1883 src[2+3*stride]=(l1 + l2+ 1)>>1;
1885 src[3+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1886 src[0+3*stride]=(l2 + l3 + 1)>>1;
1887 src[1+3*stride]=(l1 + 2*l2 + l3 + 2)>>2;
1890 void ff_pred16x16_vertical_c(uint8_t *src, int stride){
1892 const uint32_t a= ((uint32_t*)(src-stride))[0];
1893 const uint32_t b= ((uint32_t*)(src-stride))[1];
1894 const uint32_t c= ((uint32_t*)(src-stride))[2];
1895 const uint32_t d= ((uint32_t*)(src-stride))[3];
1897 for(i=0; i<16; i++){
1898 ((uint32_t*)(src+i*stride))[0]= a;
1899 ((uint32_t*)(src+i*stride))[1]= b;
1900 ((uint32_t*)(src+i*stride))[2]= c;
1901 ((uint32_t*)(src+i*stride))[3]= d;
1905 void ff_pred16x16_horizontal_c(uint8_t *src, int stride){
1908 for(i=0; i<16; i++){
1909 ((uint32_t*)(src+i*stride))[0]=
1910 ((uint32_t*)(src+i*stride))[1]=
1911 ((uint32_t*)(src+i*stride))[2]=
1912 ((uint32_t*)(src+i*stride))[3]= src[-1+i*stride]*0x01010101;
1916 void ff_pred16x16_dc_c(uint8_t *src, int stride){
1920 dc+= src[-1+i*stride];
1927 dc= 0x01010101*((dc + 16)>>5);
1929 for(i=0; i<16; i++){
1930 ((uint32_t*)(src+i*stride))[0]=
1931 ((uint32_t*)(src+i*stride))[1]=
1932 ((uint32_t*)(src+i*stride))[2]=
1933 ((uint32_t*)(src+i*stride))[3]= dc;
1937 void ff_pred16x16_left_dc_c(uint8_t *src, int stride){
1941 dc+= src[-1+i*stride];
1944 dc= 0x01010101*((dc + 8)>>4);
1946 for(i=0; i<16; i++){
1947 ((uint32_t*)(src+i*stride))[0]=
1948 ((uint32_t*)(src+i*stride))[1]=
1949 ((uint32_t*)(src+i*stride))[2]=
1950 ((uint32_t*)(src+i*stride))[3]= dc;
1954 void ff_pred16x16_top_dc_c(uint8_t *src, int stride){
1960 dc= 0x01010101*((dc + 8)>>4);
1962 for(i=0; i<16; i++){
1963 ((uint32_t*)(src+i*stride))[0]=
1964 ((uint32_t*)(src+i*stride))[1]=
1965 ((uint32_t*)(src+i*stride))[2]=
1966 ((uint32_t*)(src+i*stride))[3]= dc;
1970 void ff_pred16x16_128_dc_c(uint8_t *src, int stride){
1973 for(i=0; i<16; i++){
1974 ((uint32_t*)(src+i*stride))[0]=
1975 ((uint32_t*)(src+i*stride))[1]=
1976 ((uint32_t*)(src+i*stride))[2]=
1977 ((uint32_t*)(src+i*stride))[3]= 0x01010101U*128U;
1981 static inline void pred16x16_plane_compat_c(uint8_t *src, int stride, const int svq3){
1984 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
1985 const uint8_t * const src0 = src+7-stride;
1986 const uint8_t *src1 = src+8*stride-1;
1987 const uint8_t *src2 = src1-2*stride; // == src+6*stride-1;
1988 int H = src0[1] - src0[-1];
1989 int V = src1[0] - src2[ 0];
1990 for(k=2; k<=8; ++k) {
1991 src1 += stride; src2 -= stride;
1992 H += k*(src0[k] - src0[-k]);
1993 V += k*(src1[0] - src2[ 0]);
1996 H = ( 5*(H/4) ) / 16;
1997 V = ( 5*(V/4) ) / 16;
1999 /* required for 100% accuracy */
2000 i = H; H = V; V = i;
2002 H = ( 5*H+32 ) >> 6;
2003 V = ( 5*V+32 ) >> 6;
2006 a = 16*(src1[0] + src2[16] + 1) - 7*(V+H);
2007 for(j=16; j>0; --j) {
2010 for(i=-16; i<0; i+=4) {
2011 src[16+i] = cm[ (b ) >> 5 ];
2012 src[17+i] = cm[ (b+ H) >> 5 ];
2013 src[18+i] = cm[ (b+2*H) >> 5 ];
2014 src[19+i] = cm[ (b+3*H) >> 5 ];
2021 void ff_pred16x16_plane_c(uint8_t *src, int stride){
2022 pred16x16_plane_compat_c(src, stride, 0);
2025 void ff_pred8x8_vertical_c(uint8_t *src, int stride){
2027 const uint32_t a= ((uint32_t*)(src-stride))[0];
2028 const uint32_t b= ((uint32_t*)(src-stride))[1];
2031 ((uint32_t*)(src+i*stride))[0]= a;
2032 ((uint32_t*)(src+i*stride))[1]= b;
2036 void ff_pred8x8_horizontal_c(uint8_t *src, int stride){
2040 ((uint32_t*)(src+i*stride))[0]=
2041 ((uint32_t*)(src+i*stride))[1]= src[-1+i*stride]*0x01010101;
2045 void ff_pred8x8_128_dc_c(uint8_t *src, int stride){
2049 ((uint32_t*)(src+i*stride))[0]=
2050 ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
2054 void ff_pred8x8_left_dc_c(uint8_t *src, int stride){
2060 dc0+= src[-1+i*stride];
2061 dc2+= src[-1+(i+4)*stride];
2063 dc0= 0x01010101*((dc0 + 2)>>2);
2064 dc2= 0x01010101*((dc2 + 2)>>2);
2067 ((uint32_t*)(src+i*stride))[0]=
2068 ((uint32_t*)(src+i*stride))[1]= dc0;
2071 ((uint32_t*)(src+i*stride))[0]=
2072 ((uint32_t*)(src+i*stride))[1]= dc2;
2076 void ff_pred8x8_top_dc_c(uint8_t *src, int stride){
2082 dc0+= src[i-stride];
2083 dc1+= src[4+i-stride];
2085 dc0= 0x01010101*((dc0 + 2)>>2);
2086 dc1= 0x01010101*((dc1 + 2)>>2);
2089 ((uint32_t*)(src+i*stride))[0]= dc0;
2090 ((uint32_t*)(src+i*stride))[1]= dc1;
2093 ((uint32_t*)(src+i*stride))[0]= dc0;
2094 ((uint32_t*)(src+i*stride))[1]= dc1;
2099 void ff_pred8x8_dc_c(uint8_t *src, int stride){
2101 int dc0, dc1, dc2, dc3;
2105 dc0+= src[-1+i*stride] + src[i-stride];
2106 dc1+= src[4+i-stride];
2107 dc2+= src[-1+(i+4)*stride];
2109 dc3= 0x01010101*((dc1 + dc2 + 4)>>3);
2110 dc0= 0x01010101*((dc0 + 4)>>3);
2111 dc1= 0x01010101*((dc1 + 2)>>2);
2112 dc2= 0x01010101*((dc2 + 2)>>2);
2115 ((uint32_t*)(src+i*stride))[0]= dc0;
2116 ((uint32_t*)(src+i*stride))[1]= dc1;
2119 ((uint32_t*)(src+i*stride))[0]= dc2;
2120 ((uint32_t*)(src+i*stride))[1]= dc3;
2124 void ff_pred8x8_plane_c(uint8_t *src, int stride){
2127 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
2128 const uint8_t * const src0 = src+3-stride;
2129 const uint8_t *src1 = src+4*stride-1;
2130 const uint8_t *src2 = src1-2*stride; // == src+2*stride-1;
2131 int H = src0[1] - src0[-1];
2132 int V = src1[0] - src2[ 0];
2133 for(k=2; k<=4; ++k) {
2134 src1 += stride; src2 -= stride;
2135 H += k*(src0[k] - src0[-k]);
2136 V += k*(src1[0] - src2[ 0]);
2138 H = ( 17*H+16 ) >> 5;
2139 V = ( 17*V+16 ) >> 5;
2141 a = 16*(src1[0] + src2[8]+1) - 3*(V+H);
2142 for(j=8; j>0; --j) {
2145 src[0] = cm[ (b ) >> 5 ];
2146 src[1] = cm[ (b+ H) >> 5 ];
2147 src[2] = cm[ (b+2*H) >> 5 ];
2148 src[3] = cm[ (b+3*H) >> 5 ];
2149 src[4] = cm[ (b+4*H) >> 5 ];
2150 src[5] = cm[ (b+5*H) >> 5 ];
2151 src[6] = cm[ (b+6*H) >> 5 ];
2152 src[7] = cm[ (b+7*H) >> 5 ];
2157 #define SRC(x,y) src[(x)+(y)*stride]
2159 const int l##y = (SRC(-1,y-1) + 2*SRC(-1,y) + SRC(-1,y+1) + 2) >> 2;
2160 #define PREDICT_8x8_LOAD_LEFT \
2161 const int l0 = ((has_topleft ? SRC(-1,-1) : SRC(-1,0)) \
2162 + 2*SRC(-1,0) + SRC(-1,1) + 2) >> 2; \
2163 PL(1) PL(2) PL(3) PL(4) PL(5) PL(6) \
2164 const int l7 av_unused = (SRC(-1,6) + 3*SRC(-1,7) + 2) >> 2
2167 const int t##x = (SRC(x-1,-1) + 2*SRC(x,-1) + SRC(x+1,-1) + 2) >> 2;
2168 #define PREDICT_8x8_LOAD_TOP \
2169 const int t0 = ((has_topleft ? SRC(-1,-1) : SRC(0,-1)) \
2170 + 2*SRC(0,-1) + SRC(1,-1) + 2) >> 2; \
2171 PT(1) PT(2) PT(3) PT(4) PT(5) PT(6) \
2172 const int t7 av_unused = ((has_topright ? SRC(8,-1) : SRC(7,-1)) \
2173 + 2*SRC(7,-1) + SRC(6,-1) + 2) >> 2
2176 t##x = (SRC(x-1,-1) + 2*SRC(x,-1) + SRC(x+1,-1) + 2) >> 2;
2177 #define PREDICT_8x8_LOAD_TOPRIGHT \
2178 int t8, t9, t10, t11, t12, t13, t14, t15; \
2179 if(has_topright) { \
2180 PTR(8) PTR(9) PTR(10) PTR(11) PTR(12) PTR(13) PTR(14) \
2181 t15 = (SRC(14,-1) + 3*SRC(15,-1) + 2) >> 2; \
2182 } else t8=t9=t10=t11=t12=t13=t14=t15= SRC(7,-1);
2184 #define PREDICT_8x8_LOAD_TOPLEFT \
2185 const int lt = (SRC(-1,0) + 2*SRC(-1,-1) + SRC(0,-1) + 2) >> 2
2187 #define PREDICT_8x8_DC(v) \
2189 for( y = 0; y < 8; y++ ) { \
2190 ((uint32_t*)src)[0] = \
2191 ((uint32_t*)src)[1] = v; \
2195 static void pred8x8l_128_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2197 PREDICT_8x8_DC(0x80808080);
2199 static void pred8x8l_left_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2201 PREDICT_8x8_LOAD_LEFT;
2202 const uint32_t dc = ((l0+l1+l2+l3+l4+l5+l6+l7+4) >> 3) * 0x01010101;
2205 static void pred8x8l_top_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2207 PREDICT_8x8_LOAD_TOP;
2208 const uint32_t dc = ((t0+t1+t2+t3+t4+t5+t6+t7+4) >> 3) * 0x01010101;
2211 static void pred8x8l_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2213 PREDICT_8x8_LOAD_LEFT;
2214 PREDICT_8x8_LOAD_TOP;
2215 const uint32_t dc = ((l0+l1+l2+l3+l4+l5+l6+l7
2216 +t0+t1+t2+t3+t4+t5+t6+t7+8) >> 4) * 0x01010101;
2219 static void pred8x8l_horizontal_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2221 PREDICT_8x8_LOAD_LEFT;
2222 #define ROW(y) ((uint32_t*)(src+y*stride))[0] =\
2223 ((uint32_t*)(src+y*stride))[1] = 0x01010101 * l##y
2224 ROW(0); ROW(1); ROW(2); ROW(3); ROW(4); ROW(5); ROW(6); ROW(7);
2227 static void pred8x8l_vertical_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2230 PREDICT_8x8_LOAD_TOP;
2239 for( y = 1; y < 8; y++ )
2240 *(uint64_t*)(src+y*stride) = *(uint64_t*)src;
2242 static void pred8x8l_down_left_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2244 PREDICT_8x8_LOAD_TOP;
2245 PREDICT_8x8_LOAD_TOPRIGHT;
2246 SRC(0,0)= (t0 + 2*t1 + t2 + 2) >> 2;
2247 SRC(0,1)=SRC(1,0)= (t1 + 2*t2 + t3 + 2) >> 2;
2248 SRC(0,2)=SRC(1,1)=SRC(2,0)= (t2 + 2*t3 + t4 + 2) >> 2;
2249 SRC(0,3)=SRC(1,2)=SRC(2,1)=SRC(3,0)= (t3 + 2*t4 + t5 + 2) >> 2;
2250 SRC(0,4)=SRC(1,3)=SRC(2,2)=SRC(3,1)=SRC(4,0)= (t4 + 2*t5 + t6 + 2) >> 2;
2251 SRC(0,5)=SRC(1,4)=SRC(2,3)=SRC(3,2)=SRC(4,1)=SRC(5,0)= (t5 + 2*t6 + t7 + 2) >> 2;
2252 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;
2253 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;
2254 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;
2255 SRC(2,7)=SRC(3,6)=SRC(4,5)=SRC(5,4)=SRC(6,3)=SRC(7,2)= (t9 + 2*t10 + t11 + 2) >> 2;
2256 SRC(3,7)=SRC(4,6)=SRC(5,5)=SRC(6,4)=SRC(7,3)= (t10 + 2*t11 + t12 + 2) >> 2;
2257 SRC(4,7)=SRC(5,6)=SRC(6,5)=SRC(7,4)= (t11 + 2*t12 + t13 + 2) >> 2;
2258 SRC(5,7)=SRC(6,6)=SRC(7,5)= (t12 + 2*t13 + t14 + 2) >> 2;
2259 SRC(6,7)=SRC(7,6)= (t13 + 2*t14 + t15 + 2) >> 2;
2260 SRC(7,7)= (t14 + 3*t15 + 2) >> 2;
2262 static void pred8x8l_down_right_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2264 PREDICT_8x8_LOAD_TOP;
2265 PREDICT_8x8_LOAD_LEFT;
2266 PREDICT_8x8_LOAD_TOPLEFT;
2267 SRC(0,7)= (l7 + 2*l6 + l5 + 2) >> 2;
2268 SRC(0,6)=SRC(1,7)= (l6 + 2*l5 + l4 + 2) >> 2;
2269 SRC(0,5)=SRC(1,6)=SRC(2,7)= (l5 + 2*l4 + l3 + 2) >> 2;
2270 SRC(0,4)=SRC(1,5)=SRC(2,6)=SRC(3,7)= (l4 + 2*l3 + l2 + 2) >> 2;
2271 SRC(0,3)=SRC(1,4)=SRC(2,5)=SRC(3,6)=SRC(4,7)= (l3 + 2*l2 + l1 + 2) >> 2;
2272 SRC(0,2)=SRC(1,3)=SRC(2,4)=SRC(3,5)=SRC(4,6)=SRC(5,7)= (l2 + 2*l1 + l0 + 2) >> 2;
2273 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;
2274 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;
2275 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;
2276 SRC(2,0)=SRC(3,1)=SRC(4,2)=SRC(5,3)=SRC(6,4)=SRC(7,5)= (t0 + 2*t1 + t2 + 2) >> 2;
2277 SRC(3,0)=SRC(4,1)=SRC(5,2)=SRC(6,3)=SRC(7,4)= (t1 + 2*t2 + t3 + 2) >> 2;
2278 SRC(4,0)=SRC(5,1)=SRC(6,2)=SRC(7,3)= (t2 + 2*t3 + t4 + 2) >> 2;
2279 SRC(5,0)=SRC(6,1)=SRC(7,2)= (t3 + 2*t4 + t5 + 2) >> 2;
2280 SRC(6,0)=SRC(7,1)= (t4 + 2*t5 + t6 + 2) >> 2;
2281 SRC(7,0)= (t5 + 2*t6 + t7 + 2) >> 2;
2284 static void pred8x8l_vertical_right_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2286 PREDICT_8x8_LOAD_TOP;
2287 PREDICT_8x8_LOAD_LEFT;
2288 PREDICT_8x8_LOAD_TOPLEFT;
2289 SRC(0,6)= (l5 + 2*l4 + l3 + 2) >> 2;
2290 SRC(0,7)= (l6 + 2*l5 + l4 + 2) >> 2;
2291 SRC(0,4)=SRC(1,6)= (l3 + 2*l2 + l1 + 2) >> 2;
2292 SRC(0,5)=SRC(1,7)= (l4 + 2*l3 + l2 + 2) >> 2;
2293 SRC(0,2)=SRC(1,4)=SRC(2,6)= (l1 + 2*l0 + lt + 2) >> 2;
2294 SRC(0,3)=SRC(1,5)=SRC(2,7)= (l2 + 2*l1 + l0 + 2) >> 2;
2295 SRC(0,1)=SRC(1,3)=SRC(2,5)=SRC(3,7)= (l0 + 2*lt + t0 + 2) >> 2;
2296 SRC(0,0)=SRC(1,2)=SRC(2,4)=SRC(3,6)= (lt + t0 + 1) >> 1;
2297 SRC(1,1)=SRC(2,3)=SRC(3,5)=SRC(4,7)= (lt + 2*t0 + t1 + 2) >> 2;
2298 SRC(1,0)=SRC(2,2)=SRC(3,4)=SRC(4,6)= (t0 + t1 + 1) >> 1;
2299 SRC(2,1)=SRC(3,3)=SRC(4,5)=SRC(5,7)= (t0 + 2*t1 + t2 + 2) >> 2;
2300 SRC(2,0)=SRC(3,2)=SRC(4,4)=SRC(5,6)= (t1 + t2 + 1) >> 1;
2301 SRC(3,1)=SRC(4,3)=SRC(5,5)=SRC(6,7)= (t1 + 2*t2 + t3 + 2) >> 2;
2302 SRC(3,0)=SRC(4,2)=SRC(5,4)=SRC(6,6)= (t2 + t3 + 1) >> 1;
2303 SRC(4,1)=SRC(5,3)=SRC(6,5)=SRC(7,7)= (t2 + 2*t3 + t4 + 2) >> 2;
2304 SRC(4,0)=SRC(5,2)=SRC(6,4)=SRC(7,6)= (t3 + t4 + 1) >> 1;
2305 SRC(5,1)=SRC(6,3)=SRC(7,5)= (t3 + 2*t4 + t5 + 2) >> 2;
2306 SRC(5,0)=SRC(6,2)=SRC(7,4)= (t4 + t5 + 1) >> 1;
2307 SRC(6,1)=SRC(7,3)= (t4 + 2*t5 + t6 + 2) >> 2;
2308 SRC(6,0)=SRC(7,2)= (t5 + t6 + 1) >> 1;
2309 SRC(7,1)= (t5 + 2*t6 + t7 + 2) >> 2;
2310 SRC(7,0)= (t6 + t7 + 1) >> 1;
2312 static void pred8x8l_horizontal_down_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2314 PREDICT_8x8_LOAD_TOP;
2315 PREDICT_8x8_LOAD_LEFT;
2316 PREDICT_8x8_LOAD_TOPLEFT;
2317 SRC(0,7)= (l6 + l7 + 1) >> 1;
2318 SRC(1,7)= (l5 + 2*l6 + l7 + 2) >> 2;
2319 SRC(0,6)=SRC(2,7)= (l5 + l6 + 1) >> 1;
2320 SRC(1,6)=SRC(3,7)= (l4 + 2*l5 + l6 + 2) >> 2;
2321 SRC(0,5)=SRC(2,6)=SRC(4,7)= (l4 + l5 + 1) >> 1;
2322 SRC(1,5)=SRC(3,6)=SRC(5,7)= (l3 + 2*l4 + l5 + 2) >> 2;
2323 SRC(0,4)=SRC(2,5)=SRC(4,6)=SRC(6,7)= (l3 + l4 + 1) >> 1;
2324 SRC(1,4)=SRC(3,5)=SRC(5,6)=SRC(7,7)= (l2 + 2*l3 + l4 + 2) >> 2;
2325 SRC(0,3)=SRC(2,4)=SRC(4,5)=SRC(6,6)= (l2 + l3 + 1) >> 1;
2326 SRC(1,3)=SRC(3,4)=SRC(5,5)=SRC(7,6)= (l1 + 2*l2 + l3 + 2) >> 2;
2327 SRC(0,2)=SRC(2,3)=SRC(4,4)=SRC(6,5)= (l1 + l2 + 1) >> 1;
2328 SRC(1,2)=SRC(3,3)=SRC(5,4)=SRC(7,5)= (l0 + 2*l1 + l2 + 2) >> 2;
2329 SRC(0,1)=SRC(2,2)=SRC(4,3)=SRC(6,4)= (l0 + l1 + 1) >> 1;
2330 SRC(1,1)=SRC(3,2)=SRC(5,3)=SRC(7,4)= (lt + 2*l0 + l1 + 2) >> 2;
2331 SRC(0,0)=SRC(2,1)=SRC(4,2)=SRC(6,3)= (lt + l0 + 1) >> 1;
2332 SRC(1,0)=SRC(3,1)=SRC(5,2)=SRC(7,3)= (l0 + 2*lt + t0 + 2) >> 2;
2333 SRC(2,0)=SRC(4,1)=SRC(6,2)= (t1 + 2*t0 + lt + 2) >> 2;
2334 SRC(3,0)=SRC(5,1)=SRC(7,2)= (t2 + 2*t1 + t0 + 2) >> 2;
2335 SRC(4,0)=SRC(6,1)= (t3 + 2*t2 + t1 + 2) >> 2;
2336 SRC(5,0)=SRC(7,1)= (t4 + 2*t3 + t2 + 2) >> 2;
2337 SRC(6,0)= (t5 + 2*t4 + t3 + 2) >> 2;
2338 SRC(7,0)= (t6 + 2*t5 + t4 + 2) >> 2;
2340 static void pred8x8l_vertical_left_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2342 PREDICT_8x8_LOAD_TOP;
2343 PREDICT_8x8_LOAD_TOPRIGHT;
2344 SRC(0,0)= (t0 + t1 + 1) >> 1;
2345 SRC(0,1)= (t0 + 2*t1 + t2 + 2) >> 2;
2346 SRC(0,2)=SRC(1,0)= (t1 + t2 + 1) >> 1;
2347 SRC(0,3)=SRC(1,1)= (t1 + 2*t2 + t3 + 2) >> 2;
2348 SRC(0,4)=SRC(1,2)=SRC(2,0)= (t2 + t3 + 1) >> 1;
2349 SRC(0,5)=SRC(1,3)=SRC(2,1)= (t2 + 2*t3 + t4 + 2) >> 2;
2350 SRC(0,6)=SRC(1,4)=SRC(2,2)=SRC(3,0)= (t3 + t4 + 1) >> 1;
2351 SRC(0,7)=SRC(1,5)=SRC(2,3)=SRC(3,1)= (t3 + 2*t4 + t5 + 2) >> 2;
2352 SRC(1,6)=SRC(2,4)=SRC(3,2)=SRC(4,0)= (t4 + t5 + 1) >> 1;
2353 SRC(1,7)=SRC(2,5)=SRC(3,3)=SRC(4,1)= (t4 + 2*t5 + t6 + 2) >> 2;
2354 SRC(2,6)=SRC(3,4)=SRC(4,2)=SRC(5,0)= (t5 + t6 + 1) >> 1;
2355 SRC(2,7)=SRC(3,5)=SRC(4,3)=SRC(5,1)= (t5 + 2*t6 + t7 + 2) >> 2;
2356 SRC(3,6)=SRC(4,4)=SRC(5,2)=SRC(6,0)= (t6 + t7 + 1) >> 1;
2357 SRC(3,7)=SRC(4,5)=SRC(5,3)=SRC(6,1)= (t6 + 2*t7 + t8 + 2) >> 2;
2358 SRC(4,6)=SRC(5,4)=SRC(6,2)=SRC(7,0)= (t7 + t8 + 1) >> 1;
2359 SRC(4,7)=SRC(5,5)=SRC(6,3)=SRC(7,1)= (t7 + 2*t8 + t9 + 2) >> 2;
2360 SRC(5,6)=SRC(6,4)=SRC(7,2)= (t8 + t9 + 1) >> 1;
2361 SRC(5,7)=SRC(6,5)=SRC(7,3)= (t8 + 2*t9 + t10 + 2) >> 2;
2362 SRC(6,6)=SRC(7,4)= (t9 + t10 + 1) >> 1;
2363 SRC(6,7)=SRC(7,5)= (t9 + 2*t10 + t11 + 2) >> 2;
2364 SRC(7,6)= (t10 + t11 + 1) >> 1;
2365 SRC(7,7)= (t10 + 2*t11 + t12 + 2) >> 2;
2367 static void pred8x8l_horizontal_up_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2369 PREDICT_8x8_LOAD_LEFT;
2370 SRC(0,0)= (l0 + l1 + 1) >> 1;
2371 SRC(1,0)= (l0 + 2*l1 + l2 + 2) >> 2;
2372 SRC(0,1)=SRC(2,0)= (l1 + l2 + 1) >> 1;
2373 SRC(1,1)=SRC(3,0)= (l1 + 2*l2 + l3 + 2) >> 2;
2374 SRC(0,2)=SRC(2,1)=SRC(4,0)= (l2 + l3 + 1) >> 1;
2375 SRC(1,2)=SRC(3,1)=SRC(5,0)= (l2 + 2*l3 + l4 + 2) >> 2;
2376 SRC(0,3)=SRC(2,2)=SRC(4,1)=SRC(6,0)= (l3 + l4 + 1) >> 1;
2377 SRC(1,3)=SRC(3,2)=SRC(5,1)=SRC(7,0)= (l3 + 2*l4 + l5 + 2) >> 2;
2378 SRC(0,4)=SRC(2,3)=SRC(4,2)=SRC(6,1)= (l4 + l5 + 1) >> 1;
2379 SRC(1,4)=SRC(3,3)=SRC(5,2)=SRC(7,1)= (l4 + 2*l5 + l6 + 2) >> 2;
2380 SRC(0,5)=SRC(2,4)=SRC(4,3)=SRC(6,2)= (l5 + l6 + 1) >> 1;
2381 SRC(1,5)=SRC(3,4)=SRC(5,3)=SRC(7,2)= (l5 + 2*l6 + l7 + 2) >> 2;
2382 SRC(0,6)=SRC(2,5)=SRC(4,4)=SRC(6,3)= (l6 + l7 + 1) >> 1;
2383 SRC(1,6)=SRC(3,5)=SRC(5,4)=SRC(7,3)= (l6 + 3*l7 + 2) >> 2;
2384 SRC(0,7)=SRC(1,7)=SRC(2,6)=SRC(2,7)=SRC(3,6)=
2385 SRC(3,7)=SRC(4,5)=SRC(4,6)=SRC(4,7)=SRC(5,5)=
2386 SRC(5,6)=SRC(5,7)=SRC(6,4)=SRC(6,5)=SRC(6,6)=
2387 SRC(6,7)=SRC(7,4)=SRC(7,5)=SRC(7,6)=SRC(7,7)= l7;
2389 #undef PREDICT_8x8_LOAD_LEFT
2390 #undef PREDICT_8x8_LOAD_TOP
2391 #undef PREDICT_8x8_LOAD_TOPLEFT
2392 #undef PREDICT_8x8_LOAD_TOPRIGHT
2393 #undef PREDICT_8x8_DC
2399 static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
2400 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2401 int src_x_offset, int src_y_offset,
2402 qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
2403 MpegEncContext * const s = &h->s;
2404 const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
2405 int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
2406 const int luma_xy= (mx&3) + ((my&3)<<2);
2407 uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*h->mb_linesize;
2408 uint8_t * src_cb, * src_cr;
2409 int extra_width= h->emu_edge_width;
2410 int extra_height= h->emu_edge_height;
2412 const int full_mx= mx>>2;
2413 const int full_my= my>>2;
2414 const int pic_width = 16*s->mb_width;
2415 const int pic_height = 16*s->mb_height >> MB_MBAFF;
2417 if(!pic->data[0]) //FIXME this is unacceptable, some senseable error concealment must be done for missing reference frames
2420 if(mx&7) extra_width -= 3;
2421 if(my&7) extra_height -= 3;
2423 if( full_mx < 0-extra_width
2424 || full_my < 0-extra_height
2425 || full_mx + 16/*FIXME*/ > pic_width + extra_width
2426 || full_my + 16/*FIXME*/ > pic_height + extra_height){
2427 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);
2428 src_y= s->edge_emu_buffer + 2 + 2*h->mb_linesize;
2432 qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); //FIXME try variable height perhaps?
2434 qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
2437 if(s->flags&CODEC_FLAG_GRAY) return;
2440 // chroma offset when predicting from a field of opposite parity
2441 my += 2 * ((s->mb_y & 1) - (h->ref_cache[list][scan8[n]] & 1));
2442 emu |= (my>>3) < 0 || (my>>3) + 8 >= (pic_height>>1);
2444 src_cb= pic->data[1] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
2445 src_cr= pic->data[2] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
2448 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);
2449 src_cb= s->edge_emu_buffer;
2451 chroma_op(dest_cb, src_cb, h->mb_uvlinesize, chroma_height, mx&7, my&7);
2454 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);
2455 src_cr= s->edge_emu_buffer;
2457 chroma_op(dest_cr, src_cr, h->mb_uvlinesize, chroma_height, mx&7, my&7);
2460 static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
2461 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2462 int x_offset, int y_offset,
2463 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2464 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
2465 int list0, int list1){
2466 MpegEncContext * const s = &h->s;
2467 qpel_mc_func *qpix_op= qpix_put;
2468 h264_chroma_mc_func chroma_op= chroma_put;
2470 dest_y += 2*x_offset + 2*y_offset*h-> mb_linesize;
2471 dest_cb += x_offset + y_offset*h->mb_uvlinesize;
2472 dest_cr += x_offset + y_offset*h->mb_uvlinesize;
2473 x_offset += 8*s->mb_x;
2474 y_offset += 8*(s->mb_y >> MB_MBAFF);
2477 Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
2478 mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
2479 dest_y, dest_cb, dest_cr, x_offset, y_offset,
2480 qpix_op, chroma_op);
2483 chroma_op= chroma_avg;
2487 Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
2488 mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
2489 dest_y, dest_cb, dest_cr, x_offset, y_offset,
2490 qpix_op, chroma_op);
2494 static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
2495 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2496 int x_offset, int y_offset,
2497 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2498 h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
2499 h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
2500 int list0, int list1){
2501 MpegEncContext * const s = &h->s;
2503 dest_y += 2*x_offset + 2*y_offset*h-> mb_linesize;
2504 dest_cb += x_offset + y_offset*h->mb_uvlinesize;
2505 dest_cr += x_offset + y_offset*h->mb_uvlinesize;
2506 x_offset += 8*s->mb_x;
2507 y_offset += 8*(s->mb_y >> MB_MBAFF);
2510 /* don't optimize for luma-only case, since B-frames usually
2511 * use implicit weights => chroma too. */
2512 uint8_t *tmp_cb = s->obmc_scratchpad;
2513 uint8_t *tmp_cr = s->obmc_scratchpad + 8;
2514 uint8_t *tmp_y = s->obmc_scratchpad + 8*h->mb_uvlinesize;
2515 int refn0 = h->ref_cache[0][ scan8[n] ];
2516 int refn1 = h->ref_cache[1][ scan8[n] ];
2518 mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
2519 dest_y, dest_cb, dest_cr,
2520 x_offset, y_offset, qpix_put, chroma_put);
2521 mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
2522 tmp_y, tmp_cb, tmp_cr,
2523 x_offset, y_offset, qpix_put, chroma_put);
2525 if(h->use_weight == 2){
2526 int weight0 = h->implicit_weight[refn0][refn1];
2527 int weight1 = 64 - weight0;
2528 luma_weight_avg( dest_y, tmp_y, h-> mb_linesize, 5, weight0, weight1, 0);
2529 chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, 5, weight0, weight1, 0);
2530 chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, 5, weight0, weight1, 0);
2532 luma_weight_avg(dest_y, tmp_y, h->mb_linesize, h->luma_log2_weight_denom,
2533 h->luma_weight[0][refn0], h->luma_weight[1][refn1],
2534 h->luma_offset[0][refn0] + h->luma_offset[1][refn1]);
2535 chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
2536 h->chroma_weight[0][refn0][0], h->chroma_weight[1][refn1][0],
2537 h->chroma_offset[0][refn0][0] + h->chroma_offset[1][refn1][0]);
2538 chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
2539 h->chroma_weight[0][refn0][1], h->chroma_weight[1][refn1][1],
2540 h->chroma_offset[0][refn0][1] + h->chroma_offset[1][refn1][1]);
2543 int list = list1 ? 1 : 0;
2544 int refn = h->ref_cache[list][ scan8[n] ];
2545 Picture *ref= &h->ref_list[list][refn];
2546 mc_dir_part(h, ref, n, square, chroma_height, delta, list,
2547 dest_y, dest_cb, dest_cr, x_offset, y_offset,
2548 qpix_put, chroma_put);
2550 luma_weight_op(dest_y, h->mb_linesize, h->luma_log2_weight_denom,
2551 h->luma_weight[list][refn], h->luma_offset[list][refn]);
2552 if(h->use_weight_chroma){
2553 chroma_weight_op(dest_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
2554 h->chroma_weight[list][refn][0], h->chroma_offset[list][refn][0]);
2555 chroma_weight_op(dest_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
2556 h->chroma_weight[list][refn][1], h->chroma_offset[list][refn][1]);
2561 static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
2562 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2563 int x_offset, int y_offset,
2564 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2565 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
2566 h264_weight_func *weight_op, h264_biweight_func *weight_avg,
2567 int list0, int list1){
2568 if((h->use_weight==2 && list0 && list1
2569 && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ] != 32))
2570 || h->use_weight==1)
2571 mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
2572 x_offset, y_offset, qpix_put, chroma_put,
2573 weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
2575 mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
2576 x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
2579 static inline void prefetch_motion(H264Context *h, int list){
2580 /* fetch pixels for estimated mv 4 macroblocks ahead
2581 * optimized for 64byte cache lines */
2582 MpegEncContext * const s = &h->s;
2583 const int refn = h->ref_cache[list][scan8[0]];
2585 const int mx= (h->mv_cache[list][scan8[0]][0]>>2) + 16*s->mb_x + 8;
2586 const int my= (h->mv_cache[list][scan8[0]][1]>>2) + 16*s->mb_y;
2587 uint8_t **src= h->ref_list[list][refn].data;
2588 int off= mx + (my + (s->mb_x&3)*4)*h->mb_linesize + 64;
2589 s->dsp.prefetch(src[0]+off, s->linesize, 4);
2590 off= (mx>>1) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + 64;
2591 s->dsp.prefetch(src[1]+off, src[2]-src[1], 2);
2595 static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2596 qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
2597 qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
2598 h264_weight_func *weight_op, h264_biweight_func *weight_avg){
2599 MpegEncContext * const s = &h->s;
2600 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
2601 const int mb_type= s->current_picture.mb_type[mb_xy];
2603 assert(IS_INTER(mb_type));
2605 prefetch_motion(h, 0);
2607 if(IS_16X16(mb_type)){
2608 mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
2609 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
2610 &weight_op[0], &weight_avg[0],
2611 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2612 }else if(IS_16X8(mb_type)){
2613 mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
2614 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
2615 &weight_op[1], &weight_avg[1],
2616 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2617 mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
2618 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
2619 &weight_op[1], &weight_avg[1],
2620 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2621 }else if(IS_8X16(mb_type)){
2622 mc_part(h, 0, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 0, 0,
2623 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2624 &weight_op[2], &weight_avg[2],
2625 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2626 mc_part(h, 4, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 4, 0,
2627 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2628 &weight_op[2], &weight_avg[2],
2629 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2633 assert(IS_8X8(mb_type));
2636 const int sub_mb_type= h->sub_mb_type[i];
2638 int x_offset= (i&1)<<2;
2639 int y_offset= (i&2)<<1;
2641 if(IS_SUB_8X8(sub_mb_type)){
2642 mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2643 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2644 &weight_op[3], &weight_avg[3],
2645 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2646 }else if(IS_SUB_8X4(sub_mb_type)){
2647 mc_part(h, n , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2648 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2649 &weight_op[4], &weight_avg[4],
2650 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2651 mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
2652 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2653 &weight_op[4], &weight_avg[4],
2654 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2655 }else if(IS_SUB_4X8(sub_mb_type)){
2656 mc_part(h, n , 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2657 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2658 &weight_op[5], &weight_avg[5],
2659 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2660 mc_part(h, n+1, 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
2661 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2662 &weight_op[5], &weight_avg[5],
2663 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2666 assert(IS_SUB_4X4(sub_mb_type));
2668 int sub_x_offset= x_offset + 2*(j&1);
2669 int sub_y_offset= y_offset + (j&2);
2670 mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
2671 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2672 &weight_op[6], &weight_avg[6],
2673 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2679 prefetch_motion(h, 1);
2682 static void decode_init_vlc(void){
2683 static int done = 0;
2689 init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5,
2690 &chroma_dc_coeff_token_len [0], 1, 1,
2691 &chroma_dc_coeff_token_bits[0], 1, 1, 1);
2694 init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17,
2695 &coeff_token_len [i][0], 1, 1,
2696 &coeff_token_bits[i][0], 1, 1, 1);
2700 init_vlc(&chroma_dc_total_zeros_vlc[i], CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
2701 &chroma_dc_total_zeros_len [i][0], 1, 1,
2702 &chroma_dc_total_zeros_bits[i][0], 1, 1, 1);
2704 for(i=0; i<15; i++){
2705 init_vlc(&total_zeros_vlc[i], TOTAL_ZEROS_VLC_BITS, 16,
2706 &total_zeros_len [i][0], 1, 1,
2707 &total_zeros_bits[i][0], 1, 1, 1);
2711 init_vlc(&run_vlc[i], RUN_VLC_BITS, 7,
2712 &run_len [i][0], 1, 1,
2713 &run_bits[i][0], 1, 1, 1);
2715 init_vlc(&run7_vlc, RUN7_VLC_BITS, 16,
2716 &run_len [6][0], 1, 1,
2717 &run_bits[6][0], 1, 1, 1);
2722 * Sets the intra prediction function pointers.
2724 static void init_pred_ptrs(H264Context *h){
2725 // MpegEncContext * const s = &h->s;
2727 h->pred4x4[VERT_PRED ]= pred4x4_vertical_c;
2728 h->pred4x4[HOR_PRED ]= pred4x4_horizontal_c;
2729 h->pred4x4[DC_PRED ]= pred4x4_dc_c;
2730 h->pred4x4[DIAG_DOWN_LEFT_PRED ]= pred4x4_down_left_c;
2731 h->pred4x4[DIAG_DOWN_RIGHT_PRED]= pred4x4_down_right_c;
2732 h->pred4x4[VERT_RIGHT_PRED ]= pred4x4_vertical_right_c;
2733 h->pred4x4[HOR_DOWN_PRED ]= pred4x4_horizontal_down_c;
2734 h->pred4x4[VERT_LEFT_PRED ]= pred4x4_vertical_left_c;
2735 h->pred4x4[HOR_UP_PRED ]= pred4x4_horizontal_up_c;
2736 h->pred4x4[LEFT_DC_PRED ]= pred4x4_left_dc_c;
2737 h->pred4x4[TOP_DC_PRED ]= pred4x4_top_dc_c;
2738 h->pred4x4[DC_128_PRED ]= pred4x4_128_dc_c;
2740 h->pred8x8l[VERT_PRED ]= pred8x8l_vertical_c;
2741 h->pred8x8l[HOR_PRED ]= pred8x8l_horizontal_c;
2742 h->pred8x8l[DC_PRED ]= pred8x8l_dc_c;
2743 h->pred8x8l[DIAG_DOWN_LEFT_PRED ]= pred8x8l_down_left_c;
2744 h->pred8x8l[DIAG_DOWN_RIGHT_PRED]= pred8x8l_down_right_c;
2745 h->pred8x8l[VERT_RIGHT_PRED ]= pred8x8l_vertical_right_c;
2746 h->pred8x8l[HOR_DOWN_PRED ]= pred8x8l_horizontal_down_c;
2747 h->pred8x8l[VERT_LEFT_PRED ]= pred8x8l_vertical_left_c;
2748 h->pred8x8l[HOR_UP_PRED ]= pred8x8l_horizontal_up_c;
2749 h->pred8x8l[LEFT_DC_PRED ]= pred8x8l_left_dc_c;
2750 h->pred8x8l[TOP_DC_PRED ]= pred8x8l_top_dc_c;
2751 h->pred8x8l[DC_128_PRED ]= pred8x8l_128_dc_c;
2753 h->pred8x8[DC_PRED8x8 ]= ff_pred8x8_dc_c;
2754 h->pred8x8[VERT_PRED8x8 ]= ff_pred8x8_vertical_c;
2755 h->pred8x8[HOR_PRED8x8 ]= ff_pred8x8_horizontal_c;
2756 h->pred8x8[PLANE_PRED8x8 ]= ff_pred8x8_plane_c;
2757 h->pred8x8[LEFT_DC_PRED8x8]= ff_pred8x8_left_dc_c;
2758 h->pred8x8[TOP_DC_PRED8x8 ]= ff_pred8x8_top_dc_c;
2759 h->pred8x8[DC_128_PRED8x8 ]= ff_pred8x8_128_dc_c;
2761 h->pred16x16[DC_PRED8x8 ]= ff_pred16x16_dc_c;
2762 h->pred16x16[VERT_PRED8x8 ]= ff_pred16x16_vertical_c;
2763 h->pred16x16[HOR_PRED8x8 ]= ff_pred16x16_horizontal_c;
2764 h->pred16x16[PLANE_PRED8x8 ]= ff_pred16x16_plane_c;
2765 h->pred16x16[LEFT_DC_PRED8x8]= ff_pred16x16_left_dc_c;
2766 h->pred16x16[TOP_DC_PRED8x8 ]= ff_pred16x16_top_dc_c;
2767 h->pred16x16[DC_128_PRED8x8 ]= ff_pred16x16_128_dc_c;
2770 static void free_tables(H264Context *h){
2772 av_freep(&h->intra4x4_pred_mode);
2773 av_freep(&h->chroma_pred_mode_table);
2774 av_freep(&h->cbp_table);
2775 av_freep(&h->mvd_table[0]);
2776 av_freep(&h->mvd_table[1]);
2777 av_freep(&h->direct_table);
2778 av_freep(&h->non_zero_count);
2779 av_freep(&h->slice_table_base);
2780 av_freep(&h->top_borders[1]);
2781 av_freep(&h->top_borders[0]);
2782 h->slice_table= NULL;
2784 av_freep(&h->mb2b_xy);
2785 av_freep(&h->mb2b8_xy);
2787 av_freep(&h->s.obmc_scratchpad);
2789 for(i = 0; i < MAX_SPS_COUNT; i++)
2790 av_freep(h->sps_buffers + i);
2792 for(i = 0; i < MAX_PPS_COUNT; i++)
2793 av_freep(h->pps_buffers + i);
2796 static void init_dequant8_coeff_table(H264Context *h){
2798 const int transpose = (h->s.dsp.h264_idct8_add != ff_h264_idct8_add_c); //FIXME ugly
2799 h->dequant8_coeff[0] = h->dequant8_buffer[0];
2800 h->dequant8_coeff[1] = h->dequant8_buffer[1];
2802 for(i=0; i<2; i++ ){
2803 if(i && !memcmp(h->pps.scaling_matrix8[0], h->pps.scaling_matrix8[1], 64*sizeof(uint8_t))){
2804 h->dequant8_coeff[1] = h->dequant8_buffer[0];
2808 for(q=0; q<52; q++){
2809 int shift = ff_div6[q];
2810 int idx = ff_rem6[q];
2812 h->dequant8_coeff[i][q][transpose ? (x>>3)|((x&7)<<3) : x] =
2813 ((uint32_t)dequant8_coeff_init[idx][ dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] *
2814 h->pps.scaling_matrix8[i][x]) << shift;
2819 static void init_dequant4_coeff_table(H264Context *h){
2821 const int transpose = (h->s.dsp.h264_idct_add != ff_h264_idct_add_c); //FIXME ugly
2822 for(i=0; i<6; i++ ){
2823 h->dequant4_coeff[i] = h->dequant4_buffer[i];
2825 if(!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i], 16*sizeof(uint8_t))){
2826 h->dequant4_coeff[i] = h->dequant4_buffer[j];
2833 for(q=0; q<52; q++){
2834 int shift = ff_div6[q] + 2;
2835 int idx = ff_rem6[q];
2837 h->dequant4_coeff[i][q][transpose ? (x>>2)|((x<<2)&0xF) : x] =
2838 ((uint32_t)dequant4_coeff_init[idx][(x&1) + ((x>>2)&1)] *
2839 h->pps.scaling_matrix4[i][x]) << shift;
2844 static void init_dequant_tables(H264Context *h){
2846 init_dequant4_coeff_table(h);
2847 if(h->pps.transform_8x8_mode)
2848 init_dequant8_coeff_table(h);
2849 if(h->sps.transform_bypass){
2852 h->dequant4_coeff[i][0][x] = 1<<6;
2853 if(h->pps.transform_8x8_mode)
2856 h->dequant8_coeff[i][0][x] = 1<<6;
2863 * needs width/height
2865 static int alloc_tables(H264Context *h){
2866 MpegEncContext * const s = &h->s;
2867 const int big_mb_num= s->mb_stride * (s->mb_height+1);
2870 CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8 * sizeof(uint8_t))
2872 CHECKED_ALLOCZ(h->non_zero_count , big_mb_num * 16 * sizeof(uint8_t))
2873 CHECKED_ALLOCZ(h->slice_table_base , (big_mb_num+s->mb_stride) * sizeof(uint8_t))
2874 CHECKED_ALLOCZ(h->top_borders[0] , s->mb_width * (16+8+8) * sizeof(uint8_t))
2875 CHECKED_ALLOCZ(h->top_borders[1] , s->mb_width * (16+8+8) * sizeof(uint8_t))
2876 CHECKED_ALLOCZ(h->cbp_table, big_mb_num * sizeof(uint16_t))
2878 if( h->pps.cabac ) {
2879 CHECKED_ALLOCZ(h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t))
2880 CHECKED_ALLOCZ(h->mvd_table[0], 32*big_mb_num * sizeof(uint16_t));
2881 CHECKED_ALLOCZ(h->mvd_table[1], 32*big_mb_num * sizeof(uint16_t));
2882 CHECKED_ALLOCZ(h->direct_table, 32*big_mb_num * sizeof(uint8_t));
2885 memset(h->slice_table_base, -1, (big_mb_num+s->mb_stride) * sizeof(uint8_t));
2886 h->slice_table= h->slice_table_base + s->mb_stride*2 + 1;
2888 CHECKED_ALLOCZ(h->mb2b_xy , big_mb_num * sizeof(uint32_t));
2889 CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint32_t));
2890 for(y=0; y<s->mb_height; y++){
2891 for(x=0; x<s->mb_width; x++){
2892 const int mb_xy= x + y*s->mb_stride;
2893 const int b_xy = 4*x + 4*y*h->b_stride;
2894 const int b8_xy= 2*x + 2*y*h->b8_stride;
2896 h->mb2b_xy [mb_xy]= b_xy;
2897 h->mb2b8_xy[mb_xy]= b8_xy;
2901 s->obmc_scratchpad = NULL;
2903 if(!h->dequant4_coeff[0])
2904 init_dequant_tables(h);
2912 static void common_init(H264Context *h){
2913 MpegEncContext * const s = &h->s;
2915 s->width = s->avctx->width;
2916 s->height = s->avctx->height;
2917 s->codec_id= s->avctx->codec->id;
2921 h->dequant_coeff_pps= -1;
2922 s->unrestricted_mv=1;
2923 s->decode=1; //FIXME
2925 memset(h->pps.scaling_matrix4, 16, 6*16*sizeof(uint8_t));
2926 memset(h->pps.scaling_matrix8, 16, 2*64*sizeof(uint8_t));
2929 static int decode_init(AVCodecContext *avctx){
2930 H264Context *h= avctx->priv_data;
2931 MpegEncContext * const s = &h->s;
2933 MPV_decode_defaults(s);
2938 s->out_format = FMT_H264;
2939 s->workaround_bugs= avctx->workaround_bugs;
2942 // s->decode_mb= ff_h263_decode_mb;
2944 avctx->pix_fmt= PIX_FMT_YUV420P;
2948 if(avctx->extradata_size > 0 && avctx->extradata &&
2949 *(char *)avctx->extradata == 1){
2959 static int frame_start(H264Context *h){
2960 MpegEncContext * const s = &h->s;
2963 if(MPV_frame_start(s, s->avctx) < 0)
2965 ff_er_frame_start(s);
2967 assert(s->linesize && s->uvlinesize);
2969 for(i=0; i<16; i++){
2970 h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
2971 h->block_offset[24+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
2974 h->block_offset[16+i]=
2975 h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2976 h->block_offset[24+16+i]=
2977 h->block_offset[24+20+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2980 /* can't be in alloc_tables because linesize isn't known there.
2981 * FIXME: redo bipred weight to not require extra buffer? */
2982 if(!s->obmc_scratchpad)
2983 s->obmc_scratchpad = av_malloc(16*2*s->linesize + 8*2*s->uvlinesize);
2985 /* some macroblocks will be accessed before they're available */
2987 memset(h->slice_table, -1, (s->mb_height*s->mb_stride-1) * sizeof(uint8_t));
2989 // s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
2993 static inline void backup_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize, int simple){
2994 MpegEncContext * const s = &h->s;
2998 src_cb -= uvlinesize;
2999 src_cr -= uvlinesize;
3001 // There are two lines saved, the line above the the top macroblock of a pair,
3002 // and the line above the bottom macroblock
3003 h->left_border[0]= h->top_borders[0][s->mb_x][15];
3004 for(i=1; i<17; i++){
3005 h->left_border[i]= src_y[15+i* linesize];
3008 *(uint64_t*)(h->top_borders[0][s->mb_x]+0)= *(uint64_t*)(src_y + 16*linesize);
3009 *(uint64_t*)(h->top_borders[0][s->mb_x]+8)= *(uint64_t*)(src_y +8+16*linesize);
3011 if(simple || !(s->flags&CODEC_FLAG_GRAY)){
3012 h->left_border[17 ]= h->top_borders[0][s->mb_x][16+7];
3013 h->left_border[17+9]= h->top_borders[0][s->mb_x][24+7];
3015 h->left_border[i+17 ]= src_cb[7+i*uvlinesize];
3016 h->left_border[i+17+9]= src_cr[7+i*uvlinesize];
3018 *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+8*uvlinesize);
3019 *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+8*uvlinesize);
3023 static inline void xchg_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize, int xchg, int simple){
3024 MpegEncContext * const s = &h->s;
3031 if(h->deblocking_filter == 2) {
3032 mb_xy = s->mb_x + s->mb_y*s->mb_stride;
3033 deblock_left = h->slice_table[mb_xy] == h->slice_table[mb_xy - 1];
3034 deblock_top = h->slice_table[mb_xy] == h->slice_table[h->top_mb_xy];
3036 deblock_left = (s->mb_x > 0);
3037 deblock_top = (s->mb_y > 0);
3040 src_y -= linesize + 1;
3041 src_cb -= uvlinesize + 1;
3042 src_cr -= uvlinesize + 1;
3044 #define XCHG(a,b,t,xchg)\
3051 for(i = !deblock_top; i<17; i++){
3052 XCHG(h->left_border[i ], src_y [i* linesize], temp8, xchg);
3057 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
3058 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
3059 if(s->mb_x+1 < s->mb_width){
3060 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x+1]), *(uint64_t*)(src_y +17), temp64, 1);
3064 if(simple || !(s->flags&CODEC_FLAG_GRAY)){
3066 for(i = !deblock_top; i<9; i++){
3067 XCHG(h->left_border[i+17 ], src_cb[i*uvlinesize], temp8, xchg);
3068 XCHG(h->left_border[i+17+9], src_cr[i*uvlinesize], temp8, xchg);
3072 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
3073 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
3078 static inline void backup_pair_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){
3079 MpegEncContext * const s = &h->s;
3082 src_y -= 2 * linesize;
3083 src_cb -= 2 * uvlinesize;
3084 src_cr -= 2 * uvlinesize;
3086 // There are two lines saved, the line above the the top macroblock of a pair,
3087 // and the line above the bottom macroblock
3088 h->left_border[0]= h->top_borders[0][s->mb_x][15];
3089 h->left_border[1]= h->top_borders[1][s->mb_x][15];
3090 for(i=2; i<34; i++){
3091 h->left_border[i]= src_y[15+i* linesize];
3094 *(uint64_t*)(h->top_borders[0][s->mb_x]+0)= *(uint64_t*)(src_y + 32*linesize);
3095 *(uint64_t*)(h->top_borders[0][s->mb_x]+8)= *(uint64_t*)(src_y +8+32*linesize);
3096 *(uint64_t*)(h->top_borders[1][s->mb_x]+0)= *(uint64_t*)(src_y + 33*linesize);
3097 *(uint64_t*)(h->top_borders[1][s->mb_x]+8)= *(uint64_t*)(src_y +8+33*linesize);
3099 if(!(s->flags&CODEC_FLAG_GRAY)){
3100 h->left_border[34 ]= h->top_borders[0][s->mb_x][16+7];
3101 h->left_border[34+ 1]= h->top_borders[1][s->mb_x][16+7];
3102 h->left_border[34+18 ]= h->top_borders[0][s->mb_x][24+7];
3103 h->left_border[34+18+1]= h->top_borders[1][s->mb_x][24+7];
3104 for(i=2; i<18; i++){
3105 h->left_border[i+34 ]= src_cb[7+i*uvlinesize];
3106 h->left_border[i+34+18]= src_cr[7+i*uvlinesize];
3108 *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+16*uvlinesize);
3109 *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+16*uvlinesize);
3110 *(uint64_t*)(h->top_borders[1][s->mb_x]+16)= *(uint64_t*)(src_cb+17*uvlinesize);
3111 *(uint64_t*)(h->top_borders[1][s->mb_x]+24)= *(uint64_t*)(src_cr+17*uvlinesize);
3115 static inline void xchg_pair_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize, int xchg){
3116 MpegEncContext * const s = &h->s;
3119 int deblock_left = (s->mb_x > 0);
3120 int deblock_top = (s->mb_y > 1);
3122 tprintf(s->avctx, "xchg_pair_border: src_y:%p src_cb:%p src_cr:%p ls:%d uvls:%d\n", src_y, src_cb, src_cr, linesize, uvlinesize);
3124 src_y -= 2 * linesize + 1;
3125 src_cb -= 2 * uvlinesize + 1;
3126 src_cr -= 2 * uvlinesize + 1;
3128 #define XCHG(a,b,t,xchg)\
3135 for(i = (!deblock_top)<<1; i<34; i++){
3136 XCHG(h->left_border[i ], src_y [i* linesize], temp8, xchg);
3141 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
3142 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
3143 XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+0), *(uint64_t*)(src_y +1 +linesize), temp64, xchg);
3144 XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+8), *(uint64_t*)(src_y +9 +linesize), temp64, 1);
3145 if(s->mb_x+1 < s->mb_width){
3146 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x+1]), *(uint64_t*)(src_y +17), temp64, 1);
3147 XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x+1]), *(uint64_t*)(src_y +17 +linesize), temp64, 1);
3151 if(!(s->flags&CODEC_FLAG_GRAY)){
3153 for(i = (!deblock_top) << 1; i<18; i++){
3154 XCHG(h->left_border[i+34 ], src_cb[i*uvlinesize], temp8, xchg);
3155 XCHG(h->left_border[i+34+18], src_cr[i*uvlinesize], temp8, xchg);
3159 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
3160 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
3161 XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+16), *(uint64_t*)(src_cb+1 +uvlinesize), temp64, 1);
3162 XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+24), *(uint64_t*)(src_cr+1 +uvlinesize), temp64, 1);
3167 static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
3168 MpegEncContext * const s = &h->s;
3169 const int mb_x= s->mb_x;
3170 const int mb_y= s->mb_y;
3171 const int mb_xy= mb_x + mb_y*s->mb_stride;
3172 const int mb_type= s->current_picture.mb_type[mb_xy];
3173 uint8_t *dest_y, *dest_cb, *dest_cr;
3174 int linesize, uvlinesize /*dct_offset*/;
3176 int *block_offset = &h->block_offset[0];
3177 const unsigned int bottom = mb_y & 1;
3178 const int transform_bypass = (s->qscale == 0 && h->sps.transform_bypass), is_h264 = (simple || s->codec_id == CODEC_ID_H264);
3179 void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
3180 void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
3182 dest_y = s->current_picture.data[0] + (mb_y * 16* s->linesize ) + mb_x * 16;
3183 dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
3184 dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
3186 s->dsp.prefetch(dest_y + (s->mb_x&3)*4*s->linesize + 64, s->linesize, 4);
3187 s->dsp.prefetch(dest_cb + (s->mb_x&7)*s->uvlinesize + 64, dest_cr - dest_cb, 2);
3189 if (!simple && MB_FIELD) {
3190 linesize = h->mb_linesize = s->linesize * 2;
3191 uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
3192 block_offset = &h->block_offset[24];
3193 if(mb_y&1){ //FIXME move out of this func?
3194 dest_y -= s->linesize*15;
3195 dest_cb-= s->uvlinesize*7;
3196 dest_cr-= s->uvlinesize*7;
3200 for(list=0; list<h->list_count; list++){
3201 if(!USES_LIST(mb_type, list))
3203 if(IS_16X16(mb_type)){
3204 int8_t *ref = &h->ref_cache[list][scan8[0]];
3205 fill_rectangle(ref, 4, 4, 8, 16+*ref^(s->mb_y&1), 1);
3207 for(i=0; i<16; i+=4){
3208 //FIXME can refs be smaller than 8x8 when !direct_8x8_inference ?
3209 int ref = h->ref_cache[list][scan8[i]];
3211 fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, 16+ref^(s->mb_y&1), 1);
3217 linesize = h->mb_linesize = s->linesize;
3218 uvlinesize = h->mb_uvlinesize = s->uvlinesize;
3219 // dct_offset = s->linesize * 16;
3222 if(transform_bypass){
3224 idct_add = IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4;
3225 }else if(IS_8x8DCT(mb_type)){
3226 idct_dc_add = s->dsp.h264_idct8_dc_add;
3227 idct_add = s->dsp.h264_idct8_add;
3229 idct_dc_add = s->dsp.h264_idct_dc_add;
3230 idct_add = s->dsp.h264_idct_add;
3233 if(!simple && FRAME_MBAFF && h->deblocking_filter && IS_INTRA(mb_type)
3234 && (!bottom || !IS_INTRA(s->current_picture.mb_type[mb_xy-s->mb_stride]))){
3235 int mbt_y = mb_y&~1;
3236 uint8_t *top_y = s->current_picture.data[0] + (mbt_y * 16* s->linesize ) + mb_x * 16;
3237 uint8_t *top_cb = s->current_picture.data[1] + (mbt_y * 8 * s->uvlinesize) + mb_x * 8;
3238 uint8_t *top_cr = s->current_picture.data[2] + (mbt_y * 8 * s->uvlinesize) + mb_x * 8;
3239 xchg_pair_border(h, top_y, top_cb, top_cr, s->linesize, s->uvlinesize, 1);
3242 if (!simple && IS_INTRA_PCM(mb_type)) {
3245 // The pixels are stored in h->mb array in the same order as levels,
3246 // copy them in output in the correct order.
3247 for(i=0; i<16; i++) {
3248 for (y=0; y<4; y++) {
3249 for (x=0; x<4; x++) {
3250 *(dest_y + block_offset[i] + y*linesize + x) = h->mb[i*16+y*4+x];
3254 for(i=16; i<16+4; i++) {
3255 for (y=0; y<4; y++) {
3256 for (x=0; x<4; x++) {
3257 *(dest_cb + block_offset[i] + y*uvlinesize + x) = h->mb[i*16+y*4+x];
3261 for(i=20; i<20+4; i++) {
3262 for (y=0; y<4; y++) {
3263 for (x=0; x<4; x++) {
3264 *(dest_cr + block_offset[i] + y*uvlinesize + x) = h->mb[i*16+y*4+x];
3269 if(IS_INTRA(mb_type)){
3270 if(h->deblocking_filter && (simple || !FRAME_MBAFF))
3271 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1, simple);
3273 if(simple || !(s->flags&CODEC_FLAG_GRAY)){
3274 h->pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
3275 h->pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
3278 if(IS_INTRA4x4(mb_type)){
3279 if(simple || !s->encoding){
3280 if(IS_8x8DCT(mb_type)){
3281 for(i=0; i<16; i+=4){
3282 uint8_t * const ptr= dest_y + block_offset[i];
3283 const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
3284 const int nnz = h->non_zero_count_cache[ scan8[i] ];
3285 h->pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
3286 (h->topright_samples_available<<i)&0x4000, linesize);
3288 if(nnz == 1 && h->mb[i*16])
3289 idct_dc_add(ptr, h->mb + i*16, linesize);
3291 idct_add(ptr, h->mb + i*16, linesize);
3295 for(i=0; i<16; i++){
3296 uint8_t * const ptr= dest_y + block_offset[i];
3298 const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
3301 if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
3302 const int topright_avail= (h->topright_samples_available<<i)&0x8000;
3303 assert(mb_y || linesize <= block_offset[i]);
3304 if(!topright_avail){
3305 tr= ptr[3 - linesize]*0x01010101;
3306 topright= (uint8_t*) &tr;
3308 topright= ptr + 4 - linesize;
3312 h->pred4x4[ dir ](ptr, topright, linesize);
3313 nnz = h->non_zero_count_cache[ scan8[i] ];
3316 if(nnz == 1 && h->mb[i*16])
3317 idct_dc_add(ptr, h->mb + i*16, linesize);
3319 idct_add(ptr, h->mb + i*16, linesize);
3321 svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
3326 h->pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
3328 if(!transform_bypass)
3329 h264_luma_dc_dequant_idct_c(h->mb, s->qscale, h->dequant4_coeff[IS_INTRA(mb_type) ? 0:3][s->qscale][0]);
3331 svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
3333 if(h->deblocking_filter && (simple || !FRAME_MBAFF))
3334 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0, simple);
3336 hl_motion(h, dest_y, dest_cb, dest_cr,
3337 s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
3338 s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
3339 s->dsp.weight_h264_pixels_tab, s->dsp.biweight_h264_pixels_tab);
3343 if(!IS_INTRA4x4(mb_type)){
3345 if(IS_INTRA16x16(mb_type)){
3346 for(i=0; i<16; i++){
3347 if(h->non_zero_count_cache[ scan8[i] ])
3348 idct_add(dest_y + block_offset[i], h->mb + i*16, linesize);
3349 else if(h->mb[i*16])
3350 idct_dc_add(dest_y + block_offset[i], h->mb + i*16, linesize);
3353 const int di = IS_8x8DCT(mb_type) ? 4 : 1;
3354 for(i=0; i<16; i+=di){
3355 int nnz = h->non_zero_count_cache[ scan8[i] ];
3357 if(nnz==1 && h->mb[i*16])
3358 idct_dc_add(dest_y + block_offset[i], h->mb + i*16, linesize);
3360 idct_add(dest_y + block_offset[i], h->mb + i*16, linesize);
3365 for(i=0; i<16; i++){
3366 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
3367 uint8_t * const ptr= dest_y + block_offset[i];
3368 svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
3374 if(simple || !(s->flags&CODEC_FLAG_GRAY)){
3375 uint8_t *dest[2] = {dest_cb, dest_cr};
3376 if(transform_bypass){
3377 idct_add = idct_dc_add = s->dsp.add_pixels4;
3379 idct_add = s->dsp.h264_idct_add;
3380 idct_dc_add = s->dsp.h264_idct_dc_add;
3381 chroma_dc_dequant_idct_c(h->mb + 16*16, h->chroma_qp, h->dequant4_coeff[IS_INTRA(mb_type) ? 1:4][h->chroma_qp][0]);
3382 chroma_dc_dequant_idct_c(h->mb + 16*16+4*16, h->chroma_qp, h->dequant4_coeff[IS_INTRA(mb_type) ? 2:5][h->chroma_qp][0]);
3385 for(i=16; i<16+8; i++){
3386 if(h->non_zero_count_cache[ scan8[i] ])
3387 idct_add(dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
3388 else if(h->mb[i*16])
3389 idct_dc_add(dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
3392 for(i=16; i<16+8; i++){
3393 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
3394 uint8_t * const ptr= dest[(i&4)>>2] + block_offset[i];
3395 svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
3401 if(h->deblocking_filter) {
3402 if (!simple && FRAME_MBAFF) {
3403 //FIXME try deblocking one mb at a time?
3404 // the reduction in load/storing mvs and such might outweigh the extra backup/xchg_border
3405 const int mb_y = s->mb_y - 1;
3406 uint8_t *pair_dest_y, *pair_dest_cb, *pair_dest_cr;
3407 const int mb_xy= mb_x + mb_y*s->mb_stride;
3408 const int mb_type_top = s->current_picture.mb_type[mb_xy];
3409 const int mb_type_bottom= s->current_picture.mb_type[mb_xy+s->mb_stride];
3410 if (!bottom) return;
3411 pair_dest_y = s->current_picture.data[0] + (mb_y * 16* s->linesize ) + mb_x * 16;
3412 pair_dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
3413 pair_dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
3415 if(IS_INTRA(mb_type_top | mb_type_bottom))
3416 xchg_pair_border(h, pair_dest_y, pair_dest_cb, pair_dest_cr, s->linesize, s->uvlinesize, 0);
3418 backup_pair_border(h, pair_dest_y, pair_dest_cb, pair_dest_cr, s->linesize, s->uvlinesize);
3422 tprintf(h->s.avctx, "call mbaff filter_mb mb_x:%d mb_y:%d pair_dest_y = %p, dest_y = %p\n", mb_x, mb_y, pair_dest_y, dest_y);
3423 fill_caches(h, mb_type_top, 1); //FIXME don't fill stuff which isn't used by filter_mb
3424 h->chroma_qp = get_chroma_qp(h->pps.chroma_qp_index_offset, s->current_picture.qscale_table[mb_xy]);
3425 filter_mb(h, mb_x, mb_y, pair_dest_y, pair_dest_cb, pair_dest_cr, linesize, uvlinesize);
3428 tprintf(h->s.avctx, "call mbaff filter_mb\n");
3429 fill_caches(h, mb_type_bottom, 1); //FIXME don't fill stuff which isn't used by filter_mb
3430 h->chroma_qp = get_chroma_qp(h->pps.chroma_qp_index_offset, s->current_picture.qscale_table[mb_xy+s->mb_stride]);
3431 filter_mb(h, mb_x, mb_y+1, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
3433 tprintf(h->s.avctx, "call filter_mb\n");
3434 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, simple);
3435 fill_caches(h, mb_type, 1); //FIXME don't fill stuff which isn't used by filter_mb
3436 filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
3442 * Process a macroblock; this case avoids checks for expensive uncommon cases.
3444 static void hl_decode_mb_simple(H264Context *h){
3445 hl_decode_mb_internal(h, 1);
3449 * Process a macroblock; this handles edge cases, such as interlacing.
3451 static void av_noinline hl_decode_mb_complex(H264Context *h){
3452 hl_decode_mb_internal(h, 0);
3455 static void hl_decode_mb(H264Context *h){
3456 MpegEncContext * const s = &h->s;
3457 const int mb_x= s->mb_x;
3458 const int mb_y= s->mb_y;
3459 const int mb_xy= mb_x + mb_y*s->mb_stride;
3460 const int mb_type= s->current_picture.mb_type[mb_xy];
3461 int is_complex = FRAME_MBAFF || MB_FIELD || IS_INTRA_PCM(mb_type) || s->codec_id != CODEC_ID_H264 || (s->flags&CODEC_FLAG_GRAY) || s->encoding;
3467 hl_decode_mb_complex(h);
3468 else hl_decode_mb_simple(h);
3472 * fills the default_ref_list.
3474 static int fill_default_ref_list(H264Context *h){
3475 MpegEncContext * const s = &h->s;
3477 int smallest_poc_greater_than_current = -1;
3478 Picture sorted_short_ref[32];
3480 if(h->slice_type==B_TYPE){
3484 /* sort frame according to poc in B slice */
3485 for(out_i=0; out_i<h->short_ref_count; out_i++){
3487 int best_poc=INT_MAX;
3489 for(i=0; i<h->short_ref_count; i++){
3490 const int poc= h->short_ref[i]->poc;
3491 if(poc > limit && poc < best_poc){
3497 assert(best_i != INT_MIN);
3500 sorted_short_ref[out_i]= *h->short_ref[best_i];
3501 tprintf(h->s.avctx, "sorted poc: %d->%d poc:%d fn:%d\n", best_i, out_i, sorted_short_ref[out_i].poc, sorted_short_ref[out_i].frame_num);
3502 if (-1 == smallest_poc_greater_than_current) {
3503 if (h->short_ref[best_i]->poc >= s->current_picture_ptr->poc) {
3504 smallest_poc_greater_than_current = out_i;