1/*
2 *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
3 *
4 *  Use of this source code is governed by a BSD-style license
5 *  that can be found in the LICENSE file in the root of the source
6 *  tree. An additional intellectual property rights grant can be found
7 *  in the file PATENTS.  All contributing project authors may
8 *  be found in the AUTHORS file in the root of the source tree.
9 */
10
11
12#include <limits.h>
13#include "vpx_config.h"
14#include "vp8_rtcd.h"
15#include "vpx/vpx_integer.h"
16#include "blockd.h"
17#include "reconinter.h"
18#if CONFIG_RUNTIME_CPU_DETECT
19#include "onyxc_int.h"
20#endif
21
22void vp8_copy_mem16x16_c(
23    unsigned char *src,
24    int src_stride,
25    unsigned char *dst,
26    int dst_stride)
27{
28
29    int r;
30
31    for (r = 0; r < 16; r++)
32    {
33#if !(CONFIG_FAST_UNALIGNED)
34        dst[0] = src[0];
35        dst[1] = src[1];
36        dst[2] = src[2];
37        dst[3] = src[3];
38        dst[4] = src[4];
39        dst[5] = src[5];
40        dst[6] = src[6];
41        dst[7] = src[7];
42        dst[8] = src[8];
43        dst[9] = src[9];
44        dst[10] = src[10];
45        dst[11] = src[11];
46        dst[12] = src[12];
47        dst[13] = src[13];
48        dst[14] = src[14];
49        dst[15] = src[15];
50
51#else
52        ((uint32_t *)dst)[0] = ((uint32_t *)src)[0] ;
53        ((uint32_t *)dst)[1] = ((uint32_t *)src)[1] ;
54        ((uint32_t *)dst)[2] = ((uint32_t *)src)[2] ;
55        ((uint32_t *)dst)[3] = ((uint32_t *)src)[3] ;
56
57#endif
58        src += src_stride;
59        dst += dst_stride;
60
61    }
62
63}
64
65void vp8_copy_mem8x8_c(
66    unsigned char *src,
67    int src_stride,
68    unsigned char *dst,
69    int dst_stride)
70{
71    int r;
72
73    for (r = 0; r < 8; r++)
74    {
75#if !(CONFIG_FAST_UNALIGNED)
76        dst[0] = src[0];
77        dst[1] = src[1];
78        dst[2] = src[2];
79        dst[3] = src[3];
80        dst[4] = src[4];
81        dst[5] = src[5];
82        dst[6] = src[6];
83        dst[7] = src[7];
84#else
85        ((uint32_t *)dst)[0] = ((uint32_t *)src)[0] ;
86        ((uint32_t *)dst)[1] = ((uint32_t *)src)[1] ;
87#endif
88        src += src_stride;
89        dst += dst_stride;
90
91    }
92
93}
94
95void vp8_copy_mem8x4_c(
96    unsigned char *src,
97    int src_stride,
98    unsigned char *dst,
99    int dst_stride)
100{
101    int r;
102
103    for (r = 0; r < 4; r++)
104    {
105#if !(CONFIG_FAST_UNALIGNED)
106        dst[0] = src[0];
107        dst[1] = src[1];
108        dst[2] = src[2];
109        dst[3] = src[3];
110        dst[4] = src[4];
111        dst[5] = src[5];
112        dst[6] = src[6];
113        dst[7] = src[7];
114#else
115        ((uint32_t *)dst)[0] = ((uint32_t *)src)[0] ;
116        ((uint32_t *)dst)[1] = ((uint32_t *)src)[1] ;
117#endif
118        src += src_stride;
119        dst += dst_stride;
120
121    }
122
123}
124
125
126void vp8_build_inter_predictors_b(BLOCKD *d, int pitch, unsigned char *base_pre, int pre_stride, vp8_subpix_fn_t sppf)
127{
128    int r;
129    unsigned char *pred_ptr = d->predictor;
130    unsigned char *ptr;
131    ptr = base_pre + d->offset + (d->bmi.mv.as_mv.row >> 3) * pre_stride + (d->bmi.mv.as_mv.col >> 3);
132
133    if (d->bmi.mv.as_mv.row & 7 || d->bmi.mv.as_mv.col & 7)
134    {
135        sppf(ptr, pre_stride, d->bmi.mv.as_mv.col & 7, d->bmi.mv.as_mv.row & 7, pred_ptr, pitch);
136    }
137    else
138    {
139        for (r = 0; r < 4; r++)
140        {
141            pred_ptr[0]  = ptr[0];
142            pred_ptr[1]  = ptr[1];
143            pred_ptr[2]  = ptr[2];
144            pred_ptr[3]  = ptr[3];
145            pred_ptr     += pitch;
146            ptr         += pre_stride;
147        }
148    }
149}
150
151static void build_inter_predictors4b(MACROBLOCKD *x, BLOCKD *d, unsigned char *dst, int dst_stride, unsigned char *base_pre, int pre_stride)
152{
153    unsigned char *ptr;
154    ptr = base_pre + d->offset + (d->bmi.mv.as_mv.row >> 3) * pre_stride + (d->bmi.mv.as_mv.col >> 3);
155
156    if (d->bmi.mv.as_mv.row & 7 || d->bmi.mv.as_mv.col & 7)
157    {
158        x->subpixel_predict8x8(ptr, pre_stride, d->bmi.mv.as_mv.col & 7, d->bmi.mv.as_mv.row & 7, dst, dst_stride);
159    }
160    else
161    {
162        vp8_copy_mem8x8(ptr, pre_stride, dst, dst_stride);
163    }
164}
165
166static void build_inter_predictors2b(MACROBLOCKD *x, BLOCKD *d, unsigned char *dst, int dst_stride, unsigned char *base_pre, int pre_stride)
167{
168    unsigned char *ptr;
169    ptr = base_pre + d->offset + (d->bmi.mv.as_mv.row >> 3) * pre_stride + (d->bmi.mv.as_mv.col >> 3);
170
171    if (d->bmi.mv.as_mv.row & 7 || d->bmi.mv.as_mv.col & 7)
172    {
173        x->subpixel_predict8x4(ptr, pre_stride, d->bmi.mv.as_mv.col & 7, d->bmi.mv.as_mv.row & 7, dst, dst_stride);
174    }
175    else
176    {
177        vp8_copy_mem8x4(ptr, pre_stride, dst, dst_stride);
178    }
179}
180
181static void build_inter_predictors_b(BLOCKD *d, unsigned char *dst, int dst_stride, unsigned char *base_pre, int pre_stride, vp8_subpix_fn_t sppf)
182{
183    int r;
184    unsigned char *ptr;
185    ptr = base_pre + d->offset + (d->bmi.mv.as_mv.row >> 3) * pre_stride + (d->bmi.mv.as_mv.col >> 3);
186
187    if (d->bmi.mv.as_mv.row & 7 || d->bmi.mv.as_mv.col & 7)
188    {
189        sppf(ptr, pre_stride, d->bmi.mv.as_mv.col & 7, d->bmi.mv.as_mv.row & 7, dst, dst_stride);
190    }
191    else
192    {
193        for (r = 0; r < 4; r++)
194        {
195          dst[0]  = ptr[0];
196          dst[1]  = ptr[1];
197          dst[2]  = ptr[2];
198          dst[3]  = ptr[3];
199          dst     += dst_stride;
200          ptr     += pre_stride;
201        }
202    }
203}
204
205
206/*encoder only*/
207void vp8_build_inter16x16_predictors_mbuv(MACROBLOCKD *x)
208{
209    unsigned char *uptr, *vptr;
210    unsigned char *upred_ptr = &x->predictor[256];
211    unsigned char *vpred_ptr = &x->predictor[320];
212
213    int mv_row = x->mode_info_context->mbmi.mv.as_mv.row;
214    int mv_col = x->mode_info_context->mbmi.mv.as_mv.col;
215    int offset;
216    int pre_stride = x->pre.uv_stride;
217
218    /* calc uv motion vectors */
219    mv_row += 1 | (mv_row >> (sizeof(int) * CHAR_BIT - 1));
220    mv_col += 1 | (mv_col >> (sizeof(int) * CHAR_BIT - 1));
221    mv_row /= 2;
222    mv_col /= 2;
223    mv_row &= x->fullpixel_mask;
224    mv_col &= x->fullpixel_mask;
225
226    offset = (mv_row >> 3) * pre_stride + (mv_col >> 3);
227    uptr = x->pre.u_buffer + offset;
228    vptr = x->pre.v_buffer + offset;
229
230    if ((mv_row | mv_col) & 7)
231    {
232        x->subpixel_predict8x8(uptr, pre_stride, mv_col & 7, mv_row & 7, upred_ptr, 8);
233        x->subpixel_predict8x8(vptr, pre_stride, mv_col & 7, mv_row & 7, vpred_ptr, 8);
234    }
235    else
236    {
237        vp8_copy_mem8x8(uptr, pre_stride, upred_ptr, 8);
238        vp8_copy_mem8x8(vptr, pre_stride, vpred_ptr, 8);
239    }
240}
241
242/*encoder only*/
243void vp8_build_inter4x4_predictors_mbuv(MACROBLOCKD *x)
244{
245    int i, j;
246    int pre_stride = x->pre.uv_stride;
247    unsigned char *base_pre;
248
249    /* build uv mvs */
250    for (i = 0; i < 2; i++)
251    {
252        for (j = 0; j < 2; j++)
253        {
254            int yoffset = i * 8 + j * 2;
255            int uoffset = 16 + i * 2 + j;
256            int voffset = 20 + i * 2 + j;
257
258            int temp;
259
260            temp = x->block[yoffset  ].bmi.mv.as_mv.row
261                   + x->block[yoffset+1].bmi.mv.as_mv.row
262                   + x->block[yoffset+4].bmi.mv.as_mv.row
263                   + x->block[yoffset+5].bmi.mv.as_mv.row;
264
265            temp += 4 + ((temp >> (sizeof(temp) * CHAR_BIT - 1)) * 8);
266
267            x->block[uoffset].bmi.mv.as_mv.row = (temp / 8) & x->fullpixel_mask;
268
269            temp = x->block[yoffset  ].bmi.mv.as_mv.col
270                   + x->block[yoffset+1].bmi.mv.as_mv.col
271                   + x->block[yoffset+4].bmi.mv.as_mv.col
272                   + x->block[yoffset+5].bmi.mv.as_mv.col;
273
274            temp += 4 + ((temp >> (sizeof(temp) * CHAR_BIT - 1)) * 8);
275
276            x->block[uoffset].bmi.mv.as_mv.col = (temp / 8) & x->fullpixel_mask;
277
278            x->block[voffset].bmi.mv.as_int = x->block[uoffset].bmi.mv.as_int;
279        }
280    }
281
282    base_pre = x->pre.u_buffer;
283    for (i = 16; i < 20; i += 2)
284    {
285        BLOCKD *d0 = &x->block[i];
286        BLOCKD *d1 = &x->block[i+1];
287
288        if (d0->bmi.mv.as_int == d1->bmi.mv.as_int)
289            build_inter_predictors2b(x, d0, d0->predictor, 8, base_pre, pre_stride);
290        else
291        {
292            vp8_build_inter_predictors_b(d0, 8, base_pre, pre_stride, x->subpixel_predict);
293            vp8_build_inter_predictors_b(d1, 8, base_pre, pre_stride, x->subpixel_predict);
294        }
295    }
296
297    base_pre = x->pre.v_buffer;
298    for (i = 20; i < 24; i += 2)
299    {
300        BLOCKD *d0 = &x->block[i];
301        BLOCKD *d1 = &x->block[i+1];
302
303        if (d0->bmi.mv.as_int == d1->bmi.mv.as_int)
304            build_inter_predictors2b(x, d0, d0->predictor, 8, base_pre, pre_stride);
305        else
306        {
307            vp8_build_inter_predictors_b(d0, 8, base_pre, pre_stride, x->subpixel_predict);
308            vp8_build_inter_predictors_b(d1, 8, base_pre, pre_stride, x->subpixel_predict);
309        }
310    }
311}
312
313
314/*encoder only*/
315void vp8_build_inter16x16_predictors_mby(MACROBLOCKD *x,
316                                         unsigned char *dst_y,
317                                         int dst_ystride)
318{
319    unsigned char *ptr_base;
320    unsigned char *ptr;
321    int mv_row = x->mode_info_context->mbmi.mv.as_mv.row;
322    int mv_col = x->mode_info_context->mbmi.mv.as_mv.col;
323    int pre_stride = x->pre.y_stride;
324
325    ptr_base = x->pre.y_buffer;
326    ptr = ptr_base + (mv_row >> 3) * pre_stride + (mv_col >> 3);
327
328    if ((mv_row | mv_col) & 7)
329    {
330        x->subpixel_predict16x16(ptr, pre_stride, mv_col & 7, mv_row & 7,
331                                 dst_y, dst_ystride);
332    }
333    else
334    {
335        vp8_copy_mem16x16(ptr, pre_stride, dst_y,
336            dst_ystride);
337    }
338}
339
340static void clamp_mv_to_umv_border(MV *mv, const MACROBLOCKD *xd)
341{
342    /* If the MV points so far into the UMV border that no visible pixels
343     * are used for reconstruction, the subpel part of the MV can be
344     * discarded and the MV limited to 16 pixels with equivalent results.
345     *
346     * This limit kicks in at 19 pixels for the top and left edges, for
347     * the 16 pixels plus 3 taps right of the central pixel when subpel
348     * filtering. The bottom and right edges use 16 pixels plus 2 pixels
349     * left of the central pixel when filtering.
350     */
351    if (mv->col < (xd->mb_to_left_edge - (19 << 3)))
352        mv->col = xd->mb_to_left_edge - (16 << 3);
353    else if (mv->col > xd->mb_to_right_edge + (18 << 3))
354        mv->col = xd->mb_to_right_edge + (16 << 3);
355
356    if (mv->row < (xd->mb_to_top_edge - (19 << 3)))
357        mv->row = xd->mb_to_top_edge - (16 << 3);
358    else if (mv->row > xd->mb_to_bottom_edge + (18 << 3))
359        mv->row = xd->mb_to_bottom_edge + (16 << 3);
360}
361
362/* A version of the above function for chroma block MVs.*/
363static void clamp_uvmv_to_umv_border(MV *mv, const MACROBLOCKD *xd)
364{
365    mv->col = (2*mv->col < (xd->mb_to_left_edge - (19 << 3))) ?
366        (xd->mb_to_left_edge - (16 << 3)) >> 1 : mv->col;
367    mv->col = (2*mv->col > xd->mb_to_right_edge + (18 << 3)) ?
368        (xd->mb_to_right_edge + (16 << 3)) >> 1 : mv->col;
369
370    mv->row = (2*mv->row < (xd->mb_to_top_edge - (19 << 3))) ?
371        (xd->mb_to_top_edge - (16 << 3)) >> 1 : mv->row;
372    mv->row = (2*mv->row > xd->mb_to_bottom_edge + (18 << 3)) ?
373        (xd->mb_to_bottom_edge + (16 << 3)) >> 1 : mv->row;
374}
375
376void vp8_build_inter16x16_predictors_mb(MACROBLOCKD *x,
377                                        unsigned char *dst_y,
378                                        unsigned char *dst_u,
379                                        unsigned char *dst_v,
380                                        int dst_ystride,
381                                        int dst_uvstride)
382{
383    int offset;
384    unsigned char *ptr;
385    unsigned char *uptr, *vptr;
386
387    int_mv _16x16mv;
388
389    unsigned char *ptr_base = x->pre.y_buffer;
390    int pre_stride = x->pre.y_stride;
391
392    _16x16mv.as_int = x->mode_info_context->mbmi.mv.as_int;
393
394    if (x->mode_info_context->mbmi.need_to_clamp_mvs)
395    {
396        clamp_mv_to_umv_border(&_16x16mv.as_mv, x);
397    }
398
399    ptr = ptr_base + ( _16x16mv.as_mv.row >> 3) * pre_stride + (_16x16mv.as_mv.col >> 3);
400
401    if ( _16x16mv.as_int & 0x00070007)
402    {
403        x->subpixel_predict16x16(ptr, pre_stride, _16x16mv.as_mv.col & 7,  _16x16mv.as_mv.row & 7, dst_y, dst_ystride);
404    }
405    else
406    {
407        vp8_copy_mem16x16(ptr, pre_stride, dst_y, dst_ystride);
408    }
409
410    /* calc uv motion vectors */
411    _16x16mv.as_mv.row += 1 | (_16x16mv.as_mv.row >> (sizeof(int) * CHAR_BIT - 1));
412    _16x16mv.as_mv.col += 1 | (_16x16mv.as_mv.col >> (sizeof(int) * CHAR_BIT - 1));
413    _16x16mv.as_mv.row /= 2;
414    _16x16mv.as_mv.col /= 2;
415    _16x16mv.as_mv.row &= x->fullpixel_mask;
416    _16x16mv.as_mv.col &= x->fullpixel_mask;
417
418    pre_stride >>= 1;
419    offset = ( _16x16mv.as_mv.row >> 3) * pre_stride + (_16x16mv.as_mv.col >> 3);
420    uptr = x->pre.u_buffer + offset;
421    vptr = x->pre.v_buffer + offset;
422
423    if ( _16x16mv.as_int & 0x00070007)
424    {
425        x->subpixel_predict8x8(uptr, pre_stride, _16x16mv.as_mv.col & 7,  _16x16mv.as_mv.row & 7, dst_u, dst_uvstride);
426        x->subpixel_predict8x8(vptr, pre_stride, _16x16mv.as_mv.col & 7,  _16x16mv.as_mv.row & 7, dst_v, dst_uvstride);
427    }
428    else
429    {
430        vp8_copy_mem8x8(uptr, pre_stride, dst_u, dst_uvstride);
431        vp8_copy_mem8x8(vptr, pre_stride, dst_v, dst_uvstride);
432    }
433}
434
435static void build_inter4x4_predictors_mb(MACROBLOCKD *x)
436{
437    int i;
438    unsigned char *base_dst = x->dst.y_buffer;
439    unsigned char *base_pre = x->pre.y_buffer;
440
441    if (x->mode_info_context->mbmi.partitioning < 3)
442    {
443        BLOCKD *b;
444        int dst_stride = x->dst.y_stride;
445
446        x->block[ 0].bmi = x->mode_info_context->bmi[ 0];
447        x->block[ 2].bmi = x->mode_info_context->bmi[ 2];
448        x->block[ 8].bmi = x->mode_info_context->bmi[ 8];
449        x->block[10].bmi = x->mode_info_context->bmi[10];
450        if (x->mode_info_context->mbmi.need_to_clamp_mvs)
451        {
452            clamp_mv_to_umv_border(&x->block[ 0].bmi.mv.as_mv, x);
453            clamp_mv_to_umv_border(&x->block[ 2].bmi.mv.as_mv, x);
454            clamp_mv_to_umv_border(&x->block[ 8].bmi.mv.as_mv, x);
455            clamp_mv_to_umv_border(&x->block[10].bmi.mv.as_mv, x);
456        }
457
458        b = &x->block[ 0];
459        build_inter_predictors4b(x, b, base_dst + b->offset, dst_stride, base_pre, dst_stride);
460        b = &x->block[ 2];
461        build_inter_predictors4b(x, b, base_dst + b->offset, dst_stride, base_pre, dst_stride);
462        b = &x->block[ 8];
463        build_inter_predictors4b(x, b, base_dst + b->offset, dst_stride, base_pre, dst_stride);
464        b = &x->block[10];
465        build_inter_predictors4b(x, b, base_dst + b->offset, dst_stride, base_pre, dst_stride);
466    }
467    else
468    {
469        for (i = 0; i < 16; i += 2)
470        {
471            BLOCKD *d0 = &x->block[i];
472            BLOCKD *d1 = &x->block[i+1];
473            int dst_stride = x->dst.y_stride;
474
475            x->block[i+0].bmi = x->mode_info_context->bmi[i+0];
476            x->block[i+1].bmi = x->mode_info_context->bmi[i+1];
477            if (x->mode_info_context->mbmi.need_to_clamp_mvs)
478            {
479                clamp_mv_to_umv_border(&x->block[i+0].bmi.mv.as_mv, x);
480                clamp_mv_to_umv_border(&x->block[i+1].bmi.mv.as_mv, x);
481            }
482
483            if (d0->bmi.mv.as_int == d1->bmi.mv.as_int)
484                build_inter_predictors2b(x, d0, base_dst + d0->offset, dst_stride, base_pre, dst_stride);
485            else
486            {
487                build_inter_predictors_b(d0, base_dst + d0->offset, dst_stride, base_pre, dst_stride, x->subpixel_predict);
488                build_inter_predictors_b(d1, base_dst + d1->offset, dst_stride, base_pre, dst_stride, x->subpixel_predict);
489            }
490
491        }
492
493    }
494    base_dst = x->dst.u_buffer;
495    base_pre = x->pre.u_buffer;
496    for (i = 16; i < 20; i += 2)
497    {
498        BLOCKD *d0 = &x->block[i];
499        BLOCKD *d1 = &x->block[i+1];
500        int dst_stride = x->dst.uv_stride;
501
502        /* Note: uv mvs already clamped in build_4x4uvmvs() */
503
504        if (d0->bmi.mv.as_int == d1->bmi.mv.as_int)
505            build_inter_predictors2b(x, d0, base_dst + d0->offset, dst_stride, base_pre, dst_stride);
506        else
507        {
508            build_inter_predictors_b(d0, base_dst + d0->offset, dst_stride, base_pre, dst_stride, x->subpixel_predict);
509            build_inter_predictors_b(d1, base_dst + d1->offset, dst_stride, base_pre, dst_stride, x->subpixel_predict);
510        }
511    }
512
513    base_dst = x->dst.v_buffer;
514    base_pre = x->pre.v_buffer;
515    for (i = 20; i < 24; i += 2)
516    {
517        BLOCKD *d0 = &x->block[i];
518        BLOCKD *d1 = &x->block[i+1];
519        int dst_stride = x->dst.uv_stride;
520
521        /* Note: uv mvs already clamped in build_4x4uvmvs() */
522
523        if (d0->bmi.mv.as_int == d1->bmi.mv.as_int)
524            build_inter_predictors2b(x, d0, base_dst + d0->offset, dst_stride, base_pre, dst_stride);
525        else
526        {
527            build_inter_predictors_b(d0, base_dst + d0->offset, dst_stride, base_pre, dst_stride, x->subpixel_predict);
528            build_inter_predictors_b(d1, base_dst + d1->offset, dst_stride, base_pre, dst_stride, x->subpixel_predict);
529        }
530    }
531}
532
533static
534void build_4x4uvmvs(MACROBLOCKD *x)
535{
536    int i, j;
537
538    for (i = 0; i < 2; i++)
539    {
540        for (j = 0; j < 2; j++)
541        {
542            int yoffset = i * 8 + j * 2;
543            int uoffset = 16 + i * 2 + j;
544            int voffset = 20 + i * 2 + j;
545
546            int temp;
547
548            temp = x->mode_info_context->bmi[yoffset + 0].mv.as_mv.row
549                 + x->mode_info_context->bmi[yoffset + 1].mv.as_mv.row
550                 + x->mode_info_context->bmi[yoffset + 4].mv.as_mv.row
551                 + x->mode_info_context->bmi[yoffset + 5].mv.as_mv.row;
552
553            temp += 4 + ((temp >> (sizeof(temp) * CHAR_BIT - 1)) * 8);
554
555            x->block[uoffset].bmi.mv.as_mv.row = (temp / 8) & x->fullpixel_mask;
556
557            temp = x->mode_info_context->bmi[yoffset + 0].mv.as_mv.col
558                 + x->mode_info_context->bmi[yoffset + 1].mv.as_mv.col
559                 + x->mode_info_context->bmi[yoffset + 4].mv.as_mv.col
560                 + x->mode_info_context->bmi[yoffset + 5].mv.as_mv.col;
561
562            temp += 4 + ((temp >> (sizeof(temp) * CHAR_BIT - 1)) * 8);
563
564            x->block[uoffset].bmi.mv.as_mv.col = (temp / 8) & x->fullpixel_mask;
565
566            if (x->mode_info_context->mbmi.need_to_clamp_mvs)
567                clamp_uvmv_to_umv_border(&x->block[uoffset].bmi.mv.as_mv, x);
568
569            x->block[voffset].bmi.mv.as_int = x->block[uoffset].bmi.mv.as_int;
570        }
571    }
572}
573
574void vp8_build_inter_predictors_mb(MACROBLOCKD *xd)
575{
576    if (xd->mode_info_context->mbmi.mode != SPLITMV)
577    {
578        vp8_build_inter16x16_predictors_mb(xd, xd->dst.y_buffer,
579                                           xd->dst.u_buffer, xd->dst.v_buffer,
580                                           xd->dst.y_stride, xd->dst.uv_stride);
581    }
582    else
583    {
584        build_4x4uvmvs(xd);
585        build_inter4x4_predictors_mb(xd);
586    }
587}
588