mpegvideo: drop support for real (non-emulated) edges
[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 "dsputil.h"
29 #include "h261.h"
30 #include "mpegvideo.h"
31 #include "mjpegenc.h"
32 #include "msmpeg4.h"
33 #include <limits.h>
34
35 static void gmc1_motion(MpegEncContext *s,
36 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
37 uint8_t **ref_picture)
38 {
39 uint8_t *ptr;
40 int src_x, src_y, motion_x, motion_y;
41 ptrdiff_t offset, linesize, uvlinesize;
42 int emu = 0;
43
44 motion_x = s->sprite_offset[0][0];
45 motion_y = s->sprite_offset[0][1];
46 src_x = s->mb_x * 16 + (motion_x >> (s->sprite_warping_accuracy + 1));
47 src_y = s->mb_y * 16 + (motion_y >> (s->sprite_warping_accuracy + 1));
48 motion_x <<= (3 - s->sprite_warping_accuracy);
49 motion_y <<= (3 - s->sprite_warping_accuracy);
50 src_x = av_clip(src_x, -16, s->width);
51 if (src_x == s->width)
52 motion_x = 0;
53 src_y = av_clip(src_y, -16, s->height);
54 if (src_y == s->height)
55 motion_y = 0;
56
57 linesize = s->linesize;
58 uvlinesize = s->uvlinesize;
59
60 ptr = ref_picture[0] + src_y * linesize + src_x;
61
62 if ((unsigned)src_x >= FFMAX(s->h_edge_pos - 17, 0) ||
63 (unsigned)src_y >= FFMAX(s->v_edge_pos - 17, 0)) {
64 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr,
65 linesize, linesize,
66 17, 17,
67 src_x, src_y,
68 s->h_edge_pos, s->v_edge_pos);
69 ptr = s->edge_emu_buffer;
70 }
71
72 if ((motion_x | motion_y) & 7) {
73 s->dsp.gmc1(dest_y, ptr, linesize, 16,
74 motion_x & 15, motion_y & 15, 128 - s->no_rounding);
75 s->dsp.gmc1(dest_y + 8, ptr + 8, linesize, 16,
76 motion_x & 15, motion_y & 15, 128 - s->no_rounding);
77 } else {
78 int dxy;
79
80 dxy = ((motion_x >> 3) & 1) | ((motion_y >> 2) & 2);
81 if (s->no_rounding) {
82 s->hdsp.put_no_rnd_pixels_tab[0][dxy](dest_y, ptr, linesize, 16);
83 } else {
84 s->hdsp.put_pixels_tab[0][dxy](dest_y, ptr, linesize, 16);
85 }
86 }
87
88 if (CONFIG_GRAY && s->flags & CODEC_FLAG_GRAY)
89 return;
90
91 motion_x = s->sprite_offset[1][0];
92 motion_y = s->sprite_offset[1][1];
93 src_x = s->mb_x * 8 + (motion_x >> (s->sprite_warping_accuracy + 1));
94 src_y = s->mb_y * 8 + (motion_y >> (s->sprite_warping_accuracy + 1));
95 motion_x <<= (3 - s->sprite_warping_accuracy);
96 motion_y <<= (3 - s->sprite_warping_accuracy);
97 src_x = av_clip(src_x, -8, s->width >> 1);
98 if (src_x == s->width >> 1)
99 motion_x = 0;
100 src_y = av_clip(src_y, -8, s->height >> 1);
101 if (src_y == s->height >> 1)
102 motion_y = 0;
103
104 offset = (src_y * uvlinesize) + src_x;
105 ptr = ref_picture[1] + offset;
106 if ((unsigned)src_x >= FFMAX((s->h_edge_pos >> 1) - 9, 0) ||
107 (unsigned)src_y >= FFMAX((s->v_edge_pos >> 1) - 9, 0)) {
108 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr,
109 uvlinesize, uvlinesize,
110 9, 9,
111 src_x, src_y,
112 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
113 ptr = s->edge_emu_buffer;
114 emu = 1;
115 }
116 s->dsp.gmc1(dest_cb, ptr, uvlinesize, 8,
117 motion_x & 15, motion_y & 15, 128 - s->no_rounding);
118
119 ptr = ref_picture[2] + offset;
120 if (emu) {
121 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr,
122 uvlinesize, uvlinesize,
123 9, 9,
124 src_x, src_y,
125 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
126 ptr = s->edge_emu_buffer;
127 }
128 s->dsp.gmc1(dest_cr, ptr, uvlinesize, 8,
129 motion_x & 15, motion_y & 15, 128 - s->no_rounding);
130 }
131
132 static void gmc_motion(MpegEncContext *s,
133 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
134 uint8_t **ref_picture)
135 {
136 uint8_t *ptr;
137 int linesize, uvlinesize;
138 const int a = s->sprite_warping_accuracy;
139 int ox, oy;
140
141 linesize = s->linesize;
142 uvlinesize = s->uvlinesize;
143
144 ptr = ref_picture[0];
145
146 ox = s->sprite_offset[0][0] + s->sprite_delta[0][0] * s->mb_x * 16 +
147 s->sprite_delta[0][1] * s->mb_y * 16;
148 oy = s->sprite_offset[0][1] + s->sprite_delta[1][0] * s->mb_x * 16 +
149 s->sprite_delta[1][1] * s->mb_y * 16;
150
151 s->dsp.gmc(dest_y, ptr, linesize, 16,
152 ox, oy,
153 s->sprite_delta[0][0], s->sprite_delta[0][1],
154 s->sprite_delta[1][0], s->sprite_delta[1][1],
155 a + 1, (1 << (2 * a + 1)) - s->no_rounding,
156 s->h_edge_pos, s->v_edge_pos);
157 s->dsp.gmc(dest_y + 8, ptr, linesize, 16,
158 ox + s->sprite_delta[0][0] * 8,
159 oy + s->sprite_delta[1][0] * 8,
160 s->sprite_delta[0][0], s->sprite_delta[0][1],
161 s->sprite_delta[1][0], s->sprite_delta[1][1],
162 a + 1, (1 << (2 * a + 1)) - s->no_rounding,
163 s->h_edge_pos, s->v_edge_pos);
164
165 if (CONFIG_GRAY && s->flags & CODEC_FLAG_GRAY)
166 return;
167
168 ox = s->sprite_offset[1][0] + s->sprite_delta[0][0] * s->mb_x * 8 +
169 s->sprite_delta[0][1] * s->mb_y * 8;
170 oy = s->sprite_offset[1][1] + s->sprite_delta[1][0] * s->mb_x * 8 +
171 s->sprite_delta[1][1] * s->mb_y * 8;
172
173 ptr = ref_picture[1];
174 s->dsp.gmc(dest_cb, ptr, uvlinesize, 8,
175 ox, oy,
176 s->sprite_delta[0][0], s->sprite_delta[0][1],
177 s->sprite_delta[1][0], s->sprite_delta[1][1],
178 a + 1, (1 << (2 * a + 1)) - s->no_rounding,
179 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
180
181 ptr = ref_picture[2];
182 s->dsp.gmc(dest_cr, ptr, uvlinesize, 8,
183 ox, oy,
184 s->sprite_delta[0][0], s->sprite_delta[0][1],
185 s->sprite_delta[1][0], s->sprite_delta[1][1],
186 a + 1, (1 << (2 * a + 1)) - s->no_rounding,
187 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
188 }
189
190 static inline int hpel_motion(MpegEncContext *s,
191 uint8_t *dest, uint8_t *src,
192 int src_x, int src_y,
193 op_pixels_func *pix_op,
194 int motion_x, int motion_y)
195 {
196 int dxy = 0;
197 int emu = 0;
198
199 src_x += motion_x >> 1;
200 src_y += motion_y >> 1;
201
202 /* WARNING: do no forget half pels */
203 src_x = av_clip(src_x, -16, s->width); // FIXME unneeded for emu?
204 if (src_x != s->width)
205 dxy |= motion_x & 1;
206 src_y = av_clip(src_y, -16, s->height);
207 if (src_y != s->height)
208 dxy |= (motion_y & 1) << 1;
209 src += src_y * s->linesize + src_x;
210
211 if (s->unrestricted_mv) {
212 if ((unsigned)src_x > FFMAX(s->h_edge_pos - (motion_x & 1) - 8, 0) ||
213 (unsigned)src_y > FFMAX(s->v_edge_pos - (motion_y & 1) - 8, 0)) {
214 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, src,
215 s->linesize, s->linesize,
216 9, 9,
217 src_x, src_y, s->h_edge_pos,
218 s->v_edge_pos);
219 src = s->edge_emu_buffer;
220 emu = 1;
221 }
222 }
223 pix_op[dxy](dest, src, s->linesize, 8);
224 return emu;
225 }
226
227 static av_always_inline
228 void mpeg_motion_internal(MpegEncContext *s,
229 uint8_t *dest_y,
230 uint8_t *dest_cb,
231 uint8_t *dest_cr,
232 int field_based,
233 int bottom_field,
234 int field_select,
235 uint8_t **ref_picture,
236 op_pixels_func (*pix_op)[4],
237 int motion_x,
238 int motion_y,
239 int h,
240 int is_mpeg12,
241 int mb_y)
242 {
243 uint8_t *ptr_y, *ptr_cb, *ptr_cr;
244 int dxy, uvdxy, mx, my, src_x, src_y,
245 uvsrc_x, uvsrc_y, v_edge_pos;
246 ptrdiff_t uvlinesize, linesize;
247
248 #if 0
249 if (s->quarter_sample) {
250 motion_x >>= 1;
251 motion_y >>= 1;
252 }
253 #endif
254
255 v_edge_pos = s->v_edge_pos >> field_based;
256 linesize = s->current_picture.f.linesize[0] << field_based;
257 uvlinesize = s->current_picture.f.linesize[1] << field_based;
258
259 dxy = ((motion_y & 1) << 1) | (motion_x & 1);
260 src_x = s->mb_x * 16 + (motion_x >> 1);
261 src_y = (mb_y << (4 - field_based)) + (motion_y >> 1);
262
263 if (!is_mpeg12 && s->out_format == FMT_H263) {
264 if ((s->workaround_bugs & FF_BUG_HPEL_CHROMA) && field_based) {
265 mx = (motion_x >> 1) | (motion_x & 1);
266 my = motion_y >> 1;
267 uvdxy = ((my & 1) << 1) | (mx & 1);
268 uvsrc_x = s->mb_x * 8 + (mx >> 1);
269 uvsrc_y = (mb_y << (3 - field_based)) + (my >> 1);
270 } else {
271 uvdxy = dxy | (motion_y & 2) | ((motion_x & 2) >> 1);
272 uvsrc_x = src_x >> 1;
273 uvsrc_y = src_y >> 1;
274 }
275 // Even chroma mv's are full pel in H261
276 } else if (!is_mpeg12 && s->out_format == FMT_H261) {
277 mx = motion_x / 4;
278 my = motion_y / 4;
279 uvdxy = 0;
280 uvsrc_x = s->mb_x * 8 + mx;
281 uvsrc_y = mb_y * 8 + my;
282 } else {
283 if (s->chroma_y_shift) {
284 mx = motion_x / 2;
285 my = motion_y / 2;
286 uvdxy = ((my & 1) << 1) | (mx & 1);
287 uvsrc_x = s->mb_x * 8 + (mx >> 1);
288 uvsrc_y = (mb_y << (3 - field_based)) + (my >> 1);
289 } else {
290 if (s->chroma_x_shift) {
291 // Chroma422
292 mx = motion_x / 2;
293 uvdxy = ((motion_y & 1) << 1) | (mx & 1);
294 uvsrc_x = s->mb_x * 8 + (mx >> 1);
295 uvsrc_y = src_y;
296 } else {
297 // Chroma444
298 uvdxy = dxy;
299 uvsrc_x = src_x;
300 uvsrc_y = src_y;
301 }
302 }
303 }
304
305 ptr_y = ref_picture[0] + src_y * linesize + src_x;
306 ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x;
307 ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x;
308
309 if ((unsigned)src_x > FFMAX(s->h_edge_pos - (motion_x & 1) - 16, 0) ||
310 (unsigned)src_y > FFMAX(v_edge_pos - (motion_y & 1) - h, 0)) {
311 if (is_mpeg12 ||
312 s->codec_id == AV_CODEC_ID_MPEG2VIDEO ||
313 s->codec_id == AV_CODEC_ID_MPEG1VIDEO) {
314 av_log(s->avctx, AV_LOG_DEBUG,
315 "MPEG motion vector out of boundary (%d %d)\n", src_x,
316 src_y);
317 return;
318 }
319 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr_y,
320 s->linesize, s->linesize,
321 17, 17 + field_based,
322 src_x, src_y << field_based,
323 s->h_edge_pos, s->v_edge_pos);
324 ptr_y = s->edge_emu_buffer;
325 if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
326 uint8_t *uvbuf = s->edge_emu_buffer + 18 * s->linesize;
327 s->vdsp.emulated_edge_mc(uvbuf, ptr_cb,
328 s->uvlinesize, s->uvlinesize,
329 9, 9 + field_based,
330 uvsrc_x, uvsrc_y << field_based,
331 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
332 s->vdsp.emulated_edge_mc(uvbuf + 16, ptr_cr,
333 s->uvlinesize, s->uvlinesize,
334 9, 9 + field_based,
335 uvsrc_x, uvsrc_y << field_based,
336 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
337 ptr_cb = uvbuf;
338 ptr_cr = uvbuf + 16;
339 }
340 }
341
342 /* FIXME use this for field pix too instead of the obnoxious hack which
343 * changes picture.data */
344 if (bottom_field) {
345 dest_y += s->linesize;
346 dest_cb += s->uvlinesize;
347 dest_cr += s->uvlinesize;
348 }
349
350 if (field_select) {
351 ptr_y += s->linesize;
352 ptr_cb += s->uvlinesize;
353 ptr_cr += s->uvlinesize;
354 }
355
356 pix_op[0][dxy](dest_y, ptr_y, linesize, h);
357
358 if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
359 pix_op[s->chroma_x_shift][uvdxy]
360 (dest_cb, ptr_cb, uvlinesize, h >> s->chroma_y_shift);
361 pix_op[s->chroma_x_shift][uvdxy]
362 (dest_cr, ptr_cr, uvlinesize, h >> s->chroma_y_shift);
363 }
364 if (!is_mpeg12 && (CONFIG_H261_ENCODER || CONFIG_H261_DECODER) &&
365 s->out_format == FMT_H261) {
366 ff_h261_loop_filter(s);
367 }
368 }
369 /* apply one mpeg motion vector to the three components */
370 static void mpeg_motion(MpegEncContext *s,
371 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
372 int field_select, uint8_t **ref_picture,
373 op_pixels_func (*pix_op)[4],
374 int motion_x, int motion_y, int h, int mb_y)
375 {
376 #if !CONFIG_SMALL
377 if (s->out_format == FMT_MPEG1)
378 mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, 0, 0,
379 field_select, ref_picture, pix_op,
380 motion_x, motion_y, h, 1, mb_y);
381 else
382 #endif
383 mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, 0, 0,
384 field_select, ref_picture, pix_op,
385 motion_x, motion_y, h, 0, mb_y);
386 }
387
388 static void mpeg_motion_field(MpegEncContext *s, uint8_t *dest_y,
389 uint8_t *dest_cb, uint8_t *dest_cr,
390 int bottom_field, int field_select,
391 uint8_t **ref_picture,
392 op_pixels_func (*pix_op)[4],
393 int motion_x, int motion_y, int h, int mb_y)
394 {
395 #if !CONFIG_SMALL
396 if(s->out_format == FMT_MPEG1)
397 mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, 1,
398 bottom_field, field_select, ref_picture, pix_op,
399 motion_x, motion_y, h, 1, mb_y);
400 else
401 #endif
402 mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, 1,
403 bottom_field, field_select, ref_picture, pix_op,
404 motion_x, motion_y, h, 0, mb_y);
405 }
406
407 // FIXME move to dsputil, avg variant, 16x16 version
408 static inline void put_obmc(uint8_t *dst, uint8_t *src[5], int stride)
409 {
410 int x;
411 uint8_t *const top = src[1];
412 uint8_t *const left = src[2];
413 uint8_t *const mid = src[0];
414 uint8_t *const right = src[3];
415 uint8_t *const bottom = src[4];
416 #define OBMC_FILTER(x, t, l, m, r, b)\
417 dst[x]= (t*top[x] + l*left[x] + m*mid[x] + r*right[x] + b*bottom[x] + 4)>>3
418 #define OBMC_FILTER4(x, t, l, m, r, b)\
419 OBMC_FILTER(x , t, l, m, r, b);\
420 OBMC_FILTER(x+1 , t, l, m, r, b);\
421 OBMC_FILTER(x +stride, t, l, m, r, b);\
422 OBMC_FILTER(x+1+stride, t, l, m, r, b);
423
424 x = 0;
425 OBMC_FILTER (x , 2, 2, 4, 0, 0);
426 OBMC_FILTER (x + 1, 2, 1, 5, 0, 0);
427 OBMC_FILTER4(x + 2, 2, 1, 5, 0, 0);
428 OBMC_FILTER4(x + 4, 2, 0, 5, 1, 0);
429 OBMC_FILTER (x + 6, 2, 0, 5, 1, 0);
430 OBMC_FILTER (x + 7, 2, 0, 4, 2, 0);
431 x += stride;
432 OBMC_FILTER (x , 1, 2, 5, 0, 0);
433 OBMC_FILTER (x + 1, 1, 2, 5, 0, 0);
434 OBMC_FILTER (x + 6, 1, 0, 5, 2, 0);
435 OBMC_FILTER (x + 7, 1, 0, 5, 2, 0);
436 x += stride;
437 OBMC_FILTER4(x , 1, 2, 5, 0, 0);
438 OBMC_FILTER4(x + 2, 1, 1, 6, 0, 0);
439 OBMC_FILTER4(x + 4, 1, 0, 6, 1, 0);
440 OBMC_FILTER4(x + 6, 1, 0, 5, 2, 0);
441 x += 2 * stride;
442 OBMC_FILTER4(x , 0, 2, 5, 0, 1);
443 OBMC_FILTER4(x + 2, 0, 1, 6, 0, 1);
444 OBMC_FILTER4(x + 4, 0, 0, 6, 1, 1);
445 OBMC_FILTER4(x + 6, 0, 0, 5, 2, 1);
446 x += 2*stride;
447 OBMC_FILTER (x , 0, 2, 5, 0, 1);
448 OBMC_FILTER (x + 1, 0, 2, 5, 0, 1);
449 OBMC_FILTER4(x + 2, 0, 1, 5, 0, 2);
450 OBMC_FILTER4(x + 4, 0, 0, 5, 1, 2);
451 OBMC_FILTER (x + 6, 0, 0, 5, 2, 1);
452 OBMC_FILTER (x + 7, 0, 0, 5, 2, 1);
453 x += stride;
454 OBMC_FILTER (x , 0, 2, 4, 0, 2);
455 OBMC_FILTER (x + 1, 0, 1, 5, 0, 2);
456 OBMC_FILTER (x + 6, 0, 0, 5, 1, 2);
457 OBMC_FILTER (x + 7, 0, 0, 4, 2, 2);
458 }
459
460 /* obmc for 1 8x8 luma block */
461 static inline void obmc_motion(MpegEncContext *s,
462 uint8_t *dest, uint8_t *src,
463 int src_x, int src_y,
464 op_pixels_func *pix_op,
465 int16_t mv[5][2] /* mid top left right bottom */)
466 #define MID 0
467 {
468 int i;
469 uint8_t *ptr[5];
470
471 assert(s->quarter_sample == 0);
472
473 for (i = 0; i < 5; i++) {
474 if (i && mv[i][0] == mv[MID][0] && mv[i][1] == mv[MID][1]) {
475 ptr[i] = ptr[MID];
476 } else {
477 ptr[i] = s->obmc_scratchpad + 8 * (i & 1) +
478 s->linesize * 8 * (i >> 1);
479 hpel_motion(s, ptr[i], src, src_x, src_y, pix_op,
480 mv[i][0], mv[i][1]);
481 }
482 }
483
484 put_obmc(dest, ptr, s->linesize);
485 }
486
487 static inline void qpel_motion(MpegEncContext *s,
488 uint8_t *dest_y,
489 uint8_t *dest_cb,
490 uint8_t *dest_cr,
491 int field_based, int bottom_field,
492 int field_select, uint8_t **ref_picture,
493 op_pixels_func (*pix_op)[4],
494 qpel_mc_func (*qpix_op)[16],
495 int motion_x, int motion_y, int h)
496 {
497 uint8_t *ptr_y, *ptr_cb, *ptr_cr;
498 int dxy, uvdxy, mx, my, src_x, src_y, uvsrc_x, uvsrc_y, v_edge_pos;
499 ptrdiff_t linesize, uvlinesize;
500
501 dxy = ((motion_y & 3) << 2) | (motion_x & 3);
502
503 src_x = s->mb_x * 16 + (motion_x >> 2);
504 src_y = s->mb_y * (16 >> field_based) + (motion_y >> 2);
505
506 v_edge_pos = s->v_edge_pos >> field_based;
507 linesize = s->linesize << field_based;
508 uvlinesize = s->uvlinesize << field_based;
509
510 if (field_based) {
511 mx = motion_x / 2;
512 my = motion_y >> 1;
513 } else if (s->workaround_bugs & FF_BUG_QPEL_CHROMA2) {
514 static const int rtab[8] = { 0, 0, 1, 1, 0, 0, 0, 1 };
515 mx = (motion_x >> 1) + rtab[motion_x & 7];
516 my = (motion_y >> 1) + rtab[motion_y & 7];
517 } else if (s->workaround_bugs & FF_BUG_QPEL_CHROMA) {
518 mx = (motion_x >> 1) | (motion_x & 1);
519 my = (motion_y >> 1) | (motion_y & 1);
520 } else {
521 mx = motion_x / 2;
522 my = motion_y / 2;
523 }
524 mx = (mx >> 1) | (mx & 1);
525 my = (my >> 1) | (my & 1);
526
527 uvdxy = (mx & 1) | ((my & 1) << 1);
528 mx >>= 1;
529 my >>= 1;
530
531 uvsrc_x = s->mb_x * 8 + mx;
532 uvsrc_y = s->mb_y * (8 >> field_based) + my;
533
534 ptr_y = ref_picture[0] + src_y * linesize + src_x;
535 ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x;
536 ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x;
537
538 if ((unsigned)src_x > FFMAX(s->h_edge_pos - (motion_x & 3) - 16, 0) ||
539 (unsigned)src_y > FFMAX(v_edge_pos - (motion_y & 3) - h, 0)) {
540 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr_y,
541 s->linesize, s->linesize,
542 17, 17 + field_based,
543 src_x, src_y << field_based,
544 s->h_edge_pos, s->v_edge_pos);
545 ptr_y = s->edge_emu_buffer;
546 if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
547 uint8_t *uvbuf = s->edge_emu_buffer + 18 * s->linesize;
548 s->vdsp.emulated_edge_mc(uvbuf, ptr_cb,
549 s->uvlinesize, s->uvlinesize,
550 9, 9 + field_based,
551 uvsrc_x, uvsrc_y << field_based,
552 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
553 s->vdsp.emulated_edge_mc(uvbuf + 16, ptr_cr,
554 s->uvlinesize, s->uvlinesize,
555 9, 9 + field_based,
556 uvsrc_x, uvsrc_y << field_based,
557 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
558 ptr_cb = uvbuf;
559 ptr_cr = uvbuf + 16;
560 }
561 }
562
563 if (!field_based)
564 qpix_op[0][dxy](dest_y, ptr_y, linesize);
565 else {
566 if (bottom_field) {
567 dest_y += s->linesize;
568 dest_cb += s->uvlinesize;
569 dest_cr += s->uvlinesize;
570 }
571
572 if (field_select) {
573 ptr_y += s->linesize;
574 ptr_cb += s->uvlinesize;
575 ptr_cr += s->uvlinesize;
576 }
577 // damn interlaced mode
578 // FIXME boundary mirroring is not exactly correct here
579 qpix_op[1][dxy](dest_y, ptr_y, linesize);
580 qpix_op[1][dxy](dest_y + 8, ptr_y + 8, linesize);
581 }
582 if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
583 pix_op[1][uvdxy](dest_cr, ptr_cr, uvlinesize, h >> 1);
584 pix_op[1][uvdxy](dest_cb, ptr_cb, uvlinesize, h >> 1);
585 }
586 }
587
588 /**
589 * h263 chroma 4mv motion compensation.
590 */
591 static void chroma_4mv_motion(MpegEncContext *s,
592 uint8_t *dest_cb, uint8_t *dest_cr,
593 uint8_t **ref_picture,
594 op_pixels_func *pix_op,
595 int mx, int my)
596 {
597 uint8_t *ptr;
598 int src_x, src_y, dxy, emu = 0;
599 ptrdiff_t offset;
600
601 /* In case of 8X8, we construct a single chroma motion vector
602 * with a special rounding */
603 mx = ff_h263_round_chroma(mx);
604 my = ff_h263_round_chroma(my);
605
606 dxy = ((my & 1) << 1) | (mx & 1);
607 mx >>= 1;
608 my >>= 1;
609
610 src_x = s->mb_x * 8 + mx;
611 src_y = s->mb_y * 8 + my;
612 src_x = av_clip(src_x, -8, (s->width >> 1));
613 if (src_x == (s->width >> 1))
614 dxy &= ~1;
615 src_y = av_clip(src_y, -8, (s->height >> 1));
616 if (src_y == (s->height >> 1))
617 dxy &= ~2;
618
619 offset = src_y * s->uvlinesize + src_x;
620 ptr = ref_picture[1] + offset;
621 if ((unsigned)src_x > FFMAX((s->h_edge_pos >> 1) - (dxy & 1) - 8, 0) ||
622 (unsigned)src_y > FFMAX((s->v_edge_pos >> 1) - (dxy >> 1) - 8, 0)) {
623 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr,
624 s->uvlinesize, s->uvlinesize,
625 9, 9, src_x, src_y,
626 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
627 ptr = s->edge_emu_buffer;
628 emu = 1;
629 }
630 pix_op[dxy](dest_cb, ptr, s->uvlinesize, 8);
631
632 ptr = ref_picture[2] + offset;
633 if (emu) {
634 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr,
635 s->uvlinesize, s->uvlinesize,
636 9, 9, src_x, src_y,
637 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
638 ptr = s->edge_emu_buffer;
639 }
640 pix_op[dxy](dest_cr, ptr, s->uvlinesize, 8);
641 }
642
643 static inline void prefetch_motion(MpegEncContext *s, uint8_t **pix, int dir)
644 {
645 /* fetch pixels for estimated mv 4 macroblocks ahead
646 * optimized for 64byte cache lines */
647 const int shift = s->quarter_sample ? 2 : 1;
648 const int mx = (s->mv[dir][0][0] >> shift) + 16 * s->mb_x + 8;
649 const int my = (s->mv[dir][0][1] >> shift) + 16 * s->mb_y;
650 int off = mx + (my + (s->mb_x & 3) * 4) * s->linesize + 64;
651
652 s->vdsp.prefetch(pix[0] + off, s->linesize, 4);
653 off = (mx >> 1) + ((my >> 1) + (s->mb_x & 7)) * s->uvlinesize + 64;
654 s->vdsp.prefetch(pix[1] + off, pix[2] - pix[1], 2);
655 }
656
657 static inline void apply_obmc(MpegEncContext *s,
658 uint8_t *dest_y,
659 uint8_t *dest_cb,
660 uint8_t *dest_cr,
661 uint8_t **ref_picture,
662 op_pixels_func (*pix_op)[4])
663 {
664 LOCAL_ALIGNED_8(int16_t, mv_cache, [4], [4][2]);
665 Picture *cur_frame = &s->current_picture;
666 int mb_x = s->mb_x;
667 int mb_y = s->mb_y;
668 const int xy = mb_x + mb_y * s->mb_stride;
669 const int mot_stride = s->b8_stride;
670 const int mot_xy = mb_x * 2 + mb_y * 2 * mot_stride;
671 int mx, my, i;
672
673 assert(!s->mb_skipped);
674
675 AV_COPY32(mv_cache[1][1], cur_frame->motion_val[0][mot_xy]);
676 AV_COPY32(mv_cache[1][2], cur_frame->motion_val[0][mot_xy + 1]);
677
678 AV_COPY32(mv_cache[2][1],
679 cur_frame->motion_val[0][mot_xy + mot_stride]);
680 AV_COPY32(mv_cache[2][2],
681 cur_frame->motion_val[0][mot_xy + mot_stride + 1]);
682
683 AV_COPY32(mv_cache[3][1],
684 cur_frame->motion_val[0][mot_xy + mot_stride]);
685 AV_COPY32(mv_cache[3][2],
686 cur_frame->motion_val[0][mot_xy + mot_stride + 1]);
687
688 if (mb_y == 0 || IS_INTRA(cur_frame->mb_type[xy - s->mb_stride])) {
689 AV_COPY32(mv_cache[0][1], mv_cache[1][1]);
690 AV_COPY32(mv_cache[0][2], mv_cache[1][2]);
691 } else {
692 AV_COPY32(mv_cache[0][1],
693 cur_frame->motion_val[0][mot_xy - mot_stride]);
694 AV_COPY32(mv_cache[0][2],
695 cur_frame->motion_val[0][mot_xy - mot_stride + 1]);
696 }
697
698 if (mb_x == 0 || IS_INTRA(cur_frame->mb_type[xy - 1])) {
699 AV_COPY32(mv_cache[1][0], mv_cache[1][1]);
700 AV_COPY32(mv_cache[2][0], mv_cache[2][1]);
701 } else {
702 AV_COPY32(mv_cache[1][0], cur_frame->motion_val[0][mot_xy - 1]);
703 AV_COPY32(mv_cache[2][0],
704 cur_frame->motion_val[0][mot_xy - 1 + mot_stride]);
705 }
706
707 if (mb_x + 1 >= s->mb_width || IS_INTRA(cur_frame->mb_type[xy + 1])) {
708 AV_COPY32(mv_cache[1][3], mv_cache[1][2]);
709 AV_COPY32(mv_cache[2][3], mv_cache[2][2]);
710 } else {
711 AV_COPY32(mv_cache[1][3], cur_frame->motion_val[0][mot_xy + 2]);
712 AV_COPY32(mv_cache[2][3],
713 cur_frame->motion_val[0][mot_xy + 2 + mot_stride]);
714 }
715
716 mx = 0;
717 my = 0;
718 for (i = 0; i < 4; i++) {
719 const int x = (i & 1) + 1;
720 const int y = (i >> 1) + 1;
721 int16_t mv[5][2] = {
722 { mv_cache[y][x][0], mv_cache[y][x][1] },
723 { mv_cache[y - 1][x][0], mv_cache[y - 1][x][1] },
724 { mv_cache[y][x - 1][0], mv_cache[y][x - 1][1] },
725 { mv_cache[y][x + 1][0], mv_cache[y][x + 1][1] },
726 { mv_cache[y + 1][x][0], mv_cache[y + 1][x][1] }
727 };
728 // FIXME cleanup
729 obmc_motion(s, dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize,
730 ref_picture[0],
731 mb_x * 16 + (i & 1) * 8, mb_y * 16 + (i >> 1) * 8,
732 pix_op[1],
733 mv);
734
735 mx += mv[0][0];
736 my += mv[0][1];
737 }
738 if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY))
739 chroma_4mv_motion(s, dest_cb, dest_cr,
740 ref_picture, pix_op[1],
741 mx, my);
742 }
743
744 static inline void apply_8x8(MpegEncContext *s,
745 uint8_t *dest_y,
746 uint8_t *dest_cb,
747 uint8_t *dest_cr,
748 int dir,
749 uint8_t **ref_picture,
750 qpel_mc_func (*qpix_op)[16],
751 op_pixels_func (*pix_op)[4])
752 {
753 int dxy, mx, my, src_x, src_y;
754 int i;
755 int mb_x = s->mb_x;
756 int mb_y = s->mb_y;
757 uint8_t *ptr, *dest;
758
759 mx = 0;
760 my = 0;
761 if (s->quarter_sample) {
762 for (i = 0; i < 4; i++) {
763 int motion_x = s->mv[dir][i][0];
764 int motion_y = s->mv[dir][i][1];
765
766 dxy = ((motion_y & 3) << 2) | (motion_x & 3);
767 src_x = mb_x * 16 + (motion_x >> 2) + (i & 1) * 8;
768 src_y = mb_y * 16 + (motion_y >> 2) + (i >> 1) * 8;
769
770 /* WARNING: do no forget half pels */
771 src_x = av_clip(src_x, -16, s->width);
772 if (src_x == s->width)
773 dxy &= ~3;
774 src_y = av_clip(src_y, -16, s->height);
775 if (src_y == s->height)
776 dxy &= ~12;
777
778 ptr = ref_picture[0] + (src_y * s->linesize) + (src_x);
779 if ((unsigned)src_x > FFMAX(s->h_edge_pos - (motion_x & 3) - 8, 0) ||
780 (unsigned)src_y > FFMAX(s->v_edge_pos - (motion_y & 3) - 8, 0)) {
781 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr,
782 s->linesize, s->linesize,
783 9, 9,
784 src_x, src_y,
785 s->h_edge_pos,
786 s->v_edge_pos);
787 ptr = s->edge_emu_buffer;
788 }
789 dest = dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize;
790 qpix_op[1][dxy](dest, ptr, s->linesize);
791
792 mx += s->mv[dir][i][0] / 2;
793 my += s->mv[dir][i][1] / 2;
794 }
795 } else {
796 for (i = 0; i < 4; i++) {
797 hpel_motion(s,
798 dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize,
799 ref_picture[0],
800 mb_x * 16 + (i & 1) * 8,
801 mb_y * 16 + (i >> 1) * 8,
802 pix_op[1],
803 s->mv[dir][i][0],
804 s->mv[dir][i][1]);
805
806 mx += s->mv[dir][i][0];
807 my += s->mv[dir][i][1];
808 }
809 }
810
811 if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY))
812 chroma_4mv_motion(s, dest_cb, dest_cr,
813 ref_picture, pix_op[1], mx, my);
814 }
815
816 /**
817 * motion compensation of a single macroblock
818 * @param s context
819 * @param dest_y luma destination pointer
820 * @param dest_cb chroma cb/u destination pointer
821 * @param dest_cr chroma cr/v destination pointer
822 * @param dir direction (0->forward, 1->backward)
823 * @param ref_picture array[3] of pointers to the 3 planes of the reference picture
824 * @param pix_op halfpel motion compensation function (average or put normally)
825 * @param qpix_op qpel motion compensation function (average or put normally)
826 * the motion vectors are taken from s->mv and the MV type from s->mv_type
827 */
828 static av_always_inline void MPV_motion_internal(MpegEncContext *s,
829 uint8_t *dest_y,
830 uint8_t *dest_cb,
831 uint8_t *dest_cr,
832 int dir,
833 uint8_t **ref_picture,
834 op_pixels_func (*pix_op)[4],
835 qpel_mc_func (*qpix_op)[16],
836 int is_mpeg12)
837 {
838 int i;
839 int mb_y = s->mb_y;
840
841 prefetch_motion(s, ref_picture, dir);
842
843 if (!is_mpeg12 && s->obmc && s->pict_type != AV_PICTURE_TYPE_B) {
844 apply_obmc(s, dest_y, dest_cb, dest_cr, ref_picture, pix_op);
845 return;
846 }
847
848 switch (s->mv_type) {
849 case MV_TYPE_16X16:
850 if (s->mcsel) {
851 if (s->real_sprite_warping_points == 1) {
852 gmc1_motion(s, dest_y, dest_cb, dest_cr,
853 ref_picture);
854 } else {
855 gmc_motion(s, dest_y, dest_cb, dest_cr,
856 ref_picture);
857 }
858 } else if (!is_mpeg12 && s->quarter_sample) {
859 qpel_motion(s, dest_y, dest_cb, dest_cr,
860 0, 0, 0,
861 ref_picture, pix_op, qpix_op,
862 s->mv[dir][0][0], s->mv[dir][0][1], 16);
863 } else if (!is_mpeg12 && (CONFIG_WMV2_DECODER || CONFIG_WMV2_ENCODER) &&
864 s->mspel && s->codec_id == AV_CODEC_ID_WMV2) {
865 ff_mspel_motion(s, dest_y, dest_cb, dest_cr,
866 ref_picture, pix_op,
867 s->mv[dir][0][0], s->mv[dir][0][1], 16);
868 } else {
869 mpeg_motion(s, dest_y, dest_cb, dest_cr, 0,
870 ref_picture, pix_op,
871 s->mv[dir][0][0], s->mv[dir][0][1], 16, mb_y);
872 }
873 break;
874 case MV_TYPE_8X8:
875 if (!is_mpeg12)
876 apply_8x8(s, dest_y, dest_cb, dest_cr,
877 dir, ref_picture, qpix_op, pix_op);
878 break;
879 case MV_TYPE_FIELD:
880 if (s->picture_structure == PICT_FRAME) {
881 if (!is_mpeg12 && s->quarter_sample) {
882 for (i = 0; i < 2; i++)
883 qpel_motion(s, dest_y, dest_cb, dest_cr,
884 1, i, s->field_select[dir][i],
885 ref_picture, pix_op, qpix_op,
886 s->mv[dir][i][0], s->mv[dir][i][1], 8);
887 } else {
888 /* top field */
889 mpeg_motion_field(s, dest_y, dest_cb, dest_cr,
890 0, s->field_select[dir][0],
891 ref_picture, pix_op,
892 s->mv[dir][0][0], s->mv[dir][0][1], 8, mb_y);
893 /* bottom field */
894 mpeg_motion_field(s, dest_y, dest_cb, dest_cr,
895 1, s->field_select[dir][1],
896 ref_picture, pix_op,
897 s->mv[dir][1][0], s->mv[dir][1][1], 8, mb_y);
898 }
899 } else {
900 if (s->picture_structure != s->field_select[dir][0] + 1 &&
901 s->pict_type != AV_PICTURE_TYPE_B && !s->first_field) {
902 ref_picture = s->current_picture_ptr->f.data;
903 }
904
905 mpeg_motion(s, dest_y, dest_cb, dest_cr,
906 s->field_select[dir][0],
907 ref_picture, pix_op,
908 s->mv[dir][0][0], s->mv[dir][0][1], 16, mb_y >> 1);
909 }
910 break;
911 case MV_TYPE_16X8:
912 for (i = 0; i < 2; i++) {
913 uint8_t **ref2picture;
914
915 if (s->picture_structure == s->field_select[dir][i] + 1
916 || s->pict_type == AV_PICTURE_TYPE_B || s->first_field) {
917 ref2picture = ref_picture;
918 } else {
919 ref2picture = s->current_picture_ptr->f.data;
920 }
921
922 mpeg_motion(s, dest_y, dest_cb, dest_cr,
923 s->field_select[dir][i],
924 ref2picture, pix_op,
925 s->mv[dir][i][0], s->mv[dir][i][1] + 16 * i,
926 8, mb_y >> 1);
927
928 dest_y += 16 * s->linesize;
929 dest_cb += (16 >> s->chroma_y_shift) * s->uvlinesize;
930 dest_cr += (16 >> s->chroma_y_shift) * s->uvlinesize;
931 }
932 break;
933 case MV_TYPE_DMV:
934 if (s->picture_structure == PICT_FRAME) {
935 for (i = 0; i < 2; i++) {
936 int j;
937 for (j = 0; j < 2; j++)
938 mpeg_motion_field(s, dest_y, dest_cb, dest_cr,
939 j, j ^ i, ref_picture, pix_op,
940 s->mv[dir][2 * i + j][0],
941 s->mv[dir][2 * i + j][1], 8, mb_y);
942 pix_op = s->hdsp.avg_pixels_tab;
943 }
944 } else {
945 for (i = 0; i < 2; i++) {
946 mpeg_motion(s, dest_y, dest_cb, dest_cr,
947 s->picture_structure != i + 1,
948 ref_picture, pix_op,
949 s->mv[dir][2 * i][0], s->mv[dir][2 * i][1],
950 16, mb_y >> 1);
951
952 // after put we make avg of the same block
953 pix_op = s->hdsp.avg_pixels_tab;
954
955 /* opposite parity is always in the same frame if this is
956 * second field */
957 if (!s->first_field) {
958 ref_picture = s->current_picture_ptr->f.data;
959 }
960 }
961 }
962 break;
963 default: assert(0);
964 }
965 }
966
967 void ff_MPV_motion(MpegEncContext *s,
968 uint8_t *dest_y, uint8_t *dest_cb,
969 uint8_t *dest_cr, int dir,
970 uint8_t **ref_picture,
971 op_pixels_func (*pix_op)[4],
972 qpel_mc_func (*qpix_op)[16])
973 {
974 #if !CONFIG_SMALL
975 if (s->out_format == FMT_MPEG1)
976 MPV_motion_internal(s, dest_y, dest_cb, dest_cr, dir,
977 ref_picture, pix_op, qpix_op, 1);
978 else
979 #endif
980 MPV_motion_internal(s, dest_y, dest_cb, dest_cr, dir,
981 ref_picture, pix_op, qpix_op, 0);
982 }