encodemb.c 62.6 KB
Newer Older
Jingning Han's avatar
Jingning Han committed
1
/*
Yaowu Xu's avatar
Yaowu Xu committed
2
 * Copyright (c) 2016, Alliance for Open Media. All rights reserved
Jingning Han's avatar
Jingning Han committed
3
 *
Yaowu Xu's avatar
Yaowu Xu committed
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
 */

Yaowu Xu's avatar
Yaowu Xu committed
12 13 14
#include "./av1_rtcd.h"
#include "./aom_config.h"
#include "./aom_dsp_rtcd.h"
Jingning Han's avatar
Jingning Han committed
15

16
#include "aom_dsp/bitwriter.h"
17
#include "aom_dsp/quantize.h"
Yaowu Xu's avatar
Yaowu Xu committed
18
#include "aom_mem/aom_mem.h"
19
#include "aom_ports/mem.h"
Jingning Han's avatar
Jingning Han committed
20

21 22 23 24
#include "av1/common/idct.h"
#include "av1/common/reconinter.h"
#include "av1/common/reconintra.h"
#include "av1/common/scan.h"
Jingning Han's avatar
Jingning Han committed
25

26
#include "av1/encoder/av1_quantize.h"
27
#include "av1/encoder/encodemb.h"
28 29 30
#if CONFIG_LV_MAP
#include "av1/encoder/encodetxb.h"
#endif
31 32 33
#include "av1/encoder/hybrid_fwd_txfm.h"
#include "av1/encoder/rd.h"
#include "av1/encoder/tokenize.h"
Jingning Han's avatar
Jingning Han committed
34

35 36 37 38 39 40
#if CONFIG_PVQ
#include "av1/encoder/encint.h"
#include "av1/common/partition.h"
#include "av1/encoder/pvq_encoder.h"
#endif

41 42 43 44
#if CONFIG_CFL
#include "av1/common/cfl.h"
#endif

45 46 47
// Check if one needs to use c version subtraction.
static int check_subtract_block_size(int w, int h) { return w < 4 || h < 4; }

Angie Chiang's avatar
Angie Chiang committed
48 49 50 51
static void subtract_block(const MACROBLOCKD *xd, int rows, int cols,
                           int16_t *diff, ptrdiff_t diff_stride,
                           const uint8_t *src8, ptrdiff_t src_stride,
                           const uint8_t *pred8, ptrdiff_t pred_stride) {
52
#if !CONFIG_HIGHBITDEPTH
Angie Chiang's avatar
Angie Chiang committed
53 54
  (void)xd;
#endif
Jingning Han's avatar
Jingning Han committed
55

Angie Chiang's avatar
Angie Chiang committed
56
  if (check_subtract_block_size(rows, cols)) {
57
#if CONFIG_HIGHBITDEPTH
Angie Chiang's avatar
Angie Chiang committed
58 59 60
    if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
      aom_highbd_subtract_block_c(rows, cols, diff, diff_stride, src8,
                                  src_stride, pred8, pred_stride, xd->bd);
61 62
      return;
    }
63
#endif  // CONFIG_HIGHBITDEPTH
Angie Chiang's avatar
Angie Chiang committed
64 65
    aom_subtract_block_c(rows, cols, diff, diff_stride, src8, src_stride, pred8,
                         pred_stride);
66 67 68 69

    return;
  }

70
#if CONFIG_HIGHBITDEPTH
Angie Chiang's avatar
Angie Chiang committed
71 72 73
  if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
    aom_highbd_subtract_block(rows, cols, diff, diff_stride, src8, src_stride,
                              pred8, pred_stride, xd->bd);
Jingning Han's avatar
Jingning Han committed
74 75
    return;
  }
76
#endif  // CONFIG_HIGHBITDEPTH
Angie Chiang's avatar
Angie Chiang committed
77 78 79 80
  aom_subtract_block(rows, cols, diff, diff_stride, src8, src_stride, pred8,
                     pred_stride);
}

Angie Chiang's avatar
Angie Chiang committed
81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100
void av1_subtract_txb(MACROBLOCK *x, int plane, BLOCK_SIZE plane_bsize,
                      int blk_col, int blk_row, TX_SIZE tx_size) {
  MACROBLOCKD *const xd = &x->e_mbd;
  struct macroblock_plane *const p = &x->plane[plane];
  const struct macroblockd_plane *const pd = &x->e_mbd.plane[plane];
  const int diff_stride = block_size_wide[plane_bsize];
  const int src_stride = p->src.stride;
  const int dst_stride = pd->dst.stride;
  const int tx1d_width = tx_size_wide[tx_size];
  const int tx1d_height = tx_size_high[tx_size];
  uint8_t *dst =
      &pd->dst.buf[(blk_row * dst_stride + blk_col) << tx_size_wide_log2[0]];
  uint8_t *src =
      &p->src.buf[(blk_row * src_stride + blk_col) << tx_size_wide_log2[0]];
  int16_t *src_diff =
      &p->src_diff[(blk_row * diff_stride + blk_col) << tx_size_wide_log2[0]];
  subtract_block(xd, tx1d_height, tx1d_width, src_diff, diff_stride, src,
                 src_stride, dst, dst_stride);
}

Angie Chiang's avatar
Angie Chiang committed
101 102 103 104 105 106 107 108 109 110
void av1_subtract_plane(MACROBLOCK *x, BLOCK_SIZE bsize, int plane) {
  struct macroblock_plane *const p = &x->plane[plane];
  const struct macroblockd_plane *const pd = &x->e_mbd.plane[plane];
  const BLOCK_SIZE plane_bsize = get_plane_block_size(bsize, pd);
  const int bw = block_size_wide[plane_bsize];
  const int bh = block_size_high[plane_bsize];
  const MACROBLOCKD *xd = &x->e_mbd;

  subtract_block(xd, bh, bw, p->src_diff, bw, p->src.buf, p->src.stride,
                 pd->dst.buf, pd->dst.stride);
Jingning Han's avatar
Jingning Han committed
111 112
}

113 114
// These numbers are empirically obtained.
static const int plane_rd_mult[REF_TYPES][PLANE_TYPES] = {
Thomas Davies's avatar
Thomas Davies committed
115 116 117
#if CONFIG_EC_ADAPT
  { 10, 7 }, { 8, 5 },
#else
118
  { 10, 6 }, { 8, 6 },
Thomas Davies's avatar
Thomas Davies committed
119
#endif
120
};
Jingning Han's avatar
Jingning Han committed
121

122 123 124 125
#define UPDATE_RD_COST()                      \
  {                                           \
    rd_cost0 = RDCOST(rdmult, rate0, error0); \
    rd_cost1 = RDCOST(rdmult, rate1, error1); \
126
  }
Jingning Han's avatar
Jingning Han committed
127

128 129 130
static INLINE unsigned int get_token_bit_costs(
    unsigned int token_costs[2][COEFF_CONTEXTS][ENTROPY_TOKENS], int skip_eob,
    int ctx, int token) {
131 132 133 134
  (void)skip_eob;
  return token_costs[token == ZERO_TOKEN || token == EOB_TOKEN][ctx][token];
}

135
#if !CONFIG_LV_MAP
136

137
typedef struct av1_token_state_greedy {
138 139 140
  int16_t token;
  tran_low_t qc;
  tran_low_t dqc;
141
} av1_token_state_greedy;
142

143 144
static int optimize_b_greedy(const AV1_COMMON *cm, MACROBLOCK *mb, int plane,
                             int block, TX_SIZE tx_size, int ctx) {
145 146 147 148
  MACROBLOCKD *const xd = &mb->e_mbd;
  struct macroblock_plane *const p = &mb->plane[plane];
  struct macroblockd_plane *const pd = &xd->plane[plane];
  const int ref = is_inter_block(&xd->mi[0]->mbmi);
149
  av1_token_state_greedy tokens[MAX_TX_SQUARE + 1][2];
150 151 152 153 154 155 156 157 158 159 160 161 162 163
  uint8_t token_cache[MAX_TX_SQUARE];
  const tran_low_t *const coeff = BLOCK_OFFSET(p->coeff, block);
  tran_low_t *const qcoeff = BLOCK_OFFSET(p->qcoeff, block);
  tran_low_t *const dqcoeff = BLOCK_OFFSET(pd->dqcoeff, block);
  const int eob = p->eobs[block];
  const PLANE_TYPE plane_type = pd->plane_type;
  const int16_t *const dequant_ptr = pd->dequant;
  const uint8_t *const band_translate = get_band_translate(tx_size);
  TX_TYPE tx_type = get_tx_type(plane_type, xd, block, tx_size);
  const SCAN_ORDER *const scan_order =
      get_scan(cm, tx_size, tx_type, is_inter_block(&xd->mi[0]->mbmi));
  const int16_t *const scan = scan_order->scan;
  const int16_t *const nb = scan_order->neighbors;
  int dqv;
164
  const int shift = av1_get_tx_scale(tx_size);
165 166
#if CONFIG_AOM_QM
  int seg_id = xd->mi[0]->mbmi.segment_id;
167 168 169 170 171
  // Use a flat matrix (i.e. no weighting) for 1D and Identity transforms
  const qm_val_t *iqmatrix =
      IS_2D_TRANSFORM(tx_type)
          ? pd->seg_iqmatrix[seg_id][!ref][tx_size]
          : cm->giqmatrix[NUM_QM_LEVELS - 1][0][0][tx_size];
172 173 174 175 176 177 178 179 180 181 182 183 184 185
#endif
#if CONFIG_NEW_QUANT
  int dq = get_dq_profile_from_ctx(mb->qindex, ctx, ref, plane_type);
  const dequant_val_type_nuq *dequant_val = pd->dequant_val_nuq[dq];
#endif  // CONFIG_NEW_QUANT
  int sz = 0;
  int64_t rd_cost0, rd_cost1;
  int16_t t0, t1;
  int i, final_eob;
  const int cat6_bits = av1_get_cat6_extrabits_size(tx_size, xd->bd);
  unsigned int(*token_costs)[2][COEFF_CONTEXTS][ENTROPY_TOKENS] =
      mb->token_costs[txsize_sqr_map[tx_size]][plane_type][ref];
  const int default_eob = tx_size_2d[tx_size];

186
  assert(mb->qindex > 0);
187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219

  assert((!plane_type && !plane) || (plane_type && plane));
  assert(eob <= default_eob);

  int64_t rdmult = (mb->rdmult * plane_rd_mult[ref][plane_type]) >> 1;

  int64_t rate0, rate1;
  for (i = 0; i < eob; i++) {
    const int rc = scan[i];
    int x = qcoeff[rc];
    t0 = av1_get_token(x);

    tokens[i][0].qc = x;
    tokens[i][0].token = t0;
    tokens[i][0].dqc = dqcoeff[rc];

    token_cache[rc] = av1_pt_energy_class[t0];
  }
  tokens[eob][0].token = EOB_TOKEN;
  tokens[eob][0].qc = 0;
  tokens[eob][0].dqc = 0;
  tokens[eob][1] = tokens[eob][0];

  unsigned int(*token_costs_ptr)[2][COEFF_CONTEXTS][ENTROPY_TOKENS] =
      token_costs;

  final_eob = 0;

  int64_t eob_cost0, eob_cost1;

  const int ctx0 = ctx;
  /* Record the r-d cost */
  int64_t accu_rate = 0;
220 221 222
  // Initialized to the worst possible error for the largest transform size.
  // This ensures that it never goes negative.
  int64_t accu_error = ((int64_t)1) << 50;
223 224 225

  rate0 = get_token_bit_costs(*(token_costs_ptr + band_translate[0]), 0, ctx0,
                              EOB_TOKEN);
226
  int64_t best_block_rd_cost = RDCOST(rdmult, rate0, accu_error);
227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253

  // int64_t best_block_rd_cost_all0 = best_block_rd_cost;

  int x_prev = 1;

  for (i = 0; i < eob; i++) {
    const int rc = scan[i];
    int x = qcoeff[rc];
    sz = -(x < 0);

    int band_cur = band_translate[i];
    int ctx_cur = (i == 0) ? ctx : get_coef_context(nb, token_cache, i);
    int token_tree_sel_cur = (x_prev == 0);

    if (x == 0) {
      // no need to search when x == 0
      rate0 =
          get_token_bit_costs(*(token_costs_ptr + band_cur), token_tree_sel_cur,
                              ctx_cur, tokens[i][0].token);
      accu_rate += rate0;
      x_prev = 0;
      // accu_error does not change when x==0
    } else {
      /*  Computing distortion
       */
      // compute the distortion for the first candidate
      // and the distortion for quantizing to 0.
254
      int dx0 = abs(coeff[rc]) * (1 << shift);
255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277
#if CONFIG_HIGHBITDEPTH
      if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
        dx0 >>= xd->bd - 8;
      }
#endif
      int64_t d0 = (int64_t)dx0 * dx0;

      int x_a = x - 2 * sz - 1;
      int64_t d2, d2_a;

      int dx;

#if CONFIG_AOM_QM
      int iwt = iqmatrix[rc];
      dqv = dequant_ptr[rc != 0];
      dqv = ((iwt * (int)dqv) + (1 << (AOM_QM_BITS - 1))) >> AOM_QM_BITS;
#else
      dqv = dequant_ptr[rc != 0];
#endif

      dx = (dqcoeff[rc] - coeff[rc]) * (1 << shift);
#if CONFIG_HIGHBITDEPTH
      if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
278 279 280
        int dx_sign = dx < 0 ? 1 : 0;
        dx = abs(dx) >> (xd->bd - 8);
        if (dx_sign) dx = -dx;
281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362
      }
#endif  // CONFIG_HIGHBITDEPTH
      d2 = (int64_t)dx * dx;

      /* compute the distortion for the second candidate
       * x_a = x - 2 * sz + 1;
       */
      if (x_a != 0) {
#if CONFIG_NEW_QUANT
        dx = av1_dequant_coeff_nuq(x, dqv, dequant_val[band_translate[i]]) -
             (coeff[rc] << shift);
#if CONFIG_HIGHBITDEPTH
        if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
          dx >>= xd->bd - 8;
        }
#endif  // CONFIG_HIGHBITDEPTH
#else   // CONFIG_NEW_QUANT
#if CONFIG_HIGHBITDEPTH
        if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
          dx -= ((dqv >> (xd->bd - 8)) + sz) ^ sz;
        } else {
          dx -= (dqv + sz) ^ sz;
        }
#else
        dx -= (dqv + sz) ^ sz;
#endif  // CONFIG_HIGHBITDEPTH
#endif  // CONFIG_NEW_QUANT
        d2_a = (int64_t)dx * dx;
      } else {
        d2_a = d0;
      }
      /*  Computing rates and r-d cost
       */

      int best_x, best_eob_x;
      int64_t base_bits, next_bits0, next_bits1;
      int64_t next_eob_bits0, next_eob_bits1;

      // rate cost of x
      base_bits = av1_get_token_cost(x, &t0, cat6_bits);
      rate0 = base_bits + get_token_bit_costs(*(token_costs_ptr + band_cur),
                                              token_tree_sel_cur, ctx_cur, t0);

      base_bits = av1_get_token_cost(x_a, &t1, cat6_bits);
      rate1 = base_bits + get_token_bit_costs(*(token_costs_ptr + band_cur),
                                              token_tree_sel_cur, ctx_cur, t1);

      next_bits0 = 0;
      next_bits1 = 0;
      next_eob_bits0 = 0;
      next_eob_bits1 = 0;

      if (i < default_eob - 1) {
        int ctx_next, token_tree_sel_next;
        int band_next = band_translate[i + 1];

        token_cache[rc] = av1_pt_energy_class[t0];
        ctx_next = get_coef_context(nb, token_cache, i + 1);
        token_tree_sel_next = (x == 0);

        next_bits0 = get_token_bit_costs(*(token_costs_ptr + band_next),
                                         token_tree_sel_next, ctx_next,
                                         tokens[i + 1][0].token);
        next_eob_bits0 =
            get_token_bit_costs(*(token_costs_ptr + band_next),
                                token_tree_sel_next, ctx_next, EOB_TOKEN);

        token_cache[rc] = av1_pt_energy_class[t1];
        ctx_next = get_coef_context(nb, token_cache, i + 1);
        token_tree_sel_next = (x_a == 0);

        next_bits1 = get_token_bit_costs(*(token_costs_ptr + band_next),
                                         token_tree_sel_next, ctx_next,
                                         tokens[i + 1][0].token);

        if (x_a != 0) {
          next_eob_bits1 =
              get_token_bit_costs(*(token_costs_ptr + band_next),
                                  token_tree_sel_next, ctx_next, EOB_TOKEN);
        }
      }

363 364
      rd_cost0 = RDCOST(rdmult, (rate0 + next_bits0), d2);
      rd_cost1 = RDCOST(rdmult, (rate1 + next_bits1), d2_a);
365 366 367

      best_x = (rd_cost1 < rd_cost0);

368
      eob_cost0 = RDCOST(rdmult, (accu_rate + rate0 + next_eob_bits0),
369 370 371
                         (accu_error + d2 - d0));
      eob_cost1 = eob_cost0;
      if (x_a != 0) {
372
        eob_cost1 = RDCOST(rdmult, (accu_rate + rate1 + next_eob_bits1),
373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389
                           (accu_error + d2_a - d0));
        best_eob_x = (eob_cost1 < eob_cost0);
      } else {
        best_eob_x = 0;
      }

      int dqc, dqc_a = 0;

      dqc = dqcoeff[rc];
      if (best_x + best_eob_x) {
        if (x_a != 0) {
#if CONFIG_NEW_QUANT
          dqc_a = av1_dequant_abscoeff_nuq(abs(x_a), dqv,
                                           dequant_val[band_translate[i]]);
          dqc_a = shift ? ROUND_POWER_OF_TWO(dqc_a, shift) : dqc_a;
          if (sz) dqc_a = -dqc_a;
#else
390 391
          if (x_a < 0)
            dqc_a = -((-x_a * dqv) >> shift);
392
          else
393
            dqc_a = (x_a * dqv) >> shift;
394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416
#endif  // CONFIG_NEW_QUANT
        } else {
          dqc_a = 0;
        }  // if (x_a != 0)
      }

      // record the better quantized value
      if (best_x) {
        qcoeff[rc] = x_a;
        dqcoeff[rc] = dqc_a;

        accu_rate += rate1;
        accu_error += d2_a - d0;
        assert(d2_a <= d0);

        token_cache[rc] = av1_pt_energy_class[t1];
      } else {
        accu_rate += rate0;
        accu_error += d2 - d0;
        assert(d2 <= d0);

        token_cache[rc] = av1_pt_energy_class[t0];
      }
417
      assert(accu_error >= 0);
418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460

      x_prev = qcoeff[rc];

      // determine whether to move the eob position to i+1
      int64_t best_eob_cost_i = eob_cost0;

      tokens[i][1].token = t0;
      tokens[i][1].qc = x;
      tokens[i][1].dqc = dqc;

      if ((x_a != 0) && (best_eob_x)) {
        best_eob_cost_i = eob_cost1;

        tokens[i][1].token = t1;
        tokens[i][1].qc = x_a;
        tokens[i][1].dqc = dqc_a;
      }

      if (best_eob_cost_i < best_block_rd_cost) {
        best_block_rd_cost = best_eob_cost_i;
        final_eob = i + 1;
      }
    }  // if (x==0)
  }    // for (i)

  assert(final_eob <= eob);
  if (final_eob > 0) {
    assert(tokens[final_eob - 1][1].qc != 0);
    i = final_eob - 1;
    int rc = scan[i];
    qcoeff[rc] = tokens[i][1].qc;
    dqcoeff[rc] = tokens[i][1].dqc;
  }

  for (i = final_eob; i < eob; i++) {
    int rc = scan[i];
    qcoeff[rc] = 0;
    dqcoeff[rc] = 0;
  }

  mb->plane[plane].eobs[block] = final_eob;
  return final_eob;
}
461
#endif  // !CONFIG_LV_MAP
462 463

int av1_optimize_b(const AV1_COMMON *cm, MACROBLOCK *mb, int plane, int block,
464 465
                   BLOCK_SIZE plane_bsize, TX_SIZE tx_size,
                   const ENTROPY_CONTEXT *a, const ENTROPY_CONTEXT *l) {
466 467 468 469 470 471
  MACROBLOCKD *const xd = &mb->e_mbd;
  struct macroblock_plane *const p = &mb->plane[plane];
  const int eob = p->eobs[block];
  assert((mb->qindex == 0) ^ (xd->lossless[xd->mi[0]->mbmi.segment_id] == 0));
  if (eob == 0) return eob;
  if (xd->lossless[xd->mi[0]->mbmi.segment_id]) return eob;
Jingning Han's avatar
Jingning Han committed
472

473
#if CONFIG_PVQ
474 475
  (void)cm;
  (void)tx_size;
476 477
  (void)a;
  (void)l;
478 479
  return eob;
#endif
Jingning Han's avatar
Jingning Han committed
480

481
#if !CONFIG_LV_MAP
482
  (void)plane_bsize;
483 484 485 486
#if CONFIG_VAR_TX
  int ctx = get_entropy_context(tx_size, a, l);
#else
  int ctx = combine_entropy_contexts(*a, *l);
487
#endif  // CONFIG_VAR_TX
488
  return optimize_b_greedy(cm, mb, plane, block, tx_size, ctx);
489 490 491 492 493
#else   // !CONFIG_LV_MAP
  TXB_CTX txb_ctx;
  get_txb_ctx(plane_bsize, tx_size, plane, a, l, &txb_ctx);
  return av1_optimize_txb(cm, mb, plane, block, tx_size, &txb_ctx);
#endif  // !CONFIG_LV_MAP
494
}
495

Thomas Daede's avatar
Thomas Daede committed
496
#if !CONFIG_PVQ
497
#if CONFIG_HIGHBITDEPTH
Angie Chiang's avatar
Angie Chiang committed
498 499
typedef enum QUANT_FUNC {
  QUANT_FUNC_LOWBD = 0,
500
  QUANT_FUNC_HIGHBD = 1,
501
  QUANT_FUNC_TYPES = 2
Angie Chiang's avatar
Angie Chiang committed
502 503
} QUANT_FUNC;

504 505
static AV1_QUANT_FACADE
    quant_func_list[AV1_XFORM_QUANT_TYPES][QUANT_FUNC_TYPES] = {
506
#if !CONFIG_NEW_QUANT
507
      { av1_quantize_fp_facade, av1_highbd_quantize_fp_facade },
Yaowu Xu's avatar
Yaowu Xu committed
508 509
      { av1_quantize_b_facade, av1_highbd_quantize_b_facade },
      { av1_quantize_dc_facade, av1_highbd_quantize_dc_facade },
510
#else   // !CONFIG_NEW_QUANT
511 512 513
      { av1_quantize_fp_nuq_facade, av1_highbd_quantize_fp_nuq_facade },
      { av1_quantize_b_nuq_facade, av1_highbd_quantize_b_nuq_facade },
      { av1_quantize_dc_nuq_facade, av1_highbd_quantize_dc_nuq_facade },
514
#endif  // !CONFIG_NEW_QUANT
515 516
      { NULL, NULL }
    };
517

Thomas Daede's avatar
Thomas Daede committed
518
#else
519

Angie Chiang's avatar
Angie Chiang committed
520 521
typedef enum QUANT_FUNC {
  QUANT_FUNC_LOWBD = 0,
522
  QUANT_FUNC_TYPES = 1
Angie Chiang's avatar
Angie Chiang committed
523
} QUANT_FUNC;
Angie Chiang's avatar
Angie Chiang committed
524

525 526
static AV1_QUANT_FACADE quant_func_list[AV1_XFORM_QUANT_TYPES]
                                       [QUANT_FUNC_TYPES] = {
527
#if !CONFIG_NEW_QUANT
clang-format's avatar
clang-format committed
528 529 530
                                         { av1_quantize_fp_facade },
                                         { av1_quantize_b_facade },
                                         { av1_quantize_dc_facade },
531
#else   // !CONFIG_NEW_QUANT
532 533 534
                                         { av1_quantize_fp_nuq_facade },
                                         { av1_quantize_b_nuq_facade },
                                         { av1_quantize_dc_nuq_facade },
535
#endif  // !CONFIG_NEW_QUANT
clang-format's avatar
clang-format committed
536 537
                                         { NULL }
                                       };
538
#endif  // CONFIG_HIGHBITDEPTH
Thomas Daede's avatar
Thomas Daede committed
539
#endif  // CONFIG_PVQ
540

Angie Chiang's avatar
Angie Chiang committed
541 542
void av1_xform_quant(const AV1_COMMON *cm, MACROBLOCK *x, int plane, int block,
                     int blk_row, int blk_col, BLOCK_SIZE plane_bsize,
543 544
                     TX_SIZE tx_size, int ctx,
                     AV1_XFORM_QUANT xform_quant_idx) {
Jingning Han's avatar
Jingning Han committed
545
  MACROBLOCKD *const xd = &x->e_mbd;
546
  MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;
Yushin Cho's avatar
Yushin Cho committed
547
#if !(CONFIG_PVQ || CONFIG_DAALA_DIST)
Jingning Han's avatar
Jingning Han committed
548 549
  const struct macroblock_plane *const p = &x->plane[plane];
  const struct macroblockd_plane *const pd = &xd->plane[plane];
550 551 552 553
#else
  struct macroblock_plane *const p = &x->plane[plane];
  struct macroblockd_plane *const pd = &xd->plane[plane];
#endif
554
  PLANE_TYPE plane_type = get_plane_type(plane);
555
  TX_TYPE tx_type = get_tx_type(plane_type, xd, block, tx_size);
556
  const int is_inter = is_inter_block(mbmi);
Angie Chiang's avatar
Angie Chiang committed
557
  const SCAN_ORDER *const scan_order = get_scan(cm, tx_size, tx_type, is_inter);
Jingning Han's avatar
Jingning Han committed
558 559 560 561
  tran_low_t *const coeff = BLOCK_OFFSET(p->coeff, block);
  tran_low_t *const qcoeff = BLOCK_OFFSET(p->qcoeff, block);
  tran_low_t *const dqcoeff = BLOCK_OFFSET(pd->dqcoeff, block);
  uint16_t *const eob = &p->eobs[block];
562
  const int diff_stride = block_size_wide[plane_bsize];
563
#if CONFIG_AOM_QM
564
  int seg_id = mbmi->segment_id;
565 566 567 568 569 570 571 572
  // Use a flat matrix (i.e. no weighting) for 1D and Identity transforms
  const qm_val_t *qmatrix =
      IS_2D_TRANSFORM(tx_type) ? pd->seg_qmatrix[seg_id][!is_inter][tx_size]
                               : cm->gqmatrix[NUM_QM_LEVELS - 1][0][0][tx_size];
  const qm_val_t *iqmatrix =
      IS_2D_TRANSFORM(tx_type)
          ? pd->seg_iqmatrix[seg_id][!is_inter][tx_size]
          : cm->giqmatrix[NUM_QM_LEVELS - 1][0][0][tx_size];
573
#endif
Angie Chiang's avatar
Angie Chiang committed
574 575

  FWD_TXFM_PARAM fwd_txfm_param;
576

Yushin Cho's avatar
Yushin Cho committed
577 578 579 580
#if CONFIG_PVQ || CONFIG_DAALA_DIST
  uint8_t *dst;
  int16_t *pred;
  const int dst_stride = pd->dst.stride;
581 582
  const int txw = tx_size_wide[tx_size];
  const int txh = tx_size_high[tx_size];
Yushin Cho's avatar
Yushin Cho committed
583 584 585
  int i, j;
#endif

586 587
#if !CONFIG_PVQ
  const int tx2d_size = tx_size_2d[tx_size];
588
  QUANT_PARAM qparam;
589 590
  const int16_t *src_diff;

591 592
  src_diff =
      &p->src_diff[(blk_row * diff_stride + blk_col) << tx_size_wide_log2[0]];
593
  qparam.log_scale = av1_get_tx_scale(tx_size);
594 595 596 597 598 599 600 601
#if CONFIG_NEW_QUANT
  qparam.tx_size = tx_size;
  qparam.dq = get_dq_profile_from_ctx(x->qindex, ctx, is_inter, plane_type);
#endif  // CONFIG_NEW_QUANT
#if CONFIG_AOM_QM
  qparam.qmatrix = qmatrix;
  qparam.iqmatrix = iqmatrix;
#endif  // CONFIG_AOM_QM
602 603 604 605
#else
  tran_low_t *ref_coeff = BLOCK_OFFSET(pd->pvq_ref_coeff, block);
  int skip = 1;
  PVQ_INFO *pvq_info = NULL;
Yushin Cho's avatar
Yushin Cho committed
606 607 608
  uint8_t *src;
  int16_t *src_int16;
  const int src_stride = p->src.stride;
609

610
  (void)ctx;
611 612 613 614 615 616 617
  (void)scan_order;
  (void)qcoeff;

  if (x->pvq_coded) {
    assert(block < MAX_PVQ_BLOCKS_IN_SB);
    pvq_info = &x->pvq[block][plane];
  }
618 619 620
  src = &p->src.buf[(blk_row * src_stride + blk_col) << tx_size_wide_log2[0]];
  src_int16 =
      &p->src_int16[(blk_row * diff_stride + blk_col) << tx_size_wide_log2[0]];
Yushin Cho's avatar
Yushin Cho committed
621

622
#if CONFIG_HIGHBITDEPTH
Thomas Daede's avatar
Thomas Daede committed
623
  if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
624 625
    for (j = 0; j < txh; j++)
      for (i = 0; i < txw; i++)
Thomas Daede's avatar
Thomas Daede committed
626 627 628
        src_int16[diff_stride * j + i] =
            CONVERT_TO_SHORTPTR(src)[src_stride * j + i];
  } else {
629
#endif  // CONFIG_HIGHBITDEPTH
630 631
    for (j = 0; j < txh; j++)
      for (i = 0; i < txw; i++)
Thomas Daede's avatar
Thomas Daede committed
632
        src_int16[diff_stride * j + i] = src[src_stride * j + i];
633
#if CONFIG_HIGHBITDEPTH
Thomas Daede's avatar
Thomas Daede committed
634
  }
635
#endif  // CONFIG_HIGHBITDEPTH
Yushin Cho's avatar
Yushin Cho committed
636 637 638 639
#endif

#if CONFIG_PVQ || CONFIG_DAALA_DIST
  dst = &pd->dst.buf[(blk_row * dst_stride + blk_col) << tx_size_wide_log2[0]];
640
  pred = &pd->pred[(blk_row * diff_stride + blk_col) << tx_size_wide_log2[0]];
641

Thomas Daede's avatar
Thomas Daede committed
642 643
// copy uint8 orig and predicted block to int16 buffer
// in order to use existing VP10 transform functions
644
#if CONFIG_HIGHBITDEPTH
Thomas Daede's avatar
Thomas Daede committed
645
  if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
646 647
    for (j = 0; j < txh; j++)
      for (i = 0; i < txw; i++)
Thomas Daede's avatar
Thomas Daede committed
648 649 650
        pred[diff_stride * j + i] =
            CONVERT_TO_SHORTPTR(dst)[dst_stride * j + i];
  } else {
651
#endif  // CONFIG_HIGHBITDEPTH
652 653
    for (j = 0; j < txh; j++)
      for (i = 0; i < txw; i++)
Thomas Daede's avatar
Thomas Daede committed
654
        pred[diff_stride * j + i] = dst[dst_stride * j + i];
655
#if CONFIG_HIGHBITDEPTH
Thomas Daede's avatar
Thomas Daede committed
656
  }
657
#endif  // CONFIG_HIGHBITDEPTH
658
#endif
Yushin Cho's avatar
Yushin Cho committed
659

660
  (void)ctx;
661 662

  fwd_txfm_param.tx_type = tx_type;
Angie Chiang's avatar
Angie Chiang committed
663
  fwd_txfm_param.tx_size = tx_size;
664
  fwd_txfm_param.lossless = xd->lossless[mbmi->segment_id];
Angie Chiang's avatar
Angie Chiang committed
665

Thomas Daede's avatar
Thomas Daede committed
666
#if !CONFIG_PVQ
667
#if CONFIG_HIGHBITDEPTH
668
  fwd_txfm_param.bd = xd->bd;
Jingning Han's avatar
Jingning Han committed
669
  if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
hui su's avatar
hui su committed
670
    av1_highbd_fwd_txfm(src_diff, coeff, diff_stride, &fwd_txfm_param);
Yaowu Xu's avatar
Yaowu Xu committed
671
    if (xform_quant_idx != AV1_XFORM_QUANT_SKIP_QUANT) {
672
      if (LIKELY(!x->skip_block)) {
673
        quant_func_list[xform_quant_idx][QUANT_FUNC_HIGHBD](
674
            coeff, tx2d_size, p, qcoeff, pd, dqcoeff, eob, scan_order, &qparam);
675
      } else {
Yaowu Xu's avatar
Yaowu Xu committed
676
        av1_quantize_skip(tx2d_size, qcoeff, dqcoeff, eob);
677 678
      }
    }
679 680 681 682
#if CONFIG_LV_MAP
    p->txb_entropy_ctx[block] =
        (uint8_t)av1_get_txb_entropy_context(qcoeff, scan_order, *eob);
#endif  // CONFIG_LV_MAP
683 684
    return;
  }
685
#endif  // CONFIG_HIGHBITDEPTH
hui su's avatar
hui su committed
686
  av1_fwd_txfm(src_diff, coeff, diff_stride, &fwd_txfm_param);
Yaowu Xu's avatar
Yaowu Xu committed
687
  if (xform_quant_idx != AV1_XFORM_QUANT_SKIP_QUANT) {
688
    if (LIKELY(!x->skip_block)) {
689
      quant_func_list[xform_quant_idx][QUANT_FUNC_LOWBD](
690
          coeff, tx2d_size, p, qcoeff, pd, dqcoeff, eob, scan_order, &qparam);
691
    } else {
Yaowu Xu's avatar
Yaowu Xu committed
692
      av1_quantize_skip(tx2d_size, qcoeff, dqcoeff, eob);
Jingning Han's avatar
Jingning Han committed
693 694
    }
  }
695 696 697 698 699
#if CONFIG_LV_MAP
  p->txb_entropy_ctx[block] =
      (uint8_t)av1_get_txb_entropy_context(qcoeff, scan_order, *eob);
#endif  // CONFIG_LV_MAP
#else   // #if !CONFIG_PVQ
Angie Chiang's avatar
Angie Chiang committed
700
  (void)xform_quant_idx;
701
#if CONFIG_HIGHBITDEPTH
Thomas Daede's avatar
Thomas Daede committed
702 703
  fwd_txfm_param.bd = xd->bd;
  if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
hui su's avatar
hui su committed
704 705
    av1_highbd_fwd_txfm(src_int16, coeff, diff_stride, &fwd_txfm_param);
    av1_highbd_fwd_txfm(pred, ref_coeff, diff_stride, &fwd_txfm_param);
Thomas Daede's avatar
Thomas Daede committed
706 707
  } else {
#endif
hui su's avatar
hui su committed
708 709
    av1_fwd_txfm(src_int16, coeff, diff_stride, &fwd_txfm_param);
    av1_fwd_txfm(pred, ref_coeff, diff_stride, &fwd_txfm_param);
710
#if CONFIG_HIGHBITDEPTH
Thomas Daede's avatar
Thomas Daede committed
711 712
  }
#endif
713 714

  // PVQ for inter mode block
715
  if (!x->skip_block) {
ltrudeau's avatar
ltrudeau committed
716
    PVQ_SKIP_TYPE ac_dc_coded =
Thomas Daede's avatar
Thomas Daede committed
717
        av1_pvq_encode_helper(x,
ltrudeau's avatar
ltrudeau committed
718 719 720 721 722 723 724 725 726 727 728 729
                              coeff,        // target original vector
                              ref_coeff,    // reference vector
                              dqcoeff,      // de-quantized vector
                              eob,          // End of Block marker
                              pd->dequant,  // aom's quantizers
                              plane,        // image plane
                              tx_size,      // block size in log_2 - 2
                              tx_type,
                              &x->rate,  // rate measured
                              x->pvq_speed,
                              pvq_info);  // PVQ info for a block
    skip = ac_dc_coded == PVQ_SKIP;
730
  }
731 732 733 734
  x->pvq_skip[plane] = skip;

  if (!skip) mbmi->skip = 0;
#endif  // #if !CONFIG_PVQ
Jingning Han's avatar
Jingning Han committed
735 736
}

737
static void encode_block(int plane, int block, int blk_row, int blk_col,
738
                         BLOCK_SIZE plane_bsize, TX_SIZE tx_size, void *arg) {
Jingning Han's avatar
Jingning Han committed
739
  struct encode_b_args *const args = arg;
Angie Chiang's avatar
Angie Chiang committed
740
  AV1_COMMON *cm = args->cm;
Jingning Han's avatar
Jingning Han committed
741 742
  MACROBLOCK *const x = args->x;
  MACROBLOCKD *const xd = &x->e_mbd;
743
  int ctx;
Jingning Han's avatar
Jingning Han committed
744 745 746 747
  struct macroblock_plane *const p = &x->plane[plane];
  struct macroblockd_plane *const pd = &xd->plane[plane];
  tran_low_t *const dqcoeff = BLOCK_OFFSET(pd->dqcoeff, block);
  uint8_t *dst;
748
#if !CONFIG_PVQ
Jingning Han's avatar
Jingning Han committed
749
  ENTROPY_CONTEXT *a, *l;
750
#endif
751
#if CONFIG_VAR_TX
752
  int bw = block_size_wide[plane_bsize] >> tx_size_wide_log2[0];
753
#endif
754 755
  dst = &pd->dst
             .buf[(blk_row * pd->dst.stride + blk_col) << tx_size_wide_log2[0]];
756 757

#if !CONFIG_PVQ
758 759
  a = &args->ta[blk_col];
  l = &args->tl[blk_row];
760 761 762 763 764
#if CONFIG_VAR_TX
  ctx = get_entropy_context(tx_size, a, l);
#else
  ctx = combine_entropy_contexts(*a, *l);
#endif
765 766 767
#else
  ctx = 0;
#endif  // CONFIG_PVQ
Jingning Han's avatar
Jingning Han committed
768

Alex Converse's avatar
Alex Converse committed
769
#if CONFIG_VAR_TX
770
  // Assert not magic number (uninitialized).
771
  assert(x->blk_skip[plane][blk_row * bw + blk_col] != 234);
772

773
  if (x->blk_skip[plane][blk_row * bw + blk_col] == 0) {
774
#else
775
  {
776
#endif
Angie Chiang's avatar
Angie Chiang committed
777
    av1_xform_quant(cm, x, plane, block, blk_row, blk_col, plane_bsize, tx_size,
778
                    ctx, AV1_XFORM_QUANT_FP);
Jingning Han's avatar
Jingning Han committed
779
  }
780 781
#if CONFIG_VAR_TX
  else {
782
    p->eobs[block] = 0;
783 784
  }
#endif
785

786
#if !CONFIG_PVQ
787
  av1_optimize_b(cm, x, plane, block, plane_bsize, tx_size, a, l);
788

789 790
  av1_set_txb_context(x, plane, block, tx_size, a, l);

791
  if (p->eobs[block]) *(args->skip) = 0;
Jingning Han's avatar
Jingning Han committed
792

793
  if (p->eobs[block] == 0) return;
794 795 796 797 798 799
#else
  (void)ctx;
  if (!x->pvq_skip[plane]) *(args->skip) = 0;

  if (x->pvq_skip[plane]) return;
#endif
800
  TX_TYPE tx_type = get_tx_type(pd->plane_type, xd, block, tx_size);
801 802
  av1_inverse_transform_block(xd, dqcoeff, tx_type, tx_size, dst,
                              pd->dst.stride, p->eobs[block]);
Jingning Han's avatar
Jingning Han committed
803 804
}

805 806 807 808 809 810 811 812
#if CONFIG_VAR_TX
static void encode_block_inter(int plane, int block, int blk_row, int blk_col,
                               BLOCK_SIZE plane_bsize, TX_SIZE tx_size,
                               void *arg) {
  struct encode_b_args *const args = arg;
  MACROBLOCK *const x = args->x;
  MACROBLOCKD *const xd = &x->e_mbd;
  MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;
813
  const BLOCK_SIZE bsize = txsize_to_bsize[tx_size];
814
  const struct macroblockd_plane *const pd = &xd->plane[plane];
815 816
  const int tx_row = blk_row >> (1 - pd->subsampling_y);
  const int tx_col = blk_col >> (1 - pd->subsampling_x);
Debargha Mukherjee's avatar
Debargha Mukherjee committed
817
  TX_SIZE plane_tx_size;
818 819
  const int max_blocks_high = max_block_high(xd, plane_bsize, plane);
  const int max_blocks_wide = max_block_wide(xd, plane_bsize, plane);
820

821
  if (blk_row >= max_blocks_high || blk_col >= max_blocks_wide) return;
822

823 824 825
  plane_tx_size =
      plane ? uv_txsize_lookup[bsize][mbmi->inter_tx_size[tx_row][tx_col]][0][0]
            : mbmi->inter_tx_size[tx_row][tx_col];
Debargha Mukherjee's avatar
Debargha Mukherjee committed
826

827
  if (tx_size == plane_tx_size) {
828
    encode_block(plane, block, blk_row, blk_col, plane_bsize, tx_size, arg);
829
  } else {
830
    assert(tx_size < TX_SIZES_ALL);
831
    const TX_SIZE sub_txs = sub_tx_size_map[tx_size];
832
    assert(sub_txs < tx_size);
833 834
    // This is the square transform block partition entry point.
    int bsl = tx_size_wide_unit[sub_txs];
835 836
    int i;
    assert(bsl > 0);
837

838
    for (i = 0; i < 4; ++i) {
839 840 841
      const int offsetr = blk_row + ((i >> 1) * bsl);
      const int offsetc = blk_col + ((i & 0x01) * bsl);
      int step = tx_size_wide_unit[sub_txs] * tx_size_high_unit[sub_txs];
842

843
      if (offsetr >= max_blocks_high || offsetc >= max_blocks_wide) continue;
844

845 846 847
      encode_block_inter(plane, block, offsetr, offsetc, plane_bsize, sub_txs,
                         arg);
      block += step;
848 849 850 851 852
    }
  }
}
#endif

Angie Chiang's avatar
Angie Chiang committed
853 854 855 856 857
typedef struct encode_block_pass1_args {
  AV1_COMMON *cm;
  MACROBLOCK *x;
} encode_block_pass1_args;

858
static void encode_block_pass1(int plane, int block, int blk_row, int blk_col,
859 860
                               BLOCK_SIZE plane_bsize, TX_SIZE tx_size,
                               void *arg) {
Angie Chiang's avatar
Angie Chiang committed
861 862 863
  encode_block_pass1_args *args = (encode_block_pass1_args *)arg;
  AV1_COMMON *cm = args->cm;
  MACROBLOCK *const x = args->x;
Jingning Han's avatar
Jingning Han committed
864 865 866 867 868
  MACROBLOCKD *const xd = &x->e_mbd;
  struct macroblock_plane *const p = &x->plane[plane];
  struct macroblockd_plane *const pd = &xd->plane[plane];
  tran_low_t *const dqcoeff = BLOCK_OFFSET(pd->dqcoeff, block);
  uint8_t *dst;
869
  int ctx = 0;
870 871
  dst = &pd->dst
             .buf[(blk_row * pd->dst.stride + blk_col) << tx_size_wide_log2[0]];
Jingning Han's avatar
Jingning Han committed
872

Angie Chiang's avatar
Angie Chiang committed
873
  av1_xform_quant(cm, x, plane, block, blk_row, blk_col, plane_bsize, tx_size,
874
                  ctx, AV1_XFORM_QUANT_B);
875
#if !CONFIG_PVQ
Jingning Han's avatar
Jingning Han committed
876
  if (p->eobs[block] > 0) {
877 878 879 880 881 882 883 884
#else
  if (!x->pvq_skip[plane]) {
    {
      int tx_blk_size;
      int i, j;
      // transform block size in pixels
      tx_blk_size = tx_size_wide[tx_size];

Thomas Daede's avatar
Thomas Daede committed
885 886 887 888
// Since av1 does not have separate function which does inverse transform
// but av1_inv_txfm_add_*x*() also does addition of predicted image to
// inverse transformed image,
// pass blank dummy image to av1_inv_txfm_add_*x*(), i.e. set dst as zeros
889
#if CONFIG_HIGHBITDEPTH
Thomas Daede's avatar
Thomas Daede committed
890 891 892 893 894
      if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
        for (j = 0; j < tx_blk_size; j++)
          for (i = 0; i < tx_blk_size; i++)
            CONVERT_TO_SHORTPTR(dst)[j * pd->dst.stride + i] = 0;
      } else {
895
#endif  // CONFIG_HIGHBITDEPTH
Thomas Daede's avatar
Thomas Daede committed
896 897
        for (j = 0; j < tx_blk_size; j++)
          for (i = 0; i < tx_blk_size; i++) dst[j * pd->dst.stride + i] = 0;
898
#if CONFIG_HIGHBITDEPTH
Thomas Daede's avatar
Thomas Daede committed
899
      }
900
#endif  // CONFIG_HIGHBITDEPTH
901
    }
Jingning Han's avatar
Jingning Han committed
902
#endif  // !CONFIG_PVQ
903
#if CONFIG_HIGHBITDEPTH
Jingning Han's avatar
Jingning Han committed
904
    if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
Yaowu Xu's avatar
Yaowu Xu committed
905
      if (xd->lossless[xd->mi[0]->mbmi.segment_id]) {
Yaowu Xu's avatar
Yaowu Xu committed
906 907
        av1_highbd_iwht4x4_add(dqcoeff, dst, pd->dst.stride, p->eobs[block],
                               xd->bd);
908
      } else {
Yaowu Xu's avatar
Yaowu Xu committed
909 910
        av1_highbd_idct4x4_add(dqcoeff, dst, pd->dst.stride, p->eobs[block],
                               xd->bd);
911 912
      }
      return;
Jingning Han's avatar
Jingning Han committed
913
    }
914
#endif  //  CONFIG_HIGHBITDEPTH
Yaowu Xu's avatar
Yaowu Xu committed
915
    if (xd->lossless[xd->mi[0]->mbmi.segment_id]) {
Yaowu Xu's avatar
Yaowu Xu committed
916
      av1_iwht4x4_add(dqcoeff, dst, pd->dst.stride, p->eobs[block]);
917
    } else {
Yaowu Xu's avatar
Yaowu Xu committed
918
      av1_idct4x4_add(dqcoeff, dst, pd->dst.stride, p->eobs[block]);
919
    }