vp9_decodframe.c 56.4 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
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
56 57
static int read_is_valid(const uint8_t *start, size_t len,
                         const uint8_t *end) {
58 59 60
  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;
}

68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88
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);
}

89
static int decode_uniform(vp9_reader *r, int n) {
90 91 92 93 94 95
  int v;
  const int l = get_unsigned_bits(n);
  const int m = (1 << l) - n;
  if (!l)
    return 0;

96 97
  v = vp9_read_literal(r, l - 1);
  return v < m ?  v : (v << 1) - m + vp9_read_bit(r);
98 99
}

100
static int decode_term_subexp(vp9_reader *r, int k, int num_syms) {
101 102 103 104 105
  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) {
106
      word = decode_uniform(r, num_syms - mk) + mk;
107 108
      break;
    } else {
109
      if (vp9_read_bit(r)) {
110 111 112
        i++;
        mk += a;
      } else {
113
        word = vp9_read_literal(r, b) + mk;
114 115 116 117 118 119 120
        break;
      }
    }
  }
  return word;
}

121
static int decode_unsigned_max(vp9_reader *r, int max) {
122 123 124
  int data = 0, bit = 0, lmax = max;

  while (lmax) {
125
    data |= vp9_read_bit(r) << bit++;
126 127 128 129 130
    lmax >>= 1;
  }
  return data > max ? max : data;
}

John Koleszar's avatar
John Koleszar committed
131 132 133 134 135 136 137 138 139 140 141 142
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;
143 144
}

John Koleszar's avatar
John Koleszar committed
145 146 147
static int inv_remap_prob(int v, int m) {
  const int n = 256;
  const int modulus = MODULUS_PARAM;
148

John Koleszar's avatar
John Koleszar committed
149 150
  v = merge_index(v, n - 1, modulus);
  if ((m << 1) <= n) {
151
    return inv_recenter_nonneg(v + 1, m);
John Koleszar's avatar
John Koleszar committed
152
  } else {
153
    return n - 1 - inv_recenter_nonneg(v + 1, n - 1 - m);
John Koleszar's avatar
John Koleszar committed
154
  }
155
}
156

157 158
static vp9_prob read_prob_diff_update(vp9_reader *r, int oldp) {
  int delp = decode_term_subexp(r, SUBEXP_PARAM, 255);
159
  return (vp9_prob)inv_remap_prob(delp, oldp);
160
}
161

162
void vp9_init_de_quantizer(VP9D_COMP *pbi) {
John Koleszar's avatar
John Koleszar committed
163
  int i;
Dmitry Kovalev's avatar
Dmitry Kovalev committed
164
  int q;
165
  VP9_COMMON *const pc = &pbi->common;
John Koleszar's avatar
John Koleszar committed
166

Dmitry Kovalev's avatar
Dmitry Kovalev committed
167
  for (q = 0; q < QINDEX_RANGE; q++) {
168 169
    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
170 171 172

    /* all the ac values =; */
    for (i = 1; i < 16; i++) {
173
      const int rc = vp9_default_zig_zag1d_4x4[i];
John Koleszar's avatar
John Koleszar committed
174

Dmitry Kovalev's avatar
Dmitry Kovalev committed
175
      pc->y_dequant[q][rc] = (int16_t)vp9_ac_yquant(q);
176
      pc->uv_dequant[q][rc] = (int16_t)vp9_ac_uv_quant(q, pc->uv_ac_delta_q);
John Koleszar's avatar
John Koleszar committed
177
    }
John Koleszar's avatar
John Koleszar committed
178
  }
John Koleszar's avatar
John Koleszar committed
179 180
}

Dmitry Kovalev's avatar
Dmitry Kovalev committed
181 182 183
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)) {
184 185 186 187
    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
188 189 190 191 192 193
  } else {
    return base_qindex;
  }
}

static void mb_init_dequantizer(VP9D_COMP *pbi, MACROBLOCKD *mb) {
John Koleszar's avatar
John Koleszar committed
194 195
  int i;

Dmitry Kovalev's avatar
Dmitry Kovalev committed
196
  VP9_COMMON *const pc = &pbi->common;
197 198
  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
199
  mb->q_index = qindex;
Dmitry Kovalev's avatar
Dmitry Kovalev committed
200

Dmitry Kovalev's avatar
Dmitry Kovalev committed
201
  for (i = 0; i < 16; i++)
Dmitry Kovalev's avatar
Dmitry Kovalev committed
202
    mb->block[i].dequant = pc->y_dequant[qindex];
John Koleszar's avatar
John Koleszar committed
203

Dmitry Kovalev's avatar
Dmitry Kovalev committed
204
  for (i = 16; i < 24; i++)
Dmitry Kovalev's avatar
Dmitry Kovalev committed
205
    mb->block[i].dequant = pc->uv_dequant[qindex];
John Koleszar's avatar
John Koleszar committed
206

Dmitry Kovalev's avatar
Dmitry Kovalev committed
207
  if (mb->lossless) {
Dmitry Kovalev's avatar
Dmitry Kovalev committed
208
    assert(qindex == 0);
Yaowu Xu's avatar
Yaowu Xu committed
209 210
    mb->inv_txm4x4_1      = vp9_short_iwalsh4x4_1;
    mb->inv_txm4x4        = vp9_short_iwalsh4x4;
Dmitry Kovalev's avatar
Dmitry Kovalev committed
211 212 213 214
    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
215 216
    mb->inv_txm4x4_1      = vp9_short_idct4x4_1;
    mb->inv_txm4x4        = vp9_short_idct4x4;
Dmitry Kovalev's avatar
Dmitry Kovalev committed
217 218 219
    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
220
  }
John Koleszar's avatar
John Koleszar committed
221 222
}

223
static void decode_16x16(MACROBLOCKD *xd) {
224
  const TX_TYPE tx_type = get_tx_type_16x16(xd, 0);
225

226
  vp9_dequant_iht_add_16x16_c(tx_type, xd->plane[0].qcoeff,
227 228
                              xd->block[0].dequant, xd->plane[0].dst.buf,
                              xd->plane[0].dst.stride, xd->plane[0].eobs[0]);
229 230

  vp9_dequant_idct_add_8x8(xd->plane[1].qcoeff, xd->block[16].dequant,
231
                           xd->plane[1].dst.buf, xd->plane[1].dst.stride,
232
                           xd->plane[1].eobs[0]);
233

Dmitry Kovalev's avatar
Dmitry Kovalev committed
234
  vp9_dequant_idct_add_8x8(xd->plane[2].qcoeff, xd->block[20].dequant,
235
                           xd->plane[2].dst.buf, xd->plane[1].dst.stride,
236
                           xd->plane[2].eobs[0]);
237 238
}

239
static void decode_8x8(MACROBLOCKD *xd) {
240 241
  const MB_PREDICTION_MODE mode = xd->mode_info_context->mbmi.mode;
  // luma
242
  // if the first one is DCT_DCT assume all the rest are as well
243
  TX_TYPE tx_type = get_tx_type_8x8(xd, 0);
244
  if (tx_type != DCT_DCT || mode == I8X8_PRED) {
245 246 247 248
    int i;
    for (i = 0; i < 4; i++) {
      int ib = vp9_i8x8_block[i];
      int idx = (ib & 0x02) ? (ib + 2) : ib;
249
      int16_t *q  = BLOCK_OFFSET(xd->plane[0].qcoeff, idx, 16);
250 251
      int16_t *dq = xd->block[0].dequant;
      uint8_t *dst = *(xd->block[ib].base_dst) + xd->block[ib].dst;
252
      int stride = xd->plane[0].dst.stride;
253
      if (mode == I8X8_PRED) {
254
        BLOCKD *b = &xd->block[ib];
255
        int i8x8mode = b->bmi.as_mode.first;
256
        vp9_intra8x8_predict(xd, b, i8x8mode, dst, stride);
257
      }
258
      tx_type = get_tx_type_8x8(xd, ib);
259 260
      vp9_dequant_iht_add_8x8_c(tx_type, q, dq, dst, stride,
                                xd->plane[0].eobs[idx]);
261
    }
262
  } else {
263
    vp9_dequant_idct_add_y_block_8x8(xd->plane[0].qcoeff,
264 265
                                     xd->block[0].dequant, xd->plane[0].dst.buf,
                                     xd->plane[0].dst.stride, xd);
266 267
  }

268 269
  // chroma
  if (mode == I8X8_PRED) {
270 271 272 273 274
    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;
275

276
      b = &xd->block[16 + i];
277 278
      vp9_intra_uv4x4_predict(xd, b, i8x8mode, *(b->base_dst) + b->dst,
                              b->dst_stride);
279
      xd->itxm_add(BLOCK_OFFSET(xd->plane[1].qcoeff, i, 16),
280
                   b->dequant, *(b->base_dst) + b->dst, b->dst_stride,
John Koleszar's avatar
John Koleszar committed
281
                   xd->plane[1].eobs[i]);
282

283
      b = &xd->block[20 + i];
284 285
      vp9_intra_uv4x4_predict(xd, b, i8x8mode, *(b->base_dst) + b->dst,
                              b->dst_stride);
286
      xd->itxm_add(BLOCK_OFFSET(xd->plane[2].qcoeff, i, 16),
287
                   b->dequant, *(b->base_dst) + b->dst, b->dst_stride,
John Koleszar's avatar
John Koleszar committed
288
                   xd->plane[2].eobs[i]);
289
    }
290
  } else if (mode == SPLITMV) {
291
    xd->itxm_add_uv_block(xd->plane[1].qcoeff, xd->block[16].dequant,
292
         xd->plane[1].dst.buf, xd->plane[1].dst.stride, xd->plane[1].eobs);
293
    xd->itxm_add_uv_block(xd->plane[2].qcoeff, xd->block[16].dequant,
294
         xd->plane[2].dst.buf, xd->plane[1].dst.stride, xd->plane[2].eobs);
295
  } else {
296
    vp9_dequant_idct_add_8x8(xd->plane[1].qcoeff, xd->block[16].dequant,
297
                             xd->plane[1].dst.buf, xd->plane[1].dst.stride,
298
                             xd->plane[1].eobs[0]);
299 300

    vp9_dequant_idct_add_8x8(xd->plane[2].qcoeff, xd->block[16].dequant,
301
                             xd->plane[2].dst.buf, xd->plane[1].dst.stride,
302
                             xd->plane[2].eobs[0]);
303 304 305
  }
}

306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321
static INLINE void dequant_add_y(MACROBLOCKD *xd, TX_TYPE tx_type, int idx) {
  BLOCKD *const b = &xd->block[idx];
  struct mb_plane *const y = &xd->plane[0];
  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]);
  }
}


322
static void decode_4x4(VP9D_COMP *pbi, MACROBLOCKD *xd, vp9_reader *r) {
323
  TX_TYPE tx_type;
324
  int i = 0;
325
  const MB_PREDICTION_MODE mode = xd->mode_info_context->mbmi.mode;
326 327 328 329 330
  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
331 332
      BLOCKD *b = &xd->block[ib];
      int i8x8mode = b->bmi.as_mode.first;
333 334
      vp9_intra8x8_predict(xd, b, i8x8mode, *(b->base_dst) + b->dst,
                           b->dst_stride);
335
      for (j = 0; j < 4; j++) {
336
        tx_type = get_tx_type_4x4(xd, ib + iblock[j]);
337
        dequant_add_y(xd, tx_type, ib + iblock[j]);
338 339
      }
      b = &xd->block[16 + i];
340 341
      vp9_intra_uv4x4_predict(xd, b, i8x8mode, *(b->base_dst) + b->dst,
                              b->dst_stride);
342
      xd->itxm_add(BLOCK_OFFSET(xd->plane[1].qcoeff, i, 16),
343
                   b->dequant, *(b->base_dst) + b->dst, b->dst_stride,
John Koleszar's avatar
John Koleszar committed
344
                   xd->plane[1].eobs[i]);
345
      b = &xd->block[20 + i];
346 347
      vp9_intra_uv4x4_predict(xd, b, i8x8mode, *(b->base_dst) + b->dst,
                              b->dst_stride);
348
      xd->itxm_add(BLOCK_OFFSET(xd->plane[2].qcoeff, i, 16),
349
                   b->dequant, *(b->base_dst) + b->dst, b->dst_stride,
John Koleszar's avatar
John Koleszar committed
350
                   xd->plane[2].eobs[i]);
351
    }
Yaowu Xu's avatar
Yaowu Xu committed
352
  } else if (mode == I4X4_PRED) {
353 354
    for (i = 0; i < 16; i++) {
      BLOCKD *b = &xd->block[i];
Dmitry Kovalev's avatar
Dmitry Kovalev committed
355
      int b_mode = xd->mode_info_context->bmi[i].as_mode.first;
356 357
#if CONFIG_NEWBINTRAMODES
      xd->mode_info_context->bmi[i].as_mode.context = b->bmi.as_mode.context =
358
          vp9_find_bpred_context(xd, b);
359
      if (!xd->mode_info_context->mbmi.mb_skip_coeff)
360
        vp9_decode_coefs_4x4(pbi, xd, r, PLANE_TYPE_Y_WITH_DC, i);
361
#endif
362 363
      vp9_intra4x4_predict(xd, b, b_mode, *(b->base_dst) + b->dst,
                           b->dst_stride);
364
      tx_type = get_tx_type_4x4(xd, i);
365
      dequant_add_y(xd, tx_type, i);
366
    }
367 368
#if CONFIG_NEWBINTRAMODES
    if (!xd->mode_info_context->mbmi.mb_skip_coeff)
369
      vp9_decode_mb_tokens_4x4_uv(pbi, xd, r);
370
#endif
371
    vp9_build_intra_predictors_sbuv_s(xd, BLOCK_SIZE_MB16X16);
372
    xd->itxm_add_uv_block(xd->plane[1].qcoeff, xd->block[16].dequant,
373
         xd->plane[1].dst.buf, xd->plane[1].dst.stride, xd->plane[1].eobs);
374
    xd->itxm_add_uv_block(xd->plane[2].qcoeff, xd->block[16].dequant,
375
         xd->plane[2].dst.buf, xd->plane[1].dst.stride, xd->plane[2].eobs);
376
  } else if (mode == SPLITMV || get_tx_type_4x4(xd, 0) == DCT_DCT) {
377
    xd->itxm_add_y_block(xd->plane[0].qcoeff,
378
                          xd->block[0].dequant,
379
                          xd->plane[0].dst.buf, xd->plane[0].dst.stride, xd);
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 {
385
    for (i = 0; i < 16; i++) {
386
      tx_type = get_tx_type_4x4(xd, i);
387
      dequant_add_y(xd, tx_type, i);
388
    }
389
    xd->itxm_add_uv_block(xd->plane[1].qcoeff, xd->block[16].dequant,
390
                          xd->plane[1].dst.buf, xd->plane[1].dst.stride,
391
                          xd->plane[1].eobs);
392
    xd->itxm_add_uv_block(xd->plane[2].qcoeff, xd->block[16].dequant,
393
                          xd->plane[2].dst.buf, xd->plane[1].dst.stride,
394
                          xd->plane[2].eobs);
395 396 397
  }
}

398 399 400 401
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;
402 403 404
  int n;

  for (n = 0; n < y_count; n++) {
405 406
    const int x_idx = n & (bw - 1);
    const int y_idx = n >> bwl;
407
    const int y_offset = (y_idx * 32) * mb->plane[0].dst.stride + (x_idx * 32);
408 409
    vp9_dequant_idct_add_32x32(BLOCK_OFFSET(mb->plane[0].qcoeff, n, 1024),
                               mb->block[0].dequant ,
410 411
                               mb->plane[0].dst.buf + y_offset,
                               mb->plane[0].dst.stride,
412 413 414 415 416 417 418 419 420 421 422 423
                               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);
424 425
     const int uv_offset = (y_idx * 32) * mb->plane[1].dst.stride +
         (x_idx * 32);
426 427
     vp9_dequant_idct_add_32x32(BLOCK_OFFSET(mb->plane[1].qcoeff, n, 1024),
                                mb->block[16].dequant,
428 429 430
                                mb->plane[1].dst.buf + uv_offset,
                                mb->plane[1].dst.stride,
                                mb->plane[1].eobs[n * 64]);
431 432
     vp9_dequant_idct_add_32x32(BLOCK_OFFSET(mb->plane[2].qcoeff, n, 1024),
                                mb->block[20].dequant,
433 434 435
                                mb->plane[2].dst.buf + uv_offset,
                                mb->plane[1].dst.stride,
                                mb->plane[2].eobs[n * 64]);
436 437 438 439 440 441 442 443 444 445 446 447
  }
}

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;
448
    const int y_offset = (y_idx * 16) * mb->plane[0].dst.stride + (x_idx * 16);
449
    const TX_TYPE tx_type = get_tx_type_16x16(mb,
450
                                (y_idx * (4 * bw) + x_idx) * 4);
451 452 453
    vp9_dequant_iht_add_16x16_c(tx_type,
                                BLOCK_OFFSET(mb->plane[0].qcoeff, n, 256),
                                mb->block[0].dequant,
454 455
                                mb->plane[0].dst.buf + y_offset,
                                mb->plane[0].dst.stride,
456
                                mb->plane[0].eobs[n * 16]);
457
  }
458 459 460 461 462 463 464 465 466
}

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);
467 468

  for (n = 0; n < uv_count; n++) {
469 470
    const int x_idx = n & (bw - 1);
    const int y_idx = n >> (bwl - 1);
471
    const int uv_offset = (y_idx * 16) * mb->plane[1].dst.stride + (x_idx * 16);
472
    vp9_dequant_idct_add_16x16(BLOCK_OFFSET(mb->plane[1].qcoeff, n, 256),
473
                               mb->block[16].dequant,
474 475
                               mb->plane[1].dst.buf + uv_offset,
                               mb->plane[1].dst.stride,
John Koleszar's avatar
John Koleszar committed
476
                               mb->plane[1].eobs[n * 16]);
477
    vp9_dequant_idct_add_16x16(BLOCK_OFFSET(mb->plane[2].qcoeff, n, 256),
478
                               mb->block[20].dequant,
479 480
                               mb->plane[2].dst.buf + uv_offset,
                               mb->plane[1].dst.stride,
John Koleszar's avatar
John Koleszar committed
481
                               mb->plane[2].eobs[n * 16]);
482 483 484
  }
}

485 486 487 488
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;
489 490 491 492
  int n;

  // luma
  for (n = 0; n < y_count; n++) {
493 494
    const int x_idx = n & (bw - 1);
    const int y_idx = n >> bwl;
495
    const int y_offset = (y_idx * 8) * xd->plane[0].dst.stride + (x_idx * 8);
496
    const TX_TYPE tx_type = get_tx_type_8x8(xd,
497
                                            (y_idx * (2 * bw) + x_idx) * 2);
498 499 500 501

    vp9_dequant_iht_add_8x8_c(tx_type,
                              BLOCK_OFFSET(xd->plane[0].qcoeff, n, 64),
                              xd->block[0].dequant,
502 503
                              xd->plane[0].dst.buf + y_offset,
                              xd->plane[0].dst.stride,
504
                              xd->plane[0].eobs[n * 4]);
505
  }
506 507 508 509 510 511 512
}

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;
513 514 515

  // chroma
  for (n = 0; n < uv_count; n++) {
516 517
    const int x_idx = n & (bw - 1);
    const int y_idx = n >> (bwl - 1);
518
    const int uv_offset = (y_idx * 8) * xd->plane[1].dst.stride + (x_idx * 8);
519 520
    vp9_dequant_idct_add_8x8(BLOCK_OFFSET(xd->plane[1].qcoeff, n, 64),
                             xd->block[16].dequant,
521 522
                             xd->plane[1].dst.buf + uv_offset,
                             xd->plane[1].dst.stride,
523 524 525
                             xd->plane[1].eobs[n * 4]);
    vp9_dequant_idct_add_8x8(BLOCK_OFFSET(xd->plane[2].qcoeff, n, 64),
                             xd->block[20].dequant,
526 527
                             xd->plane[2].dst.buf + uv_offset,
                             xd->plane[1].dst.stride,
528
                             xd->plane[2].eobs[n * 4]);
529 530 531
  }
}

532 533 534 535
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;
536 537 538
  int n;

  for (n = 0; n < y_count; n++) {
539 540
    const int x_idx = n & (bw - 1);
    const int y_idx = n >> bwl;
541
    const int y_offset = (y_idx * 4) * xd->plane[0].dst.stride + (x_idx * 4);
542
    const TX_TYPE tx_type = get_tx_type_4x4(xd, n);
543
    if (tx_type == DCT_DCT) {
544 545
      xd->itxm_add(BLOCK_OFFSET(xd->plane[0].qcoeff, n, 16),
                   xd->block[0].dequant,
546
                   xd->plane[0].dst.buf + y_offset, xd->plane[0].dst.stride,
John Koleszar's avatar
John Koleszar committed
547
                   xd->plane[0].eobs[n]);
548
    } else {
549 550
      vp9_dequant_iht_add_c(tx_type,
                            BLOCK_OFFSET(xd->plane[0].qcoeff, n, 16),
551 552 553
                            xd->block[0].dequant,
                            xd->plane[0].dst.buf + y_offset,
                            xd->plane[0].dst.stride, xd->plane[0].eobs[n]);
554 555
    }
  }
556 557 558 559 560 561 562
}

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;
563 564

  for (n = 0; n < uv_count; n++) {
565 566
    const int x_idx = n & (bw - 1);
    const int y_idx = n >> (bwl - 1);
567
    const int uv_offset = (y_idx * 4) * xd->plane[1].dst.stride + (x_idx * 4);
568 569
    xd->itxm_add(BLOCK_OFFSET(xd->plane[1].qcoeff, n, 16),
        xd->block[16].dequant,
570 571
        xd->plane[1].dst.buf + uv_offset, xd->plane[1].dst.stride,
        xd->plane[1].eobs[n]);
572 573
    xd->itxm_add(BLOCK_OFFSET(xd->plane[2].qcoeff, n, 16),
        xd->block[20].dequant,
574 575
        xd->plane[2].dst.buf + uv_offset, xd->plane[1].dst.stride,
        xd->plane[2].eobs[n]);
576 577 578
  }
}

579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597
// 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,
598
                      vp9_reader *r, BLOCK_SIZE_TYPE bsize) {
599 600
  const int bwl = mb_width_log2(bsize), bhl = mb_height_log2(bsize);
  const int bw = 1 << bwl, bh = 1 << bhl;
601
  int n, eobtotal;
602
  VP9_COMMON *const pc = &pbi->common;
603
  MODE_INFO *mi = xd->mode_info_context;
604
  const int mis = pc->mode_info_stride;
605

606
  assert(mi->mbmi.sb_type == bsize);
Ronald S. Bultje's avatar
Ronald S. Bultje committed
607 608

  if (pbi->common.frame_type != KEY_FRAME)
609
    vp9_setup_interp_filters(xd, mi->mbmi.interp_filter, pc);
Ronald S. Bultje's avatar
Ronald S. Bultje committed
610

611
  // generate prediction
612 613 614
  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
615
  } else {
616
    vp9_build_inter_predictors_sb(xd, mb_row, mb_col, bsize);
Ronald S. Bultje's avatar
Ronald S. Bultje committed
617 618
  }

Yunqing Wang's avatar
Yunqing Wang committed
619 620
  if (mi->mbmi.mb_skip_coeff) {
    vp9_reset_sb_tokens_context(xd, bsize);
621
  } else {
Yunqing Wang's avatar
Yunqing Wang committed
622 623 624 625 626
    // re-initialize macroblock dequantizer before detokenization
    if (xd->segmentation_enabled)
      mb_init_dequantizer(pbi, xd);

    // dequantization and idct
627
    eobtotal = vp9_decode_tokens(pbi, xd, r, bsize);
Yunqing Wang's avatar
Yunqing Wang committed
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
    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);
      }
653
    }
654
  }
655 656
}

657 658
// TODO(jingning): Need to merge SB and MB decoding. The MB decoding currently
// couples special handles on I8x8, B_PRED, and splitmv modes.
659 660
static void decode_mb(VP9D_COMP *pbi, MACROBLOCKD *xd,
                     int mb_row, int mb_col,
661
                     vp9_reader *r) {
John Koleszar's avatar
John Koleszar committed
662
  int eobtotal = 0;
663 664
  const MB_PREDICTION_MODE mode = xd->mode_info_context->mbmi.mode;
  const int tx_size = xd->mode_info_context->mbmi.txfm_size;
665

Ronald S. Bultje's avatar
Ronald S. Bultje committed
666
  assert(!xd->mode_info_context->mbmi.sb_type);
John Koleszar's avatar
John Koleszar committed
667

Deb Mukherjee's avatar
Deb Mukherjee committed
668
  //mode = xd->mode_info_context->mbmi.mode;
669
  if (pbi->common.frame_type != KEY_FRAME)
670
    vp9_setup_interp_filters(xd, xd->mode_info_context->mbmi.interp_filter,
671
                             &pbi->common);
Gaute Strokkenes's avatar
Gaute Strokkenes committed
672

673
  // do prediction
John Koleszar's avatar
John Koleszar committed
674 675
  if (xd->mode_info_context->mbmi.ref_frame == INTRA_FRAME) {
    if (mode != I8X8_PRED) {
676
      vp9_build_intra_predictors_sbuv_s(xd, BLOCK_SIZE_MB16X16);
Yaowu Xu's avatar
Yaowu Xu committed
677
      if (mode != I4X4_PRED)
678
        vp9_build_intra_predictors_sby_s(xd, BLOCK_SIZE_MB16X16);
John Koleszar's avatar
John Koleszar committed
679
    }
John Koleszar's avatar
John Koleszar committed
680
  } else {
681
#if 0  // def DEC_DEBUG
682 683 684 685 686
  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
687
    vp9_build_inter_predictors_mb(xd, mb_row, mb_col);
John Koleszar's avatar
John Koleszar committed
688 689
  }

Yunqing Wang's avatar
Yunqing Wang committed
690 691
  if (xd->mode_info_context->mbmi.mb_skip_coeff) {
    vp9_reset_sb_tokens_context(xd, BLOCK_SIZE_MB16X16);
692
  } else {
Yunqing Wang's avatar
Yunqing Wang committed
693 694 695 696
    // re-initialize macroblock dequantizer before detokenization
    if (xd->segmentation_enabled)
      mb_init_dequantizer(pbi, xd);

697
    if (!vp9_reader_has_error(r)) {
Yunqing Wang's avatar
Yunqing Wang committed
698 699 700
#if CONFIG_NEWBINTRAMODES
    if (mode != I4X4_PRED)
#endif
701
      eobtotal = vp9_decode_tokens(pbi, xd, r, BLOCK_SIZE_MB16X16);
Yunqing Wang's avatar
Yunqing Wang committed
702 703 704 705 706 707 708
    }
  }

  if (eobtotal == 0 &&
      mode != I4X4_PRED &&
      mode != SPLITMV &&
      mode != I8X8_PRED &&
709
      !vp9_reader_has_error(r)) {
Yunqing Wang's avatar
Yunqing Wang committed
710 711 712 713 714 715 716 717
    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) {
718
      decode_16x16(xd);
Yunqing Wang's avatar
Yunqing Wang committed
719
    } else if (tx_size == TX_8X8) {
720
      decode_8x8(xd);
Yunqing Wang's avatar
Yunqing Wang committed
721
    } else {
722
      decode_4x4(pbi, xd, r);
Yunqing Wang's avatar
Yunqing Wang committed
723
    }
724
  }
Yunqing Wang's avatar
Yunqing Wang committed
725

726 727 728 729
#ifdef DEC_DEBUG
  if (dec_debug) {
    int i, j;
    printf("\n");
730 731 732 733 734 735 736
    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");
737
    printf("final y\n");
John Koleszar's avatar
John Koleszar committed
738
    for (i = 0; i < 16; i++) {
739
      for (j = 0; j < 16; j++)
740
        printf("%3d ", xd->plane[0].dst.buf[i * xd->plane[0].dst.stride + j]);
741
      printf("\n");
John Koleszar's avatar
John Koleszar committed
742
    }
743 744 745 746
    printf("\n");
    printf("final u\n");
    for (i = 0; i < 8; i++) {
      for (j = 0; j < 8; j++)
747
        printf("%3d ", xd->plane[1].dst.buf[i * xd->plane[1].dst.stride + j]);
748
      printf("\n");
749
    }
750 751 752 753
    printf("\n");
    printf("final v\n");
    for (i = 0; i < 8; i++) {
      for (j = 0; j < 8; j++)
754
        printf("%3d ", xd->plane[2].dst.buf[i <