michael@0: /*
michael@0:  *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
michael@0:  *
michael@0:  *  Use of this source code is governed by a BSD-style license
michael@0:  *  that can be found in the LICENSE file in the root of the source
michael@0:  *  tree. An additional intellectual property rights grant can be found
michael@0:  *  in the file PATENTS.  All contributing project authors may
michael@0:  *  be found in the AUTHORS file in the root of the source tree.
michael@0:  */
michael@0: 
michael@0: 
michael@0: #include <limits.h>
michael@0: #include "vpx_config.h"
michael@0: #include "vp8_rtcd.h"
michael@0: #include "vpx/vpx_integer.h"
michael@0: #include "blockd.h"
michael@0: #include "reconinter.h"
michael@0: #if CONFIG_RUNTIME_CPU_DETECT
michael@0: #include "onyxc_int.h"
michael@0: #endif
michael@0: 
michael@0: void vp8_copy_mem16x16_c(
michael@0:     unsigned char *src,
michael@0:     int src_stride,
michael@0:     unsigned char *dst,
michael@0:     int dst_stride)
michael@0: {
michael@0: 
michael@0:     int r;
michael@0: 
michael@0:     for (r = 0; r < 16; r++)
michael@0:     {
michael@0: #if !(CONFIG_FAST_UNALIGNED)
michael@0:         dst[0] = src[0];
michael@0:         dst[1] = src[1];
michael@0:         dst[2] = src[2];
michael@0:         dst[3] = src[3];
michael@0:         dst[4] = src[4];
michael@0:         dst[5] = src[5];
michael@0:         dst[6] = src[6];
michael@0:         dst[7] = src[7];
michael@0:         dst[8] = src[8];
michael@0:         dst[9] = src[9];
michael@0:         dst[10] = src[10];
michael@0:         dst[11] = src[11];
michael@0:         dst[12] = src[12];
michael@0:         dst[13] = src[13];
michael@0:         dst[14] = src[14];
michael@0:         dst[15] = src[15];
michael@0: 
michael@0: #else
michael@0:         ((uint32_t *)dst)[0] = ((uint32_t *)src)[0] ;
michael@0:         ((uint32_t *)dst)[1] = ((uint32_t *)src)[1] ;
michael@0:         ((uint32_t *)dst)[2] = ((uint32_t *)src)[2] ;
michael@0:         ((uint32_t *)dst)[3] = ((uint32_t *)src)[3] ;
michael@0: 
michael@0: #endif
michael@0:         src += src_stride;
michael@0:         dst += dst_stride;
michael@0: 
michael@0:     }
michael@0: 
michael@0: }
michael@0: 
michael@0: void vp8_copy_mem8x8_c(
michael@0:     unsigned char *src,
michael@0:     int src_stride,
michael@0:     unsigned char *dst,
michael@0:     int dst_stride)
michael@0: {
michael@0:     int r;
michael@0: 
michael@0:     for (r = 0; r < 8; r++)
michael@0:     {
michael@0: #if !(CONFIG_FAST_UNALIGNED)
michael@0:         dst[0] = src[0];
michael@0:         dst[1] = src[1];
michael@0:         dst[2] = src[2];
michael@0:         dst[3] = src[3];
michael@0:         dst[4] = src[4];
michael@0:         dst[5] = src[5];
michael@0:         dst[6] = src[6];
michael@0:         dst[7] = src[7];
michael@0: #else
michael@0:         ((uint32_t *)dst)[0] = ((uint32_t *)src)[0] ;
michael@0:         ((uint32_t *)dst)[1] = ((uint32_t *)src)[1] ;
michael@0: #endif
michael@0:         src += src_stride;
michael@0:         dst += dst_stride;
michael@0: 
michael@0:     }
michael@0: 
michael@0: }
michael@0: 
michael@0: void vp8_copy_mem8x4_c(
michael@0:     unsigned char *src,
michael@0:     int src_stride,
michael@0:     unsigned char *dst,
michael@0:     int dst_stride)
michael@0: {
michael@0:     int r;
michael@0: 
michael@0:     for (r = 0; r < 4; r++)
michael@0:     {
michael@0: #if !(CONFIG_FAST_UNALIGNED)
michael@0:         dst[0] = src[0];
michael@0:         dst[1] = src[1];
michael@0:         dst[2] = src[2];
michael@0:         dst[3] = src[3];
michael@0:         dst[4] = src[4];
michael@0:         dst[5] = src[5];
michael@0:         dst[6] = src[6];
michael@0:         dst[7] = src[7];
michael@0: #else
michael@0:         ((uint32_t *)dst)[0] = ((uint32_t *)src)[0] ;
michael@0:         ((uint32_t *)dst)[1] = ((uint32_t *)src)[1] ;
michael@0: #endif
michael@0:         src += src_stride;
michael@0:         dst += dst_stride;
michael@0: 
michael@0:     }
michael@0: 
michael@0: }
michael@0: 
michael@0: 
michael@0: void vp8_build_inter_predictors_b(BLOCKD *d, int pitch, unsigned char *base_pre, int pre_stride, vp8_subpix_fn_t sppf)
michael@0: {
michael@0:     int r;
michael@0:     unsigned char *pred_ptr = d->predictor;
michael@0:     unsigned char *ptr;
michael@0:     ptr = base_pre + d->offset + (d->bmi.mv.as_mv.row >> 3) * pre_stride + (d->bmi.mv.as_mv.col >> 3);
michael@0: 
michael@0:     if (d->bmi.mv.as_mv.row & 7 || d->bmi.mv.as_mv.col & 7)
michael@0:     {
michael@0:         sppf(ptr, pre_stride, d->bmi.mv.as_mv.col & 7, d->bmi.mv.as_mv.row & 7, pred_ptr, pitch);
michael@0:     }
michael@0:     else
michael@0:     {
michael@0:         for (r = 0; r < 4; r++)
michael@0:         {
michael@0:             pred_ptr[0]  = ptr[0];
michael@0:             pred_ptr[1]  = ptr[1];
michael@0:             pred_ptr[2]  = ptr[2];
michael@0:             pred_ptr[3]  = ptr[3];
michael@0:             pred_ptr     += pitch;
michael@0:             ptr         += pre_stride;
michael@0:         }
michael@0:     }
michael@0: }
michael@0: 
michael@0: static void build_inter_predictors4b(MACROBLOCKD *x, BLOCKD *d, unsigned char *dst, int dst_stride, unsigned char *base_pre, int pre_stride)
michael@0: {
michael@0:     unsigned char *ptr;
michael@0:     ptr = base_pre + d->offset + (d->bmi.mv.as_mv.row >> 3) * pre_stride + (d->bmi.mv.as_mv.col >> 3);
michael@0: 
michael@0:     if (d->bmi.mv.as_mv.row & 7 || d->bmi.mv.as_mv.col & 7)
michael@0:     {
michael@0:         x->subpixel_predict8x8(ptr, pre_stride, d->bmi.mv.as_mv.col & 7, d->bmi.mv.as_mv.row & 7, dst, dst_stride);
michael@0:     }
michael@0:     else
michael@0:     {
michael@0:         vp8_copy_mem8x8(ptr, pre_stride, dst, dst_stride);
michael@0:     }
michael@0: }
michael@0: 
michael@0: static void build_inter_predictors2b(MACROBLOCKD *x, BLOCKD *d, unsigned char *dst, int dst_stride, unsigned char *base_pre, int pre_stride)
michael@0: {
michael@0:     unsigned char *ptr;
michael@0:     ptr = base_pre + d->offset + (d->bmi.mv.as_mv.row >> 3) * pre_stride + (d->bmi.mv.as_mv.col >> 3);
michael@0: 
michael@0:     if (d->bmi.mv.as_mv.row & 7 || d->bmi.mv.as_mv.col & 7)
michael@0:     {
michael@0:         x->subpixel_predict8x4(ptr, pre_stride, d->bmi.mv.as_mv.col & 7, d->bmi.mv.as_mv.row & 7, dst, dst_stride);
michael@0:     }
michael@0:     else
michael@0:     {
michael@0:         vp8_copy_mem8x4(ptr, pre_stride, dst, dst_stride);
michael@0:     }
michael@0: }
michael@0: 
michael@0: static 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)
michael@0: {
michael@0:     int r;
michael@0:     unsigned char *ptr;
michael@0:     ptr = base_pre + d->offset + (d->bmi.mv.as_mv.row >> 3) * pre_stride + (d->bmi.mv.as_mv.col >> 3);
michael@0: 
michael@0:     if (d->bmi.mv.as_mv.row & 7 || d->bmi.mv.as_mv.col & 7)
michael@0:     {
michael@0:         sppf(ptr, pre_stride, d->bmi.mv.as_mv.col & 7, d->bmi.mv.as_mv.row & 7, dst, dst_stride);
michael@0:     }
michael@0:     else
michael@0:     {
michael@0:         for (r = 0; r < 4; r++)
michael@0:         {
michael@0:           dst[0]  = ptr[0];
michael@0:           dst[1]  = ptr[1];
michael@0:           dst[2]  = ptr[2];
michael@0:           dst[3]  = ptr[3];
michael@0:           dst     += dst_stride;
michael@0:           ptr     += pre_stride;
michael@0:         }
michael@0:     }
michael@0: }
michael@0: 
michael@0: 
michael@0: /*encoder only*/
michael@0: void vp8_build_inter16x16_predictors_mbuv(MACROBLOCKD *x)
michael@0: {
michael@0:     unsigned char *uptr, *vptr;
michael@0:     unsigned char *upred_ptr = &x->predictor[256];
michael@0:     unsigned char *vpred_ptr = &x->predictor[320];
michael@0: 
michael@0:     int mv_row = x->mode_info_context->mbmi.mv.as_mv.row;
michael@0:     int mv_col = x->mode_info_context->mbmi.mv.as_mv.col;
michael@0:     int offset;
michael@0:     int pre_stride = x->pre.uv_stride;
michael@0: 
michael@0:     /* calc uv motion vectors */
michael@0:     mv_row += 1 | (mv_row >> (sizeof(int) * CHAR_BIT - 1));
michael@0:     mv_col += 1 | (mv_col >> (sizeof(int) * CHAR_BIT - 1));
michael@0:     mv_row /= 2;
michael@0:     mv_col /= 2;
michael@0:     mv_row &= x->fullpixel_mask;
michael@0:     mv_col &= x->fullpixel_mask;
michael@0: 
michael@0:     offset = (mv_row >> 3) * pre_stride + (mv_col >> 3);
michael@0:     uptr = x->pre.u_buffer + offset;
michael@0:     vptr = x->pre.v_buffer + offset;
michael@0: 
michael@0:     if ((mv_row | mv_col) & 7)
michael@0:     {
michael@0:         x->subpixel_predict8x8(uptr, pre_stride, mv_col & 7, mv_row & 7, upred_ptr, 8);
michael@0:         x->subpixel_predict8x8(vptr, pre_stride, mv_col & 7, mv_row & 7, vpred_ptr, 8);
michael@0:     }
michael@0:     else
michael@0:     {
michael@0:         vp8_copy_mem8x8(uptr, pre_stride, upred_ptr, 8);
michael@0:         vp8_copy_mem8x8(vptr, pre_stride, vpred_ptr, 8);
michael@0:     }
michael@0: }
michael@0: 
michael@0: /*encoder only*/
michael@0: void vp8_build_inter4x4_predictors_mbuv(MACROBLOCKD *x)
michael@0: {
michael@0:     int i, j;
michael@0:     int pre_stride = x->pre.uv_stride;
michael@0:     unsigned char *base_pre;
michael@0: 
michael@0:     /* build uv mvs */
michael@0:     for (i = 0; i < 2; i++)
michael@0:     {
michael@0:         for (j = 0; j < 2; j++)
michael@0:         {
michael@0:             int yoffset = i * 8 + j * 2;
michael@0:             int uoffset = 16 + i * 2 + j;
michael@0:             int voffset = 20 + i * 2 + j;
michael@0: 
michael@0:             int temp;
michael@0: 
michael@0:             temp = x->block[yoffset  ].bmi.mv.as_mv.row
michael@0:                    + x->block[yoffset+1].bmi.mv.as_mv.row
michael@0:                    + x->block[yoffset+4].bmi.mv.as_mv.row
michael@0:                    + x->block[yoffset+5].bmi.mv.as_mv.row;
michael@0: 
michael@0:             temp += 4 + ((temp >> (sizeof(temp) * CHAR_BIT - 1)) * 8);
michael@0: 
michael@0:             x->block[uoffset].bmi.mv.as_mv.row = (temp / 8) & x->fullpixel_mask;
michael@0: 
michael@0:             temp = x->block[yoffset  ].bmi.mv.as_mv.col
michael@0:                    + x->block[yoffset+1].bmi.mv.as_mv.col
michael@0:                    + x->block[yoffset+4].bmi.mv.as_mv.col
michael@0:                    + x->block[yoffset+5].bmi.mv.as_mv.col;
michael@0: 
michael@0:             temp += 4 + ((temp >> (sizeof(temp) * CHAR_BIT - 1)) * 8);
michael@0: 
michael@0:             x->block[uoffset].bmi.mv.as_mv.col = (temp / 8) & x->fullpixel_mask;
michael@0: 
michael@0:             x->block[voffset].bmi.mv.as_int = x->block[uoffset].bmi.mv.as_int;
michael@0:         }
michael@0:     }
michael@0: 
michael@0:     base_pre = x->pre.u_buffer;
michael@0:     for (i = 16; i < 20; i += 2)
michael@0:     {
michael@0:         BLOCKD *d0 = &x->block[i];
michael@0:         BLOCKD *d1 = &x->block[i+1];
michael@0: 
michael@0:         if (d0->bmi.mv.as_int == d1->bmi.mv.as_int)
michael@0:             build_inter_predictors2b(x, d0, d0->predictor, 8, base_pre, pre_stride);
michael@0:         else
michael@0:         {
michael@0:             vp8_build_inter_predictors_b(d0, 8, base_pre, pre_stride, x->subpixel_predict);
michael@0:             vp8_build_inter_predictors_b(d1, 8, base_pre, pre_stride, x->subpixel_predict);
michael@0:         }
michael@0:     }
michael@0: 
michael@0:     base_pre = x->pre.v_buffer;
michael@0:     for (i = 20; i < 24; i += 2)
michael@0:     {
michael@0:         BLOCKD *d0 = &x->block[i];
michael@0:         BLOCKD *d1 = &x->block[i+1];
michael@0: 
michael@0:         if (d0->bmi.mv.as_int == d1->bmi.mv.as_int)
michael@0:             build_inter_predictors2b(x, d0, d0->predictor, 8, base_pre, pre_stride);
michael@0:         else
michael@0:         {
michael@0:             vp8_build_inter_predictors_b(d0, 8, base_pre, pre_stride, x->subpixel_predict);
michael@0:             vp8_build_inter_predictors_b(d1, 8, base_pre, pre_stride, x->subpixel_predict);
michael@0:         }
michael@0:     }
michael@0: }
michael@0: 
michael@0: 
michael@0: /*encoder only*/
michael@0: void vp8_build_inter16x16_predictors_mby(MACROBLOCKD *x,
michael@0:                                          unsigned char *dst_y,
michael@0:                                          int dst_ystride)
michael@0: {
michael@0:     unsigned char *ptr_base;
michael@0:     unsigned char *ptr;
michael@0:     int mv_row = x->mode_info_context->mbmi.mv.as_mv.row;
michael@0:     int mv_col = x->mode_info_context->mbmi.mv.as_mv.col;
michael@0:     int pre_stride = x->pre.y_stride;
michael@0: 
michael@0:     ptr_base = x->pre.y_buffer;
michael@0:     ptr = ptr_base + (mv_row >> 3) * pre_stride + (mv_col >> 3);
michael@0: 
michael@0:     if ((mv_row | mv_col) & 7)
michael@0:     {
michael@0:         x->subpixel_predict16x16(ptr, pre_stride, mv_col & 7, mv_row & 7,
michael@0:                                  dst_y, dst_ystride);
michael@0:     }
michael@0:     else
michael@0:     {
michael@0:         vp8_copy_mem16x16(ptr, pre_stride, dst_y,
michael@0:             dst_ystride);
michael@0:     }
michael@0: }
michael@0: 
michael@0: static void clamp_mv_to_umv_border(MV *mv, const MACROBLOCKD *xd)
michael@0: {
michael@0:     /* If the MV points so far into the UMV border that no visible pixels
michael@0:      * are used for reconstruction, the subpel part of the MV can be
michael@0:      * discarded and the MV limited to 16 pixels with equivalent results.
michael@0:      *
michael@0:      * This limit kicks in at 19 pixels for the top and left edges, for
michael@0:      * the 16 pixels plus 3 taps right of the central pixel when subpel
michael@0:      * filtering. The bottom and right edges use 16 pixels plus 2 pixels
michael@0:      * left of the central pixel when filtering.
michael@0:      */
michael@0:     if (mv->col < (xd->mb_to_left_edge - (19 << 3)))
michael@0:         mv->col = xd->mb_to_left_edge - (16 << 3);
michael@0:     else if (mv->col > xd->mb_to_right_edge + (18 << 3))
michael@0:         mv->col = xd->mb_to_right_edge + (16 << 3);
michael@0: 
michael@0:     if (mv->row < (xd->mb_to_top_edge - (19 << 3)))
michael@0:         mv->row = xd->mb_to_top_edge - (16 << 3);
michael@0:     else if (mv->row > xd->mb_to_bottom_edge + (18 << 3))
michael@0:         mv->row = xd->mb_to_bottom_edge + (16 << 3);
michael@0: }
michael@0: 
michael@0: /* A version of the above function for chroma block MVs.*/
michael@0: static void clamp_uvmv_to_umv_border(MV *mv, const MACROBLOCKD *xd)
michael@0: {
michael@0:     mv->col = (2*mv->col < (xd->mb_to_left_edge - (19 << 3))) ?
michael@0:         (xd->mb_to_left_edge - (16 << 3)) >> 1 : mv->col;
michael@0:     mv->col = (2*mv->col > xd->mb_to_right_edge + (18 << 3)) ?
michael@0:         (xd->mb_to_right_edge + (16 << 3)) >> 1 : mv->col;
michael@0: 
michael@0:     mv->row = (2*mv->row < (xd->mb_to_top_edge - (19 << 3))) ?
michael@0:         (xd->mb_to_top_edge - (16 << 3)) >> 1 : mv->row;
michael@0:     mv->row = (2*mv->row > xd->mb_to_bottom_edge + (18 << 3)) ?
michael@0:         (xd->mb_to_bottom_edge + (16 << 3)) >> 1 : mv->row;
michael@0: }
michael@0: 
michael@0: void vp8_build_inter16x16_predictors_mb(MACROBLOCKD *x,
michael@0:                                         unsigned char *dst_y,
michael@0:                                         unsigned char *dst_u,
michael@0:                                         unsigned char *dst_v,
michael@0:                                         int dst_ystride,
michael@0:                                         int dst_uvstride)
michael@0: {
michael@0:     int offset;
michael@0:     unsigned char *ptr;
michael@0:     unsigned char *uptr, *vptr;
michael@0: 
michael@0:     int_mv _16x16mv;
michael@0: 
michael@0:     unsigned char *ptr_base = x->pre.y_buffer;
michael@0:     int pre_stride = x->pre.y_stride;
michael@0: 
michael@0:     _16x16mv.as_int = x->mode_info_context->mbmi.mv.as_int;
michael@0: 
michael@0:     if (x->mode_info_context->mbmi.need_to_clamp_mvs)
michael@0:     {
michael@0:         clamp_mv_to_umv_border(&_16x16mv.as_mv, x);
michael@0:     }
michael@0: 
michael@0:     ptr = ptr_base + ( _16x16mv.as_mv.row >> 3) * pre_stride + (_16x16mv.as_mv.col >> 3);
michael@0: 
michael@0:     if ( _16x16mv.as_int & 0x00070007)
michael@0:     {
michael@0:         x->subpixel_predict16x16(ptr, pre_stride, _16x16mv.as_mv.col & 7,  _16x16mv.as_mv.row & 7, dst_y, dst_ystride);
michael@0:     }
michael@0:     else
michael@0:     {
michael@0:         vp8_copy_mem16x16(ptr, pre_stride, dst_y, dst_ystride);
michael@0:     }
michael@0: 
michael@0:     /* calc uv motion vectors */
michael@0:     _16x16mv.as_mv.row += 1 | (_16x16mv.as_mv.row >> (sizeof(int) * CHAR_BIT - 1));
michael@0:     _16x16mv.as_mv.col += 1 | (_16x16mv.as_mv.col >> (sizeof(int) * CHAR_BIT - 1));
michael@0:     _16x16mv.as_mv.row /= 2;
michael@0:     _16x16mv.as_mv.col /= 2;
michael@0:     _16x16mv.as_mv.row &= x->fullpixel_mask;
michael@0:     _16x16mv.as_mv.col &= x->fullpixel_mask;
michael@0: 
michael@0:     pre_stride >>= 1;
michael@0:     offset = ( _16x16mv.as_mv.row >> 3) * pre_stride + (_16x16mv.as_mv.col >> 3);
michael@0:     uptr = x->pre.u_buffer + offset;
michael@0:     vptr = x->pre.v_buffer + offset;
michael@0: 
michael@0:     if ( _16x16mv.as_int & 0x00070007)
michael@0:     {
michael@0:         x->subpixel_predict8x8(uptr, pre_stride, _16x16mv.as_mv.col & 7,  _16x16mv.as_mv.row & 7, dst_u, dst_uvstride);
michael@0:         x->subpixel_predict8x8(vptr, pre_stride, _16x16mv.as_mv.col & 7,  _16x16mv.as_mv.row & 7, dst_v, dst_uvstride);
michael@0:     }
michael@0:     else
michael@0:     {
michael@0:         vp8_copy_mem8x8(uptr, pre_stride, dst_u, dst_uvstride);
michael@0:         vp8_copy_mem8x8(vptr, pre_stride, dst_v, dst_uvstride);
michael@0:     }
michael@0: }
michael@0: 
michael@0: static void build_inter4x4_predictors_mb(MACROBLOCKD *x)
michael@0: {
michael@0:     int i;
michael@0:     unsigned char *base_dst = x->dst.y_buffer;
michael@0:     unsigned char *base_pre = x->pre.y_buffer;
michael@0: 
michael@0:     if (x->mode_info_context->mbmi.partitioning < 3)
michael@0:     {
michael@0:         BLOCKD *b;
michael@0:         int dst_stride = x->dst.y_stride;
michael@0: 
michael@0:         x->block[ 0].bmi = x->mode_info_context->bmi[ 0];
michael@0:         x->block[ 2].bmi = x->mode_info_context->bmi[ 2];
michael@0:         x->block[ 8].bmi = x->mode_info_context->bmi[ 8];
michael@0:         x->block[10].bmi = x->mode_info_context->bmi[10];
michael@0:         if (x->mode_info_context->mbmi.need_to_clamp_mvs)
michael@0:         {
michael@0:             clamp_mv_to_umv_border(&x->block[ 0].bmi.mv.as_mv, x);
michael@0:             clamp_mv_to_umv_border(&x->block[ 2].bmi.mv.as_mv, x);
michael@0:             clamp_mv_to_umv_border(&x->block[ 8].bmi.mv.as_mv, x);
michael@0:             clamp_mv_to_umv_border(&x->block[10].bmi.mv.as_mv, x);
michael@0:         }
michael@0: 
michael@0:         b = &x->block[ 0];
michael@0:         build_inter_predictors4b(x, b, base_dst + b->offset, dst_stride, base_pre, dst_stride);
michael@0:         b = &x->block[ 2];
michael@0:         build_inter_predictors4b(x, b, base_dst + b->offset, dst_stride, base_pre, dst_stride);
michael@0:         b = &x->block[ 8];
michael@0:         build_inter_predictors4b(x, b, base_dst + b->offset, dst_stride, base_pre, dst_stride);
michael@0:         b = &x->block[10];
michael@0:         build_inter_predictors4b(x, b, base_dst + b->offset, dst_stride, base_pre, dst_stride);
michael@0:     }
michael@0:     else
michael@0:     {
michael@0:         for (i = 0; i < 16; i += 2)
michael@0:         {
michael@0:             BLOCKD *d0 = &x->block[i];
michael@0:             BLOCKD *d1 = &x->block[i+1];
michael@0:             int dst_stride = x->dst.y_stride;
michael@0: 
michael@0:             x->block[i+0].bmi = x->mode_info_context->bmi[i+0];
michael@0:             x->block[i+1].bmi = x->mode_info_context->bmi[i+1];
michael@0:             if (x->mode_info_context->mbmi.need_to_clamp_mvs)
michael@0:             {
michael@0:                 clamp_mv_to_umv_border(&x->block[i+0].bmi.mv.as_mv, x);
michael@0:                 clamp_mv_to_umv_border(&x->block[i+1].bmi.mv.as_mv, x);
michael@0:             }
michael@0: 
michael@0:             if (d0->bmi.mv.as_int == d1->bmi.mv.as_int)
michael@0:                 build_inter_predictors2b(x, d0, base_dst + d0->offset, dst_stride, base_pre, dst_stride);
michael@0:             else
michael@0:             {
michael@0:                 build_inter_predictors_b(d0, base_dst + d0->offset, dst_stride, base_pre, dst_stride, x->subpixel_predict);
michael@0:                 build_inter_predictors_b(d1, base_dst + d1->offset, dst_stride, base_pre, dst_stride, x->subpixel_predict);
michael@0:             }
michael@0: 
michael@0:         }
michael@0: 
michael@0:     }
michael@0:     base_dst = x->dst.u_buffer;
michael@0:     base_pre = x->pre.u_buffer;
michael@0:     for (i = 16; i < 20; i += 2)
michael@0:     {
michael@0:         BLOCKD *d0 = &x->block[i];
michael@0:         BLOCKD *d1 = &x->block[i+1];
michael@0:         int dst_stride = x->dst.uv_stride;
michael@0: 
michael@0:         /* Note: uv mvs already clamped in build_4x4uvmvs() */
michael@0: 
michael@0:         if (d0->bmi.mv.as_int == d1->bmi.mv.as_int)
michael@0:             build_inter_predictors2b(x, d0, base_dst + d0->offset, dst_stride, base_pre, dst_stride);
michael@0:         else
michael@0:         {
michael@0:             build_inter_predictors_b(d0, base_dst + d0->offset, dst_stride, base_pre, dst_stride, x->subpixel_predict);
michael@0:             build_inter_predictors_b(d1, base_dst + d1->offset, dst_stride, base_pre, dst_stride, x->subpixel_predict);
michael@0:         }
michael@0:     }
michael@0: 
michael@0:     base_dst = x->dst.v_buffer;
michael@0:     base_pre = x->pre.v_buffer;
michael@0:     for (i = 20; i < 24; i += 2)
michael@0:     {
michael@0:         BLOCKD *d0 = &x->block[i];
michael@0:         BLOCKD *d1 = &x->block[i+1];
michael@0:         int dst_stride = x->dst.uv_stride;
michael@0: 
michael@0:         /* Note: uv mvs already clamped in build_4x4uvmvs() */
michael@0: 
michael@0:         if (d0->bmi.mv.as_int == d1->bmi.mv.as_int)
michael@0:             build_inter_predictors2b(x, d0, base_dst + d0->offset, dst_stride, base_pre, dst_stride);
michael@0:         else
michael@0:         {
michael@0:             build_inter_predictors_b(d0, base_dst + d0->offset, dst_stride, base_pre, dst_stride, x->subpixel_predict);
michael@0:             build_inter_predictors_b(d1, base_dst + d1->offset, dst_stride, base_pre, dst_stride, x->subpixel_predict);
michael@0:         }
michael@0:     }
michael@0: }
michael@0: 
michael@0: static
michael@0: void build_4x4uvmvs(MACROBLOCKD *x)
michael@0: {
michael@0:     int i, j;
michael@0: 
michael@0:     for (i = 0; i < 2; i++)
michael@0:     {
michael@0:         for (j = 0; j < 2; j++)
michael@0:         {
michael@0:             int yoffset = i * 8 + j * 2;
michael@0:             int uoffset = 16 + i * 2 + j;
michael@0:             int voffset = 20 + i * 2 + j;
michael@0: 
michael@0:             int temp;
michael@0: 
michael@0:             temp = x->mode_info_context->bmi[yoffset + 0].mv.as_mv.row
michael@0:                  + x->mode_info_context->bmi[yoffset + 1].mv.as_mv.row
michael@0:                  + x->mode_info_context->bmi[yoffset + 4].mv.as_mv.row
michael@0:                  + x->mode_info_context->bmi[yoffset + 5].mv.as_mv.row;
michael@0: 
michael@0:             temp += 4 + ((temp >> (sizeof(temp) * CHAR_BIT - 1)) * 8);
michael@0: 
michael@0:             x->block[uoffset].bmi.mv.as_mv.row = (temp / 8) & x->fullpixel_mask;
michael@0: 
michael@0:             temp = x->mode_info_context->bmi[yoffset + 0].mv.as_mv.col
michael@0:                  + x->mode_info_context->bmi[yoffset + 1].mv.as_mv.col
michael@0:                  + x->mode_info_context->bmi[yoffset + 4].mv.as_mv.col
michael@0:                  + x->mode_info_context->bmi[yoffset + 5].mv.as_mv.col;
michael@0: 
michael@0:             temp += 4 + ((temp >> (sizeof(temp) * CHAR_BIT - 1)) * 8);
michael@0: 
michael@0:             x->block[uoffset].bmi.mv.as_mv.col = (temp / 8) & x->fullpixel_mask;
michael@0: 
michael@0:             if (x->mode_info_context->mbmi.need_to_clamp_mvs)
michael@0:                 clamp_uvmv_to_umv_border(&x->block[uoffset].bmi.mv.as_mv, x);
michael@0: 
michael@0:             x->block[voffset].bmi.mv.as_int = x->block[uoffset].bmi.mv.as_int;
michael@0:         }
michael@0:     }
michael@0: }
michael@0: 
michael@0: void vp8_build_inter_predictors_mb(MACROBLOCKD *xd)
michael@0: {
michael@0:     if (xd->mode_info_context->mbmi.mode != SPLITMV)
michael@0:     {
michael@0:         vp8_build_inter16x16_predictors_mb(xd, xd->dst.y_buffer,
michael@0:                                            xd->dst.u_buffer, xd->dst.v_buffer,
michael@0:                                            xd->dst.y_stride, xd->dst.uv_stride);
michael@0:     }
michael@0:     else
michael@0:     {
michael@0:         build_4x4uvmvs(xd);
michael@0:         build_inter4x4_predictors_mb(xd);
michael@0:     }
michael@0: }