rdopt.c 193 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
/*
 *  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"

17
#include "vpx_dsp/vpx_dsp_common.h"
18 19
#include "vpx_mem/vpx_mem.h"
#include "vpx_ports/mem.h"
20
#include "vpx_ports/system_state.h"
21

22 23 24 25 26 27 28 29 30 31 32
#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"
33

34 35 36 37 38
#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"
hui su's avatar
hui su committed
39
#include "vp10/encoder/palette.h"
40 41 42 43 44
#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"
45 46 47 48 49 50 51 52 53 54 55 56 57

#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

58
#if CONFIG_EXT_TX
59
const double ext_tx_th = 0.98;
60 61
#endif

62 63 64 65 66 67 68 69 70 71
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 {
72 73 74
#if CONFIG_VAR_TX
  const VP10_COMP *cpi;
#endif
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
  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}},
};

hui su's avatar
hui su committed
141 142 143 144 145 146 147 148 149 150
static INLINE int write_uniform_cost(int n, int v) {
  int l = get_unsigned_bits(n), m = (1 << l) - n;
  if (l == 0)
    return 0;
  if (v < m)
    return (l - 1) * vp10_cost_bit(128, 0);
  else
    return l * vp10_cost_bit(128, 0);
}

151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175
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
176
static void model_rd_for_sb(VP10_COMP *cpi, BLOCK_SIZE bsize,
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
                            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.
214 215
    const int64_t low_dc_thr = VPXMIN(50, dc_thr >> 2);
    const int64_t low_ac_thr = VPXMIN(80, ac_thr >> 2);
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
    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,
355 356 357
#if CONFIG_VAR_TX
                       int coeff_ctx,
#else
358
                       ENTROPY_CONTEXT *A, ENTROPY_CONTEXT *L,
359
#endif
360 361 362 363 364 365 366 367 368 369 370 371 372 373
                       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];
374 375 376
#if CONFIG_VAR_TX
  int pt = coeff_ctx;
#else
377
  int pt = combine_entropy_contexts(*A, *L);
378
#endif
379 380 381 382 383 384 385
  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

386
#if !CONFIG_VAR_TX
387 388 389
  // 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);
390
#endif
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

  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];
      }
    }
  }

444
#if !CONFIG_VAR_TX
445 446
  // is eob first coefficient;
  *A = *L = (c > 0);
447
#endif
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

  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;
}

473
static int rate_block(int plane, int block, int blk_row, int blk_col,
474
                      TX_SIZE tx_size, struct rdcost_block_args* args) {
475 476 477 478 479 480 481 482 483 484 485 486 487 488 489
#if CONFIG_VAR_TX
  int coeff_ctx = combine_entropy_contexts(*(args->t_above + blk_col),
                                           *(args->t_left + blk_row));
  int coeff_cost = cost_coeffs(args->x, plane, block, coeff_ctx,
                               tx_size, args->so->scan, args->so->neighbors,
                               args->use_fast_coef_costing);
  const struct macroblock_plane *p = &args->x->plane[plane];
  *(args->t_above + blk_col) = !(p->eobs[block] == 0);
  *(args->t_left  + blk_row) = !(p->eobs[block] == 0);
  return coeff_cost;
#else
  return cost_coeffs(args->x, plane, block,
                     args->t_above + blk_col,
                     args->t_left + blk_row,
                     tx_size, args->so->scan, args->so->neighbors,
490
                     args->use_fast_coef_costing);
491
#endif
492 493
}

494 495
static void block_rd_txfm(int plane, int block, int blk_row, int blk_col,
                          BLOCK_SIZE plane_bsize,
496 497 498 499 500 501 502 503 504 505 506 507 508 509
                          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)) {
510 511 512 513 514 515 516
#if CONFIG_VAR_TX
    struct encode_b_args arg = {x, NULL, &mbmi->skip};
    uint8_t *dst, *src;
    int src_stride = x->plane[plane].src.stride;
    int dst_stride = xd->plane[plane].dst.stride;
    unsigned int tmp_sse;
    PREDICTION_MODE mode = (plane == 0) ?
517
        get_y_mode(xd->mi[0], block) : mbmi->uv_mode;
518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539

#if CONFIG_VP9_HIGHBITDEPTH
    vp10_encode_block_intra(plane, block, blk_row, blk_col,
                            plane_bsize, tx_size, &arg);
    dist_block(x, plane, block, tx_size, &dist, &sse);
#else
    src = &x->plane[plane].src.buf[4 * (blk_row * src_stride + blk_col)];
    dst = &xd->plane[plane].dst.buf[4 * (blk_row * dst_stride + blk_col)];
    vp10_predict_intra_block(xd, b_width_log2_lookup[plane_bsize],
                             b_height_log2_lookup[plane_bsize],
                             tx_size, mode, dst, dst_stride,
                             dst, dst_stride, blk_col, blk_row, plane);
    args->cpi->fn_ptr[txsize_to_bsize[tx_size]].vf(src, src_stride,
                                                   dst, dst_stride, &tmp_sse);
    sse = (int64_t)tmp_sse * 16;
    vp10_encode_block_intra(plane, block, blk_row, blk_col,
                            plane_bsize, tx_size, &arg);
    args->cpi->fn_ptr[txsize_to_bsize[tx_size]].vf(src, src_stride,
                                                   dst, dst_stride, &tmp_sse);
    dist = (int64_t)tmp_sse * 16;
#endif  // CONFIG_VP9_HIGHBITDEPTH
#else
540
    struct encode_b_args arg = {x, NULL, &mbmi->skip};
541 542
    vp10_encode_block_intra(plane, block, blk_row, blk_col,
                            plane_bsize, tx_size, &arg);
543
    dist_block(x, plane, block, tx_size, &dist, &sse);
544
#endif
545 546 547 548
  } 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
549 550
      vp10_xform_quant(x, plane, block, blk_row, blk_col,
                       plane_bsize, tx_size);
551 552 553 554 555 556
      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);
557 558
      vp10_xform_quant_dc(x, plane, block, blk_row, blk_col,
                          plane_bsize, tx_size);
559 560 561 562 563 564 565 566 567 568 569 570
      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;

571
        dist = VPXMAX(0, sse - dc_correct);
572 573 574 575 576 577 578 579 580 581
      }
    } 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
582
    vp10_xform_quant(x, plane, block, blk_row, blk_col, plane_bsize, tx_size);
583 584 585 586 587 588 589 590 591
    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;
  }

592
  rate = rate_block(plane, block, blk_row, blk_col, tx_size, args);
593 594 595 596
  rd1 = RDCOST(x->rdmult, x->rddiv, rate, dist);
  rd2 = RDCOST(x->rdmult, x->rddiv, 0, sse);

  // TODO(jingning): temporarily enabled only for luma component
597
  rd = VPXMIN(rd1, rd2);
598 599
  if (plane == 0)
    x->zcoeff_blk[tx_size][block] = !x->plane[plane].eobs[block] ||
600
        (rd1 > rd2 && !xd->lossless[mbmi->segment_id]);
601 602 603 604 605 606 607 608 609 610 611 612 613 614 615

  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,
616 617 618
#if CONFIG_VAR_TX
                             const VP10_COMP *cpi,
#endif
619 620 621 622 623 624 625
                             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];
626
  TX_TYPE tx_type;
627 628 629
  struct rdcost_block_args args;
  vp10_zero(args);
  args.x = x;
630 631 632
#if CONFIG_VAR_TX
  args.cpi = cpi;
#endif
633 634 635 636 637 638 639 640 641
  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
642
  tx_type = get_tx_type(pd->plane_type, xd, 0, tx_size);
643
  args.so = get_scan(tx_size, tx_type, is_inter_block(&xd->mi[0]->mbmi));
644 645

  vp10_foreach_transformed_block_in_plane(xd, bsize, plane,
646
                                          block_rd_txfm, &args);
647 648 649 650 651 652 653 654 655 656 657 658 659
  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
660
static void choose_largest_tx_size(VP10_COMP *cpi, MACROBLOCK *x,
661 662 663 664 665
                                   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
666
  VP10_COMMON *const cm = &cpi->common;
667 668 669
  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;
670
#if CONFIG_EXT_TX
671
  TX_TYPE tx_type, best_tx_type = DCT_DCT;
672 673 674 675 676
  int r, s;
  int64_t d, psse, this_rd, best_rd = INT64_MAX;
  vpx_prob skip_prob = vp10_get_skip_prob(cm, xd);
  int  s0 = vp10_cost_bit(skip_prob, 0);
  int  s1 = vp10_cost_bit(skip_prob, 1);
677 678
  int ext_tx_set;
  const int is_inter = is_inter_block(mbmi);
679
#endif  // CONFIG_EXT_TX
680

681
  mbmi->tx_size = VPXMIN(max_tx_size, largest_tx_size);
682

683
#if CONFIG_EXT_TX
684 685 686 687
  ext_tx_set = get_ext_tx_set(mbmi->tx_size, bs, is_inter);

  if (is_inter &&
      get_ext_tx_types(mbmi->tx_size, bs, is_inter) > 1 &&
Yaowu Xu's avatar
Yaowu Xu committed
688
      !xd->lossless[mbmi->segment_id]) {
689 690 691 692 693 694 695 696 697 698 699 700
    for (tx_type = 0; tx_type < TX_TYPES; ++tx_type) {
      if (is_inter) {
        if (!ext_tx_used_inter[ext_tx_set][tx_type])
          continue;
      } else {
        if (!ext_tx_used_intra[ext_tx_set][tx_type])
          continue;
      }

      mbmi->tx_type = tx_type;
      if (ext_tx_set == 1 &&
          mbmi->tx_type >= DST_ADST && mbmi->tx_type < IDTX &&
hui su's avatar
hui su committed
701 702
          best_tx_type == DCT_DCT) {
        tx_type = IDTX - 1;
703 704
        continue;
      }
705

706 707 708 709 710
      txfm_rd_in_plane(x,
#if CONFIG_VAR_TX
                       cpi,
#endif
                       &r, &d, &s,
711 712 713 714 715
                       &psse, ref_best_rd, 0, bs, mbmi->tx_size,
                       cpi->sf.use_fast_coef_costing);

      if (r == INT_MAX)
        continue;
716 717 718 719 720 721 722 723 724 725
      if (get_ext_tx_types(mbmi->tx_size, bs, is_inter) > 1) {
        if (is_inter) {
          if (ext_tx_set > 0)
            r += cpi->inter_tx_type_costs[ext_tx_set]
                                         [mbmi->tx_size][mbmi->tx_type];
        } else {
          if (ext_tx_set > 0)
            r += cpi->intra_tx_type_costs[ext_tx_set][mbmi->tx_size]
                                         [mbmi->mode][mbmi->tx_type];
        }
726
      }
727 728 729 730 731

      if (s)
        this_rd = RDCOST(x->rdmult, x->rddiv, s1, psse);
      else
        this_rd = RDCOST(x->rdmult, x->rddiv, r + s0, d);
Yaowu Xu's avatar
Yaowu Xu committed
732
      if (is_inter_block(mbmi) && !xd->lossless[mbmi->segment_id] && !s)
733 734
        this_rd = VPXMIN(this_rd, RDCOST(x->rdmult, x->rddiv, s1, psse));

hui su's avatar
hui su committed
735
      if (this_rd < ((best_tx_type == DCT_DCT) ? ext_tx_th : 1) * best_rd) {
736
        best_rd = this_rd;
hui su's avatar
hui su committed
737
        best_tx_type = mbmi->tx_type;
738 739
      }
    }
740
  }
741

hui su's avatar
hui su committed
742
  mbmi->tx_type = best_tx_type;
743
#endif  // CONFIG_EXT_TX
744

745 746 747 748 749
  txfm_rd_in_plane(x,
#if CONFIG_VAR_TX
                   cpi,
#endif
                   rate, distortion, skip,
750 751
                   sse, ref_best_rd, 0, bs,
                   mbmi->tx_size, cpi->sf.use_fast_coef_costing);
752 753

#if CONFIG_EXT_TX
754
  if (get_ext_tx_types(mbmi->tx_size, bs, is_inter) > 1 &&
Yaowu Xu's avatar
Yaowu Xu committed
755
      !xd->lossless[mbmi->segment_id] && *rate != INT_MAX) {
756 757 758 759
    int ext_tx_set = get_ext_tx_set(mbmi->tx_size, bs, is_inter);
    if (is_inter)
      *rate += cpi->inter_tx_type_costs[ext_tx_set][mbmi->tx_size]
                                       [mbmi->tx_type];
760
    else
761 762
      *rate += cpi->intra_tx_type_costs[ext_tx_set][mbmi->tx_size]
                                       [mbmi->mode][mbmi->tx_type];
763
  }
764
#endif  // CONFIG_EXT_TX
765 766
}

767 768 769 770 771 772 773 774 775 776
static void choose_smallest_tx_size(VP10_COMP *cpi, MACROBLOCK *x,
                                    int *rate, int64_t *distortion,
                                    int *skip, int64_t *sse,
                                    int64_t ref_best_rd,
                                    BLOCK_SIZE bs) {
  MACROBLOCKD *const xd = &x->e_mbd;
  MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;

  mbmi->tx_size = TX_4X4;

777 778 779 780 781
  txfm_rd_in_plane(x,
#if CONFIG_VAR_TX
                   cpi,
#endif
                   rate, distortion, skip,
782 783 784 785
                   sse, ref_best_rd, 0, bs,
                   mbmi->tx_size, cpi->sf.use_fast_coef_costing);
}

Yaowu Xu's avatar
Yaowu Xu committed
786
static void choose_tx_size_from_rd(VP10_COMP *cpi, MACROBLOCK *x,
787 788 789 790 791 792 793
                                   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
794
  VP10_COMMON *const cm = &cpi->common;
795 796 797
  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);
hui su's avatar
hui su committed
798 799 800
  int r, s;
  int64_t d, sse;
  int64_t rd = INT64_MAX;
801 802
  int n, m;
  int s0, s1;
hui su's avatar
hui su committed
803
  int64_t best_rd = INT64_MAX, last_rd = INT64_MAX;
804 805
  TX_SIZE best_tx = max_tx_size;
  int start_tx, end_tx;
hui su's avatar
hui su committed
806
  const int tx_select = cm->tx_mode == TX_MODE_SELECT;
hui su's avatar
hui su committed
807
#if CONFIG_EXT_TX
808 809
  TX_TYPE tx_type, best_tx_type = DCT_DCT;
  int ext_tx_set;
hui su's avatar
hui su committed
810
#endif  // CONFIG_EXT_TX
811
  const int is_inter = is_inter_block(mbmi);
hui su's avatar
hui su committed
812

813 814 815 816 817
  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);

hui su's avatar
hui su committed
818
  if (tx_select) {
819 820 821
    start_tx = max_tx_size;
    end_tx = 0;
  } else {
822 823
    const TX_SIZE chosen_tx_size =
        VPXMIN(max_tx_size, tx_mode_to_biggest_tx_size[cm->tx_mode]);
824 825 826 827
    start_tx = chosen_tx_size;
    end_tx = chosen_tx_size;
  }

hui su's avatar
hui su committed
828 829 830 831 832
  *distortion = INT64_MAX;
  *rate       = INT_MAX;
  *skip       = 0;
  *psse       = INT64_MAX;

833
#if CONFIG_EXT_TX
834
  for (tx_type = DCT_DCT; tx_type < TX_TYPES; ++tx_type) {
835
#endif  // CONFIG_EXT_TX
836
    last_rd = INT64_MAX;
hui su's avatar
hui su committed
837 838 839 840 841 842 843 844
    for (n = start_tx; n >= end_tx; --n) {
      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]);
      }
hui su's avatar
hui su committed
845

846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861
#if CONFIG_EXT_TX
      ext_tx_set = get_ext_tx_set(n, bs, is_inter);
      if (is_inter) {
        if (!ext_tx_used_inter[ext_tx_set][tx_type])
          continue;
      } else {
        if (!ext_tx_used_intra[ext_tx_set][tx_type])
          continue;
      }
      mbmi->tx_type = tx_type;
      if (ext_tx_set == 1 &&
          mbmi->tx_type >= DST_ADST && mbmi->tx_type < IDTX &&
          best_tx_type == DCT_DCT) {
        tx_type = IDTX - 1;
        break;
      }
862 863 864 865 866
      txfm_rd_in_plane(x,
#if CONFIG_VAR_TX
                       cpi,
#endif
                       &r, &d, &s,
hui su's avatar
hui su committed
867 868
                       &sse, ref_best_rd, 0, bs, n,
                       cpi->sf.use_fast_coef_costing);
869 870 871 872 873 874 875 876 877 878 879 880
      if (get_ext_tx_types(n, bs, is_inter) > 1 &&
          !xd->lossless[xd->mi[0]->mbmi.segment_id] &&
          r != INT_MAX) {
        if (is_inter) {
          if (ext_tx_set > 0)
            r += cpi->inter_tx_type_costs[ext_tx_set]
                                         [mbmi->tx_size][mbmi->tx_type];
        } else {
          if (ext_tx_set > 0)
            r += cpi->intra_tx_type_costs[ext_tx_set][mbmi->tx_size]
                                         [mbmi->mode][mbmi->tx_type];
        }
881
      }
882 883 884 885 886 887 888 889
#else  // CONFIG_EXT_TX
      txfm_rd_in_plane(x,
#if CONFIG_VAR_TX
                       cpi,
#endif
                       &r, &d, &s,
                       &sse, ref_best_rd, 0, bs, n,
                       cpi->sf.use_fast_coef_costing);
hui su's avatar
hui su committed
890
#endif  // CONFIG_EXT_TX
hui su's avatar
hui su committed
891

hui su's avatar
hui su committed
892 893 894 895
      if (r == INT_MAX)
        continue;

      if (s) {
896
        if (is_inter) {
hui su's avatar
hui su committed
897 898 899 900
          rd = RDCOST(x->rdmult, x->rddiv, s1, sse);
        } else {
          rd =  RDCOST(x->rdmult, x->rddiv, s1 + r_tx_size * tx_select, sse);
        }
901
      } else {
902
        rd = RDCOST(x->rdmult, x->rddiv, r + s0 + r_tx_size * tx_select, d);
903
      }
904

905 906 907 908
      if (tx_select && !(s && is_inter))
        r += r_tx_size;

      if (is_inter && !xd->lossless[xd->mi[0]->mbmi.segment_id] && !s)
hui su's avatar
hui su committed
909
        rd = VPXMIN(rd, RDCOST(x->rdmult, x->rddiv, s1, sse));
910

hui su's avatar
hui su committed
911
      // Early termination in transform size search.
912 913 914 915 916 917 918 919
      if (cpi->sf.tx_size_search_breakout &&
          (rd == INT64_MAX ||
#if CONFIG_EXT_TX
           (s == 1 && tx_type != DCT_DCT) ||
#else
           (s == 1) ||
#endif
           (n < (int) max_tx_size && rd > last_rd)))
hui su's avatar
hui su committed
920
        break;
921

hui su's avatar
hui su committed
922
      last_rd = rd;
hui su's avatar
hui su committed
923
      if (rd <
924 925
#if CONFIG_EXT_TX
          (is_inter && best_tx_type == DCT_DCT ? ext_tx_th : 1) *
hui su's avatar
hui su committed
926
#endif  // CONFIG_EXT_TX
927
          best_rd) {
hui su's avatar
hui su committed
928 929 930 931 932 933 934
        best_tx = n;
        best_rd = rd;
        *distortion = d;
        *rate       = r;
        *skip       = s;
        *psse       = sse;
#if CONFIG_EXT_TX
hui su's avatar
hui su committed
935
        best_tx_type = mbmi->tx_type;
hui su's avatar
hui su committed
936 937
#endif  // CONFIG_EXT_TX
      }
938
    }
hui su's avatar
hui su committed
939
#if CONFIG_EXT_TX
940
  }
hui su's avatar
hui su committed
941
#endif  // CONFIG_EXT_TX
942

943
  mbmi->tx_size = best_tx;
hui su's avatar
hui su committed
944
#if CONFIG_EXT_TX
hui su's avatar
hui su committed
945
  mbmi->tx_type = best_tx_type;
946 947 948 949 950
  txfm_rd_in_plane(x,
#if CONFIG_VAR_TX
                   cpi,
#endif
                   &r, &d, &s,
hui su's avatar
hui su committed
951 952 953
                   &sse, ref_best_rd, 0, bs, best_tx,
                   cpi->sf.use_fast_coef_costing);
#endif  // CONFIG_EXT_TX
954 955
}

Yaowu Xu's avatar
Yaowu Xu committed
956
static void super_block_yrd(VP10_COMP *cpi, MACROBLOCK *x, int *rate,
957 958 959 960 961 962 963 964 965
                            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);

966 967 968 969 970
  if (CONFIG_MISC_FIXES && xd->lossless[xd->mi[0]->mbmi.segment_id]) {
    choose_smallest_tx_size(cpi, x, rate, distortion, skip, ret_sse,
                            ref_best_rd, bs);
  } else if (cpi->sf.tx_size_search_method == USE_LARGESTALL ||
             xd->lossless[xd->mi[0]->mbmi.segment_id]) {
971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999
    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;
}

hui su's avatar
hui su committed
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 1041 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
void rd_pick_palette_intra_sby(VP10_COMP *cpi, MACROBLOCK *x, BLOCK_SIZE bsize,
                               int palette_ctx, int dc_mode_cost,
                               PALETTE_MODE_INFO *palette_mode_info,
                               uint8_t *best_palette_color_map,
                               TX_SIZE *best_tx, PREDICTION_MODE *mode_selected,
                               int64_t *best_rd) {
  MACROBLOCKD *const xd = &x->e_mbd;
  MODE_INFO *const mic = xd->mi[0];
  int rows = 4 * num_4x4_blocks_high_lookup[bsize];
  int cols = 4 * num_4x4_blocks_wide_lookup[bsize];
  int this_rate, this_rate_tokenonly, s;
  int64_t this_distortion, this_rd;
  int colors, n;
  int src_stride = x->plane[0].src.stride;
  uint8_t *src = x->plane[0].src.buf;

#if CONFIG_VP9_HIGHBITDEPTH
  if (cpi->common.use_highbitdepth)
    colors = vp10_count_colors_highbd(src, src_stride, rows, cols,
                                      cpi->common.bit_depth);
  else
#endif  // CONFIG_VP9_HIGHBITDEPTH
    colors = vp10_count_colors(src, src_stride, rows, cols);
  palette_mode_info->palette_size[0] = 0;

  if (colors > 1 && colors <= 64 && cpi->common.allow_screen_content_tools) {
    int r, c, i, j, k;
    int max_itr = 50;
    int color_ctx, color_idx = 0;
    int color_order[PALETTE_MAX_SIZE];
    double *data = x->palette_buffer->kmeans_data_buf;
    uint8_t *indices = x->palette_buffer->kmeans_indices_buf;
    uint8_t *pre_indices = x->palette_buffer->kmeans_pre_indices_buf;
    double centroids[PALETTE_MAX_SIZE];
    uint8_t *color_map;
    double lb, ub, val;
    PALETTE_MODE_INFO *pmi = &mic->mbmi.palette_mode_info;
#if CONFIG_VP9_HIGHBITDEPTH
    uint16_t *src16 = CONVERT_TO_SHORTPTR(src);
    if (cpi->common.use_highbitdepth)
      lb = ub = src16[0];
    else
#endif  // CONFIG_VP9_HIGHBITDEPTH
      lb = ub = src[0];

#if CONFIG_VP9_HIGHBITDEPTH
    if (cpi->common.use_highbitdepth) {
      for (r = 0; r < rows; ++r) {
        for (c = 0; c < cols; ++c) {
          val = src16[r * src_stride + c];
          data[r * cols + c] = val;
          if (val < lb)
            lb = val;
          else if (val > ub)
            ub = val;
        }
      }
    } else {
#endif  // CONFIG_VP9_HIGHBITDEPTH
      for (r = 0; r < rows; ++r) {
        for (c = 0; c < cols; ++c) {
          val = src[r * src_stride + c];
          data[r * cols + c] = val;
          if (val < lb)
            lb = val;
          else if (val > ub)
            ub = val;
        }
      }
#if CONFIG_VP9_HIGHBITDEPTH
    }
#endif  // CONFIG_VP9_HIGHBITDEPTH

    mic->mbmi.mode = DC_PRED;

    for (n = colors > PALETTE_MAX_SIZE ? PALETTE_MAX_SIZE : colors;
        n >= 2; --n) {
      for (i = 0; i < n; ++i)
        centroids[i] = lb + (2 * i + 1) * (ub - lb) / n / 2;
      vp10_k_means(data, centroids, indices, pre_indices, rows * cols,
                   n, 1, max_itr);
      vp10_insertion_sort(centroids, n);
1082 1083
      for (i = 0; i < n; ++i)
        centroids[i] = round(centroids[i]);
hui su's avatar
hui su committed
1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107
      // remove duplicates
      i = 1;
      k = n;
      while (i < k) {
        if (centroids[i] == centroids[i - 1]) {
          j = i;
          while (j < k - 1) {
            centroids[j] = centroids[j + 1];
            ++j;
          }
          --k;
        } else {
          ++i;
        }
      }

#if CONFIG_VP9_HIGHBITDEPTH
      if (cpi->common.use_highbitdepth)
        for (i = 0; i < k; ++i)
          mic->mbmi.palette_mode_info.palette_colors[i] =
              clip_pixel_highbd(round(centroids[i]), cpi->common.bit_depth);
      else
#endif  // CONFIG_VP9_HIGHBITDEPTH
        for (i = 0; i < k; ++i)
Yaowu Xu's avatar
Yaowu Xu committed
1108
          pmi->palette_colors[i] = clip_pixel((int)round(centroids[i]));
hui su's avatar
hui su committed
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 1154 1155 1156
      pmi->palette_size[0] = k;

      vp10_calc_indices(data, centroids, indices, rows * cols, k, 1);
      for (r = 0; r < rows; ++r)
        for (c = 0; c < cols; ++c)
          xd->plane[0].color_index_map[r * cols + c] = indices[r * cols + c];

      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 + dc_mode_cost +
          cpi->common.bit_depth * k * vp10_cost_bit(128, 0) +
          cpi->palette_y_size_cost[bsize - BLOCK_8X8][k - 2];
      this_rate +=
          vp10_cost_bit(vp10_default_palette_y_mode_prob[bsize - BLOCK_8X8]
                                                         [palette_ctx], 1);
      color_map = xd->plane[0].color_index_map;
      this_rate +=  write_uniform_cost(k, xd->plane[0].color_index_map[0]);
      for (i = 0; i < rows; ++i) {
        for (j = (i == 0 ? 1 : 0); j < cols; ++j) {
          color_ctx = vp10_get_palette_color_context(color_map, cols, i, j,
                                                     k, color_order);
          for (r = 0; r < k; ++r)
            if (color_map[i * cols + j] == color_order[r]) {
              color_idx = r;
              break;
            }
          assert(color_idx < k);
          this_rate +=
              cpi->palette_y_color_cost[k - 2][color_ctx][color_idx];
        }
      }
      this_rd = RDCOST(x->rdmult, x->rddiv, this_rate, this_distortion);

      if (this_rd < *best_rd) {
        *best_rd = this_rd;
        *palette_mode_info = mic->mbmi.palette_mode_info;
        memcpy(best_palette_color_map, xd->plane[0].color_index_map,
               rows * cols * sizeof(xd->plane[0].color_index_map[0]));
        *mode_selected = DC_PRED;
        *best_tx = mic->mbmi.tx_size;
      }
    }
  }
}

Yaowu Xu's avatar
Yaowu Xu committed
1157
static int64_t rd_pick_intra4x4block(VP10_COMP *cpi, MACROBLOCK *x,
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
                                     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;
hui su's avatar
hui su committed
1187
  xd->mi[0]->mbmi.palette_mode_info.palette_size[0] = 0;
1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203

#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))
1204
          continue;
1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215
      }

      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,
1216 1217
                                                                   block,
                                                                   p->src_diff);
1218 1219
          tran_low_t *const coeff = BLOCK_OFFSET(x->plane[0].coeff, block);
          xd->mi[0]->bmi[block].as_mode = mode;
1220
          vp10_predict_intra_block(xd, 1, 1, TX_4X4, mode, dst, dst_stride,
1221 1222 1223 1224
                                  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);
1225
          if (xd->lossless[xd->mi[0]->mbmi.segment_id]) {
hui su's avatar
hui su committed
1226
            TX_TYPE tx_type = get_tx_type(PLANE_TYPE_Y, xd, block, TX_4X4);
1227
            const scan_order *so = get_scan(TX_4X4, tx_type, 0);
1228 1229 1230 1231
#if CONFIG_VAR_TX
            const int coeff_ctx = combine_entropy_contexts(*(tempa + idx),
                                                           *(templ + idy));
#endif
1232
            vp10_highbd_fwd_txfm_4x4(src_diff, coeff, 8, DCT_DCT, 1);
1233
            vp10_regular_quantize_b_4x4(x, 0, block, so->scan, so->iscan);
1234 1235 1236 1237 1238 1239 1240
            ratey += cost_coeffs(x, 0, block,
#if CONFIG_VAR_TX
                                 coeff_ctx,
#else
                                 tempa + idx, templ + idy,
#endif
                                 TX_4X4,
1241 1242 1243 1244
                                 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
1245 1246
            vp10_highbd_inv_txfm_add_4x4(BLOCK_OFFSET(pd->dqcoeff, block),
                                         dst, dst_stride, p->eobs[block],
1247
                                         xd->bd, DCT_DCT, 1);
1248 1249
          } else {
            int64_t unused;
hui su's avatar
hui su committed
1250
            TX_TYPE tx_type = get_tx_type(PLANE_TYPE_Y, xd, block, TX_4X4);
1251
            const scan_order *so = get_scan(TX_4X4, tx_type, 0);
1252 1253 1254 1255
#if CONFIG_VAR_TX
            const int coeff_ctx = combine_entropy_contexts(*(tempa + idx),
                                                           *(templ + idy));
#endif
1256
            vp10_highbd_fwd_txfm_4x4(src_diff, coeff, 8, tx_type, 0);
1257
            vp10_regular_quantize_b_4x4(x, 0, block, so->scan, so->iscan);
1258 1259 1260 1261 1262 1263 1264
            ratey += cost_coeffs(x, 0, block,
#if CONFIG_VAR_TX
                                 coeff_ctx,
#else
                                 tempa + idx, templ + idy,
#endif
                                 TX_4X4,
1265 1266 1267 1268 1269 1270 1271
                                 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
1272 1273
            vp10_highbd_inv_txfm_add_4x4(BLOCK_OFFSET(pd->dqcoeff, block),
                                         dst, dst_stride, p->eobs[block],
1274
                                         xd->bd, tx_type, 0);
1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295
          }
        }
      }

      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));
        }
      }
1296
next_highbd:
1297 1298
      {}
    }
1299
    if (best_rd >= rd_thresh)
1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324
      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))
1325
        continue;
1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339
    }

    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;
1340
        vp10_predict_intra_block(xd, 1, 1, TX_4X4, mode, dst, dst_stride,
1341 1342 1343
                                dst, dst_stride, col + idx, row + idy, 0);
        vpx_subtract_block(4, 4, src_diff, 8, src, src_stride, dst, dst_stride);

1344
        if (xd->lossless[xd->mi[0]->mbmi.segment_id]) {
hui su's avatar
hui su committed
1345
          TX_TYPE tx_type = get_tx_type(PLANE_TYPE_Y, xd, block, TX_4X4);
1346
          const scan_order *so = get_scan(TX_4X4, tx_type, 0);
1347 1348 1349 1350
#if CONFIG_VAR_TX
          int coeff_ctx = combine_entropy_contexts(*(tempa + idx),
                                                   *(templ + idy));
#endif
1351
          vp10_fwd_txfm_4x4(src_diff, coeff, 8, DCT_DCT, 1);
1352
          vp10_regular_quantize_b_4x4(x, 0, block, so->scan, so->iscan);
1353 1354 1355 1356 1357 1358 1359 1360
#if CONFIG_VAR_TX
          ratey += cost_coeffs(x, 0, block, coeff_ctx, TX_4X4, so->scan,
                               so->neighbors, cpi->sf.use_fast_coef_costing);
          *(tempa + idx) = !(p->eobs[block] == 0);
          *(templ + idy) = !(p->eobs[block] == 0);
#else
          ratey += cost_coeffs(x, 0, block, tempa + idx, templ + idy,
                               TX_4X4,
1361 1362
                               so->scan, so->neighbors,
                               cpi->sf.use_fast_coef_costing);
1363
#endif
1364 1365
          if (RDCOST(x->rdmult, x->rddiv, ratey, distortion) >= best_rd)
            goto next;
hui su's avatar
hui su committed
1366
          vp10_inv_txfm_add_4x4(BLOCK_OFFSET(pd->dqcoeff, block),
1367
                                dst, dst_stride, p->eobs[block], DCT_DCT, 1);
1368 1369
        } else {
          int64_t unused;
hui su's avatar
hui su committed
1370
          TX_TYPE tx_type = get_tx_type(PLANE_TYPE_Y, xd, block, TX_4X4);
1371
          const scan_order *so = get_scan(TX_4X4, tx_type, 0);
1372 1373 1374 1375
#if CONFIG_VAR_TX
          int coeff_ctx = combine_entropy_contexts(*(tempa + idx),
                                                   *(templ + idy));
#endif
1376
          vp10_fwd_txfm_4x4(src_diff, coeff, 8, tx_type, 0);
1377
          vp10_regular_quantize_b_4x4(x, 0, block, so->scan, so->iscan);
1378 1379 1380 1381 1382 1383 1384 1385 1386 1387
#if CONFIG_VAR_TX
          ratey += cost_coeffs(x, 0, block, coeff_ctx, TX_4X4, so->scan,
                               so->neighbors, cpi->sf.use_fast_coef_costing);
          *(tempa + idx) = !(p->eobs[block] == 0);
          *(templ + idy) = !(p->eobs[block] == 0);
#else
          ratey += cost_coeffs(x, 0, block, tempa + idx, templ + idy,
                               TX_4X4, so->scan, so->neighbors,
                               cpi->sf.use_fast_coef_costing);
#endif
1388 1389 1390 1391
          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
1392
          vp10_inv_txfm_add_4x4(BLOCK_OFFSET(pd->dqcoeff, block),
1393
                                dst, dst_stride, p->eobs[block], tx_type, 0);
1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416
        }
      }
    }

    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:
    {}
  }

1417
  if (best_rd >= rd_thresh)
1418 1419 1420 1421 1422 1423 1424 1425 1426
    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
1427
static int64_t rd_pick_intra_sub_8x8_y_mode(VP10_COMP *cpi, MACROBLOCK *mb,
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 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494
                                            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
1495
static int64_t rd_pick_intra_sby_mode(VP10_COMP *cpi, MACROBLOCK *x,
1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506
                                      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;
1507
#if CONFIG_EXT_TX
hui su's avatar
hui su committed
1508
  TX_TYPE best_tx_type = DCT_DCT;
1509
#endif  // CONFIG_EXT_TX
1510
  int *bmode_costs;
hui su's avatar
hui su committed
1511
  PALETTE_MODE_INFO palette_mode_info;
1512 1513
  uint8_t *best_palette_color_map = cpi->common.allow_screen_content_tools ?
      x->palette_buffer->best_palette_color_map : NULL;
hui su's avatar
hui su committed
1514 1515 1516
  int rows = 4 * num_4x4_blocks_high_lookup[bsize];
  int cols = 4 * num_4x4_blocks_wide_lookup[bsize];
  int palette_ctx = 0;
1517 1518 1519 1520 1521 1522 1523
  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));
hui su's avatar
hui su committed
1524 1525 1526 1527 1528 1529 1530
  palette_mode_info.palette_size[0] = 0;
  mic->mbmi.palette_mode_info.palette_size[0] = 0;
  if (above_mi)
    palette_ctx += (above_mi->mbmi.palette_mode_info.palette_size[0] > 0);
  if (left_mi)
    palette_ctx += (left_mi->mbmi.palette_mode_info.palette_size[0] > 0);

1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541
  /* Y Search for intra prediction mode */
  for (mode = DC_PRED; mode <= TM_PRED; mode++) {
    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];
hui su's avatar
hui su committed
1542 1543 1544 1545
    if (cpi->common.allow_screen_content_tools && mode == DC_PRED)
      this_rate +=
          vp10_cost_bit(vp10_default_palette_y_mode_prob[bsize - BLOCK_8X8]
                                                         [palette_ctx], 0);
1546 1547 1548 1549 1550 1551
    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;
1552
#if CONFIG_EXT_TX
hui su's avatar
hui su committed
1553
      best_tx_type    = mic->mbmi.tx_type;
1554
#endif  // CONFIG_EXT_TX
1555 1556 1557 1558 1559 1560 1561
      *rate           = this_rate;
      *rate_tokenonly = this_rate_tokenonly;
      *distortion     = this_distortion;
      *skippable      = s;
    }
  }

hui su's avatar
hui su committed
1562 1563 1564 1565 1566
  if (cpi->common.allow_screen_content_tools)
    rd_pick_palette_intra_sby(cpi, x, bsize, palette_ctx, bmode_costs[DC_PRED],
                              &palette_mode_info, best_palette_color_map,
                              &best_tx, &mode_selected, &best_rd);

1567 1568
  mic->mbmi.mode = mode_selected;
  mic->mbmi.tx_size = best_tx;
1569
#if CONFIG_EXT_TX
hui su's avatar
hui su committed
1570
  mic->mbmi.tx_type = best_tx_type;
1571
#endif  // CONFIG_EXT_TX
hui su's avatar
hui su committed
1572 1573 1574 1575 1576 1577 1578 1579 1580
  mic->mbmi.palette_mode_info.palette_size[0] =
      palette_mode_info.palette_size[0];
  if (palette_mode_info.palette_size[0] > 0) {
    memcpy(mic->mbmi.palette_mode_info.palette_colors,
           palette_mode_info.palette_colors,
           PALETTE_MAX_SIZE * sizeof(palette_mode_info.palette_colors[0]));
    memcpy(xd->plane[0].color_index_map, best_palette_color_map,
           rows * cols * sizeof(best_palette_color_map[0]));
  }
1581 1582 1583 1584

  return best_rd;
}

1585
#if CONFIG_VAR_TX
1586
static void tx_block_rd_b(const VP10_COMP *cpi, MACROBLOCK *x, TX_SIZE tx_size,
1587 1588 1589 1590 1591 1592
                          int blk_row, int blk_col, int plane, int block,
                          int plane_bsize, int coeff_ctx,
                          int *rate, int64_t *dist, int64_t *bsse, int *skip) {
  MACROBLOCKD *xd = &x->e_mbd;
  const struct macroblock_plane *const p = &x->plane[plane];
  struct macroblockd_plane *const pd = &xd->plane[plane];
1593
#if CONFIG_VP9_HIGHBITDEPTH
1594 1595 1596 1597
  const int ss_txfrm_size = tx_size << 1;
  int64_t this_sse;
  int shift = tx_size == TX_32X32 ? 0 : 2;
  tran_low_t *const coeff = BLOCK_OFFSET(p->coeff, block);
1598 1599
#endif
  unsigned int tmp_sse = 0;
1600 1601 1602 1603 1604 1605
  tran_low_t *const dqcoeff = BLOCK_OFFSET(pd->dqcoeff, block);
  PLANE_TYPE plane_type = (plane == 0) ? PLANE_TYPE_Y : PLANE_TYPE_UV;
  TX_TYPE tx_type = get_tx_type(plane_type, xd, block, tx_size);
  const scan_order *const scan_order =
      get_scan(tx_size, tx_type, is_inter_block(&xd->mi[0]->mbmi));

1606 1607 1608 1609 1610 1611 1612 1613 1614 1615 1616 1617 1618 1619 1620
  BLOCK_SIZE txm_bsize = txsize_to_bsize[tx_size];
  int bh = 4 * num_4x4_blocks_wide_lookup[txm_bsize];
  int src_stride = p->src.stride;
  uint8_t *src = &p->src.buf[4 * blk_row * src_stride + 4 * blk_col];
  uint8_t *dst = &pd->dst.buf[4 * blk_row * pd->dst.stride + 4 * blk_col];
  DECLARE_ALIGNED(16, uint8_t, rec_buffer[32 * 32]);

  int max_blocks_high = num_4x4_blocks_high_lookup[plane_bsize];
  int max_blocks_wide = num_4x4_blocks_wide_lookup[plane_bsize];

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

1621 1622
  vp10_xform_quant(x, plane, block, blk_row, blk_col, plane_bsize, tx_size);

1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644
  vpx_convolve_copy(dst, pd->dst.stride, rec_buffer, 32,
                    NULL, 0, NULL, 0, bh, bh);

  if (blk_row + (bh >> 2) > max_blocks_high ||
      blk_col + (bh >> 2) > max_blocks_wide) {
    int idx, idy;
    unsigned int this_sse;
    int blocks_height = VPXMIN(bh >> 2, max_blocks_high - blk_row);
    int blocks_width  = VPXMIN(bh >> 2, max_blocks_wide - blk_col);
    for (idy = 0; idy < blocks_height; idy += 2) {
      for (idx = 0; idx < blocks_width; idx += 2) {
        cpi->fn_ptr[BLOCK_8X8].vf(src + 4 * idy * src_stride + 4 * idx,
                                  src_stride,
                                  rec_buffer + 4 * idy * 32 + 4 * idx,
                                  32, &this_sse);
        tmp_sse += this_sse;
      }
    }
  } else {
    cpi->fn_ptr[txm_bsize].vf(src, src_stride, rec_buffer, 32, &tmp_sse);
  }

1645 1646 1647
#if CONFIG_VP9_HIGHBITDEPTH
  *dist += vp10_highbd_block_error(coeff, dqcoeff, 16 << ss_txfrm_size,
                                   &this_sse, xd->bd) >> shift;
1648
  *bsse += this_sse >> shift;
1649
#else
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 1697
  *bsse += (int64_t)tmp_sse * 16;

  if (p->eobs[block] > 0) {
    switch (tx_size) {
      case TX_32X32:
        vp10_inv_txfm_add_32x32(dqcoeff, rec_buffer, 32, p->eobs[block],
                                tx_type);
        break;
      case TX_16X16:
        vp10_inv_txfm_add_16x16(dqcoeff, rec_buffer, 32, p->eobs[block],
                                tx_type);
        break;
      case TX_8X8:
        vp10_inv_txfm_add_8x8(dqcoeff, rec_buffer, 32, p->eobs[block],
                              tx_type);
        break;
      case TX_4X4:
        vp10_inv_txfm_add_4x4(dqcoeff, rec_buffer, 32, p->eobs[block],
                              tx_type,
                              xd->lossless[xd->mi[0]->mbmi.segment_id]);
        break;
      default:
        assert(0 && "Invalid transform size");
        break;
    }

    if ((bh >> 2) + blk_col > max_blocks_wide ||
        (bh >> 2) + blk_row > max_blocks_high) {
      int idx, idy;
      unsigned int this_sse;
      int blocks_height = VPXMIN(bh >> 2, max_blocks_high - blk_row);
      int blocks_width  = VPXMIN(bh >> 2, max_blocks_wide - blk_col);
      tmp_sse = 0;
      for (idy = 0; idy < blocks_height; idy += 2) {
        for (idx = 0; idx < blocks_width; idx += 2) {
          cpi->fn_ptr[BLOCK_8X8].vf(src + 4 * idy * src_stride + 4 * idx,
                                    src_stride,
                                    rec_buffer + 4 * idy * 32 + 4 * idx,
                                    32, &this_sse);
          tmp_sse += this_sse;
        }
      }
    } else {
      cpi->fn_ptr[txm_bsize].vf(src, src_stride,
                                rec_buffer, 32, &tmp_sse);
    }
  }
  *dist += (int64_t)tmp_sse * 16;
1698 1699 1700 1701 1702 1703 1704 1705 1706 1707 1708
#endif  // CONFIG_VP9_HIGHBITDEPTH

  *rate += cost_coeffs(x, plane, block, coeff_ctx, tx_size,
                       scan_order->scan, scan_order->neighbors, 0);
  *skip &= (p->eobs[block] == 0);
}

static void select_tx_block(const VP10_COMP *cpi, MACROBLOCK *x,
                            int blk_row, int blk_col, int plane, int block,
                            TX_SIZE tx_size, BLOCK_SIZE plane_bsize,
                            ENTROPY_CONTEXT *ta, ENTROPY_CONTEXT *tl,
1709
                            TXFM_CONTEXT *tx_above, TXFM_CONTEXT *tx_left,
1710
                            int *rate, int64_t *dist,
1711 1712
                            int64_t *bsse, int *skip,
                            int64_t ref_best_rd, int *is_cost_valid) {
1713 1714 1715 1716 1717 1718 1719 1720 1721 1722 1723
  MACROBLOCKD *const xd = &x->e_mbd;
  MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;
  struct macroblock_plane *const p = &x->plane[plane];
  struct macroblockd_plane *const pd = &xd->plane[plane];
  int tx_idx = (blk_row >> (1 - pd->subsampling_y)) * 8 +
               (blk_col >> (1 - pd->subsampling_x));
  int max_blocks_high = num_4x4_blocks_high_lookup[plane_bsize];
  int max_blocks_wide = num_4x4_blocks_wide_lookup[plane_bsize];
  int64_t this_rd = INT64_MAX;
  ENTROPY_CONTEXT *pta = ta + blk_col;
  ENTROPY_CONTEXT *ptl = tl + blk_row;