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 Libav.
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.
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.
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
26 #include "libavutil/internal.h"
29 #include "mpegutils.h"
30 #include "mpegvideo.h"
36 static void gmc1_motion(MpegEncContext *s,
37 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
38 uint8_t **ref_picture)
41 int src_x, src_y, motion_x, motion_y;
42 ptrdiff_t offset, linesize, uvlinesize;
45 motion_x = s->sprite_offset[0][0];
46 motion_y = s->sprite_offset[0][1];
47 src_x = s->mb_x * 16 + (motion_x >> (s->sprite_warping_accuracy + 1));
48 src_y = s->mb_y * 16 + (motion_y >> (s->sprite_warping_accuracy + 1));
49 motion_x <<= (3 - s->sprite_warping_accuracy);
50 motion_y <<= (3 - s->sprite_warping_accuracy);
51 src_x = av_clip(src_x, -16, s->width);
52 if (src_x == s->width)
54 src_y = av_clip(src_y, -16, s->height);
55 if (src_y == s->height)
58 linesize = s->linesize;
59 uvlinesize = s->uvlinesize;
61 ptr = ref_picture[0] + src_y * linesize + src_x;
63 if ((unsigned)src_x >= FFMAX(s->h_edge_pos - 17, 0) ||
64 (unsigned)src_y >= FFMAX(s->v_edge_pos - 17, 0)) {
65 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr,
69 s->h_edge_pos, s->v_edge_pos);
70 ptr = s->edge_emu_buffer;
73 if ((motion_x | motion_y) & 7) {
74 s->mdsp.gmc1(dest_y, ptr, linesize, 16,
75 motion_x & 15, motion_y & 15, 128 - s->no_rounding);
76 s->mdsp.gmc1(dest_y + 8, ptr + 8, linesize, 16,
77 motion_x & 15, motion_y & 15, 128 - s->no_rounding);
81 dxy = ((motion_x >> 3) & 1) | ((motion_y >> 2) & 2);
83 s->hdsp.put_no_rnd_pixels_tab[0][dxy](dest_y, ptr, linesize, 16);
85 s->hdsp.put_pixels_tab[0][dxy](dest_y, ptr, linesize, 16);
89 if (CONFIG_GRAY && s->flags & CODEC_FLAG_GRAY)
92 motion_x = s->sprite_offset[1][0];
93 motion_y = s->sprite_offset[1][1];
94 src_x = s->mb_x * 8 + (motion_x >> (s->sprite_warping_accuracy + 1));
95 src_y = s->mb_y * 8 + (motion_y >> (s->sprite_warping_accuracy + 1));
96 motion_x <<= (3 - s->sprite_warping_accuracy);
97 motion_y <<= (3 - s->sprite_warping_accuracy);
98 src_x = av_clip(src_x, -8, s->width >> 1);
99 if (src_x == s->width >> 1)
101 src_y = av_clip(src_y, -8, s->height >> 1);
102 if (src_y == s->height >> 1)
105 offset = (src_y * uvlinesize) + src_x;
106 ptr = ref_picture[1] + offset;
107 if ((unsigned)src_x >= FFMAX((s->h_edge_pos >> 1) - 9, 0) ||
108 (unsigned)src_y >= FFMAX((s->v_edge_pos >> 1) - 9, 0)) {
109 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr,
110 uvlinesize, uvlinesize,
113 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
114 ptr = s->edge_emu_buffer;
117 s->mdsp.gmc1(dest_cb, ptr, uvlinesize, 8,
118 motion_x & 15, motion_y & 15, 128 - s->no_rounding);
120 ptr = ref_picture[2] + offset;
122 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr,
123 uvlinesize, uvlinesize,
126 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
127 ptr = s->edge_emu_buffer;
129 s->mdsp.gmc1(dest_cr, ptr, uvlinesize, 8,
130 motion_x & 15, motion_y & 15, 128 - s->no_rounding);
133 static void gmc_motion(MpegEncContext *s,
134 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
135 uint8_t **ref_picture)
138 int linesize, uvlinesize;
139 const int a = s->sprite_warping_accuracy;
142 linesize = s->linesize;
143 uvlinesize = s->uvlinesize;
145 ptr = ref_picture[0];
147 ox = s->sprite_offset[0][0] + s->sprite_delta[0][0] * s->mb_x * 16 +
148 s->sprite_delta[0][1] * s->mb_y * 16;
149 oy = s->sprite_offset[0][1] + s->sprite_delta[1][0] * s->mb_x * 16 +
150 s->sprite_delta[1][1] * s->mb_y * 16;
152 s->mdsp.gmc(dest_y, ptr, linesize, 16,
154 s->sprite_delta[0][0], s->sprite_delta[0][1],
155 s->sprite_delta[1][0], s->sprite_delta[1][1],
156 a + 1, (1 << (2 * a + 1)) - s->no_rounding,
157 s->h_edge_pos, s->v_edge_pos);
158 s->mdsp.gmc(dest_y + 8, ptr, linesize, 16,
159 ox + s->sprite_delta[0][0] * 8,
160 oy + s->sprite_delta[1][0] * 8,
161 s->sprite_delta[0][0], s->sprite_delta[0][1],
162 s->sprite_delta[1][0], s->sprite_delta[1][1],
163 a + 1, (1 << (2 * a + 1)) - s->no_rounding,
164 s->h_edge_pos, s->v_edge_pos);
166 if (CONFIG_GRAY && s->flags & CODEC_FLAG_GRAY)
169 ox = s->sprite_offset[1][0] + s->sprite_delta[0][0] * s->mb_x * 8 +
170 s->sprite_delta[0][1] * s->mb_y * 8;
171 oy = s->sprite_offset[1][1] + s->sprite_delta[1][0] * s->mb_x * 8 +
172 s->sprite_delta[1][1] * s->mb_y * 8;
174 ptr = ref_picture[1];
175 s->mdsp.gmc(dest_cb, ptr, uvlinesize, 8,
177 s->sprite_delta[0][0], s->sprite_delta[0][1],
178 s->sprite_delta[1][0], s->sprite_delta[1][1],
179 a + 1, (1 << (2 * a + 1)) - s->no_rounding,
180 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
182 ptr = ref_picture[2];
183 s->mdsp.gmc(dest_cr, ptr, uvlinesize, 8,
185 s->sprite_delta[0][0], s->sprite_delta[0][1],
186 s->sprite_delta[1][0], s->sprite_delta[1][1],
187 a + 1, (1 << (2 * a + 1)) - s->no_rounding,
188 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
191 static inline int hpel_motion(MpegEncContext *s,
192 uint8_t *dest, uint8_t *src,
193 int src_x, int src_y,
194 op_pixels_func *pix_op,
195 int motion_x, int motion_y)
200 src_x += motion_x >> 1;
201 src_y += motion_y >> 1;
203 /* WARNING: do no forget half pels */
204 src_x = av_clip(src_x, -16, s->width); // FIXME unneeded for emu?
205 if (src_x != s->width)
207 src_y = av_clip(src_y, -16, s->height);
208 if (src_y != s->height)
209 dxy |= (motion_y & 1) << 1;
210 src += src_y * s->linesize + src_x;
212 if (s->unrestricted_mv) {
213 if ((unsigned)src_x > FFMAX(s->h_edge_pos - (motion_x & 1) - 8, 0) ||
214 (unsigned)src_y > FFMAX(s->v_edge_pos - (motion_y & 1) - 8, 0)) {
215 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, src,
216 s->linesize, s->linesize,
218 src_x, src_y, s->h_edge_pos,
220 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 *uvbuf = s->edge_emu_buffer + 18 * s->linesize;
328 s->vdsp.emulated_edge_mc(uvbuf, ptr_cb,
329 s->uvlinesize, s->uvlinesize,
331 uvsrc_x, uvsrc_y << field_based,
332 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
333 s->vdsp.emulated_edge_mc(uvbuf + 16, ptr_cr,
334 s->uvlinesize, s->uvlinesize,
336 uvsrc_x, uvsrc_y << field_based,
337 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
343 /* FIXME use this for field pix too instead of the obnoxious hack which
344 * changes picture.data */
346 dest_y += s->linesize;
347 dest_cb += s->uvlinesize;
348 dest_cr += s->uvlinesize;
352 ptr_y += s->linesize;
353 ptr_cb += s->uvlinesize;
354 ptr_cr += s->uvlinesize;
357 pix_op[0][dxy](dest_y, ptr_y, linesize, h);
359 if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
360 pix_op[s->chroma_x_shift][uvdxy]
361 (dest_cb, ptr_cb, uvlinesize, h >> s->chroma_y_shift);
362 pix_op[s->chroma_x_shift][uvdxy]
363 (dest_cr, ptr_cr, uvlinesize, h >> s->chroma_y_shift);
365 if (!is_mpeg12 && (CONFIG_H261_ENCODER || CONFIG_H261_DECODER) &&
366 s->out_format == FMT_H261) {
367 ff_h261_loop_filter(s);
370 /* apply one mpeg motion vector to the three components */
371 static void mpeg_motion(MpegEncContext *s,
372 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
373 int field_select, uint8_t **ref_picture,
374 op_pixels_func (*pix_op)[4],
375 int motion_x, int motion_y, int h, int mb_y)
378 if (s->out_format == FMT_MPEG1)
379 mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, 0, 0,
380 field_select, ref_picture, pix_op,
381 motion_x, motion_y, h, 1, mb_y);
384 mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, 0, 0,
385 field_select, ref_picture, pix_op,
386 motion_x, motion_y, h, 0, mb_y);
389 static void mpeg_motion_field(MpegEncContext *s, uint8_t *dest_y,
390 uint8_t *dest_cb, uint8_t *dest_cr,
391 int bottom_field, int field_select,
392 uint8_t **ref_picture,
393 op_pixels_func (*pix_op)[4],
394 int motion_x, int motion_y, int h, int mb_y)
397 if(s->out_format == FMT_MPEG1)
398 mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, 1,
399 bottom_field, field_select, ref_picture, pix_op,
400 motion_x, motion_y, h, 1, mb_y);
403 mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, 1,
404 bottom_field, field_select, ref_picture, pix_op,
405 motion_x, motion_y, h, 0, mb_y);
408 // FIXME: SIMDify, avg variant, 16x16 version
409 static inline void put_obmc(uint8_t *dst, uint8_t *src[5], int stride)
412 uint8_t *const top = src[1];
413 uint8_t *const left = src[2];
414 uint8_t *const mid = src[0];
415 uint8_t *const right = src[3];
416 uint8_t *const bottom = src[4];
417 #define OBMC_FILTER(x, t, l, m, r, b)\
418 dst[x]= (t*top[x] + l*left[x] + m*mid[x] + r*right[x] + b*bottom[x] + 4)>>3
419 #define OBMC_FILTER4(x, t, l, m, r, b)\
420 OBMC_FILTER(x , t, l, m, r, b);\
421 OBMC_FILTER(x+1 , t, l, m, r, b);\
422 OBMC_FILTER(x +stride, t, l, m, r, b);\
423 OBMC_FILTER(x+1+stride, t, l, m, r, b);
426 OBMC_FILTER (x , 2, 2, 4, 0, 0);
427 OBMC_FILTER (x + 1, 2, 1, 5, 0, 0);
428 OBMC_FILTER4(x + 2, 2, 1, 5, 0, 0);
429 OBMC_FILTER4(x + 4, 2, 0, 5, 1, 0);
430 OBMC_FILTER (x + 6, 2, 0, 5, 1, 0);
431 OBMC_FILTER (x + 7, 2, 0, 4, 2, 0);
433 OBMC_FILTER (x , 1, 2, 5, 0, 0);
434 OBMC_FILTER (x + 1, 1, 2, 5, 0, 0);
435 OBMC_FILTER (x + 6, 1, 0, 5, 2, 0);
436 OBMC_FILTER (x + 7, 1, 0, 5, 2, 0);
438 OBMC_FILTER4(x , 1, 2, 5, 0, 0);
439 OBMC_FILTER4(x + 2, 1, 1, 6, 0, 0);
440 OBMC_FILTER4(x + 4, 1, 0, 6, 1, 0);
441 OBMC_FILTER4(x + 6, 1, 0, 5, 2, 0);
443 OBMC_FILTER4(x , 0, 2, 5, 0, 1);
444 OBMC_FILTER4(x + 2, 0, 1, 6, 0, 1);
445 OBMC_FILTER4(x + 4, 0, 0, 6, 1, 1);
446 OBMC_FILTER4(x + 6, 0, 0, 5, 2, 1);
448 OBMC_FILTER (x , 0, 2, 5, 0, 1);
449 OBMC_FILTER (x + 1, 0, 2, 5, 0, 1);
450 OBMC_FILTER4(x + 2, 0, 1, 5, 0, 2);
451 OBMC_FILTER4(x + 4, 0, 0, 5, 1, 2);
452 OBMC_FILTER (x + 6, 0, 0, 5, 2, 1);
453 OBMC_FILTER (x + 7, 0, 0, 5, 2, 1);
455 OBMC_FILTER (x , 0, 2, 4, 0, 2);
456 OBMC_FILTER (x + 1, 0, 1, 5, 0, 2);
457 OBMC_FILTER (x + 6, 0, 0, 5, 1, 2);
458 OBMC_FILTER (x + 7, 0, 0, 4, 2, 2);
461 /* obmc for 1 8x8 luma block */
462 static inline void obmc_motion(MpegEncContext *s,
463 uint8_t *dest, uint8_t *src,
464 int src_x, int src_y,
465 op_pixels_func *pix_op,
466 int16_t mv[5][2] /* mid top left right bottom */)
472 assert(s->quarter_sample == 0);
474 for (i = 0; i < 5; i++) {
475 if (i && mv[i][0] == mv[MID][0] && mv[i][1] == mv[MID][1]) {
478 ptr[i] = s->obmc_scratchpad + 8 * (i & 1) +
479 s->linesize * 8 * (i >> 1);
480 hpel_motion(s, ptr[i], src, src_x, src_y, pix_op,
485 put_obmc(dest, ptr, s->linesize);
488 static inline void qpel_motion(MpegEncContext *s,
492 int field_based, int bottom_field,
493 int field_select, uint8_t **ref_picture,
494 op_pixels_func (*pix_op)[4],
495 qpel_mc_func (*qpix_op)[16],
496 int motion_x, int motion_y, int h)
498 uint8_t *ptr_y, *ptr_cb, *ptr_cr;
499 int dxy, uvdxy, mx, my, src_x, src_y, uvsrc_x, uvsrc_y, v_edge_pos;
500 ptrdiff_t linesize, uvlinesize;
502 dxy = ((motion_y & 3) << 2) | (motion_x & 3);
504 src_x = s->mb_x * 16 + (motion_x >> 2);
505 src_y = s->mb_y * (16 >> field_based) + (motion_y >> 2);
507 v_edge_pos = s->v_edge_pos >> field_based;
508 linesize = s->linesize << field_based;
509 uvlinesize = s->uvlinesize << field_based;
514 } else if (s->workaround_bugs & FF_BUG_QPEL_CHROMA2) {
515 static const int rtab[8] = { 0, 0, 1, 1, 0, 0, 0, 1 };
516 mx = (motion_x >> 1) + rtab[motion_x & 7];
517 my = (motion_y >> 1) + rtab[motion_y & 7];
518 } else if (s->workaround_bugs & FF_BUG_QPEL_CHROMA) {
519 mx = (motion_x >> 1) | (motion_x & 1);
520 my = (motion_y >> 1) | (motion_y & 1);
525 mx = (mx >> 1) | (mx & 1);
526 my = (my >> 1) | (my & 1);
528 uvdxy = (mx & 1) | ((my & 1) << 1);
532 uvsrc_x = s->mb_x * 8 + mx;
533 uvsrc_y = s->mb_y * (8 >> field_based) + my;
535 ptr_y = ref_picture[0] + src_y * linesize + src_x;
536 ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x;
537 ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x;
539 if ((unsigned)src_x > FFMAX(s->h_edge_pos - (motion_x & 3) - 16, 0) ||
540 (unsigned)src_y > FFMAX(v_edge_pos - (motion_y & 3) - h, 0)) {
541 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr_y,
542 s->linesize, s->linesize,
543 17, 17 + field_based,
544 src_x, src_y << field_based,
545 s->h_edge_pos, s->v_edge_pos);
546 ptr_y = s->edge_emu_buffer;
547 if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
548 uint8_t *uvbuf = s->edge_emu_buffer + 18 * s->linesize;
549 s->vdsp.emulated_edge_mc(uvbuf, ptr_cb,
550 s->uvlinesize, s->uvlinesize,
552 uvsrc_x, uvsrc_y << field_based,
553 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
554 s->vdsp.emulated_edge_mc(uvbuf + 16, ptr_cr,
555 s->uvlinesize, s->uvlinesize,
557 uvsrc_x, uvsrc_y << field_based,
558 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
565 qpix_op[0][dxy](dest_y, ptr_y, linesize);
568 dest_y += s->linesize;
569 dest_cb += s->uvlinesize;
570 dest_cr += s->uvlinesize;
574 ptr_y += s->linesize;
575 ptr_cb += s->uvlinesize;
576 ptr_cr += s->uvlinesize;
578 // damn interlaced mode
579 // FIXME boundary mirroring is not exactly correct here
580 qpix_op[1][dxy](dest_y, ptr_y, linesize);
581 qpix_op[1][dxy](dest_y + 8, ptr_y + 8, linesize);
583 if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
584 pix_op[1][uvdxy](dest_cr, ptr_cr, uvlinesize, h >> 1);
585 pix_op[1][uvdxy](dest_cb, ptr_cb, uvlinesize, h >> 1);
590 * h263 chroma 4mv motion compensation.
592 static void chroma_4mv_motion(MpegEncContext *s,
593 uint8_t *dest_cb, uint8_t *dest_cr,
594 uint8_t **ref_picture,
595 op_pixels_func *pix_op,
599 int src_x, src_y, dxy, emu = 0;
602 /* In case of 8X8, we construct a single chroma motion vector
603 * with a special rounding */
604 mx = ff_h263_round_chroma(mx);
605 my = ff_h263_round_chroma(my);
607 dxy = ((my & 1) << 1) | (mx & 1);
611 src_x = s->mb_x * 8 + mx;
612 src_y = s->mb_y * 8 + my;
613 src_x = av_clip(src_x, -8, (s->width >> 1));
614 if (src_x == (s->width >> 1))
616 src_y = av_clip(src_y, -8, (s->height >> 1));
617 if (src_y == (s->height >> 1))
620 offset = src_y * s->uvlinesize + src_x;
621 ptr = ref_picture[1] + offset;
622 if ((unsigned)src_x > FFMAX((s->h_edge_pos >> 1) - (dxy & 1) - 8, 0) ||
623 (unsigned)src_y > FFMAX((s->v_edge_pos >> 1) - (dxy >> 1) - 8, 0)) {
624 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr,
625 s->uvlinesize, s->uvlinesize,
627 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
628 ptr = s->edge_emu_buffer;
631 pix_op[dxy](dest_cb, ptr, s->uvlinesize, 8);
633 ptr = ref_picture[2] + offset;
635 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr,
636 s->uvlinesize, s->uvlinesize,
638 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
639 ptr = s->edge_emu_buffer;
641 pix_op[dxy](dest_cr, ptr, s->uvlinesize, 8);
644 static inline void prefetch_motion(MpegEncContext *s, uint8_t **pix, int dir)
646 /* fetch pixels for estimated mv 4 macroblocks ahead
647 * optimized for 64byte cache lines */
648 const int shift = s->quarter_sample ? 2 : 1;
649 const int mx = (s->mv[dir][0][0] >> shift) + 16 * s->mb_x + 8;
650 const int my = (s->mv[dir][0][1] >> shift) + 16 * s->mb_y;
651 int off = mx + (my + (s->mb_x & 3) * 4) * s->linesize + 64;
653 s->vdsp.prefetch(pix[0] + off, s->linesize, 4);
654 off = (mx >> 1) + ((my >> 1) + (s->mb_x & 7)) * s->uvlinesize + 64;
655 s->vdsp.prefetch(pix[1] + off, pix[2] - pix[1], 2);
658 static inline void apply_obmc(MpegEncContext *s,
662 uint8_t **ref_picture,
663 op_pixels_func (*pix_op)[4])
665 LOCAL_ALIGNED_8(int16_t, mv_cache, [4], [4][2]);
666 Picture *cur_frame = &s->current_picture;
669 const int xy = mb_x + mb_y * s->mb_stride;
670 const int mot_stride = s->b8_stride;
671 const int mot_xy = mb_x * 2 + mb_y * 2 * mot_stride;
674 assert(!s->mb_skipped);
676 AV_COPY32(mv_cache[1][1], cur_frame->motion_val[0][mot_xy]);
677 AV_COPY32(mv_cache[1][2], cur_frame->motion_val[0][mot_xy + 1]);
679 AV_COPY32(mv_cache[2][1],
680 cur_frame->motion_val[0][mot_xy + mot_stride]);
681 AV_COPY32(mv_cache[2][2],
682 cur_frame->motion_val[0][mot_xy + mot_stride + 1]);
684 AV_COPY32(mv_cache[3][1],
685 cur_frame->motion_val[0][mot_xy + mot_stride]);
686 AV_COPY32(mv_cache[3][2],
687 cur_frame->motion_val[0][mot_xy + mot_stride + 1]);
689 if (mb_y == 0 || IS_INTRA(cur_frame->mb_type[xy - s->mb_stride])) {
690 AV_COPY32(mv_cache[0][1], mv_cache[1][1]);
691 AV_COPY32(mv_cache[0][2], mv_cache[1][2]);
693 AV_COPY32(mv_cache[0][1],
694 cur_frame->motion_val[0][mot_xy - mot_stride]);
695 AV_COPY32(mv_cache[0][2],
696 cur_frame->motion_val[0][mot_xy - mot_stride + 1]);
699 if (mb_x == 0 || IS_INTRA(cur_frame->mb_type[xy - 1])) {
700 AV_COPY32(mv_cache[1][0], mv_cache[1][1]);
701 AV_COPY32(mv_cache[2][0], mv_cache[2][1]);
703 AV_COPY32(mv_cache[1][0], cur_frame->motion_val[0][mot_xy - 1]);
704 AV_COPY32(mv_cache[2][0],
705 cur_frame->motion_val[0][mot_xy - 1 + mot_stride]);
708 if (mb_x + 1 >= s->mb_width || IS_INTRA(cur_frame->mb_type[xy + 1])) {
709 AV_COPY32(mv_cache[1][3], mv_cache[1][2]);
710 AV_COPY32(mv_cache[2][3], mv_cache[2][2]);
712 AV_COPY32(mv_cache[1][3], cur_frame->motion_val[0][mot_xy + 2]);
713 AV_COPY32(mv_cache[2][3],
714 cur_frame->motion_val[0][mot_xy + 2 + mot_stride]);
719 for (i = 0; i < 4; i++) {
720 const int x = (i & 1) + 1;
721 const int y = (i >> 1) + 1;
723 { mv_cache[y][x][0], mv_cache[y][x][1] },
724 { mv_cache[y - 1][x][0], mv_cache[y - 1][x][1] },
725 { mv_cache[y][x - 1][0], mv_cache[y][x - 1][1] },
726 { mv_cache[y][x + 1][0], mv_cache[y][x + 1][1] },
727 { mv_cache[y + 1][x][0], mv_cache[y + 1][x][1] }
730 obmc_motion(s, dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize,
732 mb_x * 16 + (i & 1) * 8, mb_y * 16 + (i >> 1) * 8,
739 if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY))
740 chroma_4mv_motion(s, dest_cb, dest_cr,
741 ref_picture, pix_op[1],
745 static inline void apply_8x8(MpegEncContext *s,
750 uint8_t **ref_picture,
751 qpel_mc_func (*qpix_op)[16],
752 op_pixels_func (*pix_op)[4])
754 int dxy, mx, my, src_x, src_y;
762 if (s->quarter_sample) {
763 for (i = 0; i < 4; i++) {
764 int motion_x = s->mv[dir][i][0];
765 int motion_y = s->mv[dir][i][1];
767 dxy = ((motion_y & 3) << 2) | (motion_x & 3);
768 src_x = mb_x * 16 + (motion_x >> 2) + (i & 1) * 8;
769 src_y = mb_y * 16 + (motion_y >> 2) + (i >> 1) * 8;
771 /* WARNING: do no forget half pels */
772 src_x = av_clip(src_x, -16, s->width);
773 if (src_x == s->width)
775 src_y = av_clip(src_y, -16, s->height);
776 if (src_y == s->height)
779 ptr = ref_picture[0] + (src_y * s->linesize) + (src_x);
780 if ((unsigned)src_x > FFMAX(s->h_edge_pos - (motion_x & 3) - 8, 0) ||
781 (unsigned)src_y > FFMAX(s->v_edge_pos - (motion_y & 3) - 8, 0)) {
782 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr,
783 s->linesize, s->linesize,
788 ptr = s->edge_emu_buffer;
790 dest = dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize;
791 qpix_op[1][dxy](dest, ptr, s->linesize);
793 mx += s->mv[dir][i][0] / 2;
794 my += s->mv[dir][i][1] / 2;
797 for (i = 0; i < 4; i++) {
799 dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize,
801 mb_x * 16 + (i & 1) * 8,
802 mb_y * 16 + (i >> 1) * 8,
807 mx += s->mv[dir][i][0];
808 my += s->mv[dir][i][1];
812 if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY))
813 chroma_4mv_motion(s, dest_cb, dest_cr,
814 ref_picture, pix_op[1], mx, my);
818 * motion compensation of a single macroblock
820 * @param dest_y luma destination pointer
821 * @param dest_cb chroma cb/u destination pointer
822 * @param dest_cr chroma cr/v destination pointer
823 * @param dir direction (0->forward, 1->backward)
824 * @param ref_picture array[3] of pointers to the 3 planes of the reference picture
825 * @param pix_op halfpel motion compensation function (average or put normally)
826 * @param qpix_op qpel motion compensation function (average or put normally)
827 * the motion vectors are taken from s->mv and the MV type from s->mv_type
829 static av_always_inline void MPV_motion_internal(MpegEncContext *s,
834 uint8_t **ref_picture,
835 op_pixels_func (*pix_op)[4],
836 qpel_mc_func (*qpix_op)[16],
842 prefetch_motion(s, ref_picture, dir);
844 if (!is_mpeg12 && s->obmc && s->pict_type != AV_PICTURE_TYPE_B) {
845 apply_obmc(s, dest_y, dest_cb, dest_cr, ref_picture, pix_op);
849 switch (s->mv_type) {
852 if (s->real_sprite_warping_points == 1) {
853 gmc1_motion(s, dest_y, dest_cb, dest_cr,
856 gmc_motion(s, dest_y, dest_cb, dest_cr,
859 } else if (!is_mpeg12 && s->quarter_sample) {
860 qpel_motion(s, dest_y, dest_cb, dest_cr,
862 ref_picture, pix_op, qpix_op,
863 s->mv[dir][0][0], s->mv[dir][0][1], 16);
864 } else if (!is_mpeg12 && (CONFIG_WMV2_DECODER || CONFIG_WMV2_ENCODER) &&
865 s->mspel && s->codec_id == AV_CODEC_ID_WMV2) {
866 ff_mspel_motion(s, dest_y, dest_cb, dest_cr,
868 s->mv[dir][0][0], s->mv[dir][0][1], 16);
870 mpeg_motion(s, dest_y, dest_cb, dest_cr, 0,
872 s->mv[dir][0][0], s->mv[dir][0][1], 16, mb_y);
877 apply_8x8(s, dest_y, dest_cb, dest_cr,
878 dir, ref_picture, qpix_op, pix_op);
881 if (s->picture_structure == PICT_FRAME) {
882 if (!is_mpeg12 && s->quarter_sample) {
883 for (i = 0; i < 2; i++)
884 qpel_motion(s, dest_y, dest_cb, dest_cr,
885 1, i, s->field_select[dir][i],
886 ref_picture, pix_op, qpix_op,
887 s->mv[dir][i][0], s->mv[dir][i][1], 8);
890 mpeg_motion_field(s, dest_y, dest_cb, dest_cr,
891 0, s->field_select[dir][0],
893 s->mv[dir][0][0], s->mv[dir][0][1], 8, mb_y);
895 mpeg_motion_field(s, dest_y, dest_cb, dest_cr,
896 1, s->field_select[dir][1],
898 s->mv[dir][1][0], s->mv[dir][1][1], 8, mb_y);
901 if (s->picture_structure != s->field_select[dir][0] + 1 &&
902 s->pict_type != AV_PICTURE_TYPE_B && !s->first_field) {
903 ref_picture = s->current_picture_ptr->f->data;
906 mpeg_motion(s, dest_y, dest_cb, dest_cr,
907 s->field_select[dir][0],
909 s->mv[dir][0][0], s->mv[dir][0][1], 16, mb_y >> 1);
913 for (i = 0; i < 2; i++) {
914 uint8_t **ref2picture;
916 if (s->picture_structure == s->field_select[dir][i] + 1
917 || s->pict_type == AV_PICTURE_TYPE_B || s->first_field) {
918 ref2picture = ref_picture;
920 ref2picture = s->current_picture_ptr->f->data;
923 mpeg_motion(s, dest_y, dest_cb, dest_cr,
924 s->field_select[dir][i],
926 s->mv[dir][i][0], s->mv[dir][i][1] + 16 * i,
929 dest_y += 16 * s->linesize;
930 dest_cb += (16 >> s->chroma_y_shift) * s->uvlinesize;
931 dest_cr += (16 >> s->chroma_y_shift) * s->uvlinesize;
935 if (s->picture_structure == PICT_FRAME) {
936 for (i = 0; i < 2; i++) {
938 for (j = 0; j < 2; j++)
939 mpeg_motion_field(s, dest_y, dest_cb, dest_cr,
940 j, j ^ i, ref_picture, pix_op,
941 s->mv[dir][2 * i + j][0],
942 s->mv[dir][2 * i + j][1], 8, mb_y);
943 pix_op = s->hdsp.avg_pixels_tab;
946 for (i = 0; i < 2; i++) {
947 mpeg_motion(s, dest_y, dest_cb, dest_cr,
948 s->picture_structure != i + 1,
950 s->mv[dir][2 * i][0], s->mv[dir][2 * i][1],
953 // after put we make avg of the same block
954 pix_op = s->hdsp.avg_pixels_tab;
956 /* opposite parity is always in the same frame if this is
958 if (!s->first_field) {
959 ref_picture = s->current_picture_ptr->f->data;
968 void ff_MPV_motion(MpegEncContext *s,
969 uint8_t *dest_y, uint8_t *dest_cb,
970 uint8_t *dest_cr, int dir,
971 uint8_t **ref_picture,
972 op_pixels_func (*pix_op)[4],
973 qpel_mc_func (*qpix_op)[16])
976 if (s->out_format == FMT_MPEG1)
977 MPV_motion_internal(s, dest_y, dest_cb, dest_cr, dir,
978 ref_picture, pix_op, qpix_op, 1);
981 MPV_motion_internal(s, dest_y, dest_cb, dest_cr, dir,
982 ref_picture, pix_op, qpix_op, 0);