/src/libvpx/vp8/common/reconinter.c

Source (jump to first uncovered line)
/*
 *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
 *
 *  Use of this source code is governed by a BSD-style license
 *  that can be found in the LICENSE file in the root of the source
 *  tree. An additional intellectual property rights grant can be found
 *  in the file PATENTS.  All contributing project authors may
 *  be found in the AUTHORS file in the root of the source tree.
 */

#include <limits.h>
#include <string.h>

#include "vpx_config.h"
#include "vp8_rtcd.h"
#include "vpx/vpx_integer.h"
#include "blockd.h"
#include "reconinter.h"
#if CONFIG_RUNTIME_CPU_DETECT
#include "onyxc_int.h"
#endif

void vp8_copy_mem16x16_c(unsigned char *src, int src_stride, unsigned char *dst,
                         int dst_stride) {
  int r;

  for (r = 0; r < 16; ++r) {
    memcpy(dst, src, 16);

    src += src_stride;
    dst += dst_stride;
  }
}

void vp8_copy_mem8x8_c(unsigned char *src, int src_stride, unsigned char *dst,
                       int dst_stride) {
  int r;

  for (r = 0; r < 8; ++r) {
    memcpy(dst, src, 8);

    src += src_stride;
    dst += dst_stride;
  }
}

void vp8_copy_mem8x4_c(unsigned char *src, int src_stride, unsigned char *dst,
                       int dst_stride) {
  int r;

  for (r = 0; r < 4; ++r) {
    memcpy(dst, src, 8);

    src += src_stride;
    dst += dst_stride;
  }
}

void vp8_build_inter_predictors_b(BLOCKD *d, int pitch, unsigned char *base_pre,
                                  int pre_stride, vp8_subpix_fn_t sppf) {
  int r;
  unsigned char *pred_ptr = d->predictor;
  unsigned char *ptr;
  ptr = base_pre + d->offset + (d->bmi.mv.as_mv.row >> 3) * pre_stride +
        (d->bmi.mv.as_mv.col >> 3);

  if (d->bmi.mv.as_mv.row & 7 || d->bmi.mv.as_mv.col & 7) {
    sppf(ptr, pre_stride, d->bmi.mv.as_mv.col & 7, d->bmi.mv.as_mv.row & 7,
         pred_ptr, pitch);
  } else {
    for (r = 0; r < 4; ++r) {
      pred_ptr[0] = ptr[0];
      pred_ptr[1] = ptr[1];
      pred_ptr[2] = ptr[2];
      pred_ptr[3] = ptr[3];
      pred_ptr += pitch;
      ptr += pre_stride;
    }
  }
}

static void build_inter_predictors4b(MACROBLOCKD *x, BLOCKD *d,
                                     unsigned char *dst, int dst_stride,
                                     unsigned char *base_pre, int pre_stride) {
  unsigned char *ptr;
  ptr = base_pre + d->offset + (d->bmi.mv.as_mv.row >> 3) * pre_stride +
        (d->bmi.mv.as_mv.col >> 3);

  if (d->bmi.mv.as_mv.row & 7 || d->bmi.mv.as_mv.col & 7) {
    x->subpixel_predict8x8(ptr, pre_stride, d->bmi.mv.as_mv.col & 7,
                           d->bmi.mv.as_mv.row & 7, dst, dst_stride);
  } else {
    vp8_copy_mem8x8(ptr, pre_stride, dst, dst_stride);
  }
}

static void build_inter_predictors2b(MACROBLOCKD *x, BLOCKD *d,
                                     unsigned char *dst, int dst_stride,
                                     unsigned char *base_pre, int pre_stride) {
  unsigned char *ptr;
  ptr = base_pre + d->offset + (d->bmi.mv.as_mv.row >> 3) * pre_stride +
        (d->bmi.mv.as_mv.col >> 3);

  if (d->bmi.mv.as_mv.row & 7 || d->bmi.mv.as_mv.col & 7) {
    x->subpixel_predict8x4(ptr, pre_stride, d->bmi.mv.as_mv.col & 7,
                           d->bmi.mv.as_mv.row & 7, dst, dst_stride);
  } else {
    vp8_copy_mem8x4(ptr, pre_stride, dst, dst_stride);
  }
}

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) {
  int r;
  unsigned char *ptr;
  ptr = base_pre + d->offset + (d->bmi.mv.as_mv.row >> 3) * pre_stride +
        (d->bmi.mv.as_mv.col >> 3);

  if (d->bmi.mv.as_mv.row & 7 || d->bmi.mv.as_mv.col & 7) {
    sppf(ptr, pre_stride, d->bmi.mv.as_mv.col & 7, d->bmi.mv.as_mv.row & 7, dst,
         dst_stride);
  } else {
    for (r = 0; r < 4; ++r) {
      dst[0] = ptr[0];
      dst[1] = ptr[1];
      dst[2] = ptr[2];
      dst[3] = ptr[3];
      dst += dst_stride;
      ptr += pre_stride;
    }
  }
}

/*encoder only*/
void vp8_build_inter16x16_predictors_mbuv(MACROBLOCKD *x) {
  unsigned char *uptr, *vptr;
  unsigned char *upred_ptr = &x->predictor[256];
  unsigned char *vpred_ptr = &x->predictor[320];

  int mv_row = x->mode_info_context->mbmi.mv.as_mv.row;
  int mv_col = x->mode_info_context->mbmi.mv.as_mv.col;
  int offset;
  int pre_stride = x->pre.uv_stride;

  /* calc uv motion vectors */
  mv_row += 1 | (mv_row >> (sizeof(int) * CHAR_BIT - 1));
  mv_col += 1 | (mv_col >> (sizeof(int) * CHAR_BIT - 1));
  mv_row /= 2;
  mv_col /= 2;
  mv_row &= x->fullpixel_mask;
  mv_col &= x->fullpixel_mask;

  offset = (mv_row >> 3) * pre_stride + (mv_col >> 3);
  uptr = x->pre.u_buffer + offset;
  vptr = x->pre.v_buffer + offset;

  if ((mv_row | mv_col) & 7) {
    x->subpixel_predict8x8(uptr, pre_stride, mv_col & 7, mv_row & 7, upred_ptr,
                           8);
    x->subpixel_predict8x8(vptr, pre_stride, mv_col & 7, mv_row & 7, vpred_ptr,
                           8);
  } else {
    vp8_copy_mem8x8(uptr, pre_stride, upred_ptr, 8);
    vp8_copy_mem8x8(vptr, pre_stride, vpred_ptr, 8);
  }
}

/*encoder only*/
void vp8_build_inter4x4_predictors_mbuv(MACROBLOCKD *x) {
  int i, j;
  int pre_stride = x->pre.uv_stride;
  unsigned char *base_pre;

  /* build uv mvs */
  for (i = 0; i < 2; ++i) {
    for (j = 0; j < 2; ++j) {
      int yoffset = i * 8 + j * 2;
      int uoffset = 16 + i * 2 + j;
      int voffset = 20 + i * 2 + j;

      int temp;

      temp = x->block[yoffset].bmi.mv.as_mv.row +
             x->block[yoffset + 1].bmi.mv.as_mv.row +
             x->block[yoffset + 4].bmi.mv.as_mv.row +
             x->block[yoffset + 5].bmi.mv.as_mv.row;

      temp += 4 + ((temp >> (sizeof(temp) * CHAR_BIT - 1)) * 8);

      x->block[uoffset].bmi.mv.as_mv.row = (temp / 8) & x->fullpixel_mask;

      temp = x->block[yoffset].bmi.mv.as_mv.col +
             x->block[yoffset + 1].bmi.mv.as_mv.col +
             x->block[yoffset + 4].bmi.mv.as_mv.col +
             x->block[yoffset + 5].bmi.mv.as_mv.col;

      temp += 4 + ((temp >> (sizeof(temp) * CHAR_BIT - 1)) * 8);

      x->block[uoffset].bmi.mv.as_mv.col = (temp / 8) & x->fullpixel_mask;

      x->block[voffset].bmi.mv.as_int = x->block[uoffset].bmi.mv.as_int;
    }
  }

  base_pre = x->pre.u_buffer;
  for (i = 16; i < 20; i += 2) {
    BLOCKD *d0 = &x->block[i];
    BLOCKD *d1 = &x->block[i + 1];

    if (d0->bmi.mv.as_int == d1->bmi.mv.as_int) {
      build_inter_predictors2b(x, d0, d0->predictor, 8, base_pre, pre_stride);
    } else {
      vp8_build_inter_predictors_b(d0, 8, base_pre, pre_stride,
                                   x->subpixel_predict);
      vp8_build_inter_predictors_b(d1, 8, base_pre, pre_stride,
                                   x->subpixel_predict);
    }
  }

  base_pre = x->pre.v_buffer;
  for (i = 20; i < 24; i += 2) {
    BLOCKD *d0 = &x->block[i];
    BLOCKD *d1 = &x->block[i + 1];

    if (d0->bmi.mv.as_int == d1->bmi.mv.as_int) {
      build_inter_predictors2b(x, d0, d0->predictor, 8, base_pre, pre_stride);
    } else {
      vp8_build_inter_predictors_b(d0, 8, base_pre, pre_stride,
                                   x->subpixel_predict);
      vp8_build_inter_predictors_b(d1, 8, base_pre, pre_stride,
                                   x->subpixel_predict);
    }
  }
}

/*encoder only*/
void vp8_build_inter16x16_predictors_mby(MACROBLOCKD *x, unsigned char *dst_y,
                                         int dst_ystride) {
  unsigned char *ptr_base;
  unsigned char *ptr;
  int mv_row = x->mode_info_context->mbmi.mv.as_mv.row;
  int mv_col = x->mode_info_context->mbmi.mv.as_mv.col;
  int pre_stride = x->pre.y_stride;

  ptr_base = x->pre.y_buffer;
  ptr = ptr_base + (mv_row >> 3) * pre_stride + (mv_col >> 3);

  if ((mv_row | mv_col) & 7) {
    x->subpixel_predict16x16(ptr, pre_stride, mv_col & 7, mv_row & 7, dst_y,
                             dst_ystride);
  } else {
    vp8_copy_mem16x16(ptr, pre_stride, dst_y, dst_ystride);
  }
}

static void clamp_mv_to_umv_border(MV *mv, const MACROBLOCKD *xd) {
  /* If the MV points so far into the UMV border that no visible pixels
   * are used for reconstruction, the subpel part of the MV can be
   * discarded and the MV limited to 16 pixels with equivalent results.
   *
   * This limit kicks in at 19 pixels for the top and left edges, for
   * the 16 pixels plus 3 taps right of the central pixel when subpel
   * filtering. The bottom and right edges use 16 pixels plus 2 pixels
   * left of the central pixel when filtering.
   */
  if (mv->col < (xd->mb_to_left_edge - (19 << 3))) {
    mv->col = xd->mb_to_left_edge - (16 << 3);
  } else if (mv->col > xd->mb_to_right_edge + (18 << 3)) {
    mv->col = xd->mb_to_right_edge + (16 << 3);
  }

  if (mv->row < (xd->mb_to_top_edge - (19 << 3))) {
    mv->row = xd->mb_to_top_edge - (16 << 3);
  } else if (mv->row > xd->mb_to_bottom_edge + (18 << 3)) {
    mv->row = xd->mb_to_bottom_edge + (16 << 3);
  }
}

/* A version of the above function for chroma block MVs.*/
static void clamp_uvmv_to_umv_border(MV *mv, const MACROBLOCKD *xd) {
  mv->col = (2 * mv->col < (xd->mb_to_left_edge - (19 << 3)))
                ? (xd->mb_to_left_edge - (16 << 3)) >> 1
                : mv->col;
  mv->col = (2 * mv->col > xd->mb_to_right_edge + (18 << 3))
                ? (xd->mb_to_right_edge + (16 << 3)) >> 1
                : mv->col;

  mv->row = (2 * mv->row < (xd->mb_to_top_edge - (19 << 3)))
                ? (xd->mb_to_top_edge - (16 << 3)) >> 1
                : mv->row;
  mv->row = (2 * mv->row > xd->mb_to_bottom_edge + (18 << 3))
                ? (xd->mb_to_bottom_edge + (16 << 3)) >> 1
                : mv->row;
}

void vp8_build_inter16x16_predictors_mb(MACROBLOCKD *x, unsigned char *dst_y,
                                        unsigned char *dst_u,
                                        unsigned char *dst_v, int dst_ystride,
                                        int dst_uvstride) {
  int offset;
  unsigned char *ptr;
  unsigned char *uptr, *vptr;

  int_mv _16x16mv;

  unsigned char *ptr_base = x->pre.y_buffer;
  int pre_stride = x->pre.y_stride;

  _16x16mv.as_int = x->mode_info_context->mbmi.mv.as_int;

  if (x->mode_info_context->mbmi.need_to_clamp_mvs) {
    clamp_mv_to_umv_border(&_16x16mv.as_mv, x);
  }

  ptr = ptr_base + (_16x16mv.as_mv.row >> 3) * pre_stride +
        (_16x16mv.as_mv.col >> 3);

  if (_16x16mv.as_int & 0x00070007) {
    x->subpixel_predict16x16(ptr, pre_stride, _16x16mv.as_mv.col & 7,
                             _16x16mv.as_mv.row & 7, dst_y, dst_ystride);
  } else {
    vp8_copy_mem16x16(ptr, pre_stride, dst_y, dst_ystride);
  }

  /* calc uv motion vectors */
  _16x16mv.as_mv.row +=
      1 | (_16x16mv.as_mv.row >> (sizeof(int) * CHAR_BIT - 1));
  _16x16mv.as_mv.col +=
      1 | (_16x16mv.as_mv.col >> (sizeof(int) * CHAR_BIT - 1));
  _16x16mv.as_mv.row /= 2;
  _16x16mv.as_mv.col /= 2;
  _16x16mv.as_mv.row &= x->fullpixel_mask;
  _16x16mv.as_mv.col &= x->fullpixel_mask;

  if (2 * _16x16mv.as_mv.col < (x->mb_to_left_edge - (19 << 3)) ||
      2 * _16x16mv.as_mv.col > x->mb_to_right_edge + (18 << 3) ||
      2 * _16x16mv.as_mv.row < (x->mb_to_top_edge - (19 << 3)) ||
      2 * _16x16mv.as_mv.row > x->mb_to_bottom_edge + (18 << 3)) {
    return;
  }

  pre_stride >>= 1;
  offset = (_16x16mv.as_mv.row >> 3) * pre_stride + (_16x16mv.as_mv.col >> 3);
  uptr = x->pre.u_buffer + offset;
  vptr = x->pre.v_buffer + offset;

  if (_16x16mv.as_int & 0x00070007) {
    x->subpixel_predict8x8(uptr, pre_stride, _16x16mv.as_mv.col & 7,
                           _16x16mv.as_mv.row & 7, dst_u, dst_uvstride);
    x->subpixel_predict8x8(vptr, pre_stride, _16x16mv.as_mv.col & 7,
                           _16x16mv.as_mv.row & 7, dst_v, dst_uvstride);
  } else {
    vp8_copy_mem8x8(uptr, pre_stride, dst_u, dst_uvstride);
    vp8_copy_mem8x8(vptr, pre_stride, dst_v, dst_uvstride);
  }
}

static void build_inter4x4_predictors_mb(MACROBLOCKD *x) {
  int i;
  unsigned char *base_dst = x->dst.y_buffer;
  unsigned char *base_pre = x->pre.y_buffer;

  if (x->mode_info_context->mbmi.partitioning < 3) {
    BLOCKD *b;
    int dst_stride = x->dst.y_stride;

    x->block[0].bmi = x->mode_info_context->bmi[0];
    x->block[2].bmi = x->mode_info_context->bmi[2];
    x->block[8].bmi = x->mode_info_context->bmi[8];
    x->block[10].bmi = x->mode_info_context->bmi[10];
    if (x->mode_info_context->mbmi.need_to_clamp_mvs) {
      clamp_mv_to_umv_border(&x->block[0].bmi.mv.as_mv, x);
      clamp_mv_to_umv_border(&x->block[2].bmi.mv.as_mv, x);
      clamp_mv_to_umv_border(&x->block[8].bmi.mv.as_mv, x);
      clamp_mv_to_umv_border(&x->block[10].bmi.mv.as_mv, x);
    }

    b = &x->block[0];
    build_inter_predictors4b(x, b, base_dst + b->offset, dst_stride, base_pre,
                             dst_stride);
    b = &x->block[2];
    build_inter_predictors4b(x, b, base_dst + b->offset, dst_stride, base_pre,
                             dst_stride);
    b = &x->block[8];
    build_inter_predictors4b(x, b, base_dst + b->offset, dst_stride, base_pre,
                             dst_stride);
    b = &x->block[10];
    build_inter_predictors4b(x, b, base_dst + b->offset, dst_stride, base_pre,
                             dst_stride);
  } else {
    for (i = 0; i < 16; i += 2) {
      BLOCKD *d0 = &x->block[i];
      BLOCKD *d1 = &x->block[i + 1];
      int dst_stride = x->dst.y_stride;

      x->block[i + 0].bmi = x->mode_info_context->bmi[i + 0];
      x->block[i + 1].bmi = x->mode_info_context->bmi[i + 1];
      if (x->mode_info_context->mbmi.need_to_clamp_mvs) {
        clamp_mv_to_umv_border(&x->block[i + 0].bmi.mv.as_mv, x);
        clamp_mv_to_umv_border(&x->block[i + 1].bmi.mv.as_mv, x);
      }

      if (d0->bmi.mv.as_int == d1->bmi.mv.as_int) {
        build_inter_predictors2b(x, d0, base_dst + d0->offset, dst_stride,
                                 base_pre, dst_stride);
      } else {
        build_inter_predictors_b(d0, base_dst + d0->offset, dst_stride,
                                 base_pre, dst_stride, x->subpixel_predict);
        build_inter_predictors_b(d1, base_dst + d1->offset, dst_stride,
                                 base_pre, dst_stride, x->subpixel_predict);
      }
    }
  }
  base_dst = x->dst.u_buffer;
  base_pre = x->pre.u_buffer;
  for (i = 16; i < 20; i += 2) {
    BLOCKD *d0 = &x->block[i];
    BLOCKD *d1 = &x->block[i + 1];
    int dst_stride = x->dst.uv_stride;

    /* Note: uv mvs already clamped in build_4x4uvmvs() */

    if (d0->bmi.mv.as_int == d1->bmi.mv.as_int) {
      build_inter_predictors2b(x, d0, base_dst + d0->offset, dst_stride,
                               base_pre, dst_stride);
    } else {
      build_inter_predictors_b(d0, base_dst + d0->offset, dst_stride, base_pre,
                               dst_stride, x->subpixel_predict);
      build_inter_predictors_b(d1, base_dst + d1->offset, dst_stride, base_pre,
                               dst_stride, x->subpixel_predict);
    }
  }

  base_dst = x->dst.v_buffer;
  base_pre = x->pre.v_buffer;
  for (i = 20; i < 24; i += 2) {
    BLOCKD *d0 = &x->block[i];
    BLOCKD *d1 = &x->block[i + 1];
    int dst_stride = x->dst.uv_stride;

    /* Note: uv mvs already clamped in build_4x4uvmvs() */

    if (d0->bmi.mv.as_int == d1->bmi.mv.as_int) {
      build_inter_predictors2b(x, d0, base_dst + d0->offset, dst_stride,
                               base_pre, dst_stride);
    } else {
      build_inter_predictors_b(d0, base_dst + d0->offset, dst_stride, base_pre,
                               dst_stride, x->subpixel_predict);
      build_inter_predictors_b(d1, base_dst + d1->offset, dst_stride, base_pre,
                               dst_stride, x->subpixel_predict);
    }
  }
}

static void build_4x4uvmvs(MACROBLOCKD *x) {
  int i, j;

  for (i = 0; i < 2; ++i) {
    for (j = 0; j < 2; ++j) {
      int yoffset = i * 8 + j * 2;
      int uoffset = 16 + i * 2 + j;
      int voffset = 20 + i * 2 + j;

      int temp;

      temp = x->mode_info_context->bmi[yoffset + 0].mv.as_mv.row +
             x->mode_info_context->bmi[yoffset + 1].mv.as_mv.row +
             x->mode_info_context->bmi[yoffset + 4].mv.as_mv.row +
             x->mode_info_context->bmi[yoffset + 5].mv.as_mv.row;

      temp += 4 + ((temp >> (sizeof(temp) * CHAR_BIT - 1)) * 8);

      x->block[uoffset].bmi.mv.as_mv.row = (temp / 8) & x->fullpixel_mask;

      temp = x->mode_info_context->bmi[yoffset + 0].mv.as_mv.col +
             x->mode_info_context->bmi[yoffset + 1].mv.as_mv.col +
             x->mode_info_context->bmi[yoffset + 4].mv.as_mv.col +
             x->mode_info_context->bmi[yoffset + 5].mv.as_mv.col;

      temp += 4 + ((temp >> (sizeof(temp) * CHAR_BIT - 1)) * 8);

      x->block[uoffset].bmi.mv.as_mv.col = (temp / 8) & x->fullpixel_mask;

      if (x->mode_info_context->mbmi.need_to_clamp_mvs) {
        clamp_uvmv_to_umv_border(&x->block[uoffset].bmi.mv.as_mv, x);
      }

      x->block[voffset].bmi.mv.as_int = x->block[uoffset].bmi.mv.as_int;
    }
  }
}

void vp8_build_inter_predictors_mb(MACROBLOCKD *xd) {
  if (xd->mode_info_context->mbmi.mode != SPLITMV) {
    vp8_build_inter16x16_predictors_mb(xd, xd->dst.y_buffer, xd->dst.u_buffer,
                                       xd->dst.v_buffer, xd->dst.y_stride,
                                       xd->dst.uv_stride);
  } else {
    build_4x4uvmvs(xd);
    build_inter4x4_predictors_mb(xd);
  }
}

Coverage Report

Created: 2024-09-06 07:53

Line	Count	Source (jump to first uncovered line)
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		#include <limits.h>
12		#include <string.h>
13
14		#include "vpx_config.h"
15		#include "vp8_rtcd.h"
16		#include "vpx/vpx_integer.h"
17		#include "blockd.h"
18		#include "reconinter.h"
19		#if CONFIG_RUNTIME_CPU_DETECT
20		#include "onyxc_int.h"
21		#endif
22
23		void vp8_copy_mem16x16_c(unsigned char src, int src_stride, unsigned char dst,
24	0	int dst_stride) {
25	0	int r;
26
27	0	for (r = 0; r < 16; ++r) {
28	0	memcpy(dst, src, 16);
29
30	0	src += src_stride;
31	0	dst += dst_stride;
32	0	}
33	0	}
34
35		void vp8_copy_mem8x8_c(unsigned char src, int src_stride, unsigned char dst,
36	0	int dst_stride) {
37	0	int r;
38
39	0	for (r = 0; r < 8; ++r) {
40	0	memcpy(dst, src, 8);
41
42	0	src += src_stride;
43	0	dst += dst_stride;
44	0	}
45	0	}
46
47		void vp8_copy_mem8x4_c(unsigned char src, int src_stride, unsigned char dst,
48	0	int dst_stride) {
49	0	int r;
50
51	0	for (r = 0; r < 4; ++r) {
52	0	memcpy(dst, src, 8);
53
54	0	src += src_stride;
55	0	dst += dst_stride;
56	0	}
57	0	}
58
59		void vp8_build_inter_predictors_b(BLOCKD d, int pitch, unsigned char base_pre,
60	65.7M	int pre_stride, vp8_subpix_fn_t sppf) {
61	65.7M	int r;
62	65.7M	unsigned char *pred_ptr = d->predictor;
63	65.7M	unsigned char *ptr;
64	65.7M	ptr = base_pre + d->offset + (d->bmi.mv.as_mv.row >> 3) * pre_stride +
65	65.7M	(d->bmi.mv.as_mv.col >> 3);
66
67	65.7M	if (d->bmi.mv.as_mv.row & 7 \|\| d->bmi.mv.as_mv.col & 7) {
68	15.6M	sppf(ptr, pre_stride, d->bmi.mv.as_mv.col & 7, d->bmi.mv.as_mv.row & 7,
69	15.6M	pred_ptr, pitch);
70	50.1M	} else {
71	250M	for (r = 0; r < 4; ++r) {
72	200M	pred_ptr[0] = ptr[0];
73	200M	pred_ptr[1] = ptr[1];
74	200M	pred_ptr[2] = ptr[2];
75	200M	pred_ptr[3] = ptr[3];
76	200M	pred_ptr += pitch;
77	200M	ptr += pre_stride;
78	200M	}
79	50.1M	}
80	65.7M	}
81
82		static void build_inter_predictors4b(MACROBLOCKD x, BLOCKD d,
83		unsigned char *dst, int dst_stride,
84	299k	unsigned char *base_pre, int pre_stride) {
85	299k	unsigned char *ptr;
86	299k	ptr = base_pre + d->offset + (d->bmi.mv.as_mv.row >> 3) * pre_stride +
87	299k	(d->bmi.mv.as_mv.col >> 3);
88
89	299k	if (d->bmi.mv.as_mv.row & 7 \|\| d->bmi.mv.as_mv.col & 7) {
90	84.7k	x->subpixel_predict8x8(ptr, pre_stride, d->bmi.mv.as_mv.col & 7,
91	84.7k	d->bmi.mv.as_mv.row & 7, dst, dst_stride);
92	214k	} else {
93	214k	vp8_copy_mem8x8(ptr, pre_stride, dst, dst_stride);
94	214k	}
95	299k	}
96
97		static void build_inter_predictors2b(MACROBLOCKD x, BLOCKD d,
98		unsigned char *dst, int dst_stride,
99	1.03M	unsigned char *base_pre, int pre_stride) {
100	1.03M	unsigned char *ptr;
101	1.03M	ptr = base_pre + d->offset + (d->bmi.mv.as_mv.row >> 3) * pre_stride +
102	1.03M	(d->bmi.mv.as_mv.col >> 3);
103
104	1.03M	if (d->bmi.mv.as_mv.row & 7 \|\| d->bmi.mv.as_mv.col & 7) {
105	584k	x->subpixel_predict8x4(ptr, pre_stride, d->bmi.mv.as_mv.col & 7,
106	584k	d->bmi.mv.as_mv.row & 7, dst, dst_stride);
107	584k	} else {
108	455k	vp8_copy_mem8x4(ptr, pre_stride, dst, dst_stride);
109	455k	}
110	1.03M	}
111
112		static void build_inter_predictors_b(BLOCKD d, unsigned char dst,
113		int dst_stride, unsigned char *base_pre,
114	692k	int pre_stride, vp8_subpix_fn_t sppf) {
115	692k	int r;
116	692k	unsigned char *ptr;
117	692k	ptr = base_pre + d->offset + (d->bmi.mv.as_mv.row >> 3) * pre_stride +
118	692k	(d->bmi.mv.as_mv.col >> 3);
119
120	692k	if (d->bmi.mv.as_mv.row & 7 \|\| d->bmi.mv.as_mv.col & 7) {
121	406k	sppf(ptr, pre_stride, d->bmi.mv.as_mv.col & 7, d->bmi.mv.as_mv.row & 7, dst,
122	406k	dst_stride);
123	406k	} else {
124	1.43M	for (r = 0; r < 4; ++r) {
125	1.14M	dst[0] = ptr[0];
126	1.14M	dst[1] = ptr[1];
127	1.14M	dst[2] = ptr[2];
128	1.14M	dst[3] = ptr[3];
129	1.14M	dst += dst_stride;
130	1.14M	ptr += pre_stride;
131	1.14M	}
132	286k	}
133	692k	}
134
135		/encoder only/
136	2.26M	void vp8_build_inter16x16_predictors_mbuv(MACROBLOCKD *x) {
137	2.26M	unsigned char uptr, vptr;
138	2.26M	unsigned char *upred_ptr = &x->predictor[256];
139	2.26M	unsigned char *vpred_ptr = &x->predictor[320];
140
141	2.26M	int mv_row = x->mode_info_context->mbmi.mv.as_mv.row;
142	2.26M	int mv_col = x->mode_info_context->mbmi.mv.as_mv.col;
143	2.26M	int offset;
144	2.26M	int pre_stride = x->pre.uv_stride;
145
146		/* calc uv motion vectors */
147	2.26M	mv_row += 1 \| (mv_row >> (sizeof(int) * CHAR_BIT - 1));
148	2.26M	mv_col += 1 \| (mv_col >> (sizeof(int) * CHAR_BIT - 1));
149	2.26M	mv_row /= 2;
150	2.26M	mv_col /= 2;
151	2.26M	mv_row &= x->fullpixel_mask;
152	2.26M	mv_col &= x->fullpixel_mask;
153
154	2.26M	offset = (mv_row >> 3) * pre_stride + (mv_col >> 3);
155	2.26M	uptr = x->pre.u_buffer + offset;
156	2.26M	vptr = x->pre.v_buffer + offset;
157
158	2.26M	if ((mv_row \| mv_col) & 7) {
159	1.02M	x->subpixel_predict8x8(uptr, pre_stride, mv_col & 7, mv_row & 7, upred_ptr,
160	1.02M	8);
161	1.02M	x->subpixel_predict8x8(vptr, pre_stride, mv_col & 7, mv_row & 7, vpred_ptr,
162	1.02M	8);
163	1.23M	} else {
164	1.23M	vp8_copy_mem8x8(uptr, pre_stride, upred_ptr, 8);
165	1.23M	vp8_copy_mem8x8(vptr, pre_stride, vpred_ptr, 8);
166	1.23M	}
167	2.26M	}
168
169		/encoder only/
170	252k	void vp8_build_inter4x4_predictors_mbuv(MACROBLOCKD *x) {
171	252k	int i, j;
172	252k	int pre_stride = x->pre.uv_stride;
173	252k	unsigned char *base_pre;
174
175		/* build uv mvs */
176	758k	for (i = 0; i < 2; ++i) {
177	1.51M	for (j = 0; j < 2; ++j) {
178	1.01M	int yoffset = i * 8 + j * 2;
179	1.01M	int uoffset = 16 + i * 2 + j;
180	1.01M	int voffset = 20 + i * 2 + j;
181
182	1.01M	int temp;
183
184	1.01M	temp = x->block[yoffset].bmi.mv.as_mv.row +
185	1.01M	x->block[yoffset + 1].bmi.mv.as_mv.row +
186	1.01M	x->block[yoffset + 4].bmi.mv.as_mv.row +
187	1.01M	x->block[yoffset + 5].bmi.mv.as_mv.row;
188
189	1.01M	temp += 4 + ((temp >> (sizeof(temp) * CHAR_BIT - 1)) * 8);
190
191	1.01M	x->block[uoffset].bmi.mv.as_mv.row = (temp / 8) & x->fullpixel_mask;
192
193	1.01M	temp = x->block[yoffset].bmi.mv.as_mv.col +
194	1.01M	x->block[yoffset + 1].bmi.mv.as_mv.col +
195	1.01M	x->block[yoffset + 4].bmi.mv.as_mv.col +
196	1.01M	x->block[yoffset + 5].bmi.mv.as_mv.col;
197
198	1.01M	temp += 4 + ((temp >> (sizeof(temp) * CHAR_BIT - 1)) * 8);
199
200	1.01M	x->block[uoffset].bmi.mv.as_mv.col = (temp / 8) & x->fullpixel_mask;
201
202	1.01M	x->block[voffset].bmi.mv.as_int = x->block[uoffset].bmi.mv.as_int;
203	1.01M	}
204	505k	}
205
206	252k	base_pre = x->pre.u_buffer;
207	758k	for (i = 16; i < 20; i += 2) {
208	505k	BLOCKD *d0 = &x->block[i];
209	505k	BLOCKD *d1 = &x->block[i + 1];
210
211	505k	if (d0->bmi.mv.as_int == d1->bmi.mv.as_int) {
212	310k	build_inter_predictors2b(x, d0, d0->predictor, 8, base_pre, pre_stride);
213	310k	} else {
214	194k	vp8_build_inter_predictors_b(d0, 8, base_pre, pre_stride,
215	194k	x->subpixel_predict);
216	194k	vp8_build_inter_predictors_b(d1, 8, base_pre, pre_stride,
217	194k	x->subpixel_predict);
218	194k	}
219	505k	}
220
221	252k	base_pre = x->pre.v_buffer;
222	758k	for (i = 20; i < 24; i += 2) {
223	505k	BLOCKD *d0 = &x->block[i];
224	505k	BLOCKD *d1 = &x->block[i + 1];
225
226	505k	if (d0->bmi.mv.as_int == d1->bmi.mv.as_int) {
227	310k	build_inter_predictors2b(x, d0, d0->predictor, 8, base_pre, pre_stride);
228	310k	} else {
229	194k	vp8_build_inter_predictors_b(d0, 8, base_pre, pre_stride,
230	194k	x->subpixel_predict);
231	194k	vp8_build_inter_predictors_b(d1, 8, base_pre, pre_stride,
232	194k	x->subpixel_predict);
233	194k	}
234	505k	}
235	252k	}
236
237		/encoder only/
238		void vp8_build_inter16x16_predictors_mby(MACROBLOCKD x, unsigned char dst_y,
239	2.26M	int dst_ystride) {
240	2.26M	unsigned char *ptr_base;
241	2.26M	unsigned char *ptr;
242	2.26M	int mv_row = x->mode_info_context->mbmi.mv.as_mv.row;
243	2.26M	int mv_col = x->mode_info_context->mbmi.mv.as_mv.col;
244	2.26M	int pre_stride = x->pre.y_stride;
245
246	2.26M	ptr_base = x->pre.y_buffer;
247	2.26M	ptr = ptr_base + (mv_row >> 3) * pre_stride + (mv_col >> 3);
248
249	2.26M	if ((mv_row \| mv_col) & 7) {
250	560k	x->subpixel_predict16x16(ptr, pre_stride, mv_col & 7, mv_row & 7, dst_y,
251	560k	dst_ystride);
252	1.69M	} else {
253	1.69M	vp8_copy_mem16x16(ptr, pre_stride, dst_y, dst_ystride);
254	1.69M	}
255	2.26M	}
256
257	0	static void clamp_mv_to_umv_border(MV mv, const MACROBLOCKD xd) {
258		/* If the MV points so far into the UMV border that no visible pixels
259		* are used for reconstruction, the subpel part of the MV can be
260		* discarded and the MV limited to 16 pixels with equivalent results.
261		*
262		* This limit kicks in at 19 pixels for the top and left edges, for
263		* the 16 pixels plus 3 taps right of the central pixel when subpel
264		* filtering. The bottom and right edges use 16 pixels plus 2 pixels
265		* left of the central pixel when filtering.
266		*/
267	0	if (mv->col < (xd->mb_to_left_edge - (19 << 3))) {
268	0	mv->col = xd->mb_to_left_edge - (16 << 3);
269	0	} else if (mv->col > xd->mb_to_right_edge + (18 << 3)) {
270	0	mv->col = xd->mb_to_right_edge + (16 << 3);
271	0	}
272
273	0	if (mv->row < (xd->mb_to_top_edge - (19 << 3))) {
274	0	mv->row = xd->mb_to_top_edge - (16 << 3);
275	0	} else if (mv->row > xd->mb_to_bottom_edge + (18 << 3)) {
276	0	mv->row = xd->mb_to_bottom_edge + (16 << 3);
277	0	}
278	0	}
279
280		/* A version of the above function for chroma block MVs.*/
281	0	static void clamp_uvmv_to_umv_border(MV mv, const MACROBLOCKD xd) {
282	0	mv->col = (2 * mv->col < (xd->mb_to_left_edge - (19 << 3)))
283	0	? (xd->mb_to_left_edge - (16 << 3)) >> 1
284	0	: mv->col;
285	0	mv->col = (2 * mv->col > xd->mb_to_right_edge + (18 << 3))
286	0	? (xd->mb_to_right_edge + (16 << 3)) >> 1
287	0	: mv->col;
288
289	0	mv->row = (2 * mv->row < (xd->mb_to_top_edge - (19 << 3)))
290	0	? (xd->mb_to_top_edge - (16 << 3)) >> 1
291	0	: mv->row;
292	0	mv->row = (2 * mv->row > xd->mb_to_bottom_edge + (18 << 3))
293	0	? (xd->mb_to_bottom_edge + (16 << 3)) >> 1
294	0	: mv->row;
295	0	}
296
297		void vp8_build_inter16x16_predictors_mb(MACROBLOCKD x, unsigned char dst_y,
298		unsigned char *dst_u,
299		unsigned char *dst_v, int dst_ystride,
300	391k	int dst_uvstride) {
301	391k	int offset;
302	391k	unsigned char *ptr;
303	391k	unsigned char uptr, vptr;
304
305	391k	int_mv _16x16mv;
306
307	391k	unsigned char *ptr_base = x->pre.y_buffer;
308	391k	int pre_stride = x->pre.y_stride;
309
310	391k	_16x16mv.as_int = x->mode_info_context->mbmi.mv.as_int;
311
312	391k	if (x->mode_info_context->mbmi.need_to_clamp_mvs) {
313	0	clamp_mv_to_umv_border(&_16x16mv.as_mv, x);
314	0	}
315
316	391k	ptr = ptr_base + (_16x16mv.as_mv.row >> 3) * pre_stride +
317	391k	(_16x16mv.as_mv.col >> 3);
318
319	391k	if (_16x16mv.as_int & 0x00070007) {
320	148k	x->subpixel_predict16x16(ptr, pre_stride, _16x16mv.as_mv.col & 7,
321	148k	_16x16mv.as_mv.row & 7, dst_y, dst_ystride);
322	243k	} else {
323	243k	vp8_copy_mem16x16(ptr, pre_stride, dst_y, dst_ystride);
324	243k	}
325
326		/* calc uv motion vectors */
327	391k	_16x16mv.as_mv.row +=
328	391k	1 \| (_16x16mv.as_mv.row >> (sizeof(int) * CHAR_BIT - 1));
329	391k	_16x16mv.as_mv.col +=
330	391k	1 \| (_16x16mv.as_mv.col >> (sizeof(int) * CHAR_BIT - 1));
331	391k	_16x16mv.as_mv.row /= 2;
332	391k	_16x16mv.as_mv.col /= 2;
333	391k	_16x16mv.as_mv.row &= x->fullpixel_mask;
334	391k	_16x16mv.as_mv.col &= x->fullpixel_mask;
335
336	391k	if (2 * _16x16mv.as_mv.col < (x->mb_to_left_edge - (19 << 3)) \|\|
337	391k	2 * _16x16mv.as_mv.col > x->mb_to_right_edge + (18 << 3) \|\|
338	391k	2 * _16x16mv.as_mv.row < (x->mb_to_top_edge - (19 << 3)) \|\|
339	391k	2 * _16x16mv.as_mv.row > x->mb_to_bottom_edge + (18 << 3)) {
340	863	return;
341	863	}
342
343	390k	pre_stride >>= 1;
344	390k	offset = (_16x16mv.as_mv.row >> 3) * pre_stride + (_16x16mv.as_mv.col >> 3);
345	390k	uptr = x->pre.u_buffer + offset;
346	390k	vptr = x->pre.v_buffer + offset;
347
348	390k	if (_16x16mv.as_int & 0x00070007) {
349	255k	x->subpixel_predict8x8(uptr, pre_stride, _16x16mv.as_mv.col & 7,
350	255k	_16x16mv.as_mv.row & 7, dst_u, dst_uvstride);
351	255k	x->subpixel_predict8x8(vptr, pre_stride, _16x16mv.as_mv.col & 7,
352	255k	_16x16mv.as_mv.row & 7, dst_v, dst_uvstride);
353	255k	} else {
354	135k	vp8_copy_mem8x8(uptr, pre_stride, dst_u, dst_uvstride);
355	135k	vp8_copy_mem8x8(vptr, pre_stride, dst_v, dst_uvstride);
356	135k	}
357	390k	}
358
359	113k	static void build_inter4x4_predictors_mb(MACROBLOCKD *x) {
360	113k	int i;
361	113k	unsigned char *base_dst = x->dst.y_buffer;
362	113k	unsigned char *base_pre = x->pre.y_buffer;
363
364	113k	if (x->mode_info_context->mbmi.partitioning < 3) {
365	74.8k	BLOCKD *b;
366	74.8k	int dst_stride = x->dst.y_stride;
367
368	74.8k	x->block[0].bmi = x->mode_info_context->bmi[0];
369	74.8k	x->block[2].bmi = x->mode_info_context->bmi[2];
370	74.8k	x->block[8].bmi = x->mode_info_context->bmi[8];
371	74.8k	x->block[10].bmi = x->mode_info_context->bmi[10];
372	74.8k	if (x->mode_info_context->mbmi.need_to_clamp_mvs) {
373	0	clamp_mv_to_umv_border(&x->block[0].bmi.mv.as_mv, x);
374	0	clamp_mv_to_umv_border(&x->block[2].bmi.mv.as_mv, x);
375	0	clamp_mv_to_umv_border(&x->block[8].bmi.mv.as_mv, x);
376	0	clamp_mv_to_umv_border(&x->block[10].bmi.mv.as_mv, x);
377	0	}
378
379	74.8k	b = &x->block[0];
380	74.8k	build_inter_predictors4b(x, b, base_dst + b->offset, dst_stride, base_pre,
381	74.8k	dst_stride);
382	74.8k	b = &x->block[2];
383	74.8k	build_inter_predictors4b(x, b, base_dst + b->offset, dst_stride, base_pre,
384	74.8k	dst_stride);
385	74.8k	b = &x->block[8];
386	74.8k	build_inter_predictors4b(x, b, base_dst + b->offset, dst_stride, base_pre,
387	74.8k	dst_stride);
388	74.8k	b = &x->block[10];
389	74.8k	build_inter_predictors4b(x, b, base_dst + b->offset, dst_stride, base_pre,
390	74.8k	dst_stride);
391	74.8k	} else {
392	348k	for (i = 0; i < 16; i += 2) {
393	309k	BLOCKD *d0 = &x->block[i];
394	309k	BLOCKD *d1 = &x->block[i + 1];
395	309k	int dst_stride = x->dst.y_stride;
396
397	309k	x->block[i + 0].bmi = x->mode_info_context->bmi[i + 0];
398	309k	x->block[i + 1].bmi = x->mode_info_context->bmi[i + 1];
399	309k	if (x->mode_info_context->mbmi.need_to_clamp_mvs) {
400	0	clamp_mv_to_umv_border(&x->block[i + 0].bmi.mv.as_mv, x);
401	0	clamp_mv_to_umv_border(&x->block[i + 1].bmi.mv.as_mv, x);
402	0	}
403
404	309k	if (d0->bmi.mv.as_int == d1->bmi.mv.as_int) {
405	159k	build_inter_predictors2b(x, d0, base_dst + d0->offset, dst_stride,
406	159k	base_pre, dst_stride);
407	159k	} else {
408	150k	build_inter_predictors_b(d0, base_dst + d0->offset, dst_stride,
409	150k	base_pre, dst_stride, x->subpixel_predict);
410	150k	build_inter_predictors_b(d1, base_dst + d1->offset, dst_stride,
411	150k	base_pre, dst_stride, x->subpixel_predict);
412	150k	}
413	309k	}
414	38.7k	}
415	113k	base_dst = x->dst.u_buffer;
416	113k	base_pre = x->pre.u_buffer;
417	340k	for (i = 16; i < 20; i += 2) {
418	227k	BLOCKD *d0 = &x->block[i];
419	227k	BLOCKD *d1 = &x->block[i + 1];
420	227k	int dst_stride = x->dst.uv_stride;
421
422		/* Note: uv mvs already clamped in build_4x4uvmvs() */
423
424	227k	if (d0->bmi.mv.as_int == d1->bmi.mv.as_int) {
425	129k	build_inter_predictors2b(x, d0, base_dst + d0->offset, dst_stride,
426	129k	base_pre, dst_stride);
427	129k	} else {
428	97.9k	build_inter_predictors_b(d0, base_dst + d0->offset, dst_stride, base_pre,
429	97.9k	dst_stride, x->subpixel_predict);
430	97.9k	build_inter_predictors_b(d1, base_dst + d1->offset, dst_stride, base_pre,
431	97.9k	dst_stride, x->subpixel_predict);
432	97.9k	}
433	227k	}
434
435	113k	base_dst = x->dst.v_buffer;
436	113k	base_pre = x->pre.v_buffer;
437	340k	for (i = 20; i < 24; i += 2) {
438	227k	BLOCKD *d0 = &x->block[i];
439	227k	BLOCKD *d1 = &x->block[i + 1];
440	227k	int dst_stride = x->dst.uv_stride;
441
442		/* Note: uv mvs already clamped in build_4x4uvmvs() */
443
444	227k	if (d0->bmi.mv.as_int == d1->bmi.mv.as_int) {
445	129k	build_inter_predictors2b(x, d0, base_dst + d0->offset, dst_stride,
446	129k	base_pre, dst_stride);
447	129k	} else {
448	97.9k	build_inter_predictors_b(d0, base_dst + d0->offset, dst_stride, base_pre,
449	97.9k	dst_stride, x->subpixel_predict);
450	97.9k	build_inter_predictors_b(d1, base_dst + d1->offset, dst_stride, base_pre,
451	97.9k	dst_stride, x->subpixel_predict);
452	97.9k	}
453	227k	}
454	113k	}
455
456	113k	static void build_4x4uvmvs(MACROBLOCKD *x) {
457	113k	int i, j;
458
459	340k	for (i = 0; i < 2; ++i) {
460	681k	for (j = 0; j < 2; ++j) {
461	454k	int yoffset = i * 8 + j * 2;
462	454k	int uoffset = 16 + i * 2 + j;
463	454k	int voffset = 20 + i * 2 + j;
464
465	454k	int temp;
466
467	454k	temp = x->mode_info_context->bmi[yoffset + 0].mv.as_mv.row +
468	454k	x->mode_info_context->bmi[yoffset + 1].mv.as_mv.row +
469	454k	x->mode_info_context->bmi[yoffset + 4].mv.as_mv.row +
470	454k	x->mode_info_context->bmi[yoffset + 5].mv.as_mv.row;
471
472	454k	temp += 4 + ((temp >> (sizeof(temp) * CHAR_BIT - 1)) * 8);
473
474	454k	x->block[uoffset].bmi.mv.as_mv.row = (temp / 8) & x->fullpixel_mask;
475
476	454k	temp = x->mode_info_context->bmi[yoffset + 0].mv.as_mv.col +
477	454k	x->mode_info_context->bmi[yoffset + 1].mv.as_mv.col +
478	454k	x->mode_info_context->bmi[yoffset + 4].mv.as_mv.col +
479	454k	x->mode_info_context->bmi[yoffset + 5].mv.as_mv.col;
480
481	454k	temp += 4 + ((temp >> (sizeof(temp) * CHAR_BIT - 1)) * 8);
482
483	454k	x->block[uoffset].bmi.mv.as_mv.col = (temp / 8) & x->fullpixel_mask;
484
485	454k	if (x->mode_info_context->mbmi.need_to_clamp_mvs) {
486	0	clamp_uvmv_to_umv_border(&x->block[uoffset].bmi.mv.as_mv, x);
487	0	}
488
489	454k	x->block[voffset].bmi.mv.as_int = x->block[uoffset].bmi.mv.as_int;
490	454k	}
491	227k	}
492	113k	}
493
494	504k	void vp8_build_inter_predictors_mb(MACROBLOCKD *xd) {
495	504k	if (xd->mode_info_context->mbmi.mode != SPLITMV) {
496	391k	vp8_build_inter16x16_predictors_mb(xd, xd->dst.y_buffer, xd->dst.u_buffer,
497	391k	xd->dst.v_buffer, xd->dst.y_stride,
498	391k	xd->dst.uv_stride);
499	391k	} else {
500	113k	build_4x4uvmvs(xd);
501	113k	build_inter4x4_predictors_mb(xd);
502	113k	}
503	504k	}