media/libvpx/vp8/common/reconinter.c

changeset 0
6474c204b198
     1.1 --- /dev/null	Thu Jan 01 00:00:00 1970 +0000
     1.2 +++ b/media/libvpx/vp8/common/reconinter.c	Wed Dec 31 06:09:35 2014 +0100
     1.3 @@ -0,0 +1,587 @@
     1.4 +/*
     1.5 + *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
     1.6 + *
     1.7 + *  Use of this source code is governed by a BSD-style license
     1.8 + *  that can be found in the LICENSE file in the root of the source
     1.9 + *  tree. An additional intellectual property rights grant can be found
    1.10 + *  in the file PATENTS.  All contributing project authors may
    1.11 + *  be found in the AUTHORS file in the root of the source tree.
    1.12 + */
    1.13 +
    1.14 +
    1.15 +#include <limits.h>
    1.16 +#include "vpx_config.h"
    1.17 +#include "vp8_rtcd.h"
    1.18 +#include "vpx/vpx_integer.h"
    1.19 +#include "blockd.h"
    1.20 +#include "reconinter.h"
    1.21 +#if CONFIG_RUNTIME_CPU_DETECT
    1.22 +#include "onyxc_int.h"
    1.23 +#endif
    1.24 +
    1.25 +void vp8_copy_mem16x16_c(
    1.26 +    unsigned char *src,
    1.27 +    int src_stride,
    1.28 +    unsigned char *dst,
    1.29 +    int dst_stride)
    1.30 +{
    1.31 +
    1.32 +    int r;
    1.33 +
    1.34 +    for (r = 0; r < 16; r++)
    1.35 +    {
    1.36 +#if !(CONFIG_FAST_UNALIGNED)
    1.37 +        dst[0] = src[0];
    1.38 +        dst[1] = src[1];
    1.39 +        dst[2] = src[2];
    1.40 +        dst[3] = src[3];
    1.41 +        dst[4] = src[4];
    1.42 +        dst[5] = src[5];
    1.43 +        dst[6] = src[6];
    1.44 +        dst[7] = src[7];
    1.45 +        dst[8] = src[8];
    1.46 +        dst[9] = src[9];
    1.47 +        dst[10] = src[10];
    1.48 +        dst[11] = src[11];
    1.49 +        dst[12] = src[12];
    1.50 +        dst[13] = src[13];
    1.51 +        dst[14] = src[14];
    1.52 +        dst[15] = src[15];
    1.53 +
    1.54 +#else
    1.55 +        ((uint32_t *)dst)[0] = ((uint32_t *)src)[0] ;
    1.56 +        ((uint32_t *)dst)[1] = ((uint32_t *)src)[1] ;
    1.57 +        ((uint32_t *)dst)[2] = ((uint32_t *)src)[2] ;
    1.58 +        ((uint32_t *)dst)[3] = ((uint32_t *)src)[3] ;
    1.59 +
    1.60 +#endif
    1.61 +        src += src_stride;
    1.62 +        dst += dst_stride;
    1.63 +
    1.64 +    }
    1.65 +
    1.66 +}
    1.67 +
    1.68 +void vp8_copy_mem8x8_c(
    1.69 +    unsigned char *src,
    1.70 +    int src_stride,
    1.71 +    unsigned char *dst,
    1.72 +    int dst_stride)
    1.73 +{
    1.74 +    int r;
    1.75 +
    1.76 +    for (r = 0; r < 8; r++)
    1.77 +    {
    1.78 +#if !(CONFIG_FAST_UNALIGNED)
    1.79 +        dst[0] = src[0];
    1.80 +        dst[1] = src[1];
    1.81 +        dst[2] = src[2];
    1.82 +        dst[3] = src[3];
    1.83 +        dst[4] = src[4];
    1.84 +        dst[5] = src[5];
    1.85 +        dst[6] = src[6];
    1.86 +        dst[7] = src[7];
    1.87 +#else
    1.88 +        ((uint32_t *)dst)[0] = ((uint32_t *)src)[0] ;
    1.89 +        ((uint32_t *)dst)[1] = ((uint32_t *)src)[1] ;
    1.90 +#endif
    1.91 +        src += src_stride;
    1.92 +        dst += dst_stride;
    1.93 +
    1.94 +    }
    1.95 +
    1.96 +}
    1.97 +
    1.98 +void vp8_copy_mem8x4_c(
    1.99 +    unsigned char *src,
   1.100 +    int src_stride,
   1.101 +    unsigned char *dst,
   1.102 +    int dst_stride)
   1.103 +{
   1.104 +    int r;
   1.105 +
   1.106 +    for (r = 0; r < 4; r++)
   1.107 +    {
   1.108 +#if !(CONFIG_FAST_UNALIGNED)
   1.109 +        dst[0] = src[0];
   1.110 +        dst[1] = src[1];
   1.111 +        dst[2] = src[2];
   1.112 +        dst[3] = src[3];
   1.113 +        dst[4] = src[4];
   1.114 +        dst[5] = src[5];
   1.115 +        dst[6] = src[6];
   1.116 +        dst[7] = src[7];
   1.117 +#else
   1.118 +        ((uint32_t *)dst)[0] = ((uint32_t *)src)[0] ;
   1.119 +        ((uint32_t *)dst)[1] = ((uint32_t *)src)[1] ;
   1.120 +#endif
   1.121 +        src += src_stride;
   1.122 +        dst += dst_stride;
   1.123 +
   1.124 +    }
   1.125 +
   1.126 +}
   1.127 +
   1.128 +
   1.129 +void vp8_build_inter_predictors_b(BLOCKD *d, int pitch, unsigned char *base_pre, int pre_stride, vp8_subpix_fn_t sppf)
   1.130 +{
   1.131 +    int r;
   1.132 +    unsigned char *pred_ptr = d->predictor;
   1.133 +    unsigned char *ptr;
   1.134 +    ptr = base_pre + d->offset + (d->bmi.mv.as_mv.row >> 3) * pre_stride + (d->bmi.mv.as_mv.col >> 3);
   1.135 +
   1.136 +    if (d->bmi.mv.as_mv.row & 7 || d->bmi.mv.as_mv.col & 7)
   1.137 +    {
   1.138 +        sppf(ptr, pre_stride, d->bmi.mv.as_mv.col & 7, d->bmi.mv.as_mv.row & 7, pred_ptr, pitch);
   1.139 +    }
   1.140 +    else
   1.141 +    {
   1.142 +        for (r = 0; r < 4; r++)
   1.143 +        {
   1.144 +            pred_ptr[0]  = ptr[0];
   1.145 +            pred_ptr[1]  = ptr[1];
   1.146 +            pred_ptr[2]  = ptr[2];
   1.147 +            pred_ptr[3]  = ptr[3];
   1.148 +            pred_ptr     += pitch;
   1.149 +            ptr         += pre_stride;
   1.150 +        }
   1.151 +    }
   1.152 +}
   1.153 +
   1.154 +static void build_inter_predictors4b(MACROBLOCKD *x, BLOCKD *d, unsigned char *dst, int dst_stride, unsigned char *base_pre, int pre_stride)
   1.155 +{
   1.156 +    unsigned char *ptr;
   1.157 +    ptr = base_pre + d->offset + (d->bmi.mv.as_mv.row >> 3) * pre_stride + (d->bmi.mv.as_mv.col >> 3);
   1.158 +
   1.159 +    if (d->bmi.mv.as_mv.row & 7 || d->bmi.mv.as_mv.col & 7)
   1.160 +    {
   1.161 +        x->subpixel_predict8x8(ptr, pre_stride, d->bmi.mv.as_mv.col & 7, d->bmi.mv.as_mv.row & 7, dst, dst_stride);
   1.162 +    }
   1.163 +    else
   1.164 +    {
   1.165 +        vp8_copy_mem8x8(ptr, pre_stride, dst, dst_stride);
   1.166 +    }
   1.167 +}
   1.168 +
   1.169 +static void build_inter_predictors2b(MACROBLOCKD *x, BLOCKD *d, unsigned char *dst, int dst_stride, unsigned char *base_pre, int pre_stride)
   1.170 +{
   1.171 +    unsigned char *ptr;
   1.172 +    ptr = base_pre + d->offset + (d->bmi.mv.as_mv.row >> 3) * pre_stride + (d->bmi.mv.as_mv.col >> 3);
   1.173 +
   1.174 +    if (d->bmi.mv.as_mv.row & 7 || d->bmi.mv.as_mv.col & 7)
   1.175 +    {
   1.176 +        x->subpixel_predict8x4(ptr, pre_stride, d->bmi.mv.as_mv.col & 7, d->bmi.mv.as_mv.row & 7, dst, dst_stride);
   1.177 +    }
   1.178 +    else
   1.179 +    {
   1.180 +        vp8_copy_mem8x4(ptr, pre_stride, dst, dst_stride);
   1.181 +    }
   1.182 +}
   1.183 +
   1.184 +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)
   1.185 +{
   1.186 +    int r;
   1.187 +    unsigned char *ptr;
   1.188 +    ptr = base_pre + d->offset + (d->bmi.mv.as_mv.row >> 3) * pre_stride + (d->bmi.mv.as_mv.col >> 3);
   1.189 +
   1.190 +    if (d->bmi.mv.as_mv.row & 7 || d->bmi.mv.as_mv.col & 7)
   1.191 +    {
   1.192 +        sppf(ptr, pre_stride, d->bmi.mv.as_mv.col & 7, d->bmi.mv.as_mv.row & 7, dst, dst_stride);
   1.193 +    }
   1.194 +    else
   1.195 +    {
   1.196 +        for (r = 0; r < 4; r++)
   1.197 +        {
   1.198 +          dst[0]  = ptr[0];
   1.199 +          dst[1]  = ptr[1];
   1.200 +          dst[2]  = ptr[2];
   1.201 +          dst[3]  = ptr[3];
   1.202 +          dst     += dst_stride;
   1.203 +          ptr     += pre_stride;
   1.204 +        }
   1.205 +    }
   1.206 +}
   1.207 +
   1.208 +
   1.209 +/*encoder only*/
   1.210 +void vp8_build_inter16x16_predictors_mbuv(MACROBLOCKD *x)
   1.211 +{
   1.212 +    unsigned char *uptr, *vptr;
   1.213 +    unsigned char *upred_ptr = &x->predictor[256];
   1.214 +    unsigned char *vpred_ptr = &x->predictor[320];
   1.215 +
   1.216 +    int mv_row = x->mode_info_context->mbmi.mv.as_mv.row;
   1.217 +    int mv_col = x->mode_info_context->mbmi.mv.as_mv.col;
   1.218 +    int offset;
   1.219 +    int pre_stride = x->pre.uv_stride;
   1.220 +
   1.221 +    /* calc uv motion vectors */
   1.222 +    mv_row += 1 | (mv_row >> (sizeof(int) * CHAR_BIT - 1));
   1.223 +    mv_col += 1 | (mv_col >> (sizeof(int) * CHAR_BIT - 1));
   1.224 +    mv_row /= 2;
   1.225 +    mv_col /= 2;
   1.226 +    mv_row &= x->fullpixel_mask;
   1.227 +    mv_col &= x->fullpixel_mask;
   1.228 +
   1.229 +    offset = (mv_row >> 3) * pre_stride + (mv_col >> 3);
   1.230 +    uptr = x->pre.u_buffer + offset;
   1.231 +    vptr = x->pre.v_buffer + offset;
   1.232 +
   1.233 +    if ((mv_row | mv_col) & 7)
   1.234 +    {
   1.235 +        x->subpixel_predict8x8(uptr, pre_stride, mv_col & 7, mv_row & 7, upred_ptr, 8);
   1.236 +        x->subpixel_predict8x8(vptr, pre_stride, mv_col & 7, mv_row & 7, vpred_ptr, 8);
   1.237 +    }
   1.238 +    else
   1.239 +    {
   1.240 +        vp8_copy_mem8x8(uptr, pre_stride, upred_ptr, 8);
   1.241 +        vp8_copy_mem8x8(vptr, pre_stride, vpred_ptr, 8);
   1.242 +    }
   1.243 +}
   1.244 +
   1.245 +/*encoder only*/
   1.246 +void vp8_build_inter4x4_predictors_mbuv(MACROBLOCKD *x)
   1.247 +{
   1.248 +    int i, j;
   1.249 +    int pre_stride = x->pre.uv_stride;
   1.250 +    unsigned char *base_pre;
   1.251 +
   1.252 +    /* build uv mvs */
   1.253 +    for (i = 0; i < 2; i++)
   1.254 +    {
   1.255 +        for (j = 0; j < 2; j++)
   1.256 +        {
   1.257 +            int yoffset = i * 8 + j * 2;
   1.258 +            int uoffset = 16 + i * 2 + j;
   1.259 +            int voffset = 20 + i * 2 + j;
   1.260 +
   1.261 +            int temp;
   1.262 +
   1.263 +            temp = x->block[yoffset  ].bmi.mv.as_mv.row
   1.264 +                   + x->block[yoffset+1].bmi.mv.as_mv.row
   1.265 +                   + x->block[yoffset+4].bmi.mv.as_mv.row
   1.266 +                   + x->block[yoffset+5].bmi.mv.as_mv.row;
   1.267 +
   1.268 +            temp += 4 + ((temp >> (sizeof(temp) * CHAR_BIT - 1)) * 8);
   1.269 +
   1.270 +            x->block[uoffset].bmi.mv.as_mv.row = (temp / 8) & x->fullpixel_mask;
   1.271 +
   1.272 +            temp = x->block[yoffset  ].bmi.mv.as_mv.col
   1.273 +                   + x->block[yoffset+1].bmi.mv.as_mv.col
   1.274 +                   + x->block[yoffset+4].bmi.mv.as_mv.col
   1.275 +                   + x->block[yoffset+5].bmi.mv.as_mv.col;
   1.276 +
   1.277 +            temp += 4 + ((temp >> (sizeof(temp) * CHAR_BIT - 1)) * 8);
   1.278 +
   1.279 +            x->block[uoffset].bmi.mv.as_mv.col = (temp / 8) & x->fullpixel_mask;
   1.280 +
   1.281 +            x->block[voffset].bmi.mv.as_int = x->block[uoffset].bmi.mv.as_int;
   1.282 +        }
   1.283 +    }
   1.284 +
   1.285 +    base_pre = x->pre.u_buffer;
   1.286 +    for (i = 16; i < 20; i += 2)
   1.287 +    {
   1.288 +        BLOCKD *d0 = &x->block[i];
   1.289 +        BLOCKD *d1 = &x->block[i+1];
   1.290 +
   1.291 +        if (d0->bmi.mv.as_int == d1->bmi.mv.as_int)
   1.292 +            build_inter_predictors2b(x, d0, d0->predictor, 8, base_pre, pre_stride);
   1.293 +        else
   1.294 +        {
   1.295 +            vp8_build_inter_predictors_b(d0, 8, base_pre, pre_stride, x->subpixel_predict);
   1.296 +            vp8_build_inter_predictors_b(d1, 8, base_pre, pre_stride, x->subpixel_predict);
   1.297 +        }
   1.298 +    }
   1.299 +
   1.300 +    base_pre = x->pre.v_buffer;
   1.301 +    for (i = 20; i < 24; i += 2)
   1.302 +    {
   1.303 +        BLOCKD *d0 = &x->block[i];
   1.304 +        BLOCKD *d1 = &x->block[i+1];
   1.305 +
   1.306 +        if (d0->bmi.mv.as_int == d1->bmi.mv.as_int)
   1.307 +            build_inter_predictors2b(x, d0, d0->predictor, 8, base_pre, pre_stride);
   1.308 +        else
   1.309 +        {
   1.310 +            vp8_build_inter_predictors_b(d0, 8, base_pre, pre_stride, x->subpixel_predict);
   1.311 +            vp8_build_inter_predictors_b(d1, 8, base_pre, pre_stride, x->subpixel_predict);
   1.312 +        }
   1.313 +    }
   1.314 +}
   1.315 +
   1.316 +
   1.317 +/*encoder only*/
   1.318 +void vp8_build_inter16x16_predictors_mby(MACROBLOCKD *x,
   1.319 +                                         unsigned char *dst_y,
   1.320 +                                         int dst_ystride)
   1.321 +{
   1.322 +    unsigned char *ptr_base;
   1.323 +    unsigned char *ptr;
   1.324 +    int mv_row = x->mode_info_context->mbmi.mv.as_mv.row;
   1.325 +    int mv_col = x->mode_info_context->mbmi.mv.as_mv.col;
   1.326 +    int pre_stride = x->pre.y_stride;
   1.327 +
   1.328 +    ptr_base = x->pre.y_buffer;
   1.329 +    ptr = ptr_base + (mv_row >> 3) * pre_stride + (mv_col >> 3);
   1.330 +
   1.331 +    if ((mv_row | mv_col) & 7)
   1.332 +    {
   1.333 +        x->subpixel_predict16x16(ptr, pre_stride, mv_col & 7, mv_row & 7,
   1.334 +                                 dst_y, dst_ystride);
   1.335 +    }
   1.336 +    else
   1.337 +    {
   1.338 +        vp8_copy_mem16x16(ptr, pre_stride, dst_y,
   1.339 +            dst_ystride);
   1.340 +    }
   1.341 +}
   1.342 +
   1.343 +static void clamp_mv_to_umv_border(MV *mv, const MACROBLOCKD *xd)
   1.344 +{
   1.345 +    /* If the MV points so far into the UMV border that no visible pixels
   1.346 +     * are used for reconstruction, the subpel part of the MV can be
   1.347 +     * discarded and the MV limited to 16 pixels with equivalent results.
   1.348 +     *
   1.349 +     * This limit kicks in at 19 pixels for the top and left edges, for
   1.350 +     * the 16 pixels plus 3 taps right of the central pixel when subpel
   1.351 +     * filtering. The bottom and right edges use 16 pixels plus 2 pixels
   1.352 +     * left of the central pixel when filtering.
   1.353 +     */
   1.354 +    if (mv->col < (xd->mb_to_left_edge - (19 << 3)))
   1.355 +        mv->col = xd->mb_to_left_edge - (16 << 3);
   1.356 +    else if (mv->col > xd->mb_to_right_edge + (18 << 3))
   1.357 +        mv->col = xd->mb_to_right_edge + (16 << 3);
   1.358 +
   1.359 +    if (mv->row < (xd->mb_to_top_edge - (19 << 3)))
   1.360 +        mv->row = xd->mb_to_top_edge - (16 << 3);
   1.361 +    else if (mv->row > xd->mb_to_bottom_edge + (18 << 3))
   1.362 +        mv->row = xd->mb_to_bottom_edge + (16 << 3);
   1.363 +}
   1.364 +
   1.365 +/* A version of the above function for chroma block MVs.*/
   1.366 +static void clamp_uvmv_to_umv_border(MV *mv, const MACROBLOCKD *xd)
   1.367 +{
   1.368 +    mv->col = (2*mv->col < (xd->mb_to_left_edge - (19 << 3))) ?
   1.369 +        (xd->mb_to_left_edge - (16 << 3)) >> 1 : mv->col;
   1.370 +    mv->col = (2*mv->col > xd->mb_to_right_edge + (18 << 3)) ?
   1.371 +        (xd->mb_to_right_edge + (16 << 3)) >> 1 : mv->col;
   1.372 +
   1.373 +    mv->row = (2*mv->row < (xd->mb_to_top_edge - (19 << 3))) ?
   1.374 +        (xd->mb_to_top_edge - (16 << 3)) >> 1 : mv->row;
   1.375 +    mv->row = (2*mv->row > xd->mb_to_bottom_edge + (18 << 3)) ?
   1.376 +        (xd->mb_to_bottom_edge + (16 << 3)) >> 1 : mv->row;
   1.377 +}
   1.378 +
   1.379 +void vp8_build_inter16x16_predictors_mb(MACROBLOCKD *x,
   1.380 +                                        unsigned char *dst_y,
   1.381 +                                        unsigned char *dst_u,
   1.382 +                                        unsigned char *dst_v,
   1.383 +                                        int dst_ystride,
   1.384 +                                        int dst_uvstride)
   1.385 +{
   1.386 +    int offset;
   1.387 +    unsigned char *ptr;
   1.388 +    unsigned char *uptr, *vptr;
   1.389 +
   1.390 +    int_mv _16x16mv;
   1.391 +
   1.392 +    unsigned char *ptr_base = x->pre.y_buffer;
   1.393 +    int pre_stride = x->pre.y_stride;
   1.394 +
   1.395 +    _16x16mv.as_int = x->mode_info_context->mbmi.mv.as_int;
   1.396 +
   1.397 +    if (x->mode_info_context->mbmi.need_to_clamp_mvs)
   1.398 +    {
   1.399 +        clamp_mv_to_umv_border(&_16x16mv.as_mv, x);
   1.400 +    }
   1.401 +
   1.402 +    ptr = ptr_base + ( _16x16mv.as_mv.row >> 3) * pre_stride + (_16x16mv.as_mv.col >> 3);
   1.403 +
   1.404 +    if ( _16x16mv.as_int & 0x00070007)
   1.405 +    {
   1.406 +        x->subpixel_predict16x16(ptr, pre_stride, _16x16mv.as_mv.col & 7,  _16x16mv.as_mv.row & 7, dst_y, dst_ystride);
   1.407 +    }
   1.408 +    else
   1.409 +    {
   1.410 +        vp8_copy_mem16x16(ptr, pre_stride, dst_y, dst_ystride);
   1.411 +    }
   1.412 +
   1.413 +    /* calc uv motion vectors */
   1.414 +    _16x16mv.as_mv.row += 1 | (_16x16mv.as_mv.row >> (sizeof(int) * CHAR_BIT - 1));
   1.415 +    _16x16mv.as_mv.col += 1 | (_16x16mv.as_mv.col >> (sizeof(int) * CHAR_BIT - 1));
   1.416 +    _16x16mv.as_mv.row /= 2;
   1.417 +    _16x16mv.as_mv.col /= 2;
   1.418 +    _16x16mv.as_mv.row &= x->fullpixel_mask;
   1.419 +    _16x16mv.as_mv.col &= x->fullpixel_mask;
   1.420 +
   1.421 +    pre_stride >>= 1;
   1.422 +    offset = ( _16x16mv.as_mv.row >> 3) * pre_stride + (_16x16mv.as_mv.col >> 3);
   1.423 +    uptr = x->pre.u_buffer + offset;
   1.424 +    vptr = x->pre.v_buffer + offset;
   1.425 +
   1.426 +    if ( _16x16mv.as_int & 0x00070007)
   1.427 +    {
   1.428 +        x->subpixel_predict8x8(uptr, pre_stride, _16x16mv.as_mv.col & 7,  _16x16mv.as_mv.row & 7, dst_u, dst_uvstride);
   1.429 +        x->subpixel_predict8x8(vptr, pre_stride, _16x16mv.as_mv.col & 7,  _16x16mv.as_mv.row & 7, dst_v, dst_uvstride);
   1.430 +    }
   1.431 +    else
   1.432 +    {
   1.433 +        vp8_copy_mem8x8(uptr, pre_stride, dst_u, dst_uvstride);
   1.434 +        vp8_copy_mem8x8(vptr, pre_stride, dst_v, dst_uvstride);
   1.435 +    }
   1.436 +}
   1.437 +
   1.438 +static void build_inter4x4_predictors_mb(MACROBLOCKD *x)
   1.439 +{
   1.440 +    int i;
   1.441 +    unsigned char *base_dst = x->dst.y_buffer;
   1.442 +    unsigned char *base_pre = x->pre.y_buffer;
   1.443 +
   1.444 +    if (x->mode_info_context->mbmi.partitioning < 3)
   1.445 +    {
   1.446 +        BLOCKD *b;
   1.447 +        int dst_stride = x->dst.y_stride;
   1.448 +
   1.449 +        x->block[ 0].bmi = x->mode_info_context->bmi[ 0];
   1.450 +        x->block[ 2].bmi = x->mode_info_context->bmi[ 2];
   1.451 +        x->block[ 8].bmi = x->mode_info_context->bmi[ 8];
   1.452 +        x->block[10].bmi = x->mode_info_context->bmi[10];
   1.453 +        if (x->mode_info_context->mbmi.need_to_clamp_mvs)
   1.454 +        {
   1.455 +            clamp_mv_to_umv_border(&x->block[ 0].bmi.mv.as_mv, x);
   1.456 +            clamp_mv_to_umv_border(&x->block[ 2].bmi.mv.as_mv, x);
   1.457 +            clamp_mv_to_umv_border(&x->block[ 8].bmi.mv.as_mv, x);
   1.458 +            clamp_mv_to_umv_border(&x->block[10].bmi.mv.as_mv, x);
   1.459 +        }
   1.460 +
   1.461 +        b = &x->block[ 0];
   1.462 +        build_inter_predictors4b(x, b, base_dst + b->offset, dst_stride, base_pre, dst_stride);
   1.463 +        b = &x->block[ 2];
   1.464 +        build_inter_predictors4b(x, b, base_dst + b->offset, dst_stride, base_pre, dst_stride);
   1.465 +        b = &x->block[ 8];
   1.466 +        build_inter_predictors4b(x, b, base_dst + b->offset, dst_stride, base_pre, dst_stride);
   1.467 +        b = &x->block[10];
   1.468 +        build_inter_predictors4b(x, b, base_dst + b->offset, dst_stride, base_pre, dst_stride);
   1.469 +    }
   1.470 +    else
   1.471 +    {
   1.472 +        for (i = 0; i < 16; i += 2)
   1.473 +        {
   1.474 +            BLOCKD *d0 = &x->block[i];
   1.475 +            BLOCKD *d1 = &x->block[i+1];
   1.476 +            int dst_stride = x->dst.y_stride;
   1.477 +
   1.478 +            x->block[i+0].bmi = x->mode_info_context->bmi[i+0];
   1.479 +            x->block[i+1].bmi = x->mode_info_context->bmi[i+1];
   1.480 +            if (x->mode_info_context->mbmi.need_to_clamp_mvs)
   1.481 +            {
   1.482 +                clamp_mv_to_umv_border(&x->block[i+0].bmi.mv.as_mv, x);
   1.483 +                clamp_mv_to_umv_border(&x->block[i+1].bmi.mv.as_mv, x);
   1.484 +            }
   1.485 +
   1.486 +            if (d0->bmi.mv.as_int == d1->bmi.mv.as_int)
   1.487 +                build_inter_predictors2b(x, d0, base_dst + d0->offset, dst_stride, base_pre, dst_stride);
   1.488 +            else
   1.489 +            {
   1.490 +                build_inter_predictors_b(d0, base_dst + d0->offset, dst_stride, base_pre, dst_stride, x->subpixel_predict);
   1.491 +                build_inter_predictors_b(d1, base_dst + d1->offset, dst_stride, base_pre, dst_stride, x->subpixel_predict);
   1.492 +            }
   1.493 +
   1.494 +        }
   1.495 +
   1.496 +    }
   1.497 +    base_dst = x->dst.u_buffer;
   1.498 +    base_pre = x->pre.u_buffer;
   1.499 +    for (i = 16; i < 20; i += 2)
   1.500 +    {
   1.501 +        BLOCKD *d0 = &x->block[i];
   1.502 +        BLOCKD *d1 = &x->block[i+1];
   1.503 +        int dst_stride = x->dst.uv_stride;
   1.504 +
   1.505 +        /* Note: uv mvs already clamped in build_4x4uvmvs() */
   1.506 +
   1.507 +        if (d0->bmi.mv.as_int == d1->bmi.mv.as_int)
   1.508 +            build_inter_predictors2b(x, d0, base_dst + d0->offset, dst_stride, base_pre, dst_stride);
   1.509 +        else
   1.510 +        {
   1.511 +            build_inter_predictors_b(d0, base_dst + d0->offset, dst_stride, base_pre, dst_stride, x->subpixel_predict);
   1.512 +            build_inter_predictors_b(d1, base_dst + d1->offset, dst_stride, base_pre, dst_stride, x->subpixel_predict);
   1.513 +        }
   1.514 +    }
   1.515 +
   1.516 +    base_dst = x->dst.v_buffer;
   1.517 +    base_pre = x->pre.v_buffer;
   1.518 +    for (i = 20; i < 24; i += 2)
   1.519 +    {
   1.520 +        BLOCKD *d0 = &x->block[i];
   1.521 +        BLOCKD *d1 = &x->block[i+1];
   1.522 +        int dst_stride = x->dst.uv_stride;
   1.523 +
   1.524 +        /* Note: uv mvs already clamped in build_4x4uvmvs() */
   1.525 +
   1.526 +        if (d0->bmi.mv.as_int == d1->bmi.mv.as_int)
   1.527 +            build_inter_predictors2b(x, d0, base_dst + d0->offset, dst_stride, base_pre, dst_stride);
   1.528 +        else
   1.529 +        {
   1.530 +            build_inter_predictors_b(d0, base_dst + d0->offset, dst_stride, base_pre, dst_stride, x->subpixel_predict);
   1.531 +            build_inter_predictors_b(d1, base_dst + d1->offset, dst_stride, base_pre, dst_stride, x->subpixel_predict);
   1.532 +        }
   1.533 +    }
   1.534 +}
   1.535 +
   1.536 +static
   1.537 +void build_4x4uvmvs(MACROBLOCKD *x)
   1.538 +{
   1.539 +    int i, j;
   1.540 +
   1.541 +    for (i = 0; i < 2; i++)
   1.542 +    {
   1.543 +        for (j = 0; j < 2; j++)
   1.544 +        {
   1.545 +            int yoffset = i * 8 + j * 2;
   1.546 +            int uoffset = 16 + i * 2 + j;
   1.547 +            int voffset = 20 + i * 2 + j;
   1.548 +
   1.549 +            int temp;
   1.550 +
   1.551 +            temp = x->mode_info_context->bmi[yoffset + 0].mv.as_mv.row
   1.552 +                 + x->mode_info_context->bmi[yoffset + 1].mv.as_mv.row
   1.553 +                 + x->mode_info_context->bmi[yoffset + 4].mv.as_mv.row
   1.554 +                 + x->mode_info_context->bmi[yoffset + 5].mv.as_mv.row;
   1.555 +
   1.556 +            temp += 4 + ((temp >> (sizeof(temp) * CHAR_BIT - 1)) * 8);
   1.557 +
   1.558 +            x->block[uoffset].bmi.mv.as_mv.row = (temp / 8) & x->fullpixel_mask;
   1.559 +
   1.560 +            temp = x->mode_info_context->bmi[yoffset + 0].mv.as_mv.col
   1.561 +                 + x->mode_info_context->bmi[yoffset + 1].mv.as_mv.col
   1.562 +                 + x->mode_info_context->bmi[yoffset + 4].mv.as_mv.col
   1.563 +                 + x->mode_info_context->bmi[yoffset + 5].mv.as_mv.col;
   1.564 +
   1.565 +            temp += 4 + ((temp >> (sizeof(temp) * CHAR_BIT - 1)) * 8);
   1.566 +
   1.567 +            x->block[uoffset].bmi.mv.as_mv.col = (temp / 8) & x->fullpixel_mask;
   1.568 +
   1.569 +            if (x->mode_info_context->mbmi.need_to_clamp_mvs)
   1.570 +                clamp_uvmv_to_umv_border(&x->block[uoffset].bmi.mv.as_mv, x);
   1.571 +
   1.572 +            x->block[voffset].bmi.mv.as_int = x->block[uoffset].bmi.mv.as_int;
   1.573 +        }
   1.574 +    }
   1.575 +}
   1.576 +
   1.577 +void vp8_build_inter_predictors_mb(MACROBLOCKD *xd)
   1.578 +{
   1.579 +    if (xd->mode_info_context->mbmi.mode != SPLITMV)
   1.580 +    {
   1.581 +        vp8_build_inter16x16_predictors_mb(xd, xd->dst.y_buffer,
   1.582 +                                           xd->dst.u_buffer, xd->dst.v_buffer,
   1.583 +                                           xd->dst.y_stride, xd->dst.uv_stride);
   1.584 +    }
   1.585 +    else
   1.586 +    {
   1.587 +        build_4x4uvmvs(xd);
   1.588 +        build_inter4x4_predictors_mb(xd);
   1.589 +    }
   1.590 +}

mercurial