vp9_reconinter.c 12.8 KB
Newer Older
John Koleszar's avatar
John Koleszar committed
1
/*
2
 *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
John Koleszar's avatar
John Koleszar committed
3
 *
4
 *  Use of this source code is governed by a BSD-style license
5 6
 *  that can be found in the LICENSE file in the root of the source
 *  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.
John Koleszar's avatar
John Koleszar committed
9 10
 */

11
#include <assert.h>
John Koleszar's avatar
John Koleszar committed
12

13
#include "./vpx_scale_rtcd.h"
14
#include "./vpx_config.h"
15

16
#include "vpx/vpx_integer.h"
17

18 19
#include "vp9/common/vp9_blockd.h"
#include "vp9/common/vp9_reconinter.h"
20
#include "vp9/common/vp9_reconintra.h"
21

22
#if CONFIG_VP9_HIGHBITDEPTH
23
void high_inter_predictor(const uint8_t *src, int src_stride,
24 25 26 27 28 29 30
                                 uint8_t *dst, int dst_stride,
                                 const int subpel_x,
                                 const int subpel_y,
                                 const struct scale_factors *sf,
                                 int w, int h, int ref,
                                 const InterpKernel *kernel,
                                 int xs, int ys, int bd) {
31
  sf->highbd_predict[subpel_x != 0][subpel_y != 0][ref](
32 33 34 35
      src, src_stride, dst, dst_stride,
      kernel[subpel_x], xs, kernel[subpel_y], ys, w, h, bd);
}

36 37 38 39 40 41 42 43
void vp9_highbd_build_inter_predictor(const uint8_t *src, int src_stride,
                                      uint8_t *dst, int dst_stride,
                                      const MV *src_mv,
                                      const struct scale_factors *sf,
                                      int w, int h, int ref,
                                      const InterpKernel *kernel,
                                      enum mv_precision precision,
                                      int x, int y, int bd) {
44 45 46 47 48 49 50 51 52 53 54 55 56 57
  const int is_q4 = precision == MV_PRECISION_Q4;
  const MV mv_q4 = { is_q4 ? src_mv->row : src_mv->row * 2,
                     is_q4 ? src_mv->col : src_mv->col * 2 };
  MV32 mv = vp9_scale_mv(&mv_q4, x, y, sf);
  const int subpel_x = mv.col & SUBPEL_MASK;
  const int subpel_y = mv.row & SUBPEL_MASK;

  src += (mv.row >> SUBPEL_BITS) * src_stride + (mv.col >> SUBPEL_BITS);

  high_inter_predictor(src, src_stride, dst, dst_stride, subpel_x, subpel_y,
                       sf, w, h, ref, kernel, sf->x_step_q4, sf->y_step_q4, bd);
}
#endif  // CONFIG_VP9_HIGHBITDEPTH

58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78
void vp9_build_inter_predictor(const uint8_t *src, int src_stride,
                               uint8_t *dst, int dst_stride,
                               const MV *src_mv,
                               const struct scale_factors *sf,
                               int w, int h, int ref,
                               const InterpKernel *kernel,
                               enum mv_precision precision,
                               int x, int y) {
  const int is_q4 = precision == MV_PRECISION_Q4;
  const MV mv_q4 = { is_q4 ? src_mv->row : src_mv->row * 2,
                     is_q4 ? src_mv->col : src_mv->col * 2 };
  MV32 mv = vp9_scale_mv(&mv_q4, x, y, sf);
  const int subpel_x = mv.col & SUBPEL_MASK;
  const int subpel_y = mv.row & SUBPEL_MASK;

  src += (mv.row >> SUBPEL_BITS) * src_stride + (mv.col >> SUBPEL_BITS);

  inter_predictor(src, src_stride, dst, dst_stride, subpel_x, subpel_y,
                  sf, w, h, ref, kernel, sf->x_step_q4, sf->y_step_q4);
}

79 80
static INLINE int round_mv_comp_q4(int value) {
  return (value < 0 ? value - 2 : value + 2) / 4;
John Koleszar's avatar
John Koleszar committed
81 82
}

83 84 85 86 87 88 89 90 91 92
static MV mi_mv_pred_q4(const MODE_INFO *mi, int idx) {
  MV res = { round_mv_comp_q4(mi->bmi[0].as_mv[idx].as_mv.row +
                              mi->bmi[1].as_mv[idx].as_mv.row +
                              mi->bmi[2].as_mv[idx].as_mv.row +
                              mi->bmi[3].as_mv[idx].as_mv.row),
             round_mv_comp_q4(mi->bmi[0].as_mv[idx].as_mv.col +
                              mi->bmi[1].as_mv[idx].as_mv.col +
                              mi->bmi[2].as_mv[idx].as_mv.col +
                              mi->bmi[3].as_mv[idx].as_mv.col) };
  return res;
93 94
}

95 96 97 98 99 100 101 102 103 104 105 106
static INLINE int round_mv_comp_q2(int value) {
  return (value < 0 ? value - 1 : value + 1) / 2;
}

static MV mi_mv_pred_q2(const MODE_INFO *mi, int idx, int block0, int block1) {
  MV res = { round_mv_comp_q2(mi->bmi[block0].as_mv[idx].as_mv.row +
                              mi->bmi[block1].as_mv[idx].as_mv.row),
             round_mv_comp_q2(mi->bmi[block0].as_mv[idx].as_mv.col +
                              mi->bmi[block1].as_mv[idx].as_mv.col) };
  return res;
}

107
// TODO(jkoleszar): yet another mv clamping function :-(
108 109
MV clamp_mv_to_umv_border_sb(const MACROBLOCKD *xd, const MV *src_mv,
                             int bw, int bh, int ss_x, int ss_y) {
110 111 112
  // 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.
113 114 115 116
  const int spel_left = (VP9_INTERP_EXTEND + bw) << SUBPEL_BITS;
  const int spel_right = spel_left - SUBPEL_SHIFTS;
  const int spel_top = (VP9_INTERP_EXTEND + bh) << SUBPEL_BITS;
  const int spel_bottom = spel_top - SUBPEL_SHIFTS;
117
  MV clamped_mv = {
Yaowu Xu's avatar
Yaowu Xu committed
118 119
    src_mv->row * (1 << (1 - ss_y)),
    src_mv->col * (1 << (1 - ss_x))
120
  };
121 122
  assert(ss_x <= 1);
  assert(ss_y <= 1);
123

Yaowu Xu's avatar
Yaowu Xu committed
124 125 126 127 128
  clamp_mv(&clamped_mv,
           xd->mb_to_left_edge * (1 << (1 - ss_x)) - spel_left,
           xd->mb_to_right_edge * (1 << (1 - ss_x)) + spel_right,
           xd->mb_to_top_edge * (1 << (1 - ss_y)) - spel_top,
           xd->mb_to_bottom_edge * (1 << (1 - ss_y)) + spel_bottom);
129

130 131 132
  return clamped_mv;
}

133 134
MV average_split_mvs(const struct macroblockd_plane *pd,
                     const MODE_INFO *mi, int ref, int block) {
135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150
  const int ss_idx = ((pd->subsampling_x > 0) << 1) | (pd->subsampling_y > 0);
  MV res = {0, 0};
  switch (ss_idx) {
    case 0:
      res = mi->bmi[block].as_mv[ref].as_mv;
      break;
    case 1:
      res = mi_mv_pred_q2(mi, ref, block, block + 2);
      break;
    case 2:
      res = mi_mv_pred_q2(mi, ref, block, block + 1);
      break;
    case 3:
      res = mi_mv_pred_q4(mi, ref);
      break;
    default:
151
      assert(ss_idx <= 3 && ss_idx >= 0);
152 153 154 155
  }
  return res;
}

156
static void build_inter_predictors(MACROBLOCKD *xd, int plane, int block,
157 158
                                   int bw, int bh,
                                   int x, int y, int w, int h,
159
                                   int mi_x, int mi_y) {
160
  struct macroblockd_plane *const pd = &xd->plane[plane];
161
  const MODE_INFO *mi = xd->mi[0];
162 163
  const int is_compound = has_second_ref(mi);
  const InterpKernel *kernel = vp9_filter_kernels[mi->interp_filter];
164
  int ref;
165

166
  for (ref = 0; ref < 1 + is_compound; ++ref) {
167
    const struct scale_factors *const sf = &xd->block_refs[ref]->sf;
168 169
    struct buf_2d *const pre_buf = &pd->pre[ref];
    struct buf_2d *const dst_buf = &pd->dst;
170
    uint8_t *const dst = dst_buf->buf + dst_buf->stride * y + x;
171
    const MV mv = mi->sb_type < BLOCK_8X8
Yaowu Xu's avatar
Yaowu Xu committed
172
               ? average_split_mvs(pd, mi, ref, block)
173
               : mi->mv[ref].as_mv;
174 175 176 177 178

    // TODO(jkoleszar): This clamping is done in the incorrect place for the
    // scaling case. It needs to be done on the scaled MV, not the pre-scaling
    // MV. Note however that it performs the subsampling aware scaling so
    // that the result is always q4.
179 180 181 182
    // mv_precision precision is MV_PRECISION_Q4.
    const MV mv_q4 = clamp_mv_to_umv_border_sb(xd, &mv, bw, bh,
                                               pd->subsampling_x,
                                               pd->subsampling_y);
183

184 185
    uint8_t *pre;
    MV32 scaled_mv;
186
    int xs, ys, subpel_x, subpel_y;
187
    const int is_scaled = vp9_is_scaled(sf);
188

189
    if (is_scaled) {
190 191 192
      // Co-ordinate of containing block to pixel precision.
      const int x_start = (-xd->mb_to_left_edge >> (3 + pd->subsampling_x));
      const int y_start = (-xd->mb_to_top_edge >> (3 + pd->subsampling_y));
193
#if CONFIG_BETTER_HW_COMPATIBILITY
194 195
      assert(xd->mi[0]->sb_type != BLOCK_4X8 &&
             xd->mi[0]->sb_type != BLOCK_8X4);
196 197
      assert(mv_q4.row == mv.row * (1 << (1 - pd->subsampling_y)) &&
             mv_q4.col == mv.col * (1 << (1 - pd->subsampling_x)));
198
#endif
199 200 201 202 203 204 205 206 207 208
      if (plane == 0)
        pre_buf->buf = xd->block_refs[ref]->buf->y_buffer;
      else if (plane == 1)
        pre_buf->buf = xd->block_refs[ref]->buf->u_buffer;
      else
        pre_buf->buf = xd->block_refs[ref]->buf->v_buffer;

      pre_buf->buf += scaled_buffer_offset(x_start + x, y_start + y,
                                           pre_buf->stride, sf);
      pre = pre_buf->buf;
209 210 211
      scaled_mv = vp9_scale_mv(&mv_q4, mi_x + x, mi_y + y, sf);
      xs = sf->x_step_q4;
      ys = sf->y_step_q4;
212 213 214 215 216 217
    } else {
      pre = pre_buf->buf + (y * pre_buf->stride + x);
      scaled_mv.row = mv_q4.row;
      scaled_mv.col = mv_q4.col;
      xs = ys = 16;
    }
218 219 220 221
    subpel_x = scaled_mv.col & SUBPEL_MASK;
    subpel_y = scaled_mv.row & SUBPEL_MASK;
    pre += (scaled_mv.row >> SUBPEL_BITS) * pre_buf->stride
           + (scaled_mv.col >> SUBPEL_BITS);
222

223 224 225 226 227 228 229 230 231 232
#if CONFIG_VP9_HIGHBITDEPTH
    if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
      high_inter_predictor(pre, pre_buf->stride, dst, dst_buf->stride,
                           subpel_x, subpel_y, sf, w, h, ref, kernel, xs, ys,
                           xd->bd);
    } else {
      inter_predictor(pre, pre_buf->stride, dst, dst_buf->stride,
                      subpel_x, subpel_y, sf, w, h, ref, kernel, xs, ys);
    }
#else
233
    inter_predictor(pre, pre_buf->stride, dst, dst_buf->stride,
234
                    subpel_x, subpel_y, sf, w, h, ref, kernel, xs, ys);
235
#endif  // CONFIG_VP9_HIGHBITDEPTH
236 237
  }
}
238

239
static void build_inter_predictors_for_planes(MACROBLOCKD *xd, BLOCK_SIZE bsize,
240 241 242
                                              int mi_row, int mi_col,
                                              int plane_from, int plane_to) {
  int plane;
243 244
  const int mi_x = mi_col * MI_SIZE;
  const int mi_y = mi_row * MI_SIZE;
245
  for (plane = plane_from; plane <= plane_to; ++plane) {
246 247 248 249 250 251
    const BLOCK_SIZE plane_bsize = get_plane_block_size(bsize,
                                                        &xd->plane[plane]);
    const int num_4x4_w = num_4x4_blocks_wide_lookup[plane_bsize];
    const int num_4x4_h = num_4x4_blocks_high_lookup[plane_bsize];
    const int bw = 4 * num_4x4_w;
    const int bh = 4 * num_4x4_h;
252

253
    if (xd->mi[0]->sb_type < BLOCK_8X8) {
254 255
      int i = 0, x, y;
      assert(bsize == BLOCK_8X8);
256 257 258 259
      for (y = 0; y < num_4x4_h; ++y)
        for (x = 0; x < num_4x4_w; ++x)
           build_inter_predictors(xd, plane, i++, bw, bh,
                                  4 * x, 4 * y, 4, 4, mi_x, mi_y);
260
    } else {
261 262
      build_inter_predictors(xd, plane, 0, bw, bh,
                             0, 0, bw, bh, mi_x, mi_y);
263
    }
264 265 266
  }
}

267
void vp9_build_inter_predictors_sby(MACROBLOCKD *xd, int mi_row, int mi_col,
268
                                    BLOCK_SIZE bsize) {
269
  build_inter_predictors_for_planes(xd, bsize, mi_row, mi_col, 0, 0);
270
}
271 272 273 274 275 276

void vp9_build_inter_predictors_sbp(MACROBLOCKD *xd, int mi_row, int mi_col,
                                    BLOCK_SIZE bsize, int plane) {
  build_inter_predictors_for_planes(xd, bsize, mi_row, mi_col, plane, plane);
}

277
void vp9_build_inter_predictors_sbuv(MACROBLOCKD *xd, int mi_row, int mi_col,
278
                                     BLOCK_SIZE bsize) {
279 280
  build_inter_predictors_for_planes(xd, bsize, mi_row, mi_col, 1,
                                    MAX_MB_PLANE - 1);
281
}
282

283
void vp9_build_inter_predictors_sb(MACROBLOCKD *xd, int mi_row, int mi_col,
284
                                   BLOCK_SIZE bsize) {
285 286
  build_inter_predictors_for_planes(xd, bsize, mi_row, mi_col, 0,
                                    MAX_MB_PLANE - 1);
287
}
288

289
void vp9_setup_dst_planes(struct macroblockd_plane planes[MAX_MB_PLANE],
290 291
                          const YV12_BUFFER_CONFIG *src,
                          int mi_row, int mi_col) {
Yaowu Xu's avatar
Yaowu Xu committed
292 293 294 295
  uint8_t *const buffers[MAX_MB_PLANE] = { src->y_buffer, src->u_buffer,
      src->v_buffer};
  const int strides[MAX_MB_PLANE] = { src->y_stride, src->uv_stride,
      src->uv_stride};
296 297 298
  int i;

  for (i = 0; i < MAX_MB_PLANE; ++i) {
299
    struct macroblockd_plane *const pd = &planes[i];
300 301 302 303 304 305 306 307 308 309 310
    setup_pred_plane(&pd->dst, buffers[i], strides[i], mi_row, mi_col, NULL,
                     pd->subsampling_x, pd->subsampling_y);
  }
}

void vp9_setup_pre_planes(MACROBLOCKD *xd, int idx,
                          const YV12_BUFFER_CONFIG *src,
                          int mi_row, int mi_col,
                          const struct scale_factors *sf) {
  if (src != NULL) {
    int i;
Yaowu Xu's avatar
Yaowu Xu committed
311 312 313 314
    uint8_t *const buffers[MAX_MB_PLANE] = { src->y_buffer, src->u_buffer,
        src->v_buffer};
    const int strides[MAX_MB_PLANE] = { src->y_stride, src->uv_stride,
        src->uv_stride};
315 316 317 318 319 320 321
    for (i = 0; i < MAX_MB_PLANE; ++i) {
      struct macroblockd_plane *const pd = &xd->plane[i];
      setup_pred_plane(&pd->pre[idx], buffers[i], strides[i], mi_row, mi_col,
                       sf, pd->subsampling_x, pd->subsampling_y);
    }
  }
}