05375a1afa5c8540f5ad0517932196d08d113035
[libav.git] / libavcodec / mpegvideo_motion.c
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 Libav.
8 *
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.
13 *
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.
18 *
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
22 */
23
24 #include <string.h>
25
26 #include "libavutil/internal.h"
27 #include "avcodec.h"
28 #include "h261.h"
29 #include "mpegutils.h"
30 #include "mpegvideo.h"
31 #include "mjpegenc.h"
32 #include "msmpeg4.h"
33 #include "qpeldsp.h"
34 #include "wmv2.h"
35 #include <limits.h>
36
37 static void gmc1_motion(MpegEncContext *s,
38 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
39 uint8_t **ref_picture)
40 {
41 uint8_t *ptr;
42 int src_x, src_y, motion_x, motion_y;
43 ptrdiff_t offset, linesize, uvlinesize;
44 int emu = 0;
45
46 motion_x = s->sprite_offset[0][0];
47 motion_y = s->sprite_offset[0][1];
48 src_x = s->mb_x * 16 + (motion_x >> (s->sprite_warping_accuracy + 1));
49 src_y = s->mb_y * 16 + (motion_y >> (s->sprite_warping_accuracy + 1));
50 motion_x <<= (3 - s->sprite_warping_accuracy);
51 motion_y <<= (3 - s->sprite_warping_accuracy);
52 src_x = av_clip(src_x, -16, s->width);
53 if (src_x == s->width)
54 motion_x = 0;
55 src_y = av_clip(src_y, -16, s->height);
56 if (src_y == s->height)
57 motion_y = 0;
58
59 linesize = s->linesize;
60 uvlinesize = s->uvlinesize;
61
62 ptr = ref_picture[0] + src_y * linesize + src_x;
63
64 if ((unsigned)src_x >= FFMAX(s->h_edge_pos - 17, 0) ||
65 (unsigned)src_y >= FFMAX(s->v_edge_pos - 17, 0)) {
66 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, ptr,
67 linesize, linesize,
68 17, 17,
69 src_x, src_y,
70 s->h_edge_pos, s->v_edge_pos);
71 ptr = s->sc.edge_emu_buffer;
72 }
73
74 if ((motion_x | motion_y) & 7) {
75 s->mdsp.gmc1(dest_y, ptr, linesize, 16,
76 motion_x & 15, motion_y & 15, 128 - s->no_rounding);
77 s->mdsp.gmc1(dest_y + 8, ptr + 8, linesize, 16,
78 motion_x & 15, motion_y & 15, 128 - s->no_rounding);
79 } else {
80 int dxy;
81
82 dxy = ((motion_x >> 3) & 1) | ((motion_y >> 2) & 2);
83 if (s->no_rounding) {
84 s->hdsp.put_no_rnd_pixels_tab[0][dxy](dest_y, ptr, linesize, 16);
85 } else {
86 s->hdsp.put_pixels_tab[0][dxy](dest_y, ptr, linesize, 16);
87 }
88 }
89
90 if (CONFIG_GRAY && s->avctx->flags & AV_CODEC_FLAG_GRAY)
91 return;
92
93 motion_x = s->sprite_offset[1][0];
94 motion_y = s->sprite_offset[1][1];
95 src_x = s->mb_x * 8 + (motion_x >> (s->sprite_warping_accuracy + 1));
96 src_y = s->mb_y * 8 + (motion_y >> (s->sprite_warping_accuracy + 1));
97 motion_x <<= (3 - s->sprite_warping_accuracy);
98 motion_y <<= (3 - s->sprite_warping_accuracy);
99 src_x = av_clip(src_x, -8, s->width >> 1);
100 if (src_x == s->width >> 1)
101 motion_x = 0;
102 src_y = av_clip(src_y, -8, s->height >> 1);
103 if (src_y == s->height >> 1)
104 motion_y = 0;
105
106 offset = (src_y * uvlinesize) + src_x;
107 ptr = ref_picture[1] + offset;
108 if ((unsigned)src_x >= FFMAX((s->h_edge_pos >> 1) - 9, 0) ||
109 (unsigned)src_y >= FFMAX((s->v_edge_pos >> 1) - 9, 0)) {
110 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, ptr,
111 uvlinesize, uvlinesize,
112 9, 9,
113 src_x, src_y,
114 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
115 ptr = s->sc.edge_emu_buffer;
116 emu = 1;
117 }
118 s->mdsp.gmc1(dest_cb, ptr, uvlinesize, 8,
119 motion_x & 15, motion_y & 15, 128 - s->no_rounding);
120
121 ptr = ref_picture[2] + offset;
122 if (emu) {
123 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, ptr,
124 uvlinesize, uvlinesize,
125 9, 9,
126 src_x, src_y,
127 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
128 ptr = s->sc.edge_emu_buffer;
129 }
130 s->mdsp.gmc1(dest_cr, ptr, uvlinesize, 8,
131 motion_x & 15, motion_y & 15, 128 - s->no_rounding);
132 }
133
134 static void gmc_motion(MpegEncContext *s,
135 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
136 uint8_t **ref_picture)
137 {
138 uint8_t *ptr;
139 int linesize, uvlinesize;
140 const int a = s->sprite_warping_accuracy;
141 int ox, oy;
142
143 linesize = s->linesize;
144 uvlinesize = s->uvlinesize;
145
146 ptr = ref_picture[0];
147
148 ox = s->sprite_offset[0][0] + s->sprite_delta[0][0] * s->mb_x * 16 +
149 s->sprite_delta[0][1] * s->mb_y * 16;
150 oy = s->sprite_offset[0][1] + s->sprite_delta[1][0] * s->mb_x * 16 +
151 s->sprite_delta[1][1] * s->mb_y * 16;
152
153 s->mdsp.gmc(dest_y, ptr, linesize, 16,
154 ox, oy,
155 s->sprite_delta[0][0], s->sprite_delta[0][1],
156 s->sprite_delta[1][0], s->sprite_delta[1][1],
157 a + 1, (1 << (2 * a + 1)) - s->no_rounding,
158 s->h_edge_pos, s->v_edge_pos);
159 s->mdsp.gmc(dest_y + 8, ptr, linesize, 16,
160 ox + s->sprite_delta[0][0] * 8,
161 oy + s->sprite_delta[1][0] * 8,
162 s->sprite_delta[0][0], s->sprite_delta[0][1],
163 s->sprite_delta[1][0], s->sprite_delta[1][1],
164 a + 1, (1 << (2 * a + 1)) - s->no_rounding,
165 s->h_edge_pos, s->v_edge_pos);
166
167 if (CONFIG_GRAY && s->avctx->flags & AV_CODEC_FLAG_GRAY)
168 return;
169
170 ox = s->sprite_offset[1][0] + s->sprite_delta[0][0] * s->mb_x * 8 +
171 s->sprite_delta[0][1] * s->mb_y * 8;
172 oy = s->sprite_offset[1][1] + s->sprite_delta[1][0] * s->mb_x * 8 +
173 s->sprite_delta[1][1] * s->mb_y * 8;
174
175 ptr = ref_picture[1];
176 s->mdsp.gmc(dest_cb, ptr, uvlinesize, 8,
177 ox, oy,
178 s->sprite_delta[0][0], s->sprite_delta[0][1],
179 s->sprite_delta[1][0], s->sprite_delta[1][1],
180 a + 1, (1 << (2 * a + 1)) - s->no_rounding,
181 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
182
183 ptr = ref_picture[2];
184 s->mdsp.gmc(dest_cr, ptr, uvlinesize, 8,
185 ox, oy,
186 s->sprite_delta[0][0], s->sprite_delta[0][1],
187 s->sprite_delta[1][0], s->sprite_delta[1][1],
188 a + 1, (1 << (2 * a + 1)) - s->no_rounding,
189 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
190 }
191
192 static inline int hpel_motion(MpegEncContext *s,
193 uint8_t *dest, uint8_t *src,
194 int src_x, int src_y,
195 op_pixels_func *pix_op,
196 int motion_x, int motion_y)
197 {
198 int dxy = 0;
199 int emu = 0;
200
201 src_x += motion_x >> 1;
202 src_y += motion_y >> 1;
203
204 /* WARNING: do no forget half pels */
205 src_x = av_clip(src_x, -16, s->width); // FIXME unneeded for emu?
206 if (src_x != s->width)
207 dxy |= motion_x & 1;
208 src_y = av_clip(src_y, -16, s->height);
209 if (src_y != s->height)
210 dxy |= (motion_y & 1) << 1;
211 src += src_y * s->linesize + src_x;
212
213 if (s->unrestricted_mv) {
214 if ((unsigned)src_x > FFMAX(s->h_edge_pos - (motion_x & 1) - 8, 0) ||
215 (unsigned)src_y > FFMAX(s->v_edge_pos - (motion_y & 1) - 8, 0)) {
216 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, src,
217 s->linesize, s->linesize,
218 9, 9,
219 src_x, src_y, s->h_edge_pos,
220 s->v_edge_pos);
221 src = s->sc.edge_emu_buffer;
222 emu = 1;
223 }
224 }
225 pix_op[dxy](dest, src, s->linesize, 8);
226 return emu;
227 }
228
229 static av_always_inline
230 void mpeg_motion_internal(MpegEncContext *s,
231 uint8_t *dest_y,
232 uint8_t *dest_cb,
233 uint8_t *dest_cr,
234 int field_based,
235 int bottom_field,
236 int field_select,
237 uint8_t **ref_picture,
238 op_pixels_func (*pix_op)[4],
239 int motion_x,
240 int motion_y,
241 int h,
242 int is_mpeg12,
243 int mb_y)
244 {
245 uint8_t *ptr_y, *ptr_cb, *ptr_cr;
246 int dxy, uvdxy, mx, my, src_x, src_y,
247 uvsrc_x, uvsrc_y, v_edge_pos;
248 ptrdiff_t uvlinesize, linesize;
249
250 #if 0
251 if (s->quarter_sample) {
252 motion_x >>= 1;
253 motion_y >>= 1;
254 }
255 #endif
256
257 v_edge_pos = s->v_edge_pos >> field_based;
258 linesize = s->current_picture.f->linesize[0] << field_based;
259 uvlinesize = s->current_picture.f->linesize[1] << field_based;
260
261 dxy = ((motion_y & 1) << 1) | (motion_x & 1);
262 src_x = s->mb_x * 16 + (motion_x >> 1);
263 src_y = (mb_y << (4 - field_based)) + (motion_y >> 1);
264
265 if (!is_mpeg12 && s->out_format == FMT_H263) {
266 if ((s->workaround_bugs & FF_BUG_HPEL_CHROMA) && field_based) {
267 mx = (motion_x >> 1) | (motion_x & 1);
268 my = motion_y >> 1;
269 uvdxy = ((my & 1) << 1) | (mx & 1);
270 uvsrc_x = s->mb_x * 8 + (mx >> 1);
271 uvsrc_y = (mb_y << (3 - field_based)) + (my >> 1);
272 } else {
273 uvdxy = dxy | (motion_y & 2) | ((motion_x & 2) >> 1);
274 uvsrc_x = src_x >> 1;
275 uvsrc_y = src_y >> 1;
276 }
277 // Even chroma mv's are full pel in H261
278 } else if (!is_mpeg12 && s->out_format == FMT_H261) {
279 mx = motion_x / 4;
280 my = motion_y / 4;
281 uvdxy = 0;
282 uvsrc_x = s->mb_x * 8 + mx;
283 uvsrc_y = mb_y * 8 + my;
284 } else {
285 if (s->chroma_y_shift) {
286 mx = motion_x / 2;
287 my = motion_y / 2;
288 uvdxy = ((my & 1) << 1) | (mx & 1);
289 uvsrc_x = s->mb_x * 8 + (mx >> 1);
290 uvsrc_y = (mb_y << (3 - field_based)) + (my >> 1);
291 } else {
292 if (s->chroma_x_shift) {
293 // Chroma422
294 mx = motion_x / 2;
295 uvdxy = ((motion_y & 1) << 1) | (mx & 1);
296 uvsrc_x = s->mb_x * 8 + (mx >> 1);
297 uvsrc_y = src_y;
298 } else {
299 // Chroma444
300 uvdxy = dxy;
301 uvsrc_x = src_x;
302 uvsrc_y = src_y;
303 }
304 }
305 }
306
307 ptr_y = ref_picture[0] + src_y * linesize + src_x;
308 ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x;
309 ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x;
310
311 if ((unsigned)src_x > FFMAX(s->h_edge_pos - (motion_x & 1) - 16, 0) ||
312 (unsigned)src_y > FFMAX(v_edge_pos - (motion_y & 1) - h, 0)) {
313 if (is_mpeg12 ||
314 s->codec_id == AV_CODEC_ID_MPEG2VIDEO ||
315 s->codec_id == AV_CODEC_ID_MPEG1VIDEO) {
316 av_log(s->avctx, AV_LOG_DEBUG,
317 "MPEG motion vector out of boundary (%d %d)\n", src_x,
318 src_y);
319 return;
320 }
321 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, ptr_y,
322 s->linesize, s->linesize,
323 17, 17 + field_based,
324 src_x, src_y << field_based,
325 s->h_edge_pos, s->v_edge_pos);
326 ptr_y = s->sc.edge_emu_buffer;
327 if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY)) {
328 uint8_t *uvbuf = s->sc.edge_emu_buffer + 18 * s->linesize;
329 s->vdsp.emulated_edge_mc(uvbuf, ptr_cb,
330 s->uvlinesize, s->uvlinesize,
331 9, 9 + field_based,
332 uvsrc_x, uvsrc_y << field_based,
333 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
334 s->vdsp.emulated_edge_mc(uvbuf + 16, ptr_cr,
335 s->uvlinesize, s->uvlinesize,
336 9, 9 + field_based,
337 uvsrc_x, uvsrc_y << field_based,
338 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
339 ptr_cb = uvbuf;
340 ptr_cr = uvbuf + 16;
341 }
342 }
343
344 /* FIXME use this for field pix too instead of the obnoxious hack which
345 * changes picture.data */
346 if (bottom_field) {
347 dest_y += s->linesize;
348 dest_cb += s->uvlinesize;
349 dest_cr += s->uvlinesize;
350 }
351
352 if (field_select) {
353 ptr_y += s->linesize;
354 ptr_cb += s->uvlinesize;
355 ptr_cr += s->uvlinesize;
356 }
357
358 pix_op[0][dxy](dest_y, ptr_y, linesize, h);
359
360 if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY)) {
361 pix_op[s->chroma_x_shift][uvdxy]
362 (dest_cb, ptr_cb, uvlinesize, h >> s->chroma_y_shift);
363 pix_op[s->chroma_x_shift][uvdxy]
364 (dest_cr, ptr_cr, uvlinesize, h >> s->chroma_y_shift);
365 }
366 if (!is_mpeg12 && (CONFIG_H261_ENCODER || CONFIG_H261_DECODER) &&
367 s->out_format == FMT_H261) {
368 ff_h261_loop_filter(s);
369 }
370 }
371 /* apply one mpeg motion vector to the three components */
372 static void mpeg_motion(MpegEncContext *s,
373 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
374 int field_select, uint8_t **ref_picture,
375 op_pixels_func (*pix_op)[4],
376 int motion_x, int motion_y, int h, int mb_y)
377 {
378 #if !CONFIG_SMALL
379 if (s->out_format == FMT_MPEG1)
380 mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, 0, 0,
381 field_select, ref_picture, pix_op,
382 motion_x, motion_y, h, 1, mb_y);
383 else
384 #endif
385 mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, 0, 0,
386 field_select, ref_picture, pix_op,
387 motion_x, motion_y, h, 0, mb_y);
388 }
389
390 static void mpeg_motion_field(MpegEncContext *s, uint8_t *dest_y,
391 uint8_t *dest_cb, uint8_t *dest_cr,
392 int bottom_field, int field_select,
393 uint8_t **ref_picture,
394 op_pixels_func (*pix_op)[4],
395 int motion_x, int motion_y, int h, int mb_y)
396 {
397 #if !CONFIG_SMALL
398 if(s->out_format == FMT_MPEG1)
399 mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, 1,
400 bottom_field, field_select, ref_picture, pix_op,
401 motion_x, motion_y, h, 1, mb_y);
402 else
403 #endif
404 mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, 1,
405 bottom_field, field_select, ref_picture, pix_op,
406 motion_x, motion_y, h, 0, mb_y);
407 }
408
409 // FIXME: SIMDify, avg variant, 16x16 version
410 static inline void put_obmc(uint8_t *dst, uint8_t *src[5], int stride)
411 {
412 int x;
413 uint8_t *const top = src[1];
414 uint8_t *const left = src[2];
415 uint8_t *const mid = src[0];
416 uint8_t *const right = src[3];
417 uint8_t *const bottom = src[4];
418 #define OBMC_FILTER(x, t, l, m, r, b)\
419 dst[x]= (t*top[x] + l*left[x] + m*mid[x] + r*right[x] + b*bottom[x] + 4)>>3
420 #define OBMC_FILTER4(x, t, l, m, r, b)\
421 OBMC_FILTER(x , t, l, m, r, b);\
422 OBMC_FILTER(x+1 , t, l, m, r, b);\
423 OBMC_FILTER(x +stride, t, l, m, r, b);\
424 OBMC_FILTER(x+1+stride, t, l, m, r, b);
425
426 x = 0;
427 OBMC_FILTER (x , 2, 2, 4, 0, 0);
428 OBMC_FILTER (x + 1, 2, 1, 5, 0, 0);
429 OBMC_FILTER4(x + 2, 2, 1, 5, 0, 0);
430 OBMC_FILTER4(x + 4, 2, 0, 5, 1, 0);
431 OBMC_FILTER (x + 6, 2, 0, 5, 1, 0);
432 OBMC_FILTER (x + 7, 2, 0, 4, 2, 0);
433 x += stride;
434 OBMC_FILTER (x , 1, 2, 5, 0, 0);
435 OBMC_FILTER (x + 1, 1, 2, 5, 0, 0);
436 OBMC_FILTER (x + 6, 1, 0, 5, 2, 0);
437 OBMC_FILTER (x + 7, 1, 0, 5, 2, 0);
438 x += stride;
439 OBMC_FILTER4(x , 1, 2, 5, 0, 0);
440 OBMC_FILTER4(x + 2, 1, 1, 6, 0, 0);
441 OBMC_FILTER4(x + 4, 1, 0, 6, 1, 0);
442 OBMC_FILTER4(x + 6, 1, 0, 5, 2, 0);
443 x += 2 * stride;
444 OBMC_FILTER4(x , 0, 2, 5, 0, 1);
445 OBMC_FILTER4(x + 2, 0, 1, 6, 0, 1);
446 OBMC_FILTER4(x + 4, 0, 0, 6, 1, 1);
447 OBMC_FILTER4(x + 6, 0, 0, 5, 2, 1);
448 x += 2*stride;
449 OBMC_FILTER (x , 0, 2, 5, 0, 1);
450 OBMC_FILTER (x + 1, 0, 2, 5, 0, 1);
451 OBMC_FILTER4(x + 2, 0, 1, 5, 0, 2);
452 OBMC_FILTER4(x + 4, 0, 0, 5, 1, 2);
453 OBMC_FILTER (x + 6, 0, 0, 5, 2, 1);
454 OBMC_FILTER (x + 7, 0, 0, 5, 2, 1);
455 x += stride;
456 OBMC_FILTER (x , 0, 2, 4, 0, 2);
457 OBMC_FILTER (x + 1, 0, 1, 5, 0, 2);
458 OBMC_FILTER (x + 6, 0, 0, 5, 1, 2);
459 OBMC_FILTER (x + 7, 0, 0, 4, 2, 2);
460 }
461
462 /* obmc for 1 8x8 luma block */
463 static inline void obmc_motion(MpegEncContext *s,
464 uint8_t *dest, uint8_t *src,
465 int src_x, int src_y,
466 op_pixels_func *pix_op,
467 int16_t mv[5][2] /* mid top left right bottom */)
468 #define MID 0
469 {
470 int i;
471 uint8_t *ptr[5];
472
473 assert(s->quarter_sample == 0);
474
475 for (i = 0; i < 5; i++) {
476 if (i && mv[i][0] == mv[MID][0] && mv[i][1] == mv[MID][1]) {
477 ptr[i] = ptr[MID];
478 } else {
479 ptr[i] = s->sc.obmc_scratchpad + 8 * (i & 1) +
480 s->linesize * 8 * (i >> 1);
481 hpel_motion(s, ptr[i], src, src_x, src_y, pix_op,
482 mv[i][0], mv[i][1]);
483 }
484 }
485
486 put_obmc(dest, ptr, s->linesize);
487 }
488
489 static inline void qpel_motion(MpegEncContext *s,
490 uint8_t *dest_y,
491 uint8_t *dest_cb,
492 uint8_t *dest_cr,
493 int field_based, int bottom_field,
494 int field_select, uint8_t **ref_picture,
495 op_pixels_func (*pix_op)[4],
496 qpel_mc_func (*qpix_op)[16],
497 int motion_x, int motion_y, int h)
498 {
499 uint8_t *ptr_y, *ptr_cb, *ptr_cr;
500 int dxy, uvdxy, mx, my, src_x, src_y, uvsrc_x, uvsrc_y, v_edge_pos;
501 ptrdiff_t linesize, uvlinesize;
502
503 dxy = ((motion_y & 3) << 2) | (motion_x & 3);
504
505 src_x = s->mb_x * 16 + (motion_x >> 2);
506 src_y = s->mb_y * (16 >> field_based) + (motion_y >> 2);
507
508 v_edge_pos = s->v_edge_pos >> field_based;
509 linesize = s->linesize << field_based;
510 uvlinesize = s->uvlinesize << field_based;
511
512 if (field_based) {
513 mx = motion_x / 2;
514 my = motion_y >> 1;
515 } else if (s->workaround_bugs & FF_BUG_QPEL_CHROMA2) {
516 static const int rtab[8] = { 0, 0, 1, 1, 0, 0, 0, 1 };
517 mx = (motion_x >> 1) + rtab[motion_x & 7];
518 my = (motion_y >> 1) + rtab[motion_y & 7];
519 } else if (s->workaround_bugs & FF_BUG_QPEL_CHROMA) {
520 mx = (motion_x >> 1) | (motion_x & 1);
521 my = (motion_y >> 1) | (motion_y & 1);
522 } else {
523 mx = motion_x / 2;
524 my = motion_y / 2;
525 }
526 mx = (mx >> 1) | (mx & 1);
527 my = (my >> 1) | (my & 1);
528
529 uvdxy = (mx & 1) | ((my & 1) << 1);
530 mx >>= 1;
531 my >>= 1;
532
533 uvsrc_x = s->mb_x * 8 + mx;
534 uvsrc_y = s->mb_y * (8 >> field_based) + my;
535
536 ptr_y = ref_picture[0] + src_y * linesize + src_x;
537 ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x;
538 ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x;
539
540 if ((unsigned)src_x > FFMAX(s->h_edge_pos - (motion_x & 3) - 16, 0) ||
541 (unsigned)src_y > FFMAX(v_edge_pos - (motion_y & 3) - h, 0)) {
542 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, ptr_y,
543 s->linesize, s->linesize,
544 17, 17 + field_based,
545 src_x, src_y << field_based,
546 s->h_edge_pos, s->v_edge_pos);
547 ptr_y = s->sc.edge_emu_buffer;
548 if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY)) {
549 uint8_t *uvbuf = s->sc.edge_emu_buffer + 18 * s->linesize;
550 s->vdsp.emulated_edge_mc(uvbuf, ptr_cb,
551 s->uvlinesize, s->uvlinesize,
552 9, 9 + field_based,
553 uvsrc_x, uvsrc_y << field_based,
554 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
555 s->vdsp.emulated_edge_mc(uvbuf + 16, ptr_cr,
556 s->uvlinesize, s->uvlinesize,
557 9, 9 + field_based,
558 uvsrc_x, uvsrc_y << field_based,
559 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
560 ptr_cb = uvbuf;
561 ptr_cr = uvbuf + 16;
562 }
563 }
564
565 if (!field_based)
566 qpix_op[0][dxy](dest_y, ptr_y, linesize);
567 else {
568 if (bottom_field) {
569 dest_y += s->linesize;
570 dest_cb += s->uvlinesize;
571 dest_cr += s->uvlinesize;
572 }
573
574 if (field_select) {
575 ptr_y += s->linesize;
576 ptr_cb += s->uvlinesize;
577 ptr_cr += s->uvlinesize;
578 }
579 // damn interlaced mode
580 // FIXME boundary mirroring is not exactly correct here
581 qpix_op[1][dxy](dest_y, ptr_y, linesize);
582 qpix_op[1][dxy](dest_y + 8, ptr_y + 8, linesize);
583 }
584 if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY)) {
585 pix_op[1][uvdxy](dest_cr, ptr_cr, uvlinesize, h >> 1);
586 pix_op[1][uvdxy](dest_cb, ptr_cb, uvlinesize, h >> 1);
587 }
588 }
589
590 /**
591 * h263 chroma 4mv motion compensation.
592 */
593 static void chroma_4mv_motion(MpegEncContext *s,
594 uint8_t *dest_cb, uint8_t *dest_cr,
595 uint8_t **ref_picture,
596 op_pixels_func *pix_op,
597 int mx, int my)
598 {
599 uint8_t *ptr;
600 int src_x, src_y, dxy, emu = 0;
601 ptrdiff_t offset;
602
603 /* In case of 8X8, we construct a single chroma motion vector
604 * with a special rounding */
605 mx = ff_h263_round_chroma(mx);
606 my = ff_h263_round_chroma(my);
607
608 dxy = ((my & 1) << 1) | (mx & 1);
609 mx >>= 1;
610 my >>= 1;
611
612 src_x = s->mb_x * 8 + mx;
613 src_y = s->mb_y * 8 + my;
614 src_x = av_clip(src_x, -8, (s->width >> 1));
615 if (src_x == (s->width >> 1))
616 dxy &= ~1;
617 src_y = av_clip(src_y, -8, (s->height >> 1));
618 if (src_y == (s->height >> 1))
619 dxy &= ~2;
620
621 offset = src_y * s->uvlinesize + src_x;
622 ptr = ref_picture[1] + offset;
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->sc.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->sc.edge_emu_buffer;
630 emu = 1;
631 }
632 pix_op[dxy](dest_cb, ptr, s->uvlinesize, 8);
633
634 ptr = ref_picture[2] + offset;
635 if (emu) {
636 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, ptr,
637 s->uvlinesize, s->uvlinesize,
638 9, 9, src_x, src_y,
639 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
640 ptr = s->sc.edge_emu_buffer;
641 }
642 pix_op[dxy](dest_cr, ptr, s->uvlinesize, 8);
643 }
644
645 static inline void prefetch_motion(MpegEncContext *s, uint8_t **pix, int dir)
646 {
647 /* fetch pixels for estimated mv 4 macroblocks ahead
648 * optimized for 64byte cache lines */
649 const int shift = s->quarter_sample ? 2 : 1;
650 const int mx = (s->mv[dir][0][0] >> shift) + 16 * s->mb_x + 8;
651 const int my = (s->mv[dir][0][1] >> shift) + 16 * s->mb_y;
652 int off = mx + (my + (s->mb_x & 3) * 4) * s->linesize + 64;
653
654 s->vdsp.prefetch(pix[0] + off, s->linesize, 4);
655 off = (mx >> 1) + ((my >> 1) + (s->mb_x & 7)) * s->uvlinesize + 64;
656 s->vdsp.prefetch(pix[1] + off, pix[2] - pix[1], 2);
657 }
658
659 static inline void apply_obmc(MpegEncContext *s,
660 uint8_t *dest_y,
661 uint8_t *dest_cb,
662 uint8_t *dest_cr,
663 uint8_t **ref_picture,
664 op_pixels_func (*pix_op)[4])
665 {
666 LOCAL_ALIGNED_8(int16_t, mv_cache, [4], [4][2]);
667 Picture *cur_frame = &s->current_picture;
668 int mb_x = s->mb_x;
669 int mb_y = s->mb_y;
670 const int xy = mb_x + mb_y * s->mb_stride;
671 const int mot_stride = s->b8_stride;
672 const int mot_xy = mb_x * 2 + mb_y * 2 * mot_stride;
673 int mx, my, i;
674
675 assert(!s->mb_skipped);
676
677 AV_COPY32(mv_cache[1][1], cur_frame->motion_val[0][mot_xy]);
678 AV_COPY32(mv_cache[1][2], cur_frame->motion_val[0][mot_xy + 1]);
679
680 AV_COPY32(mv_cache[2][1],
681 cur_frame->motion_val[0][mot_xy + mot_stride]);
682 AV_COPY32(mv_cache[2][2],
683 cur_frame->motion_val[0][mot_xy + mot_stride + 1]);
684
685 AV_COPY32(mv_cache[3][1],
686 cur_frame->motion_val[0][mot_xy + mot_stride]);
687 AV_COPY32(mv_cache[3][2],
688 cur_frame->motion_val[0][mot_xy + mot_stride + 1]);
689
690 if (mb_y == 0 || IS_INTRA(cur_frame->mb_type[xy - s->mb_stride])) {
691 AV_COPY32(mv_cache[0][1], mv_cache[1][1]);
692 AV_COPY32(mv_cache[0][2], mv_cache[1][2]);
693 } else {
694 AV_COPY32(mv_cache[0][1],
695 cur_frame->motion_val[0][mot_xy - mot_stride]);
696 AV_COPY32(mv_cache[0][2],
697 cur_frame->motion_val[0][mot_xy - mot_stride + 1]);
698 }
699
700 if (mb_x == 0 || IS_INTRA(cur_frame->mb_type[xy - 1])) {
701 AV_COPY32(mv_cache[1][0], mv_cache[1][1]);
702 AV_COPY32(mv_cache[2][0], mv_cache[2][1]);
703 } else {
704 AV_COPY32(mv_cache[1][0], cur_frame->motion_val[0][mot_xy - 1]);
705 AV_COPY32(mv_cache[2][0],
706 cur_frame->motion_val[0][mot_xy - 1 + mot_stride]);
707 }
708
709 if (mb_x + 1 >= s->mb_width || IS_INTRA(cur_frame->mb_type[xy + 1])) {
710 AV_COPY32(mv_cache[1][3], mv_cache[1][2]);
711 AV_COPY32(mv_cache[2][3], mv_cache[2][2]);
712 } else {
713 AV_COPY32(mv_cache[1][3], cur_frame->motion_val[0][mot_xy + 2]);
714 AV_COPY32(mv_cache[2][3],
715 cur_frame->motion_val[0][mot_xy + 2 + mot_stride]);
716 }
717
718 mx = 0;
719 my = 0;
720 for (i = 0; i < 4; i++) {
721 const int x = (i & 1) + 1;
722 const int y = (i >> 1) + 1;
723 int16_t mv[5][2] = {
724 { mv_cache[y][x][0], mv_cache[y][x][1] },
725 { mv_cache[y - 1][x][0], mv_cache[y - 1][x][1] },
726 { mv_cache[y][x - 1][0], mv_cache[y][x - 1][1] },
727 { mv_cache[y][x + 1][0], mv_cache[y][x + 1][1] },
728 { mv_cache[y + 1][x][0], mv_cache[y + 1][x][1] }
729 };
730 // FIXME cleanup
731 obmc_motion(s, dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize,
732 ref_picture[0],
733 mb_x * 16 + (i & 1) * 8, mb_y * 16 + (i >> 1) * 8,
734 pix_op[1],
735 mv);
736
737 mx += mv[0][0];
738 my += mv[0][1];
739 }
740 if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY))
741 chroma_4mv_motion(s, dest_cb, dest_cr,
742 ref_picture, pix_op[1],
743 mx, my);
744 }
745
746 static inline void apply_8x8(MpegEncContext *s,
747 uint8_t *dest_y,
748 uint8_t *dest_cb,
749 uint8_t *dest_cr,
750 int dir,
751 uint8_t **ref_picture,
752 qpel_mc_func (*qpix_op)[16],
753 op_pixels_func (*pix_op)[4])
754 {
755 int dxy, mx, my, src_x, src_y;
756 int i;
757 int mb_x = s->mb_x;
758 int mb_y = s->mb_y;
759 uint8_t *ptr, *dest;
760
761 mx = 0;
762 my = 0;
763 if (s->quarter_sample) {
764 for (i = 0; i < 4; i++) {
765 int motion_x = s->mv[dir][i][0];
766 int motion_y = s->mv[dir][i][1];
767
768 dxy = ((motion_y & 3) << 2) | (motion_x & 3);
769 src_x = mb_x * 16 + (motion_x >> 2) + (i & 1) * 8;
770 src_y = mb_y * 16 + (motion_y >> 2) + (i >> 1) * 8;
771
772 /* WARNING: do no forget half pels */
773 src_x = av_clip(src_x, -16, s->width);
774 if (src_x == s->width)
775 dxy &= ~3;
776 src_y = av_clip(src_y, -16, s->height);
777 if (src_y == s->height)
778 dxy &= ~12;
779
780 ptr = ref_picture[0] + (src_y * s->linesize) + (src_x);
781 if ((unsigned)src_x > FFMAX(s->h_edge_pos - (motion_x & 3) - 8, 0) ||
782 (unsigned)src_y > FFMAX(s->v_edge_pos - (motion_y & 3) - 8, 0)) {
783 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, ptr,
784 s->linesize, s->linesize,
785 9, 9,
786 src_x, src_y,
787 s->h_edge_pos,
788 s->v_edge_pos);
789 ptr = s->sc.edge_emu_buffer;
790 }
791 dest = dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize;
792 qpix_op[1][dxy](dest, ptr, s->linesize);
793
794 mx += s->mv[dir][i][0] / 2;
795 my += s->mv[dir][i][1] / 2;
796 }
797 } else {
798 for (i = 0; i < 4; i++) {
799 hpel_motion(s,
800 dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize,
801 ref_picture[0],
802 mb_x * 16 + (i & 1) * 8,
803 mb_y * 16 + (i >> 1) * 8,
804 pix_op[1],
805 s->mv[dir][i][0],
806 s->mv[dir][i][1]);
807
808 mx += s->mv[dir][i][0];
809 my += s->mv[dir][i][1];
810 }
811 }
812
813 if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY))
814 chroma_4mv_motion(s, dest_cb, dest_cr,
815 ref_picture, pix_op[1], mx, my);
816 }
817
818 /**
819 * motion compensation of a single macroblock
820 * @param s context
821 * @param dest_y luma destination pointer
822 * @param dest_cb chroma cb/u destination pointer
823 * @param dest_cr chroma cr/v destination pointer
824 * @param dir direction (0->forward, 1->backward)
825 * @param ref_picture array[3] of pointers to the 3 planes of the reference picture
826 * @param pix_op halfpel motion compensation function (average or put normally)
827 * @param qpix_op qpel motion compensation function (average or put normally)
828 * the motion vectors are taken from s->mv and the MV type from s->mv_type
829 */
830 static av_always_inline void mpv_motion_internal(MpegEncContext *s,
831 uint8_t *dest_y,
832 uint8_t *dest_cb,
833 uint8_t *dest_cr,
834 int dir,
835 uint8_t **ref_picture,
836 op_pixels_func (*pix_op)[4],
837 qpel_mc_func (*qpix_op)[16],
838 int is_mpeg12)
839 {
840 int i;
841 int mb_y = s->mb_y;
842
843 prefetch_motion(s, ref_picture, dir);
844
845 if (!is_mpeg12 && s->obmc && s->pict_type != AV_PICTURE_TYPE_B) {
846 apply_obmc(s, dest_y, dest_cb, dest_cr, ref_picture, pix_op);
847 return;
848 }
849
850 switch (s->mv_type) {
851 case MV_TYPE_16X16:
852 if (s->mcsel) {
853 if (s->real_sprite_warping_points == 1) {
854 gmc1_motion(s, dest_y, dest_cb, dest_cr,
855 ref_picture);
856 } else {
857 gmc_motion(s, dest_y, dest_cb, dest_cr,
858 ref_picture);
859 }
860 } else if (!is_mpeg12 && s->quarter_sample) {
861 qpel_motion(s, dest_y, dest_cb, dest_cr,
862 0, 0, 0,
863 ref_picture, pix_op, qpix_op,
864 s->mv[dir][0][0], s->mv[dir][0][1], 16);
865 } else if (!is_mpeg12 && (CONFIG_WMV2_DECODER || CONFIG_WMV2_ENCODER) &&
866 s->mspel && s->codec_id == AV_CODEC_ID_WMV2) {
867 ff_mspel_motion(s, dest_y, dest_cb, dest_cr,
868 ref_picture, pix_op,
869 s->mv[dir][0][0], s->mv[dir][0][1], 16);
870 } else {
871 mpeg_motion(s, dest_y, dest_cb, dest_cr, 0,
872 ref_picture, pix_op,
873 s->mv[dir][0][0], s->mv[dir][0][1], 16, mb_y);
874 }
875 break;
876 case MV_TYPE_8X8:
877 if (!is_mpeg12)
878 apply_8x8(s, dest_y, dest_cb, dest_cr,
879 dir, ref_picture, qpix_op, pix_op);
880 break;
881 case MV_TYPE_FIELD:
882 if (s->picture_structure == PICT_FRAME) {
883 if (!is_mpeg12 && s->quarter_sample) {
884 for (i = 0; i < 2; i++)
885 qpel_motion(s, dest_y, dest_cb, dest_cr,
886 1, i, s->field_select[dir][i],
887 ref_picture, pix_op, qpix_op,
888 s->mv[dir][i][0], s->mv[dir][i][1], 8);
889 } else {
890 /* top field */
891 mpeg_motion_field(s, dest_y, dest_cb, dest_cr,
892 0, s->field_select[dir][0],
893 ref_picture, pix_op,
894 s->mv[dir][0][0], s->mv[dir][0][1], 8, mb_y);
895 /* bottom field */
896 mpeg_motion_field(s, dest_y, dest_cb, dest_cr,
897 1, s->field_select[dir][1],
898 ref_picture, pix_op,
899 s->mv[dir][1][0], s->mv[dir][1][1], 8, mb_y);
900 }
901 } else {
902 if (s->picture_structure != s->field_select[dir][0] + 1 &&
903 s->pict_type != AV_PICTURE_TYPE_B && !s->first_field) {
904 ref_picture = s->current_picture_ptr->f->data;
905 }
906
907 mpeg_motion(s, dest_y, dest_cb, dest_cr,
908 s->field_select[dir][0],
909 ref_picture, pix_op,
910 s->mv[dir][0][0], s->mv[dir][0][1], 16, mb_y >> 1);
911 }
912 break;
913 case MV_TYPE_16X8:
914 for (i = 0; i < 2; i++) {
915 uint8_t **ref2picture;
916
917 if (s->picture_structure == s->field_select[dir][i] + 1
918 || s->pict_type == AV_PICTURE_TYPE_B || s->first_field) {
919 ref2picture = ref_picture;
920 } else {
921 ref2picture = s->current_picture_ptr->f->data;
922 }
923
924 mpeg_motion(s, dest_y, dest_cb, dest_cr,
925 s->field_select[dir][i],
926 ref2picture, pix_op,
927 s->mv[dir][i][0], s->mv[dir][i][1] + 16 * i,
928 8, mb_y >> 1);
929
930 dest_y += 16 * s->linesize;
931 dest_cb += (16 >> s->chroma_y_shift) * s->uvlinesize;
932 dest_cr += (16 >> s->chroma_y_shift) * s->uvlinesize;
933 }
934 break;
935 case MV_TYPE_DMV:
936 if (s->picture_structure == PICT_FRAME) {
937 for (i = 0; i < 2; i++) {
938 int j;
939 for (j = 0; j < 2; j++)
940 mpeg_motion_field(s, dest_y, dest_cb, dest_cr,
941 j, j ^ i, ref_picture, pix_op,
942 s->mv[dir][2 * i + j][0],
943 s->mv[dir][2 * i + j][1], 8, mb_y);
944 pix_op = s->hdsp.avg_pixels_tab;
945 }
946 } else {
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: assert(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 }