lavc decoders: work with refcounted frames.
[FFMpeg-mirror/mplayer-patches.git] / libavcodec / mpegvideo_motion.c
blob0c86cfdafe645962678784df336315e629b66eaf
1 /*
2 * Copyright (c) 2000,2001 Fabrice Bellard
3 * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
5 * 4MV & hq & B-frame encoding stuff by Michael Niedermayer <michaelni@gmx.at>
7 * This file is part of Libav.
9 * Libav is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public
11 * License as published by the Free Software Foundation; either
12 * version 2.1 of the License, or (at your option) any later version.
14 * Libav is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with Libav; if not, write to the Free Software
21 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
24 #include <string.h>
26 #include "libavutil/internal.h"
27 #include "avcodec.h"
28 #include "dsputil.h"
29 #include "mpegvideo.h"
30 #include "mjpegenc.h"
31 #include "msmpeg4.h"
32 #include <limits.h>
34 static void gmc1_motion(MpegEncContext *s,
35 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
36 uint8_t **ref_picture)
38 uint8_t *ptr;
39 int offset, src_x, src_y, linesize, uvlinesize;
40 int motion_x, motion_y;
41 int emu=0;
43 motion_x= s->sprite_offset[0][0];
44 motion_y= s->sprite_offset[0][1];
45 src_x = s->mb_x * 16 + (motion_x >> (s->sprite_warping_accuracy+1));
46 src_y = s->mb_y * 16 + (motion_y >> (s->sprite_warping_accuracy+1));
47 motion_x<<=(3-s->sprite_warping_accuracy);
48 motion_y<<=(3-s->sprite_warping_accuracy);
49 src_x = av_clip(src_x, -16, s->width);
50 if (src_x == s->width)
51 motion_x =0;
52 src_y = av_clip(src_y, -16, s->height);
53 if (src_y == s->height)
54 motion_y =0;
56 linesize = s->linesize;
57 uvlinesize = s->uvlinesize;
59 ptr = ref_picture[0] + (src_y * linesize) + src_x;
61 if(s->flags&CODEC_FLAG_EMU_EDGE){
62 if( (unsigned)src_x >= FFMAX(s->h_edge_pos - 17, 0)
63 || (unsigned)src_y >= FFMAX(s->v_edge_pos - 17, 0)){
64 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr, linesize, 17, 17, src_x, src_y, s->h_edge_pos, s->v_edge_pos);
65 ptr= s->edge_emu_buffer;
69 if((motion_x|motion_y)&7){
70 s->dsp.gmc1(dest_y , ptr , linesize, 16, motion_x&15, motion_y&15, 128 - s->no_rounding);
71 s->dsp.gmc1(dest_y+8, ptr+8, linesize, 16, motion_x&15, motion_y&15, 128 - s->no_rounding);
72 }else{
73 int dxy;
75 dxy= ((motion_x>>3)&1) | ((motion_y>>2)&2);
76 if (s->no_rounding){
77 s->dsp.put_no_rnd_pixels_tab[0][dxy](dest_y, ptr, linesize, 16);
78 }else{
79 s->dsp.put_pixels_tab [0][dxy](dest_y, ptr, linesize, 16);
83 if(CONFIG_GRAY && s->flags&CODEC_FLAG_GRAY) return;
85 motion_x= s->sprite_offset[1][0];
86 motion_y= s->sprite_offset[1][1];
87 src_x = s->mb_x * 8 + (motion_x >> (s->sprite_warping_accuracy+1));
88 src_y = s->mb_y * 8 + (motion_y >> (s->sprite_warping_accuracy+1));
89 motion_x<<=(3-s->sprite_warping_accuracy);
90 motion_y<<=(3-s->sprite_warping_accuracy);
91 src_x = av_clip(src_x, -8, s->width>>1);
92 if (src_x == s->width>>1)
93 motion_x =0;
94 src_y = av_clip(src_y, -8, s->height>>1);
95 if (src_y == s->height>>1)
96 motion_y =0;
98 offset = (src_y * uvlinesize) + src_x;
99 ptr = ref_picture[1] + offset;
100 if(s->flags&CODEC_FLAG_EMU_EDGE){
101 if( (unsigned)src_x >= FFMAX((s->h_edge_pos>>1) - 9, 0)
102 || (unsigned)src_y >= FFMAX((s->v_edge_pos>>1) - 9, 0)){
103 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr, uvlinesize, 9, 9, src_x, src_y, s->h_edge_pos>>1, s->v_edge_pos>>1);
104 ptr= s->edge_emu_buffer;
105 emu=1;
108 s->dsp.gmc1(dest_cb, ptr, uvlinesize, 8, motion_x&15, motion_y&15, 128 - s->no_rounding);
110 ptr = ref_picture[2] + offset;
111 if(emu){
112 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr, uvlinesize, 9, 9, src_x, src_y, s->h_edge_pos>>1, s->v_edge_pos>>1);
113 ptr= s->edge_emu_buffer;
115 s->dsp.gmc1(dest_cr, ptr, uvlinesize, 8, motion_x&15, motion_y&15, 128 - s->no_rounding);
117 return;
120 static void gmc_motion(MpegEncContext *s,
121 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
122 uint8_t **ref_picture)
124 uint8_t *ptr;
125 int linesize, uvlinesize;
126 const int a= s->sprite_warping_accuracy;
127 int ox, oy;
129 linesize = s->linesize;
130 uvlinesize = s->uvlinesize;
132 ptr = ref_picture[0];
134 ox= s->sprite_offset[0][0] + s->sprite_delta[0][0]*s->mb_x*16 + s->sprite_delta[0][1]*s->mb_y*16;
135 oy= s->sprite_offset[0][1] + s->sprite_delta[1][0]*s->mb_x*16 + s->sprite_delta[1][1]*s->mb_y*16;
137 s->dsp.gmc(dest_y, ptr, linesize, 16,
140 s->sprite_delta[0][0], s->sprite_delta[0][1],
141 s->sprite_delta[1][0], s->sprite_delta[1][1],
142 a+1, (1<<(2*a+1)) - s->no_rounding,
143 s->h_edge_pos, s->v_edge_pos);
144 s->dsp.gmc(dest_y+8, ptr, linesize, 16,
145 ox + s->sprite_delta[0][0]*8,
146 oy + s->sprite_delta[1][0]*8,
147 s->sprite_delta[0][0], s->sprite_delta[0][1],
148 s->sprite_delta[1][0], s->sprite_delta[1][1],
149 a+1, (1<<(2*a+1)) - s->no_rounding,
150 s->h_edge_pos, s->v_edge_pos);
152 if(CONFIG_GRAY && s->flags&CODEC_FLAG_GRAY) return;
154 ox= s->sprite_offset[1][0] + s->sprite_delta[0][0]*s->mb_x*8 + s->sprite_delta[0][1]*s->mb_y*8;
155 oy= s->sprite_offset[1][1] + s->sprite_delta[1][0]*s->mb_x*8 + s->sprite_delta[1][1]*s->mb_y*8;
157 ptr = ref_picture[1];
158 s->dsp.gmc(dest_cb, ptr, uvlinesize, 8,
161 s->sprite_delta[0][0], s->sprite_delta[0][1],
162 s->sprite_delta[1][0], s->sprite_delta[1][1],
163 a+1, (1<<(2*a+1)) - s->no_rounding,
164 s->h_edge_pos>>1, s->v_edge_pos>>1);
166 ptr = ref_picture[2];
167 s->dsp.gmc(dest_cr, ptr, uvlinesize, 8,
170 s->sprite_delta[0][0], s->sprite_delta[0][1],
171 s->sprite_delta[1][0], s->sprite_delta[1][1],
172 a+1, (1<<(2*a+1)) - s->no_rounding,
173 s->h_edge_pos>>1, s->v_edge_pos>>1);
176 static inline int hpel_motion(MpegEncContext *s,
177 uint8_t *dest, uint8_t *src,
178 int src_x, int src_y,
179 op_pixels_func *pix_op,
180 int motion_x, int motion_y)
182 int dxy = 0;
183 int emu=0;
185 src_x += motion_x >> 1;
186 src_y += motion_y >> 1;
188 /* WARNING: do no forget half pels */
189 src_x = av_clip(src_x, -16, s->width); //FIXME unneeded for emu?
190 if (src_x != s->width)
191 dxy |= motion_x & 1;
192 src_y = av_clip(src_y, -16, s->height);
193 if (src_y != s->height)
194 dxy |= (motion_y & 1) << 1;
195 src += src_y * s->linesize + src_x;
197 if(s->unrestricted_mv && (s->flags&CODEC_FLAG_EMU_EDGE)){
198 if( (unsigned)src_x > FFMAX(s->h_edge_pos - (motion_x&1) - 8, 0)
199 || (unsigned)src_y > FFMAX(s->v_edge_pos - (motion_y&1) - 8, 0)){
200 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, src, s->linesize, 9, 9,
201 src_x, src_y, s->h_edge_pos, s->v_edge_pos);
202 src= s->edge_emu_buffer;
203 emu=1;
206 pix_op[dxy](dest, src, s->linesize, 8);
207 return emu;
210 static av_always_inline
211 void mpeg_motion_internal(MpegEncContext *s,
212 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
213 int field_based, int bottom_field, int field_select,
214 uint8_t **ref_picture, op_pixels_func (*pix_op)[4],
215 int motion_x, int motion_y, int h, int is_mpeg12, int mb_y)
217 uint8_t *ptr_y, *ptr_cb, *ptr_cr;
218 int dxy, uvdxy, mx, my, src_x, src_y,
219 uvsrc_x, uvsrc_y, v_edge_pos, uvlinesize, linesize;
221 #if 0
222 if(s->quarter_sample)
224 motion_x>>=1;
225 motion_y>>=1;
227 #endif
229 v_edge_pos = s->v_edge_pos >> field_based;
230 linesize = s->current_picture.f.linesize[0] << field_based;
231 uvlinesize = s->current_picture.f.linesize[1] << field_based;
233 dxy = ((motion_y & 1) << 1) | (motion_x & 1);
234 src_x = s->mb_x* 16 + (motion_x >> 1);
235 src_y =( mb_y<<(4-field_based)) + (motion_y >> 1);
237 if (!is_mpeg12 && s->out_format == FMT_H263) {
238 if((s->workaround_bugs & FF_BUG_HPEL_CHROMA) && field_based){
239 mx = (motion_x>>1)|(motion_x&1);
240 my = motion_y >>1;
241 uvdxy = ((my & 1) << 1) | (mx & 1);
242 uvsrc_x = s->mb_x* 8 + (mx >> 1);
243 uvsrc_y =( mb_y<<(3-field_based))+ (my >> 1);
244 }else{
245 uvdxy = dxy | (motion_y & 2) | ((motion_x & 2) >> 1);
246 uvsrc_x = src_x>>1;
247 uvsrc_y = src_y>>1;
249 }else if(!is_mpeg12 && s->out_format == FMT_H261){//even chroma mv's are full pel in H261
250 mx = motion_x / 4;
251 my = motion_y / 4;
252 uvdxy = 0;
253 uvsrc_x = s->mb_x*8 + mx;
254 uvsrc_y = mb_y*8 + my;
255 } else {
256 if(s->chroma_y_shift){
257 mx = motion_x / 2;
258 my = motion_y / 2;
259 uvdxy = ((my & 1) << 1) | (mx & 1);
260 uvsrc_x = s->mb_x* 8 + (mx >> 1);
261 uvsrc_y =( mb_y<<(3-field_based))+ (my >> 1);
262 } else {
263 if(s->chroma_x_shift){
264 //Chroma422
265 mx = motion_x / 2;
266 uvdxy = ((motion_y & 1) << 1) | (mx & 1);
267 uvsrc_x = s->mb_x* 8 + (mx >> 1);
268 uvsrc_y = src_y;
269 } else {
270 //Chroma444
271 uvdxy = dxy;
272 uvsrc_x = src_x;
273 uvsrc_y = src_y;
278 ptr_y = ref_picture[0] + src_y * linesize + src_x;
279 ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x;
280 ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x;
282 if( (unsigned)src_x > FFMAX(s->h_edge_pos - (motion_x&1) - 16, 0)
283 || (unsigned)src_y > FFMAX( v_edge_pos - (motion_y&1) - h , 0)){
284 if(is_mpeg12 || s->codec_id == AV_CODEC_ID_MPEG2VIDEO ||
285 s->codec_id == AV_CODEC_ID_MPEG1VIDEO){
286 av_log(s->avctx,AV_LOG_DEBUG,
287 "MPEG motion vector out of boundary (%d %d)\n", src_x, src_y);
288 return;
290 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr_y, s->linesize,
291 17, 17+field_based,
292 src_x, src_y<<field_based,
293 s->h_edge_pos, s->v_edge_pos);
294 ptr_y = s->edge_emu_buffer;
295 if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
296 uint8_t *uvbuf= s->edge_emu_buffer+18*s->linesize;
297 s->vdsp.emulated_edge_mc(uvbuf ,
298 ptr_cb, s->uvlinesize,
299 9, 9+field_based,
300 uvsrc_x, uvsrc_y<<field_based,
301 s->h_edge_pos>>1, s->v_edge_pos>>1);
302 s->vdsp.emulated_edge_mc(uvbuf+16,
303 ptr_cr, s->uvlinesize,
304 9, 9+field_based,
305 uvsrc_x, uvsrc_y<<field_based,
306 s->h_edge_pos>>1, s->v_edge_pos>>1);
307 ptr_cb= uvbuf;
308 ptr_cr= uvbuf+16;
312 if(bottom_field){ //FIXME use this for field pix too instead of the obnoxious hack which changes picture.data
313 dest_y += s->linesize;
314 dest_cb+= s->uvlinesize;
315 dest_cr+= s->uvlinesize;
318 if(field_select){
319 ptr_y += s->linesize;
320 ptr_cb+= s->uvlinesize;
321 ptr_cr+= s->uvlinesize;
324 pix_op[0][dxy](dest_y, ptr_y, linesize, h);
326 if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
327 pix_op[s->chroma_x_shift][uvdxy]
328 (dest_cb, ptr_cb, uvlinesize, h >> s->chroma_y_shift);
329 pix_op[s->chroma_x_shift][uvdxy]
330 (dest_cr, ptr_cr, uvlinesize, h >> s->chroma_y_shift);
332 if(!is_mpeg12 && (CONFIG_H261_ENCODER || CONFIG_H261_DECODER) &&
333 s->out_format == FMT_H261){
334 ff_h261_loop_filter(s);
337 /* apply one mpeg motion vector to the three components */
338 static void mpeg_motion(MpegEncContext *s,
339 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
340 int field_select, uint8_t **ref_picture,
341 op_pixels_func (*pix_op)[4],
342 int motion_x, int motion_y, int h, int mb_y)
344 #if !CONFIG_SMALL
345 if(s->out_format == FMT_MPEG1)
346 mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, 0, 0,
347 field_select, ref_picture, pix_op,
348 motion_x, motion_y, h, 1, mb_y);
349 else
350 #endif
351 mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, 0, 0,
352 field_select, ref_picture, pix_op,
353 motion_x, motion_y, h, 0, mb_y);
356 static void mpeg_motion_field(MpegEncContext *s, uint8_t *dest_y,
357 uint8_t *dest_cb, uint8_t *dest_cr,
358 int bottom_field, int field_select,
359 uint8_t **ref_picture,
360 op_pixels_func (*pix_op)[4],
361 int motion_x, int motion_y, int h, int mb_y)
363 #if !CONFIG_SMALL
364 if(s->out_format == FMT_MPEG1)
365 mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, 1,
366 bottom_field, field_select, ref_picture, pix_op,
367 motion_x, motion_y, h, 1, mb_y);
368 else
369 #endif
370 mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, 1,
371 bottom_field, field_select, ref_picture, pix_op,
372 motion_x, motion_y, h, 0, mb_y);
375 //FIXME move to dsputil, avg variant, 16x16 version
376 static inline void put_obmc(uint8_t *dst, uint8_t *src[5], int stride){
377 int x;
378 uint8_t * const top = src[1];
379 uint8_t * const left = src[2];
380 uint8_t * const mid = src[0];
381 uint8_t * const right = src[3];
382 uint8_t * const bottom= src[4];
383 #define OBMC_FILTER(x, t, l, m, r, b)\
384 dst[x]= (t*top[x] + l*left[x] + m*mid[x] + r*right[x] + b*bottom[x] + 4)>>3
385 #define OBMC_FILTER4(x, t, l, m, r, b)\
386 OBMC_FILTER(x , t, l, m, r, b);\
387 OBMC_FILTER(x+1 , t, l, m, r, b);\
388 OBMC_FILTER(x +stride, t, l, m, r, b);\
389 OBMC_FILTER(x+1+stride, t, l, m, r, b);
391 x=0;
392 OBMC_FILTER (x , 2, 2, 4, 0, 0);
393 OBMC_FILTER (x+1, 2, 1, 5, 0, 0);
394 OBMC_FILTER4(x+2, 2, 1, 5, 0, 0);
395 OBMC_FILTER4(x+4, 2, 0, 5, 1, 0);
396 OBMC_FILTER (x+6, 2, 0, 5, 1, 0);
397 OBMC_FILTER (x+7, 2, 0, 4, 2, 0);
398 x+= stride;
399 OBMC_FILTER (x , 1, 2, 5, 0, 0);
400 OBMC_FILTER (x+1, 1, 2, 5, 0, 0);
401 OBMC_FILTER (x+6, 1, 0, 5, 2, 0);
402 OBMC_FILTER (x+7, 1, 0, 5, 2, 0);
403 x+= stride;
404 OBMC_FILTER4(x , 1, 2, 5, 0, 0);
405 OBMC_FILTER4(x+2, 1, 1, 6, 0, 0);
406 OBMC_FILTER4(x+4, 1, 0, 6, 1, 0);
407 OBMC_FILTER4(x+6, 1, 0, 5, 2, 0);
408 x+= 2*stride;
409 OBMC_FILTER4(x , 0, 2, 5, 0, 1);
410 OBMC_FILTER4(x+2, 0, 1, 6, 0, 1);
411 OBMC_FILTER4(x+4, 0, 0, 6, 1, 1);
412 OBMC_FILTER4(x+6, 0, 0, 5, 2, 1);
413 x+= 2*stride;
414 OBMC_FILTER (x , 0, 2, 5, 0, 1);
415 OBMC_FILTER (x+1, 0, 2, 5, 0, 1);
416 OBMC_FILTER4(x+2, 0, 1, 5, 0, 2);
417 OBMC_FILTER4(x+4, 0, 0, 5, 1, 2);
418 OBMC_FILTER (x+6, 0, 0, 5, 2, 1);
419 OBMC_FILTER (x+7, 0, 0, 5, 2, 1);
420 x+= stride;
421 OBMC_FILTER (x , 0, 2, 4, 0, 2);
422 OBMC_FILTER (x+1, 0, 1, 5, 0, 2);
423 OBMC_FILTER (x+6, 0, 0, 5, 1, 2);
424 OBMC_FILTER (x+7, 0, 0, 4, 2, 2);
427 /* obmc for 1 8x8 luma block */
428 static inline void obmc_motion(MpegEncContext *s,
429 uint8_t *dest, uint8_t *src,
430 int src_x, int src_y,
431 op_pixels_func *pix_op,
432 int16_t mv[5][2]/* mid top left right bottom*/)
433 #define MID 0
435 int i;
436 uint8_t *ptr[5];
438 assert(s->quarter_sample==0);
440 for(i=0; i<5; i++){
441 if(i && mv[i][0]==mv[MID][0] && mv[i][1]==mv[MID][1]){
442 ptr[i]= ptr[MID];
443 }else{
444 ptr[i]= s->obmc_scratchpad + 8*(i&1) + s->linesize*8*(i>>1);
445 hpel_motion(s, ptr[i], src,
446 src_x, src_y,
447 pix_op,
448 mv[i][0], mv[i][1]);
452 put_obmc(dest, ptr, s->linesize);
455 static inline void qpel_motion(MpegEncContext *s,
456 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
457 int field_based, int bottom_field, int field_select,
458 uint8_t **ref_picture, op_pixels_func (*pix_op)[4],
459 qpel_mc_func (*qpix_op)[16],
460 int motion_x, int motion_y, int h)
462 uint8_t *ptr_y, *ptr_cb, *ptr_cr;
463 int dxy, uvdxy, mx, my, src_x, src_y, uvsrc_x, uvsrc_y, v_edge_pos, linesize, uvlinesize;
465 dxy = ((motion_y & 3) << 2) | (motion_x & 3);
466 src_x = s->mb_x * 16 + (motion_x >> 2);
467 src_y = s->mb_y * (16 >> field_based) + (motion_y >> 2);
469 v_edge_pos = s->v_edge_pos >> field_based;
470 linesize = s->linesize << field_based;
471 uvlinesize = s->uvlinesize << field_based;
473 if(field_based){
474 mx= motion_x/2;
475 my= motion_y>>1;
476 }else if(s->workaround_bugs&FF_BUG_QPEL_CHROMA2){
477 static const int rtab[8]= {0,0,1,1,0,0,0,1};
478 mx= (motion_x>>1) + rtab[motion_x&7];
479 my= (motion_y>>1) + rtab[motion_y&7];
480 }else if(s->workaround_bugs&FF_BUG_QPEL_CHROMA){
481 mx= (motion_x>>1)|(motion_x&1);
482 my= (motion_y>>1)|(motion_y&1);
483 }else{
484 mx= motion_x/2;
485 my= motion_y/2;
487 mx= (mx>>1)|(mx&1);
488 my= (my>>1)|(my&1);
490 uvdxy= (mx&1) | ((my&1)<<1);
491 mx>>=1;
492 my>>=1;
494 uvsrc_x = s->mb_x * 8 + mx;
495 uvsrc_y = s->mb_y * (8 >> field_based) + my;
497 ptr_y = ref_picture[0] + src_y * linesize + src_x;
498 ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x;
499 ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x;
501 if( (unsigned)src_x > FFMAX(s->h_edge_pos - (motion_x&3) - 16, 0)
502 || (unsigned)src_y > FFMAX( v_edge_pos - (motion_y&3) - h , 0)){
503 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr_y, s->linesize,
504 17, 17+field_based, src_x, src_y<<field_based,
505 s->h_edge_pos, s->v_edge_pos);
506 ptr_y= s->edge_emu_buffer;
507 if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
508 uint8_t *uvbuf= s->edge_emu_buffer + 18*s->linesize;
509 s->vdsp.emulated_edge_mc(uvbuf, ptr_cb, s->uvlinesize,
510 9, 9 + field_based,
511 uvsrc_x, uvsrc_y<<field_based,
512 s->h_edge_pos>>1, s->v_edge_pos>>1);
513 s->vdsp.emulated_edge_mc(uvbuf + 16, ptr_cr, s->uvlinesize,
514 9, 9 + field_based,
515 uvsrc_x, uvsrc_y<<field_based,
516 s->h_edge_pos>>1, s->v_edge_pos>>1);
517 ptr_cb= uvbuf;
518 ptr_cr= uvbuf + 16;
522 if(!field_based)
523 qpix_op[0][dxy](dest_y, ptr_y, linesize);
524 else{
525 if(bottom_field){
526 dest_y += s->linesize;
527 dest_cb+= s->uvlinesize;
528 dest_cr+= s->uvlinesize;
531 if(field_select){
532 ptr_y += s->linesize;
533 ptr_cb += s->uvlinesize;
534 ptr_cr += s->uvlinesize;
536 //damn interlaced mode
537 //FIXME boundary mirroring is not exactly correct here
538 qpix_op[1][dxy](dest_y , ptr_y , linesize);
539 qpix_op[1][dxy](dest_y+8, ptr_y+8, linesize);
541 if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
542 pix_op[1][uvdxy](dest_cr, ptr_cr, uvlinesize, h >> 1);
543 pix_op[1][uvdxy](dest_cb, ptr_cb, uvlinesize, h >> 1);
548 * h263 chroma 4mv motion compensation.
550 static void chroma_4mv_motion(MpegEncContext *s,
551 uint8_t *dest_cb, uint8_t *dest_cr,
552 uint8_t **ref_picture,
553 op_pixels_func *pix_op,
554 int mx, int my)
556 int dxy, emu=0, src_x, src_y, offset;
557 uint8_t *ptr;
559 /* In case of 8X8, we construct a single chroma motion vector
560 with a special rounding */
561 mx= ff_h263_round_chroma(mx);
562 my= ff_h263_round_chroma(my);
564 dxy = ((my & 1) << 1) | (mx & 1);
565 mx >>= 1;
566 my >>= 1;
568 src_x = s->mb_x * 8 + mx;
569 src_y = s->mb_y * 8 + my;
570 src_x = av_clip(src_x, -8, (s->width >> 1));
571 if (src_x == (s->width >> 1))
572 dxy &= ~1;
573 src_y = av_clip(src_y, -8, (s->height >> 1));
574 if (src_y == (s->height >> 1))
575 dxy &= ~2;
577 offset = src_y * s->uvlinesize + src_x;
578 ptr = ref_picture[1] + offset;
579 if(s->flags&CODEC_FLAG_EMU_EDGE){
580 if( (unsigned)src_x > FFMAX((s->h_edge_pos>>1) - (dxy &1) - 8, 0)
581 || (unsigned)src_y > FFMAX((s->v_edge_pos>>1) - (dxy>>1) - 8, 0)){
582 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr, s->uvlinesize,
583 9, 9, src_x, src_y,
584 s->h_edge_pos>>1, s->v_edge_pos>>1);
585 ptr= s->edge_emu_buffer;
586 emu=1;
589 pix_op[dxy](dest_cb, ptr, s->uvlinesize, 8);
591 ptr = ref_picture[2] + offset;
592 if(emu){
593 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr, s->uvlinesize,
594 9, 9, src_x, src_y,
595 s->h_edge_pos>>1, s->v_edge_pos>>1);
596 ptr= s->edge_emu_buffer;
598 pix_op[dxy](dest_cr, ptr, s->uvlinesize, 8);
601 static inline void prefetch_motion(MpegEncContext *s, uint8_t **pix, int dir){
602 /* fetch pixels for estimated mv 4 macroblocks ahead
603 * optimized for 64byte cache lines */
604 const int shift = s->quarter_sample ? 2 : 1;
605 const int mx= (s->mv[dir][0][0]>>shift) + 16*s->mb_x + 8;
606 const int my= (s->mv[dir][0][1]>>shift) + 16*s->mb_y;
607 int off= mx + (my + (s->mb_x&3)*4)*s->linesize + 64;
608 s->vdsp.prefetch(pix[0]+off, s->linesize, 4);
609 off= (mx>>1) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + 64;
610 s->vdsp.prefetch(pix[1]+off, pix[2]-pix[1], 2);
614 * motion compensation of a single macroblock
615 * @param s context
616 * @param dest_y luma destination pointer
617 * @param dest_cb chroma cb/u destination pointer
618 * @param dest_cr chroma cr/v destination pointer
619 * @param dir direction (0->forward, 1->backward)
620 * @param ref_picture array[3] of pointers to the 3 planes of the reference picture
621 * @param pix_op halfpel motion compensation function (average or put normally)
622 * @param qpix_op qpel motion compensation function (average or put normally)
623 * the motion vectors are taken from s->mv and the MV type from s->mv_type
625 static av_always_inline void MPV_motion_internal(MpegEncContext *s,
626 uint8_t *dest_y, uint8_t *dest_cb,
627 uint8_t *dest_cr, int dir,
628 uint8_t **ref_picture,
629 op_pixels_func (*pix_op)[4],
630 qpel_mc_func (*qpix_op)[16], int is_mpeg12)
632 int dxy, mx, my, src_x, src_y, motion_x, motion_y;
633 int mb_x, mb_y, i;
634 uint8_t *ptr, *dest;
636 mb_x = s->mb_x;
637 mb_y = s->mb_y;
639 prefetch_motion(s, ref_picture, dir);
641 if(!is_mpeg12 && s->obmc && s->pict_type != AV_PICTURE_TYPE_B){
642 LOCAL_ALIGNED_8(int16_t, mv_cache, [4], [4][2]);
643 Picture *cur_frame = &s->current_picture;
644 const int xy= s->mb_x + s->mb_y*s->mb_stride;
645 const int mot_stride= s->b8_stride;
646 const int mot_xy= mb_x*2 + mb_y*2*mot_stride;
648 assert(!s->mb_skipped);
650 AV_COPY32(mv_cache[1][1], cur_frame->motion_val[0][mot_xy ]);
651 AV_COPY32(mv_cache[1][2], cur_frame->motion_val[0][mot_xy + 1]);
653 AV_COPY32(mv_cache[2][1], cur_frame->motion_val[0][mot_xy + mot_stride ]);
654 AV_COPY32(mv_cache[2][2], cur_frame->motion_val[0][mot_xy + mot_stride + 1]);
656 AV_COPY32(mv_cache[3][1], cur_frame->motion_val[0][mot_xy + mot_stride ]);
657 AV_COPY32(mv_cache[3][2], cur_frame->motion_val[0][mot_xy + mot_stride + 1]);
659 if (mb_y == 0 || IS_INTRA(cur_frame->mb_type[xy - s->mb_stride])) {
660 AV_COPY32(mv_cache[0][1], mv_cache[1][1]);
661 AV_COPY32(mv_cache[0][2], mv_cache[1][2]);
662 }else{
663 AV_COPY32(mv_cache[0][1], cur_frame->motion_val[0][mot_xy - mot_stride ]);
664 AV_COPY32(mv_cache[0][2], cur_frame->motion_val[0][mot_xy - mot_stride + 1]);
667 if (mb_x == 0 || IS_INTRA(cur_frame->mb_type[xy - 1])) {
668 AV_COPY32(mv_cache[1][0], mv_cache[1][1]);
669 AV_COPY32(mv_cache[2][0], mv_cache[2][1]);
670 }else{
671 AV_COPY32(mv_cache[1][0], cur_frame->motion_val[0][mot_xy - 1]);
672 AV_COPY32(mv_cache[2][0], cur_frame->motion_val[0][mot_xy - 1 + mot_stride]);
675 if (mb_x + 1 >= s->mb_width || IS_INTRA(cur_frame->mb_type[xy + 1])) {
676 AV_COPY32(mv_cache[1][3], mv_cache[1][2]);
677 AV_COPY32(mv_cache[2][3], mv_cache[2][2]);
678 }else{
679 AV_COPY32(mv_cache[1][3], cur_frame->motion_val[0][mot_xy + 2]);
680 AV_COPY32(mv_cache[2][3], cur_frame->motion_val[0][mot_xy + 2 + mot_stride]);
683 mx = 0;
684 my = 0;
685 for(i=0;i<4;i++) {
686 const int x= (i&1)+1;
687 const int y= (i>>1)+1;
688 int16_t mv[5][2]= {
689 {mv_cache[y][x ][0], mv_cache[y][x ][1]},
690 {mv_cache[y-1][x][0], mv_cache[y-1][x][1]},
691 {mv_cache[y][x-1][0], mv_cache[y][x-1][1]},
692 {mv_cache[y][x+1][0], mv_cache[y][x+1][1]},
693 {mv_cache[y+1][x][0], mv_cache[y+1][x][1]}};
694 //FIXME cleanup
695 obmc_motion(s, dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize,
696 ref_picture[0],
697 mb_x * 16 + (i & 1) * 8, mb_y * 16 + (i >>1) * 8,
698 pix_op[1],
699 mv);
701 mx += mv[0][0];
702 my += mv[0][1];
704 if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY))
705 chroma_4mv_motion(s, dest_cb, dest_cr, ref_picture, pix_op[1], mx, my);
707 return;
710 switch(s->mv_type) {
711 case MV_TYPE_16X16:
712 if(s->mcsel){
713 if(s->real_sprite_warping_points==1){
714 gmc1_motion(s, dest_y, dest_cb, dest_cr,
715 ref_picture);
716 }else{
717 gmc_motion(s, dest_y, dest_cb, dest_cr,
718 ref_picture);
720 }else if(!is_mpeg12 && s->quarter_sample){
721 qpel_motion(s, dest_y, dest_cb, dest_cr,
722 0, 0, 0,
723 ref_picture, pix_op, qpix_op,
724 s->mv[dir][0][0], s->mv[dir][0][1], 16);
725 } else if (!is_mpeg12 && (CONFIG_WMV2_DECODER || CONFIG_WMV2_ENCODER) &&
726 s->mspel && s->codec_id == AV_CODEC_ID_WMV2) {
727 ff_mspel_motion(s, dest_y, dest_cb, dest_cr,
728 ref_picture, pix_op,
729 s->mv[dir][0][0], s->mv[dir][0][1], 16);
730 }else
732 mpeg_motion(s, dest_y, dest_cb, dest_cr, 0,
733 ref_picture, pix_op,
734 s->mv[dir][0][0], s->mv[dir][0][1], 16, mb_y);
736 break;
737 case MV_TYPE_8X8:
738 if (!is_mpeg12) {
739 mx = 0;
740 my = 0;
741 if(s->quarter_sample){
742 for(i=0;i<4;i++) {
743 motion_x = s->mv[dir][i][0];
744 motion_y = s->mv[dir][i][1];
746 dxy = ((motion_y & 3) << 2) | (motion_x & 3);
747 src_x = mb_x * 16 + (motion_x >> 2) + (i & 1) * 8;
748 src_y = mb_y * 16 + (motion_y >> 2) + (i >>1) * 8;
750 /* WARNING: do no forget half pels */
751 src_x = av_clip(src_x, -16, s->width);
752 if (src_x == s->width)
753 dxy &= ~3;
754 src_y = av_clip(src_y, -16, s->height);
755 if (src_y == s->height)
756 dxy &= ~12;
758 ptr = ref_picture[0] + (src_y * s->linesize) + (src_x);
759 if(s->flags&CODEC_FLAG_EMU_EDGE){
760 if( (unsigned)src_x > FFMAX(s->h_edge_pos - (motion_x&3) - 8, 0)
761 || (unsigned)src_y > FFMAX(s->v_edge_pos - (motion_y&3) - 8, 0)){
762 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr,
763 s->linesize, 9, 9,
764 src_x, src_y,
765 s->h_edge_pos, s->v_edge_pos);
766 ptr= s->edge_emu_buffer;
769 dest = dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize;
770 qpix_op[1][dxy](dest, ptr, s->linesize);
772 mx += s->mv[dir][i][0]/2;
773 my += s->mv[dir][i][1]/2;
775 }else{
776 for(i=0;i<4;i++) {
777 hpel_motion(s, dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize,
778 ref_picture[0],
779 mb_x * 16 + (i & 1) * 8, mb_y * 16 + (i >>1) * 8,
780 pix_op[1],
781 s->mv[dir][i][0], s->mv[dir][i][1]);
783 mx += s->mv[dir][i][0];
784 my += s->mv[dir][i][1];
788 if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY))
789 chroma_4mv_motion(s, dest_cb, dest_cr, ref_picture, pix_op[1], mx, my);
791 break;
792 case MV_TYPE_FIELD:
793 if (s->picture_structure == PICT_FRAME) {
794 if(!is_mpeg12 && s->quarter_sample){
795 for(i=0; i<2; i++){
796 qpel_motion(s, dest_y, dest_cb, dest_cr,
797 1, i, s->field_select[dir][i],
798 ref_picture, pix_op, qpix_op,
799 s->mv[dir][i][0], s->mv[dir][i][1], 8);
801 }else{
802 /* top field */
803 mpeg_motion_field(s, dest_y, dest_cb, dest_cr,
804 0, s->field_select[dir][0],
805 ref_picture, pix_op,
806 s->mv[dir][0][0], s->mv[dir][0][1], 8, mb_y);
807 /* bottom field */
808 mpeg_motion_field(s, dest_y, dest_cb, dest_cr,
809 1, s->field_select[dir][1],
810 ref_picture, pix_op,
811 s->mv[dir][1][0], s->mv[dir][1][1], 8, mb_y);
813 } else {
814 if(s->picture_structure != s->field_select[dir][0] + 1 && s->pict_type != AV_PICTURE_TYPE_B && !s->first_field){
815 ref_picture = s->current_picture_ptr->f.data;
818 mpeg_motion(s, dest_y, dest_cb, dest_cr,
819 s->field_select[dir][0],
820 ref_picture, pix_op,
821 s->mv[dir][0][0], s->mv[dir][0][1], 16, mb_y>>1);
823 break;
824 case MV_TYPE_16X8:
825 for(i=0; i<2; i++){
826 uint8_t ** ref2picture;
828 if(s->picture_structure == s->field_select[dir][i] + 1
829 || s->pict_type == AV_PICTURE_TYPE_B || s->first_field){
830 ref2picture= ref_picture;
831 }else{
832 ref2picture = s->current_picture_ptr->f.data;
835 mpeg_motion(s, dest_y, dest_cb, dest_cr,
836 s->field_select[dir][i],
837 ref2picture, pix_op,
838 s->mv[dir][i][0], s->mv[dir][i][1] + 16*i, 8, mb_y>>1);
840 dest_y += 16*s->linesize;
841 dest_cb+= (16>>s->chroma_y_shift)*s->uvlinesize;
842 dest_cr+= (16>>s->chroma_y_shift)*s->uvlinesize;
844 break;
845 case MV_TYPE_DMV:
846 if(s->picture_structure == PICT_FRAME){
847 for(i=0; i<2; i++){
848 int j;
849 for(j=0; j<2; j++){
850 mpeg_motion_field(s, dest_y, dest_cb, dest_cr,
851 j, j^i, ref_picture, pix_op,
852 s->mv[dir][2*i + j][0],
853 s->mv[dir][2*i + j][1], 8, mb_y);
855 pix_op = s->dsp.avg_pixels_tab;
857 }else{
858 for(i=0; i<2; i++){
859 mpeg_motion(s, dest_y, dest_cb, dest_cr,
860 s->picture_structure != i+1,
861 ref_picture, pix_op,
862 s->mv[dir][2*i][0],s->mv[dir][2*i][1],16, mb_y>>1);
864 // after put we make avg of the same block
865 pix_op=s->dsp.avg_pixels_tab;
867 //opposite parity is always in the same frame if this is second field
868 if(!s->first_field){
869 ref_picture = s->current_picture_ptr->f.data;
873 break;
874 default: assert(0);
878 void ff_MPV_motion(MpegEncContext *s,
879 uint8_t *dest_y, uint8_t *dest_cb,
880 uint8_t *dest_cr, int dir,
881 uint8_t **ref_picture,
882 op_pixels_func (*pix_op)[4],
883 qpel_mc_func (*qpix_op)[16])
885 #if !CONFIG_SMALL
886 if(s->out_format == FMT_MPEG1)
887 MPV_motion_internal(s, dest_y, dest_cb, dest_cr, dir,
888 ref_picture, pix_op, qpix_op, 1);
889 else
890 #endif
891 MPV_motion_internal(s, dest_y, dest_cb, dest_cr, dir,
892 ref_picture, pix_op, qpix_op, 0);