vp9_decodframe.c 66 KB
Newer Older
John Koleszar's avatar
John Koleszar committed
1
/*
2
 *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
John Koleszar's avatar
John Koleszar committed
3
 *
4
 *  Use of this source code is governed by a BSD-style license
5 6
 *  that can be found in the LICENSE file in the root of the source
 *  tree. An additional intellectual property rights grant can be found
7
 *  in the file PATENTS.  All contributing project authors may
8
 *  be found in the AUTHORS file in the root of the source tree.
John Koleszar's avatar
John Koleszar committed
9 10 11
 */


12
#include "vp9/decoder/vp9_onyxd_int.h"
Ronald S. Bultje's avatar
Ronald S. Bultje committed
13
#include "vp9/common/vp9_common.h"
14 15 16
#include "vp9/common/vp9_header.h"
#include "vp9/common/vp9_reconintra.h"
#include "vp9/common/vp9_reconinter.h"
Yaowu Xu's avatar
Yaowu Xu committed
17
#include "vp9/common/vp9_entropy.h"
18
#include "vp9/decoder/vp9_decodframe.h"
19
#include "vp9/decoder/vp9_detokenize.h"
20 21 22 23
#include "vp9/common/vp9_invtrans.h"
#include "vp9/common/vp9_alloccommon.h"
#include "vp9/common/vp9_entropymode.h"
#include "vp9/common/vp9_quant_common.h"
Johann's avatar
Johann committed
24
#include "vpx_scale/vpx_scale.h"
25
#include "vp9/common/vp9_setupintrarecon.h"
Scott LaVarnway's avatar
Scott LaVarnway committed
26

27
#include "vp9/decoder/vp9_decodemv.h"
28 29
#include "vp9/common/vp9_extend.h"
#include "vp9/common/vp9_modecont.h"
John Koleszar's avatar
John Koleszar committed
30
#include "vpx_mem/vpx_mem.h"
31
#include "vp9/decoder/vp9_dboolhuff.h"
John Koleszar's avatar
John Koleszar committed
32

33
#include "vp9/common/vp9_seg_common.h"
34
#include "vp9/common/vp9_tile_common.h"
35
#include "vp9_rtcd.h"
36

John Koleszar's avatar
John Koleszar committed
37 38 39
#include <assert.h>
#include <stdio.h>

40 41
#define COEFCOUNT_TESTING

42
// #define DEC_DEBUG
43 44 45 46
#ifdef DEC_DEBUG
int dec_debug = 0;
#endif

47 48 49 50 51 52 53 54 55 56 57 58 59 60
static int read_le16(const uint8_t *p) {
  return (p[1] << 8) | p[0];
}

static int read_le32(const uint8_t *p) {
  return (p[3] << 24) | (p[2] << 16) | (p[1] << 8) | p[0];
}

// len == 0 is not allowed
static int read_is_valid(const unsigned char *start, size_t len,
                         const unsigned char *end) {
  return start + len > start && start + len <= end;
}

61 62 63 64 65 66 67
static TXFM_MODE read_txfm_mode(vp9_reader *r) {
  TXFM_MODE mode = vp9_read_literal(r, 2);
  if (mode == ALLOW_32X32)
    mode += vp9_read_bit(r);
  return mode;
}

John Koleszar's avatar
John Koleszar committed
68 69 70 71 72 73 74 75 76 77 78 79
static int merge_index(int v, int n, int modulus) {
  int max1 = (n - 1 - modulus / 2) / modulus + 1;
  if (v < max1) v = v * modulus + modulus / 2;
  else {
    int w;
    v -= max1;
    w = v;
    v += (v + modulus - modulus / 2) / modulus;
    while (v % modulus == modulus / 2 ||
           w != v - (v + modulus - modulus / 2) / modulus) v++;
  }
  return v;
80 81
}

John Koleszar's avatar
John Koleszar committed
82 83 84
static int inv_remap_prob(int v, int m) {
  const int n = 256;
  const int modulus = MODULUS_PARAM;
85

John Koleszar's avatar
John Koleszar committed
86 87
  v = merge_index(v, n - 1, modulus);
  if ((m << 1) <= n) {
88
    return vp9_inv_recenter_nonneg(v + 1, m);
John Koleszar's avatar
John Koleszar committed
89
  } else {
90
    return n - 1 - vp9_inv_recenter_nonneg(v + 1, n - 1 - m);
John Koleszar's avatar
John Koleszar committed
91
  }
92
}
93

94
static vp9_prob read_prob_diff_update(vp9_reader *const bc, int oldp) {
95
  int delp = vp9_decode_term_subexp(bc, SUBEXP_PARAM, 255);
96
  return (vp9_prob)inv_remap_prob(delp, oldp);
97
}
98

99
void vp9_init_de_quantizer(VP9D_COMP *pbi) {
John Koleszar's avatar
John Koleszar committed
100
  int i;
Dmitry Kovalev's avatar
Dmitry Kovalev committed
101
  int q;
102
  VP9_COMMON *const pc = &pbi->common;
John Koleszar's avatar
John Koleszar committed
103

Dmitry Kovalev's avatar
Dmitry Kovalev committed
104 105 106
  for (q = 0; q < QINDEX_RANGE; q++) {
    pc->Y1dequant[q][0] = (int16_t)vp9_dc_quant(q, pc->y1dc_delta_q);
    pc->UVdequant[q][0] = (int16_t)vp9_dc_uv_quant(q, pc->uvdc_delta_q);
John Koleszar's avatar
John Koleszar committed
107 108 109

    /* all the ac values =; */
    for (i = 1; i < 16; i++) {
110
      const int rc = vp9_default_zig_zag1d_4x4[i];
John Koleszar's avatar
John Koleszar committed
111

Dmitry Kovalev's avatar
Dmitry Kovalev committed
112 113
      pc->Y1dequant[q][rc] = (int16_t)vp9_ac_yquant(q);
      pc->UVdequant[q][rc] = (int16_t)vp9_ac_uv_quant(q, pc->uvac_delta_q);
John Koleszar's avatar
John Koleszar committed
114
    }
John Koleszar's avatar
John Koleszar committed
115
  }
John Koleszar's avatar
John Koleszar committed
116 117
}

Dmitry Kovalev's avatar
Dmitry Kovalev committed
118 119 120 121 122 123 124 125 126 127 128 129 130 131
static int get_qindex(MACROBLOCKD *mb, int segment_id, int base_qindex) {
  // Set the Q baseline allowing for any segment level adjustment
  if (vp9_segfeature_active(mb, segment_id, SEG_LVL_ALT_Q)) {
    if (mb->mb_segment_abs_delta == SEGMENT_ABSDATA)
      return vp9_get_segdata(mb, segment_id, SEG_LVL_ALT_Q);  // Abs Value
    else
      return clamp(base_qindex + vp9_get_segdata(mb, segment_id, SEG_LVL_ALT_Q),
                   0, MAXQ);  // Delta Value
  } else {
    return base_qindex;
  }
}

static void mb_init_dequantizer(VP9D_COMP *pbi, MACROBLOCKD *mb) {
John Koleszar's avatar
John Koleszar committed
132 133
  int i;

Dmitry Kovalev's avatar
Dmitry Kovalev committed
134
  VP9_COMMON *const pc = &pbi->common;
135 136
  const int segment_id = mb->mode_info_context->mbmi.segment_id;
  const int qindex = get_qindex(mb, segment_id, pc->base_qindex);
Dmitry Kovalev's avatar
Dmitry Kovalev committed
137
  mb->q_index = qindex;
Dmitry Kovalev's avatar
Dmitry Kovalev committed
138

Dmitry Kovalev's avatar
Dmitry Kovalev committed
139 140
  for (i = 0; i < 16; i++)
    mb->block[i].dequant = pc->Y1dequant[qindex];
John Koleszar's avatar
John Koleszar committed
141

Dmitry Kovalev's avatar
Dmitry Kovalev committed
142 143
  for (i = 16; i < 24; i++)
    mb->block[i].dequant = pc->UVdequant[qindex];
John Koleszar's avatar
John Koleszar committed
144

Dmitry Kovalev's avatar
Dmitry Kovalev committed
145
  if (mb->lossless) {
Dmitry Kovalev's avatar
Dmitry Kovalev committed
146
    assert(qindex == 0);
Yaowu Xu's avatar
Yaowu Xu committed
147 148
    mb->inv_txm4x4_1      = vp9_short_iwalsh4x4_1;
    mb->inv_txm4x4        = vp9_short_iwalsh4x4;
Dmitry Kovalev's avatar
Dmitry Kovalev committed
149 150 151 152
    mb->itxm_add          = vp9_dequant_idct_add_lossless_c;
    mb->itxm_add_y_block  = vp9_dequant_idct_add_y_block_lossless_c;
    mb->itxm_add_uv_block = vp9_dequant_idct_add_uv_block_lossless_c;
  } else {
Yaowu Xu's avatar
Yaowu Xu committed
153 154
    mb->inv_txm4x4_1      = vp9_short_idct4x4_1;
    mb->inv_txm4x4        = vp9_short_idct4x4;
Dmitry Kovalev's avatar
Dmitry Kovalev committed
155 156 157
    mb->itxm_add          = vp9_dequant_idct_add;
    mb->itxm_add_y_block  = vp9_dequant_idct_add_y_block;
    mb->itxm_add_uv_block = vp9_dequant_idct_add_uv_block;
John Koleszar's avatar
John Koleszar committed
158
  }
John Koleszar's avatar
John Koleszar committed
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
#if CONFIG_CODE_NONZEROCOUNT
static void propagate_nzcs(VP9_COMMON *cm, MACROBLOCKD *xd) {
  MODE_INFO *m = xd->mode_info_context;
  BLOCK_SIZE_TYPE sb_type = m->mbmi.sb_type;
  const int mis = cm->mode_info_stride;
  int n;
  if (sb_type == BLOCK_SIZE_SB64X64) {
    for (n = 0; n < 16; ++n) {
      int i = n >> 2;
      int j = n & 3;
      if (i == 0 && j == 0) continue;
      vpx_memcpy((m + j + mis * i)->mbmi.nzcs, m->mbmi.nzcs,
                 384 * sizeof(m->mbmi.nzcs[0]));
    }
  } else if (sb_type == BLOCK_SIZE_SB32X32) {
    for (n = 0; n < 4; ++n) {
      int i = n >> 1;
      int j = n & 1;
      if (i == 0 && j == 0) continue;
      vpx_memcpy((m + j + mis * i)->mbmi.nzcs, m->mbmi.nzcs,
                 384 * sizeof(m->mbmi.nzcs[0]));
    }
  }
}
#endif

187 188 189
/* skip_recon_mb() is Modified: Instead of writing the result to predictor buffer and then copying it
 *  to dst buffer, we can write the result directly to dst buffer. This eliminates unnecessary copy.
 */
190 191
static void skip_recon_mb(VP9D_COMP *pbi, MACROBLOCKD *xd,
                          int mb_row, int mb_col) {
192 193
  MODE_INFO *m = xd->mode_info_context;
  BLOCK_SIZE_TYPE sb_type = m->mbmi.sb_type;
Dmitry Kovalev's avatar
Dmitry Kovalev committed
194

John Koleszar's avatar
John Koleszar committed
195
  if (xd->mode_info_context->mbmi.ref_frame == INTRA_FRAME) {
Dmitry Kovalev's avatar
Dmitry Kovalev committed
196
    if (sb_type == BLOCK_SIZE_SB64X64) {
Ronald S. Bultje's avatar
Ronald S. Bultje committed
197 198
      vp9_build_intra_predictors_sb64uv_s(xd);
      vp9_build_intra_predictors_sb64y_s(xd);
Dmitry Kovalev's avatar
Dmitry Kovalev committed
199
    } else if (sb_type == BLOCK_SIZE_SB32X32) {
200 201
      vp9_build_intra_predictors_sbuv_s(xd);
      vp9_build_intra_predictors_sby_s(xd);
202
    } else {
Ronald S. Bultje's avatar
Ronald S. Bultje committed
203 204
      vp9_build_intra_predictors_mbuv_s(xd);
      vp9_build_intra_predictors_mby_s(xd);
Ronald S. Bultje's avatar
Ronald S. Bultje committed
205
    }
John Koleszar's avatar
John Koleszar committed
206
  } else {
Dmitry Kovalev's avatar
Dmitry Kovalev committed
207
    if (sb_type == BLOCK_SIZE_SB64X64) {
208
      vp9_build_inter64x64_predictors_sb(xd, mb_row, mb_col);
Dmitry Kovalev's avatar
Dmitry Kovalev committed
209
    } else if (sb_type == BLOCK_SIZE_SB32X32) {
210
      vp9_build_inter32x32_predictors_sb(xd, mb_row, mb_col);
211
    } else {
212 213 214 215 216
      vp9_build_inter16x16_predictors_mb(xd,
                                         xd->dst.y_buffer,
                                         xd->dst.u_buffer,
                                         xd->dst.v_buffer,
                                         xd->dst.y_stride,
217 218
                                         xd->dst.uv_stride,
                                         mb_row, mb_col);
Ronald S. Bultje's avatar
Ronald S. Bultje committed
219
    }
John Koleszar's avatar
John Koleszar committed
220
  }
221 222 223 224
#if CONFIG_CODE_NONZEROCOUNT
  vpx_memset(m->mbmi.nzcs, 0, 384 * sizeof(m->mbmi.nzcs[0]));
  propagate_nzcs(&pbi->common, xd);
#endif
John Koleszar's avatar
John Koleszar committed
225 226
}

227 228
static void decode_16x16(VP9D_COMP *pbi, MACROBLOCKD *xd,
                         BOOL_DECODER* const bc) {
229
  const TX_TYPE tx_type = get_tx_type_16x16(xd, 0);
230
#if 0  // def DEC_DEBUG
231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247
  if (dec_debug) {
    int i;
    printf("\n");
    printf("qcoeff 16x16\n");
    for (i = 0; i < 400; i++) {
      printf("%3d ", xd->qcoeff[i]);
      if (i % 16 == 15) printf("\n");
    }
    printf("\n");
    printf("predictor\n");
    for (i = 0; i < 400; i++) {
      printf("%3d ", xd->predictor[i]);
      if (i % 16 == 15) printf("\n");
    }
  }
#endif
  if (tx_type != DCT_DCT) {
248
    vp9_ht_dequant_idct_add_16x16_c(tx_type, xd->plane[0].qcoeff,
249
                                    xd->block[0].dequant, xd->predictor,
Scott LaVarnway's avatar
Scott LaVarnway committed
250
                                    xd->dst.y_buffer, 16, xd->dst.y_stride,
251
                                    xd->eobs[0]);
252
  } else {
253
    vp9_dequant_idct_add_16x16(xd->plane[0].qcoeff, xd->block[0].dequant,
254
                               xd->predictor, xd->dst.y_buffer,
255
                               16, xd->dst.y_stride, xd->eobs[0]);
256
  }
257 258 259 260 261 262 263 264

  vp9_dequant_idct_add_8x8(xd->plane[1].qcoeff, xd->block[16].dequant,
                           xd->predictor + 16 * 16, xd->dst.u_buffer, 8,
                           xd->dst.uv_stride, xd->eobs[16]);

  vp9_dequant_idct_add_8x8(xd->plane[2].qcoeff, xd->block[16].dequant,
                           xd->predictor + 16 * 16 + 64, xd->dst.v_buffer, 8,
                           xd->dst.uv_stride, xd->eobs[20]);
265 266 267 268 269 270
}

static void decode_8x8(VP9D_COMP *pbi, MACROBLOCKD *xd,
                       BOOL_DECODER* const bc) {
  // First do Y
  // if the first one is DCT_DCT assume all the rest are as well
271
  TX_TYPE tx_type = get_tx_type_8x8(xd, 0);
272
#if 0  // def DEC_DEBUG
273 274 275 276
  if (dec_debug) {
    int i;
    printf("\n");
    printf("qcoeff 8x8\n");
277
    for (i = 0; i < 384; i++) {
278 279 280 281 282 283 284 285 286 287
      printf("%3d ", xd->qcoeff[i]);
      if (i % 16 == 15) printf("\n");
    }
  }
#endif
  if (tx_type != DCT_DCT || xd->mode_info_context->mbmi.mode == I8X8_PRED) {
    int i;
    for (i = 0; i < 4; i++) {
      int ib = vp9_i8x8_block[i];
      int idx = (ib & 0x02) ? (ib + 2) : ib;
288
      int16_t *q  = BLOCK_OFFSET(xd->plane[0].qcoeff, idx, 16);
289 290 291
      int16_t *dq = xd->block[0].dequant;
      uint8_t *pre = xd->block[ib].predictor;
      uint8_t *dst = *(xd->block[ib].base_dst) + xd->block[ib].dst;
292 293 294 295
      int stride = xd->dst.y_stride;
      BLOCKD *b = &xd->block[ib];
      if (xd->mode_info_context->mbmi.mode == I8X8_PRED) {
        int i8x8mode = b->bmi.as_mode.first;
296
        vp9_intra8x8_predict(xd, b, i8x8mode, b->predictor);
297
      }
298
      tx_type = get_tx_type_8x8(xd, ib);
299
      if (tx_type != DCT_DCT) {
Scott LaVarnway's avatar
Scott LaVarnway committed
300
        vp9_ht_dequant_idct_add_8x8_c(tx_type, q, dq, pre, dst, 16, stride,
301
                                      xd->eobs[idx]);
302 303
      } else {
        vp9_dequant_idct_add_8x8_c(q, dq, pre, dst, 16, stride,
304
                                   xd->eobs[idx]);
305 306
      }
    }
307
  } else {
308
    vp9_dequant_idct_add_y_block_8x8(xd->plane[0].qcoeff,
309 310 311 312
                                     xd->block[0].dequant,
                                     xd->predictor,
                                     xd->dst.y_buffer,
                                     xd->dst.y_stride,
313
                                     xd);
314 315 316 317 318 319 320 321 322
  }

  // Now do UV
  if (xd->mode_info_context->mbmi.mode == I8X8_PRED) {
    int i;
    for (i = 0; i < 4; i++) {
      int ib = vp9_i8x8_block[i];
      BLOCKD *b = &xd->block[ib];
      int i8x8mode = b->bmi.as_mode.first;
323

324
      b = &xd->block[16 + i];
325
      vp9_intra_uv4x4_predict(xd, b, i8x8mode, b->predictor);
326 327
      xd->itxm_add(BLOCK_OFFSET(xd->plane[1].qcoeff, i, 16),
                   b->dequant, b->predictor,
328
                   *(b->base_dst) + b->dst, 8, b->dst_stride, xd->eobs[16 + i]);
329

330
      b = &xd->block[20 + i];
331
      vp9_intra_uv4x4_predict(xd, b, i8x8mode, b->predictor);
332 333
      xd->itxm_add(BLOCK_OFFSET(xd->plane[2].qcoeff, i, 16),
                   b->dequant, b->predictor,
334
                   *(b->base_dst) + b->dst, 8, b->dst_stride, xd->eobs[20 + i]);
335 336
    }
  } else if (xd->mode_info_context->mbmi.mode == SPLITMV) {
337 338 339 340 341 342
    xd->itxm_add_uv_block(xd->plane[1].qcoeff, xd->block[16].dequant,
         xd->predictor + 16 * 16, xd->dst.u_buffer,
         xd->dst.uv_stride, xd->eobs + 16);
    xd->itxm_add_uv_block(xd->plane[2].qcoeff, xd->block[16].dequant,
         xd->predictor + 16 * 16 + 64, xd->dst.v_buffer,
         xd->dst.uv_stride, xd->eobs + 20);
343
  } else {
344 345 346 347 348 349 350
    vp9_dequant_idct_add_8x8(xd->plane[1].qcoeff, xd->block[16].dequant,
                             xd->predictor + 16 * 16, xd->dst.u_buffer, 8,
                             xd->dst.uv_stride, xd->eobs[16]);

    vp9_dequant_idct_add_8x8(xd->plane[2].qcoeff, xd->block[16].dequant,
                             xd->predictor + 16 * 16 + 64, xd->dst.v_buffer, 8,
                             xd->dst.uv_stride, xd->eobs[20]);
351
  }
352
#if 0  // def DEC_DEBUG
353 354 355 356 357 358 359 360 361 362 363 364 365 366 367
  if (dec_debug) {
    int i;
    printf("\n");
    printf("predictor\n");
    for (i = 0; i < 384; i++) {
      printf("%3d ", xd->predictor[i]);
      if (i % 16 == 15) printf("\n");
    }
  }
#endif
}

static void decode_4x4(VP9D_COMP *pbi, MACROBLOCKD *xd,
                       BOOL_DECODER* const bc) {
  TX_TYPE tx_type;
368
  int i = 0;
369
  MB_PREDICTION_MODE mode = xd->mode_info_context->mbmi.mode;
370 371 372 373 374 375 376 377 378 379 380
#if 0  // def DEC_DEBUG
  if (dec_debug) {
    int i;
    printf("\n");
    printf("predictor\n");
    for (i = 0; i < 384; i++) {
      printf("%3d ", xd->predictor[i]);
      if (i % 16 == 15) printf("\n");
    }
  }
#endif
381 382 383 384 385
  if (mode == I8X8_PRED) {
    for (i = 0; i < 4; i++) {
      int ib = vp9_i8x8_block[i];
      const int iblock[4] = {0, 1, 4, 5};
      int j;
Dmitry Kovalev's avatar
Dmitry Kovalev committed
386 387
      BLOCKD *b = &xd->block[ib];
      int i8x8mode = b->bmi.as_mode.first;
388
      vp9_intra8x8_predict(xd, b, i8x8mode, b->predictor);
389 390
      for (j = 0; j < 4; j++) {
        b = &xd->block[ib + iblock[j]];
391
        tx_type = get_tx_type_4x4(xd, ib + iblock[j]);
392
        if (tx_type != DCT_DCT) {
393 394
          vp9_ht_dequant_idct_add_c(tx_type,
              BLOCK_OFFSET(xd->plane[0].qcoeff, ib + iblock[j], 16),
395 396
                                    b->dequant, b->predictor,
                                    *(b->base_dst) + b->dst, 16,
397
                                    b->dst_stride, xd->eobs[ib + iblock[j]]);
398
        } else {
399 400
          xd->itxm_add(BLOCK_OFFSET(xd->plane[0].qcoeff, ib + iblock[j], 16),
                       b->dequant, b->predictor,
401 402
                       *(b->base_dst) + b->dst, 16, b->dst_stride,
                       xd->eobs[ib + iblock[j]]);
403 404 405
        }
      }
      b = &xd->block[16 + i];
406
      vp9_intra_uv4x4_predict(xd, b, i8x8mode, b->predictor);
407 408
      xd->itxm_add(BLOCK_OFFSET(xd->plane[1].qcoeff, i, 16),
                   b->dequant, b->predictor,
409
                   *(b->base_dst) + b->dst, 8, b->dst_stride, xd->eobs[16 + i]);
410
      b = &xd->block[20 + i];
411
      vp9_intra_uv4x4_predict(xd, b, i8x8mode, b->predictor);
412 413
      xd->itxm_add(BLOCK_OFFSET(xd->plane[2].qcoeff, i, 16),
                   b->dequant, b->predictor,
414
                   *(b->base_dst) + b->dst, 8, b->dst_stride, xd->eobs[20 + i]);
415 416 417 418
    }
  } else if (mode == B_PRED) {
    for (i = 0; i < 16; i++) {
      BLOCKD *b = &xd->block[i];
Dmitry Kovalev's avatar
Dmitry Kovalev committed
419
      int b_mode = xd->mode_info_context->bmi[i].as_mode.first;
420 421
#if CONFIG_NEWBINTRAMODES
      xd->mode_info_context->bmi[i].as_mode.context = b->bmi.as_mode.context =
422
          vp9_find_bpred_context(xd, b);
423 424
      if (!xd->mode_info_context->mbmi.mb_skip_coeff)
        vp9_decode_coefs_4x4(pbi, xd, bc, PLANE_TYPE_Y_WITH_DC, i);
425
#endif
426
      vp9_intra4x4_predict(xd, b, b_mode, b->predictor);
427
      tx_type = get_tx_type_4x4(xd, i);
428
      if (tx_type != DCT_DCT) {
429 430
        vp9_ht_dequant_idct_add_c(tx_type,
                                  BLOCK_OFFSET(xd->plane[0].qcoeff, i, 16),
431
                                  b->dequant, b->predictor,
Scott LaVarnway's avatar
Scott LaVarnway committed
432
                                  *(b->base_dst) + b->dst, 16, b->dst_stride,
433
                                  xd->eobs[i]);
434
      } else {
435 436
        xd->itxm_add(BLOCK_OFFSET(xd->plane[0].qcoeff, i, 16),
                     b->dequant, b->predictor,
437
                      *(b->base_dst) + b->dst, 16, b->dst_stride, xd->eobs[i]);
438 439
      }
    }
440 441 442 443
#if CONFIG_NEWBINTRAMODES
    if (!xd->mode_info_context->mbmi.mb_skip_coeff)
      vp9_decode_mb_tokens_4x4_uv(pbi, xd, bc);
#endif
444
    vp9_build_intra_predictors_mbuv(xd);
445 446 447 448 449 450
    xd->itxm_add_uv_block(xd->plane[1].qcoeff, xd->block[16].dequant,
         xd->predictor + 16 * 16, xd->dst.u_buffer,
         xd->dst.uv_stride, xd->eobs + 16);
    xd->itxm_add_uv_block(xd->plane[2].qcoeff, xd->block[16].dequant,
         xd->predictor + 16 * 16 + 64, xd->dst.v_buffer,
         xd->dst.uv_stride, xd->eobs + 20);
451
  } else if (mode == SPLITMV || get_tx_type_4x4(xd, 0) == DCT_DCT) {
452
    xd->itxm_add_y_block(xd->plane[0].qcoeff,
453 454 455 456
                          xd->block[0].dequant,
                          xd->predictor,
                          xd->dst.y_buffer,
                          xd->dst.y_stride,
457
                          xd);
458 459 460 461 462 463
    xd->itxm_add_uv_block(xd->plane[1].qcoeff, xd->block[16].dequant,
         xd->predictor + 16 * 16, xd->dst.u_buffer,
         xd->dst.uv_stride, xd->eobs + 16);
    xd->itxm_add_uv_block(xd->plane[2].qcoeff, xd->block[16].dequant,
         xd->predictor + 16 * 16 + 64, xd->dst.v_buffer,
         xd->dst.uv_stride, xd->eobs + 20);
464
  } else {
465
#if 0  // def DEC_DEBUG
466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481
    if (dec_debug) {
      int i;
      printf("\n");
      printf("qcoeff 4x4\n");
      for (i = 0; i < 400; i++) {
        printf("%3d ", xd->qcoeff[i]);
        if (i % 16 == 15) printf("\n");
      }
      printf("\n");
      printf("predictor\n");
      for (i = 0; i < 400; i++) {
        printf("%3d ", xd->predictor[i]);
        if (i % 16 == 15) printf("\n");
      }
    }
#endif
482 483
    for (i = 0; i < 16; i++) {
      BLOCKD *b = &xd->block[i];
484
      tx_type = get_tx_type_4x4(xd, i);
485
      if (tx_type != DCT_DCT) {
486 487
        vp9_ht_dequant_idct_add_c(tx_type,
                                  BLOCK_OFFSET(xd->plane[0].qcoeff, i, 16),
488 489
                                  b->dequant, b->predictor,
                                  *(b->base_dst) + b->dst, 16,
490
                                  b->dst_stride, xd->eobs[i]);
491
      } else {
492 493
        xd->itxm_add(BLOCK_OFFSET(xd->plane[0].qcoeff, i, 16),
                     b->dequant, b->predictor,
494
                      *(b->base_dst) + b->dst, 16, b->dst_stride, xd->eobs[i]);
495 496
      }
    }
497 498 499 500 501 502
    xd->itxm_add_uv_block(xd->plane[1].qcoeff, xd->block[16].dequant,
                          xd->predictor + 16 * 16, xd->dst.u_buffer,
                          xd->dst.uv_stride, xd->eobs + 16);
    xd->itxm_add_uv_block(xd->plane[2].qcoeff, xd->block[16].dequant,
                          xd->predictor + 16 * 16 + 64, xd->dst.v_buffer,
                          xd->dst.uv_stride, xd->eobs + 20);
503 504 505
  }
}

506
static void decode_sb_16x16(MACROBLOCKD *mb, int y_size) {
507 508 509 510 511 512 513 514 515 516 517 518
  const int y_count = y_size * y_size;
  const int uv_size = y_size / 2;
  const int uv_count = uv_size * uv_size;

  const int u_eob_offset = 16 * y_count;
  const int v_eob_offset = u_eob_offset + 16 * uv_count;
  int n;

  for (n = 0; n < y_count; n++) {
    const int x_idx = n % y_size;
    const int y_idx = n / y_size;
    const int y_offset = (y_idx * 16) * mb->dst.y_stride + (x_idx * 16);
519 520
    const TX_TYPE tx_type = get_tx_type_16x16(mb,
                                (y_idx * (4 * y_size) + x_idx) * 4);
521
    if (tx_type == DCT_DCT) {
522
      vp9_dequant_idct_add_16x16(BLOCK_OFFSET(mb->plane[0].qcoeff, n, 256),
523 524 525 526 527 528 529
                                 mb->block[0].dequant ,
                                 mb->dst.y_buffer + y_offset,
                                 mb->dst.y_buffer + y_offset,
                                 mb->dst.y_stride, mb->dst.y_stride,
                                 mb->eobs[n * 16]);
    } else {
      vp9_ht_dequant_idct_add_16x16_c(tx_type,
530
                                      BLOCK_OFFSET(mb->plane[0].qcoeff, n, 256),
531 532 533 534 535 536 537 538 539 540 541 542
                                      mb->block[0].dequant,
                                      mb->dst.y_buffer + y_offset,
                                      mb->dst.y_buffer + y_offset,
                                      mb->dst.y_stride, mb->dst.y_stride,
                                      mb->eobs[n * 16]);
    }
  }

  for (n = 0; n < uv_count; n++) {
    const int x_idx = n % uv_size;
    const int y_idx = n / uv_size;
    const int uv_offset = (y_idx * 16) * mb->dst.uv_stride + (x_idx * 16);
543
    vp9_dequant_idct_add_16x16(BLOCK_OFFSET(mb->plane[1].qcoeff, n, 256),
544 545 546 547 548
                               mb->block[16].dequant,
                               mb->dst.u_buffer + uv_offset,
                               mb->dst.u_buffer + uv_offset,
                               mb->dst.uv_stride, mb->dst.uv_stride,
                               mb->eobs[u_eob_offset + n * 16]);
549
    vp9_dequant_idct_add_16x16(BLOCK_OFFSET(mb->plane[2].qcoeff, n, 256),
550 551 552 553 554 555 556 557
                               mb->block[20].dequant,
                               mb->dst.v_buffer + uv_offset,
                               mb->dst.v_buffer + uv_offset,
                               mb->dst.uv_stride, mb->dst.uv_stride,
                               mb->eobs[v_eob_offset + n * 16]);
  }
}

558
static INLINE void decode_sb_8x8(MACROBLOCKD *xd, int y_size) {
559 560 561 562 563 564 565 566 567 568 569 570
  const int y_count = y_size * y_size;
  const int uv_size = y_size / 2;
  const int uv_count = uv_size * uv_size;

  const int u_eob_offset = 4 * y_count;
  const int v_eob_offset = u_eob_offset + 4 * uv_count;
  int n;

  // luma
  for (n = 0; n < y_count; n++) {
    const int x_idx = n % y_size;
    const int y_idx = n / y_size;
571 572
    const int y_offset = (y_idx * 8) * xd->dst.y_stride + (x_idx * 8);
    const TX_TYPE tx_type = get_tx_type_8x8(xd,
573
                                            (y_idx * (2 * y_size) + x_idx) * 2);
574
    if (tx_type == DCT_DCT) {
575 576 577 578 579 580
      vp9_dequant_idct_add_8x8_c(BLOCK_OFFSET(xd->plane[0].qcoeff, n, 64),
                                 xd->block[0].dequant,
                                 xd->dst.y_buffer + y_offset,
                                 xd->dst.y_buffer + y_offset,
                                 xd->dst.y_stride, xd->dst.y_stride,
                                 xd->eobs[n * 4]);
581
    } else {
582
      vp9_ht_dequant_idct_add_8x8_c(tx_type,
583 584 585 586 587 588
                                    BLOCK_OFFSET(xd->plane[0].qcoeff, n, 64),
                                    xd->block[0].dequant,
                                    xd->dst.y_buffer + y_offset,
                                    xd->dst.y_buffer + y_offset,
                                    xd->dst.y_stride, xd->dst.y_stride,
                                    xd->eobs[n * 4]);
589 590 591 592 593 594 595
    }
  }

  // chroma
  for (n = 0; n < uv_count; n++) {
    const int x_idx = n % uv_size;
    const int y_idx = n / uv_size;
596 597 598 599 600 601 602 603 604 605 606 607 608
    const int uv_offset = (y_idx * 8) * xd->dst.uv_stride + (x_idx * 8);
    vp9_dequant_idct_add_8x8_c(BLOCK_OFFSET(xd->plane[1].qcoeff, n, 64),
                               xd->block[16].dequant,
                               xd->dst.u_buffer + uv_offset,
                               xd->dst.u_buffer + uv_offset,
                               xd->dst.uv_stride, xd->dst.uv_stride,
                               xd->eobs[u_eob_offset + n * 4]);
    vp9_dequant_idct_add_8x8_c(BLOCK_OFFSET(xd->plane[2].qcoeff, n, 64),
                               xd->block[20].dequant,
                               xd->dst.v_buffer + uv_offset,
                               xd->dst.v_buffer + uv_offset,
                               xd->dst.uv_stride, xd->dst.uv_stride,
                               xd->eobs[v_eob_offset + n * 4]);
609 610 611 612
  }
}


613
static void decode_sb_4x4(MACROBLOCKD *xd, int y_size) {
614 615 616 617 618 619 620 621 622 623 624
  const int y_count = y_size * y_size;
  const int uv_size = y_size / 2;
  const int uv_count = uv_size * uv_size;

  const int u_eob_offset = y_count;
  const int v_eob_offset = u_eob_offset + uv_count;
  int n;

  for (n = 0; n < y_count; n++) {
    const int x_idx = n % y_size;
    const int y_idx = n / y_size;
625 626
    const int y_offset = (y_idx * 4) * xd->dst.y_stride + (x_idx * 4);
    const TX_TYPE tx_type = get_tx_type_4x4(xd, y_idx * y_size + x_idx);
627
    if (tx_type == DCT_DCT) {
628 629 630 631 632 633
      xd->itxm_add(BLOCK_OFFSET(xd->plane[0].qcoeff, n, 16),
                   xd->block[0].dequant,
                   xd->dst.y_buffer + y_offset,
                   xd->dst.y_buffer + y_offset,
                   xd->dst.y_stride, xd->dst.y_stride,
                   xd->eobs[n]);
634
    } else {
635
      vp9_ht_dequant_idct_add_c(tx_type,
636 637 638 639 640 641 642
                                BLOCK_OFFSET(xd->plane[0].qcoeff, n, 16),
                                xd->block[0].dequant,
                                xd->dst.y_buffer + y_offset,
                                xd->dst.y_buffer + y_offset,
                                xd->dst.y_stride,
                                xd->dst.y_stride,
                                xd->eobs[n]);
643 644 645 646 647 648
    }
  }

  for (n = 0; n < uv_count; n++) {
    const int x_idx = n % uv_size;
    const int y_idx = n / uv_size;
649 650 651 652 653 654 655 656 657 658 659
    const int uv_offset = (y_idx * 4) * xd->dst.uv_stride + (x_idx * 4);
    xd->itxm_add(BLOCK_OFFSET(xd->plane[1].qcoeff, n, 16),
        xd->block[16].dequant,
        xd->dst.u_buffer + uv_offset,
        xd->dst.u_buffer + uv_offset,
        xd->dst.uv_stride, xd->dst.uv_stride, xd->eobs[u_eob_offset + n]);
    xd->itxm_add(BLOCK_OFFSET(xd->plane[2].qcoeff, n, 16),
        xd->block[20].dequant,
        xd->dst.v_buffer + uv_offset,
        xd->dst.v_buffer + uv_offset,
        xd->dst.uv_stride, xd->dst.uv_stride, xd->eobs[v_eob_offset + n]);
660 661 662
  }
}

663 664
static void decode_sb64(VP9D_COMP *pbi, MACROBLOCKD *xd, int mb_row, int mb_col,
                        BOOL_DECODER* const bc) {
665
  int n, eobtotal;
666
  VP9_COMMON *const pc = &pbi->common;
667
  MODE_INFO *mi = xd->mode_info_context;
668
  const int mis = pc->mode_info_stride;
669

670
  assert(mi->mbmi.sb_type == BLOCK_SIZE_SB64X64);
Ronald S. Bultje's avatar
Ronald S. Bultje committed
671 672

  if (pbi->common.frame_type != KEY_FRAME)
673
    vp9_setup_interp_filters(xd, mi->mbmi.interp_filter, pc);
Ronald S. Bultje's avatar
Ronald S. Bultje committed
674 675 676 677 678

  // re-initialize macroblock dequantizer before detokenization
  if (xd->segmentation_enabled)
    mb_init_dequantizer(pbi, xd);

679
  if (mi->mbmi.mb_skip_coeff) {
680
    vp9_reset_sb64_tokens_context(xd);
Ronald S. Bultje's avatar
Ronald S. Bultje committed
681

682 683
    // Special case:  Force the loopfilter to skip when eobtotal and
    // mb_skip_coeff are zero.
684
    skip_recon_mb(pbi, xd, mb_row, mb_col);
Ronald S. Bultje's avatar
Ronald S. Bultje committed
685 686 687
    return;
  }

688
  // do prediction
Ronald S. Bultje's avatar
Ronald S. Bultje committed
689 690 691 692
  if (xd->mode_info_context->mbmi.ref_frame == INTRA_FRAME) {
    vp9_build_intra_predictors_sb64y_s(xd);
    vp9_build_intra_predictors_sb64uv_s(xd);
  } else {
693
    vp9_build_inter64x64_predictors_sb(xd, mb_row, mb_col);
Ronald S. Bultje's avatar
Ronald S. Bultje committed
694 695
  }

696
  // dequantization and idct
697 698
  eobtotal = vp9_decode_sb64_tokens(pbi, xd, bc);
  if (eobtotal == 0) {  // skip loopfilter
Ronald S. Bultje's avatar
Ronald S. Bultje committed
699
    for (n = 0; n < 16; n++) {
700
      const int x_idx = n & 3, y_idx = n >> 2;
Ronald S. Bultje's avatar
Ronald S. Bultje committed
701

702 703 704 705 706 707 708 709
      if (mb_col + x_idx < pc->mb_cols && mb_row + y_idx < pc->mb_rows)
        mi[y_idx * mis + x_idx].mbmi.mb_skip_coeff = mi->mbmi.mb_skip_coeff;
    }
  } else {
    switch (xd->mode_info_context->mbmi.txfm_size) {
      case TX_32X32:
        for (n = 0; n < 4; n++) {
          const int x_idx = n & 1, y_idx = n >> 1;
Dmitry Kovalev's avatar
Dmitry Kovalev committed
710
          const int y_offset = x_idx * 32 + y_idx * xd->dst.y_stride * 32;
711
          vp9_dequant_idct_add_32x32(BLOCK_OFFSET(xd->plane[0].qcoeff, n, 1024),
712
              xd->block[0].dequant,
Dmitry Kovalev's avatar
Dmitry Kovalev committed
713 714
              xd->dst.y_buffer + y_offset,
              xd->dst.y_buffer + y_offset,
715 716
              xd->dst.y_stride, xd->dst.y_stride, xd->eobs[n * 64]);
        }
717
        vp9_dequant_idct_add_32x32(xd->plane[1].qcoeff,
718 719
            xd->block[16].dequant, xd->dst.u_buffer, xd->dst.u_buffer,
            xd->dst.uv_stride, xd->dst.uv_stride, xd->eobs[256]);
720
        vp9_dequant_idct_add_32x32(xd->plane[2].qcoeff,
721 722 723
            xd->block[20].dequant, xd->dst.v_buffer, xd->dst.v_buffer,
            xd->dst.uv_stride, xd->dst.uv_stride, xd->eobs[320]);
        break;
724
      case TX_16X16:
725
        decode_sb_16x16(xd, 4);
726
        break;
727
      case TX_8X8:
728
        decode_sb_8x8(xd, 8);
729
        break;
730
      case TX_4X4:
731
        decode_sb_4x4(xd, 16);
732 733
        break;
      default: assert(0);
Ronald S. Bultje's avatar
Ronald S. Bultje committed
734 735
    }
  }
736 737 738
#if CONFIG_CODE_NONZEROCOUNT
  propagate_nzcs(&pbi->common, xd);
#endif
Ronald S. Bultje's avatar
Ronald S. Bultje committed
739 740
}

741 742
static void decode_sb32(VP9D_COMP *pbi, MACROBLOCKD *xd, int mb_row, int mb_col,
                        BOOL_DECODER* const bc) {
743
  int eobtotal;
Ronald S. Bultje's avatar
Ronald S. Bultje committed
744
  VP9_COMMON *const pc = &pbi->common;
745
  MODE_INFO *mi = xd->mode_info_context;
Ronald S. Bultje's avatar
Ronald S. Bultje committed
746 747
  const int mis = pc->mode_info_stride;

748
  assert(mi->mbmi.sb_type == BLOCK_SIZE_SB32X32);
749

750
  if (pbi->common.frame_type != KEY_FRAME)
751
    vp9_setup_interp_filters(xd, mi->mbmi.interp_filter, pc);
752

753 754 755 756
  // re-initialize macroblock dequantizer before detokenization
  if (xd->segmentation_enabled)
    mb_init_dequantizer(pbi, xd);

757
  if (mi->mbmi.mb_skip_coeff) {
758
    vp9_reset_sb_tokens_context(xd);
759

760 761
    // Special case:  Force the loopfilter to skip when eobtotal and
    // mb_skip_coeff are zero.
762
    skip_recon_mb(pbi, xd, mb_row, mb_col);
763 764 765
    return;
  }

766 767

  // do prediction
768
  if (mi->mbmi.ref_frame == INTRA_FRAME) {
769 770 771
    vp9_build_intra_predictors_sby_s(xd);
    vp9_build_intra_predictors_sbuv_s(xd);
  } else {
772
    vp9_build_inter32x32_predictors_sb(xd, mb_row, mb_col);
773 774
  }

775
  // dequantization and idct
776 777
  eobtotal = vp9_decode_sb_tokens(pbi, xd, bc);
  if (eobtotal == 0) {  // skip loopfilter
778
    mi->mbmi.mb_skip_coeff = 1;
779
    if (mb_col + 1 < pc->mb_cols)
780
      mi[1].mbmi.mb_skip_coeff = 1;
781
    if (mb_row + 1 < pc->mb_rows) {
782
      mi[mis].mbmi.mb_skip_coeff = 1;
783
      if (mb_col + 1 < pc->mb_cols)
784
        mi[mis + 1].mbmi.mb_skip_coeff = 1;
785
    }
Ronald S. Bultje's avatar
Ronald S. Bultje committed
786
  } else {
787 788
    switch (xd->mode_info_context->mbmi.txfm_size) {
      case TX_32X32:
789
        vp9_dequant_idct_add_32x32(xd->plane[0].qcoeff, xd->block[0].dequant,
790 791 792
                                   xd->dst.y_buffer, xd->dst.y_buffer,
                                   xd->dst.y_stride, xd->dst.y_stride,
                                   xd->eobs[0]);
793 794 795 796 797 798 799 800
        vp9_dequant_idct_add_16x16(xd->plane[1].qcoeff, xd->block[16].dequant,
                                   xd->dst.u_buffer, xd->dst.u_buffer,
                                   xd->dst.uv_stride, xd->dst.uv_stride,
                                   xd->eobs[64]);
        vp9_dequant_idct_add_16x16(xd->plane[2].qcoeff, xd->block[16].dequant,
                                   xd->dst.v_buffer, xd->dst.v_buffer,
                                   xd->dst.uv_stride, xd->dst.uv_stride,
                                   xd->eobs[80]);
801
        break;
802
      case TX_16X16:
803
        decode_sb_16x16(xd, 2);
804
        break;
805
      case TX_8X8:
806
        decode_sb_8x8(xd, 4);
807
        break;
808
      case TX_4X4:
809
        decode_sb_4x4(xd, 8);
810 811
        break;
      default: assert(0);
812
    }
813
  }
814 815 816
#if CONFIG_CODE_NONZEROCOUNT
  propagate_nzcs(&pbi->common, xd);
#endif
817 818
}

819 820 821
static void decode_mb(VP9D_COMP *pbi, MACROBLOCKD *xd,
                     int mb_row, int mb_col,
                     BOOL_DECODER* const bc) {
John Koleszar's avatar
John Koleszar committed
822
  int eobtotal = 0;
823 824
  const MB_PREDICTION_MODE mode = xd->mode_info_context->mbmi.mode;
  const int tx_size = xd->mode_info_context->mbmi.txfm_size;
825

Ronald S. Bultje's avatar
Ronald S. Bultje committed
826
  assert(!xd->mode_info_context->mbmi.sb_type);
John Koleszar's avatar
John Koleszar committed
827

Jingning Han's avatar
Jingning Han committed
828 829 830 831
  // re-initialize macroblock dequantizer before detokenization
  if (xd->segmentation_enabled)
    mb_init_dequantizer(pbi, xd);

John Koleszar's avatar
John Koleszar committed
832
  if (xd->mode_info_context->mbmi.mb_skip_coeff) {
833
    vp9_reset_mb_tokens_context(xd);
Ronald S. Bultje's avatar