mpegvideo: Refactor emulated_edge_mc calls
[libav.git] / libavcodec / mpegvideo_motion.c
1 /*
2 * Copyright (c) 2000,2001 Fabrice Bellard
3 * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
4 *
5 * 4MV & hq & B-frame encoding stuff by Michael Niedermayer <michaelni@gmx.at>
6 *
7 * This file is part of Libav.
8 *
9 * Libav is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public
11 * License as published by the Free Software Foundation; either
12 * version 2.1 of the License, or (at your option) any later version.
13 *
14 * Libav is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
18 *
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with Libav; if not, write to the Free Software
21 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
22 */
23
24 #include <string.h>
25
26 #include "libavutil/internal.h"
27 #include "avcodec.h"
28 #include "h261.h"
29 #include "mpegutils.h"
30 #include "mpegvideo.h"
31 #include "mjpegenc.h"
32 #include "msmpeg4.h"
33 #include "qpeldsp.h"
34 #include "wmv2.h"
35 #include <limits.h>
36
37 static void gmc1_motion(MpegEncContext *s,
38 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
39 uint8_t **ref_picture)
40 {
41 uint8_t *ptr;
42 int src_x, src_y, motion_x, motion_y;
43 ptrdiff_t offset, linesize, uvlinesize;
44 int emu = 0;
45
46 motion_x = s->sprite_offset[0][0];
47 motion_y = s->sprite_offset[0][1];
48 src_x = s->mb_x * 16 + (motion_x >> (s->sprite_warping_accuracy + 1));
49 src_y = s->mb_y * 16 + (motion_y >> (s->sprite_warping_accuracy + 1));
50 motion_x <<= (3 - s->sprite_warping_accuracy);
51 motion_y <<= (3 - s->sprite_warping_accuracy);
52 src_x = av_clip(src_x, -16, s->width);
53 if (src_x == s->width)
54 motion_x = 0;
55 src_y = av_clip(src_y, -16, s->height);
56 if (src_y == s->height)
57 motion_y = 0;
58
59 linesize = s->linesize;
60 uvlinesize = s->uvlinesize;
61
62 ptr = ref_picture[0] + src_y * linesize + src_x;
63
64 if ((unsigned)src_x >= FFMAX(s->h_edge_pos - 17, 0) ||
65 (unsigned)src_y >= FFMAX(s->v_edge_pos - 17, 0)) {
66 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, ptr,
67 linesize, linesize,
68 17, 17,
69 src_x, src_y,
70 s->h_edge_pos, s->v_edge_pos);
71 ptr = s->sc.edge_emu_buffer;
72 }
73
74 if ((motion_x | motion_y) & 7) {
75 s->mdsp.gmc1(dest_y, ptr, linesize, 16,
76 motion_x & 15, motion_y & 15, 128 - s->no_rounding);
77 s->mdsp.gmc1(dest_y + 8, ptr + 8, linesize, 16,
78 motion_x & 15, motion_y & 15, 128 - s->no_rounding);
79 } else {
80 int dxy;
81
82 dxy = ((motion_x >> 3) & 1) | ((motion_y >> 2) & 2);
83 if (s->no_rounding) {
84 s->hdsp.put_no_rnd_pixels_tab[0][dxy](dest_y, ptr, linesize, 16);
85 } else {
86 s->hdsp.put_pixels_tab[0][dxy](dest_y, ptr, linesize, 16);
87 }
88 }
89
90 if (CONFIG_GRAY && s->avctx->flags & AV_CODEC_FLAG_GRAY)
91 return;
92
93 motion_x = s->sprite_offset[1][0];
94 motion_y = s->sprite_offset[1][1];
95 src_x = s->mb_x * 8 + (motion_x >> (s->sprite_warping_accuracy + 1));
96 src_y = s->mb_y * 8 + (motion_y >> (s->sprite_warping_accuracy + 1));
97 motion_x <<= (3 - s->sprite_warping_accuracy);
98 motion_y <<= (3 - s->sprite_warping_accuracy);
99 src_x = av_clip(src_x, -8, s->width >> 1);
100 if (src_x == s->width >> 1)
101 motion_x = 0;
102 src_y = av_clip(src_y, -8, s->height >> 1);
103 if (src_y == s->height >> 1)
104 motion_y = 0;
105
106 offset = (src_y * uvlinesize) + src_x;
107 ptr = ref_picture[1] + offset;
108 if ((unsigned)src_x >= FFMAX((s->h_edge_pos >> 1) - 9, 0) ||
109 (unsigned)src_y >= FFMAX((s->v_edge_pos >> 1) - 9, 0)) {
110 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, ptr,
111 uvlinesize, uvlinesize,
112 9, 9,
113 src_x, src_y,
114 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
115 ptr = s->sc.edge_emu_buffer;
116 emu = 1;
117 }
118 s->mdsp.gmc1(dest_cb, ptr, uvlinesize, 8,
119 motion_x & 15, motion_y & 15, 128 - s->no_rounding);
120
121 ptr = ref_picture[2] + offset;
122 if (emu) {
123 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, ptr,
124 uvlinesize, uvlinesize,
125 9, 9,
126 src_x, src_y,
127 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
128 ptr = s->sc.edge_emu_buffer;
129 }
130 s->mdsp.gmc1(dest_cr, ptr, uvlinesize, 8,
131 motion_x & 15, motion_y & 15, 128 - s->no_rounding);
132 }
133
134 static void gmc_motion(MpegEncContext *s,
135 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
136 uint8_t **ref_picture)
137 {
138 uint8_t *ptr;
139 int linesize, uvlinesize;
140 const int a = s->sprite_warping_accuracy;
141 int ox, oy;
142
143 linesize = s->linesize;
144 uvlinesize = s->uvlinesize;
145
146 ptr = ref_picture[0];
147
148 ox = s->sprite_offset[0][0] + s->sprite_delta[0][0] * s->mb_x * 16 +
149 s->sprite_delta[0][1] * s->mb_y * 16;
150 oy = s->sprite_offset[0][1] + s->sprite_delta[1][0] * s->mb_x * 16 +
151 s->sprite_delta[1][1] * s->mb_y * 16;
152
153 s->mdsp.gmc(dest_y, ptr, linesize, 16,
154 ox, oy,
155 s->sprite_delta[0][0], s->sprite_delta[0][1],
156 s->sprite_delta[1][0], s->sprite_delta[1][1],
157 a + 1, (1 << (2 * a + 1)) - s->no_rounding,
158 s->h_edge_pos, s->v_edge_pos);
159 s->mdsp.gmc(dest_y + 8, ptr, linesize, 16,
160 ox + s->sprite_delta[0][0] * 8,
161 oy + s->sprite_delta[1][0] * 8,
162 s->sprite_delta[0][0], s->sprite_delta[0][1],
163 s->sprite_delta[1][0], s->sprite_delta[1][1],
164 a + 1, (1 << (2 * a + 1)) - s->no_rounding,
165 s->h_edge_pos, s->v_edge_pos);
166
167 if (CONFIG_GRAY && s->avctx->flags & AV_CODEC_FLAG_GRAY)
168 return;
169
170 ox = s->sprite_offset[1][0] + s->sprite_delta[0][0] * s->mb_x * 8 +
171 s->sprite_delta[0][1] * s->mb_y * 8;
172 oy = s->sprite_offset[1][1] + s->sprite_delta[1][0] * s->mb_x * 8 +
173 s->sprite_delta[1][1] * s->mb_y * 8;
174
175 ptr = ref_picture[1];
176 s->mdsp.gmc(dest_cb, ptr, uvlinesize, 8,
177 ox, oy,
178 s->sprite_delta[0][0], s->sprite_delta[0][1],
179 s->sprite_delta[1][0], s->sprite_delta[1][1],
180 a + 1, (1 << (2 * a + 1)) - s->no_rounding,
181 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
182
183 ptr = ref_picture[2];
184 s->mdsp.gmc(dest_cr, ptr, uvlinesize, 8,
185 ox, oy,
186 s->sprite_delta[0][0], s->sprite_delta[0][1],
187 s->sprite_delta[1][0], s->sprite_delta[1][1],
188 a + 1, (1 << (2 * a + 1)) - s->no_rounding,
189 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
190 }
191
192 static inline int hpel_motion(MpegEncContext *s,
193 uint8_t *dest, uint8_t *src,
194 int src_x, int src_y,
195 op_pixels_func *pix_op,
196 int motion_x, int motion_y)
197 {
198 int dxy = 0;
199 int emu = 0;
200
201 src_x += motion_x >> 1;
202 src_y += motion_y >> 1;
203
204 /* WARNING: do no forget half pels */
205 src_x = av_clip(src_x, -16, s->width); // FIXME unneeded for emu?
206 if (src_x != s->width)
207 dxy |= motion_x & 1;
208 src_y = av_clip(src_y, -16, s->height);
209 if (src_y != s->height)
210 dxy |= (motion_y & 1) << 1;
211 src += src_y * s->linesize + src_x;
212
213 if (s->unrestricted_mv) {
214 if ((unsigned)src_x > FFMAX(s->h_edge_pos - (motion_x & 1) - 8, 0) ||
215 (unsigned)src_y > FFMAX(s->v_edge_pos - (motion_y & 1) - 8, 0)) {
216 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, src,
217 s->linesize, s->linesize,
218 9, 9,
219 src_x, src_y, s->h_edge_pos,
220 s->v_edge_pos);
221 src = s->sc.edge_emu_buffer;
222 emu = 1;
223 }
224 }
225 pix_op[dxy](dest, src, s->linesize, 8);
226 return emu;
227 }
228
229 static av_always_inline
230 void emulated_edge_mc(MpegEncContext *s,
231 int src_x, int src_y,
232 int uvsrc_x, int uvsrc_y,
233 int field_based,
234 uint8_t **ptr_y,
235 uint8_t **ptr_cb,
236 uint8_t **ptr_cr)
237 {
238 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, *ptr_y,
239 s->linesize, s->linesize,
240 17, 17 + field_based,
241 src_x, src_y * (1 << field_based),
242 s->h_edge_pos, s->v_edge_pos);
243 *ptr_y = s->sc.edge_emu_buffer;
244 if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY)) {
245 uint8_t *uvbuf = s->sc.edge_emu_buffer + 18 * s->linesize;
246 s->vdsp.emulated_edge_mc(uvbuf, *ptr_cb,
247 s->uvlinesize, s->uvlinesize,
248 9, 9 + field_based,
249 uvsrc_x, uvsrc_y * (1 << field_based),
250 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
251 s->vdsp.emulated_edge_mc(uvbuf + 16, *ptr_cr,
252 s->uvlinesize, s->uvlinesize,
253 9, 9 + field_based,
254 uvsrc_x, uvsrc_y * (1 << field_based),
255 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
256 *ptr_cb = uvbuf;
257 *ptr_cr = uvbuf + 16;
258 }
259 }
260
261 static av_always_inline
262 void mpeg_motion_internal(MpegEncContext *s,
263 uint8_t *dest_y,
264 uint8_t *dest_cb,
265 uint8_t *dest_cr,
266 int field_based,
267 int bottom_field,
268 int field_select,
269 uint8_t **ref_picture,
270 op_pixels_func (*pix_op)[4],
271 int motion_x,
272 int motion_y,
273 int h,
274 int is_mpeg12,
275 int mb_y)
276 {
277 uint8_t *ptr_y, *ptr_cb, *ptr_cr;
278 int dxy, uvdxy, mx, my, src_x, src_y,
279 uvsrc_x, uvsrc_y, v_edge_pos;
280 ptrdiff_t uvlinesize, linesize;
281
282 #if 0
283 if (s->quarter_sample) {
284 motion_x >>= 1;
285 motion_y >>= 1;
286 }
287 #endif
288
289 v_edge_pos = s->v_edge_pos >> field_based;
290 linesize = s->current_picture.f->linesize[0] << field_based;
291 uvlinesize = s->current_picture.f->linesize[1] << field_based;
292
293 dxy = ((motion_y & 1) << 1) | (motion_x & 1);
294 src_x = s->mb_x * 16 + (motion_x >> 1);
295 src_y = (mb_y << (4 - field_based)) + (motion_y >> 1);
296
297 if (!is_mpeg12 && s->out_format == FMT_H263) {
298 if ((s->workaround_bugs & FF_BUG_HPEL_CHROMA) && field_based) {
299 mx = (motion_x >> 1) | (motion_x & 1);
300 my = motion_y >> 1;
301 uvdxy = ((my & 1) << 1) | (mx & 1);
302 uvsrc_x = s->mb_x * 8 + (mx >> 1);
303 uvsrc_y = (mb_y << (3 - field_based)) + (my >> 1);
304 } else {
305 uvdxy = dxy | (motion_y & 2) | ((motion_x & 2) >> 1);
306 uvsrc_x = src_x >> 1;
307 uvsrc_y = src_y >> 1;
308 }
309 // Even chroma mv's are full pel in H261
310 } else if (!is_mpeg12 && s->out_format == FMT_H261) {
311 mx = motion_x / 4;
312 my = motion_y / 4;
313 uvdxy = 0;
314 uvsrc_x = s->mb_x * 8 + mx;
315 uvsrc_y = mb_y * 8 + my;
316 } else {
317 if (s->chroma_y_shift) {
318 mx = motion_x / 2;
319 my = motion_y / 2;
320 uvdxy = ((my & 1) << 1) | (mx & 1);
321 uvsrc_x = s->mb_x * 8 + (mx >> 1);
322 uvsrc_y = (mb_y << (3 - field_based)) + (my >> 1);
323 } else {
324 if (s->chroma_x_shift) {
325 // Chroma422
326 mx = motion_x / 2;
327 uvdxy = ((motion_y & 1) << 1) | (mx & 1);
328 uvsrc_x = s->mb_x * 8 + (mx >> 1);
329 uvsrc_y = src_y;
330 } else {
331 // Chroma444
332 uvdxy = dxy;
333 uvsrc_x = src_x;
334 uvsrc_y = src_y;
335 }
336 }
337 }
338
339 ptr_y = ref_picture[0] + src_y * linesize + src_x;
340 ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x;
341 ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x;
342
343 if ((unsigned)src_x > FFMAX(s->h_edge_pos - (motion_x & 1) - 16, 0) ||
344 (unsigned)src_y > FFMAX(v_edge_pos - (motion_y & 1) - h, 0)) {
345 if (is_mpeg12 ||
346 s->codec_id == AV_CODEC_ID_MPEG2VIDEO ||
347 s->codec_id == AV_CODEC_ID_MPEG1VIDEO) {
348 av_log(s->avctx, AV_LOG_DEBUG,
349 "MPEG motion vector out of boundary (%d %d)\n", src_x,
350 src_y);
351 return;
352 }
353 emulated_edge_mc(s, src_x, src_y, uvsrc_x, uvsrc_y, field_based,
354 &ptr_y, &ptr_cb, &ptr_cr);
355 }
356
357 /* FIXME use this for field pix too instead of the obnoxious hack which
358 * changes picture.data */
359 if (bottom_field) {
360 dest_y += s->linesize;
361 dest_cb += s->uvlinesize;
362 dest_cr += s->uvlinesize;
363 }
364
365 if (field_select) {
366 ptr_y += s->linesize;
367 ptr_cb += s->uvlinesize;
368 ptr_cr += s->uvlinesize;
369 }
370
371 pix_op[0][dxy](dest_y, ptr_y, linesize, h);
372
373 if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY)) {
374 pix_op[s->chroma_x_shift][uvdxy]
375 (dest_cb, ptr_cb, uvlinesize, h >> s->chroma_y_shift);
376 pix_op[s->chroma_x_shift][uvdxy]
377 (dest_cr, ptr_cr, uvlinesize, h >> s->chroma_y_shift);
378 }
379 if (!is_mpeg12 && (CONFIG_H261_ENCODER || CONFIG_H261_DECODER) &&
380 s->out_format == FMT_H261) {
381 ff_h261_loop_filter(s);
382 }
383 }
384 /* apply one mpeg motion vector to the three components */
385 static void mpeg_motion(MpegEncContext *s,
386 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
387 int field_select, uint8_t **ref_picture,
388 op_pixels_func (*pix_op)[4],
389 int motion_x, int motion_y, int h, int mb_y)
390 {
391 #if !CONFIG_SMALL
392 if (s->out_format == FMT_MPEG1)
393 mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, 0, 0,
394 field_select, ref_picture, pix_op,
395 motion_x, motion_y, h, 1, mb_y);
396 else
397 #endif
398 mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, 0, 0,
399 field_select, ref_picture, pix_op,
400 motion_x, motion_y, h, 0, mb_y);
401 }
402
403 static void mpeg_motion_field(MpegEncContext *s, uint8_t *dest_y,
404 uint8_t *dest_cb, uint8_t *dest_cr,
405 int bottom_field, int field_select,
406 uint8_t **ref_picture,
407 op_pixels_func (*pix_op)[4],
408 int motion_x, int motion_y, int h, int mb_y)
409 {
410 #if !CONFIG_SMALL
411 if(s->out_format == FMT_MPEG1)
412 mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, 1,
413 bottom_field, field_select, ref_picture, pix_op,
414 motion_x, motion_y, h, 1, mb_y);
415 else
416 #endif
417 mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, 1,
418 bottom_field, field_select, ref_picture, pix_op,
419 motion_x, motion_y, h, 0, mb_y);
420 }
421
422 // FIXME: SIMDify, avg variant, 16x16 version
423 static inline void put_obmc(uint8_t *dst, uint8_t *src[5], int stride)
424 {
425 int x;
426 uint8_t *const top = src[1];
427 uint8_t *const left = src[2];
428 uint8_t *const mid = src[0];
429 uint8_t *const right = src[3];
430 uint8_t *const bottom = src[4];
431 #define OBMC_FILTER(x, t, l, m, r, b)\
432 dst[x]= (t*top[x] + l*left[x] + m*mid[x] + r*right[x] + b*bottom[x] + 4)>>3
433 #define OBMC_FILTER4(x, t, l, m, r, b)\
434 OBMC_FILTER(x , t, l, m, r, b);\
435 OBMC_FILTER(x+1 , t, l, m, r, b);\
436 OBMC_FILTER(x +stride, t, l, m, r, b);\
437 OBMC_FILTER(x+1+stride, t, l, m, r, b);
438
439 x = 0;
440 OBMC_FILTER (x , 2, 2, 4, 0, 0);
441 OBMC_FILTER (x + 1, 2, 1, 5, 0, 0);
442 OBMC_FILTER4(x + 2, 2, 1, 5, 0, 0);
443 OBMC_FILTER4(x + 4, 2, 0, 5, 1, 0);
444 OBMC_FILTER (x + 6, 2, 0, 5, 1, 0);
445 OBMC_FILTER (x + 7, 2, 0, 4, 2, 0);
446 x += stride;
447 OBMC_FILTER (x , 1, 2, 5, 0, 0);
448 OBMC_FILTER (x + 1, 1, 2, 5, 0, 0);
449 OBMC_FILTER (x + 6, 1, 0, 5, 2, 0);
450 OBMC_FILTER (x + 7, 1, 0, 5, 2, 0);
451 x += stride;
452 OBMC_FILTER4(x , 1, 2, 5, 0, 0);
453 OBMC_FILTER4(x + 2, 1, 1, 6, 0, 0);
454 OBMC_FILTER4(x + 4, 1, 0, 6, 1, 0);
455 OBMC_FILTER4(x + 6, 1, 0, 5, 2, 0);
456 x += 2 * stride;
457 OBMC_FILTER4(x , 0, 2, 5, 0, 1);
458 OBMC_FILTER4(x + 2, 0, 1, 6, 0, 1);
459 OBMC_FILTER4(x + 4, 0, 0, 6, 1, 1);
460 OBMC_FILTER4(x + 6, 0, 0, 5, 2, 1);
461 x += 2*stride;
462 OBMC_FILTER (x , 0, 2, 5, 0, 1);
463 OBMC_FILTER (x + 1, 0, 2, 5, 0, 1);
464 OBMC_FILTER4(x + 2, 0, 1, 5, 0, 2);
465 OBMC_FILTER4(x + 4, 0, 0, 5, 1, 2);
466 OBMC_FILTER (x + 6, 0, 0, 5, 2, 1);
467 OBMC_FILTER (x + 7, 0, 0, 5, 2, 1);
468 x += stride;
469 OBMC_FILTER (x , 0, 2, 4, 0, 2);
470 OBMC_FILTER (x + 1, 0, 1, 5, 0, 2);
471 OBMC_FILTER (x + 6, 0, 0, 5, 1, 2);
472 OBMC_FILTER (x + 7, 0, 0, 4, 2, 2);
473 }
474
475 /* obmc for 1 8x8 luma block */
476 static inline void obmc_motion(MpegEncContext *s,
477 uint8_t *dest, uint8_t *src,
478 int src_x, int src_y,
479 op_pixels_func *pix_op,
480 int16_t mv[5][2] /* mid top left right bottom */)
481 #define MID 0
482 {
483 int i;
484 uint8_t *ptr[5];
485
486 assert(s->quarter_sample == 0);
487
488 for (i = 0; i < 5; i++) {
489 if (i && mv[i][0] == mv[MID][0] && mv[i][1] == mv[MID][1]) {
490 ptr[i] = ptr[MID];
491 } else {
492 ptr[i] = s->sc.obmc_scratchpad + 8 * (i & 1) +
493 s->linesize * 8 * (i >> 1);
494 hpel_motion(s, ptr[i], src, src_x, src_y, pix_op,
495 mv[i][0], mv[i][1]);
496 }
497 }
498
499 put_obmc(dest, ptr, s->linesize);
500 }
501
502 static inline void qpel_motion(MpegEncContext *s,
503 uint8_t *dest_y,
504 uint8_t *dest_cb,
505 uint8_t *dest_cr,
506 int field_based, int bottom_field,
507 int field_select, uint8_t **ref_picture,
508 op_pixels_func (*pix_op)[4],
509 qpel_mc_func (*qpix_op)[16],
510 int motion_x, int motion_y, int h)
511 {
512 uint8_t *ptr_y, *ptr_cb, *ptr_cr;
513 int dxy, uvdxy, mx, my, src_x, src_y, uvsrc_x, uvsrc_y, v_edge_pos;
514 ptrdiff_t linesize, uvlinesize;
515
516 dxy = ((motion_y & 3) << 2) | (motion_x & 3);
517
518 src_x = s->mb_x * 16 + (motion_x >> 2);
519 src_y = s->mb_y * (16 >> field_based) + (motion_y >> 2);
520
521 v_edge_pos = s->v_edge_pos >> field_based;
522 linesize = s->linesize << field_based;
523 uvlinesize = s->uvlinesize << field_based;
524
525 if (field_based) {
526 mx = motion_x / 2;
527 my = motion_y >> 1;
528 } else if (s->workaround_bugs & FF_BUG_QPEL_CHROMA2) {
529 static const int rtab[8] = { 0, 0, 1, 1, 0, 0, 0, 1 };
530 mx = (motion_x >> 1) + rtab[motion_x & 7];
531 my = (motion_y >> 1) + rtab[motion_y & 7];
532 } else if (s->workaround_bugs & FF_BUG_QPEL_CHROMA) {
533 mx = (motion_x >> 1) | (motion_x & 1);
534 my = (motion_y >> 1) | (motion_y & 1);
535 } else {
536 mx = motion_x / 2;
537 my = motion_y / 2;
538 }
539 mx = (mx >> 1) | (mx & 1);
540 my = (my >> 1) | (my & 1);
541
542 uvdxy = (mx & 1) | ((my & 1) << 1);
543 mx >>= 1;
544 my >>= 1;
545
546 uvsrc_x = s->mb_x * 8 + mx;
547 uvsrc_y = s->mb_y * (8 >> field_based) + my;
548
549 ptr_y = ref_picture[0] + src_y * linesize + src_x;
550 ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x;
551 ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x;
552
553 if ((unsigned)src_x > FFMAX(s->h_edge_pos - (motion_x & 3) - 16, 0) ||
554 (unsigned)src_y > FFMAX(v_edge_pos - (motion_y & 3) - h, 0)) {
555 emulated_edge_mc(s, src_x, src_y, uvsrc_x, uvsrc_y, field_based,
556 &ptr_y, &ptr_cb, &ptr_cr);
557 }
558
559 if (!field_based)
560 qpix_op[0][dxy](dest_y, ptr_y, linesize);
561 else {
562 if (bottom_field) {
563 dest_y += s->linesize;
564 dest_cb += s->uvlinesize;
565 dest_cr += s->uvlinesize;
566 }
567
568 if (field_select) {
569 ptr_y += s->linesize;
570 ptr_cb += s->uvlinesize;
571 ptr_cr += s->uvlinesize;
572 }
573 // damn interlaced mode
574 // FIXME boundary mirroring is not exactly correct here
575 qpix_op[1][dxy](dest_y, ptr_y, linesize);
576 qpix_op[1][dxy](dest_y + 8, ptr_y + 8, linesize);
577 }
578 if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY)) {
579 pix_op[1][uvdxy](dest_cr, ptr_cr, uvlinesize, h >> 1);
580 pix_op[1][uvdxy](dest_cb, ptr_cb, uvlinesize, h >> 1);
581 }
582 }
583
584 /**
585 * h263 chroma 4mv motion compensation.
586 */
587 static void chroma_4mv_motion(MpegEncContext *s,
588 uint8_t *dest_cb, uint8_t *dest_cr,
589 uint8_t **ref_picture,
590 op_pixels_func *pix_op,
591 int mx, int my)
592 {
593 uint8_t *ptr;
594 int src_x, src_y, dxy, emu = 0;
595 ptrdiff_t offset;
596
597 /* In case of 8X8, we construct a single chroma motion vector
598 * with a special rounding */
599 mx = ff_h263_round_chroma(mx);
600 my = ff_h263_round_chroma(my);
601
602 dxy = ((my & 1) << 1) | (mx & 1);
603 mx >>= 1;
604 my >>= 1;
605
606 src_x = s->mb_x * 8 + mx;
607 src_y = s->mb_y * 8 + my;
608 src_x = av_clip(src_x, -8, (s->width >> 1));
609 if (src_x == (s->width >> 1))
610 dxy &= ~1;
611 src_y = av_clip(src_y, -8, (s->height >> 1));
612 if (src_y == (s->height >> 1))
613 dxy &= ~2;
614
615 offset = src_y * s->uvlinesize + src_x;
616 ptr = ref_picture[1] + offset;
617 if ((unsigned)src_x > FFMAX((s->h_edge_pos >> 1) - (dxy & 1) - 8, 0) ||
618 (unsigned)src_y > FFMAX((s->v_edge_pos >> 1) - (dxy >> 1) - 8, 0)) {
619 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, ptr,
620 s->uvlinesize, s->uvlinesize,
621 9, 9, src_x, src_y,
622 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
623 ptr = s->sc.edge_emu_buffer;
624 emu = 1;
625 }
626 pix_op[dxy](dest_cb, ptr, s->uvlinesize, 8);
627
628 ptr = ref_picture[2] + offset;
629 if (emu) {
630 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, ptr,
631 s->uvlinesize, s->uvlinesize,
632 9, 9, src_x, src_y,
633 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
634 ptr = s->sc.edge_emu_buffer;
635 }
636 pix_op[dxy](dest_cr, ptr, s->uvlinesize, 8);
637 }
638
639 static inline void prefetch_motion(MpegEncContext *s, uint8_t **pix, int dir)
640 {
641 /* fetch pixels for estimated mv 4 macroblocks ahead
642 * optimized for 64byte cache lines */
643 const int shift = s->quarter_sample ? 2 : 1;
644 const int mx = (s->mv[dir][0][0] >> shift) + 16 * s->mb_x + 8;
645 const int my = (s->mv[dir][0][1] >> shift) + 16 * s->mb_y;
646 int off = mx + (my + (s->mb_x & 3) * 4) * s->linesize + 64;
647
648 s->vdsp.prefetch(pix[0] + off, s->linesize, 4);
649 off = (mx >> 1) + ((my >> 1) + (s->mb_x & 7)) * s->uvlinesize + 64;
650 s->vdsp.prefetch(pix[1] + off, pix[2] - pix[1], 2);
651 }
652
653 static inline void apply_obmc(MpegEncContext *s,
654 uint8_t *dest_y,
655 uint8_t *dest_cb,
656 uint8_t *dest_cr,
657 uint8_t **ref_picture,
658 op_pixels_func (*pix_op)[4])
659 {
660 LOCAL_ALIGNED_8(int16_t, mv_cache, [4], [4][2]);
661 Picture *cur_frame = &s->current_picture;
662 int mb_x = s->mb_x;
663 int mb_y = s->mb_y;
664 const int xy = mb_x + mb_y * s->mb_stride;
665 const int mot_stride = s->b8_stride;
666 const int mot_xy = mb_x * 2 + mb_y * 2 * mot_stride;
667 int mx, my, i;
668
669 assert(!s->mb_skipped);
670
671 AV_COPY32(mv_cache[1][1], cur_frame->motion_val[0][mot_xy]);
672 AV_COPY32(mv_cache[1][2], cur_frame->motion_val[0][mot_xy + 1]);
673
674 AV_COPY32(mv_cache[2][1],
675 cur_frame->motion_val[0][mot_xy + mot_stride]);
676 AV_COPY32(mv_cache[2][2],
677 cur_frame->motion_val[0][mot_xy + mot_stride + 1]);
678
679 AV_COPY32(mv_cache[3][1],
680 cur_frame->motion_val[0][mot_xy + mot_stride]);
681 AV_COPY32(mv_cache[3][2],
682 cur_frame->motion_val[0][mot_xy + mot_stride + 1]);
683
684 if (mb_y == 0 || IS_INTRA(cur_frame->mb_type[xy - s->mb_stride])) {
685 AV_COPY32(mv_cache[0][1], mv_cache[1][1]);
686 AV_COPY32(mv_cache[0][2], mv_cache[1][2]);
687 } else {
688 AV_COPY32(mv_cache[0][1],
689 cur_frame->motion_val[0][mot_xy - mot_stride]);
690 AV_COPY32(mv_cache[0][2],
691 cur_frame->motion_val[0][mot_xy - mot_stride + 1]);
692 }
693
694 if (mb_x == 0 || IS_INTRA(cur_frame->mb_type[xy - 1])) {
695 AV_COPY32(mv_cache[1][0], mv_cache[1][1]);
696 AV_COPY32(mv_cache[2][0], mv_cache[2][1]);
697 } else {
698 AV_COPY32(mv_cache[1][0], cur_frame->motion_val[0][mot_xy - 1]);
699 AV_COPY32(mv_cache[2][0],
700 cur_frame->motion_val[0][mot_xy - 1 + mot_stride]);
701 }
702
703 if (mb_x + 1 >= s->mb_width || IS_INTRA(cur_frame->mb_type[xy + 1])) {
704 AV_COPY32(mv_cache[1][3], mv_cache[1][2]);
705 AV_COPY32(mv_cache[2][3], mv_cache[2][2]);
706 } else {
707 AV_COPY32(mv_cache[1][3], cur_frame->motion_val[0][mot_xy + 2]);
708 AV_COPY32(mv_cache[2][3],
709 cur_frame->motion_val[0][mot_xy + 2 + mot_stride]);
710 }
711
712 mx = 0;
713 my = 0;
714 for (i = 0; i < 4; i++) {
715 const int x = (i & 1) + 1;
716 const int y = (i >> 1) + 1;
717 int16_t mv[5][2] = {
718 { mv_cache[y][x][0], mv_cache[y][x][1] },
719 { mv_cache[y - 1][x][0], mv_cache[y - 1][x][1] },
720 { mv_cache[y][x - 1][0], mv_cache[y][x - 1][1] },
721 { mv_cache[y][x + 1][0], mv_cache[y][x + 1][1] },
722 { mv_cache[y + 1][x][0], mv_cache[y + 1][x][1] }
723 };
724 // FIXME cleanup
725 obmc_motion(s, dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize,
726 ref_picture[0],
727 mb_x * 16 + (i & 1) * 8, mb_y * 16 + (i >> 1) * 8,
728 pix_op[1],
729 mv);
730
731 mx += mv[0][0];
732 my += mv[0][1];
733 }
734 if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY))
735 chroma_4mv_motion(s, dest_cb, dest_cr,
736 ref_picture, pix_op[1],
737 mx, my);
738 }
739
740 static inline void apply_8x8(MpegEncContext *s,
741 uint8_t *dest_y,
742 uint8_t *dest_cb,
743 uint8_t *dest_cr,
744 int dir,
745 uint8_t **ref_picture,
746 qpel_mc_func (*qpix_op)[16],
747 op_pixels_func (*pix_op)[4])
748 {
749 int dxy, mx, my, src_x, src_y;
750 int i;
751 int mb_x = s->mb_x;
752 int mb_y = s->mb_y;
753 uint8_t *ptr, *dest;
754
755 mx = 0;
756 my = 0;
757 if (s->quarter_sample) {
758 for (i = 0; i < 4; i++) {
759 int motion_x = s->mv[dir][i][0];
760 int motion_y = s->mv[dir][i][1];
761
762 dxy = ((motion_y & 3) << 2) | (motion_x & 3);
763 src_x = mb_x * 16 + (motion_x >> 2) + (i & 1) * 8;
764 src_y = mb_y * 16 + (motion_y >> 2) + (i >> 1) * 8;
765
766 /* WARNING: do no forget half pels */
767 src_x = av_clip(src_x, -16, s->width);
768 if (src_x == s->width)
769 dxy &= ~3;
770 src_y = av_clip(src_y, -16, s->height);
771 if (src_y == s->height)
772 dxy &= ~12;
773
774 ptr = ref_picture[0] + (src_y * s->linesize) + (src_x);
775 if ((unsigned)src_x > FFMAX(s->h_edge_pos - (motion_x & 3) - 8, 0) ||
776 (unsigned)src_y > FFMAX(s->v_edge_pos - (motion_y & 3) - 8, 0)) {
777 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, ptr,
778 s->linesize, s->linesize,
779 9, 9,
780 src_x, src_y,
781 s->h_edge_pos,
782 s->v_edge_pos);
783 ptr = s->sc.edge_emu_buffer;
784 }
785 dest = dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize;
786 qpix_op[1][dxy](dest, ptr, s->linesize);
787
788 mx += s->mv[dir][i][0] / 2;
789 my += s->mv[dir][i][1] / 2;
790 }
791 } else {
792 for (i = 0; i < 4; i++) {
793 hpel_motion(s,
794 dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize,
795 ref_picture[0],
796 mb_x * 16 + (i & 1) * 8,
797 mb_y * 16 + (i >> 1) * 8,
798 pix_op[1],
799 s->mv[dir][i][0],
800 s->mv[dir][i][1]);
801
802 mx += s->mv[dir][i][0];
803 my += s->mv[dir][i][1];
804 }
805 }
806
807 if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY))
808 chroma_4mv_motion(s, dest_cb, dest_cr,
809 ref_picture, pix_op[1], mx, my);
810 }
811
812 /**
813 * motion compensation of a single macroblock
814 * @param s context
815 * @param dest_y luma destination pointer
816 * @param dest_cb chroma cb/u destination pointer
817 * @param dest_cr chroma cr/v destination pointer
818 * @param dir direction (0->forward, 1->backward)
819 * @param ref_picture array[3] of pointers to the 3 planes of the reference picture
820 * @param pix_op halfpel motion compensation function (average or put normally)
821 * @param qpix_op qpel motion compensation function (average or put normally)
822 * the motion vectors are taken from s->mv and the MV type from s->mv_type
823 */
824 static av_always_inline void mpv_motion_internal(MpegEncContext *s,
825 uint8_t *dest_y,
826 uint8_t *dest_cb,
827 uint8_t *dest_cr,
828 int dir,
829 uint8_t **ref_picture,
830 op_pixels_func (*pix_op)[4],
831 qpel_mc_func (*qpix_op)[16],
832 int is_mpeg12)
833 {
834 int i;
835 int mb_y = s->mb_y;
836
837 prefetch_motion(s, ref_picture, dir);
838
839 if (!is_mpeg12 && s->obmc && s->pict_type != AV_PICTURE_TYPE_B) {
840 apply_obmc(s, dest_y, dest_cb, dest_cr, ref_picture, pix_op);
841 return;
842 }
843
844 switch (s->mv_type) {
845 case MV_TYPE_16X16:
846 if (s->mcsel) {
847 if (s->real_sprite_warping_points == 1) {
848 gmc1_motion(s, dest_y, dest_cb, dest_cr,
849 ref_picture);
850 } else {
851 gmc_motion(s, dest_y, dest_cb, dest_cr,
852 ref_picture);
853 }
854 } else if (!is_mpeg12 && s->quarter_sample) {
855 qpel_motion(s, dest_y, dest_cb, dest_cr,
856 0, 0, 0,
857 ref_picture, pix_op, qpix_op,
858 s->mv[dir][0][0], s->mv[dir][0][1], 16);
859 } else if (!is_mpeg12 && (CONFIG_WMV2_DECODER || CONFIG_WMV2_ENCODER) &&
860 s->mspel && s->codec_id == AV_CODEC_ID_WMV2) {
861 ff_mspel_motion(s, dest_y, dest_cb, dest_cr,
862 ref_picture, pix_op,
863 s->mv[dir][0][0], s->mv[dir][0][1], 16);
864 } else {
865 mpeg_motion(s, dest_y, dest_cb, dest_cr, 0,
866 ref_picture, pix_op,
867 s->mv[dir][0][0], s->mv[dir][0][1], 16, mb_y);
868 }
869 break;
870 case MV_TYPE_8X8:
871 if (!is_mpeg12)
872 apply_8x8(s, dest_y, dest_cb, dest_cr,
873 dir, ref_picture, qpix_op, pix_op);
874 break;
875 case MV_TYPE_FIELD:
876 if (s->picture_structure == PICT_FRAME) {
877 if (!is_mpeg12 && s->quarter_sample) {
878 for (i = 0; i < 2; i++)
879 qpel_motion(s, dest_y, dest_cb, dest_cr,
880 1, i, s->field_select[dir][i],
881 ref_picture, pix_op, qpix_op,
882 s->mv[dir][i][0], s->mv[dir][i][1], 8);
883 } else {
884 /* top field */
885 mpeg_motion_field(s, dest_y, dest_cb, dest_cr,
886 0, s->field_select[dir][0],
887 ref_picture, pix_op,
888 s->mv[dir][0][0], s->mv[dir][0][1], 8, mb_y);
889 /* bottom field */
890 mpeg_motion_field(s, dest_y, dest_cb, dest_cr,
891 1, s->field_select[dir][1],
892 ref_picture, pix_op,
893 s->mv[dir][1][0], s->mv[dir][1][1], 8, mb_y);
894 }
895 } else {
896 if (s->picture_structure != s->field_select[dir][0] + 1 &&
897 s->pict_type != AV_PICTURE_TYPE_B && !s->first_field) {
898 ref_picture = s->current_picture_ptr->f->data;
899 }
900
901 mpeg_motion(s, dest_y, dest_cb, dest_cr,
902 s->field_select[dir][0],
903 ref_picture, pix_op,
904 s->mv[dir][0][0], s->mv[dir][0][1], 16, mb_y >> 1);
905 }
906 break;
907 case MV_TYPE_16X8:
908 for (i = 0; i < 2; i++) {
909 uint8_t **ref2picture;
910
911 if (s->picture_structure == s->field_select[dir][i] + 1
912 || s->pict_type == AV_PICTURE_TYPE_B || s->first_field) {
913 ref2picture = ref_picture;
914 } else {
915 ref2picture = s->current_picture_ptr->f->data;
916 }
917
918 mpeg_motion(s, dest_y, dest_cb, dest_cr,
919 s->field_select[dir][i],
920 ref2picture, pix_op,
921 s->mv[dir][i][0], s->mv[dir][i][1] + 16 * i,
922 8, mb_y >> 1);
923
924 dest_y += 16 * s->linesize;
925 dest_cb += (16 >> s->chroma_y_shift) * s->uvlinesize;
926 dest_cr += (16 >> s->chroma_y_shift) * s->uvlinesize;
927 }
928 break;
929 case MV_TYPE_DMV:
930 if (s->picture_structure == PICT_FRAME) {
931 for (i = 0; i < 2; i++) {
932 int j;
933 for (j = 0; j < 2; j++)
934 mpeg_motion_field(s, dest_y, dest_cb, dest_cr,
935 j, j ^ i, ref_picture, pix_op,
936 s->mv[dir][2 * i + j][0],
937 s->mv[dir][2 * i + j][1], 8, mb_y);
938 pix_op = s->hdsp.avg_pixels_tab;
939 }
940 } else {
941 for (i = 0; i < 2; i++) {
942 mpeg_motion(s, dest_y, dest_cb, dest_cr,
943 s->picture_structure != i + 1,
944 ref_picture, pix_op,
945 s->mv[dir][2 * i][0], s->mv[dir][2 * i][1],
946 16, mb_y >> 1);
947
948 // after put we make avg of the same block
949 pix_op = s->hdsp.avg_pixels_tab;
950
951 /* opposite parity is always in the same frame if this is
952 * second field */
953 if (!s->first_field) {
954 ref_picture = s->current_picture_ptr->f->data;
955 }
956 }
957 }
958 break;
959 default: assert(0);
960 }
961 }
962
963 void ff_mpv_motion(MpegEncContext *s,
964 uint8_t *dest_y, uint8_t *dest_cb,
965 uint8_t *dest_cr, int dir,
966 uint8_t **ref_picture,
967 op_pixels_func (*pix_op)[4],
968 qpel_mc_func (*qpix_op)[16])
969 {
970 #if !CONFIG_SMALL
971 if (s->out_format == FMT_MPEG1)
972 mpv_motion_internal(s, dest_y, dest_cb, dest_cr, dir,
973 ref_picture, pix_op, qpix_op, 1);
974 else
975 #endif
976 mpv_motion_internal(s, dest_y, dest_cb, dest_cr, dir,
977 ref_picture, pix_op, qpix_op, 0);
978 }