FFmpeg
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
mpegvideo_motion.c
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2000,2001 Fabrice Bellard
3  * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
4  *
5  * 4MV & hq & B-frame encoding stuff by Michael Niedermayer <michaelni@gmx.at>
6  *
7  * This file is part of FFmpeg.
8  *
9  * FFmpeg 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.
13  *
14  * FFmpeg 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.
18  *
19  * You should have received a copy of the GNU Lesser General Public
20  * License along with FFmpeg; if not, write to the Free Software
21  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
22  */
23 
24 #include <string.h>
25 
26 #include "libavutil/avassert.h"
27 #include "libavutil/internal.h"
28 #include "avcodec.h"
29 #include "dsputil.h"
30 #include "h261.h"
31 #include "mpegvideo.h"
32 #include "mjpegenc.h"
33 #include "msmpeg4.h"
34 #include <limits.h>
35 
36 static void gmc1_motion(MpegEncContext *s,
37  uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
38  uint8_t **ref_picture)
39 {
40  uint8_t *ptr;
41  int src_x, src_y, motion_x, motion_y;
42  ptrdiff_t offset, linesize, uvlinesize;
43  int emu = 0;
44 
45  motion_x = s->sprite_offset[0][0];
46  motion_y = s->sprite_offset[0][1];
47  src_x = s->mb_x * 16 + (motion_x >> (s->sprite_warping_accuracy + 1));
48  src_y = s->mb_y * 16 + (motion_y >> (s->sprite_warping_accuracy + 1));
49  motion_x <<= (3 - s->sprite_warping_accuracy);
50  motion_y <<= (3 - s->sprite_warping_accuracy);
51  src_x = av_clip(src_x, -16, s->width);
52  if (src_x == s->width)
53  motion_x = 0;
54  src_y = av_clip(src_y, -16, s->height);
55  if (src_y == s->height)
56  motion_y = 0;
57 
58  linesize = s->linesize;
59  uvlinesize = s->uvlinesize;
60 
61  ptr = ref_picture[0] + src_y * linesize + src_x;
62 
63  if ((unsigned)src_x >= FFMAX(s->h_edge_pos - 17, 0) ||
64  (unsigned)src_y >= FFMAX(s->v_edge_pos - 17, 0)) {
65  s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr,
66  linesize, linesize,
67  17, 17,
68  src_x, src_y,
69  s->h_edge_pos, s->v_edge_pos);
70  ptr = s->edge_emu_buffer;
71  }
72 
73  if ((motion_x | motion_y) & 7) {
74  s->dsp.gmc1(dest_y, ptr, linesize, 16,
75  motion_x & 15, motion_y & 15, 128 - s->no_rounding);
76  s->dsp.gmc1(dest_y + 8, ptr + 8, linesize, 16,
77  motion_x & 15, motion_y & 15, 128 - s->no_rounding);
78  } else {
79  int dxy;
80 
81  dxy = ((motion_x >> 3) & 1) | ((motion_y >> 2) & 2);
82  if (s->no_rounding) {
83  s->hdsp.put_no_rnd_pixels_tab[0][dxy](dest_y, ptr, linesize, 16);
84  } else {
85  s->hdsp.put_pixels_tab[0][dxy](dest_y, ptr, linesize, 16);
86  }
87  }
88 
89  if (CONFIG_GRAY && s->flags & CODEC_FLAG_GRAY)
90  return;
91 
92  motion_x = s->sprite_offset[1][0];
93  motion_y = s->sprite_offset[1][1];
94  src_x = s->mb_x * 8 + (motion_x >> (s->sprite_warping_accuracy + 1));
95  src_y = s->mb_y * 8 + (motion_y >> (s->sprite_warping_accuracy + 1));
96  motion_x <<= (3 - s->sprite_warping_accuracy);
97  motion_y <<= (3 - s->sprite_warping_accuracy);
98  src_x = av_clip(src_x, -8, s->width >> 1);
99  if (src_x == s->width >> 1)
100  motion_x = 0;
101  src_y = av_clip(src_y, -8, s->height >> 1);
102  if (src_y == s->height >> 1)
103  motion_y = 0;
104 
105  offset = (src_y * uvlinesize) + src_x;
106  ptr = ref_picture[1] + offset;
107  if ((unsigned)src_x >= FFMAX((s->h_edge_pos >> 1) - 9, 0) ||
108  (unsigned)src_y >= FFMAX((s->v_edge_pos >> 1) - 9, 0)) {
109  s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr,
110  uvlinesize, uvlinesize,
111  9, 9,
112  src_x, src_y,
113  s->h_edge_pos >> 1, s->v_edge_pos >> 1);
114  ptr = s->edge_emu_buffer;
115  emu = 1;
116  }
117  s->dsp.gmc1(dest_cb, ptr, uvlinesize, 8,
118  motion_x & 15, motion_y & 15, 128 - s->no_rounding);
119 
120  ptr = ref_picture[2] + offset;
121  if (emu) {
122  s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr,
123  uvlinesize, uvlinesize,
124  9, 9,
125  src_x, src_y,
126  s->h_edge_pos >> 1, s->v_edge_pos >> 1);
127  ptr = s->edge_emu_buffer;
128  }
129  s->dsp.gmc1(dest_cr, ptr, uvlinesize, 8,
130  motion_x & 15, motion_y & 15, 128 - s->no_rounding);
131 }
132 
133 static void gmc_motion(MpegEncContext *s,
134  uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
135  uint8_t **ref_picture)
136 {
137  uint8_t *ptr;
138  int linesize, uvlinesize;
139  const int a = s->sprite_warping_accuracy;
140  int ox, oy;
141 
142  linesize = s->linesize;
143  uvlinesize = s->uvlinesize;
144 
145  ptr = ref_picture[0];
146 
147  ox = s->sprite_offset[0][0] + s->sprite_delta[0][0] * s->mb_x * 16 +
148  s->sprite_delta[0][1] * s->mb_y * 16;
149  oy = s->sprite_offset[0][1] + s->sprite_delta[1][0] * s->mb_x * 16 +
150  s->sprite_delta[1][1] * s->mb_y * 16;
151 
152  s->dsp.gmc(dest_y, ptr, linesize, 16,
153  ox, oy,
154  s->sprite_delta[0][0], s->sprite_delta[0][1],
155  s->sprite_delta[1][0], s->sprite_delta[1][1],
156  a + 1, (1 << (2 * a + 1)) - s->no_rounding,
157  s->h_edge_pos, s->v_edge_pos);
158  s->dsp.gmc(dest_y + 8, ptr, linesize, 16,
159  ox + s->sprite_delta[0][0] * 8,
160  oy + s->sprite_delta[1][0] * 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, s->v_edge_pos);
165 
166  if (CONFIG_GRAY && s->flags & CODEC_FLAG_GRAY)
167  return;
168 
169  ox = s->sprite_offset[1][0] + s->sprite_delta[0][0] * s->mb_x * 8 +
170  s->sprite_delta[0][1] * s->mb_y * 8;
171  oy = s->sprite_offset[1][1] + s->sprite_delta[1][0] * s->mb_x * 8 +
172  s->sprite_delta[1][1] * s->mb_y * 8;
173 
174  ptr = ref_picture[1];
175  s->dsp.gmc(dest_cb, ptr, uvlinesize, 8,
176  ox, oy,
177  s->sprite_delta[0][0], s->sprite_delta[0][1],
178  s->sprite_delta[1][0], s->sprite_delta[1][1],
179  a + 1, (1 << (2 * a + 1)) - s->no_rounding,
180  s->h_edge_pos >> 1, s->v_edge_pos >> 1);
181 
182  ptr = ref_picture[2];
183  s->dsp.gmc(dest_cr, ptr, uvlinesize, 8,
184  ox, oy,
185  s->sprite_delta[0][0], s->sprite_delta[0][1],
186  s->sprite_delta[1][0], s->sprite_delta[1][1],
187  a + 1, (1 << (2 * a + 1)) - s->no_rounding,
188  s->h_edge_pos >> 1, s->v_edge_pos >> 1);
189 }
190 
191 static inline int hpel_motion(MpegEncContext *s,
192  uint8_t *dest, uint8_t *src,
193  int src_x, int src_y,
194  op_pixels_func *pix_op,
195  int motion_x, int motion_y)
196 {
197  int dxy = 0;
198  int emu = 0;
199 
200  src_x += motion_x >> 1;
201  src_y += motion_y >> 1;
202 
203  /* WARNING: do no forget half pels */
204  src_x = av_clip(src_x, -16, s->width); // FIXME unneeded for emu?
205  if (src_x != s->width)
206  dxy |= motion_x & 1;
207  src_y = av_clip(src_y, -16, s->height);
208  if (src_y != s->height)
209  dxy |= (motion_y & 1) << 1;
210  src += src_y * s->linesize + src_x;
211 
212  if ((unsigned)src_x > FFMAX(s->h_edge_pos - (motion_x & 1) - 8, 0) ||
213  (unsigned)src_y > FFMAX(s->v_edge_pos - (motion_y & 1) - 8, 0)) {
214  s->vdsp.emulated_edge_mc(s->edge_emu_buffer, src,
215  s->linesize, s->linesize,
216  9, 9,
217  src_x, src_y,
218  s->h_edge_pos, s->v_edge_pos);
219  src = s->edge_emu_buffer;
220  emu = 1;
221  }
222  pix_op[dxy](dest, src, s->linesize, 8);
223  return emu;
224 }
225 
226 static av_always_inline
227 void mpeg_motion_internal(MpegEncContext *s,
228  uint8_t *dest_y,
229  uint8_t *dest_cb,
230  uint8_t *dest_cr,
231  int field_based,
232  int bottom_field,
233  int field_select,
234  uint8_t **ref_picture,
235  op_pixels_func (*pix_op)[4],
236  int motion_x,
237  int motion_y,
238  int h,
239  int is_mpeg12,
240  int mb_y)
241 {
242  uint8_t *ptr_y, *ptr_cb, *ptr_cr;
243  int dxy, uvdxy, mx, my, src_x, src_y,
244  uvsrc_x, uvsrc_y, v_edge_pos;
245  ptrdiff_t uvlinesize, linesize;
246 
247 #if 0
248  if (s->quarter_sample) {
249  motion_x >>= 1;
250  motion_y >>= 1;
251  }
252 #endif
253 
254  v_edge_pos = s->v_edge_pos >> field_based;
255  linesize = s->current_picture.f.linesize[0] << field_based;
256  uvlinesize = s->current_picture.f.linesize[1] << field_based;
257 
258  dxy = ((motion_y & 1) << 1) | (motion_x & 1);
259  src_x = s->mb_x * 16 + (motion_x >> 1);
260  src_y = (mb_y << (4 - field_based)) + (motion_y >> 1);
261 
262  if (!is_mpeg12 && s->out_format == FMT_H263) {
263  if ((s->workaround_bugs & FF_BUG_HPEL_CHROMA) && field_based) {
264  mx = (motion_x >> 1) | (motion_x & 1);
265  my = motion_y >> 1;
266  uvdxy = ((my & 1) << 1) | (mx & 1);
267  uvsrc_x = s->mb_x * 8 + (mx >> 1);
268  uvsrc_y = (mb_y << (3 - field_based)) + (my >> 1);
269  } else {
270  uvdxy = dxy | (motion_y & 2) | ((motion_x & 2) >> 1);
271  uvsrc_x = src_x >> 1;
272  uvsrc_y = src_y >> 1;
273  }
274  // Even chroma mv's are full pel in H261
275  } else if (!is_mpeg12 && s->out_format == FMT_H261) {
276  mx = motion_x / 4;
277  my = motion_y / 4;
278  uvdxy = 0;
279  uvsrc_x = s->mb_x * 8 + mx;
280  uvsrc_y = mb_y * 8 + my;
281  } else {
282  if (s->chroma_y_shift) {
283  mx = motion_x / 2;
284  my = motion_y / 2;
285  uvdxy = ((my & 1) << 1) | (mx & 1);
286  uvsrc_x = s->mb_x * 8 + (mx >> 1);
287  uvsrc_y = (mb_y << (3 - field_based)) + (my >> 1);
288  } else {
289  if (s->chroma_x_shift) {
290  // Chroma422
291  mx = motion_x / 2;
292  uvdxy = ((motion_y & 1) << 1) | (mx & 1);
293  uvsrc_x = s->mb_x * 8 + (mx >> 1);
294  uvsrc_y = src_y;
295  } else {
296  // Chroma444
297  uvdxy = dxy;
298  uvsrc_x = src_x;
299  uvsrc_y = src_y;
300  }
301  }
302  }
303 
304  ptr_y = ref_picture[0] + src_y * linesize + src_x;
305  ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x;
306  ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x;
307 
308  if ((unsigned)src_x > FFMAX(s->h_edge_pos - (motion_x & 1) - 16, 0) ||
309  (unsigned)src_y > FFMAX( v_edge_pos - (motion_y & 1) - h , 0)) {
310  if (is_mpeg12 ||
311  s->codec_id == AV_CODEC_ID_MPEG2VIDEO ||
312  s->codec_id == AV_CODEC_ID_MPEG1VIDEO) {
313  av_log(s->avctx, AV_LOG_DEBUG,
314  "MPEG motion vector out of boundary (%d %d)\n", src_x,
315  src_y);
316  return;
317  }
318  s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr_y,
319  s->linesize, s->linesize,
320  17, 17 + field_based,
321  src_x, src_y << field_based,
322  s->h_edge_pos, s->v_edge_pos);
323  ptr_y = s->edge_emu_buffer;
324  if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
325  uint8_t *uvbuf = s->edge_emu_buffer + 18 * s->linesize;
326  s->vdsp.emulated_edge_mc(uvbuf, ptr_cb,
327  s->uvlinesize, s->uvlinesize,
328  9, 9 + field_based,
329  uvsrc_x, uvsrc_y << field_based,
330  s->h_edge_pos >> 1, s->v_edge_pos >> 1);
331  s->vdsp.emulated_edge_mc(uvbuf + 16, ptr_cr,
332  s->uvlinesize, s->uvlinesize,
333  9, 9 + field_based,
334  uvsrc_x, uvsrc_y << field_based,
335  s->h_edge_pos >> 1, s->v_edge_pos >> 1);
336  ptr_cb = uvbuf;
337  ptr_cr = uvbuf + 16;
338  }
339  }
340 
341  /* FIXME use this for field pix too instead of the obnoxious hack which
342  * changes picture.data */
343  if (bottom_field) {
344  dest_y += s->linesize;
345  dest_cb += s->uvlinesize;
346  dest_cr += s->uvlinesize;
347  }
348 
349  if (field_select) {
350  ptr_y += s->linesize;
351  ptr_cb += s->uvlinesize;
352  ptr_cr += s->uvlinesize;
353  }
354 
355  pix_op[0][dxy](dest_y, ptr_y, linesize, h);
356 
357  if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
358  pix_op[s->chroma_x_shift][uvdxy]
359  (dest_cb, ptr_cb, uvlinesize, h >> s->chroma_y_shift);
360  pix_op[s->chroma_x_shift][uvdxy]
361  (dest_cr, ptr_cr, uvlinesize, h >> s->chroma_y_shift);
362  }
363  if (!is_mpeg12 && (CONFIG_H261_ENCODER || CONFIG_H261_DECODER) &&
364  s->out_format == FMT_H261) {
365  ff_h261_loop_filter(s);
366  }
367 }
368 /* apply one mpeg motion vector to the three components */
369 static void mpeg_motion(MpegEncContext *s,
370  uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
371  int field_select, uint8_t **ref_picture,
372  op_pixels_func (*pix_op)[4],
373  int motion_x, int motion_y, int h, int mb_y)
374 {
375 #if !CONFIG_SMALL
376  if (s->out_format == FMT_MPEG1)
377  mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, 0, 0,
378  field_select, ref_picture, pix_op,
379  motion_x, motion_y, h, 1, mb_y);
380  else
381 #endif
382  mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, 0, 0,
383  field_select, ref_picture, pix_op,
384  motion_x, motion_y, h, 0, mb_y);
385 }
386 
387 static void mpeg_motion_field(MpegEncContext *s, uint8_t *dest_y,
388  uint8_t *dest_cb, uint8_t *dest_cr,
389  int bottom_field, int field_select,
390  uint8_t **ref_picture,
391  op_pixels_func (*pix_op)[4],
392  int motion_x, int motion_y, int h, int mb_y)
393 {
394 #if !CONFIG_SMALL
395  if (s->out_format == FMT_MPEG1)
396  mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, 1,
397  bottom_field, field_select, ref_picture, pix_op,
398  motion_x, motion_y, h, 1, mb_y);
399  else
400 #endif
401  mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, 1,
402  bottom_field, field_select, ref_picture, pix_op,
403  motion_x, motion_y, h, 0, mb_y);
404 }
405 
406 // FIXME move to dsputil, avg variant, 16x16 version
407 static inline void put_obmc(uint8_t *dst, uint8_t *src[5], int stride)
408 {
409  int x;
410  uint8_t *const top = src[1];
411  uint8_t *const left = src[2];
412  uint8_t *const mid = src[0];
413  uint8_t *const right = src[3];
414  uint8_t *const bottom = src[4];
415 #define OBMC_FILTER(x, t, l, m, r, b)\
416  dst[x]= (t*top[x] + l*left[x] + m*mid[x] + r*right[x] + b*bottom[x] + 4)>>3
417 #define OBMC_FILTER4(x, t, l, m, r, b)\
418  OBMC_FILTER(x , t, l, m, r, b);\
419  OBMC_FILTER(x+1 , t, l, m, r, b);\
420  OBMC_FILTER(x +stride, t, l, m, r, b);\
421  OBMC_FILTER(x+1+stride, t, l, m, r, b);
422 
423  x = 0;
424  OBMC_FILTER (x , 2, 2, 4, 0, 0);
425  OBMC_FILTER (x + 1, 2, 1, 5, 0, 0);
426  OBMC_FILTER4(x + 2, 2, 1, 5, 0, 0);
427  OBMC_FILTER4(x + 4, 2, 0, 5, 1, 0);
428  OBMC_FILTER (x + 6, 2, 0, 5, 1, 0);
429  OBMC_FILTER (x + 7, 2, 0, 4, 2, 0);
430  x += stride;
431  OBMC_FILTER (x , 1, 2, 5, 0, 0);
432  OBMC_FILTER (x + 1, 1, 2, 5, 0, 0);
433  OBMC_FILTER (x + 6, 1, 0, 5, 2, 0);
434  OBMC_FILTER (x + 7, 1, 0, 5, 2, 0);
435  x += stride;
436  OBMC_FILTER4(x , 1, 2, 5, 0, 0);
437  OBMC_FILTER4(x + 2, 1, 1, 6, 0, 0);
438  OBMC_FILTER4(x + 4, 1, 0, 6, 1, 0);
439  OBMC_FILTER4(x + 6, 1, 0, 5, 2, 0);
440  x += 2 * stride;
441  OBMC_FILTER4(x , 0, 2, 5, 0, 1);
442  OBMC_FILTER4(x + 2, 0, 1, 6, 0, 1);
443  OBMC_FILTER4(x + 4, 0, 0, 6, 1, 1);
444  OBMC_FILTER4(x + 6, 0, 0, 5, 2, 1);
445  x += 2*stride;
446  OBMC_FILTER (x , 0, 2, 5, 0, 1);
447  OBMC_FILTER (x + 1, 0, 2, 5, 0, 1);
448  OBMC_FILTER4(x + 2, 0, 1, 5, 0, 2);
449  OBMC_FILTER4(x + 4, 0, 0, 5, 1, 2);
450  OBMC_FILTER (x + 6, 0, 0, 5, 2, 1);
451  OBMC_FILTER (x + 7, 0, 0, 5, 2, 1);
452  x += stride;
453  OBMC_FILTER (x , 0, 2, 4, 0, 2);
454  OBMC_FILTER (x + 1, 0, 1, 5, 0, 2);
455  OBMC_FILTER (x + 6, 0, 0, 5, 1, 2);
456  OBMC_FILTER (x + 7, 0, 0, 4, 2, 2);
457 }
458 
459 /* obmc for 1 8x8 luma block */
460 static inline void obmc_motion(MpegEncContext *s,
461  uint8_t *dest, uint8_t *src,
462  int src_x, int src_y,
463  op_pixels_func *pix_op,
464  int16_t mv[5][2] /* mid top left right bottom */)
465 #define MID 0
466 {
467  int i;
468  uint8_t *ptr[5];
469 
470  av_assert2(s->quarter_sample == 0);
471 
472  for (i = 0; i < 5; i++) {
473  if (i && mv[i][0] == mv[MID][0] && mv[i][1] == mv[MID][1]) {
474  ptr[i] = ptr[MID];
475  } else {
476  ptr[i] = s->obmc_scratchpad + 8 * (i & 1) +
477  s->linesize * 8 * (i >> 1);
478  hpel_motion(s, ptr[i], src, src_x, src_y, pix_op,
479  mv[i][0], mv[i][1]);
480  }
481  }
482 
483  put_obmc(dest, ptr, s->linesize);
484 }
485 
486 static inline void qpel_motion(MpegEncContext *s,
487  uint8_t *dest_y,
488  uint8_t *dest_cb,
489  uint8_t *dest_cr,
490  int field_based, int bottom_field,
491  int field_select, uint8_t **ref_picture,
492  op_pixels_func (*pix_op)[4],
493  qpel_mc_func (*qpix_op)[16],
494  int motion_x, int motion_y, int h)
495 {
496  uint8_t *ptr_y, *ptr_cb, *ptr_cr;
497  int dxy, uvdxy, mx, my, src_x, src_y, uvsrc_x, uvsrc_y, v_edge_pos;
498  ptrdiff_t linesize, uvlinesize;
499 
500  dxy = ((motion_y & 3) << 2) | (motion_x & 3);
501 
502  src_x = s->mb_x * 16 + (motion_x >> 2);
503  src_y = s->mb_y * (16 >> field_based) + (motion_y >> 2);
504 
505  v_edge_pos = s->v_edge_pos >> field_based;
506  linesize = s->linesize << field_based;
507  uvlinesize = s->uvlinesize << field_based;
508 
509  if (field_based) {
510  mx = motion_x / 2;
511  my = motion_y >> 1;
512  } else if (s->workaround_bugs & FF_BUG_QPEL_CHROMA2) {
513  static const int rtab[8] = { 0, 0, 1, 1, 0, 0, 0, 1 };
514  mx = (motion_x >> 1) + rtab[motion_x & 7];
515  my = (motion_y >> 1) + rtab[motion_y & 7];
516  } else if (s->workaround_bugs & FF_BUG_QPEL_CHROMA) {
517  mx = (motion_x >> 1) | (motion_x & 1);
518  my = (motion_y >> 1) | (motion_y & 1);
519  } else {
520  mx = motion_x / 2;
521  my = motion_y / 2;
522  }
523  mx = (mx >> 1) | (mx & 1);
524  my = (my >> 1) | (my & 1);
525 
526  uvdxy = (mx & 1) | ((my & 1) << 1);
527  mx >>= 1;
528  my >>= 1;
529 
530  uvsrc_x = s->mb_x * 8 + mx;
531  uvsrc_y = s->mb_y * (8 >> field_based) + my;
532 
533  ptr_y = ref_picture[0] + src_y * linesize + src_x;
534  ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x;
535  ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x;
536 
537  if ((unsigned)src_x > FFMAX(s->h_edge_pos - (motion_x & 3) - 16, 0) ||
538  (unsigned)src_y > FFMAX( v_edge_pos - (motion_y & 3) - h, 0)) {
539  s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr_y,
540  s->linesize, s->linesize,
541  17, 17 + field_based,
542  src_x, src_y << field_based,
543  s->h_edge_pos, s->v_edge_pos);
544  ptr_y = s->edge_emu_buffer;
545  if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
546  uint8_t *uvbuf = s->edge_emu_buffer + 18 * s->linesize;
547  s->vdsp.emulated_edge_mc(uvbuf, ptr_cb,
548  s->uvlinesize, s->uvlinesize,
549  9, 9 + field_based,
550  uvsrc_x, uvsrc_y << field_based,
551  s->h_edge_pos >> 1, s->v_edge_pos >> 1);
552  s->vdsp.emulated_edge_mc(uvbuf + 16, ptr_cr,
553  s->uvlinesize, s->uvlinesize,
554  9, 9 + field_based,
555  uvsrc_x, uvsrc_y << field_based,
556  s->h_edge_pos >> 1, s->v_edge_pos >> 1);
557  ptr_cb = uvbuf;
558  ptr_cr = uvbuf + 16;
559  }
560  }
561 
562  if (!field_based)
563  qpix_op[0][dxy](dest_y, ptr_y, linesize);
564  else {
565  if (bottom_field) {
566  dest_y += s->linesize;
567  dest_cb += s->uvlinesize;
568  dest_cr += s->uvlinesize;
569  }
570 
571  if (field_select) {
572  ptr_y += s->linesize;
573  ptr_cb += s->uvlinesize;
574  ptr_cr += s->uvlinesize;
575  }
576  // damn interlaced mode
577  // FIXME boundary mirroring is not exactly correct here
578  qpix_op[1][dxy](dest_y, ptr_y, linesize);
579  qpix_op[1][dxy](dest_y + 8, ptr_y + 8, linesize);
580  }
581  if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
582  pix_op[1][uvdxy](dest_cr, ptr_cr, uvlinesize, h >> 1);
583  pix_op[1][uvdxy](dest_cb, ptr_cb, uvlinesize, h >> 1);
584  }
585 }
586 
587 /**
588  * h263 chroma 4mv motion compensation.
589  */
590 static void chroma_4mv_motion(MpegEncContext *s,
591  uint8_t *dest_cb, uint8_t *dest_cr,
592  uint8_t **ref_picture,
593  op_pixels_func *pix_op,
594  int mx, int my)
595 {
596  uint8_t *ptr;
597  int src_x, src_y, dxy, emu = 0;
598  ptrdiff_t offset;
599 
600  /* In case of 8X8, we construct a single chroma motion vector
601  * with a special rounding */
602  mx = ff_h263_round_chroma(mx);
603  my = ff_h263_round_chroma(my);
604 
605  dxy = ((my & 1) << 1) | (mx & 1);
606  mx >>= 1;
607  my >>= 1;
608 
609  src_x = s->mb_x * 8 + mx;
610  src_y = s->mb_y * 8 + my;
611  src_x = av_clip(src_x, -8, (s->width >> 1));
612  if (src_x == (s->width >> 1))
613  dxy &= ~1;
614  src_y = av_clip(src_y, -8, (s->height >> 1));
615  if (src_y == (s->height >> 1))
616  dxy &= ~2;
617 
618  offset = src_y * s->uvlinesize + src_x;
619  ptr = ref_picture[1] + offset;
620  if ((unsigned)src_x > FFMAX((s->h_edge_pos >> 1) - (dxy & 1) - 8, 0) ||
621  (unsigned)src_y > FFMAX((s->v_edge_pos >> 1) - (dxy >> 1) - 8, 0)) {
622  s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr,
623  s->uvlinesize, s->uvlinesize,
624  9, 9, src_x, src_y,
625  s->h_edge_pos >> 1, s->v_edge_pos >> 1);
626  ptr = s->edge_emu_buffer;
627  emu = 1;
628  }
629  pix_op[dxy](dest_cb, ptr, s->uvlinesize, 8);
630 
631  ptr = ref_picture[2] + offset;
632  if (emu) {
633  s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr,
634  s->uvlinesize, s->uvlinesize,
635  9, 9, src_x, src_y,
636  s->h_edge_pos >> 1, s->v_edge_pos >> 1);
637  ptr = s->edge_emu_buffer;
638  }
639  pix_op[dxy](dest_cr, ptr, s->uvlinesize, 8);
640 }
641 
642 static inline void prefetch_motion(MpegEncContext *s, uint8_t **pix, int dir)
643 {
644  /* fetch pixels for estimated mv 4 macroblocks ahead
645  * optimized for 64byte cache lines */
646  const int shift = s->quarter_sample ? 2 : 1;
647  const int mx = (s->mv[dir][0][0] >> shift) + 16 * s->mb_x + 8;
648  const int my = (s->mv[dir][0][1] >> shift) + 16 * s->mb_y;
649  int off = mx + (my + (s->mb_x & 3) * 4) * s->linesize + 64;
650 
651  s->vdsp.prefetch(pix[0] + off, s->linesize, 4);
652  off = (mx >> 1) + ((my >> 1) + (s->mb_x & 7)) * s->uvlinesize + 64;
653  s->vdsp.prefetch(pix[1] + off, pix[2] - pix[1], 2);
654 }
655 
656 static inline void apply_obmc(MpegEncContext *s,
657  uint8_t *dest_y,
658  uint8_t *dest_cb,
659  uint8_t *dest_cr,
660  uint8_t **ref_picture,
661  op_pixels_func (*pix_op)[4])
662 {
663  LOCAL_ALIGNED_8(int16_t, mv_cache, [4], [4][2]);
664  Picture *cur_frame = &s->current_picture;
665  int mb_x = s->mb_x;
666  int mb_y = s->mb_y;
667  const int xy = mb_x + mb_y * s->mb_stride;
668  const int mot_stride = s->b8_stride;
669  const int mot_xy = mb_x * 2 + mb_y * 2 * mot_stride;
670  int mx, my, i;
671 
672  av_assert2(!s->mb_skipped);
673 
674  AV_COPY32(mv_cache[1][1], cur_frame->motion_val[0][mot_xy]);
675  AV_COPY32(mv_cache[1][2], cur_frame->motion_val[0][mot_xy + 1]);
676 
677  AV_COPY32(mv_cache[2][1],
678  cur_frame->motion_val[0][mot_xy + mot_stride]);
679  AV_COPY32(mv_cache[2][2],
680  cur_frame->motion_val[0][mot_xy + mot_stride + 1]);
681 
682  AV_COPY32(mv_cache[3][1],
683  cur_frame->motion_val[0][mot_xy + mot_stride]);
684  AV_COPY32(mv_cache[3][2],
685  cur_frame->motion_val[0][mot_xy + mot_stride + 1]);
686 
687  if (mb_y == 0 || IS_INTRA(cur_frame->mb_type[xy - s->mb_stride])) {
688  AV_COPY32(mv_cache[0][1], mv_cache[1][1]);
689  AV_COPY32(mv_cache[0][2], mv_cache[1][2]);
690  } else {
691  AV_COPY32(mv_cache[0][1],
692  cur_frame->motion_val[0][mot_xy - mot_stride]);
693  AV_COPY32(mv_cache[0][2],
694  cur_frame->motion_val[0][mot_xy - mot_stride + 1]);
695  }
696 
697  if (mb_x == 0 || IS_INTRA(cur_frame->mb_type[xy - 1])) {
698  AV_COPY32(mv_cache[1][0], mv_cache[1][1]);
699  AV_COPY32(mv_cache[2][0], mv_cache[2][1]);
700  } else {
701  AV_COPY32(mv_cache[1][0], cur_frame->motion_val[0][mot_xy - 1]);
702  AV_COPY32(mv_cache[2][0],
703  cur_frame->motion_val[0][mot_xy - 1 + mot_stride]);
704  }
705 
706  if (mb_x + 1 >= s->mb_width || IS_INTRA(cur_frame->mb_type[xy + 1])) {
707  AV_COPY32(mv_cache[1][3], mv_cache[1][2]);
708  AV_COPY32(mv_cache[2][3], mv_cache[2][2]);
709  } else {
710  AV_COPY32(mv_cache[1][3], cur_frame->motion_val[0][mot_xy + 2]);
711  AV_COPY32(mv_cache[2][3],
712  cur_frame->motion_val[0][mot_xy + 2 + mot_stride]);
713  }
714 
715  mx = 0;
716  my = 0;
717  for (i = 0; i < 4; i++) {
718  const int x = (i & 1) + 1;
719  const int y = (i >> 1) + 1;
720  int16_t mv[5][2] = {
721  { mv_cache[y][x][0], mv_cache[y][x][1] },
722  { mv_cache[y - 1][x][0], mv_cache[y - 1][x][1] },
723  { mv_cache[y][x - 1][0], mv_cache[y][x - 1][1] },
724  { mv_cache[y][x + 1][0], mv_cache[y][x + 1][1] },
725  { mv_cache[y + 1][x][0], mv_cache[y + 1][x][1] }
726  };
727  // FIXME cleanup
728  obmc_motion(s, dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize,
729  ref_picture[0],
730  mb_x * 16 + (i & 1) * 8, mb_y * 16 + (i >> 1) * 8,
731  pix_op[1],
732  mv);
733 
734  mx += mv[0][0];
735  my += mv[0][1];
736  }
737  if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY))
738  chroma_4mv_motion(s, dest_cb, dest_cr,
739  ref_picture, pix_op[1],
740  mx, my);
741 }
742 
743 static inline void apply_8x8(MpegEncContext *s,
744  uint8_t *dest_y,
745  uint8_t *dest_cb,
746  uint8_t *dest_cr,
747  int dir,
748  uint8_t **ref_picture,
749  qpel_mc_func (*qpix_op)[16],
750  op_pixels_func (*pix_op)[4])
751 {
752  int dxy, mx, my, src_x, src_y;
753  int i;
754  int mb_x = s->mb_x;
755  int mb_y = s->mb_y;
756  uint8_t *ptr, *dest;
757 
758  mx = 0;
759  my = 0;
760  if (s->quarter_sample) {
761  for (i = 0; i < 4; i++) {
762  int motion_x = s->mv[dir][i][0];
763  int motion_y = s->mv[dir][i][1];
764 
765  dxy = ((motion_y & 3) << 2) | (motion_x & 3);
766  src_x = mb_x * 16 + (motion_x >> 2) + (i & 1) * 8;
767  src_y = mb_y * 16 + (motion_y >> 2) + (i >> 1) * 8;
768 
769  /* WARNING: do no forget half pels */
770  src_x = av_clip(src_x, -16, s->width);
771  if (src_x == s->width)
772  dxy &= ~3;
773  src_y = av_clip(src_y, -16, s->height);
774  if (src_y == s->height)
775  dxy &= ~12;
776 
777  ptr = ref_picture[0] + (src_y * s->linesize) + (src_x);
778  if ((unsigned)src_x > FFMAX(s->h_edge_pos - (motion_x & 3) - 8, 0) ||
779  (unsigned)src_y > FFMAX(s->v_edge_pos - (motion_y & 3) - 8, 0)) {
780  s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr,
781  s->linesize, s->linesize,
782  9, 9,
783  src_x, src_y,
784  s->h_edge_pos,
785  s->v_edge_pos);
786  ptr = s->edge_emu_buffer;
787  }
788  dest = dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize;
789  qpix_op[1][dxy](dest, ptr, s->linesize);
790 
791  mx += s->mv[dir][i][0] / 2;
792  my += s->mv[dir][i][1] / 2;
793  }
794  } else {
795  for (i = 0; i < 4; i++) {
796  hpel_motion(s,
797  dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize,
798  ref_picture[0],
799  mb_x * 16 + (i & 1) * 8,
800  mb_y * 16 + (i >> 1) * 8,
801  pix_op[1],
802  s->mv[dir][i][0],
803  s->mv[dir][i][1]);
804 
805  mx += s->mv[dir][i][0];
806  my += s->mv[dir][i][1];
807  }
808  }
809 
810  if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY))
811  chroma_4mv_motion(s, dest_cb, dest_cr,
812  ref_picture, pix_op[1], mx, my);
813 }
814 
815 /**
816  * motion compensation of a single macroblock
817  * @param s context
818  * @param dest_y luma destination pointer
819  * @param dest_cb chroma cb/u destination pointer
820  * @param dest_cr chroma cr/v destination pointer
821  * @param dir direction (0->forward, 1->backward)
822  * @param ref_picture array[3] of pointers to the 3 planes of the reference picture
823  * @param pix_op halfpel motion compensation function (average or put normally)
824  * @param qpix_op qpel motion compensation function (average or put normally)
825  * the motion vectors are taken from s->mv and the MV type from s->mv_type
826  */
827 static av_always_inline void MPV_motion_internal(MpegEncContext *s,
828  uint8_t *dest_y,
829  uint8_t *dest_cb,
830  uint8_t *dest_cr,
831  int dir,
832  uint8_t **ref_picture,
833  op_pixels_func (*pix_op)[4],
834  qpel_mc_func (*qpix_op)[16],
835  int is_mpeg12)
836 {
837  int i;
838  int mb_y = s->mb_y;
839 
840  prefetch_motion(s, ref_picture, dir);
841 
842  if (!is_mpeg12 && s->obmc && s->pict_type != AV_PICTURE_TYPE_B) {
843  apply_obmc(s, dest_y, dest_cb, dest_cr, ref_picture, pix_op);
844  return;
845  }
846 
847  switch (s->mv_type) {
848  case MV_TYPE_16X16:
849  if (s->mcsel) {
850  if (s->real_sprite_warping_points == 1) {
851  gmc1_motion(s, dest_y, dest_cb, dest_cr,
852  ref_picture);
853  } else {
854  gmc_motion(s, dest_y, dest_cb, dest_cr,
855  ref_picture);
856  }
857  } else if (!is_mpeg12 && s->quarter_sample) {
858  qpel_motion(s, dest_y, dest_cb, dest_cr,
859  0, 0, 0,
860  ref_picture, pix_op, qpix_op,
861  s->mv[dir][0][0], s->mv[dir][0][1], 16);
862  } else if (!is_mpeg12 && (CONFIG_WMV2_DECODER || CONFIG_WMV2_ENCODER) &&
863  s->mspel && s->codec_id == AV_CODEC_ID_WMV2) {
864  ff_mspel_motion(s, dest_y, dest_cb, dest_cr,
865  ref_picture, pix_op,
866  s->mv[dir][0][0], s->mv[dir][0][1], 16);
867  } else {
868  mpeg_motion(s, dest_y, dest_cb, dest_cr, 0,
869  ref_picture, pix_op,
870  s->mv[dir][0][0], s->mv[dir][0][1], 16, mb_y);
871  }
872  break;
873  case MV_TYPE_8X8:
874  if (!is_mpeg12)
875  apply_8x8(s, dest_y, dest_cb, dest_cr,
876  dir, ref_picture, qpix_op, pix_op);
877  break;
878  case MV_TYPE_FIELD:
879  if (s->picture_structure == PICT_FRAME) {
880  if (!is_mpeg12 && s->quarter_sample) {
881  for (i = 0; i < 2; i++)
882  qpel_motion(s, dest_y, dest_cb, dest_cr,
883  1, i, s->field_select[dir][i],
884  ref_picture, pix_op, qpix_op,
885  s->mv[dir][i][0], s->mv[dir][i][1], 8);
886  } else {
887  /* top field */
888  mpeg_motion_field(s, dest_y, dest_cb, dest_cr,
889  0, s->field_select[dir][0],
890  ref_picture, pix_op,
891  s->mv[dir][0][0], s->mv[dir][0][1], 8, mb_y);
892  /* bottom field */
893  mpeg_motion_field(s, dest_y, dest_cb, dest_cr,
894  1, s->field_select[dir][1],
895  ref_picture, pix_op,
896  s->mv[dir][1][0], s->mv[dir][1][1], 8, mb_y);
897  }
898  } else {
899  if ( s->picture_structure != s->field_select[dir][0] + 1 && s->pict_type != AV_PICTURE_TYPE_B && !s->first_field
900  || !ref_picture[0]) {
901  ref_picture = s->current_picture_ptr->f.data;
902  }
903 
904  mpeg_motion(s, dest_y, dest_cb, dest_cr,
905  s->field_select[dir][0],
906  ref_picture, pix_op,
907  s->mv[dir][0][0], s->mv[dir][0][1], 16, mb_y >> 1);
908  }
909  break;
910  case MV_TYPE_16X8:
911  for (i = 0; i < 2; i++) {
912  uint8_t **ref2picture;
913 
914  if ((s->picture_structure == s->field_select[dir][i] + 1
915  || s->pict_type == AV_PICTURE_TYPE_B || s->first_field) && ref_picture[0]) {
916  ref2picture = ref_picture;
917  } else {
918  ref2picture = s->current_picture_ptr->f.data;
919  }
920 
921  mpeg_motion(s, dest_y, dest_cb, dest_cr,
922  s->field_select[dir][i],
923  ref2picture, pix_op,
924  s->mv[dir][i][0], s->mv[dir][i][1] + 16 * i,
925  8, mb_y >> 1);
926 
927  dest_y += 16 * s->linesize;
928  dest_cb += (16 >> s->chroma_y_shift) * s->uvlinesize;
929  dest_cr += (16 >> s->chroma_y_shift) * s->uvlinesize;
930  }
931  break;
932  case MV_TYPE_DMV:
933  if (s->picture_structure == PICT_FRAME) {
934  for (i = 0; i < 2; i++) {
935  int j;
936  for (j = 0; j < 2; j++)
937  mpeg_motion_field(s, dest_y, dest_cb, dest_cr,
938  j, j ^ i, ref_picture, pix_op,
939  s->mv[dir][2 * i + j][0],
940  s->mv[dir][2 * i + j][1], 8, mb_y);
941  pix_op = s->hdsp.avg_pixels_tab;
942  }
943  } else {
944  if (!ref_picture[0]) {
945  ref_picture = s->current_picture_ptr->f.data;
946  }
947  for (i = 0; i < 2; i++) {
948  mpeg_motion(s, dest_y, dest_cb, dest_cr,
949  s->picture_structure != i + 1,
950  ref_picture, pix_op,
951  s->mv[dir][2 * i][0], s->mv[dir][2 * i][1],
952  16, mb_y >> 1);
953 
954  // after put we make avg of the same block
955  pix_op = s->hdsp.avg_pixels_tab;
956 
957  /* opposite parity is always in the same frame if this is
958  * second field */
959  if (!s->first_field) {
960  ref_picture = s->current_picture_ptr->f.data;
961  }
962  }
963  }
964  break;
965  default: av_assert2(0);
966  }
967 }
968 
969 void ff_MPV_motion(MpegEncContext *s,
970  uint8_t *dest_y, uint8_t *dest_cb,
971  uint8_t *dest_cr, int dir,
972  uint8_t **ref_picture,
973  op_pixels_func (*pix_op)[4],
974  qpel_mc_func (*qpix_op)[16])
975 {
976 #if !CONFIG_SMALL
977  if (s->out_format == FMT_MPEG1)
978  MPV_motion_internal(s, dest_y, dest_cb, dest_cr, dir,
979  ref_picture, pix_op, qpix_op, 1);
980  else
981 #endif
982  MPV_motion_internal(s, dest_y, dest_cb, dest_cr, dir,
983  ref_picture, pix_op, qpix_op, 0);
984 }
Generated on Sun Mar 23 2014 23:50:00 for FFmpeg by   doxygen 1.8.2