summaryrefslogtreecommitdiff
path: root/libavcodec/mpegvideo_motion.c (plain)
blob: c913504a657f228be18600605fbdc7a3e767334b
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 "h261.h"
30#include "mpegutils.h"
31#include "mpegvideo.h"
32#include "mjpegenc.h"
33#include "msmpeg4.h"
34#include "qpeldsp.h"
35#include "wmv2.h"
36#include <limits.h>
37
38static void gmc1_motion(MpegEncContext *s,
39 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
40 uint8_t **ref_picture)
41{
42 uint8_t *ptr;
43 int src_x, src_y, motion_x, motion_y;
44 ptrdiff_t offset, linesize, uvlinesize;
45 int emu = 0;
46
47 motion_x = s->sprite_offset[0][0];
48 motion_y = s->sprite_offset[0][1];
49 src_x = s->mb_x * 16 + (motion_x >> (s->sprite_warping_accuracy + 1));
50 src_y = s->mb_y * 16 + (motion_y >> (s->sprite_warping_accuracy + 1));
51 motion_x *= 1 << (3 - s->sprite_warping_accuracy);
52 motion_y *= 1 << (3 - s->sprite_warping_accuracy);
53 src_x = av_clip(src_x, -16, s->width);
54 if (src_x == s->width)
55 motion_x = 0;
56 src_y = av_clip(src_y, -16, s->height);
57 if (src_y == s->height)
58 motion_y = 0;
59
60 linesize = s->linesize;
61 uvlinesize = s->uvlinesize;
62
63 ptr = ref_picture[0] + src_y * linesize + src_x;
64
65 if ((unsigned)src_x >= FFMAX(s->h_edge_pos - 17, 0) ||
66 (unsigned)src_y >= FFMAX(s->v_edge_pos - 17, 0)) {
67 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, ptr,
68 linesize, linesize,
69 17, 17,
70 src_x, src_y,
71 s->h_edge_pos, s->v_edge_pos);
72 ptr = s->sc.edge_emu_buffer;
73 }
74
75 if ((motion_x | motion_y) & 7) {
76 s->mdsp.gmc1(dest_y, ptr, linesize, 16,
77 motion_x & 15, motion_y & 15, 128 - s->no_rounding);
78 s->mdsp.gmc1(dest_y + 8, ptr + 8, linesize, 16,
79 motion_x & 15, motion_y & 15, 128 - s->no_rounding);
80 } else {
81 int dxy;
82
83 dxy = ((motion_x >> 3) & 1) | ((motion_y >> 2) & 2);
84 if (s->no_rounding) {
85 s->hdsp.put_no_rnd_pixels_tab[0][dxy](dest_y, ptr, linesize, 16);
86 } else {
87 s->hdsp.put_pixels_tab[0][dxy](dest_y, ptr, linesize, 16);
88 }
89 }
90
91 if (CONFIG_GRAY && s->avctx->flags & AV_CODEC_FLAG_GRAY)
92 return;
93
94 motion_x = s->sprite_offset[1][0];
95 motion_y = s->sprite_offset[1][1];
96 src_x = s->mb_x * 8 + (motion_x >> (s->sprite_warping_accuracy + 1));
97 src_y = s->mb_y * 8 + (motion_y >> (s->sprite_warping_accuracy + 1));
98 motion_x *= 1 << (3 - s->sprite_warping_accuracy);
99 motion_y *= 1 << (3 - s->sprite_warping_accuracy);
100 src_x = av_clip(src_x, -8, s->width >> 1);
101 if (src_x == s->width >> 1)
102 motion_x = 0;
103 src_y = av_clip(src_y, -8, s->height >> 1);
104 if (src_y == s->height >> 1)
105 motion_y = 0;
106
107 offset = (src_y * uvlinesize) + src_x;
108 ptr = ref_picture[1] + offset;
109 if ((unsigned)src_x >= FFMAX((s->h_edge_pos >> 1) - 9, 0) ||
110 (unsigned)src_y >= FFMAX((s->v_edge_pos >> 1) - 9, 0)) {
111 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, ptr,
112 uvlinesize, uvlinesize,
113 9, 9,
114 src_x, src_y,
115 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
116 ptr = s->sc.edge_emu_buffer;
117 emu = 1;
118 }
119 s->mdsp.gmc1(dest_cb, ptr, uvlinesize, 8,
120 motion_x & 15, motion_y & 15, 128 - s->no_rounding);
121
122 ptr = ref_picture[2] + offset;
123 if (emu) {
124 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, ptr,
125 uvlinesize, uvlinesize,
126 9, 9,
127 src_x, src_y,
128 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
129 ptr = s->sc.edge_emu_buffer;
130 }
131 s->mdsp.gmc1(dest_cr, ptr, uvlinesize, 8,
132 motion_x & 15, motion_y & 15, 128 - s->no_rounding);
133}
134
135static void gmc_motion(MpegEncContext *s,
136 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
137 uint8_t **ref_picture)
138{
139 uint8_t *ptr;
140 int linesize, uvlinesize;
141 const int a = s->sprite_warping_accuracy;
142 int ox, oy;
143
144 linesize = s->linesize;
145 uvlinesize = s->uvlinesize;
146
147 ptr = ref_picture[0];
148
149 ox = s->sprite_offset[0][0] + s->sprite_delta[0][0] * s->mb_x * 16 +
150 s->sprite_delta[0][1] * s->mb_y * 16;
151 oy = s->sprite_offset[0][1] + s->sprite_delta[1][0] * s->mb_x * 16 +
152 s->sprite_delta[1][1] * s->mb_y * 16;
153
154 s->mdsp.gmc(dest_y, ptr, linesize, 16,
155 ox, oy,
156 s->sprite_delta[0][0], s->sprite_delta[0][1],
157 s->sprite_delta[1][0], s->sprite_delta[1][1],
158 a + 1, (1 << (2 * a + 1)) - s->no_rounding,
159 s->h_edge_pos, s->v_edge_pos);
160 s->mdsp.gmc(dest_y + 8, ptr, linesize, 16,
161 ox + s->sprite_delta[0][0] * 8,
162 oy + s->sprite_delta[1][0] * 8,
163 s->sprite_delta[0][0], s->sprite_delta[0][1],
164 s->sprite_delta[1][0], s->sprite_delta[1][1],
165 a + 1, (1 << (2 * a + 1)) - s->no_rounding,
166 s->h_edge_pos, s->v_edge_pos);
167
168 if (CONFIG_GRAY && s->avctx->flags & AV_CODEC_FLAG_GRAY)
169 return;
170
171 ox = s->sprite_offset[1][0] + s->sprite_delta[0][0] * s->mb_x * 8 +
172 s->sprite_delta[0][1] * s->mb_y * 8;
173 oy = s->sprite_offset[1][1] + s->sprite_delta[1][0] * s->mb_x * 8 +
174 s->sprite_delta[1][1] * s->mb_y * 8;
175
176 ptr = ref_picture[1];
177 s->mdsp.gmc(dest_cb, ptr, uvlinesize, 8,
178 ox, oy,
179 s->sprite_delta[0][0], s->sprite_delta[0][1],
180 s->sprite_delta[1][0], s->sprite_delta[1][1],
181 a + 1, (1 << (2 * a + 1)) - s->no_rounding,
182 (s->h_edge_pos + 1) >> 1, (s->v_edge_pos + 1) >> 1);
183
184 ptr = ref_picture[2];
185 s->mdsp.gmc(dest_cr, ptr, uvlinesize, 8,
186 ox, oy,
187 s->sprite_delta[0][0], s->sprite_delta[0][1],
188 s->sprite_delta[1][0], s->sprite_delta[1][1],
189 a + 1, (1 << (2 * a + 1)) - s->no_rounding,
190 (s->h_edge_pos + 1) >> 1, (s->v_edge_pos + 1) >> 1);
191}
192
193static inline int hpel_motion(MpegEncContext *s,
194 uint8_t *dest, uint8_t *src,
195 int src_x, int src_y,
196 op_pixels_func *pix_op,
197 int motion_x, int motion_y)
198{
199 int dxy = 0;
200 int emu = 0;
201
202 src_x += motion_x >> 1;
203 src_y += motion_y >> 1;
204
205 /* WARNING: do no forget half pels */
206 src_x = av_clip(src_x, -16, s->width); // FIXME unneeded for emu?
207 if (src_x != s->width)
208 dxy |= motion_x & 1;
209 src_y = av_clip(src_y, -16, s->height);
210 if (src_y != s->height)
211 dxy |= (motion_y & 1) << 1;
212 src += src_y * s->linesize + src_x;
213
214 if ((unsigned)src_x >= FFMAX(s->h_edge_pos - (motion_x & 1) - 7, 0) ||
215 (unsigned)src_y >= FFMAX(s->v_edge_pos - (motion_y & 1) - 7, 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,
220 s->h_edge_pos, s->v_edge_pos);
221 src = s->sc.edge_emu_buffer;
222 emu = 1;
223 }
224 pix_op[dxy](dest, src, s->linesize, 8);
225 return emu;
226}
227
228static av_always_inline
229void 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 v_edge_pos = s->v_edge_pos >> field_based;
250 linesize = s->current_picture.f->linesize[0] << field_based;
251 uvlinesize = s->current_picture.f->linesize[1] << field_based;
252
253 dxy = ((motion_y & 1) << 1) | (motion_x & 1);
254 src_x = s->mb_x * 16 + (motion_x >> 1);
255 src_y = (mb_y << (4 - field_based)) + (motion_y >> 1);
256
257 if (!is_mpeg12 && s->out_format == FMT_H263) {
258 if ((s->workaround_bugs & FF_BUG_HPEL_CHROMA) && field_based) {
259 mx = (motion_x >> 1) | (motion_x & 1);
260 my = motion_y >> 1;
261 uvdxy = ((my & 1) << 1) | (mx & 1);
262 uvsrc_x = s->mb_x * 8 + (mx >> 1);
263 uvsrc_y = (mb_y << (3 - field_based)) + (my >> 1);
264 } else {
265 uvdxy = dxy | (motion_y & 2) | ((motion_x & 2) >> 1);
266 uvsrc_x = src_x >> 1;
267 uvsrc_y = src_y >> 1;
268 }
269 // Even chroma mv's are full pel in H261
270 } else if (!is_mpeg12 && s->out_format == FMT_H261) {
271 mx = motion_x / 4;
272 my = motion_y / 4;
273 uvdxy = 0;
274 uvsrc_x = s->mb_x * 8 + mx;
275 uvsrc_y = mb_y * 8 + my;
276 } else {
277 if (s->chroma_y_shift) {
278 mx = motion_x / 2;
279 my = motion_y / 2;
280 uvdxy = ((my & 1) << 1) | (mx & 1);
281 uvsrc_x = s->mb_x * 8 + (mx >> 1);
282 uvsrc_y = (mb_y << (3 - field_based)) + (my >> 1);
283 } else {
284 if (s->chroma_x_shift) {
285 // Chroma422
286 mx = motion_x / 2;
287 uvdxy = ((motion_y & 1) << 1) | (mx & 1);
288 uvsrc_x = s->mb_x * 8 + (mx >> 1);
289 uvsrc_y = src_y;
290 } else {
291 // Chroma444
292 uvdxy = dxy;
293 uvsrc_x = src_x;
294 uvsrc_y = src_y;
295 }
296 }
297 }
298
299 ptr_y = ref_picture[0] + src_y * linesize + src_x;
300 ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x;
301 ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x;
302
303 if ((unsigned)src_x >= FFMAX(s->h_edge_pos - (motion_x & 1) - 15 , 0) ||
304 (unsigned)src_y >= FFMAX( v_edge_pos - (motion_y & 1) - h + 1, 0)) {
305 if (is_mpeg12 ||
306 s->codec_id == AV_CODEC_ID_MPEG2VIDEO ||
307 s->codec_id == AV_CODEC_ID_MPEG1VIDEO) {
308 av_log(s->avctx, AV_LOG_DEBUG,
309 "MPEG motion vector out of boundary (%d %d)\n", src_x,
310 src_y);
311 return;
312 }
313 src_y = (unsigned)src_y << field_based;
314 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, ptr_y,
315 s->linesize, s->linesize,
316 17, 17 + field_based,
317 src_x, src_y,
318 s->h_edge_pos, s->v_edge_pos);
319 ptr_y = s->sc.edge_emu_buffer;
320 if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY)) {
321 uint8_t *ubuf = s->sc.edge_emu_buffer + 18 * s->linesize;
322 uint8_t *vbuf = ubuf + 10 * s->uvlinesize;
323 if (s->workaround_bugs & FF_BUG_IEDGE)
324 vbuf -= s->uvlinesize;
325 uvsrc_y = (unsigned)uvsrc_y << field_based;
326 s->vdsp.emulated_edge_mc(ubuf, ptr_cb,
327 s->uvlinesize, s->uvlinesize,
328 9, 9 + field_based,
329 uvsrc_x, uvsrc_y,
330 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
331 s->vdsp.emulated_edge_mc(vbuf, ptr_cr,
332 s->uvlinesize, s->uvlinesize,
333 9, 9 + field_based,
334 uvsrc_x, uvsrc_y,
335 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
336 ptr_cb = ubuf;
337 ptr_cr = vbuf;
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->avctx->flags & AV_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 */
369static 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
387static 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: SIMDify, avg variant, 16x16 version
407static 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 */
460static 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->sc.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
486static 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) - 15 , 0) ||
538 (unsigned)src_y >= FFMAX( v_edge_pos - (motion_y & 3) - h + 1, 0)) {
539 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, ptr_y,
540 s->linesize, s->linesize,
541 17, 17 + field_based,
542 src_x, src_y * (1 << field_based),
543 s->h_edge_pos, s->v_edge_pos);
544 ptr_y = s->sc.edge_emu_buffer;
545 if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY)) {
546 uint8_t *ubuf = s->sc.edge_emu_buffer + 18 * s->linesize;
547 uint8_t *vbuf = ubuf + 10 * s->uvlinesize;
548 if (s->workaround_bugs & FF_BUG_IEDGE)
549 vbuf -= s->uvlinesize;
550 s->vdsp.emulated_edge_mc(ubuf, ptr_cb,
551 s->uvlinesize, s->uvlinesize,
552 9, 9 + field_based,
553 uvsrc_x, uvsrc_y * (1 << field_based),
554 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
555 s->vdsp.emulated_edge_mc(vbuf, ptr_cr,
556 s->uvlinesize, s->uvlinesize,
557 9, 9 + field_based,
558 uvsrc_x, uvsrc_y * (1 << field_based),
559 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
560 ptr_cb = ubuf;
561 ptr_cr = vbuf;
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 * H.263 chroma 4mv motion compensation.
592 */
593static 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) - 7, 0) ||
624 (unsigned)src_y >= FFMAX((s->v_edge_pos >> 1) - (dxy >> 1) - 7, 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
645static 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
659static 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 av_assert2(!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
746static 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) - 7, 0) ||
782 (unsigned)src_y >= FFMAX(s->v_edge_pos - (motion_y & 3) - 7, 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 */
830static 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 && s->pict_type != AV_PICTURE_TYPE_B && !s->first_field
903 || !ref_picture[0]) {
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) && ref_picture[0]) {
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 if (!ref_picture[0]) {
948 ref_picture = s->current_picture_ptr->f->data;
949 }
950 for (i = 0; i < 2; i++) {
951 mpeg_motion(s, dest_y, dest_cb, dest_cr,
952 s->picture_structure != i + 1,
953 ref_picture, pix_op,
954 s->mv[dir][2 * i][0], s->mv[dir][2 * i][1],
955 16, mb_y >> 1);
956
957 // after put we make avg of the same block
958 pix_op = s->hdsp.avg_pixels_tab;
959
960 /* opposite parity is always in the same frame if this is
961 * second field */
962 if (!s->first_field) {
963 ref_picture = s->current_picture_ptr->f->data;
964 }
965 }
966 }
967 break;
968 default: av_assert2(0);
969 }
970}
971
972void ff_mpv_motion(MpegEncContext *s,
973 uint8_t *dest_y, uint8_t *dest_cb,
974 uint8_t *dest_cr, int dir,
975 uint8_t **ref_picture,
976 op_pixels_func (*pix_op)[4],
977 qpel_mc_func (*qpix_op)[16])
978{
979#if !CONFIG_SMALL
980 if (s->out_format == FMT_MPEG1)
981 mpv_motion_internal(s, dest_y, dest_cb, dest_cr, dir,
982 ref_picture, pix_op, qpix_op, 1);
983 else
984#endif
985 mpv_motion_internal(s, dest_y, dest_cb, dest_cr, dir,
986 ref_picture, pix_op, qpix_op, 0);
987}
988
generated by cgit at 2024-05-17 05:05:30 (GMT)