encodemb.c 61.6 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
static INLINE unsigned int get_token_bit_costs(
    unsigned int token_costs[2][COEFF_CONTEXTS][ENTROPY_TOKENS], int skip_eob,
    int ctx, int token) {
125 126 127 128
  (void)skip_eob;
  return token_costs[token == ZERO_TOKEN || token == EOB_TOKEN][ctx][token];
}

129
#if !CONFIG_LV_MAP
130

131 132
static int optimize_b_greedy(const AV1_COMMON *cm, MACROBLOCK *mb, int plane,
                             int block, TX_SIZE tx_size, int ctx) {
133 134 135 136 137 138 139 140 141 142 143 144 145 146
  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);
  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 =
Angie Chiang's avatar
Angie Chiang committed
147
      get_scan(cm, tx_size, tx_type, &xd->mi[0]->mbmi);
148 149 150
  const int16_t *const scan = scan_order->scan;
  const int16_t *const nb = scan_order->neighbors;
  int dqv;
151
  const int shift = av1_get_tx_scale(tx_size);
152 153
#if CONFIG_AOM_QM
  int seg_id = xd->mi[0]->mbmi.segment_id;
154 155 156 157 158
  // 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];
159 160 161 162 163 164 165 166 167 168 169 170 171 172
#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];

173
  assert(mb->qindex > 0);
174 175 176 177 178 179 180 181 182

  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];
183
    token_cache[rc] = av1_pt_energy_class[av1_get_token(qcoeff[rc])];
184 185 186 187 188 189 190 191
  }

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

  final_eob = 0;

  int64_t eob_cost0, eob_cost1;
192 193
  tran_low_t before_best_eob_qc = 0;
  tran_low_t before_best_eob_dqc = 0;
194 195 196 197

  const int ctx0 = ctx;
  /* Record the r-d cost */
  int64_t accu_rate = 0;
198 199 200
  // 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;
201 202 203

  rate0 = get_token_bit_costs(*(token_costs_ptr + band_translate[0]), 0, ctx0,
                              EOB_TOKEN);
204
  int64_t best_block_rd_cost = RDCOST(rdmult, rate0, accu_error);
205 206 207 208 209 210 211 212 213 214 215 216 217 218

  // 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
219 220 221
      int token = av1_get_token(x);
      rate0 = get_token_bit_costs(*(token_costs_ptr + band_cur),
                                  token_tree_sel_cur, ctx_cur, token);
222 223 224 225 226 227 228 229
      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.
230
      int dx0 = abs(coeff[rc]) * (1 << shift);
231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253
#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) {
254 255 256
        int dx_sign = dx < 0 ? 1 : 0;
        dx = abs(dx) >> (xd->bd - 8);
        if (dx_sign) dx = -dx;
257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311
      }
#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];
312 313
        int token_next =
            i + 1 != eob ? av1_get_token(qcoeff[scan[i + 1]]) : EOB_TOKEN;
314 315 316 317 318

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

319 320 321
        next_bits0 =
            get_token_bit_costs(*(token_costs_ptr + band_next),
                                token_tree_sel_next, ctx_next, token_next);
322 323 324 325 326 327 328 329
        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);

330 331 332
        next_bits1 =
            get_token_bit_costs(*(token_costs_ptr + band_next),
                                token_tree_sel_next, ctx_next, token_next);
333 334 335 336 337 338 339 340

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

341 342
      rd_cost0 = RDCOST(rdmult, (rate0 + next_bits0), d2);
      rd_cost1 = RDCOST(rdmult, (rate1 + next_bits1), d2_a);
343 344 345

      best_x = (rd_cost1 < rd_cost0);

346
      eob_cost0 = RDCOST(rdmult, (accu_rate + rate0 + next_eob_bits0),
347 348 349
                         (accu_error + d2 - d0));
      eob_cost1 = eob_cost0;
      if (x_a != 0) {
350
        eob_cost1 = RDCOST(rdmult, (accu_rate + rate1 + next_eob_bits1),
351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367
                           (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
368 369
          if (x_a < 0)
            dqc_a = -((-x_a * dqv) >> shift);
370
          else
371
            dqc_a = (x_a * dqv) >> shift;
372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394
#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];
      }
395
      assert(accu_error >= 0);
396 397 398 399

      x_prev = qcoeff[rc];

      // determine whether to move the eob position to i+1
400 401
      int use_a = (x_a != 0) && (best_eob_x);
      int64_t best_eob_cost_i = use_a ? eob_cost1 : eob_cost0;
402 403 404 405

      if (best_eob_cost_i < best_block_rd_cost) {
        best_block_rd_cost = best_eob_cost_i;
        final_eob = i + 1;
406 407 408 409 410 411 412
        if (use_a) {
          before_best_eob_qc = x_a;
          before_best_eob_dqc = dqc_a;
        } else {
          before_best_eob_qc = x;
          before_best_eob_dqc = dqc;
        }
413 414 415 416 417 418
      }
    }  // if (x==0)
  }    // for (i)

  assert(final_eob <= eob);
  if (final_eob > 0) {
419
    assert(before_best_eob_qc != 0);
420 421
    i = final_eob - 1;
    int rc = scan[i];
422 423
    qcoeff[rc] = before_best_eob_qc;
    dqcoeff[rc] = before_best_eob_dqc;
424 425 426 427 428 429 430 431 432 433 434
  }

  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;
}
435
#endif  // !CONFIG_LV_MAP
436 437

int av1_optimize_b(const AV1_COMMON *cm, MACROBLOCK *mb, int plane, int block,
438 439
                   BLOCK_SIZE plane_bsize, TX_SIZE tx_size,
                   const ENTROPY_CONTEXT *a, const ENTROPY_CONTEXT *l) {
440 441 442 443 444 445
  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
446

447
#if CONFIG_PVQ
448 449
  (void)cm;
  (void)tx_size;
450 451
  (void)a;
  (void)l;
452 453
  return eob;
#endif
Jingning Han's avatar
Jingning Han committed
454

455
#if !CONFIG_LV_MAP
456
  (void)plane_bsize;
457 458 459 460
#if CONFIG_VAR_TX
  int ctx = get_entropy_context(tx_size, a, l);
#else
  int ctx = combine_entropy_contexts(*a, *l);
461
#endif  // CONFIG_VAR_TX
462
  return optimize_b_greedy(cm, mb, plane, block, tx_size, ctx);
463 464 465 466 467
#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
468
}
469

Thomas Daede's avatar
Thomas Daede committed
470
#if !CONFIG_PVQ
Angie Chiang's avatar
Angie Chiang committed
471 472
typedef enum QUANT_FUNC {
  QUANT_FUNC_LOWBD = 0,
473
  QUANT_FUNC_HIGHBD = 1,
474
  QUANT_FUNC_TYPES = 2
Angie Chiang's avatar
Angie Chiang committed
475 476
} QUANT_FUNC;

477 478
static AV1_QUANT_FACADE
    quant_func_list[AV1_XFORM_QUANT_TYPES][QUANT_FUNC_TYPES] = {
479
#if !CONFIG_NEW_QUANT
480
      { av1_quantize_fp_facade, av1_highbd_quantize_fp_facade },
Yaowu Xu's avatar
Yaowu Xu committed
481 482
      { av1_quantize_b_facade, av1_highbd_quantize_b_facade },
      { av1_quantize_dc_facade, av1_highbd_quantize_dc_facade },
483
#else   // !CONFIG_NEW_QUANT
484 485 486
      { 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 },
487
#endif  // !CONFIG_NEW_QUANT
488 489
      { NULL, NULL }
    };
490
#endif  // !CONFIG_PVQ
491

492 493 494 495
typedef void (*fwdTxfmFunc)(const int16_t *diff, tran_low_t *coeff, int stride,
                            FWD_TXFM_PARAM *param);
static const fwdTxfmFunc fwd_txfm_func[2] = { av1_fwd_txfm,
                                              av1_highbd_fwd_txfm };
496

Angie Chiang's avatar
Angie Chiang committed
497 498
void av1_xform_quant(const AV1_COMMON *cm, MACROBLOCK *x, int plane, int block,
                     int blk_row, int blk_col, BLOCK_SIZE plane_bsize,
499 500
                     TX_SIZE tx_size, int ctx,
                     AV1_XFORM_QUANT xform_quant_idx) {
Jingning Han's avatar
Jingning Han committed
501
  MACROBLOCKD *const xd = &x->e_mbd;
502
  MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;
Yushin Cho's avatar
Yushin Cho committed
503
#if !(CONFIG_PVQ || CONFIG_DAALA_DIST)
Jingning Han's avatar
Jingning Han committed
504 505
  const struct macroblock_plane *const p = &x->plane[plane];
  const struct macroblockd_plane *const pd = &xd->plane[plane];
506 507 508 509
#else
  struct macroblock_plane *const p = &x->plane[plane];
  struct macroblockd_plane *const pd = &xd->plane[plane];
#endif
510
  PLANE_TYPE plane_type = get_plane_type(plane);
511
  TX_TYPE tx_type = get_tx_type(plane_type, xd, block, tx_size);
Angie Chiang's avatar
Angie Chiang committed
512 513

#if CONFIG_AOM_QM || CONFIG_NEW_QUANT
514
  const int is_inter = is_inter_block(mbmi);
Angie Chiang's avatar
Angie Chiang committed
515 516 517
#endif

  const SCAN_ORDER *const scan_order = get_scan(cm, tx_size, tx_type, mbmi);
Jingning Han's avatar
Jingning Han committed
518 519 520 521
  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];
522
  const int diff_stride = block_size_wide[plane_bsize];
523
#if CONFIG_AOM_QM
524
  int seg_id = mbmi->segment_id;
525 526 527 528 529 530 531 532
  // 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];
533
#endif
Angie Chiang's avatar
Angie Chiang committed
534 535

  FWD_TXFM_PARAM fwd_txfm_param;
536

Yushin Cho's avatar
Yushin Cho committed
537 538 539 540
#if CONFIG_PVQ || CONFIG_DAALA_DIST
  uint8_t *dst;
  int16_t *pred;
  const int dst_stride = pd->dst.stride;
541 542
  const int txw = tx_size_wide[tx_size];
  const int txh = tx_size_high[tx_size];
Yushin Cho's avatar
Yushin Cho committed
543 544 545
  int i, j;
#endif

546 547
#if !CONFIG_PVQ
  const int tx2d_size = tx_size_2d[tx_size];
548
  QUANT_PARAM qparam;
549 550
  const int16_t *src_diff;

551 552
  src_diff =
      &p->src_diff[(blk_row * diff_stride + blk_col) << tx_size_wide_log2[0]];
553
  qparam.log_scale = av1_get_tx_scale(tx_size);
554 555 556 557 558 559 560 561
#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
562 563 564 565
#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
566 567 568
  uint8_t *src;
  int16_t *src_int16;
  const int src_stride = p->src.stride;
569

570
  (void)ctx;
571 572 573 574 575 576 577
  (void)scan_order;
  (void)qcoeff;

  if (x->pvq_coded) {
    assert(block < MAX_PVQ_BLOCKS_IN_SB);
    pvq_info = &x->pvq[block][plane];
  }
578 579 580
  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
581

582
#if CONFIG_HIGHBITDEPTH
Thomas Daede's avatar
Thomas Daede committed
583
  if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
584 585
    for (j = 0; j < txh; j++)
      for (i = 0; i < txw; i++)
Thomas Daede's avatar
Thomas Daede committed
586 587 588
        src_int16[diff_stride * j + i] =
            CONVERT_TO_SHORTPTR(src)[src_stride * j + i];
  } else {
589
#endif  // CONFIG_HIGHBITDEPTH
590 591
    for (j = 0; j < txh; j++)
      for (i = 0; i < txw; i++)
Thomas Daede's avatar
Thomas Daede committed
592
        src_int16[diff_stride * j + i] = src[src_stride * j + i];
593
#if CONFIG_HIGHBITDEPTH
Thomas Daede's avatar
Thomas Daede committed
594
  }
595
#endif  // CONFIG_HIGHBITDEPTH
Yushin Cho's avatar
Yushin Cho committed
596 597 598 599
#endif

#if CONFIG_PVQ || CONFIG_DAALA_DIST
  dst = &pd->dst.buf[(blk_row * dst_stride + blk_col) << tx_size_wide_log2[0]];
600
  pred = &pd->pred[(blk_row * diff_stride + blk_col) << tx_size_wide_log2[0]];
601

Thomas Daede's avatar
Thomas Daede committed
602 603
// copy uint8 orig and predicted block to int16 buffer
// in order to use existing VP10 transform functions
604
#if CONFIG_HIGHBITDEPTH
Thomas Daede's avatar
Thomas Daede committed
605
  if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
606 607
    for (j = 0; j < txh; j++)
      for (i = 0; i < txw; i++)
Thomas Daede's avatar
Thomas Daede committed
608 609 610
        pred[diff_stride * j + i] =
            CONVERT_TO_SHORTPTR(dst)[dst_stride * j + i];
  } else {
611
#endif  // CONFIG_HIGHBITDEPTH
612 613
    for (j = 0; j < txh; j++)
      for (i = 0; i < txw; i++)
Thomas Daede's avatar
Thomas Daede committed
614
        pred[diff_stride * j + i] = dst[dst_stride * j + i];
615
#if CONFIG_HIGHBITDEPTH
Thomas Daede's avatar
Thomas Daede committed
616
  }
617
#endif  // CONFIG_HIGHBITDEPTH
618
#endif
Yushin Cho's avatar
Yushin Cho committed
619

620
  (void)ctx;
621 622

  fwd_txfm_param.tx_type = tx_type;
Angie Chiang's avatar
Angie Chiang committed
623
  fwd_txfm_param.tx_size = tx_size;
624
  fwd_txfm_param.lossless = xd->lossless[mbmi->segment_id];
Angie Chiang's avatar
Angie Chiang committed
625

Thomas Daede's avatar
Thomas Daede committed
626
#if !CONFIG_PVQ
627
  fwd_txfm_param.bd = xd->bd;
628 629 630
  const int is_hbd = get_bitdepth_data_path_index(xd);
  fwd_txfm_func[is_hbd](src_diff, coeff, diff_stride, &fwd_txfm_param);

Yaowu Xu's avatar
Yaowu Xu committed
631
  if (xform_quant_idx != AV1_XFORM_QUANT_SKIP_QUANT) {
632
    if (LIKELY(!x->skip_block)) {
633
      quant_func_list[xform_quant_idx][is_hbd](
634
          coeff, tx2d_size, p, qcoeff, pd, dqcoeff, eob, scan_order, &qparam);
635
    } else {
Yaowu Xu's avatar
Yaowu Xu committed
636
      av1_quantize_skip(tx2d_size, qcoeff, dqcoeff, eob);
Jingning Han's avatar
Jingning Han committed
637 638
    }
  }
639 640 641 642
#if CONFIG_LV_MAP
  p->txb_entropy_ctx[block] =
      (uint8_t)av1_get_txb_entropy_context(qcoeff, scan_order, *eob);
#endif  // CONFIG_LV_MAP
643 644
  return;
#else  // CONFIG_PVQ
Angie Chiang's avatar
Angie Chiang committed
645
  (void)xform_quant_idx;
646
#if CONFIG_HIGHBITDEPTH
Thomas Daede's avatar
Thomas Daede committed
647 648
  fwd_txfm_param.bd = xd->bd;
  if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
hui su's avatar
hui su committed
649 650
    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
651 652
  } else {
#endif
hui su's avatar
hui su committed
653 654
    av1_fwd_txfm(src_int16, coeff, diff_stride, &fwd_txfm_param);
    av1_fwd_txfm(pred, ref_coeff, diff_stride, &fwd_txfm_param);
655
#if CONFIG_HIGHBITDEPTH
Thomas Daede's avatar
Thomas Daede committed
656 657
  }
#endif
658 659

  // PVQ for inter mode block
660
  if (!x->skip_block) {
ltrudeau's avatar
ltrudeau committed
661
    PVQ_SKIP_TYPE ac_dc_coded =
Thomas Daede's avatar
Thomas Daede committed
662
        av1_pvq_encode_helper(x,
ltrudeau's avatar
ltrudeau committed
663 664 665 666 667 668 669 670 671 672 673 674
                              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;
675
  }
676 677 678 679
  x->pvq_skip[plane] = skip;

  if (!skip) mbmi->skip = 0;
#endif  // #if !CONFIG_PVQ
Jingning Han's avatar
Jingning Han committed
680 681
}

682
static void encode_block(int plane, int block, int blk_row, int blk_col,
683
                         BLOCK_SIZE plane_bsize, TX_SIZE tx_size, void *arg) {
Jingning Han's avatar
Jingning Han committed
684
  struct encode_b_args *const args = arg;
Angie Chiang's avatar
Angie Chiang committed
685
  AV1_COMMON *cm = args->cm;
Jingning Han's avatar
Jingning Han committed
686 687
  MACROBLOCK *const x = args->x;
  MACROBLOCKD *const xd = &x->e_mbd;
688
  int ctx;
Jingning Han's avatar
Jingning Han committed
689 690 691 692
  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;
693
#if !CONFIG_PVQ
Jingning Han's avatar
Jingning Han committed
694
  ENTROPY_CONTEXT *a, *l;
695
#endif
696
#if CONFIG_VAR_TX
697
  int bw = block_size_wide[plane_bsize] >> tx_size_wide_log2[0];
698
#endif
699 700
  dst = &pd->dst
             .buf[(blk_row * pd->dst.stride + blk_col) << tx_size_wide_log2[0]];
701 702

#if !CONFIG_PVQ
703 704
  a = &args->ta[blk_col];
  l = &args->tl[blk_row];
705 706 707 708 709
#if CONFIG_VAR_TX
  ctx = get_entropy_context(tx_size, a, l);
#else
  ctx = combine_entropy_contexts(*a, *l);
#endif
710 711 712
#else
  ctx = 0;
#endif  // CONFIG_PVQ
Jingning Han's avatar
Jingning Han committed
713

Alex Converse's avatar
Alex Converse committed
714
#if CONFIG_VAR_TX
715
  // Assert not magic number (uninitialized).
716
  assert(x->blk_skip[plane][blk_row * bw + blk_col] != 234);
717

718
  if (x->blk_skip[plane][blk_row * bw + blk_col] == 0) {
719
#else
720
  {
721
#endif
Angie Chiang's avatar
Angie Chiang committed
722
    av1_xform_quant(cm, x, plane, block, blk_row, blk_col, plane_bsize, tx_size,
723
                    ctx, AV1_XFORM_QUANT_FP);
Jingning Han's avatar
Jingning Han committed
724
  }
725 726
#if CONFIG_VAR_TX
  else {
727
    p->eobs[block] = 0;
728 729
  }
#endif
730

731
#if !CONFIG_PVQ
732
  av1_optimize_b(cm, x, plane, block, plane_bsize, tx_size, a, l);
733

734 735
  av1_set_txb_context(x, plane, block, tx_size, a, l);

736
  if (p->eobs[block]) *(args->skip) = 0;
Jingning Han's avatar
Jingning Han committed
737

738
  if (p->eobs[block] == 0) return;
739 740 741 742 743 744
#else
  (void)ctx;
  if (!x->pvq_skip[plane]) *(args->skip) = 0;

  if (x->pvq_skip[plane]) return;
#endif
745
  TX_TYPE tx_type = get_tx_type(pd->plane_type, xd, block, tx_size);
746 747
  av1_inverse_transform_block(xd, dqcoeff, tx_type, tx_size, dst,
                              pd->dst.stride, p->eobs[block]);
Jingning Han's avatar
Jingning Han committed
748 749
}

750 751 752 753 754 755 756 757
#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;
758
  const BLOCK_SIZE bsize = txsize_to_bsize[tx_size];
759
  const struct macroblockd_plane *const pd = &xd->plane[plane];
760 761
  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
762
  TX_SIZE plane_tx_size;
763 764
  const int max_blocks_high = max_block_high(xd, plane_bsize, plane);
  const int max_blocks_wide = max_block_wide(xd, plane_bsize, plane);
765

766
  if (blk_row >= max_blocks_high || blk_col >= max_blocks_wide) return;
767

768 769 770
  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
771

772
  if (tx_size == plane_tx_size) {
773
    encode_block(plane, block, blk_row, blk_col, plane_bsize, tx_size, arg);
774
  } else {
775
    assert(tx_size < TX_SIZES_ALL);
776
    const TX_SIZE sub_txs = sub_tx_size_map[tx_size];
777
    assert(sub_txs < tx_size);
778 779
    // This is the square transform block partition entry point.
    int bsl = tx_size_wide_unit[sub_txs];
780 781
    int i;
    assert(bsl > 0);
782

783
    for (i = 0; i < 4; ++i) {
784 785 786
      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];
787

788
      if (offsetr >= max_blocks_high || offsetc >= max_blocks_wide) continue;
789

790 791 792
      encode_block_inter(plane, block, offsetr, offsetc, plane_bsize, sub_txs,
                         arg);
      block += step;
793 794 795 796 797
    }
  }
}
#endif

Angie Chiang's avatar
Angie Chiang committed
798 799 800 801 802
typedef struct encode_block_pass1_args {
  AV1_COMMON *cm;
  MACROBLOCK *x;
} encode_block_pass1_args;

803
static void encode_block_pass1(int plane, int block, int blk_row, int blk_col,
804 805
                               BLOCK_SIZE plane_bsize, TX_SIZE tx_size,
                               void *arg) {
Angie Chiang's avatar
Angie Chiang committed
806 807 808
  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
809 810 811 812 813
  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;
814
  int ctx = 0;
815 816
  dst = &pd->dst
             .buf[(blk_row * pd->dst.stride + blk_col) << tx_size_wide_log2[0]];
Jingning Han's avatar
Jingning Han committed
817

Angie Chiang's avatar
Angie Chiang committed
818
  av1_xform_quant(cm, x, plane, block, blk_row, blk_col, plane_bsize, tx_size,
819
                  ctx, AV1_XFORM_QUANT_B);
820
#if !CONFIG_PVQ
Jingning Han's avatar
Jingning Han committed
821
  if (p->eobs[block] > 0) {
822 823 824 825 826 827 828 829
#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
830 831 832 833
// 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
834
#if CONFIG_HIGHBITDEPTH
Thomas Daede's avatar
Thomas Daede committed
835 836 837 838 839
      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 {
840
#endif  // CONFIG_HIGHBITDEPTH
Thomas Daede's avatar
Thomas Daede committed
841 842
        for (j = 0; j < tx_blk_size; j++)
          for (i = 0; i < tx_blk_size; i++) dst[j * pd->dst.stride + i] = 0;
843
#if CONFIG_HIGHBITDEPTH
Thomas Daede's avatar
Thomas Daede committed
844
      }
845
#endif  // CONFIG_HIGHBITDEPTH
846
    }
Jingning Han's avatar
Jingning Han committed
847
#endif  // !CONFIG_PVQ
848
#if CONFIG_HIGHBITDEPTH
Jingning Han's avatar
Jingning Han committed
849
    if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
850 851 852
      av1_highbd_inv_txfm_add_4x4(dqcoeff, dst, pd->dst.stride, p->eobs[block],
                                  xd->bd, DCT_DCT,
                                  xd->lossless[xd->mi[0]->mbmi.segment_id]);
853
      return;
Jingning Han's avatar
Jingning Han committed
854
    }
855
#endif  //  CONFIG_HIGHBITDEPTH
Yaowu Xu's avatar
Yaowu Xu committed
856
    if (xd->lossless[xd->mi[0]->mbmi.segment_id]) {
Yaowu Xu's avatar
Yaowu Xu committed
857
      av1_iwht4x4_add(dqcoeff, dst, pd->dst.stride, p->eobs[block]);
858
    } else {
Yaowu Xu's avatar
Yaowu Xu committed
859
      av1_idct4x4_add(dqcoeff, dst, pd->dst.stride, p->eobs[block]);
860
    }
Jingning Han's avatar
Jingning Han committed
861 862 863
  }
}

Angie Chiang's avatar
Angie Chiang committed
864 865
void av1_encode_sby_pass1(AV1_COMMON *cm, MACROBLOCK *x, BLOCK_SIZE bsize) {
  encode_block_pass1_args args = { cm, x };
Yaowu Xu's avatar
Yaowu Xu committed
866 867
  av1_subtract_plane(x, bsize, 0);
  av1_foreach_transformed_block_in_plane(&x->e_mbd, bsize, 0,
Angie Chiang's avatar
Angie Chiang committed
868
                                         encode_block_pass1, &args);
Jingning Han's avatar
Jingning Han committed
869 870
}

Yaowu Xu's avatar
Yaowu Xu committed
871 872
void av1_encode_sb(AV1_COMMON *cm, MACROBLOCK *x, BLOCK_SIZE bsize, int mi_row,
                   int mi_col) {
Jingning Han's avatar
Jingning Han committed
873 874 875
  MACROBLOCKD *const xd = &x->e_mbd;
  struct optimize_ctx ctx;
  MB_MODE_INFO *mbmi = &xd->mi[0]->mbmi;
Angie Chiang's avatar
Angie Chiang committed
876
  struct encode_b_args arg = { cm, x, &ctx, &mbmi->skip, NULL, NULL, 1 };
Jingning Han's avatar
Jingning Han committed
877 878 879 880
  int plane;

  mbmi->skip = 1;

881
  if (x->skip) return;
Jingning Han's avatar
Jingning Han committed
882 883

  for (plane = 0; plane < MAX_MB_PLANE; ++plane) {
Jingning Han's avatar
Jingning Han committed
884
#if CONFIG_CB4X4 && !CONFIG_CHROMA_2X2
885 886 887 888 889
    const int subsampling_x = xd->plane[plane].subsampling_x;
    const int subsampling_y = xd->plane[plane].subsampling_y;

    if (!is_chroma_reference(mi_row, mi_col, bsize, subsampling_x,
                             subsampling_y))
890
      continue;
891 892

    bsize = scale_chroma_bsize(bsize, subsampling_x, subsampling_y);
893 894 895 896 897
#else
    (void)mi_row;
    (void)mi_col;
#endif

898 899 900 901
#if CONFIG_VAR_TX
    // TODO(jingning): Clean this up.
    const struct macroblockd_plane *const pd = &xd->plane[plane];
    const BLOCK_SIZE plane_bsize = get_plane_block_size(bsize, pd);
902 903
    const int mi_width = block_size_wide[plane_bsize] >> tx_size_wide_log2[0];
    const int mi_height = block_size_high[plane_bsize] >> tx_size_wide_log2[0];
904
    const TX_SIZE max_tx_size = get_vartx_max_txsize(mbmi, plane_bsize);