decodeframe.c 152 KB
Newer Older
Jingning Han's avatar
Jingning Han committed
1
/*
2
 * Copyright (c) 2016, Alliance for Open Media. All rights reserved
Jingning Han's avatar
Jingning Han committed
3
 *
4 5 6 7 8 9
 * This source code is subject to the terms of the BSD 2 Clause License and
 * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
 * was not distributed with this source code in the LICENSE file, you can
 * obtain it at www.aomedia.org/license/software. If the Alliance for Open
 * Media Patent License 1.0 was not distributed with this source code in the
 * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
Jingning Han's avatar
Jingning Han committed
10 11 12 13 14
 */

#include <assert.h>
#include <stdlib.h>  // qsort()

15 16 17
#include "./aom_config.h"
#include "./aom_dsp_rtcd.h"
#include "./aom_scale_rtcd.h"
Jingning Han's avatar
Jingning Han committed
18
#include "./av1_rtcd.h"
Jingning Han's avatar
Jingning Han committed
19

20
#include "aom/aom_codec.h"
21
#include "aom_dsp/aom_dsp_common.h"
Jingning Han's avatar
Jingning Han committed
22 23
#include "aom_dsp/bitreader.h"
#include "aom_dsp/bitreader_buffer.h"
24
#include "aom_mem/aom_mem.h"
25 26
#include "aom_ports/mem.h"
#include "aom_ports/mem_ops.h"
27 28
#include "aom_scale/aom_scale.h"
#include "aom_util/aom_thread.h"
Jingning Han's avatar
Jingning Han committed
29

30
#include "av1/common/alloccommon.h"
31
#if CONFIG_CLPF
Steinar Midtskogen's avatar
Steinar Midtskogen committed
32
#include "aom/aom_image.h"
33
#include "av1/common/clpf.h"
34
#endif
35
#include "av1/common/common.h"
Yaowu Xu's avatar
Yaowu Xu committed
36
#if CONFIG_DERING
37
#include "av1/common/dering.h"
Yaowu Xu's avatar
Yaowu Xu committed
38
#endif  // CONFIG_DERING
39 40 41 42 43 44
#include "av1/common/entropy.h"
#include "av1/common/entropymode.h"
#include "av1/common/idct.h"
#include "av1/common/pred_common.h"
#include "av1/common/quant_common.h"
#include "av1/common/reconinter.h"
Jingning Han's avatar
Jingning Han committed
45
#include "av1/common/reconintra.h"
46
#include "av1/common/seg_common.h"
Jingning Han's avatar
Jingning Han committed
47
#include "av1/common/thread_common.h"
48
#include "av1/common/tile_common.h"
49

50 51 52
#include "av1/decoder/decodeframe.h"
#include "av1/decoder/decodemv.h"
#include "av1/decoder/decoder.h"
Jingning Han's avatar
Jingning Han committed
53
#include "av1/decoder/detokenize.h"
54
#include "av1/decoder/dsubexp.h"
Jingning Han's avatar
Jingning Han committed
55

56
#define MAX_AV1_HEADER_SIZE 80
Jingning Han's avatar
Jingning Han committed
57

58
static int is_compound_reference_allowed(const AV1_COMMON *cm) {
Jingning Han's avatar
Jingning Han committed
59
  int i;
60
  if (frame_is_intra_only(cm)) return 0;
61
  for (i = 1; i < INTER_REFS_PER_FRAME; ++i)
62
    if (cm->ref_frame_sign_bias[i + 1] != cm->ref_frame_sign_bias[1]) return 1;
Jingning Han's avatar
Jingning Han committed
63 64 65 66

  return 0;
}

67
static void setup_compound_reference_mode(AV1_COMMON *cm) {
68
#if CONFIG_EXT_REFS
69
  cm->comp_fwd_ref[0] = LAST_FRAME;
70 71 72 73
  cm->comp_fwd_ref[1] = LAST2_FRAME;
  cm->comp_fwd_ref[2] = LAST3_FRAME;
  cm->comp_fwd_ref[3] = GOLDEN_FRAME;

74 75
  cm->comp_bwd_ref[0] = BWDREF_FRAME;
  cm->comp_bwd_ref[1] = ALTREF_FRAME;
76
#else
Jingning Han's avatar
Jingning Han committed
77
  if (cm->ref_frame_sign_bias[LAST_FRAME] ==
78
      cm->ref_frame_sign_bias[GOLDEN_FRAME]) {
Jingning Han's avatar
Jingning Han committed
79 80 81 82
    cm->comp_fixed_ref = ALTREF_FRAME;
    cm->comp_var_ref[0] = LAST_FRAME;
    cm->comp_var_ref[1] = GOLDEN_FRAME;
  } else if (cm->ref_frame_sign_bias[LAST_FRAME] ==
83
             cm->ref_frame_sign_bias[ALTREF_FRAME]) {
Jingning Han's avatar
Jingning Han committed
84 85 86 87 88 89 90 91
    cm->comp_fixed_ref = GOLDEN_FRAME;
    cm->comp_var_ref[0] = LAST_FRAME;
    cm->comp_var_ref[1] = ALTREF_FRAME;
  } else {
    cm->comp_fixed_ref = LAST_FRAME;
    cm->comp_var_ref[0] = GOLDEN_FRAME;
    cm->comp_var_ref[1] = ALTREF_FRAME;
  }
92
#endif  // CONFIG_EXT_REFS
Jingning Han's avatar
Jingning Han committed
93 94 95 96 97 98
}

static int read_is_valid(const uint8_t *start, size_t len, const uint8_t *end) {
  return len != 0 && len <= (size_t)(end - start);
}

99 100
static int decode_unsigned_max(struct aom_read_bit_buffer *rb, int max) {
  const int data = aom_rb_read_literal(rb, get_unsigned_bits(max));
Jingning Han's avatar
Jingning Han committed
101 102 103
  return data > max ? max : data;
}

104 105
static TX_MODE read_tx_mode(struct aom_read_bit_buffer *rb) {
  return aom_rb_read_bit(rb) ? TX_MODE_SELECT : aom_rb_read_literal(rb, 2);
106
}
Jingning Han's avatar
Jingning Han committed
107

108
static void read_switchable_interp_probs(FRAME_CONTEXT *fc, aom_reader *r) {
Jingning Han's avatar
Jingning Han committed
109 110 111
  int i, j;
  for (j = 0; j < SWITCHABLE_FILTER_CONTEXTS; ++j)
    for (i = 0; i < SWITCHABLE_FILTERS - 1; ++i)
112
      av1_diff_update_prob(r, &fc->switchable_interp_prob[j][i]);
Jingning Han's avatar
Jingning Han committed
113 114
}

115
static void read_inter_mode_probs(FRAME_CONTEXT *fc, aom_reader *r) {
116 117 118
  int i;
#if CONFIG_REF_MV
  for (i = 0; i < NEWMV_MODE_CONTEXTS; ++i)
119
    av1_diff_update_prob(r, &fc->newmv_prob[i]);
120
  for (i = 0; i < ZEROMV_MODE_CONTEXTS; ++i)
121
    av1_diff_update_prob(r, &fc->zeromv_prob[i]);
122
  for (i = 0; i < REFMV_MODE_CONTEXTS; ++i)
123
    av1_diff_update_prob(r, &fc->refmv_prob[i]);
124
  for (i = 0; i < DRL_MODE_CONTEXTS; ++i)
125
    av1_diff_update_prob(r, &fc->drl_prob[i]);
Yue Chen's avatar
Yue Chen committed
126
#if CONFIG_EXT_INTER
127
  av1_diff_update_prob(r, &fc->new2mv_prob);
Yue Chen's avatar
Yue Chen committed
128
#endif  // CONFIG_EXT_INTER
129 130
#else
  int j;
Jingning Han's avatar
Jingning Han committed
131 132
  for (i = 0; i < INTER_MODE_CONTEXTS; ++i)
    for (j = 0; j < INTER_MODES - 1; ++j)
133
      av1_diff_update_prob(r, &fc->inter_mode_probs[i][j]);
134
#endif
Jingning Han's avatar
Jingning Han committed
135 136
}

137
#if CONFIG_EXT_INTER
138
static void read_inter_compound_mode_probs(FRAME_CONTEXT *fc, aom_reader *r) {
139
  int i, j;
140
  if (aom_read(r, GROUP_DIFF_UPDATE_PROB)) {
141 142
    for (j = 0; j < INTER_MODE_CONTEXTS; ++j) {
      for (i = 0; i < INTER_COMPOUND_MODES - 1; ++i) {
143
        av1_diff_update_prob(r, &fc->inter_compound_mode_probs[j][i]);
144 145 146 147 148 149
      }
    }
  }
}
#endif  // CONFIG_EXT_INTER

150
static REFERENCE_MODE read_frame_reference_mode(
151
    const AV1_COMMON *cm, struct aom_read_bit_buffer *rb) {
152
  if (is_compound_reference_allowed(cm)) {
153
    return aom_rb_read_bit(rb)
154
               ? REFERENCE_MODE_SELECT
155
               : (aom_rb_read_bit(rb) ? COMPOUND_REFERENCE : SINGLE_REFERENCE);
156 157 158 159
  } else {
    return SINGLE_REFERENCE;
  }
}
Jingning Han's avatar
Jingning Han committed
160

161
static void read_frame_reference_mode_probs(AV1_COMMON *cm, aom_reader *r) {
Jingning Han's avatar
Jingning Han committed
162
  FRAME_CONTEXT *const fc = cm->fc;
163
  int i, j;
Jingning Han's avatar
Jingning Han committed
164 165 166

  if (cm->reference_mode == REFERENCE_MODE_SELECT)
    for (i = 0; i < COMP_INTER_CONTEXTS; ++i)
167
      av1_diff_update_prob(r, &fc->comp_inter_prob[i]);
Jingning Han's avatar
Jingning Han committed
168

169
  if (cm->reference_mode != COMPOUND_REFERENCE) {
Jingning Han's avatar
Jingning Han committed
170
    for (i = 0; i < REF_CONTEXTS; ++i) {
171
      for (j = 0; j < (SINGLE_REFS - 1); ++j) {
172
        av1_diff_update_prob(r, &fc->single_ref_prob[i][j]);
173
      }
Jingning Han's avatar
Jingning Han committed
174
    }
175
  }
Jingning Han's avatar
Jingning Han committed
176

177 178
  if (cm->reference_mode != SINGLE_REFERENCE) {
    for (i = 0; i < REF_CONTEXTS; ++i) {
179
#if CONFIG_EXT_REFS
180
      for (j = 0; j < (FWD_REFS - 1); ++j)
181
        av1_diff_update_prob(r, &fc->comp_ref_prob[i][j]);
182
      for (j = 0; j < (BWD_REFS - 1); ++j)
183
        av1_diff_update_prob(r, &fc->comp_bwdref_prob[i][j]);
184 185
#else
      for (j = 0; j < (COMP_REFS - 1); ++j)
186
        av1_diff_update_prob(r, &fc->comp_ref_prob[i][j]);
187
#endif  // CONFIG_EXT_REFS
188 189
    }
  }
Jingning Han's avatar
Jingning Han committed
190 191
}

192
static void update_mv_probs(aom_prob *p, int n, aom_reader *r) {
Jingning Han's avatar
Jingning Han committed
193
  int i;
194
  for (i = 0; i < n; ++i) av1_diff_update_prob(r, &p[i]);
Jingning Han's avatar
Jingning Han committed
195 196
}

197
static void read_mv_probs(nmv_context *ctx, int allow_hp, aom_reader *r) {
Jingning Han's avatar
Jingning Han committed
198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225
  int i, j;

  update_mv_probs(ctx->joints, MV_JOINTS - 1, r);

  for (i = 0; i < 2; ++i) {
    nmv_component *const comp_ctx = &ctx->comps[i];
    update_mv_probs(&comp_ctx->sign, 1, r);
    update_mv_probs(comp_ctx->classes, MV_CLASSES - 1, r);
    update_mv_probs(comp_ctx->class0, CLASS0_SIZE - 1, r);
    update_mv_probs(comp_ctx->bits, MV_OFFSET_BITS, r);
  }

  for (i = 0; i < 2; ++i) {
    nmv_component *const comp_ctx = &ctx->comps[i];
    for (j = 0; j < CLASS0_SIZE; ++j)
      update_mv_probs(comp_ctx->class0_fp[j], MV_FP_SIZE - 1, r);
    update_mv_probs(comp_ctx->fp, 3, r);
  }

  if (allow_hp) {
    for (i = 0; i < 2; ++i) {
      nmv_component *const comp_ctx = &ctx->comps[i];
      update_mv_probs(&comp_ctx->class0_hp, 1, r);
      update_mv_probs(&comp_ctx->hp, 1, r);
    }
  }
}

226
static void inverse_transform_block(MACROBLOCKD *xd, int plane,
227
                                    const TX_TYPE tx_type,
228 229
                                    const TX_SIZE tx_size, uint8_t *dst,
                                    int stride, int eob) {
Jingning Han's avatar
Jingning Han committed
230 231 232
  struct macroblockd_plane *const pd = &xd->plane[plane];
  if (eob > 0) {
    tran_low_t *const dqcoeff = pd->dqcoeff;
Angie Chiang's avatar
Angie Chiang committed
233 234 235 236
    INV_TXFM_PARAM inv_txfm_param;
    inv_txfm_param.tx_type = tx_type;
    inv_txfm_param.tx_size = tx_size;
    inv_txfm_param.eob = eob;
237
    inv_txfm_param.lossless = xd->lossless[xd->mi[0]->mbmi.segment_id];
Angie Chiang's avatar
Angie Chiang committed
238

239
#if CONFIG_AOM_HIGHBITDEPTH
Jingning Han's avatar
Jingning Han committed
240
    if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
Angie Chiang's avatar
Angie Chiang committed
241 242
      inv_txfm_param.bd = xd->bd;
      highbd_inv_txfm_add(dqcoeff, dst, stride, &inv_txfm_param);
Jingning Han's avatar
Jingning Han committed
243
    } else {
244
#endif  // CONFIG_AOM_HIGHBITDEPTH
Angie Chiang's avatar
Angie Chiang committed
245
      inv_txfm_add(dqcoeff, dst, stride, &inv_txfm_param);
246
#if CONFIG_AOM_HIGHBITDEPTH
Jingning Han's avatar
Jingning Han committed
247
    }
248
#endif  // CONFIG_AOM_HIGHBITDEPTH
Jingning Han's avatar
Jingning Han committed
249 250 251 252

    if (eob == 1) {
      dqcoeff[0] = 0;
    } else {
253
      if (tx_type == DCT_DCT && tx_size <= TX_16X16 && eob <= 10)
254
        memset(dqcoeff, 0, 4 * 4 * num_4x4_blocks_wide_txsize_lookup[tx_size] *
255
                               sizeof(dqcoeff[0]));
256 257
#if CONFIG_EXT_TX
      else
258
        memset(dqcoeff, 0, get_tx2d_size(tx_size) * sizeof(dqcoeff[0]));
259
#else
Jingning Han's avatar
Jingning Han committed
260 261 262
      else if (tx_size == TX_32X32 && eob <= 34)
        memset(dqcoeff, 0, 256 * sizeof(dqcoeff[0]));
      else
263
        memset(dqcoeff, 0, get_tx2d_size(tx_size) * sizeof(dqcoeff[0]));
264
#endif
Jingning Han's avatar
Jingning Han committed
265 266 267 268 269
    }
  }
}

static void predict_and_reconstruct_intra_block(MACROBLOCKD *const xd,
Alex Converse's avatar
Alex Converse committed
270 271 272
#if CONFIG_ANS
                                                struct AnsDecoder *const r,
#else
273
                                                aom_reader *r,
Alex Converse's avatar
Alex Converse committed
274
#endif  // CONFIG_ANS
Jingning Han's avatar
Jingning Han committed
275
                                                MB_MODE_INFO *const mbmi,
276
                                                int plane, int row, int col,
Jingning Han's avatar
Jingning Han committed
277 278 279
                                                TX_SIZE tx_size) {
  struct macroblockd_plane *const pd = &xd->plane[plane];
  PREDICTION_MODE mode = (plane == 0) ? mbmi->mode : mbmi->uv_mode;
hui su's avatar
hui su committed
280
  PLANE_TYPE plane_type = (plane == 0) ? PLANE_TYPE_Y : PLANE_TYPE_UV;
Jingning Han's avatar
Jingning Han committed
281
  uint8_t *dst;
hui su's avatar
hui su committed
282
  int block_idx = (row << 1) + col;
Jingning Han's avatar
Jingning Han committed
283 284 285
  dst = &pd->dst.buf[4 * row * pd->dst.stride + 4 * col];

  if (mbmi->sb_type < BLOCK_8X8)
286
    if (plane == 0) mode = xd->mi[0]->bmi[(row << 1) + col].as_mode;
Jingning Han's avatar
Jingning Han committed
287

288 289
  av1_predict_intra_block(xd, pd->n4_wl, pd->n4_hl, tx_size, mode, dst,
                          pd->dst.stride, dst, pd->dst.stride, col, row, plane);
Jingning Han's avatar
Jingning Han committed
290 291

  if (!mbmi->skip) {
hui su's avatar
hui su committed
292
    TX_TYPE tx_type = get_tx_type(plane_type, xd, block_idx, tx_size);
293
    const scan_order *sc = get_scan(tx_size, tx_type, 0);
294 295
    const int eob = av1_decode_block_tokens(xd, plane, sc, col, row, tx_size,
                                            tx_type, r, mbmi->segment_id);
296 297
    inverse_transform_block(xd, plane, tx_type, tx_size, dst, pd->dst.stride,
                            eob);
Jingning Han's avatar
Jingning Han committed
298 299 300
  }
}

301
#if CONFIG_VAR_TX
302
static void decode_reconstruct_tx(MACROBLOCKD *const xd, aom_reader *r,
303 304 305 306
                                  MB_MODE_INFO *const mbmi, int plane,
                                  BLOCK_SIZE plane_bsize, int block,
                                  int blk_row, int blk_col, TX_SIZE tx_size,
                                  int *eob_total) {
307
  const struct macroblockd_plane *const pd = &xd->plane[plane];
308
  const BLOCK_SIZE bsize = txsize_to_bsize[tx_size];
309 310
  const int tx_row = blk_row >> (1 - pd->subsampling_y);
  const int tx_col = blk_col >> (1 - pd->subsampling_x);
311
  const TX_SIZE plane_tx_size =
312
      plane ? uv_txsize_lookup[bsize][mbmi->inter_tx_size[tx_row][tx_col]][0][0]
313
            : mbmi->inter_tx_size[tx_row][tx_col];
314 315 316 317 318 319 320 321
  int max_blocks_high = num_4x4_blocks_high_lookup[plane_bsize];
  int max_blocks_wide = num_4x4_blocks_wide_lookup[plane_bsize];

  if (xd->mb_to_bottom_edge < 0)
    max_blocks_high += xd->mb_to_bottom_edge >> (5 + pd->subsampling_y);
  if (xd->mb_to_right_edge < 0)
    max_blocks_wide += xd->mb_to_right_edge >> (5 + pd->subsampling_x);

322
  if (blk_row >= max_blocks_high || blk_col >= max_blocks_wide) return;
323

324
  if (tx_size == plane_tx_size) {
325
    PLANE_TYPE plane_type = (plane == 0) ? PLANE_TYPE_Y : PLANE_TYPE_UV;
326 327
    TX_TYPE tx_type = get_tx_type(plane_type, xd, block, plane_tx_size);
    const scan_order *sc = get_scan(plane_tx_size, tx_type, 1);
328
    const int eob =
329 330
        av1_decode_block_tokens(xd, plane, sc, blk_col, blk_row, plane_tx_size,
                                tx_type, r, mbmi->segment_id);
331 332
    inverse_transform_block(
        xd, plane, tx_type, plane_tx_size,
333
        &pd->dst.buf[4 * blk_row * pd->dst.stride + 4 * blk_col],
334
        pd->dst.stride, eob);
335 336 337 338 339 340 341 342 343 344 345
    *eob_total += eob;
  } else {
    int bsl = b_width_log2_lookup[bsize];
    int i;

    assert(bsl > 0);
    --bsl;

    for (i = 0; i < 4; ++i) {
      const int offsetr = blk_row + ((i >> 1) << bsl);
      const int offsetc = blk_col + ((i & 0x01) << bsl);
346
      int step = num_4x4_blocks_txsize_lookup[tx_size - 1];
347

348
      if (offsetr >= max_blocks_high || offsetc >= max_blocks_wide) continue;
349 350 351 352 353 354

      decode_reconstruct_tx(xd, r, mbmi, plane, plane_bsize, block + i * step,
                            offsetr, offsetc, tx_size - 1, eob_total);
    }
  }
}
355
#endif  // CONFIG_VAR_TX
356

357
#if !CONFIG_VAR_TX || CONFIG_SUPERTX || (CONFIG_EXT_TX && CONFIG_RECT_TX)
Alex Converse's avatar
Alex Converse committed
358 359 360 361
static int reconstruct_inter_block(MACROBLOCKD *const xd,
#if CONFIG_ANS
                                   struct AnsDecoder *const r,
#else
362
                                   aom_reader *r,
Alex Converse's avatar
Alex Converse committed
363
#endif
364 365
                                   int segment_id, int plane, int row, int col,
                                   TX_SIZE tx_size) {
Jingning Han's avatar
Jingning Han committed
366
  struct macroblockd_plane *const pd = &xd->plane[plane];
hui su's avatar
hui su committed
367 368
  PLANE_TYPE plane_type = (plane == 0) ? PLANE_TYPE_Y : PLANE_TYPE_UV;
  int block_idx = (row << 1) + col;
hui su's avatar
hui su committed
369
  TX_TYPE tx_type = get_tx_type(plane_type, xd, block_idx, tx_size);
370
  const scan_order *sc = get_scan(tx_size, tx_type, 1);
371 372
  const int eob = av1_decode_block_tokens(xd, plane, sc, col, row, tx_size,
                                          tx_type, r, segment_id);
Jingning Han's avatar
Jingning Han committed
373

374 375 376
  inverse_transform_block(xd, plane, tx_type, tx_size,
                          &pd->dst.buf[4 * row * pd->dst.stride + 4 * col],
                          pd->dst.stride, eob);
Jingning Han's avatar
Jingning Han committed
377 378
  return eob;
}
379
#endif  // !CONFIG_VAR_TX || CONFIG_SUPER_TX
Jingning Han's avatar
Jingning Han committed
380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400

static INLINE void dec_reset_skip_context(MACROBLOCKD *xd) {
  int i;
  for (i = 0; i < MAX_MB_PLANE; i++) {
    struct macroblockd_plane *const pd = &xd->plane[i];
    memset(pd->above_context, 0, sizeof(ENTROPY_CONTEXT) * pd->n4_w);
    memset(pd->left_context, 0, sizeof(ENTROPY_CONTEXT) * pd->n4_h);
  }
}

static void set_plane_n4(MACROBLOCKD *const xd, int bw, int bh, int bwl,
                         int bhl) {
  int i;
  for (i = 0; i < MAX_MB_PLANE; i++) {
    xd->plane[i].n4_w = (bw << 1) >> xd->plane[i].subsampling_x;
    xd->plane[i].n4_h = (bh << 1) >> xd->plane[i].subsampling_y;
    xd->plane[i].n4_wl = bwl - xd->plane[i].subsampling_x;
    xd->plane[i].n4_hl = bhl - xd->plane[i].subsampling_y;
  }
}

401
static MB_MODE_INFO *set_offsets(AV1_COMMON *const cm, MACROBLOCKD *const xd,
Jingning Han's avatar
Jingning Han committed
402
                                 BLOCK_SIZE bsize, int mi_row, int mi_col,
403 404
                                 int bw, int bh, int x_mis, int y_mis, int bwl,
                                 int bhl) {
Jingning Han's avatar
Jingning Han committed
405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422
  const int offset = mi_row * cm->mi_stride + mi_col;
  int x, y;
  const TileInfo *const tile = &xd->tile;

  xd->mi = cm->mi_grid_visible + offset;
  xd->mi[0] = &cm->mi[offset];
  // TODO(slavarnway): Generate sb_type based on bwl and bhl, instead of
  // passing bsize from decode_partition().
  xd->mi[0]->mbmi.sb_type = bsize;
  for (y = 0; y < y_mis; ++y)
    for (x = !y; x < x_mis; ++x) {
      xd->mi[y * cm->mi_stride + x] = xd->mi[0];
    }

  set_plane_n4(xd, bw, bh, bwl, bhl);

  set_skip_context(xd, mi_row, mi_col);

423 424 425 426
#if CONFIG_VAR_TX
  xd->max_tx_size = max_txsize_lookup[bsize];
#endif

Jingning Han's avatar
Jingning Han committed
427 428 429 430
  // Distance of Mb to the various image edges. These are specified to 8th pel
  // as they are always compared to values that are in 1/8th pel units
  set_mi_row_col(xd, tile, mi_row, bh, mi_col, bw, cm->mi_rows, cm->mi_cols);

431
  av1_setup_dst_planes(xd->plane, get_frame_new_buffer(cm), mi_row, mi_col);
Jingning Han's avatar
Jingning Han committed
432 433 434
  return &xd->mi[0]->mbmi;
}

435
#if CONFIG_SUPERTX
436
static MB_MODE_INFO *set_offsets_extend(AV1_COMMON *const cm,
437 438
                                        MACROBLOCKD *const xd,
                                        const TileInfo *const tile,
439 440 441
                                        BLOCK_SIZE bsize_pred, int mi_row_pred,
                                        int mi_col_pred, int mi_row_ori,
                                        int mi_col_ori) {
442 443 444 445 446 447 448 449 450 451
  // Used in supertx
  // (mi_row_ori, mi_col_ori): location for mv
  // (mi_row_pred, mi_col_pred, bsize_pred): region to predict
  const int bw = num_8x8_blocks_wide_lookup[bsize_pred];
  const int bh = num_8x8_blocks_high_lookup[bsize_pred];
  const int offset = mi_row_ori * cm->mi_stride + mi_col_ori;
  const int bwl = b_width_log2_lookup[bsize_pred];
  const int bhl = b_height_log2_lookup[bsize_pred];
  xd->mi = cm->mi_grid_visible + offset;
  xd->mi[0] = cm->mi + offset;
452 453
  set_mi_row_col(xd, tile, mi_row_pred, bh, mi_col_pred, bw, cm->mi_rows,
                 cm->mi_cols);
454

455 456
  xd->up_available = (mi_row_ori > tile->mi_row_start);
  xd->left_available = (mi_col_ori > tile->mi_col_start);
457 458 459 460 461 462

  set_plane_n4(xd, bw, bh, bwl, bhl);

  return &xd->mi[0]->mbmi;
}

463 464 465
static MB_MODE_INFO *set_mb_offsets(AV1_COMMON *const cm, MACROBLOCKD *const xd,
                                    BLOCK_SIZE bsize, int mi_row, int mi_col,
                                    int bw, int bh, int x_mis, int y_mis) {
466 467 468 469 470 471 472 473
  const int offset = mi_row * cm->mi_stride + mi_col;
  const TileInfo *const tile = &xd->tile;
  int x, y;

  xd->mi = cm->mi_grid_visible + offset;
  xd->mi[0] = cm->mi + offset;
  xd->mi[0]->mbmi.sb_type = bsize;
  for (y = 0; y < y_mis; ++y)
474
    for (x = !y; x < x_mis; ++x) xd->mi[y * cm->mi_stride + x] = xd->mi[0];
475 476 477 478 479

  set_mi_row_col(xd, tile, mi_row, bh, mi_col, bw, cm->mi_rows, cm->mi_cols);
  return &xd->mi[0]->mbmi;
}

480
static void set_offsets_topblock(AV1_COMMON *const cm, MACROBLOCKD *const xd,
481 482
                                 const TileInfo *const tile, BLOCK_SIZE bsize,
                                 int mi_row, int mi_col) {
483 484 485 486 487 488 489 490 491 492 493 494 495
  const int bw = num_8x8_blocks_wide_lookup[bsize];
  const int bh = num_8x8_blocks_high_lookup[bsize];
  const int offset = mi_row * cm->mi_stride + mi_col;
  const int bwl = b_width_log2_lookup[bsize];
  const int bhl = b_height_log2_lookup[bsize];

  xd->mi = cm->mi_grid_visible + offset;
  xd->mi[0] = cm->mi + offset;

  set_plane_n4(xd, bw, bh, bwl, bhl);

  set_mi_row_col(xd, tile, mi_row, bh, mi_col, bw, cm->mi_rows, cm->mi_cols);

496
  av1_setup_dst_planes(xd->plane, get_frame_new_buffer(cm), mi_row, mi_col);
497 498
}

499
static void set_param_topblock(AV1_COMMON *const cm, MACROBLOCKD *const xd,
500
                               BLOCK_SIZE bsize, int mi_row, int mi_col,
501
                               int txfm, int skip) {
502 503
  const int bw = num_8x8_blocks_wide_lookup[bsize];
  const int bh = num_8x8_blocks_high_lookup[bsize];
504 505
  const int x_mis = AOMMIN(bw, cm->mi_cols - mi_col);
  const int y_mis = AOMMIN(bh, cm->mi_rows - mi_row);
506 507 508 509 510 511 512 513 514 515 516
  const int offset = mi_row * cm->mi_stride + mi_col;
  int x, y;

  xd->mi = cm->mi_grid_visible + offset;
  xd->mi[0] = cm->mi + offset;

  for (y = 0; y < y_mis; ++y)
    for (x = 0; x < x_mis; ++x) {
      xd->mi[y * cm->mi_stride + x]->mbmi.skip = skip;
      xd->mi[y * cm->mi_stride + x]->mbmi.tx_type = txfm;
    }
517 518
#if CONFIG_VAR_TX
  xd->above_txfm_context = cm->above_txfm_context + mi_col;
519
  xd->left_txfm_context =
520
      xd->left_txfm_context_buffer + (mi_row & MAX_MIB_MASK);
521
  set_txfm_ctxs(xd->mi[0]->mbmi.tx_size, bw, bh, xd);
522
#endif
523 524
}

525
static void set_ref(AV1_COMMON *const cm, MACROBLOCKD *const xd, int idx,
526
                    int mi_row, int mi_col) {
527 528 529
  MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;
  RefBuffer *ref_buffer = &cm->frame_refs[mbmi->ref_frame[idx] - LAST_FRAME];
  xd->block_refs[idx] = ref_buffer;
530 531
  if (!av1_is_valid_scale(&ref_buffer->sf))
    aom_internal_error(&cm->error, AOM_CODEC_UNSUP_BITSTREAM,
532
                       "Invalid scale factors");
533 534
  av1_setup_pre_planes(xd, idx, ref_buffer->buf, mi_row, mi_col,
                       &ref_buffer->sf);
535 536 537 538
  xd->corrupted |= ref_buffer->buf->corrupted;
}

static void dec_predict_b_extend(
539
    AV1Decoder *const pbi, MACROBLOCKD *const xd, const TileInfo *const tile,
540 541 542
    int block, int mi_row_ori, int mi_col_ori, int mi_row_pred, int mi_col_pred,
    int mi_row_top, int mi_col_top, uint8_t *dst_buf[3], int dst_stride[3],
    BLOCK_SIZE bsize_top, BLOCK_SIZE bsize_pred, int b_sub8x8, int bextend) {
543 544 545 546 547 548 549 550 551 552 553 554
  // Used in supertx
  // (mi_row_ori, mi_col_ori): location for mv
  // (mi_row_pred, mi_col_pred, bsize_pred): region to predict
  // (mi_row_top, mi_col_top, bsize_top): region of the top partition size
  // block: sub location of sub8x8 blocks
  // b_sub8x8: 1: ori is sub8x8; 0: ori is not sub8x8
  // bextend: 1: region to predict is an extension of ori; 0: not
  int r = (mi_row_pred - mi_row_top) * MI_SIZE;
  int c = (mi_col_pred - mi_col_top) * MI_SIZE;
  const int mi_width_top = num_8x8_blocks_wide_lookup[bsize_top];
  const int mi_height_top = num_8x8_blocks_high_lookup[bsize_top];
  MB_MODE_INFO *mbmi;
555
  AV1_COMMON *const cm = &pbi->common;
556 557 558

  if (mi_row_pred < mi_row_top || mi_col_pred < mi_col_top ||
      mi_row_pred >= mi_row_top + mi_height_top ||
559 560
      mi_col_pred >= mi_col_top + mi_width_top || mi_row_pred >= cm->mi_rows ||
      mi_col_pred >= cm->mi_cols)
561 562
    return;

563
  mbmi = set_offsets_extend(cm, xd, tile, bsize_pred, mi_row_pred, mi_col_pred,
564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586
                            mi_row_ori, mi_col_ori);
  set_ref(cm, xd, 0, mi_row_pred, mi_col_pred);
  if (has_second_ref(&xd->mi[0]->mbmi))
    set_ref(cm, xd, 1, mi_row_pred, mi_col_pred);

  if (!bextend) {
    mbmi->tx_size = b_width_log2_lookup[bsize_top];
  }

  xd->plane[0].dst.stride = dst_stride[0];
  xd->plane[1].dst.stride = dst_stride[1];
  xd->plane[2].dst.stride = dst_stride[2];
  xd->plane[0].dst.buf = dst_buf[0] +
                         (r >> xd->plane[0].subsampling_y) * dst_stride[0] +
                         (c >> xd->plane[0].subsampling_x);
  xd->plane[1].dst.buf = dst_buf[1] +
                         (r >> xd->plane[1].subsampling_y) * dst_stride[1] +
                         (c >> xd->plane[1].subsampling_x);
  xd->plane[2].dst.buf = dst_buf[2] +
                         (r >> xd->plane[2].subsampling_y) * dst_stride[2] +
                         (c >> xd->plane[2].subsampling_x);

  if (!b_sub8x8)
587
    av1_build_inter_predictors_sb_extend(xd,
588
#if CONFIG_EXT_INTER
589
                                         mi_row_ori, mi_col_ori,
590
#endif  // CONFIG_EXT_INTER
591
                                         mi_row_pred, mi_col_pred, bsize_pred);
592
  else
593
    av1_build_inter_predictors_sb_sub8x8_extend(xd,
594
#if CONFIG_EXT_INTER
595
                                                mi_row_ori, mi_col_ori,
596
#endif  // CONFIG_EXT_INTER
597 598
                                                mi_row_pred, mi_col_pred,
                                                bsize_pred, block);
599 600
}

601
static void dec_extend_dir(AV1Decoder *const pbi, MACROBLOCKD *const xd,
602
                           const TileInfo *const tile, int block,
603 604 605
                           BLOCK_SIZE bsize, BLOCK_SIZE top_bsize, int mi_row,
                           int mi_col, int mi_row_top, int mi_col_top,
                           uint8_t *dst_buf[3], int dst_stride[3], int dir) {
606 607 608 609 610 611 612 613 614 615 616
  // dir: 0-lower, 1-upper, 2-left, 3-right
  //      4-lowerleft, 5-upperleft, 6-lowerright, 7-upperright
  const int mi_width = num_8x8_blocks_wide_lookup[bsize];
  const int mi_height = num_8x8_blocks_high_lookup[bsize];
  int xss = xd->plane[1].subsampling_x;
  int yss = xd->plane[1].subsampling_y;
  int b_sub8x8 = (bsize < BLOCK_8X8) ? 1 : 0;
  BLOCK_SIZE extend_bsize;
  int unit, mi_row_pred, mi_col_pred;

  if (dir == 0 || dir == 1) {
617 618 619
    extend_bsize = (mi_width == 1 || bsize < BLOCK_8X8 || xss < yss)
                       ? BLOCK_8X8
                       : BLOCK_16X8;
620 621 622 623
    unit = num_8x8_blocks_wide_lookup[extend_bsize];
    mi_row_pred = mi_row + ((dir == 0) ? mi_height : -1);
    mi_col_pred = mi_col;

624 625 626
    dec_predict_b_extend(pbi, xd, tile, block, mi_row, mi_col, mi_row_pred,
                         mi_col_pred, mi_row_top, mi_col_top, dst_buf,
                         dst_stride, top_bsize, extend_bsize, b_sub8x8, 1);
627 628 629 630

    if (mi_width > unit) {
      int i;
      assert(!b_sub8x8);
631
      for (i = 0; i < mi_width / unit - 1; i++) {
632
        mi_col_pred += unit;
633 634 635
        dec_predict_b_extend(pbi, xd, tile, block, mi_row, mi_col, mi_row_pred,
                             mi_col_pred, mi_row_top, mi_col_top, dst_buf,
                             dst_stride, top_bsize, extend_bsize, b_sub8x8, 1);
636 637 638
      }
    }
  } else if (dir == 2 || dir == 3) {
639 640 641
    extend_bsize = (mi_height == 1 || bsize < BLOCK_8X8 || yss < xss)
                       ? BLOCK_8X8
                       : BLOCK_8X16;
642 643 644 645
    unit = num_8x8_blocks_high_lookup[extend_bsize];
    mi_row_pred = mi_row;
    mi_col_pred = mi_col + ((dir == 3) ? mi_width : -1);

646 647 648
    dec_predict_b_extend(pbi, xd, tile, block, mi_row, mi_col, mi_row_pred,
                         mi_col_pred, mi_row_top, mi_col_top, dst_buf,
                         dst_stride, top_bsize, extend_bsize, b_sub8x8, 1);
649 650 651

    if (mi_height > unit) {
      int i;
652
      for (i = 0; i < mi_height / unit - 1; i++) {
653
        mi_row_pred += unit;
654 655 656
        dec_predict_b_extend(pbi, xd, tile, block, mi_row, mi_col, mi_row_pred,
                             mi_col_pred, mi_row_top, mi_col_top, dst_buf,
                             dst_stride, top_bsize, extend_bsize, b_sub8x8, 1);
657 658 659 660 661 662
      }
    }
  } else {
    extend_bsize = BLOCK_8X8;
    mi_row_pred = mi_row + ((dir == 4 || dir == 6) ? mi_height : -1);
    mi_col_pred = mi_col + ((dir == 6 || dir == 7) ? mi_width : -1);
663 664 665
    dec_predict_b_extend(pbi, xd, tile, block, mi_row, mi_col, mi_row_pred,
                         mi_col_pred, mi_row_top, mi_col_top, dst_buf,
                         dst_stride, top_bsize, extend_bsize, b_sub8x8, 1);
666 667 668
  }
}

669
static void dec_extend_all(AV1Decoder *const pbi, MACROBLOCKD *const xd,
670
                           const TileInfo *const tile, int block,
671 672 673
                           BLOCK_SIZE bsize, BLOCK_SIZE top_bsize, int mi_row,
                           int mi_col, int mi_row_top, int mi_col_top,
                           uint8_t *dst_buf[3], int dst_stride[3]) {
674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691
  dec_extend_dir(pbi, xd, tile, block, bsize, top_bsize, mi_row, mi_col,
                 mi_row_top, mi_col_top, dst_buf, dst_stride, 0);
  dec_extend_dir(pbi, xd, tile, block, bsize, top_bsize, mi_row, mi_col,
                 mi_row_top, mi_col_top, dst_buf, dst_stride, 1);
  dec_extend_dir(pbi, xd, tile, block, bsize, top_bsize, mi_row, mi_col,
                 mi_row_top, mi_col_top, dst_buf, dst_stride, 2);
  dec_extend_dir(pbi, xd, tile, block, bsize, top_bsize, mi_row, mi_col,
                 mi_row_top, mi_col_top, dst_buf, dst_stride, 3);
  dec_extend_dir(pbi, xd, tile, block, bsize, top_bsize, mi_row, mi_col,
                 mi_row_top, mi_col_top, dst_buf, dst_stride, 4);
  dec_extend_dir(pbi, xd, tile, block, bsize, top_bsize, mi_row, mi_col,
                 mi_row_top, mi_col_top, dst_buf, dst_stride, 5);
  dec_extend_dir(pbi, xd, tile, block, bsize, top_bsize, mi_row, mi_col,
                 mi_row_top, mi_col_top, dst_buf, dst_stride, 6);
  dec_extend_dir(pbi, xd, tile, block, bsize, top_bsize, mi_row, mi_col,
                 mi_row_top, mi_col_top, dst_buf, dst_stride, 7);
}

692
static void dec_predict_sb_complex(AV1Decoder *const pbi, MACROBLOCKD *const xd,
693 694
                                   const TileInfo *const tile, int mi_row,
                                   int mi_col, int mi_row_top, int mi_col_top,
695 696
                                   BLOCK_SIZE bsize, BLOCK_SIZE top_bsize,
                                   uint8_t *dst_buf[3], int dst_stride[3]) {
697
  const AV1_COMMON *const cm = &pbi->common;
698 699 700
  const int hbs = num_8x8_blocks_wide_lookup[bsize] / 2;
  const PARTITION_TYPE partition = get_partition(cm, mi_row, mi_col, bsize);
  const BLOCK_SIZE subsize = get_subsize(bsize, partition);
701
#if CONFIG_EXT_PARTITION_TYPES
702
  const BLOCK_SIZE bsize2 = get_subsize(bsize, PARTITION_SPLIT);
703
#endif
704 705
  int i;
  const int mi_offset = mi_row * cm->mi_stride + mi_col;
706 707
  uint8_t *dst_buf1[3], *dst_buf2[3], *dst_buf3[3];

708 709 710 711 712 713
  DECLARE_ALIGNED(16, uint8_t, tmp_buf1[MAX_MB_PLANE * MAX_TX_SQUARE * 2]);
  DECLARE_ALIGNED(16, uint8_t, tmp_buf2[MAX_MB_PLANE * MAX_TX_SQUARE * 2]);
  DECLARE_ALIGNED(16, uint8_t, tmp_buf3[MAX_MB_PLANE * MAX_TX_SQUARE * 2]);
  int dst_stride1[3] = { MAX_TX_SIZE, MAX_TX_SIZE, MAX_TX_SIZE };
  int dst_stride2[3] = { MAX_TX_SIZE, MAX_TX_SIZE, MAX_TX_SIZE };
  int dst_stride3[3] = { MAX_TX_SIZE, MAX_TX_SIZE, MAX_TX_SIZE };
714

715
#if CONFIG_AOM_HIGHBITDEPTH
716 717 718
  if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
    int len = sizeof(uint16_t);
    dst_buf1[0] = CONVERT_TO_BYTEPTR(tmp_buf1);
719 720
    dst_buf1[1] = CONVERT_TO_BYTEPTR(tmp_buf1 + MAX_TX_SQUARE * len);
    dst_buf1[2] = CONVERT_TO_BYTEPTR(tmp_buf1 + 2 * MAX_TX_SQUARE * len);
721
    dst_buf2[0] = CONVERT_TO_BYTEPTR(tmp_buf2);
722 723
    dst_buf2[1] = CONVERT_TO_BYTEPTR(tmp_buf2 + MAX_TX_SQUARE * len);
    dst_buf2[2] = CONVERT_TO_BYTEPTR(tmp_buf2 + 2 * MAX_TX_SQUARE * len);
724
    dst_buf3[0] = CONVERT_TO_BYTEPTR(tmp_buf3);
725 726
    dst_buf3[1] = CONVERT_TO_BYTEPTR(tmp_buf3 + MAX_TX_SQUARE * len);
    dst_buf3[2] = CONVERT_TO_BYTEPTR(tmp_buf3 + 2 * MAX_TX_SQUARE * len);
727 728 729
  } else {
#endif
    dst_buf1[0] = tmp_buf1;
730 731
    dst_buf1[1] = tmp_buf1 + MAX_TX_SQUARE;
    dst_buf1[2] = tmp_buf1 + 2 * MAX_TX_SQUARE;
732
    dst_buf2[0] = tmp_buf2;
733 734
    dst_buf2[1] = tmp_buf2 + MAX_TX_SQUARE;
    dst_buf2[2] = tmp_buf2 + 2 * MAX_TX_SQUARE;
735
    dst_buf3[0] = tmp_buf3;
736 737
    dst_buf3[1] = tmp_buf3 + MAX_TX_SQUARE;
    dst_buf3[2] = tmp_buf3 + 2 * MAX_TX_SQUARE;
738
#if CONFIG_AOM_HIGHBITDEPTH
739 740 741
  }
#endif

742
  if (mi_row >= cm->mi_rows || mi_col >= cm->mi_cols) return;
743

744 745
  xd->mi = cm->mi_grid_visible + mi_offset;
  xd->mi[0] = cm->mi + mi_offset;
746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782

  for (i = 0; i < MAX_MB_PLANE; i++) {
    xd->plane[i].dst.buf = dst_buf[i];
    xd->plane[i].dst.stride = dst_stride[i];
  }

  switch (partition) {
    case PARTITION_NONE:
      assert(bsize < top_bsize);
      dec_predict_b_extend(pbi, xd, tile, 0, mi_row, mi_col, mi_row, mi_col,
                           mi_row_top, mi_col_top, dst_buf, dst_stride,
                           top_bsize, bsize, 0, 0);
      dec_extend_all(pbi, xd, tile, 0, bsize, top_bsize, mi_row, mi_col,
                     mi_row_top, mi_col_top, dst_buf, dst_stride);
      break;
    case PARTITION_HORZ:
      if (bsize == BLOCK_8X8) {
        // For sub8x8, predict in 8x8 unit
        // First half
        dec_predict_b_extend(pbi, xd, tile, 0, mi_row, mi_col, mi_row, mi_col,
                             mi_row_top, mi_col_top, dst_buf, dst_stride,
                             top_bsize, BLOCK_8X8, 1, 0);
        if (bsize < top_bsize)
          dec_extend_all(pbi, xd, tile, 0, subsize, top_bsize, mi_row, mi_col,
                         mi_row_top, mi_col_top, dst_buf, dst_stride);

        // Second half
        dec_predict_b_extend(pbi, xd, tile, 2, mi_row, mi_col, mi_row, mi_col,
                             mi_row_top, mi_col_top, dst_buf1, dst_stride1,
                             top_bsize, BLOCK_8X8, 1, 1);
        if (bsize < top_bsize)
          dec_extend_all(pbi, xd, tile, 2, subsize, top_bsize, mi_row, mi_col,
                         mi_row_top, mi_col_top, dst_buf1, dst_stride1);

        // weighted average to smooth the boundary
        xd->plane[0].dst.buf = dst_buf[0];
        xd->plane[0].dst.stride = dst_stride[0];
783
        av1_build_masked_inter_predictor_complex(
784 785 786
            xd, dst_buf[0], dst_stride[0], dst_buf1[0], dst_stride1[0], mi_row,
            mi_col, mi_row_top, mi_col_top, bsize, top_bsize, PARTITION_HORZ,
            0);
787 788 789 790 791 792 793 794 795 796 797 798 799 800 801
      } else {
        // First half
        dec_predict_b_extend(pbi, xd, tile, 0, mi_row, mi_col, mi_row, mi_col,
                             mi_row_top, mi_col_top, dst_buf, dst_stride,
                             top_bsize, subsize, 0, 0);
        if (bsize < top_bsize)
          dec_extend_all(pbi, xd, tile, 0, subsize, top_bsize, mi_row, mi_col,
                         mi_row_top, mi_col_top, dst_buf, dst_stride);
        else
          dec_extend_dir(pbi, xd, tile, 0, subsize, top_bsize, mi_row, mi_col,
                         mi_row_top, mi_col_top, dst_buf, dst_stride, 0);

        if (mi_row + hbs < cm->mi_rows) {
          // Second half
          dec_predict_b_extend(pbi, xd, tile, 0, mi_row + hbs, mi_col,
802 803
                               mi_row + hbs, mi_col, mi_row_top, mi_col_top,
                               dst_buf1, dst_stride1, top_bsize, subsize, 0, 0);
804
          if (bsize < top_bsize)
805 806 807
            dec_extend_all(pbi, xd, tile, 0, subsize, top_bsize, mi_row + hbs,
                           mi_col, mi_row_top, mi_col_top, dst_buf1,
                           dst_stride1);
808
          else
809 810 811
            dec_extend_dir(pbi, xd, tile, 0, subsize, top_bsize, mi_row + hbs,
                           mi_col, mi_row_top, mi_col_top, dst_buf1,
                           dst_stride1, 1);
812 813 814 815 816

          // weighted average to smooth the boundary
          for (i = 0; i < MAX_MB_PLANE; i++) {
            xd->plane[i].dst.buf = dst_buf[i];
            xd->plane[i].dst.stride = dst_stride[i];
817
            av1_build_masked_inter_predictor_complex(
818
                xd, dst_buf[i], dst_stride[i], dst_buf1[i], dst_stride1[i],
819 820
                mi_row, mi_col, mi_row_top, mi_col_top, bsize, top_bsize,
                PARTITION_HORZ, i);
821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845
          }
        }
      }
      break;
    case PARTITION_VERT:
      if (bsize == BLOCK_8X8) {
        // First half
        dec_predict_b_extend(pbi, xd, tile, 0, mi_row, mi_col, mi_row, mi_col,
                             mi_row_top, mi_col_top, dst_buf, dst_stride,
                             top_bsize, BLOCK_8X8, 1, 0);
        if (bsize < top_bsize)
          dec_extend_all(pbi, xd, tile, 0, subsize, top_bsize, mi_row, mi_col,
                         mi_row_top, mi_col_top, dst_buf, dst_stride);

        // Second half
        dec_predict_b_extend(pbi, xd, tile, 1, mi_row, mi_col, mi_row, mi_col,
                             mi_row_top, mi_col_top, dst_buf1, dst_stride1,
                             top_bsize, BLOCK_8X8, 1, 1);
        if (bsize < top_bsize)
          dec_extend_all(pbi, xd, tile, 1, subsize, top_bsize, mi_row, mi_col,
                         mi_row_top, mi_col_top, dst_buf1, dst_stride1);

        // Smooth
        xd->plane[0].dst.buf = dst_buf[0];
        xd->plane[0].dst.stride = dst_stride[0];
846
        av1_build_masked_inter_predictor_complex(
847 848 849
            xd, dst_buf[0], dst_stride[0], dst_buf1[0], dst_stride1[0], mi_row,
            mi_col, mi_row_top, mi_col_top, bsize, top_bsize, PARTITION_VERT,
            0);
850 851 852 853 854 855 856 857 858 859 860 861 862 863
      } else {
        // First half
        dec_predict_b_extend(pbi, xd, tile, 0, mi_row, mi_col, mi_row, mi_col,
                             mi_row_top, mi_col_top, dst_buf, dst_stride,
                             top_bsize, subsize, 0, 0);
        if (bsize < top_bsize)
          dec_extend_all(pbi, xd, tile, 0, subsize, top_bsize, mi_row, mi_col,
                         mi_row_top, mi_col_top, dst_buf, dst_stride);
        else
          dec_extend_dir(pbi, xd, tile, 0, subsize, top_bsize, mi_row, mi_col,
                         mi_row_top, mi_col_top, dst_buf, dst_stride, 3);

        // Second half
        if (mi_col + hbs < cm->mi_cols) {
864 865 866
          dec_predict_b_extend(pbi, xd, tile, 0, mi_row, mi_col + hbs, mi_row,
                               mi_col + hbs, mi_row_top, mi_col_top, dst_buf1,
                               dst_stride1, top_bsize, subsize, 0, 0);
867
          if (bsize < top_bsize)
868 869 870
            dec_extend_all(pbi, xd, tile, 0, subsize, top_bsize, mi_row,
                           mi_col + hbs, mi_row_top, mi_col_top, dst_buf1,
                           dst_stride1);
871
          else
872 873 874
            dec_extend_dir(pbi, xd, tile, 0, subsize, top_bsize, mi_row,
                           mi_col + hbs, mi_row_top, mi_col_top, dst_buf1,
                           dst_stride1, 2);
875 876 877 878 879

          // Smooth
          for (i = 0; i < MAX_MB_PLANE; i++) {
            xd->plane[i].dst.buf = dst_buf[i];
            xd->plane[i].dst.stride = dst_stride[i];
880
            av1_build_masked_inter_predictor_complex(
881
                xd, dst_buf[i], dst_stride[i], dst_buf1[i], dst_stride1[i],
882 883
                mi_row, mi_col, mi_row_top, mi_col_top, bsize, top_bsize,
                PARTITION_VERT, i);
884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912
          }
        }
      }
      break;
    case PARTITION_SPLIT:
      if (bsize == BLOCK_8X8) {
        dec_predict_b_extend(pbi, xd, tile, 0, mi_row, mi_col, mi_row, mi_col,
                             mi_row_top, mi_col_top, dst_buf, dst_stride,
                             top_bsize, BLOCK_8X8, 1, 0);
        dec_predict_b_extend(pbi, xd, tile, 1, mi_row, mi_col, mi_row, mi_col,
                             mi_row_top, mi_col_top, dst_buf1, dst_stride1,
                             top_bsize, BLOCK_8X8, 1, 1);
        dec_predict_b_extend(pbi, xd, tile, 2, mi_row, mi_col, mi_row, mi_col,
                             mi_row_top, mi_col_top, dst_buf2, dst_stride2,
                             top_bsize, BLOCK_8X8, 1, 1);
        dec_predict_b_extend(pbi, xd, tile, 3, mi_row, mi_col, mi_row, mi_col,
                             mi_row_top, mi_col_top, dst_buf3, dst_stride3,
                             top_bsize, BLOCK_8X8, 1, 1);
        if (bsize < top_bsize) {
          dec_extend_all(pbi, xd, tile, 0, subsize, top_bsize, mi_row, mi_col,
                         mi_row_top, mi_col_top, dst_buf, dst_stride);
          dec_extend_all(pbi, xd, tile, 1, subsize, top_bsize, mi_row, mi_col,
                         mi_row_top, mi_col_top, dst_buf1, dst_stride1);
          dec_extend_all(pbi, xd, tile, 2, subsize, top_bsize, mi_row, mi_col,
                         mi_row_top, mi_col_top, dst_buf2, dst_stride2);
          dec_extend_all(pbi, xd, tile, 3, subsize, top_bsize, mi_row, mi_col,
                         mi_row_top, mi_col_top, dst_buf3, dst_stride3);
        }
      } else {
913 914 915
        dec_predict_sb_complex(pbi, xd, tile, mi_row, mi_col, mi_row_top,
                               mi_col_top, subsize, top_bsize, dst_buf,
                               dst_stride);
916 917 918 919 920 921 922 923 924 925 926 927 928
        if (mi_row < cm->mi_rows && mi_col + hbs < cm->mi_cols)
          dec_predict_sb_complex(pbi, xd, tile, mi_row, mi_col + hbs,
                                 mi_row_top, mi_col_top, subsize, top_bsize,
                                 dst_buf1, dst_stride1);
        if (mi_row + hbs < cm->mi_rows && mi_col < cm->mi_cols)
          dec_predict_sb_complex(pbi, xd, tile, mi_row + hbs, mi_col,
                                 mi_row_top, mi_col_top, subsize, top_bsize,
                                 dst_buf2, dst_stride2);
        if (mi_row + hbs < cm->mi_rows && mi_col + hbs < cm->mi_cols)
          dec_predict_sb_complex(pbi, xd, tile, mi_row + hbs, mi_col + hbs,
                                 mi_row_top, mi_col_top, subsize, top_bsize,
                                 dst_buf3, dst_stride3);
      }
929 930 931 932
      for (i = 0; i < MAX_MB_PLANE; i++) {
        if (bsize == BLOCK_8X8 && i != 0)
          continue;  // Skip <4x4 chroma smoothing
        if (mi_row < cm->mi_rows && mi_col + hbs < cm->mi_cols) {
933
          av1_build_masked_inter_predictor_complex(
934 935 936 937
              xd, dst_buf[i], dst_stride[i], dst_buf1[i], dst_stride1[i],
              mi_row, mi_col, mi_row_top, mi_col_top, bsize, top_bsize,
              PARTITION_VERT, i);
          if (mi_row + hbs < cm->mi_rows) {
938
            av1_build_masked_inter_predictor_complex(
939 940 941
                xd, dst_buf2[i], dst_stride2[i], dst_buf3[i], dst_stride3[i],
                mi_row, mi_col, mi_row_top, mi_col_top, bsize, top_bsize,
                PARTITION_VERT, i);
942
            av1_build_masked_inter_predictor_complex(
943 944 945
                xd, dst_buf[i], dst_stride[i], dst_buf2[i], dst_stride2[i],
                mi_row, mi_col, mi_row_top, mi_col_top, bsize, top_bsize,
                PARTITION_HORZ, i);
946
          }
947
        } else if (mi_row + hbs < cm->mi_rows && mi_col < cm->mi_cols) {
948
          av1_build_masked_inter_predictor_complex(
949 950 951
              xd, dst_buf[i], dst_stride[i], dst_buf2[i], dst_stride2[i],
              mi_row, mi_col, mi_row_top, mi_col_top, bsize, top_bsize,
              PARTITION_HORZ, i);
952
        }
953
      }
954
      break;
955 956 957 958 959 960 961 962 963
#if CONFIG_EXT_PARTITION_TYPES
    case PARTITION_HORZ_A:
      dec_predict_b_extend(pbi, xd, tile, 0, mi_row, mi_col, mi_row, mi_col,
                           mi_row_top, mi_col_top, dst_buf, dst_stride,
                           top_bsize, bsize2, 0, 0);
      dec_extend_all(pbi, xd, tile, 0, bsize2, top_bsize, mi_row, mi_col,
                     mi_row_top, mi_col_top, dst_buf, dst_stride);

      dec_predict_b_extend(pbi, xd, tile, 0, mi_row, mi_col + hbs, mi_row,
964 965
                           mi_col + hbs, mi_row_top, mi_col_top, dst_buf1,
                           dst_stride1, top_bsize, bsize2, 0, 0);
966 967 968
      dec_extend_all(pbi, xd, tile, 0, bsize2, top_bsize, mi_row, mi_col + hbs,
                     mi_row_top, mi_col_top, dst_buf1, dst_stride1);

969 970 971
      dec_predict_b_extend(pbi, xd, tile, 0, mi_row + hbs, mi_col, mi_row + hbs,
                           mi_col, mi_row_top, mi_col_top, dst_buf2,
                           dst_stride2, top_bsize, subsize, 0, 0);
972
      if (bsize < top_bsize)
973 974
        dec_extend_all(pbi, xd, tile, 0, subsize, top_bsize, mi_row + hbs,
                       mi_col, mi_row_top, mi_col_top, dst_buf2, dst_stride2);
975
      else
976 977 978
        dec_extend_dir(pbi, xd, tile, 0, subsize, top_bsize, mi_row + hbs,
                       mi_col, mi_row_top, mi_col_top, dst_buf2, dst_stride2,
                       1);
979 980 981 982

      for (i = 0; i < MAX_MB_PLANE; i++) {
        xd->plane[i].dst.buf = dst_buf[i];
        xd->plane[i].dst.stride = dst_stride[i];
983
        av1_build_masked_inter_predictor_complex(
984 985 986
            xd, dst_buf[i], dst_stride[i], dst_buf1[i], dst_stride1[i], mi_row,
            mi_col, mi_row_top, mi_col_top, bsize, top_bsize, PARTITION_VERT,
            i);
987 988
      }
      for (i = 0; i < MAX_MB_PLANE; i++) {
989
        av1_build_masked_inter_predictor_complex(
990 991 992
            xd, dst_buf[i], dst_stride[i], dst_buf2[i], dst_stride2[i], mi_row,
            mi_col, mi_row_top, mi_col_top, bsize, top_bsize, PARTITION_HORZ,
            i);
993 994 995 996 997 998 999 1000 1001 1002
      }
      break;
    case PARTITION_VERT_A:

      dec_predict_b_extend(pbi, xd, tile, 0, mi_row, mi_col, mi_row, mi_col,
                           mi_row_top, mi_col_top, dst_buf, dst_stride,
                           top_bsize, bsize2, 0, 0);
      dec_extend_all(pbi, xd, tile, 0, bsize2, top_bsize, mi_row, mi_col,
                     mi_row_top, mi_col_top, dst_buf, dst_stride);

1003 1004 1005
      dec_predict_b_extend(pbi, xd, tile, 0, mi_row + hbs, mi_col, mi_row + hbs,
                           mi_col, mi_row_top, mi_col_top, dst_buf1,
                           dst_stride1, top_bsize, bsize2, 0, 0);
1006 1007 1008
      dec_extend_all(pbi, xd, tile, 0, bsize2, top_bsize, mi_row + hbs, mi_col,
                     mi_row_top, mi_col_top, dst_buf1, dst_stride1);

1009 1010 1011
      dec_predict_b_extend(pbi, xd, tile, 0, mi_row, mi_col + hbs, mi_row,
                           mi_col + hbs, mi_row_top, mi_col_top, dst_buf2,
                           dst_stride2, top_bsize, subsize, 0, 0);
1012
      if (bsize < top_bsize)
1013 1014 1015
        dec_extend_all(pbi, xd, tile, 0, subsize, top_bsize, mi_row,
                       mi_col + hbs, mi_row_top, mi_col_top, dst_buf2,
                       dst_stride2);
1016
      else
1017 1018 1019
        dec_extend_dir(pbi, xd, tile, 0, subsize, top_bsize, mi_row,
                       mi_col + hbs, mi_row_top, mi_col_top, dst_buf2,
                       dst_stride2, 2);
1020 1021 1022 1023

      for (i = 0; i < MAX_MB_PLANE; i++) {
        xd->plane[i].dst.buf = dst_buf[i];
        xd->plane[i].dst.stride = dst_stride[i];
1024
        av1_build_masked_inter_predictor_complex(
1025 1026 1027
            xd, dst_buf[i], dst_stride[i], dst_buf1[i], dst_stride1[i], mi_row,
            mi_col, mi_row_top, mi_col_top, bsize, top_bsize, PARTITION_HORZ,
            i);
1028 1029
      }
      for (i = 0; i < MAX_MB_PLANE; i++) {
1030
        av1_build_masked_inter_predictor_complex(
1031 1032 1033
            xd, dst_buf[i], dst_stride[i], dst_buf2[i], dst_stride2[i], mi_row,
            mi_col, mi_row_top, mi_col_top, bsize, top_bsize, PARTITION_VERT,
            i);