rdopt.c 157 KB
Newer Older
Jingning Han's avatar
Jingning Han committed
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
/*
 *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
 *
 *  Use of this source code is governed by a BSD-style license
 *  that can be found in the LICENSE file in the root of the source
 *  tree. An additional intellectual property rights grant can be found
 *  in the file PATENTS.  All contributing project authors may
 *  be found in the AUTHORS file in the root of the source tree.
 */

#include <assert.h>
#include <math.h>

#include "./vp10_rtcd.h"
#include "./vpx_dsp_rtcd.h"

#include "vpx_mem/vpx_mem.h"
#include "vpx_ports/mem.h"
19
#include "vpx_ports/system_state.h"
Jingning Han's avatar
Jingning Han committed
20

21 22 23 24 25 26 27 28 29 30 31
#include "vp10/common/common.h"
#include "vp10/common/entropy.h"
#include "vp10/common/entropymode.h"
#include "vp10/common/idct.h"
#include "vp10/common/mvref_common.h"
#include "vp10/common/pred_common.h"
#include "vp10/common/quant_common.h"
#include "vp10/common/reconinter.h"
#include "vp10/common/reconintra.h"
#include "vp10/common/scan.h"
#include "vp10/common/seg_common.h"
Jingning Han's avatar
Jingning Han committed
32

33 34 35 36 37 38 39 40 41 42
#include "vp10/encoder/cost.h"
#include "vp10/encoder/encodemb.h"
#include "vp10/encoder/encodemv.h"
#include "vp10/encoder/encoder.h"
#include "vp10/encoder/mcomp.h"
#include "vp10/encoder/quantize.h"
#include "vp10/encoder/ratectrl.h"
#include "vp10/encoder/rd.h"
#include "vp10/encoder/rdopt.h"
#include "vp10/encoder/aq_variance.h"
Jingning Han's avatar
Jingning Han committed
43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156

#define LAST_FRAME_MODE_MASK    ((1 << GOLDEN_FRAME) | (1 << ALTREF_FRAME) | \
                                 (1 << INTRA_FRAME))
#define GOLDEN_FRAME_MODE_MASK  ((1 << LAST_FRAME) | (1 << ALTREF_FRAME) | \
                                 (1 << INTRA_FRAME))
#define ALT_REF_MODE_MASK       ((1 << LAST_FRAME) | (1 << GOLDEN_FRAME) | \
                                 (1 << INTRA_FRAME))

#define SECOND_REF_FRAME_MASK   ((1 << ALTREF_FRAME) | 0x01)

#define MIN_EARLY_TERM_INDEX    3
#define NEW_MV_DISCOUNT_FACTOR  8

typedef struct {
  PREDICTION_MODE mode;
  MV_REFERENCE_FRAME ref_frame[2];
} MODE_DEFINITION;

typedef struct {
  MV_REFERENCE_FRAME ref_frame[2];
} REF_DEFINITION;

struct rdcost_block_args {
  MACROBLOCK *x;
  ENTROPY_CONTEXT t_above[16];
  ENTROPY_CONTEXT t_left[16];
  int this_rate;
  int64_t this_dist;
  int64_t this_sse;
  int64_t this_rd;
  int64_t best_rd;
  int exit_early;
  int use_fast_coef_costing;
  const scan_order *so;
  uint8_t skippable;
};

#define LAST_NEW_MV_INDEX 6
static const MODE_DEFINITION vp10_mode_order[MAX_MODES] = {
  {NEARESTMV, {LAST_FRAME,   NONE}},
  {NEARESTMV, {ALTREF_FRAME, NONE}},
  {NEARESTMV, {GOLDEN_FRAME, NONE}},

  {DC_PRED,   {INTRA_FRAME,  NONE}},

  {NEWMV,     {LAST_FRAME,   NONE}},
  {NEWMV,     {ALTREF_FRAME, NONE}},
  {NEWMV,     {GOLDEN_FRAME, NONE}},

  {NEARMV,    {LAST_FRAME,   NONE}},
  {NEARMV,    {ALTREF_FRAME, NONE}},
  {NEARMV,    {GOLDEN_FRAME, NONE}},

  {ZEROMV,    {LAST_FRAME,   NONE}},
  {ZEROMV,    {GOLDEN_FRAME, NONE}},
  {ZEROMV,    {ALTREF_FRAME, NONE}},

  {NEARESTMV, {LAST_FRAME,   ALTREF_FRAME}},
  {NEARESTMV, {GOLDEN_FRAME, ALTREF_FRAME}},

  {TM_PRED,   {INTRA_FRAME,  NONE}},

  {NEARMV,    {LAST_FRAME,   ALTREF_FRAME}},
  {NEWMV,     {LAST_FRAME,   ALTREF_FRAME}},
  {NEARMV,    {GOLDEN_FRAME, ALTREF_FRAME}},
  {NEWMV,     {GOLDEN_FRAME, ALTREF_FRAME}},

  {ZEROMV,    {LAST_FRAME,   ALTREF_FRAME}},
  {ZEROMV,    {GOLDEN_FRAME, ALTREF_FRAME}},

  {H_PRED,    {INTRA_FRAME,  NONE}},
  {V_PRED,    {INTRA_FRAME,  NONE}},
  {D135_PRED, {INTRA_FRAME,  NONE}},
  {D207_PRED, {INTRA_FRAME,  NONE}},
  {D153_PRED, {INTRA_FRAME,  NONE}},
  {D63_PRED,  {INTRA_FRAME,  NONE}},
  {D117_PRED, {INTRA_FRAME,  NONE}},
  {D45_PRED,  {INTRA_FRAME,  NONE}},
};

static const REF_DEFINITION vp10_ref_order[MAX_REFS] = {
  {{LAST_FRAME,   NONE}},
  {{GOLDEN_FRAME, NONE}},
  {{ALTREF_FRAME, NONE}},
  {{LAST_FRAME,   ALTREF_FRAME}},
  {{GOLDEN_FRAME, ALTREF_FRAME}},
  {{INTRA_FRAME,  NONE}},
};

static void swap_block_ptr(MACROBLOCK *x, PICK_MODE_CONTEXT *ctx,
                           int m, int n, int min_plane, int max_plane) {
  int i;

  for (i = min_plane; i < max_plane; ++i) {
    struct macroblock_plane *const p = &x->plane[i];
    struct macroblockd_plane *const pd = &x->e_mbd.plane[i];

    p->coeff    = ctx->coeff_pbuf[i][m];
    p->qcoeff   = ctx->qcoeff_pbuf[i][m];
    pd->dqcoeff = ctx->dqcoeff_pbuf[i][m];
    p->eobs     = ctx->eobs_pbuf[i][m];

    ctx->coeff_pbuf[i][m]   = ctx->coeff_pbuf[i][n];
    ctx->qcoeff_pbuf[i][m]  = ctx->qcoeff_pbuf[i][n];
    ctx->dqcoeff_pbuf[i][m] = ctx->dqcoeff_pbuf[i][n];
    ctx->eobs_pbuf[i][m]    = ctx->eobs_pbuf[i][n];

    ctx->coeff_pbuf[i][n]   = p->coeff;
    ctx->qcoeff_pbuf[i][n]  = p->qcoeff;
    ctx->dqcoeff_pbuf[i][n] = pd->dqcoeff;
    ctx->eobs_pbuf[i][n]    = p->eobs;
  }
}

Yaowu Xu's avatar
Yaowu Xu committed
157
static void model_rd_for_sb(VP10_COMP *cpi, BLOCK_SIZE bsize,
Jingning Han's avatar
Jingning Han committed
158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559
                            MACROBLOCK *x, MACROBLOCKD *xd,
                            int *out_rate_sum, int64_t *out_dist_sum,
                            int *skip_txfm_sb, int64_t *skip_sse_sb) {
  // Note our transform coeffs are 8 times an orthogonal transform.
  // Hence quantizer step is also 8 times. To get effective quantizer
  // we need to divide by 8 before sending to modeling function.
  int i;
  int64_t rate_sum = 0;
  int64_t dist_sum = 0;
  const int ref = xd->mi[0]->mbmi.ref_frame[0];
  unsigned int sse;
  unsigned int var = 0;
  unsigned int sum_sse = 0;
  int64_t total_sse = 0;
  int skip_flag = 1;
  const int shift = 6;
  int rate;
  int64_t dist;
  const int dequant_shift =
#if CONFIG_VP9_HIGHBITDEPTH
      (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) ?
          xd->bd - 5 :
#endif  // CONFIG_VP9_HIGHBITDEPTH
          3;

  x->pred_sse[ref] = 0;

  for (i = 0; i < MAX_MB_PLANE; ++i) {
    struct macroblock_plane *const p = &x->plane[i];
    struct macroblockd_plane *const pd = &xd->plane[i];
    const BLOCK_SIZE bs = get_plane_block_size(bsize, pd);
    const TX_SIZE max_tx_size = max_txsize_lookup[bs];
    const BLOCK_SIZE unit_size = txsize_to_bsize[max_tx_size];
    const int64_t dc_thr = p->quant_thred[0] >> shift;
    const int64_t ac_thr = p->quant_thred[1] >> shift;
    // The low thresholds are used to measure if the prediction errors are
    // low enough so that we can skip the mode search.
    const int64_t low_dc_thr = MIN(50, dc_thr >> 2);
    const int64_t low_ac_thr = MIN(80, ac_thr >> 2);
    int bw = 1 << (b_width_log2_lookup[bs] - b_width_log2_lookup[unit_size]);
    int bh = 1 << (b_height_log2_lookup[bs] - b_width_log2_lookup[unit_size]);
    int idx, idy;
    int lw = b_width_log2_lookup[unit_size] + 2;
    int lh = b_height_log2_lookup[unit_size] + 2;

    sum_sse = 0;

    for (idy = 0; idy < bh; ++idy) {
      for (idx = 0; idx < bw; ++idx) {
        uint8_t *src = p->src.buf + (idy * p->src.stride << lh) + (idx << lw);
        uint8_t *dst = pd->dst.buf + (idy * pd->dst.stride << lh) + (idx << lh);
        int block_idx = (idy << 1) + idx;
        int low_err_skip = 0;

        var = cpi->fn_ptr[unit_size].vf(src, p->src.stride,
                                        dst, pd->dst.stride, &sse);
        x->bsse[(i << 2) + block_idx] = sse;
        sum_sse += sse;

        x->skip_txfm[(i << 2) + block_idx] = SKIP_TXFM_NONE;
        if (!x->select_tx_size) {
          // Check if all ac coefficients can be quantized to zero.
          if (var < ac_thr || var == 0) {
            x->skip_txfm[(i << 2) + block_idx] = SKIP_TXFM_AC_ONLY;

            // Check if dc coefficient can be quantized to zero.
            if (sse - var < dc_thr || sse == var) {
              x->skip_txfm[(i << 2) + block_idx] = SKIP_TXFM_AC_DC;

              if (!sse || (var < low_ac_thr && sse - var < low_dc_thr))
                low_err_skip = 1;
            }
          }
        }

        if (skip_flag && !low_err_skip)
          skip_flag = 0;

        if (i == 0)
          x->pred_sse[ref] += sse;
      }
    }

    total_sse += sum_sse;

    // Fast approximate the modelling function.
    if (cpi->sf.simple_model_rd_from_var) {
      int64_t rate;
      const int64_t square_error = sum_sse;
      int quantizer = (pd->dequant[1] >> dequant_shift);

      if (quantizer < 120)
        rate = (square_error * (280 - quantizer)) >> 8;
      else
        rate = 0;
      dist = (square_error * quantizer) >> 8;
      rate_sum += rate;
      dist_sum += dist;
    } else {
      vp10_model_rd_from_var_lapndz(sum_sse, num_pels_log2_lookup[bs],
                                   pd->dequant[1] >> dequant_shift,
                                   &rate, &dist);
      rate_sum += rate;
      dist_sum += dist;
    }
  }

  *skip_txfm_sb = skip_flag;
  *skip_sse_sb = total_sse << 4;
  *out_rate_sum = (int)rate_sum;
  *out_dist_sum = dist_sum << 4;
}

int64_t vp10_block_error_c(const tran_low_t *coeff, const tran_low_t *dqcoeff,
                          intptr_t block_size, int64_t *ssz) {
  int i;
  int64_t error = 0, sqcoeff = 0;

  for (i = 0; i < block_size; i++) {
    const int diff = coeff[i] - dqcoeff[i];
    error +=  diff * diff;
    sqcoeff += coeff[i] * coeff[i];
  }

  *ssz = sqcoeff;
  return error;
}

int64_t vp10_block_error_fp_c(const int16_t *coeff, const int16_t *dqcoeff,
                             int block_size) {
  int i;
  int64_t error = 0;

  for (i = 0; i < block_size; i++) {
    const int diff = coeff[i] - dqcoeff[i];
    error +=  diff * diff;
  }

  return error;
}

#if CONFIG_VP9_HIGHBITDEPTH
int64_t vp10_highbd_block_error_c(const tran_low_t *coeff,
                                 const tran_low_t *dqcoeff,
                                 intptr_t block_size,
                                 int64_t *ssz, int bd) {
  int i;
  int64_t error = 0, sqcoeff = 0;
  int shift = 2 * (bd - 8);
  int rounding = shift > 0 ? 1 << (shift - 1) : 0;

  for (i = 0; i < block_size; i++) {
    const int64_t diff = coeff[i] - dqcoeff[i];
    error +=  diff * diff;
    sqcoeff += (int64_t)coeff[i] * (int64_t)coeff[i];
  }
  assert(error >= 0 && sqcoeff >= 0);
  error = (error + rounding) >> shift;
  sqcoeff = (sqcoeff + rounding) >> shift;

  *ssz = sqcoeff;
  return error;
}
#endif  // CONFIG_VP9_HIGHBITDEPTH

/* The trailing '0' is a terminator which is used inside cost_coeffs() to
 * decide whether to include cost of a trailing EOB node or not (i.e. we
 * can skip this if the last coefficient in this transform block, e.g. the
 * 16th coefficient in a 4x4 block or the 64th coefficient in a 8x8 block,
 * were non-zero). */
static const int16_t band_counts[TX_SIZES][8] = {
  { 1, 2, 3, 4,  3,   16 - 13, 0 },
  { 1, 2, 3, 4, 11,   64 - 21, 0 },
  { 1, 2, 3, 4, 11,  256 - 21, 0 },
  { 1, 2, 3, 4, 11, 1024 - 21, 0 },
};
static int cost_coeffs(MACROBLOCK *x,
                       int plane, int block,
                       ENTROPY_CONTEXT *A, ENTROPY_CONTEXT *L,
                       TX_SIZE tx_size,
                       const int16_t *scan, const int16_t *nb,
                       int use_fast_coef_costing) {
  MACROBLOCKD *const xd = &x->e_mbd;
  MB_MODE_INFO *mbmi = &xd->mi[0]->mbmi;
  const struct macroblock_plane *p = &x->plane[plane];
  const struct macroblockd_plane *pd = &xd->plane[plane];
  const PLANE_TYPE type = pd->plane_type;
  const int16_t *band_count = &band_counts[tx_size][1];
  const int eob = p->eobs[block];
  const tran_low_t *const qcoeff = BLOCK_OFFSET(p->qcoeff, block);
  unsigned int (*token_costs)[2][COEFF_CONTEXTS][ENTROPY_TOKENS] =
                   x->token_costs[tx_size][type][is_inter_block(mbmi)];
  uint8_t token_cache[32 * 32];
  int pt = combine_entropy_contexts(*A, *L);
  int c, cost;
#if CONFIG_VP9_HIGHBITDEPTH
  const int16_t *cat6_high_cost = vp10_get_high_cost_table(xd->bd);
#else
  const int16_t *cat6_high_cost = vp10_get_high_cost_table(8);
#endif

  // Check for consistency of tx_size with mode info
  assert(type == PLANE_TYPE_Y ? mbmi->tx_size == tx_size
                              : get_uv_tx_size(mbmi, pd) == tx_size);

  if (eob == 0) {
    // single eob token
    cost = token_costs[0][0][pt][EOB_TOKEN];
    c = 0;
  } else {
    int band_left = *band_count++;

    // dc token
    int v = qcoeff[0];
    int16_t prev_t;
    EXTRABIT e;
    vp10_get_token_extra(v, &prev_t, &e);
    cost = (*token_costs)[0][pt][prev_t] +
        vp10_get_cost(prev_t, e, cat6_high_cost);

    token_cache[0] = vp10_pt_energy_class[prev_t];
    ++token_costs;

    // ac tokens
    for (c = 1; c < eob; c++) {
      const int rc = scan[c];
      int16_t t;

      v = qcoeff[rc];
      vp10_get_token_extra(v, &t, &e);
      if (use_fast_coef_costing) {
        cost += (*token_costs)[!prev_t][!prev_t][t] +
            vp10_get_cost(t, e, cat6_high_cost);
      } else {
        pt = get_coef_context(nb, token_cache, c);
        cost += (*token_costs)[!prev_t][pt][t] +
            vp10_get_cost(t, e, cat6_high_cost);
        token_cache[rc] = vp10_pt_energy_class[t];
      }
      prev_t = t;
      if (!--band_left) {
        band_left = *band_count++;
        ++token_costs;
      }
    }

    // eob token
    if (band_left) {
      if (use_fast_coef_costing) {
        cost += (*token_costs)[0][!prev_t][EOB_TOKEN];
      } else {
        pt = get_coef_context(nb, token_cache, c);
        cost += (*token_costs)[0][pt][EOB_TOKEN];
      }
    }
  }

  // is eob first coefficient;
  *A = *L = (c > 0);

  return cost;
}

static void dist_block(MACROBLOCK *x, int plane, int block, TX_SIZE tx_size,
                       int64_t *out_dist, int64_t *out_sse) {
  const int ss_txfrm_size = tx_size << 1;
  MACROBLOCKD* const xd = &x->e_mbd;
  const struct macroblock_plane *const p = &x->plane[plane];
  const struct macroblockd_plane *const pd = &xd->plane[plane];
  int64_t this_sse;
  int shift = tx_size == TX_32X32 ? 0 : 2;
  tran_low_t *const coeff = BLOCK_OFFSET(p->coeff, block);
  tran_low_t *const dqcoeff = BLOCK_OFFSET(pd->dqcoeff, block);
#if CONFIG_VP9_HIGHBITDEPTH
  const int bd = (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) ? xd->bd : 8;
  *out_dist = vp10_highbd_block_error(coeff, dqcoeff, 16 << ss_txfrm_size,
                                     &this_sse, bd) >> shift;
#else
  *out_dist = vp10_block_error(coeff, dqcoeff, 16 << ss_txfrm_size,
                              &this_sse) >> shift;
#endif  // CONFIG_VP9_HIGHBITDEPTH
  *out_sse = this_sse >> shift;

  if (x->skip_encode && !is_inter_block(&xd->mi[0]->mbmi)) {
    // TODO(jingning): tune the model to better capture the distortion.
    int64_t p = (pd->dequant[1] * pd->dequant[1] *
                    (1 << ss_txfrm_size)) >>
#if CONFIG_VP9_HIGHBITDEPTH
                        (shift + 2 + (bd - 8) * 2);
#else
                        (shift + 2);
#endif  // CONFIG_VP9_HIGHBITDEPTH
    *out_dist += (p >> 4);
    *out_sse  += p;
  }
}

static int rate_block(int plane, int block, BLOCK_SIZE plane_bsize,
                      TX_SIZE tx_size, struct rdcost_block_args* args) {
  int x_idx, y_idx;
  txfrm_block_to_raster_xy(plane_bsize, tx_size, block, &x_idx, &y_idx);

  return cost_coeffs(args->x, plane, block, args->t_above + x_idx,
                     args->t_left + y_idx, tx_size,
                     args->so->scan, args->so->neighbors,
                     args->use_fast_coef_costing);
}

static void block_rd_txfm(int plane, int block, BLOCK_SIZE plane_bsize,
                          TX_SIZE tx_size, void *arg) {
  struct rdcost_block_args *args = arg;
  MACROBLOCK *const x = args->x;
  MACROBLOCKD *const xd = &x->e_mbd;
  MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;
  int64_t rd1, rd2, rd;
  int rate;
  int64_t dist;
  int64_t sse;

  if (args->exit_early)
    return;

  if (!is_inter_block(mbmi)) {
    struct encode_b_args arg = {x, NULL, &mbmi->skip};
    vp10_encode_block_intra(plane, block, plane_bsize, tx_size, &arg);
    dist_block(x, plane, block, tx_size, &dist, &sse);
  } else if (max_txsize_lookup[plane_bsize] == tx_size) {
    if (x->skip_txfm[(plane << 2) + (block >> (tx_size << 1))] ==
        SKIP_TXFM_NONE) {
      // full forward transform and quantization
      vp10_xform_quant(x, plane, block, plane_bsize, tx_size);
      dist_block(x, plane, block, tx_size, &dist, &sse);
    } else if (x->skip_txfm[(plane << 2) + (block >> (tx_size << 1))] ==
               SKIP_TXFM_AC_ONLY) {
      // compute DC coefficient
      tran_low_t *const coeff   = BLOCK_OFFSET(x->plane[plane].coeff, block);
      tran_low_t *const dqcoeff = BLOCK_OFFSET(xd->plane[plane].dqcoeff, block);
      vp10_xform_quant_dc(x, plane, block, plane_bsize, tx_size);
      sse  = x->bsse[(plane << 2) + (block >> (tx_size << 1))] << 4;
      dist = sse;
      if (x->plane[plane].eobs[block]) {
        const int64_t orig_sse = (int64_t)coeff[0] * coeff[0];
        const int64_t resd_sse = coeff[0] - dqcoeff[0];
        int64_t dc_correct = orig_sse - resd_sse * resd_sse;
#if CONFIG_VP9_HIGHBITDEPTH
        dc_correct >>= ((xd->bd - 8) * 2);
#endif
        if (tx_size != TX_32X32)
          dc_correct >>= 2;

        dist = MAX(0, sse - dc_correct);
      }
    } else {
      // SKIP_TXFM_AC_DC
      // skip forward transform
      x->plane[plane].eobs[block] = 0;
      sse  = x->bsse[(plane << 2) + (block >> (tx_size << 1))] << 4;
      dist = sse;
    }
  } else {
    // full forward transform and quantization
    vp10_xform_quant(x, plane, block, plane_bsize, tx_size);
    dist_block(x, plane, block, tx_size, &dist, &sse);
  }

  rd = RDCOST(x->rdmult, x->rddiv, 0, dist);
  if (args->this_rd + rd > args->best_rd) {
    args->exit_early = 1;
    return;
  }

  rate = rate_block(plane, block, plane_bsize, tx_size, args);
  rd1 = RDCOST(x->rdmult, x->rddiv, rate, dist);
  rd2 = RDCOST(x->rdmult, x->rddiv, 0, sse);

  // TODO(jingning): temporarily enabled only for luma component
  rd = MIN(rd1, rd2);
  if (plane == 0)
    x->zcoeff_blk[tx_size][block] = !x->plane[plane].eobs[block] ||
                                    (rd1 > rd2 && !xd->lossless);

  args->this_rate += rate;
  args->this_dist += dist;
  args->this_sse += sse;
  args->this_rd += rd;

  if (args->this_rd > args->best_rd) {
    args->exit_early = 1;
    return;
  }

  args->skippable &= !x->plane[plane].eobs[block];
}

static void txfm_rd_in_plane(MACROBLOCK *x,
                             int *rate, int64_t *distortion,
                             int *skippable, int64_t *sse,
                             int64_t ref_best_rd, int plane,
                             BLOCK_SIZE bsize, TX_SIZE tx_size,
                             int use_fast_coef_casting) {
  MACROBLOCKD *const xd = &x->e_mbd;
  const struct macroblockd_plane *const pd = &xd->plane[plane];
hui su's avatar
hui su committed
560
  TX_TYPE tx_type;
Jingning Han's avatar
Jingning Han committed
561 562 563 564 565 566 567 568 569 570 571 572
  struct rdcost_block_args args;
  vp10_zero(args);
  args.x = x;
  args.best_rd = ref_best_rd;
  args.use_fast_coef_costing = use_fast_coef_casting;
  args.skippable = 1;

  if (plane == 0)
    xd->mi[0]->mbmi.tx_size = tx_size;

  vp10_get_entropy_contexts(bsize, tx_size, pd, args.t_above, args.t_left);

hui su's avatar
hui su committed
573
  tx_type = get_tx_type(pd->plane_type, xd, 0, tx_size);
hui su's avatar
hui su committed
574
  args.so = get_scan(tx_size, tx_type);
Jingning Han's avatar
Jingning Han committed
575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590

  vp10_foreach_transformed_block_in_plane(xd, bsize, plane,
                                         block_rd_txfm, &args);
  if (args.exit_early) {
    *rate       = INT_MAX;
    *distortion = INT64_MAX;
    *sse        = INT64_MAX;
    *skippable  = 0;
  } else {
    *distortion = args.this_dist;
    *rate       = args.this_rate;
    *sse        = args.this_sse;
    *skippable  = args.skippable;
  }
}

Yaowu Xu's avatar
Yaowu Xu committed
591
static void choose_largest_tx_size(VP10_COMP *cpi, MACROBLOCK *x,
Jingning Han's avatar
Jingning Han committed
592 593 594 595 596
                                   int *rate, int64_t *distortion,
                                   int *skip, int64_t *sse,
                                   int64_t ref_best_rd,
                                   BLOCK_SIZE bs) {
  const TX_SIZE max_tx_size = max_txsize_lookup[bs];
Yaowu Xu's avatar
Yaowu Xu committed
597
  VP10_COMMON *const cm = &cpi->common;
Jingning Han's avatar
Jingning Han committed
598 599 600 601 602 603 604 605 606 607 608
  const TX_SIZE largest_tx_size = tx_mode_to_biggest_tx_size[cm->tx_mode];
  MACROBLOCKD *const xd = &x->e_mbd;
  MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;

  mbmi->tx_size = MIN(max_tx_size, largest_tx_size);

  txfm_rd_in_plane(x, rate, distortion, skip,
                   sse, ref_best_rd, 0, bs,
                   mbmi->tx_size, cpi->sf.use_fast_coef_costing);
}

Yaowu Xu's avatar
Yaowu Xu committed
609
static void choose_tx_size_from_rd(VP10_COMP *cpi, MACROBLOCK *x,
Jingning Han's avatar
Jingning Han committed
610 611 612 613 614 615 616
                                   int *rate,
                                   int64_t *distortion,
                                   int *skip,
                                   int64_t *psse,
                                   int64_t ref_best_rd,
                                   BLOCK_SIZE bs) {
  const TX_SIZE max_tx_size = max_txsize_lookup[bs];
Yaowu Xu's avatar
Yaowu Xu committed
617
  VP10_COMMON *const cm = &cpi->common;
Jingning Han's avatar
Jingning Han committed
618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648
  MACROBLOCKD *const xd = &x->e_mbd;
  MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;
  vpx_prob skip_prob = vp10_get_skip_prob(cm, xd);
  int r[TX_SIZES][2], s[TX_SIZES];
  int64_t d[TX_SIZES], sse[TX_SIZES];
  int64_t rd[TX_SIZES][2] = {{INT64_MAX, INT64_MAX},
                             {INT64_MAX, INT64_MAX},
                             {INT64_MAX, INT64_MAX},
                             {INT64_MAX, INT64_MAX}};
  int n, m;
  int s0, s1;
  int64_t best_rd = INT64_MAX;
  TX_SIZE best_tx = max_tx_size;
  int start_tx, end_tx;

  const vpx_prob *tx_probs = get_tx_probs2(max_tx_size, xd, &cm->fc->tx_probs);
  assert(skip_prob > 0);
  s0 = vp10_cost_bit(skip_prob, 0);
  s1 = vp10_cost_bit(skip_prob, 1);

  if (cm->tx_mode == TX_MODE_SELECT) {
    start_tx = max_tx_size;
    end_tx = 0;
  } else {
    TX_SIZE chosen_tx_size = MIN(max_tx_size,
                                 tx_mode_to_biggest_tx_size[cm->tx_mode]);
    start_tx = chosen_tx_size;
    end_tx = chosen_tx_size;
  }

  for (n = start_tx; n >= end_tx; n--) {
649 650 651 652 653 654 655
    int r_tx_size = 0;
    for (m = 0; m <= n - (n == (int) max_tx_size); m++) {
      if (m == n)
        r_tx_size += vp10_cost_zero(tx_probs[m]);
      else
        r_tx_size += vp10_cost_one(tx_probs[m]);
    }
Jingning Han's avatar
Jingning Han committed
656 657 658 659 660
    txfm_rd_in_plane(x, &r[n][0], &d[n], &s[n],
                     &sse[n], ref_best_rd, 0, bs, n,
                     cpi->sf.use_fast_coef_costing);
    r[n][1] = r[n][0];
    if (r[n][0] < INT_MAX) {
661
      r[n][1] += r_tx_size;
Jingning Han's avatar
Jingning Han committed
662
    }
663
    if (d[n] == INT64_MAX || r[n][0] == INT_MAX) {
Jingning Han's avatar
Jingning Han committed
664 665
      rd[n][0] = rd[n][1] = INT64_MAX;
    } else if (s[n]) {
666 667 668 669 670 671
      if (is_inter_block(mbmi)) {
        rd[n][0] = rd[n][1] = RDCOST(x->rdmult, x->rddiv, s1, sse[n]);
      } else {
        rd[n][0] = RDCOST(x->rdmult, x->rddiv, s1, sse[n]);
        rd[n][1] = RDCOST(x->rdmult, x->rddiv, s1 + r_tx_size, sse[n]);
      }
Jingning Han's avatar
Jingning Han committed
672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696
    } else {
      rd[n][0] = RDCOST(x->rdmult, x->rddiv, r[n][0] + s0, d[n]);
      rd[n][1] = RDCOST(x->rdmult, x->rddiv, r[n][1] + s0, d[n]);
    }

    // Early termination in transform size search.
    if (cpi->sf.tx_size_search_breakout &&
        (rd[n][1] == INT64_MAX ||
        (n < (int) max_tx_size && rd[n][1] > rd[n + 1][1]) ||
        s[n] == 1))
      break;

    if (rd[n][1] < best_rd) {
      best_tx = n;
      best_rd = rd[n][1];
    }
  }
  mbmi->tx_size = best_tx;

  *distortion = d[mbmi->tx_size];
  *rate       = r[mbmi->tx_size][cm->tx_mode == TX_MODE_SELECT];
  *skip       = s[mbmi->tx_size];
  *psse       = sse[mbmi->tx_size];
}

Yaowu Xu's avatar
Yaowu Xu committed
697
static void super_block_yrd(VP10_COMP *cpi, MACROBLOCK *x, int *rate,
Jingning Han's avatar
Jingning Han committed
698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736
                            int64_t *distortion, int *skip,
                            int64_t *psse, BLOCK_SIZE bs,
                            int64_t ref_best_rd) {
  MACROBLOCKD *xd = &x->e_mbd;
  int64_t sse;
  int64_t *ret_sse = psse ? psse : &sse;

  assert(bs == xd->mi[0]->mbmi.sb_type);

  if (cpi->sf.tx_size_search_method == USE_LARGESTALL || xd->lossless) {
    choose_largest_tx_size(cpi, x, rate, distortion, skip, ret_sse, ref_best_rd,
                           bs);
  } else {
    choose_tx_size_from_rd(cpi, x, rate, distortion, skip, ret_sse,
                           ref_best_rd, bs);
  }
}

static int conditional_skipintra(PREDICTION_MODE mode,
                                 PREDICTION_MODE best_intra_mode) {
  if (mode == D117_PRED &&
      best_intra_mode != V_PRED &&
      best_intra_mode != D135_PRED)
    return 1;
  if (mode == D63_PRED &&
      best_intra_mode != V_PRED &&
      best_intra_mode != D45_PRED)
    return 1;
  if (mode == D207_PRED &&
      best_intra_mode != H_PRED &&
      best_intra_mode != D45_PRED)
    return 1;
  if (mode == D153_PRED &&
      best_intra_mode != H_PRED &&
      best_intra_mode != D135_PRED)
    return 1;
  return 0;
}

Yaowu Xu's avatar
Yaowu Xu committed
737
static int64_t rd_pick_intra4x4block(VP10_COMP *cpi, MACROBLOCK *x,
Jingning Han's avatar
Jingning Han committed
738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806
                                     int row, int col,
                                     PREDICTION_MODE *best_mode,
                                     const int *bmode_costs,
                                     ENTROPY_CONTEXT *a, ENTROPY_CONTEXT *l,
                                     int *bestrate, int *bestratey,
                                     int64_t *bestdistortion,
                                     BLOCK_SIZE bsize, int64_t rd_thresh) {
  PREDICTION_MODE mode;
  MACROBLOCKD *const xd = &x->e_mbd;
  int64_t best_rd = rd_thresh;
  struct macroblock_plane *p = &x->plane[0];
  struct macroblockd_plane *pd = &xd->plane[0];
  const int src_stride = p->src.stride;
  const int dst_stride = pd->dst.stride;
  const uint8_t *src_init = &p->src.buf[row * 4 * src_stride + col * 4];
  uint8_t *dst_init = &pd->dst.buf[row * 4 * src_stride + col * 4];
  ENTROPY_CONTEXT ta[2], tempa[2];
  ENTROPY_CONTEXT tl[2], templ[2];
  const int num_4x4_blocks_wide = num_4x4_blocks_wide_lookup[bsize];
  const int num_4x4_blocks_high = num_4x4_blocks_high_lookup[bsize];
  int idx, idy;
  uint8_t best_dst[8 * 8];
#if CONFIG_VP9_HIGHBITDEPTH
  uint16_t best_dst16[8 * 8];
#endif

  memcpy(ta, a, sizeof(ta));
  memcpy(tl, l, sizeof(tl));
  xd->mi[0]->mbmi.tx_size = TX_4X4;

#if CONFIG_VP9_HIGHBITDEPTH
  if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
    for (mode = DC_PRED; mode <= TM_PRED; ++mode) {
      int64_t this_rd;
      int ratey = 0;
      int64_t distortion = 0;
      int rate = bmode_costs[mode];

      if (!(cpi->sf.intra_y_mode_mask[TX_4X4] & (1 << mode)))
        continue;

      // Only do the oblique modes if the best so far is
      // one of the neighboring directional modes
      if (cpi->sf.mode_search_skip_flags & FLAG_SKIP_INTRA_DIRMISMATCH) {
        if (conditional_skipintra(mode, *best_mode))
            continue;
      }

      memcpy(tempa, ta, sizeof(ta));
      memcpy(templ, tl, sizeof(tl));

      for (idy = 0; idy < num_4x4_blocks_high; ++idy) {
        for (idx = 0; idx < num_4x4_blocks_wide; ++idx) {
          const int block = (row + idy) * 2 + (col + idx);
          const uint8_t *const src = &src_init[idx * 4 + idy * 4 * src_stride];
          uint8_t *const dst = &dst_init[idx * 4 + idy * 4 * dst_stride];
          int16_t *const src_diff = vp10_raster_block_offset_int16(BLOCK_8X8,
                                                                  block,
                                                                  p->src_diff);
          tran_low_t *const coeff = BLOCK_OFFSET(x->plane[0].coeff, block);
          xd->mi[0]->bmi[block].as_mode = mode;
          vp10_predict_intra_block(xd, 1, TX_4X4, mode,
                                  x->skip_encode ? src : dst,
                                  x->skip_encode ? src_stride : dst_stride,
                                  dst, dst_stride,
                                  col + idx, row + idy, 0);
          vpx_highbd_subtract_block(4, 4, src_diff, 8, src, src_stride,
                                    dst, dst_stride, xd->bd);
          if (xd->lossless) {
hui su's avatar
hui su committed
807
            TX_TYPE tx_type = get_tx_type(PLANE_TYPE_Y, xd, block, TX_4X4);
hui su's avatar
hui su committed
808
            const scan_order *so = get_scan(TX_4X4, tx_type);
hui su's avatar
hui su committed
809 810
            vp10_highbd_fwd_txfm_4x4(src_diff, coeff, 8, DCT_DCT,
                                     vp10_highbd_fwht4x4);
Jingning Han's avatar
Jingning Han committed
811 812 813 814 815 816
            vp10_regular_quantize_b_4x4(x, 0, block, so->scan, so->iscan);
            ratey += cost_coeffs(x, 0, block, tempa + idx, templ + idy, TX_4X4,
                                 so->scan, so->neighbors,
                                 cpi->sf.use_fast_coef_costing);
            if (RDCOST(x->rdmult, x->rddiv, ratey, distortion) >= best_rd)
              goto next_highbd;
hui su's avatar
hui su committed
817 818 819 820
            vp10_highbd_inv_txfm_add_4x4(BLOCK_OFFSET(pd->dqcoeff, block),
                                         dst, dst_stride, p->eobs[block],
                                         xd->bd, DCT_DCT,
                                         vp10_highbd_iwht4x4_add);
Jingning Han's avatar
Jingning Han committed
821 822
          } else {
            int64_t unused;
hui su's avatar
hui su committed
823
            TX_TYPE tx_type = get_tx_type(PLANE_TYPE_Y, xd, block, TX_4X4);
hui su's avatar
hui su committed
824
            const scan_order *so = get_scan(TX_4X4, tx_type);
hui su's avatar
hui su committed
825 826
            vp10_highbd_fwd_txfm_4x4(src_diff, coeff, 8, tx_type,
                                     vpx_highbd_fdct4x4);
Jingning Han's avatar
Jingning Han committed
827 828 829 830 831 832 833 834 835
            vp10_regular_quantize_b_4x4(x, 0, block, so->scan, so->iscan);
            ratey += cost_coeffs(x, 0, block, tempa + idx, templ + idy, TX_4X4,
                                 so->scan, so->neighbors,
                                 cpi->sf.use_fast_coef_costing);
            distortion += vp10_highbd_block_error(
                coeff, BLOCK_OFFSET(pd->dqcoeff, block),
                16, &unused, xd->bd) >> 2;
            if (RDCOST(x->rdmult, x->rddiv, ratey, distortion) >= best_rd)
              goto next_highbd;
hui su's avatar
hui su committed
836 837 838 839
            vp10_highbd_inv_txfm_add_4x4(BLOCK_OFFSET(pd->dqcoeff, block),
                                         dst, dst_stride, p->eobs[block],
                                         xd->bd, tx_type,
                                         vp10_highbd_idct4x4_add);
Jingning Han's avatar
Jingning Han committed
840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911
          }
        }
      }

      rate += ratey;
      this_rd = RDCOST(x->rdmult, x->rddiv, rate, distortion);

      if (this_rd < best_rd) {
        *bestrate = rate;
        *bestratey = ratey;
        *bestdistortion = distortion;
        best_rd = this_rd;
        *best_mode = mode;
        memcpy(a, tempa, sizeof(tempa));
        memcpy(l, templ, sizeof(templ));
        for (idy = 0; idy < num_4x4_blocks_high * 4; ++idy) {
          memcpy(best_dst16 + idy * 8,
                 CONVERT_TO_SHORTPTR(dst_init + idy * dst_stride),
                 num_4x4_blocks_wide * 4 * sizeof(uint16_t));
        }
      }
    next_highbd:
      {}
    }
    if (best_rd >= rd_thresh || x->skip_encode)
      return best_rd;

    for (idy = 0; idy < num_4x4_blocks_high * 4; ++idy) {
      memcpy(CONVERT_TO_SHORTPTR(dst_init + idy * dst_stride),
             best_dst16 + idy * 8,
             num_4x4_blocks_wide * 4 * sizeof(uint16_t));
    }

    return best_rd;
  }
#endif  // CONFIG_VP9_HIGHBITDEPTH

  for (mode = DC_PRED; mode <= TM_PRED; ++mode) {
    int64_t this_rd;
    int ratey = 0;
    int64_t distortion = 0;
    int rate = bmode_costs[mode];

    if (!(cpi->sf.intra_y_mode_mask[TX_4X4] & (1 << mode)))
      continue;

    // Only do the oblique modes if the best so far is
    // one of the neighboring directional modes
    if (cpi->sf.mode_search_skip_flags & FLAG_SKIP_INTRA_DIRMISMATCH) {
      if (conditional_skipintra(mode, *best_mode))
          continue;
    }

    memcpy(tempa, ta, sizeof(ta));
    memcpy(templ, tl, sizeof(tl));

    for (idy = 0; idy < num_4x4_blocks_high; ++idy) {
      for (idx = 0; idx < num_4x4_blocks_wide; ++idx) {
        const int block = (row + idy) * 2 + (col + idx);
        const uint8_t *const src = &src_init[idx * 4 + idy * 4 * src_stride];
        uint8_t *const dst = &dst_init[idx * 4 + idy * 4 * dst_stride];
        int16_t *const src_diff =
            vp10_raster_block_offset_int16(BLOCK_8X8, block, p->src_diff);
        tran_low_t *const coeff = BLOCK_OFFSET(x->plane[0].coeff, block);
        xd->mi[0]->bmi[block].as_mode = mode;
        vp10_predict_intra_block(xd, 1, TX_4X4, mode,
                                x->skip_encode ? src : dst,
                                x->skip_encode ? src_stride : dst_stride,
                                dst, dst_stride, col + idx, row + idy, 0);
        vpx_subtract_block(4, 4, src_diff, 8, src, src_stride, dst, dst_stride);

        if (xd->lossless) {
hui su's avatar
hui su committed
912
          TX_TYPE tx_type = get_tx_type(PLANE_TYPE_Y, xd, block, TX_4X4);
hui su's avatar
hui su committed
913
          const scan_order *so = get_scan(TX_4X4, tx_type);
hui su's avatar
hui su committed
914
          vp10_fwd_txfm_4x4(src_diff, coeff, 8, DCT_DCT, vp10_fwht4x4);
Jingning Han's avatar
Jingning Han committed
915 916 917 918 919 920
          vp10_regular_quantize_b_4x4(x, 0, block, so->scan, so->iscan);
          ratey += cost_coeffs(x, 0, block, tempa + idx, templ + idy, TX_4X4,
                               so->scan, so->neighbors,
                               cpi->sf.use_fast_coef_costing);
          if (RDCOST(x->rdmult, x->rddiv, ratey, distortion) >= best_rd)
            goto next;
hui su's avatar
hui su committed
921 922 923
          vp10_inv_txfm_add_4x4(BLOCK_OFFSET(pd->dqcoeff, block),
                                dst, dst_stride, p->eobs[block], DCT_DCT,
                                vp10_iwht4x4_add);
Jingning Han's avatar
Jingning Han committed
924 925
        } else {
          int64_t unused;
hui su's avatar
hui su committed
926
          TX_TYPE tx_type = get_tx_type(PLANE_TYPE_Y, xd, block, TX_4X4);
hui su's avatar
hui su committed
927
          const scan_order *so = get_scan(TX_4X4, tx_type);
hui su's avatar
hui su committed
928
          vp10_fwd_txfm_4x4(src_diff, coeff, 8, tx_type, vpx_fdct4x4);
Jingning Han's avatar
Jingning Han committed
929 930 931 932 933 934 935 936
          vp10_regular_quantize_b_4x4(x, 0, block, so->scan, so->iscan);
          ratey += cost_coeffs(x, 0, block, tempa + idx, templ + idy, TX_4X4,
                             so->scan, so->neighbors,
                             cpi->sf.use_fast_coef_costing);
          distortion += vp10_block_error(coeff, BLOCK_OFFSET(pd->dqcoeff, block),
                                        16, &unused) >> 2;
          if (RDCOST(x->rdmult, x->rddiv, ratey, distortion) >= best_rd)
            goto next;
hui su's avatar
hui su committed
937 938 939
          vp10_inv_txfm_add_4x4(BLOCK_OFFSET(pd->dqcoeff, block),
                                dst, dst_stride, p->eobs[block], tx_type,
                                vp10_idct4x4_add);
Jingning Han's avatar
Jingning Han committed
940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972
        }
      }
    }

    rate += ratey;
    this_rd = RDCOST(x->rdmult, x->rddiv, rate, distortion);

    if (this_rd < best_rd) {
      *bestrate = rate;
      *bestratey = ratey;
      *bestdistortion = distortion;
      best_rd = this_rd;
      *best_mode = mode;
      memcpy(a, tempa, sizeof(tempa));
      memcpy(l, templ, sizeof(templ));
      for (idy = 0; idy < num_4x4_blocks_high * 4; ++idy)
        memcpy(best_dst + idy * 8, dst_init + idy * dst_stride,
               num_4x4_blocks_wide * 4);
    }
  next:
    {}
  }

  if (best_rd >= rd_thresh || x->skip_encode)
    return best_rd;

  for (idy = 0; idy < num_4x4_blocks_high * 4; ++idy)
    memcpy(dst_init + idy * dst_stride, best_dst + idy * 8,
           num_4x4_blocks_wide * 4);

  return best_rd;
}

Yaowu Xu's avatar
Yaowu Xu committed
973
static int64_t rd_pick_intra_sub_8x8_y_mode(VP10_COMP *cpi, MACROBLOCK *mb,
Jingning Han's avatar
Jingning Han committed
974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 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
                                            int *rate, int *rate_y,
                                            int64_t *distortion,
                                            int64_t best_rd) {
  int i, j;
  const MACROBLOCKD *const xd = &mb->e_mbd;
  MODE_INFO *const mic = xd->mi[0];
  const MODE_INFO *above_mi = xd->above_mi;
  const MODE_INFO *left_mi = xd->left_mi;
  const BLOCK_SIZE bsize = xd->mi[0]->mbmi.sb_type;
  const int num_4x4_blocks_wide = num_4x4_blocks_wide_lookup[bsize];
  const int num_4x4_blocks_high = num_4x4_blocks_high_lookup[bsize];
  int idx, idy;
  int cost = 0;
  int64_t total_distortion = 0;
  int tot_rate_y = 0;
  int64_t total_rd = 0;
  ENTROPY_CONTEXT t_above[4], t_left[4];
  const int *bmode_costs = cpi->mbmode_cost;

  memcpy(t_above, xd->plane[0].above_context, sizeof(t_above));
  memcpy(t_left, xd->plane[0].left_context, sizeof(t_left));

  // Pick modes for each sub-block (of size 4x4, 4x8, or 8x4) in an 8x8 block.
  for (idy = 0; idy < 2; idy += num_4x4_blocks_high) {
    for (idx = 0; idx < 2; idx += num_4x4_blocks_wide) {
      PREDICTION_MODE best_mode = DC_PRED;
      int r = INT_MAX, ry = INT_MAX;
      int64_t d = INT64_MAX, this_rd = INT64_MAX;
      i = idy * 2 + idx;
      if (cpi->common.frame_type == KEY_FRAME) {
        const PREDICTION_MODE A = vp10_above_block_mode(mic, above_mi, i);
        const PREDICTION_MODE L = vp10_left_block_mode(mic, left_mi, i);

        bmode_costs  = cpi->y_mode_costs[A][L];
      }

      this_rd = rd_pick_intra4x4block(cpi, mb, idy, idx, &best_mode,
                                      bmode_costs, t_above + idx, t_left + idy,
                                      &r, &ry, &d, bsize, best_rd - total_rd);
      if (this_rd >= best_rd - total_rd)
        return INT64_MAX;

      total_rd += this_rd;
      cost += r;
      total_distortion += d;
      tot_rate_y += ry;

      mic->bmi[i].as_mode = best_mode;
      for (j = 1; j < num_4x4_blocks_high; ++j)
        mic->bmi[i + j * 2].as_mode = best_mode;
      for (j = 1; j < num_4x4_blocks_wide; ++j)
        mic->bmi[i + j].as_mode = best_mode;

      if (total_rd >= best_rd)
        return INT64_MAX;
    }
  }

  *rate = cost;
  *rate_y = tot_rate_y;
  *distortion = total_distortion;
  mic->mbmi.mode = mic->bmi[3].as_mode;

  return RDCOST(mb->rdmult, mb->rddiv, cost, total_distortion);
}

// This function is used only for intra_only frames
Yaowu Xu's avatar
Yaowu Xu committed
1041
static int64_t rd_pick_intra_sby_mode(VP10_COMP *cpi, MACROBLOCK *x,
Jingning Han's avatar
Jingning Han committed
1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102
                                      int *rate, int *rate_tokenonly,
                                      int64_t *distortion, int *skippable,
                                      BLOCK_SIZE bsize,
                                      int64_t best_rd) {
  PREDICTION_MODE mode;
  PREDICTION_MODE mode_selected = DC_PRED;
  MACROBLOCKD *const xd = &x->e_mbd;
  MODE_INFO *const mic = xd->mi[0];
  int this_rate, this_rate_tokenonly, s;
  int64_t this_distortion, this_rd;
  TX_SIZE best_tx = TX_4X4;
  int *bmode_costs;
  const MODE_INFO *above_mi = xd->above_mi;
  const MODE_INFO *left_mi = xd->left_mi;
  const PREDICTION_MODE A = vp10_above_block_mode(mic, above_mi, 0);
  const PREDICTION_MODE L = vp10_left_block_mode(mic, left_mi, 0);
  bmode_costs = cpi->y_mode_costs[A][L];

  memset(x->skip_txfm, SKIP_TXFM_NONE, sizeof(x->skip_txfm));
  /* Y Search for intra prediction mode */
  for (mode = DC_PRED; mode <= TM_PRED; mode++) {

    if (cpi->sf.use_nonrd_pick_mode) {
      // These speed features are turned on in hybrid non-RD and RD mode
      // for key frame coding in the context of real-time setting.
      if (conditional_skipintra(mode, mode_selected))
          continue;
      if (*skippable)
        break;
    }

    mic->mbmi.mode = mode;

    super_block_yrd(cpi, x, &this_rate_tokenonly, &this_distortion,
        &s, NULL, bsize, best_rd);

    if (this_rate_tokenonly == INT_MAX)
      continue;

    this_rate = this_rate_tokenonly + bmode_costs[mode];
    this_rd = RDCOST(x->rdmult, x->rddiv, this_rate, this_distortion);

    if (this_rd < best_rd) {
      mode_selected   = mode;
      best_rd         = this_rd;
      best_tx         = mic->mbmi.tx_size;
      *rate           = this_rate;
      *rate_tokenonly = this_rate_tokenonly;
      *distortion     = this_distortion;
      *skippable      = s;
    }
  }

  mic->mbmi.mode = mode_selected;
  mic->mbmi.tx_size = best_tx;

  return best_rd;
}

// Return value 0: early termination triggered, no valid rd cost available;
//              1: rd cost values are valid.
Yaowu Xu's avatar
Yaowu Xu committed
1103
static int super_block_uvrd(const VP10_COMP *cpi, MACROBLOCK *x,
Jingning Han's avatar
Jingning Han committed
1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153
                            int *rate, int64_t *distortion, int *skippable,
                            int64_t *sse, BLOCK_SIZE bsize,
                            int64_t ref_best_rd) {
  MACROBLOCKD *const xd = &x->e_mbd;
  MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;
  const TX_SIZE uv_tx_size = get_uv_tx_size(mbmi, &xd->plane[1]);
  int plane;
  int pnrate = 0, pnskip = 1;
  int64_t pndist = 0, pnsse = 0;
  int is_cost_valid = 1;

  if (ref_best_rd < 0)
    is_cost_valid = 0;

  if (is_inter_block(mbmi) && is_cost_valid) {
    int plane;
    for (plane = 1; plane < MAX_MB_PLANE; ++plane)
      vp10_subtract_plane(x, bsize, plane);
  }

  *rate = 0;
  *distortion = 0;
  *sse = 0;
  *skippable = 1;

  for (plane = 1; plane < MAX_MB_PLANE; ++plane) {
    txfm_rd_in_plane(x, &pnrate, &pndist, &pnskip, &pnsse,
                     ref_best_rd, plane, bsize, uv_tx_size,
                     cpi->sf.use_fast_coef_costing);
    if (pnrate == INT_MAX) {
      is_cost_valid = 0;
      break;
    }
    *rate += pnrate;
    *distortion += pndist;
    *sse += pnsse;
    *skippable &= pnskip;
  }

  if (!is_cost_valid) {
    // reset cost value
    *rate = INT_MAX;
    *distortion = INT64_MAX;
    *sse = INT64_MAX;
    *skippable = 0;
  }

  return is_cost_valid;
}

Yaowu Xu's avatar
Yaowu Xu committed
1154
static int64_t rd_pick_intra_sbuv_mode(VP10_COMP *cpi, MACROBLOCK *x,
Jingning Han's avatar
Jingning Han committed
1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195
                                       PICK_MODE_CONTEXT *ctx,
                                       int *rate, int *rate_tokenonly,
                                       int64_t *distortion, int *skippable,
                                       BLOCK_SIZE bsize, TX_SIZE max_tx_size) {
  MACROBLOCKD *xd = &x->e_mbd;
  PREDICTION_MODE mode;
  PREDICTION_MODE mode_selected = DC_PRED;
  int64_t best_rd = INT64_MAX, this_rd;
  int this_rate_tokenonly, this_rate, s;
  int64_t this_distortion, this_sse;

  memset(x->skip_txfm, SKIP_TXFM_NONE, sizeof(x->skip_txfm));
  for (mode = DC_PRED; mode <= TM_PRED; ++mode) {
    if (!(cpi->sf.intra_uv_mode_mask[max_tx_size] & (1 << mode)))
      continue;

    xd->mi[0]->mbmi.uv_mode = mode;

    if (!super_block_uvrd(cpi, x, &this_rate_tokenonly,
                          &this_distortion, &s, &this_sse, bsize, best_rd))
      continue;
    this_rate = this_rate_tokenonly +
                cpi->intra_uv_mode_cost[cpi->common.frame_type][mode];
    this_rd = RDCOST(x->rdmult, x->rddiv, this_rate, this_distortion);

    if (this_rd < best_rd) {
      mode_selected   = mode;
      best_rd         = this_rd;
      *rate           = this_rate;
      *rate_tokenonly = this_rate_tokenonly;
      *distortion     = this_distortion;
      *skippable      = s;
      if (!x->select_tx_size)
        swap_block_ptr(x, ctx, 2, 0, 1, MAX_MB_PLANE);
    }
  }

  xd->mi[0]->mbmi.uv_mode = mode_selected;
  return best_rd;
}

Yaowu Xu's avatar
Yaowu Xu committed
1196
static int64_t rd_sbuv_dcpred(const VP10_COMP *cpi, MACROBLOCK *x,
Jingning Han's avatar
Jingning Han committed
1197 1198 1199
                              int *rate, int *rate_tokenonly,
                              int64_t *distortion, int *skippable,
                              BLOCK_SIZE bsize) {
Yaowu Xu's avatar
Yaowu Xu committed
1200
  const VP10_COMMON *cm = &cpi->common;
Jingning Han's avatar
Jingning Han committed
1201 1202 1203 1204 1205 1206 1207 1208 1209 1210
  int64_t unused;

  x->e_mbd.mi[0]->mbmi.uv_mode = DC_PRED;
  memset(x->skip_txfm, SKIP_TXFM_NONE, sizeof(x->skip_txfm));
  super_block_uvrd(cpi, x, rate_tokenonly, distortion,
                   skippable, &unused, bsize, INT64_MAX);
  *rate = *rate_tokenonly + cpi->intra_uv_mode_cost[cm->frame_type][DC_PRED];
  return RDCOST(x->rdmult, x->rddiv, *rate, *distortion);
}

Yaowu Xu's avatar
Yaowu Xu committed
1211
static void choose_intra_uv_mode(VP10_COMP *cpi, MACROBLOCK *const x,
Jingning Han's avatar
Jingning Han committed
1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231
                                 PICK_MODE_CONTEXT *ctx,
                                 BLOCK_SIZE bsize, TX_SIZE max_tx_size,
                                 int *rate_uv, int *rate_uv_tokenonly,
                                 int64_t *dist_uv, int *skip_uv,
                                 PREDICTION_MODE *mode_uv) {
  // Use an estimated rd for uv_intra based on DC_PRED if the
  // appropriate speed flag is set.
  if (cpi->sf.use_uv_intra_rd_estimate) {
    rd_sbuv_dcpred(cpi, x, rate_uv, rate_uv_tokenonly, dist_uv,
                   skip_uv, bsize < BLOCK_8X8 ? BLOCK_8X8 : bsize);
  // Else do a proper rd search for each possible transform size that may
  // be considered in the main rd loop.
  } else {
    rd_pick_intra_sbuv_mode(cpi, x, ctx,
                            rate_uv, rate_uv_tokenonly, dist_uv, skip_uv,
                            bsize < BLOCK_8X8 ? BLOCK_8X8 : bsize, max_tx_size);
  }
  *mode_uv = x->e_mbd.mi[0]->mbmi.uv_mode;
}

Yaowu Xu's avatar
Yaowu Xu committed
1232
static int cost_mv_ref(const VP10_COMP *cpi, PREDICTION_MODE mode,
Jingning Han's avatar
Jingning Han committed
1233 1234 1235 1236 1237
                       int mode_context) {
  assert(is_inter_mode(mode));
  return cpi->inter_mode_cost[mode_context][INTER_OFFSET(mode)];
}

Yaowu Xu's avatar
Yaowu Xu committed
1238
static int set_and_cost_bmi_mvs(VP10_COMP *cpi, MACROBLOCK *x, MACROBLOCKD *xd,
Jingning Han's avatar
Jingning Han committed
1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293
                                int i,
                                PREDICTION_MODE mode, int_mv this_mv[2],
                                int_mv frame_mv[MB_MODE_COUNT][MAX_REF_FRAMES],
                                int_mv seg_mvs[MAX_REF_FRAMES],
                                int_mv *best_ref_mv[2], const int *mvjcost,
                                int *mvcost[2]) {
  MODE_INFO *const mic = xd->mi[0];
  const MB_MODE_INFO *const mbmi = &mic->mbmi;
  const MB_MODE_INFO_EXT *const mbmi_ext = x->mbmi_ext;
  int thismvcost = 0;
  int idx, idy;
  const int num_4x4_blocks_wide = num_4x4_blocks_wide_lookup[mbmi->sb_type];
  const int num_4x4_blocks_high = num_4x4_blocks_high_lookup[mbmi->sb_type];
  const int is_compound = has_second_ref(mbmi);

  switch (mode) {
    case NEWMV:
      this_mv[0].as_int = seg_mvs[mbmi->ref_frame[0]].as_int;
      thismvcost += vp10_mv_bit_cost(&this_mv[0].as_mv, &best_ref_mv[0]->as_mv,
                                    mvjcost, mvcost, MV_COST_WEIGHT_SUB);
      if (is_compound) {
        this_mv[1].as_int = seg_mvs[mbmi->ref_frame[1]].as_int;
        thismvcost += vp10_mv_bit_cost(&this_mv[1].as_mv, &best_ref_mv[1]->as_mv,
                                      mvjcost, mvcost, MV_COST_WEIGHT_SUB);
      }
      break;
    case NEARMV:
    case NEARESTMV:
      this_mv[0].as_int = frame_mv[mode][mbmi->ref_frame[0]].as_int;
      if (is_compound)
        this_mv[1].as_int = frame_mv[mode][mbmi->ref_frame[1]].as_int;
      break;
    case ZEROMV:
      this_mv[0].as_int = 0;
      if (is_compound)
        this_mv[1].as_int = 0;
      break;
    default:
      break;
  }

  mic->bmi[i].as_mv[0].as_int = this_mv[0].as_int;
  if (is_compound)
    mic->bmi[i].as_mv[1].as_int = this_mv[1].as_int;

  mic->bmi[i].as_mode = mode;

  for (idy = 0; idy < num_4x4_blocks_high; ++idy)
    for (idx = 0; idx < num_4x4_blocks_wide; ++idx)
      memmove(&mic->bmi[i + idy * 2 + idx], &mic->bmi[i], sizeof(mic->bmi[i]));

  return cost_mv_ref(cpi, mode, mbmi_ext->mode_context[mbmi->ref_frame[0]]) +
            thismvcost;
}

Yaowu Xu's avatar
Yaowu Xu committed
1294
static int64_t encode_inter_mb_segment(VP10_COMP *cpi,
Jingning Han's avatar
Jingning Han committed
1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318
                                       MACROBLOCK *x,
                                       int64_t best_yrd,
                                       int i,
                                       int *labelyrate,
                                       int64_t *distortion, int64_t *sse,
                                       ENTROPY_CONTEXT *ta,
                                       ENTROPY_CONTEXT *tl,
                                       int mi_row, int mi_col) {
  int k;
  MACROBLOCKD *xd = &x->e_mbd;
  struct macroblockd_plane *const pd = &xd->plane[0];
  struct macroblock_plane *const p = &x->plane[0];
  MODE_INFO *const mi = xd->mi[0];
  const BLOCK_SIZE plane_bsize = get_plane_block_size(mi->mbmi.sb_type, pd);
  const int width = 4 * num_4x4_blocks_wide_lookup[plane_bsize];
  const int height = 4 * num_4x4_blocks_high_lookup[plane_bsize];
  int idx, idy;

  const uint8_t *const src =
      &p->src.buf[vp10_raster_block_offset(BLOCK_8X8, i, p->src.stride)];
  uint8_t *const dst = &pd->dst.buf[vp10_raster_block_offset(BLOCK_8X8, i,
                                                            pd->dst.stride)];
  int64_t thisdistortion = 0, thissse = 0;
  int thisrate = 0, ref;
hui su's avatar
hui su committed
1319
  TX_TYPE tx_type = get_tx_type(PLANE_TYPE_Y, xd, i, TX_4X4);
hui su's avatar
hui su committed
1320
  const scan_order *so = get_scan(TX_4X4, tx_type);
Jingning Han's avatar
Jingning Han committed
1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478
  const int is_compound = has_second_ref(&mi->mbmi);
  const InterpKernel *kernel = vp10_filter_kernels[mi->mbmi.interp_filter];

  for (ref = 0; ref < 1 + is_compound; ++ref) {
    const uint8_t *pre = &pd->pre[ref].buf[vp10_raster_block_offset(BLOCK_8X8, i,
                                               pd->pre[ref].stride)];
#if CONFIG_VP9_HIGHBITDEPTH
  if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
    vp10_highbd_build_inter_predictor(pre, pd->pre[ref].stride,
                                     dst, pd->dst.stride,
                                     &mi->bmi[i].as_mv[ref].as_mv,
                                     &xd->block_refs[ref]->sf, width, height,
                                     ref, kernel, MV_PRECISION_Q3,
                                     mi_col * MI_SIZE + 4 * (i % 2),
                                     mi_row * MI_SIZE + 4 * (i / 2), xd->bd);
  } else {
    vp10_build_inter_predictor(pre, pd->pre[ref].stride,
                              dst, pd->dst.stride,
                              &mi->bmi[i].as_mv[ref].as_mv,
                              &xd->block_refs[ref]->sf, width, height, ref,
                              kernel, MV_PRECISION_Q3,
                              mi_col * MI_SIZE + 4 * (i % 2),
                              mi_row * MI_SIZE + 4 * (i / 2));
  }
#else
    vp10_build_inter_predictor(pre, pd->pre[ref].stride,
                              dst, pd->dst.stride,
                              &mi->bmi[i].as_mv[ref].as_mv,
                              &xd->block_refs[ref]->sf, width, height, ref,
                              kernel, MV_PRECISION_Q3,
                              mi_col * MI_SIZE + 4 * (i % 2),
                              mi_row * MI_SIZE + 4 * (i / 2));
#endif  // CONFIG_VP9_HIGHBITDEPTH
  }

#if CONFIG_VP9_HIGHBITDEPTH
  if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
    vpx_highbd_subtract_block(
        height, width, vp10_raster_block_offset_int16(BLOCK_8X8, i, p->src_diff),
        8, src, p->src.stride, dst, pd->dst.stride, xd->bd);
  } else {
    vpx_subtract_block(
        height, width, vp10_raster_block_offset_int16(BLOCK_8X8, i, p->src_diff),
        8, src, p->src.stride, dst, pd->dst.stride);
  }
#else
  vpx_subtract_block(height, width,
                     vp10_raster_block_offset_int16(BLOCK_8X8, i, p->src_diff),
                     8, src, p->src.stride, dst, pd->dst.stride);
#endif  // CONFIG_VP9_HIGHBITDEPTH

  k = i;
  for (idy = 0; idy < height / 4; ++idy) {
    for (idx = 0; idx < width / 4; ++idx) {
      int64_t ssz, rd, rd1, rd2;
      tran_low_t* coeff;

      k += (idy * 2 + idx);
      coeff = BLOCK_OFFSET(p->coeff, k);
      x->fwd_txm4x4(vp10_raster_block_offset_int16(BLOCK_8X8, k, p->src_diff),
                    coeff, 8);
      vp10_regular_quantize_b_4x4(x, 0, k, so->scan, so->iscan);
#if CONFIG_VP9_HIGHBITDEPTH
      if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
        thisdistortion += vp10_highbd_block_error(coeff,
                                                 BLOCK_OFFSET(pd->dqcoeff, k),
                                                 16, &ssz, xd->bd);
      } else {
        thisdistortion += vp10_block_error(coeff, BLOCK_OFFSET(pd->dqcoeff, k),
                                          16, &ssz);
      }
#else
      thisdistortion += vp10_block_error(coeff, BLOCK_OFFSET(pd->dqcoeff, k),
                                        16, &ssz);
#endif  // CONFIG_VP9_HIGHBITDEPTH
      thissse += ssz;
      thisrate += cost_coeffs(x, 0, k, ta + (k & 1), tl + (k >> 1), TX_4X4,
                              so->scan, so->neighbors,
                              cpi->sf.use_fast_coef_costing);
      rd1 = RDCOST(x->rdmult, x->rddiv, thisrate, thisdistortion >> 2);
      rd2 = RDCOST(x->rdmult, x->rddiv, 0, thissse >> 2);
      rd = MIN(rd1, rd2);
      if (rd >= best_yrd)
        return INT64_MAX;
    }
  }

  *distortion = thisdistortion >> 2;
  *labelyrate = thisrate;
  *sse = thissse >> 2;

  return RDCOST(x->rdmult, x->rddiv, *labelyrate, *distortion);
}

typedef struct {
  int eobs;
  int brate;
  int byrate;
  int64_t bdist;
  int64_t bsse;
  int64_t brdcost;
  int_mv mvs[2];
  ENTROPY_CONTEXT ta[2];
  ENTROPY_CONTEXT tl[2];
} SEG_RDSTAT;

typedef struct {
  int_mv *ref_mv[2];
  int_mv mvp;

  int64_t segment_rd;
  int r;
  int64_t d;
  int64_t sse;
  int segment_yrate;
  PREDICTION_MODE modes[4];
  SEG_RDSTAT rdstat[4][INTER_MODES];
  int mvthresh;
} BEST_SEG_INFO;

static INLINE int mv_check_bounds(const MACROBLOCK *x, const MV *mv) {
  return (mv->row >> 3) < x->mv_row_min ||
         (mv->row >> 3) > x->mv_row_max ||
         (mv->col >> 3) < x->mv_col_min ||
         (mv->col >> 3) > x->mv_col_max;
}

static INLINE void mi_buf_shift(MACROBLOCK *x, int i) {
  MB_MODE_INFO *const mbmi = &x->e_mbd.mi[0]->mbmi;
  struct macroblock_plane *const p = &x->plane[0];
  struct macroblockd_plane *const pd = &x->e_mbd.plane[0];

  p->src.buf = &p->src.buf[vp10_raster_block_offset(BLOCK_8X8, i,
                                                   p->src.stride)];
  assert(((intptr_t)pd->pre[0].buf & 0x7) == 0);
  pd->pre[0].buf = &pd->pre[0].buf[vp10_raster_block_offset(BLOCK_8X8, i,
                                                           pd->pre[0].stride)];
  if (has_second_ref(mbmi))
    pd->pre[1].buf = &pd->pre[1].buf[vp10_raster_block_offset(BLOCK_8X8, i,
                                                           pd->pre[1].stride)];
}

static INLINE void mi_buf_restore(MACROBLOCK *x, struct buf_2d orig_src,
                                  struct buf_2d orig_pre[2]) {
  MB_MODE_INFO *mbmi = &x->e_mbd.mi[0]->mbmi;
  x->plane[0].src = orig_src;
  x->e_mbd.plane[0].pre[0] = orig_pre[0];
  if (has_second_ref(mbmi))
    x->e_mbd.plane[0].pre[1] = orig_pre[1];
}

static INLINE int mv_has_subpel(const MV *mv) {
  return (mv->row & 0x0F) || (mv->col & 0x0F);
}

// Check if NEARESTMV/NEARMV/ZEROMV is the cheapest way encode zero motion.
// TODO(aconverse): Find out if this is still productive then clean up or remove
static int check_best_zero_mv(
Yaowu Xu's avatar
Yaowu Xu committed
1479
    const VP10_COMP *cpi, const uint8_t mode_context[MAX_REF_FRAMES],
Jingning Han's avatar
Jingning Han committed
1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512
    int_mv frame_mv[MB_MODE_COUNT][MAX_REF_FRAMES], int this_mode,
    const MV_REFERENCE_FRAME ref_frames[2]) {
  if ((this_mode == NEARMV || this_mode == NEARESTMV || this_mode == ZEROMV) &&
      frame_mv[this_mode][ref_frames[0]].as_int == 0 &&
      (ref_frames[1] == NONE ||
       frame_mv[this_mode][ref_frames[1]].as_int == 0)) {
    int rfc = mode_context[ref_frames[0]];
    int c1 = cost_mv_ref(cpi, NEARMV, rfc);
    int c2 = cost_mv_ref(cpi, NEARESTMV, rfc);
    int c3 = cost_mv_ref(cpi, ZEROMV, rfc);

    if (this_mode == NEARMV) {
      if (c1 > c3) return 0;
    } else if (this_mode == NEARESTMV) {
      if (c2 > c3) return 0;
    } else {
      assert(this_mode == ZEROMV);
      if (ref_frames[1] == NONE) {
        if ((c3 >= c2 && frame_mv[NEARESTMV][ref_frames[0]].as_int == 0) ||
            (c3 >= c1 && frame_mv[NEARMV][ref_frames[0]].as_int == 0))
          return 0;
      } else {
        if ((c3 >= c2 && frame_mv[NEARESTMV][ref_frames[0]].as_int == 0 &&
             frame_mv[NEARESTMV][ref_frames[1]].as_int == 0) ||
            (c3 >= c1 && frame_mv[NEARMV][ref_frames[0]].as_int == 0 &&
             frame_mv[NEARMV][ref_frames[1]].as_int == 0))
          return 0;
      }
    }
  }
  return 1;
}

Yaowu Xu's avatar
Yaowu Xu committed
1513
static void joint_motion_search(VP10_COMP *cpi, MACROBLOCK *x,
Jingning Han's avatar
Jingning Han committed
1514 1515 1516 1517 1518
                                BLOCK_SIZE bsize,
                                int_mv *frame_mv,
                                int mi_row, int mi_col,
                                int_mv single_newmv[MAX_REF_FRAMES],
                                int *rate_mv) {
Yaowu Xu's avatar
Yaowu Xu committed
1519
  const VP10_COMMON *const cm = &cpi->common;
Jingning Han's avatar
Jingning Han committed
1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606 1607 1608 1609 1610 1611 1612 1613 1614 1615 1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661 1662 1663 1664 1665 1666 1667 1668 1669 1670 1671 1672 1673 1674 1675 1676 1677 1678 1679 1680 1681 1682 1683 1684 1685 1686 1687 1688 1689 1690 1691 1692 1693 1694 1695 1696
  const int pw = 4 * num_4x4_blocks_wide_lookup[bsize];
  const int ph = 4 * num_4x4_blocks_high_lookup[bsize];
  MACROBLOCKD *xd = &x->e_mbd;
  MB_MODE_INFO *mbmi = &xd->mi[0]->mbmi;
  const int refs[2] = {mbmi->ref_frame[0],
                       mbmi->ref_frame[1] < 0 ? 0 : mbmi->ref_frame[1]};
  int_mv ref_mv[2];
  int ite, ref;
  const InterpKernel *kernel = vp10_filter_kernels[mbmi->interp_filter];
  struct scale_factors sf;

  // Do joint motion search in compound mode to get more accurate mv.
  struct buf_2d backup_yv12[2][MAX_MB_PLANE];
  int last_besterr[2] = {INT_MAX, INT_MAX};
  const YV12_BUFFER_CONFIG *const scaled_ref_frame[2] = {
    vp10_get_scaled_ref_frame(cpi, mbmi->ref_frame[0]),
    vp10_get_scaled_ref_frame(cpi, mbmi->ref_frame[1])
  };

  // Prediction buffer from second frame.
#if CONFIG_VP9_HIGHBITDEPTH
  DECLARE_ALIGNED(16, uint16_t, second_pred_alloc_16[64 * 64]);
  uint8_t *second_pred;
#else
  DECLARE_ALIGNED(16, uint8_t, second_pred[64 * 64]);
#endif  // CONFIG_VP9_HIGHBITDEPTH

  for (ref = 0; ref < 2; ++ref) {
    ref_mv[ref] = x->mbmi_ext->ref_mvs[refs[ref]][0];

    if (scaled_ref_frame[ref]) {
      int i;
      // Swap out the reference frame for a version that's been scaled to
      // match the resolution of the current frame, allowing the existing
      // motion search code to be used without additional modifications.
      for (i = 0; i < MAX_MB_PLANE; i++)
        backup_yv12[ref][i] = xd->plane[i].pre[ref];
      vp10_setup_pre_planes(xd, ref, scaled_ref_frame[ref], mi_row, mi_col,
                           NULL);
    }

    frame_mv[refs[ref]].as_int = single_newmv[refs[ref]].as_int;
  }

  // Since we have scaled the reference frames to match the size of the current
  // frame we must use a unit scaling factor during mode selection.
#if CONFIG_VP9_HIGHBITDEPTH
  vp10_setup_scale_factors_for_frame(&sf, cm->width, cm->height,
                                    cm->width, cm->height,
                                    cm->use_highbitdepth);
#else
  vp10_setup_scale_factors_for_frame(&sf, cm->width, cm->height,
                                    cm->width, cm->height);
#endif  // CONFIG_VP9_HIGHBITDEPTH

  // Allow joint search multiple times iteratively for each reference frame
  // and break out of the search loop if it couldn't find a better mv.
  for (ite = 0; ite < 4; ite++) {
    struct buf_2d ref_yv12[2];
    int bestsme = INT_MAX;
    int sadpb = x->sadperbit16;
    MV tmp_mv;
    int search_range = 3;

    int tmp_col_min = x->mv_col_min;
    int tmp_col_max = x->mv_col_max;
    int tmp_row_min = x->mv_row_min;
    int tmp_row_max = x->mv_row_max;
    int id = ite % 2;  // Even iterations search in the first reference frame,
                       // odd iterations search in the second. The predictor
                       // found for the 'other' reference frame is factored in.

    // Initialized here because of compiler problem in Visual Studio.
    ref_yv12[0] = xd->plane[0].pre[0];
    ref_yv12[1] = xd->plane[0].pre[1];

    // Get the prediction block from the 'other' reference frame.
#if CONFIG_VP9_HIGHBITDEPTH
    if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
      second_pred = CONVERT_TO_BYTEPTR(second_pred_alloc_16);
      vp10_highbd_build_inter_predictor(ref_yv12[!id].buf,
                                       ref_yv12[!id].stride,
                                       second_pred, pw,
                                       &frame_mv[refs[!id]].as_mv,
                                       &sf, pw, ph, 0,
                                       kernel, MV_PRECISION_Q3,
                                       mi_col * MI_SIZE, mi_row * MI_SIZE,
                                       xd->bd);
    } else {
      second_pred = (uint8_t *)second_pred_alloc_16;
      vp10_build_inter_predictor(ref_yv12[!id].buf,
                                ref_yv12[!id].stride,
                                second_pred, pw,
                                &frame_mv[refs[!id]].as_mv,
                                &sf, pw, ph, 0,
                                kernel, MV_PRECISION_Q3,
                                mi_col * MI_SIZE, mi_row * MI_SIZE);
    }
#else
    vp10_build_inter_predictor(ref_yv12[!id].buf,
                              ref_yv12[!id].stride,
                              second_pred, pw,
                              &frame_mv[refs[!id]].as_mv,
                              &sf, pw, ph, 0,
                              kernel, MV_PRECISION_Q3,
                              mi_col * MI_SIZE, mi_row * MI_SIZE);
#endif  // CONFIG_VP9_HIGHBITDEPTH

    // Do compound motion search on the current reference frame.
    if (id)
      xd->plane[0].pre[0] = ref_yv12[id];
    vp10_set_mv_search_range(x, &ref_mv[id].as_mv);

    // Use the mv result from the single mode as mv predictor.
    tmp_mv = frame_mv[refs[id]].as_mv;

    tmp_mv.col >>= 3;
    tmp_mv.row >>= 3;

    // Small-range full-pixel motion search.
    bestsme = vp10_refining_search_8p_c(x, &tmp_mv, sadpb,
                                       search_range,
                                       &cpi->fn_ptr[bsize],
                                       &ref_mv[id].as_mv, second_pred);
    if (bestsme < INT_MAX)
      bestsme = vp10_get_mvpred_av_var(x, &tmp_mv, &ref_mv[id].as_mv,
                                      second_pred, &cpi->fn_ptr[bsize], 1);

    x->mv_col_min = tmp_col_min;
    x->mv_col_max = tmp_col_max;
    x->mv_row_min = tmp_row_min;
    x->mv_row_max = tmp_row_max;

    if (bestsme < INT_MAX) {
      int dis; /* TODO: use dis in distortion calculation later. */
      unsigned int sse;
      bestsme = cpi->find_fractional_mv_step(
          x, &tmp_mv,
          &ref_mv[id].as_mv,
          cpi->common.allow_high_precision_mv,
          x->errorperbit,
          &cpi->fn_ptr[bsize],
          0, cpi->sf.mv.subpel_iters_per_step,
          NULL,
          x->nmvjointcost, x->mvcost,
          &dis, &sse, second_pred,
          pw, ph);
    }

    // Restore the pointer to the first (possibly scaled) prediction buffer.
    if (id)
      xd->plane[0].pre[0] = ref_yv12[0];

    if (bestsme < last_besterr[id]) {
      frame_mv[refs[id]].as_mv = tmp_mv;
      last_besterr[id] = bestsme;
    } else {
      break;
    }
  }

  *rate_mv = 0;

  for (ref = 0; ref < 2; ++ref) {
    if (scaled_ref_frame[ref]) {
      // Restore the prediction frame pointers to their unscaled versions.
      int i;
      for (i = 0; i < MAX_MB_PLANE; i++)
        xd->plane[i].pre[ref] = backup_yv12[ref][i];
    }

    *rate_mv += vp10_mv_bit_cost(&frame_mv[refs[ref]].as_mv,
                                &x->mbmi_ext->ref_mvs[refs[ref]][0].as_mv,
                                x->nmvjointcost, x->mvcost, MV_COST_WEIGHT);
  }
}

Yaowu Xu's avatar
Yaowu Xu committed
1697
static int64_t rd_pick_best_sub8x8_mode(VP10_COMP *cpi, MACROBLOCK *x,
Jingning Han's avatar
Jingning Han committed
1698 1699 1700 1701 1702 1703 1704 1705 1706 1707 1708 1709 1710 1711 1712 1713 1714 1715 1716
                                        int_mv *best_ref_mv,
                                        int_mv *second_best_ref_mv,
                                        int64_t best_rd, int *returntotrate,
                                        int *returnyrate,
                                        int64_t *returndistortion,
                                        int *skippable, int64_t *psse,
                                        int mvthresh,
                                        int_mv seg_mvs[4][MAX_REF_FRAMES],
                                        BEST_SEG_INFO *bsi_buf, int filter_idx,
                                        int mi_row, int mi_col) {
  int i;
  BEST_SEG_INFO *bsi = bsi_buf + filter_idx;
  MACROBLOCKD *xd = &x->e_mbd;
  MODE_INFO *mi = xd->mi[0];
  MB_MODE_INFO *mbmi = &mi->mbmi;
  int mode_idx;
  int k, br = 0, idx, idy;
  int64_t bd = 0, block_sse = 0;
  PREDICTION_MODE this_mode;
Yaowu Xu's avatar
Yaowu Xu committed
1717
  VP10_COMMON *cm = &cpi->common;
Jingning Han's avatar
Jingning Han committed
1718 1719 1720 1721 1722 1723 1724 1725 1726 1727 1728 1729 1730 1731 1732 1733 1734 1735 1736 1737 1738 1739 1740 1741 1742 1743 1744 1745 1746 1747 1748 1749 1750 1751 1752 1753 1754 1755 1756 1757 1758 1759 1760 1761 1762 1763 1764 1765 1766 1767 1768 1769 1770 1771 1772 1773 1774 1775 1776 1777 1778 1779 1780 1781 1782 1783 1784 1785 1786 1787 1788 1789 1790 1791 1792 1793 1794 1795 1796 1797 1798 1799 1800 1801 1802 1803 1804 1805 1806 1807 1808 1809 1810 1811 1812 1813 1814 1815 1816 1817 1818 1819 1820 1821 1822 1823 1824 1825 1826 1827 1828 1829 1830 1831 1832 1833 1834 1835 1836 1837 1838 1839 1840 1841 1842 1843 1844 1845 1846 1847 1848 1849 1850 1851 1852 1853 1854 1855 1856 1857 1858 1859 1860 1861 1862 1863 1864 1865 1866 1867 1868 1869 1870 1871 1872 1873 1874 1875 1876 1877 1878 1879 1880 1881 1882 1883 1884 1885 1886 1887 1888 1889 1890 1891 1892 1893 1894 1895 1896 1897 1898 1899 1900 1901 1902 1903 1904 1905 1906 1907 1908 1909 1910 1911 1912 1913 1914 1915 1916 1917 1918 1919 1920 1921 1922 1923 1924 1925 1926 1927 1928 1929 1930 1931 1932 1933 1934 1935 1936 1937 1938 1939 1940