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 (s->flags & CODEC_FLAG_EMU_EDGE) {
213  if ((unsigned)src_x > FFMAX(s->h_edge_pos - (motion_x & 1) - 8, 0) ||
214  (unsigned)src_y > FFMAX(s->v_edge_pos - (motion_y & 1) - 8, 0)) {
215  s->vdsp.emulated_edge_mc(s->edge_emu_buffer, src,
216  s->linesize, s->linesize,
217  9, 9,
218  src_x, src_y,
219  s->h_edge_pos, s->v_edge_pos);
220  src = s->edge_emu_buffer;
221  emu = 1;
222  }
223  }
224  pix_op[dxy](dest, src, s->linesize, 8);
225  return emu;
226 }
227 
228 static av_always_inline
229 void mpeg_motion_internal(MpegEncContext *s,
230  uint8_t *dest_y,
231  uint8_t *dest_cb,
232  uint8_t *dest_cr,
233  int field_based,
234  int bottom_field,
235  int field_select,
236  uint8_t **ref_picture,
237  op_pixels_func (*pix_op)[4],
238  int motion_x,
239  int motion_y,
240  int h,
241  int is_mpeg12,
242  int mb_y)
243 {
244  uint8_t *ptr_y, *ptr_cb, *ptr_cr;
245  int dxy, uvdxy, mx, my, src_x, src_y,
246  uvsrc_x, uvsrc_y, v_edge_pos;
247  ptrdiff_t uvlinesize, linesize;
248 
249 #if 0
250  if (s->quarter_sample) {
251  motion_x >>= 1;
252  motion_y >>= 1;
253  }
254 #endif
255 
256  v_edge_pos = s->v_edge_pos >> field_based;
257  linesize = s->current_picture.f.linesize[0] << field_based;
258  uvlinesize = s->current_picture.f.linesize[1] << field_based;
259 
260  dxy = ((motion_y & 1) << 1) | (motion_x & 1);
261  src_x = s->mb_x * 16 + (motion_x >> 1);
262  src_y = (mb_y << (4 - field_based)) + (motion_y >> 1);
263 
264  if (!is_mpeg12 && s->out_format == FMT_H263) {
265  if ((s->workaround_bugs & FF_BUG_HPEL_CHROMA) && field_based) {
266  mx = (motion_x >> 1) | (motion_x & 1);
267  my = motion_y >> 1;
268  uvdxy = ((my & 1) << 1) | (mx & 1);
269  uvsrc_x = s->mb_x * 8 + (mx >> 1);
270  uvsrc_y = (mb_y << (3 - field_based)) + (my >> 1);
271  } else {
272  uvdxy = dxy | (motion_y & 2) | ((motion_x & 2) >> 1);
273  uvsrc_x = src_x >> 1;
274  uvsrc_y = src_y >> 1;
275  }
276  // Even chroma mv's are full pel in H261
277  } else if (!is_mpeg12 && s->out_format == FMT_H261) {
278  mx = motion_x / 4;
279  my = motion_y / 4;
280  uvdxy = 0;
281  uvsrc_x = s->mb_x * 8 + mx;
282  uvsrc_y = mb_y * 8 + my;
283  } else {
284  if (s->chroma_y_shift) {
285  mx = motion_x / 2;
286  my = motion_y / 2;
287  uvdxy = ((my & 1) << 1) | (mx & 1);
288  uvsrc_x = s->mb_x * 8 + (mx >> 1);
289  uvsrc_y = (mb_y << (3 - field_based)) + (my >> 1);
290  } else {
291  if (s->chroma_x_shift) {
292  // Chroma422
293  mx = motion_x / 2;
294  uvdxy = ((motion_y & 1) << 1) | (mx & 1);
295  uvsrc_x = s->mb_x * 8 + (mx >> 1);
296  uvsrc_y = src_y;
297  } else {
298  // Chroma444
299  uvdxy = dxy;
300  uvsrc_x = src_x;
301  uvsrc_y = src_y;
302  }
303  }
304  }
305 
306  ptr_y = ref_picture[0] + src_y * linesize + src_x;
307  ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x;
308  ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x;
309 
310  if ((unsigned)src_x > FFMAX(s->h_edge_pos - (motion_x & 1) - 16, 0) ||
311  (unsigned)src_y > FFMAX( v_edge_pos - (motion_y & 1) - h , 0)) {
312  if (is_mpeg12 ||
313  s->codec_id == AV_CODEC_ID_MPEG2VIDEO ||
314  s->codec_id == AV_CODEC_ID_MPEG1VIDEO) {
315  av_log(s->avctx, AV_LOG_DEBUG,
316  "MPEG motion vector out of boundary (%d %d)\n", src_x,
317  src_y);
318  return;
319  }
320  s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr_y,
321  s->linesize, s->linesize,
322  17, 17 + field_based,
323  src_x, src_y << field_based,
324  s->h_edge_pos, s->v_edge_pos);
325  ptr_y = s->edge_emu_buffer;
326  if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
327  uint8_t *uvbuf = s->edge_emu_buffer + 18 * s->linesize;
328  s->vdsp.emulated_edge_mc(uvbuf, ptr_cb,
329  s->uvlinesize, s->uvlinesize,
330  9, 9 + field_based,
331  uvsrc_x, uvsrc_y << field_based,
332  s->h_edge_pos >> 1, s->v_edge_pos >> 1);
333  s->vdsp.emulated_edge_mc(uvbuf + 16, ptr_cr,
334  s->uvlinesize, s->uvlinesize,
335  9, 9 + field_based,
336  uvsrc_x, uvsrc_y << field_based,
337  s->h_edge_pos >> 1, s->v_edge_pos >> 1);
338  ptr_cb = uvbuf;
339  ptr_cr = uvbuf + 16;
340  }
341  }
342 
343  /* FIXME use this for field pix too instead of the obnoxious hack which
344  * changes picture.data */
345  if (bottom_field) {
346  dest_y += s->linesize;
347  dest_cb += s->uvlinesize;
348  dest_cr += s->uvlinesize;
349  }
350 
351  if (field_select) {
352  ptr_y += s->linesize;
353  ptr_cb += s->uvlinesize;
354  ptr_cr += s->uvlinesize;
355  }
356 
357  pix_op[0][dxy](dest_y, ptr_y, linesize, h);
358 
359  if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
360  pix_op[s->chroma_x_shift][uvdxy]
361  (dest_cb, ptr_cb, uvlinesize, h >> s->chroma_y_shift);
362  pix_op[s->chroma_x_shift][uvdxy]
363  (dest_cr, ptr_cr, uvlinesize, h >> s->chroma_y_shift);
364  }
365  if (!is_mpeg12 && (CONFIG_H261_ENCODER || CONFIG_H261_DECODER) &&
366  s->out_format == FMT_H261) {
367  ff_h261_loop_filter(s);
368  }
369 }
370 /* apply one mpeg motion vector to the three components */
371 static void mpeg_motion(MpegEncContext *s,
372  uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
373  int field_select, uint8_t **ref_picture,
374  op_pixels_func (*pix_op)[4],
375  int motion_x, int motion_y, int h, int mb_y)
376 {
377 #if !CONFIG_SMALL
378  if (s->out_format == FMT_MPEG1)
379  mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, 0, 0,
380  field_select, ref_picture, pix_op,
381  motion_x, motion_y, h, 1, mb_y);
382  else
383 #endif
384  mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, 0, 0,
385  field_select, ref_picture, pix_op,
386  motion_x, motion_y, h, 0, mb_y);
387 }
388 
389 static void mpeg_motion_field(MpegEncContext *s, uint8_t *dest_y,
390  uint8_t *dest_cb, uint8_t *dest_cr,
391  int bottom_field, int field_select,
392  uint8_t **ref_picture,
393  op_pixels_func (*pix_op)[4],
394  int motion_x, int motion_y, int h, int mb_y)
395 {
396 #if !CONFIG_SMALL
397  if (s->out_format == FMT_MPEG1)
398  mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, 1,
399  bottom_field, field_select, ref_picture, pix_op,
400  motion_x, motion_y, h, 1, mb_y);
401  else
402 #endif
403  mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, 1,
404  bottom_field, field_select, ref_picture, pix_op,
405  motion_x, motion_y, h, 0, mb_y);
406 }
407 
408 // FIXME move to dsputil, avg variant, 16x16 version
409 static inline void put_obmc(uint8_t *dst, uint8_t *src[5], int stride)
410 {
411  int x;
412  uint8_t *const top = src[1];
413  uint8_t *const left = src[2];
414  uint8_t *const mid = src[0];
415  uint8_t *const right = src[3];
416  uint8_t *const bottom = src[4];
417 #define OBMC_FILTER(x, t, l, m, r, b)\
418  dst[x]= (t*top[x] + l*left[x] + m*mid[x] + r*right[x] + b*bottom[x] + 4)>>3
419 #define OBMC_FILTER4(x, t, l, m, r, b)\
420  OBMC_FILTER(x , t, l, m, r, b);\
421  OBMC_FILTER(x+1 , t, l, m, r, b);\
422  OBMC_FILTER(x +stride, t, l, m, r, b);\
423  OBMC_FILTER(x+1+stride, t, l, m, r, b);
424 
425  x = 0;
426  OBMC_FILTER (x , 2, 2, 4, 0, 0);
427  OBMC_FILTER (x + 1, 2, 1, 5, 0, 0);
428  OBMC_FILTER4(x + 2, 2, 1, 5, 0, 0);
429  OBMC_FILTER4(x + 4, 2, 0, 5, 1, 0);
430  OBMC_FILTER (x + 6, 2, 0, 5, 1, 0);
431  OBMC_FILTER (x + 7, 2, 0, 4, 2, 0);
432  x += stride;
433  OBMC_FILTER (x , 1, 2, 5, 0, 0);
434  OBMC_FILTER (x + 1, 1, 2, 5, 0, 0);
435  OBMC_FILTER (x + 6, 1, 0, 5, 2, 0);
436  OBMC_FILTER (x + 7, 1, 0, 5, 2, 0);
437  x += stride;
438  OBMC_FILTER4(x , 1, 2, 5, 0, 0);
439  OBMC_FILTER4(x + 2, 1, 1, 6, 0, 0);
440  OBMC_FILTER4(x + 4, 1, 0, 6, 1, 0);
441  OBMC_FILTER4(x + 6, 1, 0, 5, 2, 0);
442  x += 2 * stride;
443  OBMC_FILTER4(x , 0, 2, 5, 0, 1);
444  OBMC_FILTER4(x + 2, 0, 1, 6, 0, 1);
445  OBMC_FILTER4(x + 4, 0, 0, 6, 1, 1);
446  OBMC_FILTER4(x + 6, 0, 0, 5, 2, 1);
447  x += 2*stride;
448  OBMC_FILTER (x , 0, 2, 5, 0, 1);
449  OBMC_FILTER (x + 1, 0, 2, 5, 0, 1);
450  OBMC_FILTER4(x + 2, 0, 1, 5, 0, 2);
451  OBMC_FILTER4(x + 4, 0, 0, 5, 1, 2);
452  OBMC_FILTER (x + 6, 0, 0, 5, 2, 1);
453  OBMC_FILTER (x + 7, 0, 0, 5, 2, 1);
454  x += stride;
455  OBMC_FILTER (x , 0, 2, 4, 0, 2);
456  OBMC_FILTER (x + 1, 0, 1, 5, 0, 2);
457  OBMC_FILTER (x + 6, 0, 0, 5, 1, 2);
458  OBMC_FILTER (x + 7, 0, 0, 4, 2, 2);
459 }
460 
461 /* obmc for 1 8x8 luma block */
462 static inline void obmc_motion(MpegEncContext *s,
463  uint8_t *dest, uint8_t *src,
464  int src_x, int src_y,
465  op_pixels_func *pix_op,
466  int16_t mv[5][2] /* mid top left right bottom */)
467 #define MID 0
468 {
469  int i;
470  uint8_t *ptr[5];
471 
472  av_assert2(s->quarter_sample == 0);
473 
474  for (i = 0; i < 5; i++) {
475  if (i && mv[i][0] == mv[MID][0] && mv[i][1] == mv[MID][1]) {
476  ptr[i] = ptr[MID];
477  } else {
478  ptr[i] = s->obmc_scratchpad + 8 * (i & 1) +
479  s->linesize * 8 * (i >> 1);
480  hpel_motion(s, ptr[i], src, src_x, src_y, pix_op,
481  mv[i][0], mv[i][1]);
482  }
483  }
484 
485  put_obmc(dest, ptr, s->linesize);
486 }
487 
488 static inline void qpel_motion(MpegEncContext *s,
489  uint8_t *dest_y,
490  uint8_t *dest_cb,
491  uint8_t *dest_cr,
492  int field_based, int bottom_field,
493  int field_select, uint8_t **ref_picture,
494  op_pixels_func (*pix_op)[4],
495  qpel_mc_func (*qpix_op)[16],
496  int motion_x, int motion_y, int h)
497 {
498  uint8_t *ptr_y, *ptr_cb, *ptr_cr;
499  int dxy, uvdxy, mx, my, src_x, src_y, uvsrc_x, uvsrc_y, v_edge_pos;
500  ptrdiff_t linesize, uvlinesize;
501 
502  dxy = ((motion_y & 3) << 2) | (motion_x & 3);
503 
504  src_x = s->mb_x * 16 + (motion_x >> 2);
505  src_y = s->mb_y * (16 >> field_based) + (motion_y >> 2);
506 
507  v_edge_pos = s->v_edge_pos >> field_based;
508  linesize = s->linesize << field_based;
509  uvlinesize = s->uvlinesize << field_based;
510 
511  if (field_based) {
512  mx = motion_x / 2;
513  my = motion_y >> 1;
514  } else if (s->workaround_bugs & FF_BUG_QPEL_CHROMA2) {
515  static const int rtab[8] = { 0, 0, 1, 1, 0, 0, 0, 1 };
516  mx = (motion_x >> 1) + rtab[motion_x & 7];
517  my = (motion_y >> 1) + rtab[motion_y & 7];
518  } else if (s->workaround_bugs & FF_BUG_QPEL_CHROMA) {
519  mx = (motion_x >> 1) | (motion_x & 1);
520  my = (motion_y >> 1) | (motion_y & 1);
521  } else {
522  mx = motion_x / 2;
523  my = motion_y / 2;
524  }
525  mx = (mx >> 1) | (mx & 1);
526  my = (my >> 1) | (my & 1);
527 
528  uvdxy = (mx & 1) | ((my & 1) << 1);
529  mx >>= 1;
530  my >>= 1;
531 
532  uvsrc_x = s->mb_x * 8 + mx;
533  uvsrc_y = s->mb_y * (8 >> field_based) + my;
534 
535  ptr_y = ref_picture[0] + src_y * linesize + src_x;
536  ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x;
537  ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x;
538 
539  if ((unsigned)src_x > FFMAX(s->h_edge_pos - (motion_x & 3) - 16, 0) ||
540  (unsigned)src_y > FFMAX( v_edge_pos - (motion_y & 3) - h, 0)) {
541  s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr_y,
542  s->linesize, s->linesize,
543  17, 17 + field_based,
544  src_x, src_y << field_based,
545  s->h_edge_pos, s->v_edge_pos);
546  ptr_y = s->edge_emu_buffer;
547  if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
548  uint8_t *uvbuf = s->edge_emu_buffer + 18 * s->linesize;
549  s->vdsp.emulated_edge_mc(uvbuf, ptr_cb,
550  s->uvlinesize, s->uvlinesize,
551  9, 9 + field_based,
552  uvsrc_x, uvsrc_y << field_based,
553  s->h_edge_pos >> 1, s->v_edge_pos >> 1);
554  s->vdsp.emulated_edge_mc(uvbuf + 16, ptr_cr,
555  s->uvlinesize, s->uvlinesize,
556  9, 9 + field_based,
557  uvsrc_x, uvsrc_y << field_based,
558  s->h_edge_pos >> 1, s->v_edge_pos >> 1);
559  ptr_cb = uvbuf;
560  ptr_cr = uvbuf + 16;
561  }
562  }
563 
564  if (!field_based)
565  qpix_op[0][dxy](dest_y, ptr_y, linesize);
566  else {
567  if (bottom_field) {
568  dest_y += s->linesize;
569  dest_cb += s->uvlinesize;
570  dest_cr += s->uvlinesize;
571  }
572 
573  if (field_select) {
574  ptr_y += s->linesize;
575  ptr_cb += s->uvlinesize;
576  ptr_cr += s->uvlinesize;
577  }
578  // damn interlaced mode
579  // FIXME boundary mirroring is not exactly correct here
580  qpix_op[1][dxy](dest_y, ptr_y, linesize);
581  qpix_op[1][dxy](dest_y + 8, ptr_y + 8, linesize);
582  }
583  if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
584  pix_op[1][uvdxy](dest_cr, ptr_cr, uvlinesize, h >> 1);
585  pix_op[1][uvdxy](dest_cb, ptr_cb, uvlinesize, h >> 1);
586  }
587 }
588 
589 /**
590  * h263 chroma 4mv motion compensation.
591  */
592 static void chroma_4mv_motion(MpegEncContext *s,
593  uint8_t *dest_cb, uint8_t *dest_cr,
594  uint8_t **ref_picture,
595  op_pixels_func *pix_op,
596  int mx, int my)
597 {
598  uint8_t *ptr;
599  int src_x, src_y, dxy, emu = 0;
600  ptrdiff_t offset;
601 
602  /* In case of 8X8, we construct a single chroma motion vector
603  * with a special rounding */
604  mx = ff_h263_round_chroma(mx);
605  my = ff_h263_round_chroma(my);
606 
607  dxy = ((my & 1) << 1) | (mx & 1);
608  mx >>= 1;
609  my >>= 1;
610 
611  src_x = s->mb_x * 8 + mx;
612  src_y = s->mb_y * 8 + my;
613  src_x = av_clip(src_x, -8, (s->width >> 1));
614  if (src_x == (s->width >> 1))
615  dxy &= ~1;
616  src_y = av_clip(src_y, -8, (s->height >> 1));
617  if (src_y == (s->height >> 1))
618  dxy &= ~2;
619 
620  offset = src_y * s->uvlinesize + src_x;
621  ptr = ref_picture[1] + offset;
622  if (s->flags & CODEC_FLAG_EMU_EDGE) {
623  if ((unsigned)src_x > FFMAX((s->h_edge_pos >> 1) - (dxy & 1) - 8, 0) ||
624  (unsigned)src_y > FFMAX((s->v_edge_pos >> 1) - (dxy >> 1) - 8, 0)) {
625  s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr,
626  s->uvlinesize, s->uvlinesize,
627  9, 9, src_x, src_y,
628  s->h_edge_pos >> 1, s->v_edge_pos >> 1);
629  ptr = s->edge_emu_buffer;
630  emu = 1;
631  }
632  }
633  pix_op[dxy](dest_cb, ptr, s->uvlinesize, 8);
634 
635  ptr = ref_picture[2] + offset;
636  if (emu) {
637  s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr,
638  s->uvlinesize, s->uvlinesize,
639  9, 9, src_x, src_y,
640  s->h_edge_pos >> 1, s->v_edge_pos >> 1);
641  ptr = s->edge_emu_buffer;
642  }
643  pix_op[dxy](dest_cr, ptr, s->uvlinesize, 8);
644 }
645 
646 static inline void prefetch_motion(MpegEncContext *s, uint8_t **pix, int dir)
647 {
648  /* fetch pixels for estimated mv 4 macroblocks ahead
649  * optimized for 64byte cache lines */
650  const int shift = s->quarter_sample ? 2 : 1;
651  const int mx = (s->mv[dir][0][0] >> shift) + 16 * s->mb_x + 8;
652  const int my = (s->mv[dir][0][1] >> shift) + 16 * s->mb_y;
653  int off = mx + (my + (s->mb_x & 3) * 4) * s->linesize + 64;
654 
655  s->vdsp.prefetch(pix[0] + off, s->linesize, 4);
656  off = (mx >> 1) + ((my >> 1) + (s->mb_x & 7)) * s->uvlinesize + 64;
657  s->vdsp.prefetch(pix[1] + off, pix[2] - pix[1], 2);
658 }
659 
660 static inline void apply_obmc(MpegEncContext *s,
661  uint8_t *dest_y,
662  uint8_t *dest_cb,
663  uint8_t *dest_cr,
664  uint8_t **ref_picture,
665  op_pixels_func (*pix_op)[4])
666 {
667  LOCAL_ALIGNED_8(int16_t, mv_cache, [4], [4][2]);
668  Picture *cur_frame = &s->current_picture;
669  int mb_x = s->mb_x;
670  int mb_y = s->mb_y;
671  const int xy = mb_x + mb_y * s->mb_stride;
672  const int mot_stride = s->b8_stride;
673  const int mot_xy = mb_x * 2 + mb_y * 2 * mot_stride;
674  int mx, my, i;
675 
676  av_assert2(!s->mb_skipped);
677 
678  AV_COPY32(mv_cache[1][1], cur_frame->motion_val[0][mot_xy]);
679  AV_COPY32(mv_cache[1][2], cur_frame->motion_val[0][mot_xy + 1]);
680 
681  AV_COPY32(mv_cache[2][1],
682  cur_frame->motion_val[0][mot_xy + mot_stride]);
683  AV_COPY32(mv_cache[2][2],
684  cur_frame->motion_val[0][mot_xy + mot_stride + 1]);
685 
686  AV_COPY32(mv_cache[3][1],
687  cur_frame->motion_val[0][mot_xy + mot_stride]);
688  AV_COPY32(mv_cache[3][2],
689  cur_frame->motion_val[0][mot_xy + mot_stride + 1]);
690 
691  if (mb_y == 0 || IS_INTRA(cur_frame->mb_type[xy - s->mb_stride])) {
692  AV_COPY32(mv_cache[0][1], mv_cache[1][1]);
693  AV_COPY32(mv_cache[0][2], mv_cache[1][2]);
694  } else {
695  AV_COPY32(mv_cache[0][1],
696  cur_frame->motion_val[0][mot_xy - mot_stride]);
697  AV_COPY32(mv_cache[0][2],
698  cur_frame->motion_val[0][mot_xy - mot_stride + 1]);
699  }
700 
701  if (mb_x == 0 || IS_INTRA(cur_frame->mb_type[xy - 1])) {
702  AV_COPY32(mv_cache[1][0], mv_cache[1][1]);
703  AV_COPY32(mv_cache[2][0], mv_cache[2][1]);
704  } else {
705  AV_COPY32(mv_cache[1][0], cur_frame->motion_val[0][mot_xy - 1]);
706  AV_COPY32(mv_cache[2][0],
707  cur_frame->motion_val[0][mot_xy - 1 + mot_stride]);
708  }
709 
710  if (mb_x + 1 >= s->mb_width || IS_INTRA(cur_frame->mb_type[xy + 1])) {
711  AV_COPY32(mv_cache[1][3], mv_cache[1][2]);
712  AV_COPY32(mv_cache[2][3], mv_cache[2][2]);
713  } else {
714  AV_COPY32(mv_cache[1][3], cur_frame->motion_val[0][mot_xy + 2]);
715  AV_COPY32(mv_cache[2][3],
716  cur_frame->motion_val[0][mot_xy + 2 + mot_stride]);
717  }
718 
719  mx = 0;
720  my = 0;
721  for (i = 0; i < 4; i++) {
722  const int x = (i & 1) + 1;
723  const int y = (i >> 1) + 1;
724  int16_t mv[5][2] = {
725  { mv_cache[y][x][0], mv_cache[y][x][1] },
726  { mv_cache[y - 1][x][0], mv_cache[y - 1][x][1] },
727  { mv_cache[y][x - 1][0], mv_cache[y][x - 1][1] },
728  { mv_cache[y][x + 1][0], mv_cache[y][x + 1][1] },
729  { mv_cache[y + 1][x][0], mv_cache[y + 1][x][1] }
730  };
731  // FIXME cleanup
732  obmc_motion(s, dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize,
733  ref_picture[0],
734  mb_x * 16 + (i & 1) * 8, mb_y * 16 + (i >> 1) * 8,
735  pix_op[1],
736  mv);
737 
738  mx += mv[0][0];
739  my += mv[0][1];
740  }
741  if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY))
742  chroma_4mv_motion(s, dest_cb, dest_cr,
743  ref_picture, pix_op[1],
744  mx, my);
745 }
746 
747 static inline void apply_8x8(MpegEncContext *s,
748  uint8_t *dest_y,
749  uint8_t *dest_cb,
750  uint8_t *dest_cr,
751  int dir,
752  uint8_t **ref_picture,
753  qpel_mc_func (*qpix_op)[16],
754  op_pixels_func (*pix_op)[4])
755 {
756  int dxy, mx, my, src_x, src_y;
757  int i;
758  int mb_x = s->mb_x;
759  int mb_y = s->mb_y;
760  uint8_t *ptr, *dest;
761 
762  mx = 0;
763  my = 0;
764  if (s->quarter_sample) {
765  for (i = 0; i < 4; i++) {
766  int motion_x = s->mv[dir][i][0];
767  int motion_y = s->mv[dir][i][1];
768 
769  dxy = ((motion_y & 3) << 2) | (motion_x & 3);
770  src_x = mb_x * 16 + (motion_x >> 2) + (i & 1) * 8;
771  src_y = mb_y * 16 + (motion_y >> 2) + (i >> 1) * 8;
772 
773  /* WARNING: do no forget half pels */
774  src_x = av_clip(src_x, -16, s->width);
775  if (src_x == s->width)
776  dxy &= ~3;
777  src_y = av_clip(src_y, -16, s->height);
778  if (src_y == s->height)
779  dxy &= ~12;
780 
781  ptr = ref_picture[0] + (src_y * s->linesize) + (src_x);
782  if (s->flags & CODEC_FLAG_EMU_EDGE) {
783  if ((unsigned)src_x > FFMAX(s->h_edge_pos - (motion_x & 3) - 8, 0) ||
784  (unsigned)src_y > FFMAX(s->v_edge_pos - (motion_y & 3) - 8, 0)) {
785  s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr,
786  s->linesize, s->linesize,
787  9, 9,
788  src_x, src_y,
789  s->h_edge_pos,
790  s->v_edge_pos);
791  ptr = s->edge_emu_buffer;
792  }
793  }
794  dest = dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize;
795  qpix_op[1][dxy](dest, ptr, s->linesize);
796 
797  mx += s->mv[dir][i][0] / 2;
798  my += s->mv[dir][i][1] / 2;
799  }
800  } else {
801  for (i = 0; i < 4; i++) {
802  hpel_motion(s,
803  dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize,
804  ref_picture[0],
805  mb_x * 16 + (i & 1) * 8,
806  mb_y * 16 + (i >> 1) * 8,
807  pix_op[1],
808  s->mv[dir][i][0],
809  s->mv[dir][i][1]);
810 
811  mx += s->mv[dir][i][0];
812  my += s->mv[dir][i][1];
813  }
814  }
815 
816  if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY))
817  chroma_4mv_motion(s, dest_cb, dest_cr,
818  ref_picture, pix_op[1], mx, my);
819 }
820 
821 /**
822  * motion compensation of a single macroblock
823  * @param s context
824  * @param dest_y luma destination pointer
825  * @param dest_cb chroma cb/u destination pointer
826  * @param dest_cr chroma cr/v destination pointer
827  * @param dir direction (0->forward, 1->backward)
828  * @param ref_picture array[3] of pointers to the 3 planes of the reference picture
829  * @param pix_op halfpel motion compensation function (average or put normally)
830  * @param qpix_op qpel motion compensation function (average or put normally)
831  * the motion vectors are taken from s->mv and the MV type from s->mv_type
832  */
833 static av_always_inline void MPV_motion_internal(MpegEncContext *s,
834  uint8_t *dest_y,
835  uint8_t *dest_cb,
836  uint8_t *dest_cr,
837  int dir,
838  uint8_t **ref_picture,
839  op_pixels_func (*pix_op)[4],
840  qpel_mc_func (*qpix_op)[16],
841  int is_mpeg12)
842 {
843  int i;
844  int mb_y = s->mb_y;
845 
846  prefetch_motion(s, ref_picture, dir);
847 
848  if (!is_mpeg12 && s->obmc && s->pict_type != AV_PICTURE_TYPE_B) {
849  apply_obmc(s, dest_y, dest_cb, dest_cr, ref_picture, pix_op);
850  return;
851  }
852 
853  switch (s->mv_type) {
854  case MV_TYPE_16X16:
855  if (s->mcsel) {
856  if (s->real_sprite_warping_points == 1) {
857  gmc1_motion(s, dest_y, dest_cb, dest_cr,
858  ref_picture);
859  } else {
860  gmc_motion(s, dest_y, dest_cb, dest_cr,
861  ref_picture);
862  }
863  } else if (!is_mpeg12 && s->quarter_sample) {
864  qpel_motion(s, dest_y, dest_cb, dest_cr,
865  0, 0, 0,
866  ref_picture, pix_op, qpix_op,
867  s->mv[dir][0][0], s->mv[dir][0][1], 16);
868  } else if (!is_mpeg12 && (CONFIG_WMV2_DECODER || CONFIG_WMV2_ENCODER) &&
869  s->mspel && s->codec_id == AV_CODEC_ID_WMV2) {
870  ff_mspel_motion(s, dest_y, dest_cb, dest_cr,
871  ref_picture, pix_op,
872  s->mv[dir][0][0], s->mv[dir][0][1], 16);
873  } else {
874  mpeg_motion(s, dest_y, dest_cb, dest_cr, 0,
875  ref_picture, pix_op,
876  s->mv[dir][0][0], s->mv[dir][0][1], 16, mb_y);
877  }
878  break;
879  case MV_TYPE_8X8:
880  if (!is_mpeg12)
881  apply_8x8(s, dest_y, dest_cb, dest_cr,
882  dir, ref_picture, qpix_op, pix_op);
883  break;
884  case MV_TYPE_FIELD:
885  if (s->picture_structure == PICT_FRAME) {
886  if (!is_mpeg12 && s->quarter_sample) {
887  for (i = 0; i < 2; i++)
888  qpel_motion(s, dest_y, dest_cb, dest_cr,
889  1, i, s->field_select[dir][i],
890  ref_picture, pix_op, qpix_op,
891  s->mv[dir][i][0], s->mv[dir][i][1], 8);
892  } else {
893  /* top field */
894  mpeg_motion_field(s, dest_y, dest_cb, dest_cr,
895  0, s->field_select[dir][0],
896  ref_picture, pix_op,
897  s->mv[dir][0][0], s->mv[dir][0][1], 8, mb_y);
898  /* bottom field */
899  mpeg_motion_field(s, dest_y, dest_cb, dest_cr,
900  1, s->field_select[dir][1],
901  ref_picture, pix_op,
902  s->mv[dir][1][0], s->mv[dir][1][1], 8, mb_y);
903  }
904  } else {
905  if ( s->picture_structure != s->field_select[dir][0] + 1 && s->pict_type != AV_PICTURE_TYPE_B && !s->first_field
906  || !ref_picture[0]) {
907  ref_picture = s->current_picture_ptr->f.data;
908  }
909 
910  mpeg_motion(s, dest_y, dest_cb, dest_cr,
911  s->field_select[dir][0],
912  ref_picture, pix_op,
913  s->mv[dir][0][0], s->mv[dir][0][1], 16, mb_y >> 1);
914  }
915  break;
916  case MV_TYPE_16X8:
917  for (i = 0; i < 2; i++) {
918  uint8_t **ref2picture;
919 
920  if ((s->picture_structure == s->field_select[dir][i] + 1
921  || s->pict_type == AV_PICTURE_TYPE_B || s->first_field) && ref_picture[0]) {
922  ref2picture = ref_picture;
923  } else {
924  ref2picture = s->current_picture_ptr->f.data;
925  }
926 
927  mpeg_motion(s, dest_y, dest_cb, dest_cr,
928  s->field_select[dir][i],
929  ref2picture, pix_op,
930  s->mv[dir][i][0], s->mv[dir][i][1] + 16 * i,
931  8, mb_y >> 1);
932 
933  dest_y += 16 * s->linesize;
934  dest_cb += (16 >> s->chroma_y_shift) * s->uvlinesize;
935  dest_cr += (16 >> s->chroma_y_shift) * s->uvlinesize;
936  }
937  break;
938  case MV_TYPE_DMV:
939  if (s->picture_structure == PICT_FRAME) {
940  for (i = 0; i < 2; i++) {
941  int j;
942  for (j = 0; j < 2; j++)
943  mpeg_motion_field(s, dest_y, dest_cb, dest_cr,
944  j, j ^ i, ref_picture, pix_op,
945  s->mv[dir][2 * i + j][0],
946  s->mv[dir][2 * i + j][1], 8, mb_y);
947  pix_op = s->hdsp.avg_pixels_tab;
948  }
949  } else {
950  if (!ref_picture[0]) {
951  ref_picture = s->current_picture_ptr->f.data;
952  }
953  for (i = 0; i < 2; i++) {
954  mpeg_motion(s, dest_y, dest_cb, dest_cr,
955  s->picture_structure != i + 1,
956  ref_picture, pix_op,
957  s->mv[dir][2 * i][0], s->mv[dir][2 * i][1],
958  16, mb_y >> 1);
959 
960  // after put we make avg of the same block
961  pix_op = s->hdsp.avg_pixels_tab;
962 
963  /* opposite parity is always in the same frame if this is
964  * second field */
965  if (!s->first_field) {
966  ref_picture = s->current_picture_ptr->f.data;
967  }
968  }
969  }
970  break;
971  default: av_assert2(0);
972  }
973 }
974 
975 void ff_MPV_motion(MpegEncContext *s,
976  uint8_t *dest_y, uint8_t *dest_cb,
977  uint8_t *dest_cr, int dir,
978  uint8_t **ref_picture,
979  op_pixels_func (*pix_op)[4],
980  qpel_mc_func (*qpix_op)[16])
981 {
982 #if !CONFIG_SMALL
983  if (s->out_format == FMT_MPEG1)
984  MPV_motion_internal(s, dest_y, dest_cb, dest_cr, dir,
985  ref_picture, pix_op, qpix_op, 1);
986  else
987 #endif
988  MPV_motion_internal(s, dest_y, dest_cb, dest_cr, dir,
989  ref_picture, pix_op, qpix_op, 0);
990 }
Generated on Sat Dec 7 2013 19:21:00 for FFmpeg by   doxygen 1.8.2