2 * Copyright (c) 2000,2001 Fabrice Bellard
3 * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
5 * 4MV & hq & B-frame encoding stuff by Michael Niedermayer <michaelni@gmx.at>
7 * This file is part of FFmpeg.
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.
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.
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
26 #include "libavutil/avassert.h"
27 #include "libavutil/internal.h"
31 #include "mpegutils.h"
32 #include "mpegvideo.h"
38 static void gmc1_motion(MpegEncContext *s,
39 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
40 uint8_t **ref_picture)
43 int src_x, src_y, motion_x, motion_y;
44 ptrdiff_t offset, linesize, uvlinesize;
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 <<= (3 - s->sprite_warping_accuracy);
52 motion_y <<= (3 - s->sprite_warping_accuracy);
53 src_x = av_clip(src_x, -16, s->width);
54 if (src_x == s->width)
56 src_y = av_clip(src_y, -16, s->height);
57 if (src_y == s->height)
60 linesize = s->linesize;
61 uvlinesize = s->uvlinesize;
63 ptr = ref_picture[0] + src_y * linesize + src_x;
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->edge_emu_buffer, ptr,
71 s->h_edge_pos, s->v_edge_pos);
72 ptr = s->edge_emu_buffer;
75 if ((motion_x | motion_y) & 7) {
76 s->dsp.gmc1(dest_y, ptr, linesize, 16,
77 motion_x & 15, motion_y & 15, 128 - s->no_rounding);
78 s->dsp.gmc1(dest_y + 8, ptr + 8, linesize, 16,
79 motion_x & 15, motion_y & 15, 128 - s->no_rounding);
83 dxy = ((motion_x >> 3) & 1) | ((motion_y >> 2) & 2);
85 s->hdsp.put_no_rnd_pixels_tab[0][dxy](dest_y, ptr, linesize, 16);
87 s->hdsp.put_pixels_tab[0][dxy](dest_y, ptr, linesize, 16);
91 if (CONFIG_GRAY && s->flags & CODEC_FLAG_GRAY)
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 <<= (3 - s->sprite_warping_accuracy);
99 motion_y <<= (3 - s->sprite_warping_accuracy);
100 src_x = av_clip(src_x, -8, s->width >> 1);
101 if (src_x == s->width >> 1)
103 src_y = av_clip(src_y, -8, s->height >> 1);
104 if (src_y == s->height >> 1)
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->edge_emu_buffer, ptr,
112 uvlinesize, uvlinesize,
115 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
116 ptr = s->edge_emu_buffer;
119 s->dsp.gmc1(dest_cb, ptr, uvlinesize, 8,
120 motion_x & 15, motion_y & 15, 128 - s->no_rounding);
122 ptr = ref_picture[2] + offset;
124 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr,
125 uvlinesize, uvlinesize,
128 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
129 ptr = s->edge_emu_buffer;
131 s->dsp.gmc1(dest_cr, ptr, uvlinesize, 8,
132 motion_x & 15, motion_y & 15, 128 - s->no_rounding);
135 static void gmc_motion(MpegEncContext *s,
136 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
137 uint8_t **ref_picture)
140 int linesize, uvlinesize;
141 const int a = s->sprite_warping_accuracy;
144 linesize = s->linesize;
145 uvlinesize = s->uvlinesize;
147 ptr = ref_picture[0];
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;
154 s->dsp.gmc(dest_y, ptr, linesize, 16,
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->dsp.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);
168 if (CONFIG_GRAY && s->flags & CODEC_FLAG_GRAY)
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;
176 ptr = ref_picture[1];
177 s->dsp.gmc(dest_cb, ptr, uvlinesize, 8,
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, s->v_edge_pos >> 1);
184 ptr = ref_picture[2];
185 s->dsp.gmc(dest_cr, ptr, uvlinesize, 8,
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, s->v_edge_pos >> 1);
193 static 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)
202 src_x += motion_x >> 1;
203 src_y += motion_y >> 1;
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)
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;
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->edge_emu_buffer, src,
217 s->linesize, s->linesize,
220 s->h_edge_pos, s->v_edge_pos);
221 src = s->edge_emu_buffer;
224 pix_op[dxy](dest, src, s->linesize, 8);
228 static av_always_inline
229 void mpeg_motion_internal(MpegEncContext *s,
236 uint8_t **ref_picture,
237 op_pixels_func (*pix_op)[4],
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;
250 if (s->quarter_sample) {
256 v_edge_pos = s->v_edge_pos >> field_based;
257 linesize = s->current_picture.f->linesize[0] << field_based;
258 uvlinesize = s->current_picture.f->linesize[1] << field_based;
260 dxy = ((motion_y & 1) << 1) | (motion_x & 1);
261 src_x = s->mb_x * 16 + (motion_x >> 1);
262 src_y = (mb_y << (4 - field_based)) + (motion_y >> 1);
264 if (!is_mpeg12 && s->out_format == FMT_H263) {
265 if ((s->workaround_bugs & FF_BUG_HPEL_CHROMA) && field_based) {
266 mx = (motion_x >> 1) | (motion_x & 1);
268 uvdxy = ((my & 1) << 1) | (mx & 1);
269 uvsrc_x = s->mb_x * 8 + (mx >> 1);
270 uvsrc_y = (mb_y << (3 - field_based)) + (my >> 1);
272 uvdxy = dxy | (motion_y & 2) | ((motion_x & 2) >> 1);
273 uvsrc_x = src_x >> 1;
274 uvsrc_y = src_y >> 1;
276 // Even chroma mv's are full pel in H261
277 } else if (!is_mpeg12 && s->out_format == FMT_H261) {
281 uvsrc_x = s->mb_x * 8 + mx;
282 uvsrc_y = mb_y * 8 + my;
284 if (s->chroma_y_shift) {
287 uvdxy = ((my & 1) << 1) | (mx & 1);
288 uvsrc_x = s->mb_x * 8 + (mx >> 1);
289 uvsrc_y = (mb_y << (3 - field_based)) + (my >> 1);
291 if (s->chroma_x_shift) {
294 uvdxy = ((motion_y & 1) << 1) | (mx & 1);
295 uvsrc_x = s->mb_x * 8 + (mx >> 1);
306 ptr_y = ref_picture[0] + src_y * linesize + src_x;
307 ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x;
308 ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x;
310 if ((unsigned)src_x > FFMAX(s->h_edge_pos - (motion_x & 1) - 16, 0) ||
311 (unsigned)src_y > FFMAX( v_edge_pos - (motion_y & 1) - h , 0)) {
313 s->codec_id == AV_CODEC_ID_MPEG2VIDEO ||
314 s->codec_id == AV_CODEC_ID_MPEG1VIDEO) {
315 av_log(s->avctx, AV_LOG_DEBUG,
316 "MPEG motion vector out of boundary (%d %d)\n", src_x,
320 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr_y,
321 s->linesize, s->linesize,
322 17, 17 + field_based,
323 src_x, src_y << field_based,
324 s->h_edge_pos, s->v_edge_pos);
325 ptr_y = s->edge_emu_buffer;
326 if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
327 uint8_t *ubuf = s->edge_emu_buffer + 18 * s->linesize;
328 uint8_t *vbuf = ubuf + 9 * s->uvlinesize;
329 s->vdsp.emulated_edge_mc(ubuf, ptr_cb,
330 s->uvlinesize, s->uvlinesize,
332 uvsrc_x, uvsrc_y << field_based,
333 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
334 s->vdsp.emulated_edge_mc(vbuf, ptr_cr,
335 s->uvlinesize, s->uvlinesize,
337 uvsrc_x, uvsrc_y << field_based,
338 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
344 /* FIXME use this for field pix too instead of the obnoxious hack which
345 * changes picture.data */
347 dest_y += s->linesize;
348 dest_cb += s->uvlinesize;
349 dest_cr += s->uvlinesize;
353 ptr_y += s->linesize;
354 ptr_cb += s->uvlinesize;
355 ptr_cr += s->uvlinesize;
358 pix_op[0][dxy](dest_y, ptr_y, linesize, h);
360 if (!CONFIG_GRAY || !(s->flags & 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);
366 if (!is_mpeg12 && (CONFIG_H261_ENCODER || CONFIG_H261_DECODER) &&
367 s->out_format == FMT_H261) {
368 ff_h261_loop_filter(s);
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)
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);
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);
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)
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);
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);
409 // FIXME: SIMDify, avg variant, 16x16 version
410 static inline void put_obmc(uint8_t *dst, uint8_t *src[5], int stride)
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);
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);
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);
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);
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);
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);
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);
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 */)
473 av_assert2(s->quarter_sample == 0);
475 for (i = 0; i < 5; i++) {
476 if (i && mv[i][0] == mv[MID][0] && mv[i][1] == mv[MID][1]) {
479 ptr[i] = s->obmc_scratchpad + 8 * (i & 1) +
480 s->linesize * 8 * (i >> 1);
481 hpel_motion(s, ptr[i], src, src_x, src_y, pix_op,
486 put_obmc(dest, ptr, s->linesize);
489 static inline void qpel_motion(MpegEncContext *s,
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)
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;
503 dxy = ((motion_y & 3) << 2) | (motion_x & 3);
505 src_x = s->mb_x * 16 + (motion_x >> 2);
506 src_y = s->mb_y * (16 >> field_based) + (motion_y >> 2);
508 v_edge_pos = s->v_edge_pos >> field_based;
509 linesize = s->linesize << field_based;
510 uvlinesize = s->uvlinesize << field_based;
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);
526 mx = (mx >> 1) | (mx & 1);
527 my = (my >> 1) | (my & 1);
529 uvdxy = (mx & 1) | ((my & 1) << 1);
533 uvsrc_x = s->mb_x * 8 + mx;
534 uvsrc_y = s->mb_y * (8 >> field_based) + my;
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;
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->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->edge_emu_buffer;
548 if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
549 uint8_t *ubuf = s->edge_emu_buffer + 18 * s->linesize;
550 uint8_t *vbuf = ubuf + 9 * s->uvlinesize;
551 s->vdsp.emulated_edge_mc(ubuf, ptr_cb,
552 s->uvlinesize, s->uvlinesize,
554 uvsrc_x, uvsrc_y << field_based,
555 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
556 s->vdsp.emulated_edge_mc(vbuf, ptr_cr,
557 s->uvlinesize, s->uvlinesize,
559 uvsrc_x, uvsrc_y << field_based,
560 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
567 qpix_op[0][dxy](dest_y, ptr_y, linesize);
570 dest_y += s->linesize;
571 dest_cb += s->uvlinesize;
572 dest_cr += s->uvlinesize;
576 ptr_y += s->linesize;
577 ptr_cb += s->uvlinesize;
578 ptr_cr += s->uvlinesize;
580 // damn interlaced mode
581 // FIXME boundary mirroring is not exactly correct here
582 qpix_op[1][dxy](dest_y, ptr_y, linesize);
583 qpix_op[1][dxy](dest_y + 8, ptr_y + 8, linesize);
585 if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
586 pix_op[1][uvdxy](dest_cr, ptr_cr, uvlinesize, h >> 1);
587 pix_op[1][uvdxy](dest_cb, ptr_cb, uvlinesize, h >> 1);
592 * h263 chroma 4mv motion compensation.
594 static void chroma_4mv_motion(MpegEncContext *s,
595 uint8_t *dest_cb, uint8_t *dest_cr,
596 uint8_t **ref_picture,
597 op_pixels_func *pix_op,
601 int src_x, src_y, dxy, emu = 0;
604 /* In case of 8X8, we construct a single chroma motion vector
605 * with a special rounding */
606 mx = ff_h263_round_chroma(mx);
607 my = ff_h263_round_chroma(my);
609 dxy = ((my & 1) << 1) | (mx & 1);
613 src_x = s->mb_x * 8 + mx;
614 src_y = s->mb_y * 8 + my;
615 src_x = av_clip(src_x, -8, (s->width >> 1));
616 if (src_x == (s->width >> 1))
618 src_y = av_clip(src_y, -8, (s->height >> 1));
619 if (src_y == (s->height >> 1))
622 offset = src_y * s->uvlinesize + src_x;
623 ptr = ref_picture[1] + offset;
624 if ((unsigned)src_x > FFMAX((s->h_edge_pos >> 1) - (dxy & 1) - 8, 0) ||
625 (unsigned)src_y > FFMAX((s->v_edge_pos >> 1) - (dxy >> 1) - 8, 0)) {
626 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr,
627 s->uvlinesize, s->uvlinesize,
629 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
630 ptr = s->edge_emu_buffer;
633 pix_op[dxy](dest_cb, ptr, s->uvlinesize, 8);
635 ptr = ref_picture[2] + offset;
637 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr,
638 s->uvlinesize, s->uvlinesize,
640 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
641 ptr = s->edge_emu_buffer;
643 pix_op[dxy](dest_cr, ptr, s->uvlinesize, 8);
646 static inline void prefetch_motion(MpegEncContext *s, uint8_t **pix, int dir)
648 /* fetch pixels for estimated mv 4 macroblocks ahead
649 * optimized for 64byte cache lines */
650 const int shift = s->quarter_sample ? 2 : 1;
651 const int mx = (s->mv[dir][0][0] >> shift) + 16 * s->mb_x + 8;
652 const int my = (s->mv[dir][0][1] >> shift) + 16 * s->mb_y;
653 int off = mx + (my + (s->mb_x & 3) * 4) * s->linesize + 64;
655 s->vdsp.prefetch(pix[0] + off, s->linesize, 4);
656 off = (mx >> 1) + ((my >> 1) + (s->mb_x & 7)) * s->uvlinesize + 64;
657 s->vdsp.prefetch(pix[1] + off, pix[2] - pix[1], 2);
660 static inline void apply_obmc(MpegEncContext *s,
664 uint8_t **ref_picture,
665 op_pixels_func (*pix_op)[4])
667 LOCAL_ALIGNED_8(int16_t, mv_cache, [4], [4][2]);
668 Picture *cur_frame = &s->current_picture;
671 const int xy = mb_x + mb_y * s->mb_stride;
672 const int mot_stride = s->b8_stride;
673 const int mot_xy = mb_x * 2 + mb_y * 2 * mot_stride;
676 av_assert2(!s->mb_skipped);
678 AV_COPY32(mv_cache[1][1], cur_frame->motion_val[0][mot_xy]);
679 AV_COPY32(mv_cache[1][2], cur_frame->motion_val[0][mot_xy + 1]);
681 AV_COPY32(mv_cache[2][1],
682 cur_frame->motion_val[0][mot_xy + mot_stride]);
683 AV_COPY32(mv_cache[2][2],
684 cur_frame->motion_val[0][mot_xy + mot_stride + 1]);
686 AV_COPY32(mv_cache[3][1],
687 cur_frame->motion_val[0][mot_xy + mot_stride]);
688 AV_COPY32(mv_cache[3][2],
689 cur_frame->motion_val[0][mot_xy + mot_stride + 1]);
691 if (mb_y == 0 || IS_INTRA(cur_frame->mb_type[xy - s->mb_stride])) {
692 AV_COPY32(mv_cache[0][1], mv_cache[1][1]);
693 AV_COPY32(mv_cache[0][2], mv_cache[1][2]);
695 AV_COPY32(mv_cache[0][1],
696 cur_frame->motion_val[0][mot_xy - mot_stride]);
697 AV_COPY32(mv_cache[0][2],
698 cur_frame->motion_val[0][mot_xy - mot_stride + 1]);
701 if (mb_x == 0 || IS_INTRA(cur_frame->mb_type[xy - 1])) {
702 AV_COPY32(mv_cache[1][0], mv_cache[1][1]);
703 AV_COPY32(mv_cache[2][0], mv_cache[2][1]);
705 AV_COPY32(mv_cache[1][0], cur_frame->motion_val[0][mot_xy - 1]);
706 AV_COPY32(mv_cache[2][0],
707 cur_frame->motion_val[0][mot_xy - 1 + mot_stride]);
710 if (mb_x + 1 >= s->mb_width || IS_INTRA(cur_frame->mb_type[xy + 1])) {
711 AV_COPY32(mv_cache[1][3], mv_cache[1][2]);
712 AV_COPY32(mv_cache[2][3], mv_cache[2][2]);
714 AV_COPY32(mv_cache[1][3], cur_frame->motion_val[0][mot_xy + 2]);
715 AV_COPY32(mv_cache[2][3],
716 cur_frame->motion_val[0][mot_xy + 2 + mot_stride]);
721 for (i = 0; i < 4; i++) {
722 const int x = (i & 1) + 1;
723 const int y = (i >> 1) + 1;
725 { mv_cache[y][x][0], mv_cache[y][x][1] },
726 { mv_cache[y - 1][x][0], mv_cache[y - 1][x][1] },
727 { mv_cache[y][x - 1][0], mv_cache[y][x - 1][1] },
728 { mv_cache[y][x + 1][0], mv_cache[y][x + 1][1] },
729 { mv_cache[y + 1][x][0], mv_cache[y + 1][x][1] }
732 obmc_motion(s, dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize,
734 mb_x * 16 + (i & 1) * 8, mb_y * 16 + (i >> 1) * 8,
741 if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY))
742 chroma_4mv_motion(s, dest_cb, dest_cr,
743 ref_picture, pix_op[1],
747 static inline void apply_8x8(MpegEncContext *s,
752 uint8_t **ref_picture,
753 qpel_mc_func (*qpix_op)[16],
754 op_pixels_func (*pix_op)[4])
756 int dxy, mx, my, src_x, src_y;
764 if (s->quarter_sample) {
765 for (i = 0; i < 4; i++) {
766 int motion_x = s->mv[dir][i][0];
767 int motion_y = s->mv[dir][i][1];
769 dxy = ((motion_y & 3) << 2) | (motion_x & 3);
770 src_x = mb_x * 16 + (motion_x >> 2) + (i & 1) * 8;
771 src_y = mb_y * 16 + (motion_y >> 2) + (i >> 1) * 8;
773 /* WARNING: do no forget half pels */
774 src_x = av_clip(src_x, -16, s->width);
775 if (src_x == s->width)
777 src_y = av_clip(src_y, -16, s->height);
778 if (src_y == s->height)
781 ptr = ref_picture[0] + (src_y * s->linesize) + (src_x);
782 if ((unsigned)src_x > FFMAX(s->h_edge_pos - (motion_x & 3) - 8, 0) ||
783 (unsigned)src_y > FFMAX(s->v_edge_pos - (motion_y & 3) - 8, 0)) {
784 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr,
785 s->linesize, s->linesize,
790 ptr = s->edge_emu_buffer;
792 dest = dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize;
793 qpix_op[1][dxy](dest, ptr, s->linesize);
795 mx += s->mv[dir][i][0] / 2;
796 my += s->mv[dir][i][1] / 2;
799 for (i = 0; i < 4; i++) {
801 dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize,
803 mb_x * 16 + (i & 1) * 8,
804 mb_y * 16 + (i >> 1) * 8,
809 mx += s->mv[dir][i][0];
810 my += s->mv[dir][i][1];
814 if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY))
815 chroma_4mv_motion(s, dest_cb, dest_cr,
816 ref_picture, pix_op[1], mx, my);
820 * motion compensation of a single macroblock
822 * @param dest_y luma destination pointer
823 * @param dest_cb chroma cb/u destination pointer
824 * @param dest_cr chroma cr/v destination pointer
825 * @param dir direction (0->forward, 1->backward)
826 * @param ref_picture array[3] of pointers to the 3 planes of the reference picture
827 * @param pix_op halfpel motion compensation function (average or put normally)
828 * @param qpix_op qpel motion compensation function (average or put normally)
829 * the motion vectors are taken from s->mv and the MV type from s->mv_type
831 static av_always_inline void MPV_motion_internal(MpegEncContext *s,
836 uint8_t **ref_picture,
837 op_pixels_func (*pix_op)[4],
838 qpel_mc_func (*qpix_op)[16],
844 prefetch_motion(s, ref_picture, dir);
846 if (!is_mpeg12 && s->obmc && s->pict_type != AV_PICTURE_TYPE_B) {
847 apply_obmc(s, dest_y, dest_cb, dest_cr, ref_picture, pix_op);
851 switch (s->mv_type) {
854 if (s->real_sprite_warping_points == 1) {
855 gmc1_motion(s, dest_y, dest_cb, dest_cr,
858 gmc_motion(s, dest_y, dest_cb, dest_cr,
861 } else if (!is_mpeg12 && s->quarter_sample) {
862 qpel_motion(s, dest_y, dest_cb, dest_cr,
864 ref_picture, pix_op, qpix_op,
865 s->mv[dir][0][0], s->mv[dir][0][1], 16);
866 } else if (!is_mpeg12 && (CONFIG_WMV2_DECODER || CONFIG_WMV2_ENCODER) &&
867 s->mspel && s->codec_id == AV_CODEC_ID_WMV2) {
868 ff_mspel_motion(s, dest_y, dest_cb, dest_cr,
870 s->mv[dir][0][0], s->mv[dir][0][1], 16);
872 mpeg_motion(s, dest_y, dest_cb, dest_cr, 0,
874 s->mv[dir][0][0], s->mv[dir][0][1], 16, mb_y);
879 apply_8x8(s, dest_y, dest_cb, dest_cr,
880 dir, ref_picture, qpix_op, pix_op);
883 if (s->picture_structure == PICT_FRAME) {
884 if (!is_mpeg12 && s->quarter_sample) {
885 for (i = 0; i < 2; i++)
886 qpel_motion(s, dest_y, dest_cb, dest_cr,
887 1, i, s->field_select[dir][i],
888 ref_picture, pix_op, qpix_op,
889 s->mv[dir][i][0], s->mv[dir][i][1], 8);
892 mpeg_motion_field(s, dest_y, dest_cb, dest_cr,
893 0, s->field_select[dir][0],
895 s->mv[dir][0][0], s->mv[dir][0][1], 8, mb_y);
897 mpeg_motion_field(s, dest_y, dest_cb, dest_cr,
898 1, s->field_select[dir][1],
900 s->mv[dir][1][0], s->mv[dir][1][1], 8, mb_y);
903 if ( s->picture_structure != s->field_select[dir][0] + 1 && s->pict_type != AV_PICTURE_TYPE_B && !s->first_field
904 || !ref_picture[0]) {
905 ref_picture = s->current_picture_ptr->f->data;
908 mpeg_motion(s, dest_y, dest_cb, dest_cr,
909 s->field_select[dir][0],
911 s->mv[dir][0][0], s->mv[dir][0][1], 16, mb_y >> 1);
915 for (i = 0; i < 2; i++) {
916 uint8_t **ref2picture;
918 if ((s->picture_structure == s->field_select[dir][i] + 1
919 || s->pict_type == AV_PICTURE_TYPE_B || s->first_field) && ref_picture[0]) {
920 ref2picture = ref_picture;
922 ref2picture = s->current_picture_ptr->f->data;
925 mpeg_motion(s, dest_y, dest_cb, dest_cr,
926 s->field_select[dir][i],
928 s->mv[dir][i][0], s->mv[dir][i][1] + 16 * i,
931 dest_y += 16 * s->linesize;
932 dest_cb += (16 >> s->chroma_y_shift) * s->uvlinesize;
933 dest_cr += (16 >> s->chroma_y_shift) * s->uvlinesize;
937 if (s->picture_structure == PICT_FRAME) {
938 for (i = 0; i < 2; i++) {
940 for (j = 0; j < 2; j++)
941 mpeg_motion_field(s, dest_y, dest_cb, dest_cr,
942 j, j ^ i, ref_picture, pix_op,
943 s->mv[dir][2 * i + j][0],
944 s->mv[dir][2 * i + j][1], 8, mb_y);
945 pix_op = s->hdsp.avg_pixels_tab;
948 if (!ref_picture[0]) {
949 ref_picture = s->current_picture_ptr->f->data;
951 for (i = 0; i < 2; i++) {
952 mpeg_motion(s, dest_y, dest_cb, dest_cr,
953 s->picture_structure != i + 1,
955 s->mv[dir][2 * i][0], s->mv[dir][2 * i][1],
958 // after put we make avg of the same block
959 pix_op = s->hdsp.avg_pixels_tab;
961 /* opposite parity is always in the same frame if this is
963 if (!s->first_field) {
964 ref_picture = s->current_picture_ptr->f->data;
969 default: av_assert2(0);
973 void ff_MPV_motion(MpegEncContext *s,
974 uint8_t *dest_y, uint8_t *dest_cb,
975 uint8_t *dest_cr, int dir,
976 uint8_t **ref_picture,
977 op_pixels_func (*pix_op)[4],
978 qpel_mc_func (*qpix_op)[16])
981 if (s->out_format == FMT_MPEG1)
982 MPV_motion_internal(s, dest_y, dest_cb, dest_cr, dir,
983 ref_picture, pix_op, qpix_op, 1);
986 MPV_motion_internal(s, dest_y, dest_cb, dest_cr, dir,
987 ref_picture, pix_op, qpix_op, 0);