vp9_decodframe.c 53.6 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

209 210 211
  mb->plane[0].dequant = pc->y_dequant[qindex];
  for (i = 1; i < MAX_MB_PLANE; i++)
    mb->plane[i].dequant = pc->uv_dequant[qindex];
John Koleszar's avatar
John Koleszar committed
212

Dmitry Kovalev's avatar
Dmitry Kovalev committed
213
  if (mb->lossless) {
Dmitry Kovalev's avatar
Dmitry Kovalev committed
214
    assert(qindex == 0);
Yaowu Xu's avatar
Yaowu Xu committed
215 216
    mb->inv_txm4x4_1      = vp9_short_iwalsh4x4_1;
    mb->inv_txm4x4        = vp9_short_iwalsh4x4;
217 218 219
    mb->itxm_add          = vp9_idct_add_lossless_c;
    mb->itxm_add_y_block  = vp9_idct_add_y_block_lossless_c;
    mb->itxm_add_uv_block = vp9_idct_add_uv_block_lossless_c;
Dmitry Kovalev's avatar
Dmitry Kovalev committed
220
  } else {
Yaowu Xu's avatar
Yaowu Xu committed
221 222
    mb->inv_txm4x4_1      = vp9_short_idct4x4_1;
    mb->inv_txm4x4        = vp9_short_idct4x4;
223 224 225
    mb->itxm_add          = vp9_idct_add;
    mb->itxm_add_y_block  = vp9_idct_add_y_block;
    mb->itxm_add_uv_block = vp9_idct_add_uv_block;
John Koleszar's avatar
John Koleszar committed
226
  }
John Koleszar's avatar
John Koleszar committed
227 228
}

229
static void decode_16x16(MACROBLOCKD *xd) {
230
  const TX_TYPE tx_type = get_tx_type_16x16(xd, 0);
231

232 233
  vp9_iht_add_16x16_c(tx_type, xd->plane[0].qcoeff, xd->plane[0].dst.buf,
                      xd->plane[0].dst.stride, xd->plane[0].eobs[0]);
234

235 236
  vp9_idct_add_8x8(xd->plane[1].qcoeff, xd->plane[1].dst.buf,
                   xd->plane[1].dst.stride, xd->plane[1].eobs[0]);
237

238 239
  vp9_idct_add_8x8(xd->plane[2].qcoeff, xd->plane[2].dst.buf,
                   xd->plane[1].dst.stride, xd->plane[2].eobs[0]);
240 241
}

242
static void decode_8x8(MACROBLOCKD *xd) {
243 244
  const MB_PREDICTION_MODE mode = xd->mode_info_context->mbmi.mode;
  // luma
245
  // if the first one is DCT_DCT assume all the rest are as well
246
  TX_TYPE tx_type = get_tx_type_8x8(xd, 0);
247
  if (tx_type != DCT_DCT || mode == I8X8_PRED) {
248 249 250 251
    int i;
    for (i = 0; i < 4; i++) {
      int ib = vp9_i8x8_block[i];
      int idx = (ib & 0x02) ? (ib + 2) : ib;
252
      int16_t *q  = BLOCK_OFFSET(xd->plane[0].qcoeff, idx, 16);
253
      uint8_t *dst = *(xd->block[ib].base_dst) + xd->block[ib].dst;
254
      int stride = xd->plane[0].dst.stride;
255
      if (mode == I8X8_PRED) {
256
        BLOCKD *b = &xd->block[ib];
257
        int i8x8mode = b->bmi.as_mode.first;
258
        vp9_intra8x8_predict(xd, b, i8x8mode, dst, stride);
259
      }
260
      tx_type = get_tx_type_8x8(xd, ib);
261
      vp9_iht_add_8x8_c(tx_type, q, dst, stride, xd->plane[0].eobs[idx]);
262
    }
263
  } else {
264 265
    vp9_idct_add_y_block_8x8(xd->plane[0].qcoeff, 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->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->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) {
291 292 293 294
    xd->itxm_add_uv_block(xd->plane[1].qcoeff, xd->plane[1].dst.buf,
        xd->plane[1].dst.stride, xd->plane[1].eobs);
    xd->itxm_add_uv_block(xd->plane[2].qcoeff, xd->plane[2].dst.buf,
        xd->plane[1].dst.stride, xd->plane[2].eobs);
295
  } else {
296 297
    vp9_idct_add_8x8(xd->plane[1].qcoeff, xd->plane[1].dst.buf,
                     xd->plane[1].dst.stride, xd->plane[1].eobs[0]);
298

299 300
    vp9_idct_add_8x8(xd->plane[2].qcoeff, xd->plane[2].dst.buf,
                     xd->plane[1].dst.stride, xd->plane[2].eobs[0]);
301 302 303
  }
}

304 305
static INLINE void dequant_add_y(MACROBLOCKD *xd, TX_TYPE tx_type, int idx) {
  BLOCKD *const b = &xd->block[idx];
306
  struct macroblockd_plane *const y = &xd->plane[0];
307
  if (tx_type != DCT_DCT) {
308 309
    vp9_iht_add_c(tx_type, BLOCK_OFFSET(y->qcoeff, idx, 16),
                  *(b->base_dst) + b->dst, b->dst_stride, y->eobs[idx]);
310
  } else {
311
    xd->itxm_add(BLOCK_OFFSET(y->qcoeff, idx, 16), *(b->base_dst) + b->dst,
312 313 314 315 316
                 b->dst_stride, y->eobs[idx]);
  }
}


317
static void decode_4x4(VP9D_COMP *pbi, MACROBLOCKD *xd, vp9_reader *r) {
318
  TX_TYPE tx_type;
319
  int i = 0;
320
  const MB_PREDICTION_MODE mode = xd->mode_info_context->mbmi.mode;
321 322 323 324 325
  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
326 327
      BLOCKD *b = &xd->block[ib];
      int i8x8mode = b->bmi.as_mode.first;
328 329
      vp9_intra8x8_predict(xd, b, i8x8mode, *(b->base_dst) + b->dst,
                           b->dst_stride);
330
      for (j = 0; j < 4; j++) {
331
        tx_type = get_tx_type_4x4(xd, ib + iblock[j]);
332
        dequant_add_y(xd, tx_type, ib + iblock[j]);
333 334
      }
      b = &xd->block[16 + i];
335 336
      vp9_intra_uv4x4_predict(xd, b, i8x8mode, *(b->base_dst) + b->dst,
                              b->dst_stride);
337
      xd->itxm_add(BLOCK_OFFSET(xd->plane[1].qcoeff, i, 16),
338
                   *(b->base_dst) + b->dst, b->dst_stride,
John Koleszar's avatar
John Koleszar committed
339
                   xd->plane[1].eobs[i]);
340
      b = &xd->block[20 + i];
341 342
      vp9_intra_uv4x4_predict(xd, b, i8x8mode, *(b->base_dst) + b->dst,
                              b->dst_stride);
343
      xd->itxm_add(BLOCK_OFFSET(xd->plane[2].qcoeff, i, 16),
344
                   *(b->base_dst) + b->dst, b->dst_stride,
John Koleszar's avatar
John Koleszar committed
345
                   xd->plane[2].eobs[i]);
346
    }
Yaowu Xu's avatar
Yaowu Xu committed
347
  } else if (mode == I4X4_PRED) {
348 349
    for (i = 0; i < 16; i++) {
      BLOCKD *b = &xd->block[i];
Dmitry Kovalev's avatar
Dmitry Kovalev committed
350
      int b_mode = xd->mode_info_context->bmi[i].as_mode.first;
351 352
#if CONFIG_NEWBINTRAMODES
      xd->mode_info_context->bmi[i].as_mode.context = b->bmi.as_mode.context =
353
          vp9_find_bpred_context(xd, b);
354
      if (!xd->mode_info_context->mbmi.mb_skip_coeff)
355
        vp9_decode_coefs_4x4(pbi, xd, r, PLANE_TYPE_Y_WITH_DC, i);
356
#endif
357 358
      vp9_intra4x4_predict(xd, b, b_mode, *(b->base_dst) + b->dst,
                           b->dst_stride);
359
      tx_type = get_tx_type_4x4(xd, i);
360
      dequant_add_y(xd, tx_type, i);
361
    }
362 363
#if CONFIG_NEWBINTRAMODES
    if (!xd->mode_info_context->mbmi.mb_skip_coeff)
364
      vp9_decode_mb_tokens_4x4_uv(pbi, xd, r);
365
#endif
366
    vp9_build_intra_predictors_sbuv_s(xd, BLOCK_SIZE_MB16X16);
367 368 369 370
    xd->itxm_add_uv_block(xd->plane[1].qcoeff, xd->plane[1].dst.buf,
        xd->plane[1].dst.stride, xd->plane[1].eobs);
    xd->itxm_add_uv_block(xd->plane[2].qcoeff, xd->plane[2].dst.buf,
        xd->plane[1].dst.stride, xd->plane[2].eobs);
371
  } else if (mode == SPLITMV || get_tx_type_4x4(xd, 0) == DCT_DCT) {
372 373 374 375 376 377
    xd->itxm_add_y_block(xd->plane[0].qcoeff, xd->plane[0].dst.buf,
        xd->plane[0].dst.stride, xd);
    xd->itxm_add_uv_block(xd->plane[1].qcoeff, xd->plane[1].dst.buf,
        xd->plane[1].dst.stride, xd->plane[1].eobs);
    xd->itxm_add_uv_block(xd->plane[2].qcoeff, xd->plane[2].dst.buf,
        xd->plane[1].dst.stride, xd->plane[2].eobs);
378
  } else {
379
    for (i = 0; i < 16; i++) {
380
      tx_type = get_tx_type_4x4(xd, i);
381
      dequant_add_y(xd, tx_type, i);
382
    }
383 384 385 386
    xd->itxm_add_uv_block(xd->plane[1].qcoeff, xd->plane[1].dst.buf,
                          xd->plane[1].dst.stride, xd->plane[1].eobs);
    xd->itxm_add_uv_block(xd->plane[2].qcoeff, xd->plane[2].dst.buf,
                          xd->plane[1].dst.stride, xd->plane[2].eobs);
387 388 389
  }
}

390
static INLINE void decode_sby_32x32(MACROBLOCKD *mb, BLOCK_SIZE_TYPE bsize) {
391 392
  const int bwl = b_width_log2(bsize) - 3, bw = 1 << bwl;
  const int bhl = b_height_log2(bsize) - 3, bh = 1 << bhl;
393
  const int y_count = bw * bh;
394 395 396
  int n;

  for (n = 0; n < y_count; n++) {
397 398
    const int x_idx = n & (bw - 1);
    const int y_idx = n >> bwl;
399
    const int y_offset = (y_idx * 32) * mb->plane[0].dst.stride + (x_idx * 32);
400 401 402 403
    vp9_idct_add_32x32(BLOCK_OFFSET(mb->plane[0].qcoeff, n, 1024),
                       mb->plane[0].dst.buf + y_offset,
                       mb->plane[0].dst.stride,
                       mb->plane[0].eobs[n * 64]);
404 405 406 407
  }
}

static INLINE void decode_sbuv_32x32(MACROBLOCKD *mb, BLOCK_SIZE_TYPE bsize) {
408 409
  const int bwl = b_width_log2(bsize) - 3, bw = (1 << bwl) / 2;
  const int bhl = b_height_log2(bsize) - 3, bh = (1 << bhl) / 2;
410 411 412 413 414
  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);
415 416
     const int uv_offset = (y_idx * 32) * mb->plane[1].dst.stride +
         (x_idx * 32);
417 418 419 420 421 422 423 424
     vp9_idct_add_32x32(BLOCK_OFFSET(mb->plane[1].qcoeff, n, 1024),
                        mb->plane[1].dst.buf + uv_offset,
                        mb->plane[1].dst.stride,
                        mb->plane[1].eobs[n * 64]);
     vp9_idct_add_32x32(BLOCK_OFFSET(mb->plane[2].qcoeff, n, 1024),
                        mb->plane[2].dst.buf + uv_offset,
                        mb->plane[1].dst.stride,
                        mb->plane[2].eobs[n * 64]);
425 426 427 428
  }
}

static INLINE void decode_sby_16x16(MACROBLOCKD *mb, BLOCK_SIZE_TYPE bsize) {
429 430
  const int bwl = b_width_log2(bsize) - 2, bw = 1 << bwl;
  const int bhl = b_height_log2(bsize) - 2, bh = 1 << bhl;
431 432 433 434 435 436
  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;
437
    const int y_offset = (y_idx * 16) * mb->plane[0].dst.stride + (x_idx * 16);
438
    const TX_TYPE tx_type = get_tx_type_16x16(mb,
439
                                (y_idx * (4 * bw) + x_idx) * 4);
440 441 442
    vp9_iht_add_16x16_c(tx_type, BLOCK_OFFSET(mb->plane[0].qcoeff, n, 256),
                        mb->plane[0].dst.buf + y_offset,
                        mb->plane[0].dst.stride, mb->plane[0].eobs[n * 16]);
443
  }
444 445 446
}

static INLINE void decode_sbuv_16x16(MACROBLOCKD *mb, BLOCK_SIZE_TYPE bsize) {
447 448
  const int bwl = b_width_log2(bsize) - 2, bw = (1 << bwl) / 2;
  const int bhl = b_height_log2(bsize) - 2, bh = (1 << bhl) / 2;
449 450 451 452
  const int uv_count = bw * bh;
  int n;

  assert(bsize >= BLOCK_SIZE_SB32X32);
453 454

  for (n = 0; n < uv_count; n++) {
455 456
    const int x_idx = n & (bw - 1);
    const int y_idx = n >> (bwl - 1);
457
    const int uv_offset = (y_idx * 16) * mb->plane[1].dst.stride + (x_idx * 16);
458 459 460 461 462 463
    vp9_idct_add_16x16(BLOCK_OFFSET(mb->plane[1].qcoeff, n, 256),
                       mb->plane[1].dst.buf + uv_offset,
                       mb->plane[1].dst.stride, mb->plane[1].eobs[n * 16]);
    vp9_idct_add_16x16(BLOCK_OFFSET(mb->plane[2].qcoeff, n, 256),
                       mb->plane[2].dst.buf + uv_offset,
                       mb->plane[1].dst.stride, mb->plane[2].eobs[n * 16]);
464 465 466
  }
}

467
static INLINE void decode_sby_8x8(MACROBLOCKD *xd, BLOCK_SIZE_TYPE bsize) {
468 469
  const int bwl = b_width_log2(bsize)  - 1, bw = 1 << bwl;
  const int bhl = b_height_log2(bsize) - 1, bh = 1 << bhl;
470
  const int y_count = bw * bh;
471 472 473 474
  int n;

  // luma
  for (n = 0; n < y_count; n++) {
475 476
    const int x_idx = n & (bw - 1);
    const int y_idx = n >> bwl;
477
    const int y_offset = (y_idx * 8) * xd->plane[0].dst.stride + (x_idx * 8);
478
    const TX_TYPE tx_type = get_tx_type_8x8(xd,
479
                                            (y_idx * (2 * bw) + x_idx) * 2);
480

481 482 483
    vp9_iht_add_8x8_c(tx_type, BLOCK_OFFSET(xd->plane[0].qcoeff, n, 64),
                      xd->plane[0].dst.buf + y_offset, xd->plane[0].dst.stride,
                      xd->plane[0].eobs[n * 4]);
484
  }
485 486 487
}

static INLINE void decode_sbuv_8x8(MACROBLOCKD *xd, BLOCK_SIZE_TYPE bsize) {
488 489
  const int bwl = b_width_log2(bsize)  - 1, bw = 1 << (bwl - 1);
  const int bhl = b_height_log2(bsize) - 1, bh = 1 << (bhl - 1);
490 491
  const int uv_count = bw * bh;
  int n;
492 493 494

  // chroma
  for (n = 0; n < uv_count; n++) {
495 496
    const int x_idx = n & (bw - 1);
    const int y_idx = n >> (bwl - 1);
497
    const int uv_offset = (y_idx * 8) * xd->plane[1].dst.stride + (x_idx * 8);
498 499 500 501 502 503
    vp9_idct_add_8x8(BLOCK_OFFSET(xd->plane[1].qcoeff, n, 64),
                     xd->plane[1].dst.buf + uv_offset, xd->plane[1].dst.stride,
                     xd->plane[1].eobs[n * 4]);
    vp9_idct_add_8x8(BLOCK_OFFSET(xd->plane[2].qcoeff, n, 64),
                     xd->plane[2].dst.buf + uv_offset, xd->plane[1].dst.stride,
                     xd->plane[2].eobs[n * 4]);
504 505 506
  }
}

507
static INLINE void decode_sby_4x4(MACROBLOCKD *xd, BLOCK_SIZE_TYPE bsize) {
508 509
  const int bwl = b_width_log2(bsize), bw = 1 << bwl;
  const int bhl = b_height_log2(bsize), bh = 1 << bhl;
510
  const int y_count = bw * bh;
511 512 513
  int n;

  for (n = 0; n < y_count; n++) {
514 515
    const int x_idx = n & (bw - 1);
    const int y_idx = n >> bwl;
516
    const int y_offset = (y_idx * 4) * xd->plane[0].dst.stride + (x_idx * 4);
517
    const TX_TYPE tx_type = get_tx_type_4x4(xd, n);
518
    if (tx_type == DCT_DCT) {
519
      xd->itxm_add(BLOCK_OFFSET(xd->plane[0].qcoeff, n, 16),
520
                   xd->plane[0].dst.buf + y_offset, xd->plane[0].dst.stride,
John Koleszar's avatar
John Koleszar committed
521
                   xd->plane[0].eobs[n]);
522
    } else {
523 524 525
      vp9_iht_add_c(tx_type, BLOCK_OFFSET(xd->plane[0].qcoeff, n, 16),
                    xd->plane[0].dst.buf + y_offset, xd->plane[0].dst.stride,
                    xd->plane[0].eobs[n]);
526 527
    }
  }
528 529 530
}

static INLINE void decode_sbuv_4x4(MACROBLOCKD *xd, BLOCK_SIZE_TYPE bsize) {
531 532
  const int bwl = b_width_log2(bsize), bw = 1 << (bwl - 1);
  const int bhl = b_height_log2(bsize), bh = 1 << (bhl - 1);
533 534
  const int uv_count = bw * bh;
  int n;
535 536

  for (n = 0; n < uv_count; n++) {
537 538
    const int x_idx = n & (bw - 1);
    const int y_idx = n >> (bwl - 1);
539
    const int uv_offset = (y_idx * 4) * xd->plane[1].dst.stride + (x_idx * 4);
540
    xd->itxm_add(BLOCK_OFFSET(xd->plane[1].qcoeff, n, 16),
541 542
        xd->plane[1].dst.buf + uv_offset, xd->plane[1].dst.stride,
        xd->plane[1].eobs[n]);
543
    xd->itxm_add(BLOCK_OFFSET(xd->plane[2].qcoeff, n, 16),
544 545
        xd->plane[2].dst.buf + uv_offset, xd->plane[1].dst.stride,
        xd->plane[2].eobs[n]);
546 547 548
  }
}

549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567
// 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,
568
                      vp9_reader *r, BLOCK_SIZE_TYPE bsize) {
569 570
  const int bwl = mb_width_log2(bsize), bhl = mb_height_log2(bsize);
  const int bw = 1 << bwl, bh = 1 << bhl;
571
  int n, eobtotal;
572
  VP9_COMMON *const pc = &pbi->common;
573
  MODE_INFO *mi = xd->mode_info_context;
574
  const int mis = pc->mode_info_stride;
575

576
  assert(mi->mbmi.sb_type == bsize);
Ronald S. Bultje's avatar
Ronald S. Bultje committed
577 578

  if (pbi->common.frame_type != KEY_FRAME)
579
    vp9_setup_interp_filters(xd, mi->mbmi.interp_filter, pc);
Ronald S. Bultje's avatar
Ronald S. Bultje committed
580

581
  // generate prediction
582 583 584
  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
585
  } else {
586
    vp9_build_inter_predictors_sb(xd, mb_row, mb_col, bsize);
Ronald S. Bultje's avatar
Ronald S. Bultje committed
587 588
  }

Yunqing Wang's avatar
Yunqing Wang committed
589 590
  if (mi->mbmi.mb_skip_coeff) {
    vp9_reset_sb_tokens_context(xd, bsize);
591
  } else {
Yunqing Wang's avatar
Yunqing Wang committed
592 593 594 595 596
    // re-initialize macroblock dequantizer before detokenization
    if (xd->segmentation_enabled)
      mb_init_dequantizer(pbi, xd);

    // dequantization and idct
597
    eobtotal = vp9_decode_tokens(pbi, xd, r, bsize);
Yunqing Wang's avatar
Yunqing Wang committed
598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622
    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);
      }
623
    }
624
  }
625 626
}

627 628
// TODO(jingning): Need to merge SB and MB decoding. The MB decoding currently
// couples special handles on I8x8, B_PRED, and splitmv modes.
629 630
static void decode_mb(VP9D_COMP *pbi, MACROBLOCKD *xd,
                     int mb_row, int mb_col,
631
                     vp9_reader *r) {
John Koleszar's avatar
John Koleszar committed
632
  int eobtotal = 0;
633 634
  const MB_PREDICTION_MODE mode = xd->mode_info_context->mbmi.mode;
  const int tx_size = xd->mode_info_context->mbmi.txfm_size;
635

636
  assert(xd->mode_info_context->mbmi.sb_type == BLOCK_SIZE_MB16X16);
John Koleszar's avatar
John Koleszar committed
637

Deb Mukherjee's avatar
Deb Mukherjee committed
638
  //mode = xd->mode_info_context->mbmi.mode;
639
  if (pbi->common.frame_type != KEY_FRAME)
640
    vp9_setup_interp_filters(xd, xd->mode_info_context->mbmi.interp_filter,
641
                             &pbi->common);
Gaute Strokkenes's avatar
Gaute Strokkenes committed
642

643
  // do prediction
John Koleszar's avatar
John Koleszar committed
644 645
  if (xd->mode_info_context->mbmi.ref_frame == INTRA_FRAME) {
    if (mode != I8X8_PRED) {
646
      vp9_build_intra_predictors_sbuv_s(xd, BLOCK_SIZE_MB16X16);
Yaowu Xu's avatar
Yaowu Xu committed
647
      if (mode != I4X4_PRED)
648
        vp9_build_intra_predictors_sby_s(xd, BLOCK_SIZE_MB16X16);
John Koleszar's avatar
John Koleszar committed
649
    }
John Koleszar's avatar
John Koleszar committed
650
  } else {
651
#if 0  // def DEC_DEBUG
652 653 654 655 656
  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
657
    vp9_build_inter_predictors_sb(xd, mb_row, mb_col, BLOCK_SIZE_MB16X16);
John Koleszar's avatar
John Koleszar committed
658 659
  }

Yunqing Wang's avatar
Yunqing Wang committed
660 661
  if (xd->mode_info_context->mbmi.mb_skip_coeff) {
    vp9_reset_sb_tokens_context(xd, BLOCK_SIZE_MB16X16);
662
  } else {
Yunqing Wang's avatar
Yunqing Wang committed
663 664 665 666
    // re-initialize macroblock dequantizer before detokenization
    if (xd->segmentation_enabled)
      mb_init_dequantizer(pbi, xd);

667
    if (!vp9_reader_has_error(r)) {
Yunqing Wang's avatar
Yunqing Wang committed
668 669 670
#if CONFIG_NEWBINTRAMODES
    if (mode != I4X4_PRED)
#endif
671
      eobtotal = vp9_decode_tokens(pbi, xd, r, BLOCK_SIZE_MB16X16);
Yunqing Wang's avatar
Yunqing Wang committed
672 673 674 675 676 677 678
    }
  }

  if (eobtotal == 0 &&
      mode != I4X4_PRED &&
      mode != SPLITMV &&
      mode != I8X8_PRED &&
679
      !vp9_reader_has_error(r)) {
Yunqing Wang's avatar
Yunqing Wang committed
680 681 682 683 684 685 686 687
    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) {
688
      decode_16x16(xd);
Yunqing Wang's avatar
Yunqing Wang committed
689
    } else if (tx_size == TX_8X8) {
690
      decode_8x8(xd);
Yunqing Wang's avatar
Yunqing Wang committed
691
    } else {
692
      decode_4x4(pbi, xd, r);
Yunqing Wang's avatar
Yunqing Wang committed
693
    }
694
  }
Yunqing Wang's avatar
Yunqing Wang committed
695

696 697 698 699
#ifdef DEC_DEBUG
  if (dec_debug) {
    int i, j;
    printf("\n");
700 701 702 703 704 705 706
    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");
707
    printf("final y\n");
John Koleszar's avatar
John Koleszar committed
708
    for (i = 0; i < 16; i++) {
709
      for (j = 0; j < 16; j++)
710
        printf("%3d ", xd->plane[0].dst.buf[i * xd->plane[0].dst.stride + j]);
711
      printf("\n");
John Koleszar's avatar
John Koleszar committed
712
    }
713 714 715 716
    printf("\n");
    printf("final u\n");
    for (i = 0; i < 8; i++) {
      for (j = 0; j < 8; j++)
717
        printf("%3d ", xd->plane[1].dst.buf[i * xd->plane[1].dst.stride + j]);
718
      printf("\n");
719
    }
720 721 722 723
    printf("\n");
    printf("final v\n");
    for (i = 0; i < 8; i++) {
      for (j = 0; j < 8; j++)
724
        printf("%3d ", xd->plane[2].dst.buf[i * xd->plane[1].dst.stride + j]);
725
      printf("\n");
726
    }
727
    fflush(stdout);
John Koleszar's avatar
John Koleszar committed
728
  }
729
#endif
Yaowu Xu's avatar
Yaowu Xu committed
730
}
John Koleszar's avatar
John Koleszar committed
731

732 733 734 735
static int get_delta_q(vp9_reader *r, int *dq) {
  const int old_value = *dq;

  if (vp9_read_bit(r)) {  // Update bit
736 737
    const int value = vp9_read_literal(r, 4);
    *dq = vp9_read_and_apply_sign(r, value);
John Koleszar's avatar
John Koleszar committed
738
  }
John Koleszar's avatar
John Koleszar committed
739

740
  // Trigger a quantizer update if the delta-q value has changed
741
  return old_value != *dq;
John Koleszar's avatar
John Koleszar committed
742 743
}

744
static void set_offsets(VP9D_COMP *pbi, BLOCK_SIZE_TYPE bsize,
Ronald S. Bultje's avatar
Ronald S. Bultje committed
745
                        int mb_row, int mb_col) {
746 747
  const int bh = 1 << mb_height_log2(bsize);
  const int bw = 1 << mb_width_log2(bsize);
Ronald S. Bultje's avatar
Ronald S. Bultje committed
748 749
  VP9_COMMON *const cm = &pbi->common;
  MACROBLOCKD *const xd = &pbi->mb;
John Koleszar's avatar
John Koleszar committed
750

751 752 753 754 755 756
  const int mb_idx = mb_row * cm->mode_info_stride + mb_col;
  const YV12_BUFFER_CONFIG *dst_fb = &cm->yv12_fb[cm->new_fb_idx];
  const int recon_yoffset = (16 * mb_row) * dst_fb->y_stride + (16 * mb_col);
  const int recon_uvoffset = (8 * mb_row) * dst_fb->uv_stride + (8 * mb_col);

  xd->mode_info_context = cm->mi + mb_idx;
757
  xd->mode_info_context->mbmi.sb_type = bsize;
758 759 760
  xd->prev_mode_info_context = cm->prev_mi + mb_idx;
  xd->above_context = cm->above_context + mb_col;
  xd->left_context = cm->left_context + mb_row % 4;
761 762
  xd->above_seg_context = cm->above_seg_context + mb_col;
  xd->left_seg_context  = cm->left_seg_context + (mb_row & 3);
Ronald S. Bultje's avatar
Ronald S. Bultje committed
763

764 765
  // Distance of Mb to the various image edges. These are specified to 8th pel
  // as they are always compared to values that are in 1/8th pel units
766
  set_mb_row_col(cm, xd, mb_row, bh, mb_col, bw);
Ronald S. Bultje's avatar
Ronald S. Bultje committed
767

768 769 770
  xd->plane[0].dst.buf = dst_fb->y_buffer + recon_yoffset;
  xd->plane[1].dst.buf = dst_fb->u_buffer + recon_uvoffset;
  xd->plane[2].dst.buf = dst_fb->v_buffer + recon_uvoffset;
Ronald S. Bultje's avatar
Ronald S. Bultje committed
771
}
John Koleszar's avatar
John Koleszar committed
772

773
static void set_refs(VP9D_COMP *pbi, int mb_row, int mb_col) {
Ronald S. Bultje's avatar
Ronald S. Bultje committed
774 775
  VP9_COMMON *const cm = &pbi->common;
  MACROBLOCKD *const xd = &pbi->mb;
776
  MB_MODE_INFO *const mbmi = &xd->mode_info_context->mbmi;
Ronald S. Bultje's avatar
Ronald S. Bultje committed
777 778

  if (mbmi->ref_frame > INTRA_FRAME) {
779
    // Select the appropriate reference frame for this MB