vp9_decodframe.c 55.2 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
// #define DEC_DEBUG
41 42 43 44
#ifdef DEC_DEBUG
int dec_debug = 0;
#endif

45 46 47 48 49 50 51 52 53
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
54 55
static int read_is_valid(const uint8_t *start, size_t len,
                         const uint8_t *end) {
56 57 58
  return start + len > start && start + len <= end;
}

59 60 61 62 63 64 65 66 67 68 69 70 71 72
static void setup_txfm_mode(VP9_COMMON *pc, int lossless, vp9_reader *r) {
  if (lossless) {
    pc->txfm_mode = ONLY_4X4;
  } else {
    pc->txfm_mode = vp9_read_literal(r, 2);
    if (pc->txfm_mode == ALLOW_32X32)
      pc->txfm_mode += vp9_read_bit(r);

    if (pc->txfm_mode == TX_MODE_SELECT) {
      pc->prob_tx[0] = vp9_read_prob(r);
      pc->prob_tx[1] = vp9_read_prob(r);
      pc->prob_tx[2] = vp9_read_prob(r);
    }
  }
73 74
}

75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95
static int get_unsigned_bits(unsigned int num_values) {
  int cat = 0;
  if (num_values <= 1)
    return 0;
  num_values--;
  while (num_values > 0) {
    cat++;
    num_values >>= 1;
  }
  return cat;
}

static int inv_recenter_nonneg(int v, int m) {
  if (v > (m << 1))
    return v;
  else if ((v & 1) == 0)
    return (v >> 1) + m;
  else
    return m - ((v + 1) >> 1);
}

96
static int decode_uniform(vp9_reader *r, int n) {
97 98 99 100 101 102
  int v;
  const int l = get_unsigned_bits(n);
  const int m = (1 << l) - n;
  if (!l)
    return 0;

103 104
  v = vp9_read_literal(r, l - 1);
  return v < m ?  v : (v << 1) - m + vp9_read_bit(r);
105 106
}

107
static int decode_term_subexp(vp9_reader *r, int k, int num_syms) {
108 109 110 111 112
  int i = 0, mk = 0, word;
  while (1) {
    const int b = i ? k + i - 1 : k;
    const int a = 1 << b;
    if (num_syms <= mk + 3 * a) {
113
      word = decode_uniform(r, num_syms - mk) + mk;
114 115
      break;
    } else {
116
      if (vp9_read_bit(r)) {
117 118 119
        i++;
        mk += a;
      } else {
120
        word = vp9_read_literal(r, b) + mk;
121 122 123 124 125 126 127
        break;
      }
    }
  }
  return word;
}

128
static int decode_unsigned_max(vp9_reader *r, int max) {
129 130 131
  int data = 0, bit = 0, lmax = max;

  while (lmax) {
132
    data |= vp9_read_bit(r) << bit++;
133 134 135 136 137
    lmax >>= 1;
  }
  return data > max ? max : data;
}

John Koleszar's avatar
John Koleszar committed
138 139 140 141 142 143 144 145 146 147 148 149
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;
150 151
}

John Koleszar's avatar
John Koleszar committed
152 153 154
static int inv_remap_prob(int v, int m) {
  const int n = 256;
  const int modulus = MODULUS_PARAM;
155

John Koleszar's avatar
John Koleszar committed
156 157
  v = merge_index(v, n - 1, modulus);
  if ((m << 1) <= n) {
158
    return inv_recenter_nonneg(v + 1, m);
John Koleszar's avatar
John Koleszar committed
159
  } else {
160
    return n - 1 - inv_recenter_nonneg(v + 1, n - 1 - m);
John Koleszar's avatar
John Koleszar committed
161
  }
162
}
163

164 165
static vp9_prob read_prob_diff_update(vp9_reader *r, int oldp) {
  int delp = decode_term_subexp(r, SUBEXP_PARAM, 255);
166
  return (vp9_prob)inv_remap_prob(delp, oldp);
167
}
168

169
void vp9_init_de_quantizer(VP9D_COMP *pbi) {
John Koleszar's avatar
John Koleszar committed
170
  int i;
Dmitry Kovalev's avatar
Dmitry Kovalev committed
171
  int q;
172
  VP9_COMMON *const pc = &pbi->common;
John Koleszar's avatar
John Koleszar committed
173

Dmitry Kovalev's avatar
Dmitry Kovalev committed
174
  for (q = 0; q < QINDEX_RANGE; q++) {
175
    // DC value
176 177
    pc->y_dequant[q][0] = (int16_t)vp9_dc_quant(q, pc->y_dc_delta_q);
    pc->uv_dequant[q][0] = (int16_t)vp9_dc_uv_quant(q, pc->uv_dc_delta_q);
John Koleszar's avatar
John Koleszar committed
178

179
    // AC values
John Koleszar's avatar
John Koleszar committed
180
    for (i = 1; i < 16; i++) {
181
      const int rc = vp9_default_zig_zag1d_4x4[i];
John Koleszar's avatar
John Koleszar committed
182

Dmitry Kovalev's avatar
Dmitry Kovalev committed
183
      pc->y_dequant[q][rc] = (int16_t)vp9_ac_yquant(q);
184
      pc->uv_dequant[q][rc] = (int16_t)vp9_ac_uv_quant(q, pc->uv_ac_delta_q);
John Koleszar's avatar
John Koleszar committed
185
    }
John Koleszar's avatar
John Koleszar committed
186
  }
John Koleszar's avatar
John Koleszar committed
187 188
}

Dmitry Kovalev's avatar
Dmitry Kovalev committed
189 190 191
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)) {
192 193 194 195
    const int data = vp9_get_segdata(mb, segment_id, SEG_LVL_ALT_Q);
    return mb->mb_segment_abs_delta == SEGMENT_ABSDATA ?
               data :  // Abs value
               clamp(base_qindex + data, 0, MAXQ);  // Delta value
Dmitry Kovalev's avatar
Dmitry Kovalev committed
196 197 198 199 200 201
  } else {
    return base_qindex;
  }
}

static void mb_init_dequantizer(VP9D_COMP *pbi, MACROBLOCKD *mb) {
John Koleszar's avatar
John Koleszar committed
202 203
  int i;

Dmitry Kovalev's avatar
Dmitry Kovalev committed
204
  VP9_COMMON *const pc = &pbi->common;
205 206
  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
207
  mb->q_index = qindex;
Dmitry Kovalev's avatar
Dmitry Kovalev committed
208

Dmitry Kovalev's avatar
Dmitry Kovalev committed
209
  for (i = 0; i < 16; i++)
Dmitry Kovalev's avatar
Dmitry Kovalev committed
210
    mb->block[i].dequant = pc->y_dequant[qindex];
John Koleszar's avatar
John Koleszar committed
211

Dmitry Kovalev's avatar
Dmitry Kovalev committed
212
  for (i = 16; i < 24; i++)
Dmitry Kovalev's avatar
Dmitry Kovalev committed
213
    mb->block[i].dequant = pc->uv_dequant[qindex];
John Koleszar's avatar
John Koleszar committed
214

Dmitry Kovalev's avatar
Dmitry Kovalev committed
215
  if (mb->lossless) {
Dmitry Kovalev's avatar
Dmitry Kovalev committed
216
    assert(qindex == 0);
Yaowu Xu's avatar
Yaowu Xu committed
217 218
    mb->inv_txm4x4_1      = vp9_short_iwalsh4x4_1;
    mb->inv_txm4x4        = vp9_short_iwalsh4x4;
Dmitry Kovalev's avatar
Dmitry Kovalev committed
219 220 221 222
    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
223 224
    mb->inv_txm4x4_1      = vp9_short_idct4x4_1;
    mb->inv_txm4x4        = vp9_short_idct4x4;
Dmitry Kovalev's avatar
Dmitry Kovalev committed
225 226 227
    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
228
  }
John Koleszar's avatar
John Koleszar committed
229 230
}

231
static void decode_16x16(MACROBLOCKD *xd) {
232
  const TX_TYPE tx_type = get_tx_type_16x16(xd, 0);
233

234
  vp9_dequant_iht_add_16x16_c(tx_type, xd->plane[0].qcoeff,
235 236
                              xd->block[0].dequant, xd->plane[0].dst.buf,
                              xd->plane[0].dst.stride, xd->plane[0].eobs[0]);
237 238

  vp9_dequant_idct_add_8x8(xd->plane[1].qcoeff, xd->block[16].dequant,
239
                           xd->plane[1].dst.buf, xd->plane[1].dst.stride,
240
                           xd->plane[1].eobs[0]);
241

Dmitry Kovalev's avatar
Dmitry Kovalev committed
242
  vp9_dequant_idct_add_8x8(xd->plane[2].qcoeff, xd->block[20].dequant,
243
                           xd->plane[2].dst.buf, xd->plane[1].dst.stride,
244
                           xd->plane[2].eobs[0]);
245 246
}

247
static void decode_8x8(MACROBLOCKD *xd) {
248 249
  const MB_PREDICTION_MODE mode = xd->mode_info_context->mbmi.mode;
  // luma
250
  // if the first one is DCT_DCT assume all the rest are as well
251
  TX_TYPE tx_type = get_tx_type_8x8(xd, 0);
252
  if (tx_type != DCT_DCT || mode == I8X8_PRED) {
253 254 255 256
    int i;
    for (i = 0; i < 4; i++) {
      int ib = vp9_i8x8_block[i];
      int idx = (ib & 0x02) ? (ib + 2) : ib;
257
      int16_t *q  = BLOCK_OFFSET(xd->plane[0].qcoeff, idx, 16);
258 259
      int16_t *dq = xd->block[0].dequant;
      uint8_t *dst = *(xd->block[ib].base_dst) + xd->block[ib].dst;
260
      int stride = xd->plane[0].dst.stride;
261
      if (mode == I8X8_PRED) {
262
        BLOCKD *b = &xd->block[ib];
263
        int i8x8mode = b->bmi.as_mode.first;
264
        vp9_intra8x8_predict(xd, b, i8x8mode, dst, stride);
265
      }
266
      tx_type = get_tx_type_8x8(xd, ib);
267 268
      vp9_dequant_iht_add_8x8_c(tx_type, q, dq, dst, stride,
                                xd->plane[0].eobs[idx]);
269
    }
270
  } else {
271
    vp9_dequant_idct_add_y_block_8x8(xd->plane[0].qcoeff,
272 273
                                     xd->block[0].dequant, xd->plane[0].dst.buf,
                                     xd->plane[0].dst.stride, xd);
274 275
  }

276 277
  // chroma
  if (mode == I8X8_PRED) {
278 279 280 281 282
    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;
283

284
      b = &xd->block[16 + i];
285 286
      vp9_intra_uv4x4_predict(xd, b, i8x8mode, *(b->base_dst) + b->dst,
                              b->dst_stride);
287
      xd->itxm_add(BLOCK_OFFSET(xd->plane[1].qcoeff, i, 16),
288
                   b->dequant, *(b->base_dst) + b->dst, b->dst_stride,
John Koleszar's avatar
John Koleszar committed
289
                   xd->plane[1].eobs[i]);
290

291
      b = &xd->block[20 + i];
292 293
      vp9_intra_uv4x4_predict(xd, b, i8x8mode, *(b->base_dst) + b->dst,
                              b->dst_stride);
294
      xd->itxm_add(BLOCK_OFFSET(xd->plane[2].qcoeff, i, 16),
295
                   b->dequant, *(b->base_dst) + b->dst, b->dst_stride,
John Koleszar's avatar
John Koleszar committed
296
                   xd->plane[2].eobs[i]);
297
    }
298
  } else if (mode == SPLITMV) {
299
    xd->itxm_add_uv_block(xd->plane[1].qcoeff, xd->block[16].dequant,
300
         xd->plane[1].dst.buf, xd->plane[1].dst.stride, xd->plane[1].eobs);
301
    xd->itxm_add_uv_block(xd->plane[2].qcoeff, xd->block[16].dequant,
302
         xd->plane[2].dst.buf, xd->plane[1].dst.stride, xd->plane[2].eobs);
303
  } else {
304
    vp9_dequant_idct_add_8x8(xd->plane[1].qcoeff, xd->block[16].dequant,
305
                             xd->plane[1].dst.buf, xd->plane[1].dst.stride,
306
                             xd->plane[1].eobs[0]);
307 308

    vp9_dequant_idct_add_8x8(xd->plane[2].qcoeff, xd->block[16].dequant,
309
                             xd->plane[2].dst.buf, xd->plane[1].dst.stride,
310
                             xd->plane[2].eobs[0]);
311 312 313
  }
}

314 315
static INLINE void dequant_add_y(MACROBLOCKD *xd, TX_TYPE tx_type, int idx) {
  BLOCKD *const b = &xd->block[idx];
316
  struct macroblockd_plane *const y = &xd->plane[0];
317 318 319 320 321 322 323 324 325 326 327 328 329
  if (tx_type != DCT_DCT) {
    vp9_dequant_iht_add_c(tx_type,
                          BLOCK_OFFSET(y->qcoeff, idx, 16),
                          b->dequant, *(b->base_dst) + b->dst,
                          b->dst_stride, y->eobs[idx]);
  } else {
    xd->itxm_add(BLOCK_OFFSET(y->qcoeff, idx, 16),
                 b->dequant, *(b->base_dst) + b->dst,
                 b->dst_stride, y->eobs[idx]);
  }
}


330
static void decode_4x4(VP9D_COMP *pbi, MACROBLOCKD *xd, vp9_reader *r) {
331
  TX_TYPE tx_type;
332
  int i = 0;
333
  const MB_PREDICTION_MODE mode = xd->mode_info_context->mbmi.mode;
334 335 336 337 338
  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
339 340
      BLOCKD *b = &xd->block[ib];
      int i8x8mode = b->bmi.as_mode.first;
341 342
      vp9_intra8x8_predict(xd, b, i8x8mode, *(b->base_dst) + b->dst,
                           b->dst_stride);
343
      for (j = 0; j < 4; j++) {
344
        tx_type = get_tx_type_4x4(xd, ib + iblock[j]);
345
        dequant_add_y(xd, tx_type, ib + iblock[j]);
346 347
      }
      b = &xd->block[16 + i];
348 349
      vp9_intra_uv4x4_predict(xd, b, i8x8mode, *(b->base_dst) + b->dst,
                              b->dst_stride);
350
      xd->itxm_add(BLOCK_OFFSET(xd->plane[1].qcoeff, i, 16),
351
                   b->dequant, *(b->base_dst) + b->dst, b->dst_stride,
John Koleszar's avatar
John Koleszar committed
352
                   xd->plane[1].eobs[i]);
353
      b = &xd->block[20 + i];
354 355
      vp9_intra_uv4x4_predict(xd, b, i8x8mode, *(b->base_dst) + b->dst,
                              b->dst_stride);
356
      xd->itxm_add(BLOCK_OFFSET(xd->plane[2].qcoeff, i, 16),
357
                   b->dequant, *(b->base_dst) + b->dst, b->dst_stride,
John Koleszar's avatar
John Koleszar committed
358
                   xd->plane[2].eobs[i]);
359
    }
Yaowu Xu's avatar
Yaowu Xu committed
360
  } else if (mode == I4X4_PRED) {
361 362
    for (i = 0; i < 16; i++) {
      BLOCKD *b = &xd->block[i];
Dmitry Kovalev's avatar
Dmitry Kovalev committed
363
      int b_mode = xd->mode_info_context->bmi[i].as_mode.first;
364 365
#if CONFIG_NEWBINTRAMODES
      xd->mode_info_context->bmi[i].as_mode.context = b->bmi.as_mode.context =
366
          vp9_find_bpred_context(xd, b);
367
      if (!xd->mode_info_context->mbmi.mb_skip_coeff)
368
        vp9_decode_coefs_4x4(pbi, xd, r, PLANE_TYPE_Y_WITH_DC, i);
369
#endif
370 371
      vp9_intra4x4_predict(xd, b, b_mode, *(b->base_dst) + b->dst,
                           b->dst_stride);
372
      tx_type = get_tx_type_4x4(xd, i);
373
      dequant_add_y(xd, tx_type, i);
374
    }
375 376
#if CONFIG_NEWBINTRAMODES
    if (!xd->mode_info_context->mbmi.mb_skip_coeff)
377
      vp9_decode_mb_tokens_4x4_uv(pbi, xd, r);
378
#endif
379
    vp9_build_intra_predictors_sbuv_s(xd, BLOCK_SIZE_MB16X16);
380
    xd->itxm_add_uv_block(xd->plane[1].qcoeff, xd->block[16].dequant,
381
         xd->plane[1].dst.buf, xd->plane[1].dst.stride, xd->plane[1].eobs);
382
    xd->itxm_add_uv_block(xd->plane[2].qcoeff, xd->block[16].dequant,
383
         xd->plane[2].dst.buf, xd->plane[1].dst.stride, xd->plane[2].eobs);
384
  } else if (mode == SPLITMV || get_tx_type_4x4(xd, 0) == DCT_DCT) {
385
    xd->itxm_add_y_block(xd->plane[0].qcoeff,
386
                          xd->block[0].dequant,
387
                          xd->plane[0].dst.buf, xd->plane[0].dst.stride, xd);
388
    xd->itxm_add_uv_block(xd->plane[1].qcoeff, xd->block[16].dequant,
389
         xd->plane[1].dst.buf, xd->plane[1].dst.stride, xd->plane[1].eobs);
390
    xd->itxm_add_uv_block(xd->plane[2].qcoeff, xd->block[16].dequant,
391
         xd->plane[2].dst.buf, xd->plane[1].dst.stride, xd->plane[2].eobs);
392
  } else {
393
    for (i = 0; i < 16; i++) {
394
      tx_type = get_tx_type_4x4(xd, i);
395
      dequant_add_y(xd, tx_type, i);
396
    }
397
    xd->itxm_add_uv_block(xd->plane[1].qcoeff, xd->block[16].dequant,
398
                          xd->plane[1].dst.buf, xd->plane[1].dst.stride,
399
                          xd->plane[1].eobs);
400
    xd->itxm_add_uv_block(xd->plane[2].qcoeff, xd->block[16].dequant,
401
                          xd->plane[2].dst.buf, xd->plane[1].dst.stride,
402
                          xd->plane[2].eobs);
403 404 405
  }
}

406 407 408 409
static INLINE void decode_sby_32x32(MACROBLOCKD *mb, BLOCK_SIZE_TYPE bsize) {
  const int bwl = mb_width_log2(bsize) - 1, bw = 1 << bwl;
  const int bhl = mb_height_log2(bsize) - 1, bh = 1 << bhl;
  const int y_count = bw * bh;
410 411 412
  int n;

  for (n = 0; n < y_count; n++) {
413 414
    const int x_idx = n & (bw - 1);
    const int y_idx = n >> bwl;
415
    const int y_offset = (y_idx * 32) * mb->plane[0].dst.stride + (x_idx * 32);
416 417
    vp9_dequant_idct_add_32x32(BLOCK_OFFSET(mb->plane[0].qcoeff, n, 1024),
                               mb->block[0].dequant ,
418 419
                               mb->plane[0].dst.buf + y_offset,
                               mb->plane[0].dst.stride,
420 421 422 423 424 425 426 427 428 429 430 431
                               mb->plane[0].eobs[n * 64]);
  }
}

static INLINE void decode_sbuv_32x32(MACROBLOCKD *mb, BLOCK_SIZE_TYPE bsize) {
  const int bwl = mb_width_log2(bsize) - 1, bw = (1 << bwl) / 2;
  const int bhl = mb_height_log2(bsize) - 1, bh = (1 << bhl) / 2;
  const int uv_count = bw * bh;
  int n;
  for (n = 0; n < uv_count; n++) {
     const int x_idx = n & (bw - 1);
     const int y_idx = n >> (bwl - 1);
432 433
     const int uv_offset = (y_idx * 32) * mb->plane[1].dst.stride +
         (x_idx * 32);
434 435
     vp9_dequant_idct_add_32x32(BLOCK_OFFSET(mb->plane[1].qcoeff, n, 1024),
                                mb->block[16].dequant,
436 437 438
                                mb->plane[1].dst.buf + uv_offset,
                                mb->plane[1].dst.stride,
                                mb->plane[1].eobs[n * 64]);
439 440
     vp9_dequant_idct_add_32x32(BLOCK_OFFSET(mb->plane[2].qcoeff, n, 1024),
                                mb->block[20].dequant,
441 442 443
                                mb->plane[2].dst.buf + uv_offset,
                                mb->plane[1].dst.stride,
                                mb->plane[2].eobs[n * 64]);
444 445 446 447 448 449 450 451 452 453 454 455
  }
}

static INLINE void decode_sby_16x16(MACROBLOCKD *mb, BLOCK_SIZE_TYPE bsize) {
  const int bwl = mb_width_log2(bsize), bw = 1 << bwl;
  const int bhl = mb_height_log2(bsize), bh = 1 << bhl;
  const int y_count = bw * bh;
  int n;

  for (n = 0; n < y_count; n++) {
    const int x_idx = n & (bw - 1);
    const int y_idx = n >> bwl;
456
    const int y_offset = (y_idx * 16) * mb->plane[0].dst.stride + (x_idx * 16);
457
    const TX_TYPE tx_type = get_tx_type_16x16(mb,
458
                                (y_idx * (4 * bw) + x_idx) * 4);
459 460 461
    vp9_dequant_iht_add_16x16_c(tx_type,
                                BLOCK_OFFSET(mb->plane[0].qcoeff, n, 256),
                                mb->block[0].dequant,
462 463
                                mb->plane[0].dst.buf + y_offset,
                                mb->plane[0].dst.stride,
464
                                mb->plane[0].eobs[n * 16]);
465
  }
466 467 468 469 470 471 472 473 474
}

static INLINE void decode_sbuv_16x16(MACROBLOCKD *mb, BLOCK_SIZE_TYPE bsize) {
  const int bwl = mb_width_log2(bsize), bw = (1 << bwl) / 2;
  const int bhl = mb_height_log2(bsize), bh = (1 << bhl) / 2;
  const int uv_count = bw * bh;
  int n;

  assert(bsize >= BLOCK_SIZE_SB32X32);
475 476

  for (n = 0; n < uv_count; n++) {
477 478
    const int x_idx = n & (bw - 1);
    const int y_idx = n >> (bwl - 1);
479
    const int uv_offset = (y_idx * 16) * mb->plane[1].dst.stride + (x_idx * 16);
480
    vp9_dequant_idct_add_16x16(BLOCK_OFFSET(mb->plane[1].qcoeff, n, 256),
481
                               mb->block[16].dequant,
482 483
                               mb->plane[1].dst.buf + uv_offset,
                               mb->plane[1].dst.stride,
John Koleszar's avatar
John Koleszar committed
484
                               mb->plane[1].eobs[n * 16]);
485
    vp9_dequant_idct_add_16x16(BLOCK_OFFSET(mb->plane[2].qcoeff, n, 256),
486
                               mb->block[20].dequant,
487 488
                               mb->plane[2].dst.buf + uv_offset,
                               mb->plane[1].dst.stride,
John Koleszar's avatar
John Koleszar committed
489
                               mb->plane[2].eobs[n * 16]);
490 491 492
  }
}

493 494 495 496
static INLINE void decode_sby_8x8(MACROBLOCKD *xd, BLOCK_SIZE_TYPE bsize) {
  const int bwl = mb_width_log2(bsize)  + 1, bw = 1 << bwl;
  const int bhl = mb_height_log2(bsize) + 1, bh = 1 << bhl;
  const int y_count = bw * bh;
497 498 499 500
  int n;

  // luma
  for (n = 0; n < y_count; n++) {
501 502
    const int x_idx = n & (bw - 1);
    const int y_idx = n >> bwl;
503
    const int y_offset = (y_idx * 8) * xd->plane[0].dst.stride + (x_idx * 8);
504
    const TX_TYPE tx_type = get_tx_type_8x8(xd,
505
                                            (y_idx * (2 * bw) + x_idx) * 2);
506 507 508 509

    vp9_dequant_iht_add_8x8_c(tx_type,
                              BLOCK_OFFSET(xd->plane[0].qcoeff, n, 64),
                              xd->block[0].dequant,
510 511
                              xd->plane[0].dst.buf + y_offset,
                              xd->plane[0].dst.stride,
512
                              xd->plane[0].eobs[n * 4]);
513
  }
514 515 516 517 518 519 520
}

static INLINE void decode_sbuv_8x8(MACROBLOCKD *xd, BLOCK_SIZE_TYPE bsize) {
  const int bwl = mb_width_log2(bsize)  + 1, bw = 1 << (bwl - 1);
  const int bhl = mb_height_log2(bsize) + 1, bh = 1 << (bhl - 1);
  const int uv_count = bw * bh;
  int n;
521 522 523

  // chroma
  for (n = 0; n < uv_count; n++) {
524 525
    const int x_idx = n & (bw - 1);
    const int y_idx = n >> (bwl - 1);
526
    const int uv_offset = (y_idx * 8) * xd->plane[1].dst.stride + (x_idx * 8);
527 528
    vp9_dequant_idct_add_8x8(BLOCK_OFFSET(xd->plane[1].qcoeff, n, 64),
                             xd->block[16].dequant,
529 530
                             xd->plane[1].dst.buf + uv_offset,
                             xd->plane[1].dst.stride,
531 532 533
                             xd->plane[1].eobs[n * 4]);
    vp9_dequant_idct_add_8x8(BLOCK_OFFSET(xd->plane[2].qcoeff, n, 64),
                             xd->block[20].dequant,
534 535
                             xd->plane[2].dst.buf + uv_offset,
                             xd->plane[1].dst.stride,
536
                             xd->plane[2].eobs[n * 4]);
537 538 539
  }
}

540 541 542 543
static INLINE void decode_sby_4x4(MACROBLOCKD *xd, BLOCK_SIZE_TYPE bsize) {
  const int bwl = mb_width_log2(bsize)  + 2, bw = 1 << bwl;
  const int bhl = mb_height_log2(bsize) + 2, bh = 1 << bhl;
  const int y_count = bw * bh;
544 545 546
  int n;

  for (n = 0; n < y_count; n++) {
547 548
    const int x_idx = n & (bw - 1);
    const int y_idx = n >> bwl;
549
    const int y_offset = (y_idx * 4) * xd->plane[0].dst.stride + (x_idx * 4);
550
    const TX_TYPE tx_type = get_tx_type_4x4(xd, n);
551
    if (tx_type == DCT_DCT) {
552 553
      xd->itxm_add(BLOCK_OFFSET(xd->plane[0].qcoeff, n, 16),
                   xd->block[0].dequant,
554
                   xd->plane[0].dst.buf + y_offset, xd->plane[0].dst.stride,
John Koleszar's avatar
John Koleszar committed
555
                   xd->plane[0].eobs[n]);
556
    } else {
557 558
      vp9_dequant_iht_add_c(tx_type,
                            BLOCK_OFFSET(xd->plane[0].qcoeff, n, 16),
559 560 561
                            xd->block[0].dequant,
                            xd->plane[0].dst.buf + y_offset,
                            xd->plane[0].dst.stride, xd->plane[0].eobs[n]);
562 563
    }
  }
564 565 566 567 568 569 570
}

static INLINE void decode_sbuv_4x4(MACROBLOCKD *xd, BLOCK_SIZE_TYPE bsize) {
  const int bwl = mb_width_log2(bsize)  + 2, bw = 1 << (bwl - 1);
  const int bhl = mb_height_log2(bsize) + 2, bh = 1 << (bhl - 1);
  const int uv_count = bw * bh;
  int n;
571 572

  for (n = 0; n < uv_count; n++) {
573 574
    const int x_idx = n & (bw - 1);
    const int y_idx = n >> (bwl - 1);
575
    const int uv_offset = (y_idx * 4) * xd->plane[1].dst.stride + (x_idx * 4);
576 577
    xd->itxm_add(BLOCK_OFFSET(xd->plane[1].qcoeff, n, 16),
        xd->block[16].dequant,
578 579
        xd->plane[1].dst.buf + uv_offset, xd->plane[1].dst.stride,
        xd->plane[1].eobs[n]);
580 581
    xd->itxm_add(BLOCK_OFFSET(xd->plane[2].qcoeff, n, 16),
        xd->block[20].dequant,
582 583
        xd->plane[2].dst.buf + uv_offset, xd->plane[1].dst.stride,
        xd->plane[2].eobs[n]);
584 585 586
  }
}

587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605
// TODO(jingning): combine luma and chroma dequantization and inverse
// transform into a single function looping over planes.
static void decode_sb_32x32(MACROBLOCKD *mb, BLOCK_SIZE_TYPE bsize) {
  decode_sby_32x32(mb, bsize);
  if (bsize == BLOCK_SIZE_SB64X64)
    decode_sbuv_32x32(mb, bsize);
  else
    decode_sbuv_16x16(mb, bsize);
}

static void decode_sb_16x16(MACROBLOCKD *mb, BLOCK_SIZE_TYPE bsize) {
  decode_sby_16x16(mb, bsize);
  if (bsize >= BLOCK_SIZE_SB32X32)
    decode_sbuv_16x16(mb, bsize);
  else
    decode_sbuv_8x8(mb, bsize);
}

static void decode_sb(VP9D_COMP *pbi, MACROBLOCKD *xd, int mb_row, int mb_col,
606
                      vp9_reader *r, BLOCK_SIZE_TYPE bsize) {
607 608
  const int bwl = mb_width_log2(bsize), bhl = mb_height_log2(bsize);
  const int bw = 1 << bwl, bh = 1 << bhl;
609
  int n, eobtotal;
610
  VP9_COMMON *const pc = &pbi->common;
611
  MODE_INFO *mi = xd->mode_info_context;
612
  const int mis = pc->mode_info_stride;
613

614
  assert(mi->mbmi.sb_type == bsize);
Ronald S. Bultje's avatar
Ronald S. Bultje committed
615 616

  if (pbi->common.frame_type != KEY_FRAME)
617
    vp9_setup_interp_filters(xd, mi->mbmi.interp_filter, pc);
Ronald S. Bultje's avatar
Ronald S. Bultje committed
618

619
  // generate prediction
620 621 622
  if (xd->mode_info_context->mbmi.ref_frame == INTRA_FRAME) {
    vp9_build_intra_predictors_sby_s(xd, bsize);
    vp9_build_intra_predictors_sbuv_s(xd, bsize);
Ronald S. Bultje's avatar
Ronald S. Bultje committed
623
  } else {
624
    vp9_build_inter_predictors_sb(xd, mb_row, mb_col, bsize);
Ronald S. Bultje's avatar
Ronald S. Bultje committed
625 626
  }

Yunqing Wang's avatar
Yunqing Wang committed
627 628
  if (mi->mbmi.mb_skip_coeff) {
    vp9_reset_sb_tokens_context(xd, bsize);
629
  } else {
Yunqing Wang's avatar
Yunqing Wang committed
630 631 632 633 634
    // re-initialize macroblock dequantizer before detokenization
    if (xd->segmentation_enabled)
      mb_init_dequantizer(pbi, xd);

    // dequantization and idct
635
    eobtotal = vp9_decode_tokens(pbi, xd, r, bsize);
Yunqing Wang's avatar
Yunqing Wang committed
636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660
    if (eobtotal == 0) {  // skip loopfilter
      for (n = 0; n < bw * bh; n++) {
        const int x_idx = n & (bw - 1), y_idx = n >> bwl;

        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 = 1;
      }
    } else {
      switch (xd->mode_info_context->mbmi.txfm_size) {
        case TX_32X32:
          decode_sb_32x32(xd, bsize);
          break;
        case TX_16X16:
          decode_sb_16x16(xd, bsize);
          break;
        case TX_8X8:
          decode_sby_8x8(xd, bsize);
          decode_sbuv_8x8(xd, bsize);
          break;
        case TX_4X4:
          decode_sby_4x4(xd, bsize);
          decode_sbuv_4x4(xd, bsize);
          break;
        default: assert(0);
      }
661
    }
662
  }
663 664
}

665 666
// TODO(jingning): Need to merge SB and MB decoding. The MB decoding currently
// couples special handles on I8x8, B_PRED, and splitmv modes.
667 668
static void decode_mb(VP9D_COMP *pbi, MACROBLOCKD *xd,
                     int mb_row, int mb_col,
669
                     vp9_reader *r) {
John Koleszar's avatar
John Koleszar committed
670
  int eobtotal = 0;
671 672
  const MB_PREDICTION_MODE mode = xd->mode_info_context->mbmi.mode;
  const int tx_size = xd->mode_info_context->mbmi.txfm_size;
673

674
  assert(xd->mode_info_context->mbmi.sb_type == BLOCK_SIZE_MB16X16);
John Koleszar's avatar
John Koleszar committed
675

Deb Mukherjee's avatar
Deb Mukherjee committed
676
  //mode = xd->mode_info_context->mbmi.mode;
677
  if (pbi->common.frame_type != KEY_FRAME)
678
    vp9_setup_interp_filters(xd, xd->mode_info_context->mbmi.interp_filter,
679
                             &pbi->common);
Gaute Strokkenes's avatar
Gaute Strokkenes committed
680

681
  // do prediction
John Koleszar's avatar
John Koleszar committed
682 683
  if (xd->mode_info_context->mbmi.ref_frame == INTRA_FRAME) {
    if (mode != I8X8_PRED) {
684
      vp9_build_intra_predictors_sbuv_s(xd, BLOCK_SIZE_MB16X16);
Yaowu Xu's avatar
Yaowu Xu committed
685
      if (mode != I4X4_PRED)
686
        vp9_build_intra_predictors_sby_s(xd, BLOCK_SIZE_MB16X16);
John Koleszar's avatar
John Koleszar committed
687
    }
John Koleszar's avatar
John Koleszar committed
688
  } else {
689
#if 0  // def DEC_DEBUG
690 691 692 693 694
  if (dec_debug)
    printf("Decoding mb:  %d %d interp %d\n",
           xd->mode_info_context->mbmi.mode, tx_size,
           xd->mode_info_context->mbmi.interp_filter);
#endif
695
    vp9_build_inter_predictors_sb(xd, mb_row, mb_col, BLOCK_SIZE_MB16X16);
John Koleszar's avatar
John Koleszar committed
696 697
  }

Yunqing Wang's avatar
Yunqing Wang committed
698 699
  if (xd->mode_info_context->mbmi.mb_skip_coeff) {
    vp9_reset_sb_tokens_context(xd, BLOCK_SIZE_MB16X16);
700
  } else {
Yunqing Wang's avatar
Yunqing Wang committed
701 702 703 704
    // re-initialize macroblock dequantizer before detokenization
    if (xd->segmentation_enabled)
      mb_init_dequantizer(pbi, xd);

705
    if (!vp9_reader_has_error(r)) {
Yunqing Wang's avatar
Yunqing Wang committed
706 707 708
#if CONFIG_NEWBINTRAMODES
    if (mode != I4X4_PRED)
#endif
709
      eobtotal = vp9_decode_tokens(pbi, xd, r, BLOCK_SIZE_MB16X16);
Yunqing Wang's avatar
Yunqing Wang committed
710 711 712 713 714 715 716
    }
  }

  if (eobtotal == 0 &&
      mode != I4X4_PRED &&
      mode != SPLITMV &&
      mode != I8X8_PRED &&
717
      !vp9_reader_has_error(r)) {
Yunqing Wang's avatar
Yunqing Wang committed
718 719 720 721 722 723 724 725
    xd->mode_info_context->mbmi.mb_skip_coeff = 1;
  } else {
#if 0  // def DEC_DEBUG
  if (dec_debug)
    printf("Decoding mb:  %d %d\n", xd->mode_info_context->mbmi.mode, tx_size);
#endif

    if (tx_size == TX_16X16) {
726
      decode_16x16(xd);
Yunqing Wang's avatar
Yunqing Wang committed
727
    } else if (tx_size == TX_8X8) {
728
      decode_8x8(xd);
Yunqing Wang's avatar
Yunqing Wang committed
729
    } else {
730
      decode_4x4(pbi, xd, r);
Yunqing Wang's avatar
Yunqing Wang committed
731
    }
732
  }
Yunqing Wang's avatar
Yunqing Wang committed
733

734 735 736 737
#ifdef DEC_DEBUG
  if (dec_debug) {
    int i, j;
    printf("\n");
738 739 740 741 742 743 744
    printf("predictor y\n");
    for (i = 0; i < 16; i++) {
      for (j = 0; j < 16; j++)
        printf("%3d ", xd->predictor[i * 16 + j]);
      printf("\n");
    }
    printf("\n");
745
    printf("final y\n");
John Koleszar's avatar
John Koleszar committed
746
    for (i = 0; i < 16; i++) {
747
      for (j = 0; j < 16; j++)
748
        printf("%3d ", xd->plane[0].dst.buf[i * xd->plane[0].dst.stride + j]);
749
      printf("\n");
John Koleszar's avatar
John Koleszar committed
750
    }
751 752 753 754
    printf("\n");
    printf("final u\n");
    for (i = 0; i < 8; i++) {
      for (j = 0; j < 8; j++)
755
        printf("%3d ", xd->plane[1].dst.buf[i * xd->plane[1].dst.stride + j]);
756
      printf("\n");
757
    }
758 759 760 761
    printf("\n");
    printf("final v\n");
    for (i = 0; i < 8; i++) {
      for (j = 0; j < 8; j++)
762
        printf("%3d ", xd->plane[2].dst.buf[i * xd->plane[1].dst.stride + j]);
763
      printf("\n");
764
    }
765
    fflush(stdout);
John Koleszar's avatar
John Koleszar committed
766
  }
767
#endif
Yaowu Xu's avatar
Yaowu Xu committed
768
}
John Koleszar's avatar
John Koleszar committed
769

770 771 772 773
static int get_delta_q(vp9_reader *r, int *dq) {
  const int old_value = *dq;

  if (vp9_read_bit(r)) {  // Update bit
774 775
    const int value = vp9_read_literal(r, 4);
    *dq = vp9_read_and_apply_sign(r, value);
John Koleszar's avatar
John Koleszar committed