1 /*
2 * Copyright (c) 2000,2001 Fabrice Bellard
3 * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
4 *
5 * 4MV & hq & B-frame encoding stuff by Michael Niedermayer <michaelni@gmx.at>
6 *
7 * This file is part of FFmpeg.
8 *
9 * FFmpeg is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public
11 * License as published by the Free Software Foundation; either
12 * version 2.1 of the License, or (at your option) any later version.
13 *
14 * FFmpeg is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
18 *
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with FFmpeg; if not, write to the Free Software
21 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
22 */
23
24 #include <string.h>
25
26 #include "libavutil/avassert.h"
27 #include "libavutil/internal.h"
28 #include "avcodec.h"
29 #include "h261.h"
30 #include "mpegutils.h"
31 #include "mpegvideo.h"
32 #include "mjpegenc.h"
33 #include "msmpeg4.h"
34 #include "qpeldsp.h"
35 #include "wmv2.h"
36 #include <limits.h>
37
gmc1_motion(MpegEncContext * s,uint8_t * dest_y,uint8_t * dest_cb,uint8_t * dest_cr,uint8_t ** ref_picture)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)
41 {
42 uint8_t *ptr;
43 int src_x, src_y, motion_x, motion_y;
44 ptrdiff_t offset, linesize, uvlinesize;
45 int emu = 0;
46
47 motion_x = s->sprite_offset[0][0];
48 motion_y = s->sprite_offset[0][1];
49 src_x = s->mb_x * 16 + (motion_x >> (s->sprite_warping_accuracy + 1));
50 src_y = s->mb_y * 16 + (motion_y >> (s->sprite_warping_accuracy + 1));
51 motion_x *= 1 << (3 - s->sprite_warping_accuracy);
52 motion_y *= 1 << (3 - s->sprite_warping_accuracy);
53 src_x = av_clip(src_x, -16, s->width);
54 if (src_x == s->width)
55 motion_x = 0;
56 src_y = av_clip(src_y, -16, s->height);
57 if (src_y == s->height)
58 motion_y = 0;
59
60 linesize = s->linesize;
61 uvlinesize = s->uvlinesize;
62
63 ptr = ref_picture[0] + src_y * linesize + src_x;
64
65 if ((unsigned)src_x >= FFMAX(s->h_edge_pos - 17, 0) ||
66 (unsigned)src_y >= FFMAX(s->v_edge_pos - 17, 0)) {
67 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, ptr,
68 linesize, linesize,
69 17, 17,
70 src_x, src_y,
71 s->h_edge_pos, s->v_edge_pos);
72 ptr = s->sc.edge_emu_buffer;
73 }
74
75 if ((motion_x | motion_y) & 7) {
76 s->mdsp.gmc1(dest_y, ptr, linesize, 16,
77 motion_x & 15, motion_y & 15, 128 - s->no_rounding);
78 s->mdsp.gmc1(dest_y + 8, ptr + 8, linesize, 16,
79 motion_x & 15, motion_y & 15, 128 - s->no_rounding);
80 } else {
81 int dxy;
82
83 dxy = ((motion_x >> 3) & 1) | ((motion_y >> 2) & 2);
84 if (s->no_rounding) {
85 s->hdsp.put_no_rnd_pixels_tab[0][dxy](dest_y, ptr, linesize, 16);
86 } else {
87 s->hdsp.put_pixels_tab[0][dxy](dest_y, ptr, linesize, 16);
88 }
89 }
90
91 if (CONFIG_GRAY && s->avctx->flags & AV_CODEC_FLAG_GRAY)
92 return;
93
94 motion_x = s->sprite_offset[1][0];
95 motion_y = s->sprite_offset[1][1];
96 src_x = s->mb_x * 8 + (motion_x >> (s->sprite_warping_accuracy + 1));
97 src_y = s->mb_y * 8 + (motion_y >> (s->sprite_warping_accuracy + 1));
98 motion_x *= 1 << (3 - s->sprite_warping_accuracy);
99 motion_y *= 1 << (3 - s->sprite_warping_accuracy);
100 src_x = av_clip(src_x, -8, s->width >> 1);
101 if (src_x == s->width >> 1)
102 motion_x = 0;
103 src_y = av_clip(src_y, -8, s->height >> 1);
104 if (src_y == s->height >> 1)
105 motion_y = 0;
106
107 offset = (src_y * uvlinesize) + src_x;
108 ptr = ref_picture[1] + offset;
109 if ((unsigned)src_x >= FFMAX((s->h_edge_pos >> 1) - 9, 0) ||
110 (unsigned)src_y >= FFMAX((s->v_edge_pos >> 1) - 9, 0)) {
111 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, ptr,
112 uvlinesize, uvlinesize,
113 9, 9,
114 src_x, src_y,
115 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
116 ptr = s->sc.edge_emu_buffer;
117 emu = 1;
118 }
119 s->mdsp.gmc1(dest_cb, ptr, uvlinesize, 8,
120 motion_x & 15, motion_y & 15, 128 - s->no_rounding);
121
122 ptr = ref_picture[2] + offset;
123 if (emu) {
124 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, ptr,
125 uvlinesize, uvlinesize,
126 9, 9,
127 src_x, src_y,
128 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
129 ptr = s->sc.edge_emu_buffer;
130 }
131 s->mdsp.gmc1(dest_cr, ptr, uvlinesize, 8,
132 motion_x & 15, motion_y & 15, 128 - s->no_rounding);
133 }
134
gmc_motion(MpegEncContext * s,uint8_t * dest_y,uint8_t * dest_cb,uint8_t * dest_cr,uint8_t ** ref_picture)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)
138 {
139 uint8_t *ptr;
140 int linesize, uvlinesize;
141 const int a = s->sprite_warping_accuracy;
142 int ox, oy;
143
144 linesize = s->linesize;
145 uvlinesize = s->uvlinesize;
146
147 ptr = ref_picture[0];
148
149 ox = s->sprite_offset[0][0] + s->sprite_delta[0][0] * s->mb_x * 16 +
150 s->sprite_delta[0][1] * s->mb_y * 16;
151 oy = s->sprite_offset[0][1] + s->sprite_delta[1][0] * s->mb_x * 16 +
152 s->sprite_delta[1][1] * s->mb_y * 16;
153
154 s->mdsp.gmc(dest_y, ptr, linesize, 16,
155 ox, oy,
156 s->sprite_delta[0][0], s->sprite_delta[0][1],
157 s->sprite_delta[1][0], s->sprite_delta[1][1],
158 a + 1, (1 << (2 * a + 1)) - s->no_rounding,
159 s->h_edge_pos, s->v_edge_pos);
160 s->mdsp.gmc(dest_y + 8, ptr, linesize, 16,
161 ox + s->sprite_delta[0][0] * 8,
162 oy + s->sprite_delta[1][0] * 8,
163 s->sprite_delta[0][0], s->sprite_delta[0][1],
164 s->sprite_delta[1][0], s->sprite_delta[1][1],
165 a + 1, (1 << (2 * a + 1)) - s->no_rounding,
166 s->h_edge_pos, s->v_edge_pos);
167
168 if (CONFIG_GRAY && s->avctx->flags & AV_CODEC_FLAG_GRAY)
169 return;
170
171 ox = s->sprite_offset[1][0] + s->sprite_delta[0][0] * s->mb_x * 8 +
172 s->sprite_delta[0][1] * s->mb_y * 8;
173 oy = s->sprite_offset[1][1] + s->sprite_delta[1][0] * s->mb_x * 8 +
174 s->sprite_delta[1][1] * s->mb_y * 8;
175
176 ptr = ref_picture[1];
177 s->mdsp.gmc(dest_cb, ptr, uvlinesize, 8,
178 ox, oy,
179 s->sprite_delta[0][0], s->sprite_delta[0][1],
180 s->sprite_delta[1][0], s->sprite_delta[1][1],
181 a + 1, (1 << (2 * a + 1)) - s->no_rounding,
182 (s->h_edge_pos + 1) >> 1, (s->v_edge_pos + 1) >> 1);
183
184 ptr = ref_picture[2];
185 s->mdsp.gmc(dest_cr, ptr, uvlinesize, 8,
186 ox, oy,
187 s->sprite_delta[0][0], s->sprite_delta[0][1],
188 s->sprite_delta[1][0], s->sprite_delta[1][1],
189 a + 1, (1 << (2 * a + 1)) - s->no_rounding,
190 (s->h_edge_pos + 1) >> 1, (s->v_edge_pos + 1) >> 1);
191 }
192
hpel_motion(MpegEncContext * s,uint8_t * dest,uint8_t * src,int src_x,int src_y,op_pixels_func * pix_op,int motion_x,int motion_y)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)
198 {
199 int dxy = 0;
200 int emu = 0;
201
202 src_x += motion_x >> 1;
203 src_y += motion_y >> 1;
204
205 /* WARNING: do no forget half pels */
206 src_x = av_clip(src_x, -16, s->width); // FIXME unneeded for emu?
207 if (src_x != s->width)
208 dxy |= motion_x & 1;
209 src_y = av_clip(src_y, -16, s->height);
210 if (src_y != s->height)
211 dxy |= (motion_y & 1) << 1;
212 src += src_y * s->linesize + src_x;
213
214 if ((unsigned)src_x >= FFMAX(s->h_edge_pos - (motion_x & 1) - 7, 0) ||
215 (unsigned)src_y >= FFMAX(s->v_edge_pos - (motion_y & 1) - 7, 0)) {
216 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, src,
217 s->linesize, s->linesize,
218 9, 9,
219 src_x, src_y,
220 s->h_edge_pos, s->v_edge_pos);
221 src = s->sc.edge_emu_buffer;
222 emu = 1;
223 }
224 pix_op[dxy](dest, src, s->linesize, 8);
225 return emu;
226 }
227
228 static av_always_inline
mpeg_motion_internal(MpegEncContext * s,uint8_t * dest_y,uint8_t * dest_cb,uint8_t * dest_cr,int field_based,int bottom_field,int field_select,uint8_t ** ref_picture,op_pixels_func (* pix_op)[4],int motion_x,int motion_y,int h,int is_mpeg12,int is_16x8,int mb_y)229 void mpeg_motion_internal(MpegEncContext *s,
230 uint8_t *dest_y,
231 uint8_t *dest_cb,
232 uint8_t *dest_cr,
233 int field_based,
234 int bottom_field,
235 int field_select,
236 uint8_t **ref_picture,
237 op_pixels_func (*pix_op)[4],
238 int motion_x,
239 int motion_y,
240 int h,
241 int is_mpeg12,
242 int is_16x8,
243 int mb_y)
244 {
245 uint8_t *ptr_y, *ptr_cb, *ptr_cr;
246 int dxy, uvdxy, mx, my, src_x, src_y,
247 uvsrc_x, uvsrc_y, v_edge_pos, block_y_half;
248 ptrdiff_t uvlinesize, linesize;
249
250 v_edge_pos = s->v_edge_pos >> field_based;
251 linesize = s->current_picture.f->linesize[0] << field_based;
252 uvlinesize = s->current_picture.f->linesize[1] << field_based;
253 block_y_half = (field_based | is_16x8);
254
255 dxy = ((motion_y & 1) << 1) | (motion_x & 1);
256 src_x = s->mb_x * 16 + (motion_x >> 1);
257 src_y = (mb_y << (4 - block_y_half)) + (motion_y >> 1);
258
259 if (!is_mpeg12 && s->out_format == FMT_H263) {
260 if ((s->workaround_bugs & FF_BUG_HPEL_CHROMA) && field_based) {
261 mx = (motion_x >> 1) | (motion_x & 1);
262 my = motion_y >> 1;
263 uvdxy = ((my & 1) << 1) | (mx & 1);
264 uvsrc_x = s->mb_x * 8 + (mx >> 1);
265 uvsrc_y = (mb_y << (3 - block_y_half)) + (my >> 1);
266 } else {
267 uvdxy = dxy | (motion_y & 2) | ((motion_x & 2) >> 1);
268 uvsrc_x = src_x >> 1;
269 uvsrc_y = src_y >> 1;
270 }
271 // Even chroma mv's are full pel in H261
272 } else if (!is_mpeg12 && s->out_format == FMT_H261) {
273 mx = motion_x / 4;
274 my = motion_y / 4;
275 uvdxy = 0;
276 uvsrc_x = s->mb_x * 8 + mx;
277 uvsrc_y = mb_y * 8 + my;
278 } else {
279 if (s->chroma_y_shift) {
280 mx = motion_x / 2;
281 my = motion_y / 2;
282 uvdxy = ((my & 1) << 1) | (mx & 1);
283 uvsrc_x = s->mb_x * 8 + (mx >> 1);
284 uvsrc_y = (mb_y << (3 - block_y_half)) + (my >> 1);
285 } else {
286 if (s->chroma_x_shift) {
287 // Chroma422
288 mx = motion_x / 2;
289 uvdxy = ((motion_y & 1) << 1) | (mx & 1);
290 uvsrc_x = s->mb_x * 8 + (mx >> 1);
291 uvsrc_y = src_y;
292 } else {
293 // Chroma444
294 uvdxy = dxy;
295 uvsrc_x = src_x;
296 uvsrc_y = src_y;
297 }
298 }
299 }
300
301 ptr_y = ref_picture[0] + src_y * linesize + src_x;
302 ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x;
303 ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x;
304
305 if ((unsigned)src_x >= FFMAX(s->h_edge_pos - (motion_x & 1) - 15 , 0) ||
306 (unsigned)src_y >= FFMAX( v_edge_pos - (motion_y & 1) - h + 1, 0)) {
307 if (is_mpeg12 ||
308 s->codec_id == AV_CODEC_ID_MPEG2VIDEO ||
309 s->codec_id == AV_CODEC_ID_MPEG1VIDEO) {
310 av_log(s->avctx, AV_LOG_DEBUG,
311 "MPEG motion vector out of boundary (%d %d)\n", src_x,
312 src_y);
313 return;
314 }
315 src_y = (unsigned)src_y << field_based;
316 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, ptr_y,
317 s->linesize, s->linesize,
318 17, 17 + field_based,
319 src_x, src_y,
320 s->h_edge_pos, s->v_edge_pos);
321 ptr_y = s->sc.edge_emu_buffer;
322 if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY)) {
323 uint8_t *ubuf = s->sc.edge_emu_buffer + 18 * s->linesize;
324 uint8_t *vbuf = ubuf + 10 * s->uvlinesize;
325 if (s->workaround_bugs & FF_BUG_IEDGE)
326 vbuf -= s->uvlinesize;
327 uvsrc_y = (unsigned)uvsrc_y << field_based;
328 s->vdsp.emulated_edge_mc(ubuf, ptr_cb,
329 s->uvlinesize, s->uvlinesize,
330 9, 9 + field_based,
331 uvsrc_x, uvsrc_y,
332 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
333 s->vdsp.emulated_edge_mc(vbuf, ptr_cr,
334 s->uvlinesize, s->uvlinesize,
335 9, 9 + field_based,
336 uvsrc_x, uvsrc_y,
337 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
338 ptr_cb = ubuf;
339 ptr_cr = vbuf;
340 }
341 }
342
343 /* FIXME use this for field pix too instead of the obnoxious hack which
344 * changes picture.data */
345 if (bottom_field) {
346 dest_y += s->linesize;
347 dest_cb += s->uvlinesize;
348 dest_cr += s->uvlinesize;
349 }
350
351 if (field_select) {
352 ptr_y += s->linesize;
353 ptr_cb += s->uvlinesize;
354 ptr_cr += s->uvlinesize;
355 }
356
357 pix_op[0][dxy](dest_y, ptr_y, linesize, h);
358
359 if (!CONFIG_GRAY || !(s->avctx->flags & AV_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);
364 }
365 if (!is_mpeg12 && (CONFIG_H261_ENCODER || CONFIG_H261_DECODER) &&
366 s->out_format == FMT_H261) {
367 ff_h261_loop_filter(s);
368 }
369 }
370 /* apply one mpeg motion vector to the three components */
mpeg_motion(MpegEncContext * s,uint8_t * dest_y,uint8_t * dest_cb,uint8_t * dest_cr,int field_select,uint8_t ** ref_picture,op_pixels_func (* pix_op)[4],int motion_x,int motion_y,int h,int is_16x8,int mb_y)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 is_16x8, int mb_y)
376 {
377 #if !CONFIG_SMALL
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, is_16x8, mb_y);
382 else
383 #endif
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, is_16x8, mb_y);
387 }
388
mpeg_motion_field(MpegEncContext * s,uint8_t * dest_y,uint8_t * dest_cb,uint8_t * dest_cr,int bottom_field,int field_select,uint8_t ** ref_picture,op_pixels_func (* pix_op)[4],int motion_x,int motion_y,int h,int 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)
395 {
396 #if !CONFIG_SMALL
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, 0, mb_y);
401 else
402 #endif
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, 0, mb_y);
406 }
407
408 // FIXME: SIMDify, avg variant, 16x16 version
put_obmc(uint8_t * dst,uint8_t * src[5],int stride)409 static inline void put_obmc(uint8_t *dst, uint8_t *src[5], int stride)
410 {
411 int x;
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);
424
425 x = 0;
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);
432 x += stride;
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);
437 x += stride;
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);
442 x += 2 * stride;
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);
447 x += 2*stride;
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);
454 x += stride;
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);
459 }
460
461 /* obmc for 1 8x8 luma block */
obmc_motion(MpegEncContext * s,uint8_t * dest,uint8_t * src,int src_x,int src_y,op_pixels_func * pix_op,int16_t mv[5][2])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 */)
467 #define MID 0
468 {
469 int i;
470 uint8_t *ptr[5];
471
472 av_assert2(s->quarter_sample == 0);
473
474 for (i = 0; i < 5; i++) {
475 if (i && mv[i][0] == mv[MID][0] && mv[i][1] == mv[MID][1]) {
476 ptr[i] = ptr[MID];
477 } else {
478 ptr[i] = s->sc.obmc_scratchpad + 8 * (i & 1) +
479 s->linesize * 8 * (i >> 1);
480 hpel_motion(s, ptr[i], src, src_x, src_y, pix_op,
481 mv[i][0], mv[i][1]);
482 }
483 }
484
485 put_obmc(dest, ptr, s->linesize);
486 }
487
qpel_motion(MpegEncContext * s,uint8_t * dest_y,uint8_t * dest_cb,uint8_t * dest_cr,int field_based,int bottom_field,int field_select,uint8_t ** ref_picture,op_pixels_func (* pix_op)[4],qpel_mc_func (* qpix_op)[16],int motion_x,int motion_y,int h)488 static inline void qpel_motion(MpegEncContext *s,
489 uint8_t *dest_y,
490 uint8_t *dest_cb,
491 uint8_t *dest_cr,
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)
497 {
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;
501
502 dxy = ((motion_y & 3) << 2) | (motion_x & 3);
503
504 src_x = s->mb_x * 16 + (motion_x >> 2);
505 src_y = s->mb_y * (16 >> field_based) + (motion_y >> 2);
506
507 v_edge_pos = s->v_edge_pos >> field_based;
508 linesize = s->linesize << field_based;
509 uvlinesize = s->uvlinesize << field_based;
510
511 if (field_based) {
512 mx = motion_x / 2;
513 my = motion_y >> 1;
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);
521 } else {
522 mx = motion_x / 2;
523 my = motion_y / 2;
524 }
525 mx = (mx >> 1) | (mx & 1);
526 my = (my >> 1) | (my & 1);
527
528 uvdxy = (mx & 1) | ((my & 1) << 1);
529 mx >>= 1;
530 my >>= 1;
531
532 uvsrc_x = s->mb_x * 8 + mx;
533 uvsrc_y = s->mb_y * (8 >> field_based) + my;
534
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;
538
539 if ((unsigned)src_x >= FFMAX(s->h_edge_pos - (motion_x & 3) - 15 , 0) ||
540 (unsigned)src_y >= FFMAX( v_edge_pos - (motion_y & 3) - h + 1, 0)) {
541 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, ptr_y,
542 s->linesize, s->linesize,
543 17, 17 + field_based,
544 src_x, src_y * (1 << field_based),
545 s->h_edge_pos, s->v_edge_pos);
546 ptr_y = s->sc.edge_emu_buffer;
547 if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY)) {
548 uint8_t *ubuf = s->sc.edge_emu_buffer + 18 * s->linesize;
549 uint8_t *vbuf = ubuf + 10 * s->uvlinesize;
550 if (s->workaround_bugs & FF_BUG_IEDGE)
551 vbuf -= s->uvlinesize;
552 s->vdsp.emulated_edge_mc(ubuf, ptr_cb,
553 s->uvlinesize, s->uvlinesize,
554 9, 9 + field_based,
555 uvsrc_x, uvsrc_y * (1 << field_based),
556 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
557 s->vdsp.emulated_edge_mc(vbuf, ptr_cr,
558 s->uvlinesize, s->uvlinesize,
559 9, 9 + field_based,
560 uvsrc_x, uvsrc_y * (1 << field_based),
561 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
562 ptr_cb = ubuf;
563 ptr_cr = vbuf;
564 }
565 }
566
567 if (!field_based)
568 qpix_op[0][dxy](dest_y, ptr_y, linesize);
569 else {
570 if (bottom_field) {
571 dest_y += s->linesize;
572 dest_cb += s->uvlinesize;
573 dest_cr += s->uvlinesize;
574 }
575
576 if (field_select) {
577 ptr_y += s->linesize;
578 ptr_cb += s->uvlinesize;
579 ptr_cr += s->uvlinesize;
580 }
581 // damn interlaced mode
582 // FIXME boundary mirroring is not exactly correct here
583 qpix_op[1][dxy](dest_y, ptr_y, linesize);
584 qpix_op[1][dxy](dest_y + 8, ptr_y + 8, linesize);
585 }
586 if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY)) {
587 pix_op[1][uvdxy](dest_cr, ptr_cr, uvlinesize, h >> 1);
588 pix_op[1][uvdxy](dest_cb, ptr_cb, uvlinesize, h >> 1);
589 }
590 }
591
592 /**
593 * H.263 chroma 4mv motion compensation.
594 */
chroma_4mv_motion(MpegEncContext * s,uint8_t * dest_cb,uint8_t * dest_cr,uint8_t ** ref_picture,op_pixels_func * pix_op,int mx,int my)595 static void chroma_4mv_motion(MpegEncContext *s,
596 uint8_t *dest_cb, uint8_t *dest_cr,
597 uint8_t **ref_picture,
598 op_pixels_func *pix_op,
599 int mx, int my)
600 {
601 uint8_t *ptr;
602 int src_x, src_y, dxy, emu = 0;
603 ptrdiff_t offset;
604
605 /* In case of 8X8, we construct a single chroma motion vector
606 * with a special rounding */
607 mx = ff_h263_round_chroma(mx);
608 my = ff_h263_round_chroma(my);
609
610 dxy = ((my & 1) << 1) | (mx & 1);
611 mx >>= 1;
612 my >>= 1;
613
614 src_x = s->mb_x * 8 + mx;
615 src_y = s->mb_y * 8 + my;
616 src_x = av_clip(src_x, -8, (s->width >> 1));
617 if (src_x == (s->width >> 1))
618 dxy &= ~1;
619 src_y = av_clip(src_y, -8, (s->height >> 1));
620 if (src_y == (s->height >> 1))
621 dxy &= ~2;
622
623 offset = src_y * s->uvlinesize + src_x;
624 ptr = ref_picture[1] + offset;
625 if ((unsigned)src_x >= FFMAX((s->h_edge_pos >> 1) - (dxy & 1) - 7, 0) ||
626 (unsigned)src_y >= FFMAX((s->v_edge_pos >> 1) - (dxy >> 1) - 7, 0)) {
627 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, ptr,
628 s->uvlinesize, s->uvlinesize,
629 9, 9, src_x, src_y,
630 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
631 ptr = s->sc.edge_emu_buffer;
632 emu = 1;
633 }
634 pix_op[dxy](dest_cb, ptr, s->uvlinesize, 8);
635
636 ptr = ref_picture[2] + offset;
637 if (emu) {
638 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, ptr,
639 s->uvlinesize, s->uvlinesize,
640 9, 9, src_x, src_y,
641 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
642 ptr = s->sc.edge_emu_buffer;
643 }
644 pix_op[dxy](dest_cr, ptr, s->uvlinesize, 8);
645 }
646
prefetch_motion(MpegEncContext * s,uint8_t ** pix,int dir)647 static inline void prefetch_motion(MpegEncContext *s, uint8_t **pix, int dir)
648 {
649 /* fetch pixels for estimated mv 4 macroblocks ahead
650 * optimized for 64byte cache lines */
651 const int shift = s->quarter_sample ? 2 : 1;
652 const int mx = (s->mv[dir][0][0] >> shift) + 16 * s->mb_x + 8;
653 const int my = (s->mv[dir][0][1] >> shift) + 16 * s->mb_y;
654 int off = mx + (my + (s->mb_x & 3) * 4) * s->linesize + 64;
655
656 s->vdsp.prefetch(pix[0] + off, s->linesize, 4);
657 off = (mx >> 1) + ((my >> 1) + (s->mb_x & 7)) * s->uvlinesize + 64;
658 s->vdsp.prefetch(pix[1] + off, pix[2] - pix[1], 2);
659 }
660
apply_obmc(MpegEncContext * s,uint8_t * dest_y,uint8_t * dest_cb,uint8_t * dest_cr,uint8_t ** ref_picture,op_pixels_func (* pix_op)[4])661 static inline void apply_obmc(MpegEncContext *s,
662 uint8_t *dest_y,
663 uint8_t *dest_cb,
664 uint8_t *dest_cr,
665 uint8_t **ref_picture,
666 op_pixels_func (*pix_op)[4])
667 {
668 LOCAL_ALIGNED_8(int16_t, mv_cache, [4], [4][2]);
669 Picture *cur_frame = &s->current_picture;
670 int mb_x = s->mb_x;
671 int mb_y = s->mb_y;
672 const int xy = mb_x + mb_y * s->mb_stride;
673 const int mot_stride = s->b8_stride;
674 const int mot_xy = mb_x * 2 + mb_y * 2 * mot_stride;
675 int mx, my, i;
676
677 av_assert2(!s->mb_skipped);
678
679 AV_COPY32(mv_cache[1][1], cur_frame->motion_val[0][mot_xy]);
680 AV_COPY32(mv_cache[1][2], cur_frame->motion_val[0][mot_xy + 1]);
681
682 AV_COPY32(mv_cache[2][1],
683 cur_frame->motion_val[0][mot_xy + mot_stride]);
684 AV_COPY32(mv_cache[2][2],
685 cur_frame->motion_val[0][mot_xy + mot_stride + 1]);
686
687 AV_COPY32(mv_cache[3][1],
688 cur_frame->motion_val[0][mot_xy + mot_stride]);
689 AV_COPY32(mv_cache[3][2],
690 cur_frame->motion_val[0][mot_xy + mot_stride + 1]);
691
692 if (mb_y == 0 || IS_INTRA(cur_frame->mb_type[xy - s->mb_stride])) {
693 AV_COPY32(mv_cache[0][1], mv_cache[1][1]);
694 AV_COPY32(mv_cache[0][2], mv_cache[1][2]);
695 } else {
696 AV_COPY32(mv_cache[0][1],
697 cur_frame->motion_val[0][mot_xy - mot_stride]);
698 AV_COPY32(mv_cache[0][2],
699 cur_frame->motion_val[0][mot_xy - mot_stride + 1]);
700 }
701
702 if (mb_x == 0 || IS_INTRA(cur_frame->mb_type[xy - 1])) {
703 AV_COPY32(mv_cache[1][0], mv_cache[1][1]);
704 AV_COPY32(mv_cache[2][0], mv_cache[2][1]);
705 } else {
706 AV_COPY32(mv_cache[1][0], cur_frame->motion_val[0][mot_xy - 1]);
707 AV_COPY32(mv_cache[2][0],
708 cur_frame->motion_val[0][mot_xy - 1 + mot_stride]);
709 }
710
711 if (mb_x + 1 >= s->mb_width || IS_INTRA(cur_frame->mb_type[xy + 1])) {
712 AV_COPY32(mv_cache[1][3], mv_cache[1][2]);
713 AV_COPY32(mv_cache[2][3], mv_cache[2][2]);
714 } else {
715 AV_COPY32(mv_cache[1][3], cur_frame->motion_val[0][mot_xy + 2]);
716 AV_COPY32(mv_cache[2][3],
717 cur_frame->motion_val[0][mot_xy + 2 + mot_stride]);
718 }
719
720 mx = 0;
721 my = 0;
722 for (i = 0; i < 4; i++) {
723 const int x = (i & 1) + 1;
724 const int y = (i >> 1) + 1;
725 int16_t mv[5][2] = {
726 { mv_cache[y][x][0], mv_cache[y][x][1] },
727 { mv_cache[y - 1][x][0], mv_cache[y - 1][x][1] },
728 { mv_cache[y][x - 1][0], mv_cache[y][x - 1][1] },
729 { mv_cache[y][x + 1][0], mv_cache[y][x + 1][1] },
730 { mv_cache[y + 1][x][0], mv_cache[y + 1][x][1] }
731 };
732 // FIXME cleanup
733 obmc_motion(s, dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize,
734 ref_picture[0],
735 mb_x * 16 + (i & 1) * 8, mb_y * 16 + (i >> 1) * 8,
736 pix_op[1],
737 mv);
738
739 mx += mv[0][0];
740 my += mv[0][1];
741 }
742 if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY))
743 chroma_4mv_motion(s, dest_cb, dest_cr,
744 ref_picture, pix_op[1],
745 mx, my);
746 }
747
apply_8x8(MpegEncContext * s,uint8_t * dest_y,uint8_t * dest_cb,uint8_t * dest_cr,int dir,uint8_t ** ref_picture,qpel_mc_func (* qpix_op)[16],op_pixels_func (* pix_op)[4])748 static inline void apply_8x8(MpegEncContext *s,
749 uint8_t *dest_y,
750 uint8_t *dest_cb,
751 uint8_t *dest_cr,
752 int dir,
753 uint8_t **ref_picture,
754 qpel_mc_func (*qpix_op)[16],
755 op_pixels_func (*pix_op)[4])
756 {
757 int dxy, mx, my, src_x, src_y;
758 int i;
759 int mb_x = s->mb_x;
760 int mb_y = s->mb_y;
761 uint8_t *ptr, *dest;
762
763 mx = 0;
764 my = 0;
765 if (s->quarter_sample) {
766 for (i = 0; i < 4; i++) {
767 int motion_x = s->mv[dir][i][0];
768 int motion_y = s->mv[dir][i][1];
769
770 dxy = ((motion_y & 3) << 2) | (motion_x & 3);
771 src_x = mb_x * 16 + (motion_x >> 2) + (i & 1) * 8;
772 src_y = mb_y * 16 + (motion_y >> 2) + (i >> 1) * 8;
773
774 /* WARNING: do no forget half pels */
775 src_x = av_clip(src_x, -16, s->width);
776 if (src_x == s->width)
777 dxy &= ~3;
778 src_y = av_clip(src_y, -16, s->height);
779 if (src_y == s->height)
780 dxy &= ~12;
781
782 ptr = ref_picture[0] + (src_y * s->linesize) + (src_x);
783 if ((unsigned)src_x >= FFMAX(s->h_edge_pos - (motion_x & 3) - 7, 0) ||
784 (unsigned)src_y >= FFMAX(s->v_edge_pos - (motion_y & 3) - 7, 0)) {
785 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, ptr,
786 s->linesize, s->linesize,
787 9, 9,
788 src_x, src_y,
789 s->h_edge_pos,
790 s->v_edge_pos);
791 ptr = s->sc.edge_emu_buffer;
792 }
793 dest = dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize;
794 qpix_op[1][dxy](dest, ptr, s->linesize);
795
796 mx += s->mv[dir][i][0] / 2;
797 my += s->mv[dir][i][1] / 2;
798 }
799 } else {
800 for (i = 0; i < 4; i++) {
801 hpel_motion(s,
802 dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize,
803 ref_picture[0],
804 mb_x * 16 + (i & 1) * 8,
805 mb_y * 16 + (i >> 1) * 8,
806 pix_op[1],
807 s->mv[dir][i][0],
808 s->mv[dir][i][1]);
809
810 mx += s->mv[dir][i][0];
811 my += s->mv[dir][i][1];
812 }
813 }
814
815 if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY))
816 chroma_4mv_motion(s, dest_cb, dest_cr,
817 ref_picture, pix_op[1], mx, my);
818 }
819
820 /**
821 * motion compensation of a single macroblock
822 * @param s context
823 * @param dest_y luma destination pointer
824 * @param dest_cb chroma cb/u destination pointer
825 * @param dest_cr chroma cr/v destination pointer
826 * @param dir direction (0->forward, 1->backward)
827 * @param ref_picture array[3] of pointers to the 3 planes of the reference picture
828 * @param pix_op halfpel motion compensation function (average or put normally)
829 * @param qpix_op qpel motion compensation function (average or put normally)
830 * the motion vectors are taken from s->mv and the MV type from s->mv_type
831 */
mpv_motion_internal(MpegEncContext * s,uint8_t * dest_y,uint8_t * dest_cb,uint8_t * dest_cr,int dir,uint8_t ** ref_picture,op_pixels_func (* pix_op)[4],qpel_mc_func (* qpix_op)[16],int is_mpeg12)832 static av_always_inline void mpv_motion_internal(MpegEncContext *s,
833 uint8_t *dest_y,
834 uint8_t *dest_cb,
835 uint8_t *dest_cr,
836 int dir,
837 uint8_t **ref_picture,
838 op_pixels_func (*pix_op)[4],
839 qpel_mc_func (*qpix_op)[16],
840 int is_mpeg12)
841 {
842 int i;
843 int mb_y = s->mb_y;
844
845 prefetch_motion(s, ref_picture, dir);
846
847 if (!is_mpeg12 && s->obmc && s->pict_type != AV_PICTURE_TYPE_B) {
848 apply_obmc(s, dest_y, dest_cb, dest_cr, ref_picture, pix_op);
849 return;
850 }
851
852 switch (s->mv_type) {
853 case MV_TYPE_16X16:
854 if (s->mcsel) {
855 if (s->real_sprite_warping_points == 1) {
856 gmc1_motion(s, dest_y, dest_cb, dest_cr,
857 ref_picture);
858 } else {
859 gmc_motion(s, dest_y, dest_cb, dest_cr,
860 ref_picture);
861 }
862 } else if (!is_mpeg12 && s->quarter_sample) {
863 qpel_motion(s, dest_y, dest_cb, dest_cr,
864 0, 0, 0,
865 ref_picture, pix_op, qpix_op,
866 s->mv[dir][0][0], s->mv[dir][0][1], 16);
867 } else if (!is_mpeg12 && (CONFIG_WMV2_DECODER || CONFIG_WMV2_ENCODER) &&
868 s->mspel && s->codec_id == AV_CODEC_ID_WMV2) {
869 ff_mspel_motion(s, dest_y, dest_cb, dest_cr,
870 ref_picture, pix_op,
871 s->mv[dir][0][0], s->mv[dir][0][1], 16);
872 } else {
873 mpeg_motion(s, dest_y, dest_cb, dest_cr, 0,
874 ref_picture, pix_op,
875 s->mv[dir][0][0], s->mv[dir][0][1], 16, 0, mb_y);
876 }
877 break;
878 case MV_TYPE_8X8:
879 if (!is_mpeg12)
880 apply_8x8(s, dest_y, dest_cb, dest_cr,
881 dir, ref_picture, qpix_op, pix_op);
882 break;
883 case MV_TYPE_FIELD:
884 if (s->picture_structure == PICT_FRAME) {
885 if (!is_mpeg12 && s->quarter_sample) {
886 for (i = 0; i < 2; i++)
887 qpel_motion(s, dest_y, dest_cb, dest_cr,
888 1, i, s->field_select[dir][i],
889 ref_picture, pix_op, qpix_op,
890 s->mv[dir][i][0], s->mv[dir][i][1], 8);
891 } else {
892 /* top field */
893 mpeg_motion_field(s, dest_y, dest_cb, dest_cr,
894 0, s->field_select[dir][0],
895 ref_picture, pix_op,
896 s->mv[dir][0][0], s->mv[dir][0][1], 8, mb_y);
897 /* bottom field */
898 mpeg_motion_field(s, dest_y, dest_cb, dest_cr,
899 1, s->field_select[dir][1],
900 ref_picture, pix_op,
901 s->mv[dir][1][0], s->mv[dir][1][1], 8, mb_y);
902 }
903 } else {
904 if ( s->picture_structure != s->field_select[dir][0] + 1 && s->pict_type != AV_PICTURE_TYPE_B && !s->first_field
905 || !ref_picture[0]) {
906 ref_picture = s->current_picture_ptr->f->data;
907 }
908
909 mpeg_motion(s, dest_y, dest_cb, dest_cr,
910 s->field_select[dir][0],
911 ref_picture, pix_op,
912 s->mv[dir][0][0], s->mv[dir][0][1], 16, 0, mb_y >> 1);
913 }
914 break;
915 case MV_TYPE_16X8:
916 for (i = 0; i < 2; i++) {
917 uint8_t **ref2picture;
918
919 if ((s->picture_structure == s->field_select[dir][i] + 1
920 || s->pict_type == AV_PICTURE_TYPE_B || s->first_field) && ref_picture[0]) {
921 ref2picture = ref_picture;
922 } else {
923 ref2picture = s->current_picture_ptr->f->data;
924 }
925
926 mpeg_motion(s, dest_y, dest_cb, dest_cr,
927 s->field_select[dir][i],
928 ref2picture, pix_op,
929 s->mv[dir][i][0], s->mv[dir][i][1],
930 8, 1, (mb_y & ~1) + i);
931
932 dest_y += 16 * s->linesize;
933 dest_cb += (16 >> s->chroma_y_shift) * s->uvlinesize;
934 dest_cr += (16 >> s->chroma_y_shift) * s->uvlinesize;
935 }
936 break;
937 case MV_TYPE_DMV:
938 if (s->picture_structure == PICT_FRAME) {
939 for (i = 0; i < 2; i++) {
940 int j;
941 for (j = 0; j < 2; j++)
942 mpeg_motion_field(s, dest_y, dest_cb, dest_cr,
943 j, j ^ i, ref_picture, pix_op,
944 s->mv[dir][2 * i + j][0],
945 s->mv[dir][2 * i + j][1], 8, mb_y);
946 pix_op = s->hdsp.avg_pixels_tab;
947 }
948 } else {
949 if (!ref_picture[0]) {
950 ref_picture = s->current_picture_ptr->f->data;
951 }
952 for (i = 0; i < 2; i++) {
953 mpeg_motion(s, dest_y, dest_cb, dest_cr,
954 s->picture_structure != i + 1,
955 ref_picture, pix_op,
956 s->mv[dir][2 * i][0], s->mv[dir][2 * i][1],
957 16, 0, mb_y >> 1);
958
959 // after put we make avg of the same block
960 pix_op = s->hdsp.avg_pixels_tab;
961
962 /* opposite parity is always in the same frame if this is
963 * second field */
964 if (!s->first_field) {
965 ref_picture = s->current_picture_ptr->f->data;
966 }
967 }
968 }
969 break;
970 default: av_assert2(0);
971 }
972 }
973
ff_mpv_motion(MpegEncContext * s,uint8_t * dest_y,uint8_t * dest_cb,uint8_t * dest_cr,int dir,uint8_t ** ref_picture,op_pixels_func (* pix_op)[4],qpel_mc_func (* qpix_op)[16])974 void ff_mpv_motion(MpegEncContext *s,
975 uint8_t *dest_y, uint8_t *dest_cb,
976 uint8_t *dest_cr, int dir,
977 uint8_t **ref_picture,
978 op_pixels_func (*pix_op)[4],
979 qpel_mc_func (*qpix_op)[16])
980 {
981 #if !CONFIG_SMALL
982 if (s->out_format == FMT_MPEG1)
983 mpv_motion_internal(s, dest_y, dest_cb, dest_cr, dir,
984 ref_picture, pix_op, qpix_op, 1);
985 else
986 #endif
987 mpv_motion_internal(s, dest_y, dest_cb, dest_cr, dir,
988 ref_picture, pix_op, qpix_op, 0);
989 }
990