decodeframe.c 166 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
#include "av1/common/entropy.h"
#include "av1/common/entropymode.h"
41
#include "av1/common/entropymv.h"
42 43 44 45
#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
46
#include "av1/common/reconintra.h"
47
#include "av1/common/seg_common.h"
Jingning Han's avatar
Jingning Han committed
48
#include "av1/common/thread_common.h"
49
#include "av1/common/tile_common.h"
50

51 52 53
#include "av1/decoder/decodeframe.h"
#include "av1/decoder/decodemv.h"
#include "av1/decoder/decoder.h"
Jingning Han's avatar
Jingning Han committed
54
#include "av1/decoder/detokenize.h"
55
#include "av1/decoder/dsubexp.h"
Jingning Han's avatar
Jingning Han committed
56

57
#define MAX_AV1_HEADER_SIZE 80
Michael Bebenita's avatar
Michael Bebenita committed
58
#define ACCT_STR __func__
Jingning Han's avatar
Jingning Han committed
59

60 61 62 63 64 65 66 67 68 69
#if CONFIG_PVQ
#include "av1/decoder/pvq_decoder.h"
#include "av1/encoder/encodemb.h"

#include "aom_dsp/entdec.h"
#include "av1/common/partition.h"
#include "av1/decoder/decint.h"
#include "av1/encoder/hybrid_fwd_txfm.h"
#endif

70 71 72 73 74 75 76 77
static struct aom_read_bit_buffer *init_read_bit_buffer(
    AV1Decoder *pbi, struct aom_read_bit_buffer *rb, const uint8_t *data,
    const uint8_t *data_end, uint8_t clear_data[MAX_AV1_HEADER_SIZE]);
static int read_compressed_header(AV1Decoder *pbi, const uint8_t *data,
                                  size_t partition_size);
static size_t read_uncompressed_header(AV1Decoder *pbi,
                                       struct aom_read_bit_buffer *rb);

78
static int is_compound_reference_allowed(const AV1_COMMON *cm) {
Jingning Han's avatar
Jingning Han committed
79
  int i;
80
  if (frame_is_intra_only(cm)) return 0;
81
  for (i = 1; i < INTER_REFS_PER_FRAME; ++i)
82
    if (cm->ref_frame_sign_bias[i + 1] != cm->ref_frame_sign_bias[1]) return 1;
Jingning Han's avatar
Jingning Han committed
83 84 85 86

  return 0;
}

87
static void setup_compound_reference_mode(AV1_COMMON *cm) {
88
#if CONFIG_EXT_REFS
89
  cm->comp_fwd_ref[0] = LAST_FRAME;
90 91 92 93
  cm->comp_fwd_ref[1] = LAST2_FRAME;
  cm->comp_fwd_ref[2] = LAST3_FRAME;
  cm->comp_fwd_ref[3] = GOLDEN_FRAME;

94 95
  cm->comp_bwd_ref[0] = BWDREF_FRAME;
  cm->comp_bwd_ref[1] = ALTREF_FRAME;
96
#else
Jingning Han's avatar
Jingning Han committed
97
  if (cm->ref_frame_sign_bias[LAST_FRAME] ==
98
      cm->ref_frame_sign_bias[GOLDEN_FRAME]) {
Jingning Han's avatar
Jingning Han committed
99 100 101 102
    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] ==
103
             cm->ref_frame_sign_bias[ALTREF_FRAME]) {
Jingning Han's avatar
Jingning Han committed
104 105 106 107 108 109 110 111
    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;
  }
112
#endif  // CONFIG_EXT_REFS
Jingning Han's avatar
Jingning Han committed
113 114 115 116 117 118
}

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

119 120
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
121 122 123
  return data > max ? max : data;
}

124 125
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);
126
}
Jingning Han's avatar
Jingning Han committed
127

Yaowu Xu's avatar
Yaowu Xu committed
128 129 130 131 132 133 134 135
static void read_tx_size_probs(FRAME_CONTEXT *fc, aom_reader *r) {
  int i, j, k;
  for (i = 0; i < MAX_TX_DEPTH; ++i)
    for (j = 0; j < TX_SIZE_CONTEXTS; ++j)
      for (k = 0; k < i + 1; ++k)
        av1_diff_update_prob(r, &fc->tx_size_probs[i][j][k], ACCT_STR);
}

136
#if !CONFIG_EC_ADAPT
137
static void read_switchable_interp_probs(FRAME_CONTEXT *fc, aom_reader *r) {
Jingning Han's avatar
Jingning Han committed
138
  int i, j;
139
  for (j = 0; j < SWITCHABLE_FILTER_CONTEXTS; ++j) {
Jingning Han's avatar
Jingning Han committed
140
    for (i = 0; i < SWITCHABLE_FILTERS - 1; ++i)
Michael Bebenita's avatar
Michael Bebenita committed
141
      av1_diff_update_prob(r, &fc->switchable_interp_prob[j][i], ACCT_STR);
142
  }
Jingning Han's avatar
Jingning Han committed
143
}
144
#endif
Jingning Han's avatar
Jingning Han committed
145

146
static void read_inter_mode_probs(FRAME_CONTEXT *fc, aom_reader *r) {
147
#if CONFIG_REF_MV
148
  int i;
149
  for (i = 0; i < NEWMV_MODE_CONTEXTS; ++i)
Michael Bebenita's avatar
Michael Bebenita committed
150
    av1_diff_update_prob(r, &fc->newmv_prob[i], ACCT_STR);
151
  for (i = 0; i < ZEROMV_MODE_CONTEXTS; ++i)
Michael Bebenita's avatar
Michael Bebenita committed
152
    av1_diff_update_prob(r, &fc->zeromv_prob[i], ACCT_STR);
153
  for (i = 0; i < REFMV_MODE_CONTEXTS; ++i)
Michael Bebenita's avatar
Michael Bebenita committed
154
    av1_diff_update_prob(r, &fc->refmv_prob[i], ACCT_STR);
155
  for (i = 0; i < DRL_MODE_CONTEXTS; ++i)
Michael Bebenita's avatar
Michael Bebenita committed
156
    av1_diff_update_prob(r, &fc->drl_prob[i], ACCT_STR);
Yue Chen's avatar
Yue Chen committed
157
#if CONFIG_EXT_INTER
Michael Bebenita's avatar
Michael Bebenita committed
158
  av1_diff_update_prob(r, &fc->new2mv_prob, ACCT_STR);
Yue Chen's avatar
Yue Chen committed
159
#endif  // CONFIG_EXT_INTER
160
#else
161
#if !CONFIG_EC_ADAPT
162
  int i, j;
163
  for (i = 0; i < INTER_MODE_CONTEXTS; ++i) {
Jingning Han's avatar
Jingning Han committed
164
    for (j = 0; j < INTER_MODES - 1; ++j)
Michael Bebenita's avatar
Michael Bebenita committed
165
      av1_diff_update_prob(r, &fc->inter_mode_probs[i][j], ACCT_STR);
166
  }
167 168 169
#else
  (void)fc;
  (void)r;
170
#endif
171
#endif
Jingning Han's avatar
Jingning Han committed
172 173
}

174
#if !CONFIG_EC_ADAPT
175
#if CONFIG_EXT_INTER
176
static void read_inter_compound_mode_probs(FRAME_CONTEXT *fc, aom_reader *r) {
177
  int i, j;
Michael Bebenita's avatar
Michael Bebenita committed
178
  if (aom_read(r, GROUP_DIFF_UPDATE_PROB, ACCT_STR)) {
179 180
    for (j = 0; j < INTER_MODE_CONTEXTS; ++j) {
      for (i = 0; i < INTER_COMPOUND_MODES - 1; ++i) {
Michael Bebenita's avatar
Michael Bebenita committed
181
        av1_diff_update_prob(r, &fc->inter_compound_mode_probs[j][i], ACCT_STR);
182 183 184 185 186
      }
    }
  }
}
#endif  // CONFIG_EXT_INTER
187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206
#if !CONFIG_EXT_TX
static void read_ext_tx_probs(FRAME_CONTEXT *fc, aom_reader *r) {
  int i, j, k;
  if (aom_read(r, GROUP_DIFF_UPDATE_PROB, ACCT_STR)) {
    for (i = TX_4X4; i < EXT_TX_SIZES; ++i) {
      for (j = 0; j < TX_TYPES; ++j) {
        for (k = 0; k < TX_TYPES - 1; ++k)
          av1_diff_update_prob(r, &fc->intra_ext_tx_prob[i][j][k], ACCT_STR);
      }
    }
  }
  if (aom_read(r, GROUP_DIFF_UPDATE_PROB, ACCT_STR)) {
    for (i = TX_4X4; i < EXT_TX_SIZES; ++i) {
      for (k = 0; k < TX_TYPES - 1; ++k)
        av1_diff_update_prob(r, &fc->inter_ext_tx_prob[i][k], ACCT_STR);
    }
  }
}
#endif
#endif
207

208
static REFERENCE_MODE read_frame_reference_mode(
209
    const AV1_COMMON *cm, struct aom_read_bit_buffer *rb) {
210
  if (is_compound_reference_allowed(cm)) {
211
    return aom_rb_read_bit(rb)
212
               ? REFERENCE_MODE_SELECT
213
               : (aom_rb_read_bit(rb) ? COMPOUND_REFERENCE : SINGLE_REFERENCE);
214 215 216 217
  } else {
    return SINGLE_REFERENCE;
  }
}
Jingning Han's avatar
Jingning Han committed
218

219
static void read_frame_reference_mode_probs(AV1_COMMON *cm, aom_reader *r) {
Jingning Han's avatar
Jingning Han committed
220
  FRAME_CONTEXT *const fc = cm->fc;
221
  int i, j;
Jingning Han's avatar
Jingning Han committed
222 223 224

  if (cm->reference_mode == REFERENCE_MODE_SELECT)
    for (i = 0; i < COMP_INTER_CONTEXTS; ++i)
Michael Bebenita's avatar
Michael Bebenita committed
225
      av1_diff_update_prob(r, &fc->comp_inter_prob[i], ACCT_STR);
Jingning Han's avatar
Jingning Han committed
226

227
  if (cm->reference_mode != COMPOUND_REFERENCE) {
Jingning Han's avatar
Jingning Han committed
228
    for (i = 0; i < REF_CONTEXTS; ++i) {
229
      for (j = 0; j < (SINGLE_REFS - 1); ++j) {
Michael Bebenita's avatar
Michael Bebenita committed
230
        av1_diff_update_prob(r, &fc->single_ref_prob[i][j], ACCT_STR);
231
      }
Jingning Han's avatar
Jingning Han committed
232
    }
233
  }
Jingning Han's avatar
Jingning Han committed
234

235 236
  if (cm->reference_mode != SINGLE_REFERENCE) {
    for (i = 0; i < REF_CONTEXTS; ++i) {
237
#if CONFIG_EXT_REFS
238
      for (j = 0; j < (FWD_REFS - 1); ++j)
Michael Bebenita's avatar
Michael Bebenita committed
239
        av1_diff_update_prob(r, &fc->comp_ref_prob[i][j], ACCT_STR);
240
      for (j = 0; j < (BWD_REFS - 1); ++j)
Michael Bebenita's avatar
Michael Bebenita committed
241
        av1_diff_update_prob(r, &fc->comp_bwdref_prob[i][j], ACCT_STR);
242 243
#else
      for (j = 0; j < (COMP_REFS - 1); ++j)
Michael Bebenita's avatar
Michael Bebenita committed
244
        av1_diff_update_prob(r, &fc->comp_ref_prob[i][j], ACCT_STR);
245
#endif  // CONFIG_EXT_REFS
246 247
    }
  }
Jingning Han's avatar
Jingning Han committed
248 249
}

250
static void update_mv_probs(aom_prob *p, int n, aom_reader *r) {
Jingning Han's avatar
Jingning Han committed
251
  int i;
Michael Bebenita's avatar
Michael Bebenita committed
252
  for (i = 0; i < n; ++i) av1_diff_update_prob(r, &p[i], ACCT_STR);
Jingning Han's avatar
Jingning Han committed
253 254
}

255
static void read_mv_probs(nmv_context *ctx, int allow_hp, aom_reader *r) {
256
  int i;
Jingning Han's avatar
Jingning Han committed
257

Yaowu Xu's avatar
Yaowu Xu committed
258
#if !CONFIG_EC_ADAPT
259
  int j;
Jingning Han's avatar
Jingning Han committed
260 261 262 263 264 265 266 267 268 269 270
  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];
271
    for (j = 0; j < CLASS0_SIZE; ++j) {
Jingning Han's avatar
Jingning Han committed
272
      update_mv_probs(comp_ctx->class0_fp[j], MV_FP_SIZE - 1, r);
273
    }
274
    update_mv_probs(comp_ctx->fp, MV_FP_SIZE - 1, r);
Jingning Han's avatar
Jingning Han committed
275
  }
276
#endif  // !CONFIG_EC_ADAPT
Jingning Han's avatar
Jingning Han committed
277 278 279 280 281 282 283 284 285 286

  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);
    }
  }
}

287
static void inverse_transform_block(MACROBLOCKD *xd, int plane,
288
                                    const TX_TYPE tx_type,
289
                                    const TX_SIZE tx_size, uint8_t *dst,
290
                                    int stride, int16_t scan_line, int eob) {
Jingning Han's avatar
Jingning Han committed
291
  struct macroblockd_plane *const pd = &xd->plane[plane];
292 293 294 295 296 297
  tran_low_t *const dqcoeff = pd->dqcoeff;
  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;
  inv_txfm_param.lossless = xd->lossless[xd->mi[0]->mbmi.segment_id];
Angie Chiang's avatar
Angie Chiang committed
298

299
#if CONFIG_AOM_HIGHBITDEPTH
300 301 302 303
  if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
    inv_txfm_param.bd = xd->bd;
    highbd_inv_txfm_add(dqcoeff, dst, stride, &inv_txfm_param);
  } else {
304
#endif  // CONFIG_AOM_HIGHBITDEPTH
305
    inv_txfm_add(dqcoeff, dst, stride, &inv_txfm_param);
306
#if CONFIG_AOM_HIGHBITDEPTH
Jingning Han's avatar
Jingning Han committed
307
  }
308 309
#endif  // CONFIG_AOM_HIGHBITDEPTH
  memset(dqcoeff, 0, (scan_line + 1) * sizeof(dqcoeff[0]));
Jingning Han's avatar
Jingning Han committed
310 311
}

312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446
#if CONFIG_PVQ
static int av1_pvq_decode_helper(od_dec_ctx *dec, int16_t *ref_coeff,
                                 int16_t *dqcoeff, int16_t *quant, int pli,
                                 int bs, TX_TYPE tx_type, int xdec,
                                 int ac_dc_coded) {
  unsigned int flags;  // used for daala's stream analyzer.
  int off;
  const int is_keyframe = 0;
  const int has_dc_skip = 1;
  int quant_shift = bs == TX_32X32 ? 1 : 0;
  // DC quantizer for PVQ
  int pvq_dc_quant;
  int lossless = (quant[0] == 0);
  const int blk_size = tx_size_wide[bs];
  int eob = 0;
  int i;
  // TODO(yushin) : To enable activity masking,
  // int use_activity_masking = dec->use_activity_masking;
  int use_activity_masking = 0;

  DECLARE_ALIGNED(16, int16_t, dqcoeff_pvq[OD_BSIZE_MAX * OD_BSIZE_MAX]);
  DECLARE_ALIGNED(16, int16_t, ref_coeff_pvq[OD_BSIZE_MAX * OD_BSIZE_MAX]);

  od_coeff ref_int32[OD_BSIZE_MAX * OD_BSIZE_MAX];
  od_coeff out_int32[OD_BSIZE_MAX * OD_BSIZE_MAX];

  od_raster_to_coding_order(ref_coeff_pvq, blk_size, tx_type, ref_coeff,
                            blk_size);

  if (lossless)
    pvq_dc_quant = 1;
  else {
    // TODO(yushin): Enable this for activity masking,
    // when pvq_qm_q4 is available in AOM.
    // pvq_dc_quant = OD_MAXI(1, quant*
    // dec->state.pvq_qm_q4[pli][od_qm_get_index(bs, 0)] >> 4);
    pvq_dc_quant = OD_MAXI(1, quant[0] >> quant_shift);
  }

  off = od_qm_offset(bs, xdec);

  // copy int16 inputs to int32
  for (i = 0; i < blk_size * blk_size; i++) ref_int32[i] = ref_coeff_pvq[i];

  od_pvq_decode(dec, ref_int32, out_int32, (int)quant[1] >> quant_shift, pli,
                bs, OD_PVQ_BETA[use_activity_masking][pli][bs],
                OD_ROBUST_STREAM,
                is_keyframe, &flags, ac_dc_coded, dec->state.qm + off,
                dec->state.qm_inv + off);

  // copy int32 result back to int16
  for (i = 0; i < blk_size * blk_size; i++) dqcoeff_pvq[i] = out_int32[i];

  if (!has_dc_skip || dqcoeff_pvq[0]) {
    dqcoeff_pvq[0] =
        has_dc_skip + generic_decode(dec->ec, &dec->state.adapt.model_dc[pli],
                                     -1, &dec->state.adapt.ex_dc[pli][bs][0], 2,
                                     "dc:mag");
    if (dqcoeff_pvq[0])
      dqcoeff_pvq[0] *= od_ec_dec_bits(dec->ec, 1, "dc:sign") ? -1 : 1;
  }
  dqcoeff_pvq[0] = dqcoeff_pvq[0] * pvq_dc_quant + ref_coeff_pvq[0];

  od_coding_order_to_raster(dqcoeff, blk_size, tx_type, dqcoeff_pvq, blk_size);

  eob = blk_size * blk_size;

  return eob;
}

static int av1_pvq_decode_helper2(
    MACROBLOCKD *const xd, MB_MODE_INFO *const mbmi, int plane, int row,
    int col, TX_SIZE tx_size, TX_TYPE tx_type ) {
  struct macroblockd_plane *const pd = &xd->plane[plane];
  // transform block size in pixels
  int tx_blk_size = tx_size_wide[tx_size];
  int i, j;
  tran_low_t *pvq_ref_coeff = pd->pvq_ref_coeff;
  const int diff_stride = tx_blk_size;
  int16_t *pred = pd->pred;
  tran_low_t *const dqcoeff = pd->dqcoeff;
  int ac_dc_coded;  // bit0: DC coded, bit1 : AC coded
  uint8_t *dst;
  int eob;

  eob = 0;
  dst = &pd->dst.buf[4 * row * pd->dst.stride + 4 * col];

  // decode ac/dc coded flag. bit0: DC coded, bit1 : AC coded
  // NOTE : we don't use 5 symbols for luma here in aom codebase,
  // since block partition is taken care of by aom.
  // So, only AC/DC skip info is coded
  ac_dc_coded = od_decode_cdf_adapt(
      xd->daala_dec.ec,
      xd->daala_dec.state.adapt.skip_cdf[2 * tx_size + (plane != 0)], 4,
      xd->daala_dec.state.adapt.skip_increment, "skip");

  if (ac_dc_coded) {
    int xdec = pd->subsampling_x;
    int seg_id = mbmi->segment_id;
    int16_t *quant;
    FWD_TXFM_PARAM fwd_txfm_param;

    for (j = 0; j < tx_blk_size; j++)
      for (i = 0; i < tx_blk_size; i++) {
        pred[diff_stride * j + i] = dst[pd->dst.stride * j + i];
      }

    fwd_txfm_param.tx_type = tx_type;
    fwd_txfm_param.tx_size = tx_size;
    fwd_txfm_param.fwd_txfm_opt = FWD_TXFM_OPT_NORMAL;
    fwd_txfm_param.rd_transform = 0;
    fwd_txfm_param.lossless = xd->lossless[seg_id];

    fwd_txfm(pred, pvq_ref_coeff, diff_stride, &fwd_txfm_param);

    quant = &pd->seg_dequant[seg_id][0];  // aom's quantizer

    eob = av1_pvq_decode_helper(&xd->daala_dec, pvq_ref_coeff, dqcoeff, quant,
                                plane, tx_size, tx_type, xdec, ac_dc_coded);

    // Since av1 does not have separate inverse transform
    // but also contains adding to predicted image,
    // pass blank dummy image to av1_inv_txfm_add_*x*(), i.e. set dst as zeros
    for (j = 0; j < tx_blk_size; j++)
      for (i = 0; i < tx_blk_size; i++) dst[j * pd->dst.stride + i] = 0;

    inverse_transform_block(xd, plane, tx_type, tx_size, dst,
                            pd->dst.stride, eob);
  }

  return eob;
}
#endif

Angie Chiang's avatar
Angie Chiang committed
447 448
static void predict_and_reconstruct_intra_block(AV1_COMMON *cm,
                                                MACROBLOCKD *const xd,
Alex Converse's avatar
Alex Converse committed
449 450 451
#if CONFIG_ANS
                                                struct AnsDecoder *const r,
#else
452
                                                aom_reader *r,
Alex Converse's avatar
Alex Converse committed
453
#endif  // CONFIG_ANS
Jingning Han's avatar
Jingning Han committed
454
                                                MB_MODE_INFO *const mbmi,
455
                                                int plane, int row, int col,
Jingning Han's avatar
Jingning Han committed
456 457 458
                                                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
459
  PLANE_TYPE plane_type = (plane == 0) ? PLANE_TYPE_Y : PLANE_TYPE_UV;
Jingning Han's avatar
Jingning Han committed
460
  uint8_t *dst;
hui su's avatar
hui su committed
461
  int block_idx = (row << 1) + col;
462 463 464 465
#if CONFIG_PVQ
  (void)cm;
  (void)r;
#endif
Jingning Han's avatar
Jingning Han committed
466 467 468
  dst = &pd->dst.buf[4 * row * pd->dst.stride + 4 * col];

  if (mbmi->sb_type < BLOCK_8X8)
469
    if (plane == 0) mode = xd->mi[0]->bmi[(row << 1) + col].as_mode;
Jingning Han's avatar
Jingning Han committed
470

471
  av1_predict_intra_block(xd, pd->width, pd->height, tx_size, mode, dst,
472
                          pd->dst.stride, dst, pd->dst.stride, col, row, plane);
Jingning Han's avatar
Jingning Han committed
473 474

  if (!mbmi->skip) {
hui su's avatar
hui su committed
475
    TX_TYPE tx_type = get_tx_type(plane_type, xd, block_idx, tx_size);
476
#if !CONFIG_PVQ
Angie Chiang's avatar
Angie Chiang committed
477
    const SCAN_ORDER *scan_order = get_scan(cm, tx_size, tx_type, 0);
478 479 480 481
    int16_t max_scan_line = 0;
    const int eob =
        av1_decode_block_tokens(xd, plane, scan_order, col, row, tx_size,
                                tx_type, &max_scan_line, r, mbmi->segment_id);
Angie Chiang's avatar
Angie Chiang committed
482 483 484
#if CONFIG_ADAPT_SCAN
    av1_update_scan_count_facade(cm, tx_size, tx_type, pd->dqcoeff, eob);
#endif
485 486 487
    if (eob)
      inverse_transform_block(xd, plane, tx_type, tx_size, dst, pd->dst.stride,
                              max_scan_line, eob);
488 489 490
#else
    av1_pvq_decode_helper2(xd, mbmi, plane, row, col, tx_size, tx_type);
#endif
Jingning Han's avatar
Jingning Han committed
491 492 493
  }
}

494
#if CONFIG_VAR_TX
Angie Chiang's avatar
Angie Chiang committed
495 496
static void decode_reconstruct_tx(AV1_COMMON *cm, MACROBLOCKD *const xd,
                                  aom_reader *r, MB_MODE_INFO *const mbmi,
497
                                  int plane, BLOCK_SIZE plane_bsize,
498 499
                                  int blk_row, int blk_col, TX_SIZE tx_size,
                                  int *eob_total) {
500
  const struct macroblockd_plane *const pd = &xd->plane[plane];
501
  const BLOCK_SIZE bsize = txsize_to_bsize[tx_size];
502 503
  const int tx_row = blk_row >> (1 - pd->subsampling_y);
  const int tx_col = blk_col >> (1 - pd->subsampling_x);
504
  const TX_SIZE plane_tx_size =
505
      plane ? uv_txsize_lookup[bsize][mbmi->inter_tx_size[tx_row][tx_col]][0][0]
506
            : mbmi->inter_tx_size[tx_row][tx_col];
507
  // Scale to match transform block unit.
508 509
  const int max_blocks_high = max_block_high(xd, plane_bsize, plane);
  const int max_blocks_wide = max_block_wide(xd, plane_bsize, plane);
510

511
  if (blk_row >= max_blocks_high || blk_col >= max_blocks_wide) return;
512

513
  if (tx_size == plane_tx_size) {
514
    PLANE_TYPE plane_type = (plane == 0) ? PLANE_TYPE_Y : PLANE_TYPE_UV;
515 516
    int block_idx = (blk_row << 1) + blk_col;
    TX_TYPE tx_type = get_tx_type(plane_type, xd, block_idx, plane_tx_size);
Angie Chiang's avatar
Angie Chiang committed
517
    const SCAN_ORDER *sc = get_scan(cm, plane_tx_size, tx_type, 1);
518
    int16_t max_scan_line = 0;
519
    const int eob =
520
        av1_decode_block_tokens(xd, plane, sc, blk_col, blk_row, plane_tx_size,
521
                                tx_type, &max_scan_line, r, mbmi->segment_id);
522 523
    inverse_transform_block(
        xd, plane, tx_type, plane_tx_size,
524
        &pd->dst.buf[4 * blk_row * pd->dst.stride + 4 * blk_col],
525
        pd->dst.stride, max_scan_line, eob);
526 527
    *eob_total += eob;
  } else {
528 529
    const TX_SIZE sub_txs = sub_tx_size_map[tx_size];
    const int bsl = tx_size_wide_unit[sub_txs];
530 531 532 533 534
    int i;

    assert(bsl > 0);

    for (i = 0; i < 4; ++i) {
535 536
      const int offsetr = blk_row + (i >> 1) * bsl;
      const int offsetc = blk_col + (i & 0x01) * bsl;
537

538
      if (offsetr >= max_blocks_high || offsetc >= max_blocks_wide) continue;
539

540
      decode_reconstruct_tx(cm, xd, r, mbmi, plane, plane_bsize, offsetr,
541
                            offsetc, sub_txs, eob_total);
542 543 544
    }
  }
}
545
#endif  // CONFIG_VAR_TX
546

547
#if !CONFIG_VAR_TX || CONFIG_SUPERTX || (CONFIG_EXT_TX && CONFIG_RECT_TX)
Angie Chiang's avatar
Angie Chiang committed
548
static int reconstruct_inter_block(AV1_COMMON *cm, MACROBLOCKD *const xd,
Alex Converse's avatar
Alex Converse committed
549 550 551
#if CONFIG_ANS
                                   struct AnsDecoder *const r,
#else
552
                                   aom_reader *r,
Alex Converse's avatar
Alex Converse committed
553
#endif
554 555
                                   int segment_id, int plane, int row, int col,
                                   TX_SIZE tx_size) {
Jingning Han's avatar
Jingning Han committed
556
  struct macroblockd_plane *const pd = &xd->plane[plane];
hui su's avatar
hui su committed
557 558
  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
559
  TX_TYPE tx_type = get_tx_type(plane_type, xd, block_idx, tx_size);
560 561 562 563 564 565 566
#if CONFIG_PVQ
  int eob;
  (void)cm;
  (void)r;
#endif

#if !CONFIG_PVQ
Angie Chiang's avatar
Angie Chiang committed
567
  const SCAN_ORDER *scan_order = get_scan(cm, tx_size, tx_type, 1);
568 569 570 571
  int16_t max_scan_line = 0;
  const int eob =
      av1_decode_block_tokens(xd, plane, scan_order, col, row, tx_size, tx_type,
                              &max_scan_line, r, segment_id);
Angie Chiang's avatar
Angie Chiang committed
572 573 574
#if CONFIG_ADAPT_SCAN
  av1_update_scan_count_facade(cm, tx_size, tx_type, pd->dqcoeff, eob);
#endif
575 576 577 578
  if (eob)
    inverse_transform_block(xd, plane, tx_type, tx_size,
                            &pd->dst.buf[4 * row * pd->dst.stride + 4 * col],
                            pd->dst.stride, max_scan_line, eob);
579 580 581
#else
  eob = av1_pvq_decode_helper2(xd, mbmi, plane, row, col, tx_size, tx_type);
#endif
Jingning Han's avatar
Jingning Han committed
582 583
  return eob;
}
584
#endif  // !CONFIG_VAR_TX || CONFIG_SUPER_TX
Jingning Han's avatar
Jingning Han committed
585 586 587 588 589 590 591 592 593 594

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);
  }
}

595
static MB_MODE_INFO *set_offsets(AV1_COMMON *const cm, MACROBLOCKD *const xd,
Jingning Han's avatar
Jingning Han committed
596
                                 BLOCK_SIZE bsize, int mi_row, int mi_col,
597 598
                                 int bw, int bh, int x_mis, int y_mis, int bwl,
                                 int bhl) {
Jingning Han's avatar
Jingning Han committed
599 600 601 602 603 604 605 606 607
  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;
608 609 610 611
#if CONFIG_RD_DEBUG
  xd->mi[0]->mbmi.mi_row = mi_row;
  xd->mi[0]->mbmi.mi_col = mi_col;
#endif
Jingning Han's avatar
Jingning Han committed
612
  for (y = 0; y < y_mis; ++y)
Jingning Han's avatar
Jingning Han committed
613
    for (x = !y; x < x_mis; ++x) xd->mi[y * cm->mi_stride + x] = xd->mi[0];
Jingning Han's avatar
Jingning Han committed
614 615 616 617

  set_plane_n4(xd, bw, bh, bwl, bhl);
  set_skip_context(xd, mi_row, mi_col);

618 619 620 621
#if CONFIG_VAR_TX
  xd->max_tx_size = max_txsize_lookup[bsize];
#endif

Jingning Han's avatar
Jingning Han committed
622 623 624 625
  // 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);

626
  av1_setup_dst_planes(xd->plane, get_frame_new_buffer(cm), mi_row, mi_col);
Jingning Han's avatar
Jingning Han committed
627 628 629
  return &xd->mi[0]->mbmi;
}

630
#if CONFIG_SUPERTX
631
static MB_MODE_INFO *set_offsets_extend(AV1_COMMON *const cm,
632 633
                                        MACROBLOCKD *const xd,
                                        const TileInfo *const tile,
634 635 636
                                        BLOCK_SIZE bsize_pred, int mi_row_pred,
                                        int mi_col_pred, int mi_row_ori,
                                        int mi_col_ori) {
637 638 639 640 641 642 643 644 645 646
  // 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;
647 648
  set_mi_row_col(xd, tile, mi_row_pred, bh, mi_col_pred, bw, cm->mi_rows,
                 cm->mi_cols);
649

650 651
  xd->up_available = (mi_row_ori > tile->mi_row_start);
  xd->left_available = (mi_col_ori > tile->mi_col_start);
652 653 654 655 656 657

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

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

658 659 660
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) {
661 662 663 664 665 666 667 668
  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)
669
    for (x = !y; x < x_mis; ++x) xd->mi[y * cm->mi_stride + x] = xd->mi[0];
670 671 672 673 674

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

675
static void set_offsets_topblock(AV1_COMMON *const cm, MACROBLOCKD *const xd,
676 677
                                 const TileInfo *const tile, BLOCK_SIZE bsize,
                                 int mi_row, int mi_col) {
678 679 680 681 682 683 684 685 686 687 688 689 690
  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);

691
  av1_setup_dst_planes(xd->plane, get_frame_new_buffer(cm), mi_row, mi_col);
692 693
}

694
static void set_param_topblock(AV1_COMMON *const cm, MACROBLOCKD *const xd,
695
                               BLOCK_SIZE bsize, int mi_row, int mi_col,
696
                               int txfm, int skip) {
697 698
  const int bw = num_8x8_blocks_wide_lookup[bsize];
  const int bh = num_8x8_blocks_high_lookup[bsize];
699 700
  const int x_mis = AOMMIN(bw, cm->mi_cols - mi_col);
  const int y_mis = AOMMIN(bh, cm->mi_rows - mi_row);
701 702 703 704 705 706 707 708 709 710 711
  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;
    }
712 713
#if CONFIG_VAR_TX
  xd->above_txfm_context = cm->above_txfm_context + mi_col;
714
  xd->left_txfm_context =
715
      xd->left_txfm_context_buffer + (mi_row & MAX_MIB_MASK);
716
  set_txfm_ctxs(xd->mi[0]->mbmi.tx_size, bw, bh, xd);
717
#endif
718 719
}

720
static void set_ref(AV1_COMMON *const cm, MACROBLOCKD *const xd, int idx,
721
                    int mi_row, int mi_col) {
722 723 724
  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;
725 726
  if (!av1_is_valid_scale(&ref_buffer->sf))
    aom_internal_error(&cm->error, AOM_CODEC_UNSUP_BITSTREAM,
727
                       "Invalid scale factors");
728 729
  av1_setup_pre_planes(xd, idx, ref_buffer->buf, mi_row, mi_col,
                       &ref_buffer->sf);
730 731 732 733
  xd->corrupted |= ref_buffer->buf->corrupted;
}

static void dec_predict_b_extend(
734
    AV1Decoder *const pbi, MACROBLOCKD *const xd, const TileInfo *const tile,
735 736 737
    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) {
738 739 740 741 742 743 744 745 746 747 748 749
  // 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;
750
  AV1_COMMON *const cm = &pbi->common;
751 752 753

  if (mi_row_pred < mi_row_top || mi_col_pred < mi_col_top ||
      mi_row_pred >= mi_row_top + mi_height_top ||
754 755
      mi_col_pred >= mi_col_top + mi_width_top || mi_row_pred >= cm->mi_rows ||
      mi_col_pred >= cm->mi_cols)
756 757
    return;

758
  mbmi = set_offsets_extend(cm, xd, tile, bsize_pred, mi_row_pred, mi_col_pred,
759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781
                            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)
782
    av1_build_inter_predictors_sb_extend(xd,
783
#if CONFIG_EXT_INTER
784
                                         mi_row_ori, mi_col_ori,
785
#endif  // CONFIG_EXT_INTER
786
                                         mi_row_pred, mi_col_pred, bsize_pred);
787
  else
788
    av1_build_inter_predictors_sb_sub8x8_extend(xd,
789
#if CONFIG_EXT_INTER
790
                                                mi_row_ori, mi_col_ori,
791
#endif  // CONFIG_EXT_INTER
792 793
                                                mi_row_pred, mi_col_pred,
                                                bsize_pred, block);
794 795
}

796
static void dec_extend_dir(AV1Decoder *const pbi, MACROBLOCKD *const xd,
797
                           const TileInfo *const tile, int block,
798 799 800
                           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) {
801 802 803 804 805 806 807 808 809 810 811
  // 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) {
812 813 814
    extend_bsize = (mi_width == 1 || bsize < BLOCK_8X8 || xss < yss)
                       ? BLOCK_8X8
                       : BLOCK_16X8;
815 816 817 818
    unit = num_8x8_blocks_wide_lookup[extend_bsize];
    mi_row_pred = mi_row + ((dir == 0) ? mi_height : -1);
    mi_col_pred = mi_col;

819 820 821
    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);
822 823 824 825

    if (mi_width > unit) {
      int i;
      assert(!b_sub8x8);
826
      for (i = 0; i < mi_width / unit - 1; i++) {
827
        mi_col_pred += unit;
828 829 830
        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);
831 832 833
      }
    }
  } else if (dir == 2 || dir == 3) {
834 835 836
    extend_bsize = (mi_height == 1 || bsize < BLOCK_8X8 || yss < xss)
                       ? BLOCK_8X8
                       : BLOCK_8X16;
837 838 839 840
    unit = num_8x8_blocks_high_lookup[extend_bsize];
    mi_row_pred = mi_row;
    mi_col_pred = mi_col + ((dir == 3) ? mi_width : -1);

841 842 843
    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);
844 845 846

    if (mi_height > unit) {
      int i;
847
      for (i = 0; i < mi_height / unit - 1; i++) {
848
        mi_row_pred += unit;
849 850 851
        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);
852 853 854 855 856 857
      }
    }
  } 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);
858 859 860
    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);
861 862 863
  }
}

864
static void dec_extend_all(AV1Decoder *const pbi, MACROBLOCKD *const xd,
865
                           const TileInfo *const tile, int block,
866 867 868
                           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]) {
869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886
  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);
}

887
static void dec_predict_sb_complex(AV1Decoder *const pbi, MACROBLOCKD *const xd,
888 889
                                   const TileInfo *const tile, int mi_row,
                                   int mi_col, int mi_row_top, int mi_col_top,
890 891
                                   BLOCK_SIZE bsize, BLOCK_SIZE top_bsize,
                                   uint8_t *dst_buf[3], int dst_stride[3]) {
892
  const AV1_COMMON *const cm = &pbi->common;
893 894 895
  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);
896
#if CONFIG_EXT_PARTITION_TYPES
897
  const BLOCK_SIZE bsize2 = get_subsize(bsize, PARTITION_SPLIT);
898
#endif
899 900
  int i;
  const int mi_offset = mi_row * cm->mi_stride + mi_col;
901 902
  uint8_t *dst_buf1[3], *dst_buf2[3], *dst_buf3[3];

903 904 905 906 907 908
  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 };
909

910
#if CONFIG_AOM_HIGHBITDEPTH
911 912 913
  if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
    int len = sizeof(uint16_t);
    dst_buf1[0] = CONVERT_TO_BYTEPTR(tmp_buf1);
914 915
    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);
916
    dst_buf2[0] = CONVERT_TO_BYTEPTR(tmp_buf2);
917 918
    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);
919
    dst_buf3[0] = CONVERT_TO_BYTEPTR(tmp_buf3);
920 921
    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);
922 923 924
  } else {
#endif
    dst_buf1[0] = tmp_buf1;
925 926
    dst_buf1[1] = tmp_buf1 + MAX_TX_SQUARE;
    dst_buf1[2] = tmp_buf1 + 2 * MAX_TX_SQUARE;
927
    dst_buf2[0] = tmp_buf2;
928 929
    dst_buf2[1] = tmp_buf2 + MAX_TX_SQUARE;
    dst_buf2[2] = tmp_buf2 + 2 * MAX_TX_SQUARE;
930
    dst_buf3[0] = tmp_buf3;
931 932
    dst_buf3[1] = tmp_buf3 + MAX_TX_SQUARE;
    dst_buf3[2] = tmp_buf3 + 2 * MAX_TX_SQUARE;
933
#if CONFIG_AOM_HIGHBITDEPTH
934 935 936
  }
#endif

937
  if (mi_row >= cm->mi_rows || mi_col >= cm->mi_cols) return;
938

939 940
  xd->mi = cm->mi_grid_visible + mi_offset;
  xd->mi[0] = cm->mi + mi_offset;
941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977

  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];
978
        av1_build_masked_inter_predictor_complex(
979 980 981
            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);
982 983 984 985 986 987 988 989 990 991 992 993 994 995 996
      } 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,
997 998
                               mi_row + hbs, mi_col, mi_row_top, mi_col_top,
                               dst_buf1, dst_stride1, top_bsize, subsize, 0, 0);
999
          if (bsize < top_bsize)
1000 1001 1002
            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);
1003
          else
1004 1005 1006
            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);
1007 1008 1009 1010 1011

          // 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];
1012
            av1_build_masked_inter_predictor_complex(
1013
                xd, dst_buf[i], dst_stride[i], dst_buf1[i], dst_stride1[i],
1014 1015
                mi_row, mi_col, mi_row_top, mi_col_top, bsize, top_bsize,
                PARTITION_HORZ, i);
1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040
          }
        }
      }
      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];
1041
        av1_build_masked_inter_predictor_complex(
1042 1043 1044
            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);
1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058
      } 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) {
1059 1060 1061
          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);
1062
          if (bsize < top_bsize)
1063 1064 1065
            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);
1066
          else
1067 1068 1069
            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);
1070 1071 1072 1073 1074

          // 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];
1075
            av1_build_masked_inter_predictor_complex(