encodemb.c 40.9 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
/*
 *  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 "./vp10_rtcd.h"
#include "./vpx_config.h"
#include "./vpx_dsp_rtcd.h"

#include "vpx_dsp/quantize.h"
#include "vpx_mem/vpx_mem.h"
#include "vpx_ports/mem.h"

20 21 22 23
#include "vp10/common/idct.h"
#include "vp10/common/reconinter.h"
#include "vp10/common/reconintra.h"
#include "vp10/common/scan.h"
24

25 26 27
#include "vp10/encoder/encodemb.h"
#include "vp10/encoder/rd.h"
#include "vp10/encoder/tokenize.h"
28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 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

struct optimize_ctx {
  ENTROPY_CONTEXT ta[MAX_MB_PLANE][16];
  ENTROPY_CONTEXT tl[MAX_MB_PLANE][16];
};

void vp10_subtract_plane(MACROBLOCK *x, BLOCK_SIZE bsize, int plane) {
  struct macroblock_plane *const p = &x->plane[plane];
  const struct macroblockd_plane *const pd = &x->e_mbd.plane[plane];
  const BLOCK_SIZE plane_bsize = get_plane_block_size(bsize, pd);
  const int bw = 4 * num_4x4_blocks_wide_lookup[plane_bsize];
  const int bh = 4 * num_4x4_blocks_high_lookup[plane_bsize];

#if CONFIG_VP9_HIGHBITDEPTH
  if (x->e_mbd.cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
    vpx_highbd_subtract_block(bh, bw, p->src_diff, bw, p->src.buf,
                              p->src.stride, pd->dst.buf, pd->dst.stride,
                              x->e_mbd.bd);
    return;
  }
#endif  // CONFIG_VP9_HIGHBITDEPTH
  vpx_subtract_block(bh, bw, p->src_diff, bw, p->src.buf, p->src.stride,
                     pd->dst.buf, pd->dst.stride);
}

#define RDTRUNC(RM, DM, R, D) ((128 + (R) * (RM)) & 0xFF)

typedef struct vp10_token_state {
  int           rate;
  int           error;
  int           next;
  int16_t       token;
  short         qc;
} vp10_token_state;

// TODO(jimbankoski): experiment to find optimal RD numbers.
static const int plane_rd_mult[PLANE_TYPES] = { 4, 2 };

#define UPDATE_RD_COST()\
{\
  rd_cost0 = RDCOST(rdmult, rddiv, rate0, error0);\
  rd_cost1 = RDCOST(rdmult, rddiv, rate1, error1);\
  if (rd_cost0 == rd_cost1) {\
    rd_cost0 = RDTRUNC(rdmult, rddiv, rate0, error0);\
    rd_cost1 = RDTRUNC(rdmult, rddiv, rate1, error1);\
  }\
}

// This function is a place holder for now but may ultimately need
// to scan previous tokens to work out the correct context.
static int trellis_get_coeff_context(const int16_t *scan,
                                     const int16_t *nb,
                                     int idx, int token,
                                     uint8_t *token_cache) {
  int bak = token_cache[scan[idx]], pt;
  token_cache[scan[idx]] = vp10_pt_energy_class[token];
  pt = get_coef_context(nb, token_cache, idx + 1);
  token_cache[scan[idx]] = bak;
  return pt;
}

static int optimize_b(MACROBLOCK *mb, int plane, int block,
                      TX_SIZE tx_size, int ctx) {
  MACROBLOCKD *const xd = &mb->e_mbd;
  struct macroblock_plane *const p = &mb->plane[plane];
  struct macroblockd_plane *const pd = &xd->plane[plane];
  const int ref = is_inter_block(&xd->mi[0]->mbmi);
  vp10_token_state tokens[1025][2];
  unsigned best_index[1025][2];
  uint8_t token_cache[1024];
  const tran_low_t *const coeff = BLOCK_OFFSET(mb->plane[plane].coeff, block);
  tran_low_t *const qcoeff = BLOCK_OFFSET(p->qcoeff, block);
  tran_low_t *const dqcoeff = BLOCK_OFFSET(pd->dqcoeff, block);
  const int eob = p->eobs[block];
  const PLANE_TYPE type = pd->plane_type;
  const int default_eob = 16 << (tx_size << 1);
  const int mul = 1 + (tx_size == TX_32X32);
  const int16_t *dequant_ptr = pd->dequant;
  const uint8_t *const band_translate = get_band_translate(tx_size);
107 108
  TX_TYPE tx_type = get_tx_type(type, xd, block);
  const scan_order *const so = get_scan(tx_size, tx_type);
109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326
  const int16_t *const scan = so->scan;
  const int16_t *const nb = so->neighbors;
  int next = eob, sz = 0;
  int64_t rdmult = mb->rdmult * plane_rd_mult[type], rddiv = mb->rddiv;
  int64_t rd_cost0, rd_cost1;
  int rate0, rate1, error0, error1;
  int16_t t0, t1;
  EXTRABIT e0;
  int best, band, pt, i, final_eob;
#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

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

  /* Now set up a Viterbi trellis to evaluate alternative roundings. */
  if (!ref)
    rdmult = (rdmult * 9) >> 4;

  /* Initialize the sentinel node of the trellis. */
  tokens[eob][0].rate = 0;
  tokens[eob][0].error = 0;
  tokens[eob][0].next = default_eob;
  tokens[eob][0].token = EOB_TOKEN;
  tokens[eob][0].qc = 0;
  tokens[eob][1] = tokens[eob][0];

  for (i = 0; i < eob; i++)
    token_cache[scan[i]] =
        vp10_pt_energy_class[vp10_get_token(qcoeff[scan[i]])];

  for (i = eob; i-- > 0;) {
    int base_bits, d2, dx;
    const int rc = scan[i];
    int x = qcoeff[rc];
    /* Only add a trellis state for non-zero coefficients. */
    if (x) {
      int shortcut = 0;
      error0 = tokens[next][0].error;
      error1 = tokens[next][1].error;
      /* Evaluate the first possibility for this state. */
      rate0 = tokens[next][0].rate;
      rate1 = tokens[next][1].rate;
      vp10_get_token_extra(x, &t0, &e0);
      /* Consider both possible successor states. */
      if (next < default_eob) {
        band = band_translate[i + 1];
        pt = trellis_get_coeff_context(scan, nb, i, t0, token_cache);
        rate0 += mb->token_costs[tx_size][type][ref][band][0][pt]
                                [tokens[next][0].token];
        rate1 += mb->token_costs[tx_size][type][ref][band][0][pt]
                                [tokens[next][1].token];
      }
      UPDATE_RD_COST();
      /* And pick the best. */
      best = rd_cost1 < rd_cost0;
      base_bits = vp10_get_cost(t0, e0, cat6_high_cost);
      dx = mul * (dqcoeff[rc] - coeff[rc]);
#if CONFIG_VP9_HIGHBITDEPTH
      if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
        dx >>= xd->bd - 8;
      }
#endif  // CONFIG_VP9_HIGHBITDEPTH
      d2 = dx * dx;
      tokens[i][0].rate = base_bits + (best ? rate1 : rate0);
      tokens[i][0].error = d2 + (best ? error1 : error0);
      tokens[i][0].next = next;
      tokens[i][0].token = t0;
      tokens[i][0].qc = x;
      best_index[i][0] = best;

      /* Evaluate the second possibility for this state. */
      rate0 = tokens[next][0].rate;
      rate1 = tokens[next][1].rate;

      if ((abs(x) * dequant_ptr[rc != 0] > abs(coeff[rc]) * mul) &&
          (abs(x) * dequant_ptr[rc != 0] < abs(coeff[rc]) * mul +
                                               dequant_ptr[rc != 0]))
        shortcut = 1;
      else
        shortcut = 0;

      if (shortcut) {
        sz = -(x < 0);
        x -= 2 * sz + 1;
      }

      /* Consider both possible successor states. */
      if (!x) {
        /* If we reduced this coefficient to zero, check to see if
         *  we need to move the EOB back here.
         */
        t0 = tokens[next][0].token == EOB_TOKEN ? EOB_TOKEN : ZERO_TOKEN;
        t1 = tokens[next][1].token == EOB_TOKEN ? EOB_TOKEN : ZERO_TOKEN;
        e0 = 0;
      } else {
        vp10_get_token_extra(x, &t0, &e0);
        t1 = t0;
      }
      if (next < default_eob) {
        band = band_translate[i + 1];
        if (t0 != EOB_TOKEN) {
          pt = trellis_get_coeff_context(scan, nb, i, t0, token_cache);
          rate0 += mb->token_costs[tx_size][type][ref][band][!x][pt]
                                  [tokens[next][0].token];
        }
        if (t1 != EOB_TOKEN) {
          pt = trellis_get_coeff_context(scan, nb, i, t1, token_cache);
          rate1 += mb->token_costs[tx_size][type][ref][band][!x][pt]
                                  [tokens[next][1].token];
        }
      }

      UPDATE_RD_COST();
      /* And pick the best. */
      best = rd_cost1 < rd_cost0;
      base_bits = vp10_get_cost(t0, e0, cat6_high_cost);

      if (shortcut) {
#if CONFIG_VP9_HIGHBITDEPTH
        if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
          dx -= ((dequant_ptr[rc != 0] >> (xd->bd - 8)) + sz) ^ sz;
        } else {
          dx -= (dequant_ptr[rc != 0] + sz) ^ sz;
        }
#else
        dx -= (dequant_ptr[rc != 0] + sz) ^ sz;
#endif  // CONFIG_VP9_HIGHBITDEPTH
        d2 = dx * dx;
      }
      tokens[i][1].rate = base_bits + (best ? rate1 : rate0);
      tokens[i][1].error = d2 + (best ? error1 : error0);
      tokens[i][1].next = next;
      tokens[i][1].token = best ? t1 : t0;
      tokens[i][1].qc = x;
      best_index[i][1] = best;
      /* Finally, make this the new head of the trellis. */
      next = i;
    } else {
      /* There's no choice to make for a zero coefficient, so we don't
       *  add a new trellis node, but we do need to update the costs.
       */
      band = band_translate[i + 1];
      t0 = tokens[next][0].token;
      t1 = tokens[next][1].token;
      /* Update the cost of each path if we're past the EOB token. */
      if (t0 != EOB_TOKEN) {
        tokens[next][0].rate +=
            mb->token_costs[tx_size][type][ref][band][1][0][t0];
        tokens[next][0].token = ZERO_TOKEN;
      }
      if (t1 != EOB_TOKEN) {
        tokens[next][1].rate +=
            mb->token_costs[tx_size][type][ref][band][1][0][t1];
        tokens[next][1].token = ZERO_TOKEN;
      }
      best_index[i][0] = best_index[i][1] = 0;
      /* Don't update next, because we didn't add a new node. */
    }
  }

  /* Now pick the best path through the whole trellis. */
  band = band_translate[i + 1];
  rate0 = tokens[next][0].rate;
  rate1 = tokens[next][1].rate;
  error0 = tokens[next][0].error;
  error1 = tokens[next][1].error;
  t0 = tokens[next][0].token;
  t1 = tokens[next][1].token;
  rate0 += mb->token_costs[tx_size][type][ref][band][0][ctx][t0];
  rate1 += mb->token_costs[tx_size][type][ref][band][0][ctx][t1];
  UPDATE_RD_COST();
  best = rd_cost1 < rd_cost0;
  final_eob = -1;
  memset(qcoeff, 0, sizeof(*qcoeff) * (16 << (tx_size * 2)));
  memset(dqcoeff, 0, sizeof(*dqcoeff) * (16 << (tx_size * 2)));
  for (i = next; i < eob; i = next) {
    const int x = tokens[i][best].qc;
    const int rc = scan[i];
    if (x) {
      final_eob = i;
    }

    qcoeff[rc] = x;
    dqcoeff[rc] = (x * dequant_ptr[rc != 0]) / mul;

    next = tokens[i][best].next;
    best = best_index[i][best];
  }
  final_eob++;

  mb->plane[plane].eobs[block] = final_eob;
  return final_eob;
}

static INLINE void fdct32x32(int rd_transform,
                             const int16_t *src, tran_low_t *dst,
                             int src_stride) {
  if (rd_transform)
    vpx_fdct32x32_rd(src, dst, src_stride);
  else
    vpx_fdct32x32(src, dst, src_stride);
}

#if CONFIG_VP9_HIGHBITDEPTH
static INLINE void highbd_fdct32x32(int rd_transform, const int16_t *src,
                                    tran_low_t *dst, int src_stride) {
  if (rd_transform)
    vpx_highbd_fdct32x32_rd(src, dst, src_stride);
  else
    vpx_highbd_fdct32x32(src, dst, src_stride);
}
#endif  // CONFIG_VP9_HIGHBITDEPTH

void vp10_xform_quant_fp(MACROBLOCK *x, int plane, int block,
327 328
                         int blk_row, int blk_col,
                         BLOCK_SIZE plane_bsize, TX_SIZE tx_size) {
329 330 331
  MACROBLOCKD *const xd = &x->e_mbd;
  const struct macroblock_plane *const p = &x->plane[plane];
  const struct macroblockd_plane *const pd = &xd->plane[plane];
332 333 334
  PLANE_TYPE plane_type = (plane == 0) ? PLANE_TYPE_Y : PLANE_TYPE_UV;
  TX_TYPE tx_type = get_tx_type(plane_type, xd, block);
  const scan_order *const scan_order = get_scan(tx_size, tx_type);
335 336 337 338 339 340
  tran_low_t *const coeff = BLOCK_OFFSET(p->coeff, block);
  tran_low_t *const qcoeff = BLOCK_OFFSET(p->qcoeff, block);
  tran_low_t *const dqcoeff = BLOCK_OFFSET(pd->dqcoeff, block);
  uint16_t *const eob = &p->eobs[block];
  const int diff_stride = 4 * num_4x4_blocks_wide_lookup[plane_bsize];
  const int16_t *src_diff;
341
  src_diff = &p->src_diff[4 * (blk_row * diff_stride + blk_col)];
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

#if CONFIG_VP9_HIGHBITDEPTH
  if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
    switch (tx_size) {
      case TX_32X32:
        highbd_fdct32x32(x->use_lp32x32fdct, src_diff, coeff, diff_stride);
        vp10_highbd_quantize_fp_32x32(coeff, 1024, x->skip_block, p->zbin,
                                     p->round_fp, p->quant_fp, p->quant_shift,
                                     qcoeff, dqcoeff, pd->dequant,
                                     eob, scan_order->scan,
                                     scan_order->iscan);
        break;
      case TX_16X16:
        vpx_highbd_fdct16x16(src_diff, coeff, diff_stride);
        vp10_highbd_quantize_fp(coeff, 256, x->skip_block, p->zbin, p->round_fp,
                               p->quant_fp, p->quant_shift, qcoeff, dqcoeff,
                               pd->dequant, eob,
                               scan_order->scan, scan_order->iscan);
        break;
      case TX_8X8:
        vpx_highbd_fdct8x8(src_diff, coeff, diff_stride);
        vp10_highbd_quantize_fp(coeff, 64, x->skip_block, p->zbin, p->round_fp,
                               p->quant_fp, p->quant_shift, qcoeff, dqcoeff,
                               pd->dequant, eob,
                               scan_order->scan, scan_order->iscan);
        break;
      case TX_4X4:
369
        if (xd->lossless[xd->mi[0]->mbmi.segment_id]) {
370 371 372 373
          vp10_highbd_fwht4x4(src_diff, coeff, diff_stride);
        } else {
          vpx_highbd_fdct4x4(src_diff, coeff, diff_stride);
        }
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
        vp10_highbd_quantize_fp(coeff, 16, x->skip_block, p->zbin, p->round_fp,
                               p->quant_fp, p->quant_shift, qcoeff, dqcoeff,
                               pd->dequant, eob,
                               scan_order->scan, scan_order->iscan);
        break;
      default:
        assert(0);
    }
    return;
  }
#endif  // CONFIG_VP9_HIGHBITDEPTH

  switch (tx_size) {
    case TX_32X32:
      fdct32x32(x->use_lp32x32fdct, src_diff, coeff, diff_stride);
      vp10_quantize_fp_32x32(coeff, 1024, x->skip_block, p->zbin, p->round_fp,
                            p->quant_fp, p->quant_shift, qcoeff, dqcoeff,
                            pd->dequant, eob, scan_order->scan,
                            scan_order->iscan);
      break;
    case TX_16X16:
      vpx_fdct16x16(src_diff, coeff, diff_stride);
      vp10_quantize_fp(coeff, 256, x->skip_block, p->zbin, p->round_fp,
                      p->quant_fp, p->quant_shift, qcoeff, dqcoeff,
                      pd->dequant, eob,
                      scan_order->scan, scan_order->iscan);
      break;
    case TX_8X8:
      vp10_fdct8x8_quant(src_diff, diff_stride, coeff, 64,
                        x->skip_block, p->zbin, p->round_fp,
                        p->quant_fp, p->quant_shift, qcoeff, dqcoeff,
                        pd->dequant, eob,
                        scan_order->scan, scan_order->iscan);
      break;
    case TX_4X4:
409
      if (xd->lossless[xd->mi[0]->mbmi.segment_id]) {
410 411 412 413
        vp10_fwht4x4(src_diff, coeff, diff_stride);
      } else {
        vpx_fdct4x4(src_diff, coeff, diff_stride);
      }
414 415 416 417 418 419 420 421 422 423 424 425
      vp10_quantize_fp(coeff, 16, x->skip_block, p->zbin, p->round_fp,
                      p->quant_fp, p->quant_shift, qcoeff, dqcoeff,
                      pd->dequant, eob,
                      scan_order->scan, scan_order->iscan);
      break;
    default:
      assert(0);
      break;
  }
}

void vp10_xform_quant_dc(MACROBLOCK *x, int plane, int block,
426 427
                         int blk_row, int blk_col,
                         BLOCK_SIZE plane_bsize, TX_SIZE tx_size) {
428 429 430 431 432 433 434 435 436
  MACROBLOCKD *const xd = &x->e_mbd;
  const struct macroblock_plane *const p = &x->plane[plane];
  const struct macroblockd_plane *const pd = &xd->plane[plane];
  tran_low_t *const coeff = BLOCK_OFFSET(p->coeff, block);
  tran_low_t *const qcoeff = BLOCK_OFFSET(p->qcoeff, block);
  tran_low_t *const dqcoeff = BLOCK_OFFSET(pd->dqcoeff, block);
  uint16_t *const eob = &p->eobs[block];
  const int diff_stride = 4 * num_4x4_blocks_wide_lookup[plane_bsize];
  const int16_t *src_diff;
437
  src_diff = &p->src_diff[4 * (blk_row * diff_stride + blk_col)];
438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460

#if CONFIG_VP9_HIGHBITDEPTH
  if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
    switch (tx_size) {
      case TX_32X32:
        vpx_highbd_fdct32x32_1(src_diff, coeff, diff_stride);
        vpx_highbd_quantize_dc_32x32(coeff, x->skip_block, p->round,
                                     p->quant_fp[0], qcoeff, dqcoeff,
                                     pd->dequant[0], eob);
        break;
      case TX_16X16:
        vpx_highbd_fdct16x16_1(src_diff, coeff, diff_stride);
        vpx_highbd_quantize_dc(coeff, 256, x->skip_block, p->round,
                               p->quant_fp[0], qcoeff, dqcoeff,
                               pd->dequant[0], eob);
        break;
      case TX_8X8:
        vpx_highbd_fdct8x8_1(src_diff, coeff, diff_stride);
        vpx_highbd_quantize_dc(coeff, 64, x->skip_block, p->round,
                               p->quant_fp[0], qcoeff, dqcoeff,
                               pd->dequant[0], eob);
        break;
      case TX_4X4:
461
        if (xd->lossless[xd->mi[0]->mbmi.segment_id]) {
462 463 464 465
          vp10_highbd_fwht4x4(src_diff, coeff, diff_stride);
        } else {
          vpx_highbd_fdct4x4(src_diff, coeff, diff_stride);
        }
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
        vpx_highbd_quantize_dc(coeff, 16, x->skip_block, p->round,
                               p->quant_fp[0], qcoeff, dqcoeff,
                               pd->dequant[0], eob);
        break;
      default:
        assert(0);
    }
    return;
  }
#endif  // CONFIG_VP9_HIGHBITDEPTH

  switch (tx_size) {
    case TX_32X32:
      vpx_fdct32x32_1(src_diff, coeff, diff_stride);
      vpx_quantize_dc_32x32(coeff, x->skip_block, p->round,
                            p->quant_fp[0], qcoeff, dqcoeff,
                            pd->dequant[0], eob);
      break;
    case TX_16X16:
      vpx_fdct16x16_1(src_diff, coeff, diff_stride);
      vpx_quantize_dc(coeff, 256, x->skip_block, p->round,
                     p->quant_fp[0], qcoeff, dqcoeff,
                     pd->dequant[0], eob);
      break;
    case TX_8X8:
      vpx_fdct8x8_1(src_diff, coeff, diff_stride);
      vpx_quantize_dc(coeff, 64, x->skip_block, p->round,
                      p->quant_fp[0], qcoeff, dqcoeff,
                      pd->dequant[0], eob);
      break;
    case TX_4X4:
497
      if (xd->lossless[xd->mi[0]->mbmi.segment_id]) {
498 499 500 501
        vp10_fwht4x4(src_diff, coeff, diff_stride);
      } else {
        vpx_fdct4x4(src_diff, coeff, diff_stride);
      }
502 503 504 505 506 507 508 509 510 511
      vpx_quantize_dc(coeff, 16, x->skip_block, p->round,
                      p->quant_fp[0], qcoeff, dqcoeff,
                      pd->dequant[0], eob);
      break;
    default:
      assert(0);
      break;
  }
}

512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529
void vp10_fwd_txfm_4x4(const int16_t *src_diff, tran_low_t *coeff,
                       int diff_stride, TX_TYPE tx_type, int lossless) {
  if (lossless) {
    vp10_fwht4x4(src_diff, coeff, diff_stride);
  } else {
    switch (tx_type) {
      case DCT_DCT:
        vpx_fdct4x4(src_diff, coeff, diff_stride);
        break;
      case ADST_DCT:
      case DCT_ADST:
      case ADST_ADST:
        vp10_fht4x4(src_diff, coeff, diff_stride, tx_type);
        break;
      default:
        assert(0);
        break;
    }
hui su's avatar
hui su committed
530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582
  }
}

static void fwd_txfm_8x8(const int16_t *src_diff, tran_low_t *coeff,
                         int diff_stride, TX_TYPE tx_type) {
  switch (tx_type) {
    case DCT_DCT:
    case ADST_DCT:
    case DCT_ADST:
    case ADST_ADST:
      vp10_fht8x8(src_diff, coeff, diff_stride, tx_type);
      break;
    default:
      assert(0);
      break;
  }
}

static void fwd_txfm_16x16(const int16_t *src_diff, tran_low_t *coeff,
                           int diff_stride, TX_TYPE tx_type) {
  switch (tx_type) {
    case DCT_DCT:
    case ADST_DCT:
    case DCT_ADST:
    case ADST_ADST:
      vp10_fht16x16(src_diff, coeff, diff_stride, tx_type);
      break;
    default:
      assert(0);
      break;
  }
}

static void fwd_txfm_32x32(int rd_transform, const int16_t *src_diff,
                           tran_low_t *coeff, int diff_stride,
                           TX_TYPE tx_type) {
  switch (tx_type) {
    case DCT_DCT:
      fdct32x32(rd_transform, src_diff, coeff, diff_stride);
      break;
    case ADST_DCT:
    case DCT_ADST:
    case ADST_ADST:
      assert(0);
      break;
    default:
      assert(0);
      break;
  }
}

#if CONFIG_VP9_HIGHBITDEPTH
void vp10_highbd_fwd_txfm_4x4(const int16_t *src_diff, tran_low_t *coeff,
583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600
                              int diff_stride, TX_TYPE tx_type, int lossless) {
  if (lossless) {
    assert(tx_type == DCT_DCT);
    vp10_highbd_fwht4x4(src_diff, coeff, diff_stride);
  } else {
    switch (tx_type) {
      case DCT_DCT:
        vpx_highbd_fdct4x4(src_diff, coeff, diff_stride);
        break;
      case ADST_DCT:
      case DCT_ADST:
      case ADST_ADST:
        vp10_highbd_fht4x4(src_diff, coeff, diff_stride, tx_type);
        break;
      default:
        assert(0);
        break;
    }
hui su's avatar
hui su committed
601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656
  }
}

static void highbd_fwd_txfm_8x8(const int16_t *src_diff, tran_low_t *coeff,
                         int diff_stride, TX_TYPE tx_type) {
  switch (tx_type) {
    case DCT_DCT:
      vpx_highbd_fdct8x8(src_diff, coeff, diff_stride);
      break;
    case ADST_DCT:
    case DCT_ADST:
    case ADST_ADST:
      vp10_highbd_fht8x8(src_diff, coeff, diff_stride, tx_type);
      break;
    default:
      assert(0);
      break;
  }
}

static void highbd_fwd_txfm_16x16(const int16_t *src_diff, tran_low_t *coeff,
                           int diff_stride, TX_TYPE tx_type) {
  switch (tx_type) {
    case DCT_DCT:
      vpx_highbd_fdct16x16(src_diff, coeff, diff_stride);
      break;
    case ADST_DCT:
    case DCT_ADST:
    case ADST_ADST:
      vp10_highbd_fht16x16(src_diff, coeff, diff_stride, tx_type);
      break;
    default:
      assert(0);
      break;
  }
}

static void highbd_fwd_txfm_32x32(int rd_transform, const int16_t *src_diff,
                                  tran_low_t *coeff, int diff_stride,
                                  TX_TYPE tx_type) {
  switch (tx_type) {
    case DCT_DCT:
      highbd_fdct32x32(rd_transform, src_diff, coeff, diff_stride);
      break;
    case ADST_DCT:
    case DCT_ADST:
    case ADST_ADST:
      assert(0);
      break;
    default:
      assert(0);
      break;
  }
}
#endif  // CONFIG_VP9_HIGHBITDEPTH

657
void vp10_xform_quant(MACROBLOCK *x, int plane, int block,
658 659
                      int blk_row, int blk_col,
                      BLOCK_SIZE plane_bsize, TX_SIZE tx_size) {
660 661 662
  MACROBLOCKD *const xd = &x->e_mbd;
  const struct macroblock_plane *const p = &x->plane[plane];
  const struct macroblockd_plane *const pd = &xd->plane[plane];
663 664 665
  PLANE_TYPE plane_type = (plane == 0) ? PLANE_TYPE_Y : PLANE_TYPE_UV;
  TX_TYPE tx_type = get_tx_type(plane_type, xd, block);
  const scan_order *const scan_order = get_scan(tx_size, tx_type);
666 667 668 669 670 671
  tran_low_t *const coeff = BLOCK_OFFSET(p->coeff, block);
  tran_low_t *const qcoeff = BLOCK_OFFSET(p->qcoeff, block);
  tran_low_t *const dqcoeff = BLOCK_OFFSET(pd->dqcoeff, block);
  uint16_t *const eob = &p->eobs[block];
  const int diff_stride = 4 * num_4x4_blocks_wide_lookup[plane_bsize];
  const int16_t *src_diff;
672
  src_diff = &p->src_diff[4 * (blk_row * diff_stride + blk_col)];
673 674 675 676 677

#if CONFIG_VP9_HIGHBITDEPTH
  if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
     switch (tx_size) {
      case TX_32X32:
hui su's avatar
hui su committed
678 679
        highbd_fwd_txfm_32x32(x->use_lp32x32fdct, src_diff, coeff, diff_stride,
                         tx_type);
680 681 682 683 684 685
        vpx_highbd_quantize_b_32x32(coeff, 1024, x->skip_block, p->zbin,
                                    p->round, p->quant, p->quant_shift, qcoeff,
                                    dqcoeff, pd->dequant, eob,
                                    scan_order->scan, scan_order->iscan);
        break;
      case TX_16X16:
hui su's avatar
hui su committed
686
        highbd_fwd_txfm_16x16(src_diff, coeff, diff_stride, tx_type);
687 688 689 690 691 692
        vpx_highbd_quantize_b(coeff, 256, x->skip_block, p->zbin, p->round,
                              p->quant, p->quant_shift, qcoeff, dqcoeff,
                              pd->dequant, eob,
                              scan_order->scan, scan_order->iscan);
        break;
      case TX_8X8:
hui su's avatar
hui su committed
693
        highbd_fwd_txfm_8x8(src_diff, coeff, diff_stride, tx_type);
694 695 696 697 698 699
        vpx_highbd_quantize_b(coeff, 64, x->skip_block, p->zbin, p->round,
                              p->quant, p->quant_shift, qcoeff, dqcoeff,
                              pd->dequant, eob,
                              scan_order->scan, scan_order->iscan);
        break;
      case TX_4X4:
hui su's avatar
hui su committed
700
        vp10_highbd_fwd_txfm_4x4(src_diff, coeff, diff_stride, tx_type,
701
                                 xd->lossless[xd->mi[0]->mbmi.segment_id]);
702 703 704 705 706 707 708 709 710 711 712 713 714 715
        vpx_highbd_quantize_b(coeff, 16, x->skip_block, p->zbin, p->round,
                              p->quant, p->quant_shift, qcoeff, dqcoeff,
                              pd->dequant, eob,
                              scan_order->scan, scan_order->iscan);
        break;
      default:
        assert(0);
    }
    return;
  }
#endif  // CONFIG_VP9_HIGHBITDEPTH

  switch (tx_size) {
    case TX_32X32:
hui su's avatar
hui su committed
716
      fwd_txfm_32x32(x->use_lp32x32fdct, src_diff, coeff, diff_stride, tx_type);
717 718 719 720 721 722
      vpx_quantize_b_32x32(coeff, 1024, x->skip_block, p->zbin, p->round,
                           p->quant, p->quant_shift, qcoeff, dqcoeff,
                           pd->dequant, eob, scan_order->scan,
                           scan_order->iscan);
      break;
    case TX_16X16:
hui su's avatar
hui su committed
723
      fwd_txfm_16x16(src_diff, coeff, diff_stride, tx_type);
724 725 726 727 728 729
      vpx_quantize_b(coeff, 256, x->skip_block, p->zbin, p->round,
                     p->quant, p->quant_shift, qcoeff, dqcoeff,
                     pd->dequant, eob,
                     scan_order->scan, scan_order->iscan);
      break;
    case TX_8X8:
hui su's avatar
hui su committed
730
      fwd_txfm_8x8(src_diff, coeff, diff_stride, tx_type);
731 732 733 734 735 736
      vpx_quantize_b(coeff, 64, x->skip_block, p->zbin, p->round,
                     p->quant, p->quant_shift, qcoeff, dqcoeff,
                     pd->dequant, eob,
                     scan_order->scan, scan_order->iscan);
      break;
    case TX_4X4:
737 738
      vp10_fwd_txfm_4x4(src_diff, coeff, diff_stride, tx_type,
                        xd->lossless[xd->mi[0]->mbmi.segment_id]);
739 740 741 742 743 744 745 746 747 748 749
      vpx_quantize_b(coeff, 16, x->skip_block, p->zbin, p->round,
                     p->quant, p->quant_shift, qcoeff, dqcoeff,
                     pd->dequant, eob,
                     scan_order->scan, scan_order->iscan);
      break;
    default:
      assert(0);
      break;
  }
}

750 751
static void encode_block(int plane, int block, int blk_row, int blk_col,
                         BLOCK_SIZE plane_bsize,
752 753 754 755 756 757 758 759 760 761
                         TX_SIZE tx_size, void *arg) {
  struct encode_b_args *const args = arg;
  MACROBLOCK *const x = args->x;
  MACROBLOCKD *const xd = &x->e_mbd;
  struct optimize_ctx *const ctx = args->ctx;
  struct macroblock_plane *const p = &x->plane[plane];
  struct macroblockd_plane *const pd = &xd->plane[plane];
  tran_low_t *const dqcoeff = BLOCK_OFFSET(pd->dqcoeff, block);
  uint8_t *dst;
  ENTROPY_CONTEXT *a, *l;
hui su's avatar
hui su committed
762
  TX_TYPE tx_type = get_tx_type(pd->plane_type, xd, block);
763 764 765
  dst = &pd->dst.buf[4 * blk_row * pd->dst.stride + 4 * blk_col];
  a = &ctx->ta[plane][blk_col];
  l = &ctx->tl[plane][blk_row];
766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783

  // TODO(jingning): per transformed block zero forcing only enabled for
  // luma component. will integrate chroma components as well.
  if (x->zcoeff_blk[tx_size][block] && plane == 0) {
    p->eobs[block] = 0;
    *a = *l = 0;
    return;
  }

  if (!x->skip_recode) {
    if (x->quant_fp) {
      // Encoding process for rtc mode
      if (x->skip_txfm[0] == SKIP_TXFM_AC_DC && plane == 0) {
        // skip forward transform
        p->eobs[block] = 0;
        *a = *l = 0;
        return;
      } else {
784 785
        vp10_xform_quant_fp(x, plane, block, blk_row, blk_col,
                            plane_bsize, tx_size);
786 787 788 789 790 791
      }
    } else {
      if (max_txsize_lookup[plane_bsize] == tx_size) {
        int txfm_blk_index = (plane << 2) + (block >> (tx_size << 1));
        if (x->skip_txfm[txfm_blk_index] == SKIP_TXFM_NONE) {
          // full forward transform and quantization
792 793
          vp10_xform_quant(x, plane, block, blk_row, blk_col,
                           plane_bsize, tx_size);
794 795
        } else if (x->skip_txfm[txfm_blk_index] == SKIP_TXFM_AC_ONLY) {
          // fast path forward transform and quantization
796 797
          vp10_xform_quant_dc(x, plane, block, blk_row, blk_col,
                              plane_bsize, tx_size);
798 799 800 801 802 803 804
        } else {
          // skip forward transform
          p->eobs[block] = 0;
          *a = *l = 0;
          return;
        }
      } else {
805 806
        vp10_xform_quant(x, plane, block, blk_row, blk_col,
                         plane_bsize, tx_size);
807 808 809 810 811 812 813 814 815 816 817 818 819 820
      }
    }
  }

  if (x->optimize && (!x->skip_recode || !x->skip_optimize)) {
    const int ctx = combine_entropy_contexts(*a, *l);
    *a = *l = optimize_b(x, plane, block, tx_size, ctx) > 0;
  } else {
    *a = *l = p->eobs[block] > 0;
  }

  if (p->eobs[block])
    *(args->skip) = 0;

821
  if (p->eobs[block] == 0)
822 823 824 825 826
    return;
#if CONFIG_VP9_HIGHBITDEPTH
  if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
    switch (tx_size) {
      case TX_32X32:
hui su's avatar
hui su committed
827 828
        vp10_highbd_inv_txfm_add_32x32(dqcoeff, dst, pd->dst.stride,
                                       p->eobs[block], xd->bd, tx_type);
829 830
        break;
      case TX_16X16:
hui su's avatar
hui su committed
831 832
        vp10_highbd_inv_txfm_add_16x16(dqcoeff, dst, pd->dst.stride,
                                       p->eobs[block], xd->bd, tx_type);
833 834
        break;
      case TX_8X8:
hui su's avatar
hui su committed
835 836
        vp10_highbd_inv_txfm_add_8x8(dqcoeff, dst, pd->dst.stride,
                                     p->eobs[block], xd->bd, tx_type);
837 838 839 840 841
        break;
      case TX_4X4:
        // this is like vp10_short_idct4x4 but has a special case around eob<=1
        // which is significant (not just an optimization) for the lossless
        // case.
hui su's avatar
hui su committed
842 843
        vp10_highbd_inv_txfm_add_4x4(dqcoeff, dst, pd->dst.stride,
                                     p->eobs[block], xd->bd, tx_type,
844
                                     xd->lossless[xd->mi[0]->mbmi.segment_id]);
845 846 847
        break;
      default:
        assert(0 && "Invalid transform size");
hui su's avatar
hui su committed
848
        break;
849
    }
hui su's avatar
hui su committed
850

851 852 853 854 855 856
    return;
  }
#endif  // CONFIG_VP9_HIGHBITDEPTH

  switch (tx_size) {
    case TX_32X32:
hui su's avatar
hui su committed
857 858
      vp10_inv_txfm_add_32x32(dqcoeff, dst, pd->dst.stride, p->eobs[block],
                              tx_type);
859 860
      break;
    case TX_16X16:
hui su's avatar
hui su committed
861 862
      vp10_inv_txfm_add_16x16(dqcoeff, dst, pd->dst.stride, p->eobs[block],
                              tx_type);
863 864
      break;
    case TX_8X8:
hui su's avatar
hui su committed
865 866
      vp10_inv_txfm_add_8x8(dqcoeff, dst, pd->dst.stride, p->eobs[block],
                            tx_type);
867 868 869 870 871
      break;
    case TX_4X4:
      // this is like vp10_short_idct4x4 but has a special case around eob<=1
      // which is significant (not just an optimization) for the lossless
      // case.
hui su's avatar
hui su committed
872
      vp10_inv_txfm_add_4x4(dqcoeff, dst, pd->dst.stride, p->eobs[block],
873
                            tx_type, xd->lossless[xd->mi[0]->mbmi.segment_id]);
874 875 876 877 878 879 880
      break;
    default:
      assert(0 && "Invalid transform size");
      break;
  }
}

881 882
static void encode_block_pass1(int plane, int block, int blk_row, int blk_col,
                               BLOCK_SIZE plane_bsize,
883 884 885 886 887 888 889
                               TX_SIZE tx_size, void *arg) {
  MACROBLOCK *const x = (MACROBLOCK *)arg;
  MACROBLOCKD *const xd = &x->e_mbd;
  struct macroblock_plane *const p = &x->plane[plane];
  struct macroblockd_plane *const pd = &xd->plane[plane];
  tran_low_t *const dqcoeff = BLOCK_OFFSET(pd->dqcoeff, block);
  uint8_t *dst;
890
  dst = &pd->dst.buf[4 * blk_row * pd->dst.stride + 4 * blk_col];
891

892
  vp10_xform_quant(x, plane, block, blk_row, blk_col, plane_bsize, tx_size);
893 894 895 896

  if (p->eobs[block] > 0) {
#if CONFIG_VP9_HIGHBITDEPTH
    if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
897
      if (xd->lossless[0]) {
898 899 900 901 902 903 904
        vp10_highbd_iwht4x4_add(dqcoeff, dst, pd->dst.stride,
                                p->eobs[block], xd->bd);
      } else {
        vp10_highbd_idct4x4_add(dqcoeff, dst, pd->dst.stride,
                                p->eobs[block], xd->bd);
      }
      return;
905 906
    }
#endif  // CONFIG_VP9_HIGHBITDEPTH
907
    if (xd->lossless[0]) {
908 909 910 911
      vp10_iwht4x4_add(dqcoeff, dst, pd->dst.stride, p->eobs[block]);
    } else {
      vp10_idct4x4_add(dqcoeff, dst, pd->dst.stride, p->eobs[block]);
    }
912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948
  }
}

void vp10_encode_sby_pass1(MACROBLOCK *x, BLOCK_SIZE bsize) {
  vp10_subtract_plane(x, bsize, 0);
  vp10_foreach_transformed_block_in_plane(&x->e_mbd, bsize, 0,
                                         encode_block_pass1, x);
}

void vp10_encode_sb(MACROBLOCK *x, BLOCK_SIZE bsize) {
  MACROBLOCKD *const xd = &x->e_mbd;
  struct optimize_ctx ctx;
  MB_MODE_INFO *mbmi = &xd->mi[0]->mbmi;
  struct encode_b_args arg = {x, &ctx, &mbmi->skip};
  int plane;

  mbmi->skip = 1;

  if (x->skip)
    return;

  for (plane = 0; plane < MAX_MB_PLANE; ++plane) {
    if (!x->skip_recode)
      vp10_subtract_plane(x, bsize, plane);

    if (x->optimize && (!x->skip_recode || !x->skip_optimize)) {
      const struct macroblockd_plane* const pd = &xd->plane[plane];
      const TX_SIZE tx_size = plane ? get_uv_tx_size(mbmi, pd) : mbmi->tx_size;
      vp10_get_entropy_contexts(bsize, tx_size, pd,
                               ctx.ta[plane], ctx.tl[plane]);
    }

    vp10_foreach_transformed_block_in_plane(xd, bsize, plane, encode_block,
                                           &arg);
  }
}

949 950 951
void vp10_encode_block_intra(int plane, int block, int blk_row, int blk_col,
                             BLOCK_SIZE plane_bsize,
                             TX_SIZE tx_size, void *arg) {
952 953 954 955 956 957 958 959 960
  struct encode_b_args* const args = arg;
  MACROBLOCK *const x = args->x;
  MACROBLOCKD *const xd = &x->e_mbd;
  MB_MODE_INFO *mbmi = &xd->mi[0]->mbmi;
  struct macroblock_plane *const p = &x->plane[plane];
  struct macroblockd_plane *const pd = &xd->plane[plane];
  tran_low_t *coeff = BLOCK_OFFSET(p->coeff, block);
  tran_low_t *qcoeff = BLOCK_OFFSET(p->qcoeff, block);
  tran_low_t *dqcoeff = BLOCK_OFFSET(pd->dqcoeff, block);
961 962 963
  PLANE_TYPE plane_type = (plane == 0) ? PLANE_TYPE_Y : PLANE_TYPE_UV;
  TX_TYPE tx_type = get_tx_type(plane_type, xd, block);
  const scan_order *const scan_order = get_scan(tx_size, tx_type);
964 965
  PREDICTION_MODE mode;
  const int bwl = b_width_log2_lookup[plane_bsize];
966
  const int bhl = b_height_log2_lookup[plane_bsize];
967 968 969 970 971 972
  const int diff_stride = 4 * (1 << bwl);
  uint8_t *src, *dst;
  int16_t *src_diff;
  uint16_t *eob = &p->eobs[block];
  const int src_stride = p->src.stride;
  const int dst_stride = pd->dst.stride;
973 974 975
  dst = &pd->dst.buf[4 * (blk_row * dst_stride + blk_col)];
  src = &p->src.buf[4 * (blk_row * src_stride + blk_col)];
  src_diff = &p->src_diff[4 * (blk_row * diff_stride + blk_col)];
976

977
  mode = plane == 0 ? get_y_mode(xd->mi[0], block) : mbmi->uv_mode;
978
  vp10_predict_intra_block(xd, bwl, bhl, tx_size, mode, dst, dst_stride,
979
                          dst, dst_stride, blk_col, blk_row, plane);
980 981 982 983 984 985 986 987

#if CONFIG_VP9_HIGHBITDEPTH
  if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
    switch (tx_size) {
      case TX_32X32:
        if (!x->skip_recode) {
          vpx_highbd_subtract_block(32, 32, src_diff, diff_stride,
                                    src, src_stride, dst, dst_stride, xd->bd);
hui su's avatar
hui su committed
988 989
          highbd_fwd_txfm_32x32(x->use_lp32x32fdct, src_diff, coeff,
                                diff_stride, tx_type);
990 991 992 993 994
          vpx_highbd_quantize_b_32x32(coeff, 1024, x->skip_block, p->zbin,
                                      p->round, p->quant, p->quant_shift,
                                      qcoeff, dqcoeff, pd->dequant, eob,
                                      scan_order->scan, scan_order->iscan);
        }
995
        if (*eob)
hui su's avatar
hui su committed
996 997
          vp10_highbd_inv_txfm_add_32x32(dqcoeff, dst, dst_stride, *eob, xd->bd,
                                         tx_type);
998 999 1000 1001 1002
        break;
      case TX_16X16:
        if (!x->skip_recode) {
          vpx_highbd_subtract_block(16, 16, src_diff, diff_stride,
                                    src, src_stride, dst, dst_stride, xd->bd);
hui su's avatar
hui su committed
1003
          highbd_fwd_txfm_16x16(src_diff, coeff, diff_stride, tx_type);
1004 1005 1006 1007 1008
          vpx_highbd_quantize_b(coeff, 256, x->skip_block, p->zbin, p->round,
                                p->quant, p->quant_shift, qcoeff, dqcoeff,
                                pd->dequant, eob,
                                scan_order->scan, scan_order->iscan);
        }
1009
        if (*eob)
hui su's avatar
hui su committed
1010 1011
          vp10_highbd_inv_txfm_add_16x16(dqcoeff, dst, dst_stride, *eob, xd->bd,
                                         tx_type);
1012 1013 1014 1015 1016
        break;
      case TX_8X8:
        if (!x->skip_recode) {
          vpx_highbd_subtract_block(8, 8, src_diff, diff_stride,
                                    src, src_stride, dst, dst_stride, xd->bd);
hui su's avatar
hui su committed
1017
          highbd_fwd_txfm_8x8(src_diff, coeff, diff_stride, tx_type);
1018 1019 1020 1021 1022
          vpx_highbd_quantize_b(coeff, 64, x->skip_block, p->zbin, p->round,
                                p->quant, p->quant_shift, qcoeff, dqcoeff,
                                pd->dequant, eob,
                                scan_order->scan, scan_order->iscan);
        }
1023
        if (*eob)
hui su's avatar
hui su committed
1024 1025
          vp10_highbd_inv_txfm_add_8x8(dqcoeff, dst, dst_stride, *eob, xd->bd,
                                       tx_type);
1026 1027 1028 1029 1030
        break;
      case TX_4X4:
        if (!x->skip_recode) {
          vpx_highbd_subtract_block(4, 4, src_diff, diff_stride,
                                    src, src_stride, dst, dst_stride, xd->bd);
hui su's avatar
hui su committed
1031
          vp10_highbd_fwd_txfm_4x4(src_diff, coeff, diff_stride, tx_type,
1032
                                   xd->lossless[mbmi->segment_id]);
1033 1034 1035 1036 1037 1038
          vpx_highbd_quantize_b(coeff, 16, x->skip_block, p->zbin, p->round,
                                p->quant, p->quant_shift, qcoeff, dqcoeff,
                                pd->dequant, eob,
                                scan_order->scan, scan_order->iscan);
        }

1039
        if (*eob)
hui su's avatar
hui su committed
1040 1041 1042 1043
          // this is like vp10_short_idct4x4 but has a special case around
          // eob<=1 which is significant (not just an optimization) for the
          // lossless case.
          vp10_highbd_inv_txfm_add_4x4(dqcoeff, dst, dst_stride, *eob, xd->bd,
1044
                                       tx_type, xd->lossless[mbmi->segment_id]);
1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060
        break;
      default:
        assert(0);
        return;
    }
    if (*eob)
      *(args->skip) = 0;
    return;
  }
#endif  // CONFIG_VP9_HIGHBITDEPTH

  switch (tx_size) {
    case TX_32X32:
      if (!x->skip_recode) {
        vpx_subtract_block(32, 32, src_diff, diff_stride,
                           src, src_stride, dst, dst_stride);
hui su's avatar
hui su committed
1061 1062
        fwd_txfm_32x32(x->use_lp32x32fdct, src_diff, coeff, diff_stride,
                       tx_type);
1063 1064 1065 1066 1067
        vpx_quantize_b_32x32(coeff, 1024, x->skip_block, p->zbin, p->round,
                             p->quant, p->quant_shift, qcoeff, dqcoeff,
                             pd->dequant, eob, scan_order->scan,
                             scan_order->iscan);
      }
1068
      if (*eob)
hui su's avatar
hui su committed
1069
        vp10_inv_txfm_add_32x32(dqcoeff, dst, dst_stride, *eob, tx_type);
1070 1071 1072 1073 1074
      break;
    case TX_16X16:
      if (!x->skip_recode) {
        vpx_subtract_block(16, 16, src_diff, diff_stride,
                           src, src_stride, dst, dst_stride);
hui su's avatar
hui su committed
1075
        fwd_txfm_16x16(src_diff, coeff, diff_stride, tx_type);
1076 1077 1078 1079 1080
        vpx_quantize_b(coeff, 256, x->skip_block, p->zbin, p->round,
                       p->quant, p->quant_shift, qcoeff, dqcoeff,
                       pd->dequant, eob, scan_order->scan,
                       scan_order->iscan);
      }
1081
      if (*eob)
hui su's avatar
hui su committed
1082
        vp10_inv_txfm_add_16x16(dqcoeff, dst, dst_stride, *eob, tx_type);
1083 1084 1085 1086 1087
      break;
    case TX_8X8:
      if (!x->skip_recode) {
        vpx_subtract_block(8, 8, src_diff, diff_stride,
                           src, src_stride, dst, dst_stride);
hui su's avatar
hui su committed
1088
        fwd_txfm_8x8(src_diff, coeff, diff_stride, tx_type);
1089 1090 1091 1092 1093
        vpx_quantize_b(coeff, 64, x->skip_block, p->zbin, p->round, p->quant,
                       p->quant_shift, qcoeff, dqcoeff,
                       pd->dequant, eob, scan_order->scan,
                       scan_order->iscan);
      }
1094
      if (*eob)
hui su's avatar
hui su committed
1095
        vp10_inv_txfm_add_8x8(dqcoeff, dst, dst_stride, *eob, tx_type);
1096 1097 1098 1099 1100
      break;
    case TX_4X4:
      if (!x->skip_recode) {
        vpx_subtract_block(4, 4, src_diff, diff_stride,
                           src, src_stride, dst, dst_stride);
1101 1102
        vp10_fwd_txfm_4x4(src_diff, coeff, diff_stride, tx_type,
                          xd->lossless[mbmi->segment_id]);
1103 1104 1105 1106 1107 1108
        vpx_quantize_b(coeff, 16, x->skip_block, p->zbin, p->round, p->quant,
                       p->quant_shift, qcoeff, dqcoeff,
                       pd->dequant, eob, scan_order->scan,
                       scan_order->iscan);
      }

1109
      if (*eob) {
hui su's avatar
hui su committed
1110 1111 1112 1113
        // this is like vp10_short_idct4x4 but has a special case around eob<=1
        // which is significant (not just an optimization) for the lossless
        // case.
        vp10_inv_txfm_add_4x4(dqcoeff, dst, dst_stride, *eob, tx_type,
1114
                              xd->lossless[mbmi->segment_id]);
1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129
      }
      break;
    default:
      assert(0);
      break;
  }
  if (*eob)
    *(args->skip) = 0;
}

void vp10_encode_intra_block_plane(MACROBLOCK *x, BLOCK_SIZE bsize, int plane) {
  const MACROBLOCKD *const xd = &x->e_mbd;
  struct encode_b_args arg = {x, NULL, &xd->mi[0]->mbmi.skip};

  vp10_foreach_transformed_block_in_plane(xd, bsize, plane,
1130
                                          vp10_encode_block_intra, &arg);
1131
}