Coverage Report

Created: 2026-08-13 07:23

next uncovered line (L), next uncovered region (R), next uncovered branch (B)
/work/vvenc/source/Lib/EncoderLib/EncModeCtrl.cpp
Line
Count
Source
1
/* -----------------------------------------------------------------------------
2
The copyright in this software is being made available under the Clear BSD
3
License, included below. No patent rights, trademark rights and/or 
4
other Intellectual Property Rights other than the copyrights concerning 
5
the Software are granted under this license.
6
7
The Clear BSD License
8
9
Copyright (c) 2019-2026, Fraunhofer-Gesellschaft zur Förderung der angewandten Forschung e.V. & The VVenC Authors.
10
All rights reserved.
11
12
Redistribution and use in source and binary forms, with or without modification,
13
are permitted (subject to the limitations in the disclaimer below) provided that
14
the following conditions are met:
15
16
     * Redistributions of source code must retain the above copyright notice,
17
     this list of conditions and the following disclaimer.
18
19
     * Redistributions in binary form must reproduce the above copyright
20
     notice, this list of conditions and the following disclaimer in the
21
     documentation and/or other materials provided with the distribution.
22
23
     * Neither the name of the copyright holder nor the names of its
24
     contributors may be used to endorse or promote products derived from this
25
     software without specific prior written permission.
26
27
NO EXPRESS OR IMPLIED LICENSES TO ANY PARTY'S PATENT RIGHTS ARE GRANTED BY
28
THIS LICENSE. THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND
29
CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
30
LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A
31
PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR
32
CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
33
EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
34
PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
35
BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER
36
IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
37
ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
38
POSSIBILITY OF SUCH DAMAGE.
39
40
41
------------------------------------------------------------------------------------------- */
42
43
44
/** \file     EncModeCtrl.cpp
45
    \brief    Encoder controller for trying out specific modes
46
*/
47
48
#include "EncCu.h"
49
50
#include "EncModeCtrl.h"
51
#include "CommonLib/RdCost.h"
52
#include "CommonLib/CodingStructure.h"
53
#include "CommonLib/Picture.h"
54
#include "CommonLib/UnitTools.h"
55
#include "CommonLib/dtrace_next.h"
56
57
#include <cmath>
58
59
//! \ingroup EncoderLib
60
//! \{
61
62
namespace vvenc {
63
64
void CacheBlkInfoCtrl::create(int ctuSize)
65
19.2k
{
66
19.2k
  const unsigned numPos = MAX_CU_SIZE >> MIN_CU_LOG2;
67
19.2k
  const int maxSizeIdx  = MAX_CU_SIZE_IDX - MIN_CU_LOG2;
68
69
  //static constexpr size_t numCu = 7921;
70
71
19.2k
  size_t numCu = 0;
72
  
73
134k
  for( int wIdx = 0; wIdx < maxSizeIdx; wIdx++ )
74
115k
  {
75
806k
    for( int hIdx = 0; hIdx < maxSizeIdx; hIdx++ )
76
691k
    {
77
22.8M
      for( unsigned y = 0; y < numPos; y++ )
78
22.1M
      {
79
729M
        for( unsigned x = 0; x < numPos; x++ )
80
707M
        {
81
          // a block of width W might be offset of N * W + 1/2 W (bcs of TT), same for H
82
          // W = 1 << ( wIdx + 2 )
83
          // 1/2 W = 1 << ( wIdx + 1 )
84
          // remainder of (N+1/2)*W -> x & ( ( 1 << ( wIdx + 1 ) ) - 1 )
85
  
86
707M
          if( (x + (1 << (wIdx)) <= (ctuSize >> MIN_CU_LOG2))
87
497M
              && (y + (1 << (hIdx)) <= (ctuSize >> MIN_CU_LOG2))
88
349M
              && (((x << MIN_CU_LOG2) & ((1 << (wIdx + MIN_CU_LOG2 - 1)) - 1)) == 0)
89
230M
              && (((y << MIN_CU_LOG2) & ((1 << (hIdx + MIN_CU_LOG2 - 1)) - 1)) == 0) )
90
152M
          {
91
152M
            numCu++;
92
152M
          }
93
707M
        }
94
22.1M
      }
95
691k
    }
96
115k
  }
97
98
19.2k
  m_codedCUInfoBuf = new CodedCUInfo[numCu];
99
19.2k
  CodedCUInfo* cuInfo = m_codedCUInfoBuf;
100
101
134k
  for( int wIdx = 0; wIdx < maxSizeIdx; wIdx++ )
102
115k
  {
103
806k
    for( int hIdx = 0; hIdx < maxSizeIdx; hIdx++ )
104
691k
    {
105
22.8M
      for( unsigned y = 0; y < numPos; y++ )
106
22.1M
      {
107
729M
        for( unsigned x = 0; x < numPos; x++ )
108
707M
        {
109
          // a block of width W might be offset of N * W + 1/2 W (bcs of TT), same for H
110
          // W = 1 << ( wIdx + 2 )
111
          // 1/2 W = 1 << ( wIdx + 1 )
112
          // remainder of (N+1/2)*W -> x & ( ( 1 << ( wIdx + 1 ) ) - 1 )
113
114
707M
          if(( x + (1<<(wIdx)) <= ( ctuSize >> MIN_CU_LOG2 ) )
115
497M
            && ( y + (1<<(hIdx)) <= ( ctuSize >> MIN_CU_LOG2 ) )
116
349M
            && ( ( ( x << MIN_CU_LOG2 ) & ((1 << (wIdx + MIN_CU_LOG2 - 1)) - 1) ) == 0 )
117
230M
            && ( ( ( y << MIN_CU_LOG2 ) & ((1 << (hIdx + MIN_CU_LOG2 - 1)) - 1) ) == 0 ) )
118
152M
          {
119
152M
            m_codedCUInfo[wIdx][hIdx][x][y] = cuInfo++;
120
152M
            m_codedCUInfo[wIdx][hIdx][x][y]->poc       = -1;
121
152M
            m_codedCUInfo[wIdx][hIdx][x][y]->ctuRsAddr = -1;
122
152M
          }
123
555M
          else
124
555M
          {
125
555M
            m_codedCUInfo[wIdx][hIdx][x][y] = nullptr;
126
555M
          }
127
707M
        }
128
22.1M
      }
129
691k
    }
130
115k
  }
131
19.2k
}
132
133
void CacheBlkInfoCtrl::destroy()
134
19.2k
{
135
19.2k
  delete[] m_codedCUInfoBuf;
136
19.2k
  m_codedCUInfoBuf = nullptr;
137
19.2k
}
138
139
void CacheBlkInfoCtrl::init( const Slice &slice )
140
3.84k
{
141
3.84k
  m_pcv = slice.pps->pcv;
142
3.84k
}
143
144
void CacheBlkInfoCtrl::initBlk( const UnitArea& area, int poc )
145
130k
{
146
130k
  unsigned idx1, idx2, idx3, idx4;
147
130k
  getAreaIdxNew( area.Y(), *m_pcv, idx1, idx2, idx3, idx4 );
148
149
130k
  const int ctuRsAddr = getCtuAddr( area.lumaPos(), *m_pcv );
150
130k
  CodedCUInfo* cuInfo = m_codedCUInfo[idx1][idx2][idx3][idx4];
151
152
130k
  if( cuInfo->poc != poc || cuInfo->ctuRsAddr != ctuRsAddr )
153
93.8k
  {
154
93.8k
    GCC_WARNING_DISABLE_class_memaccess
155
93.8k
    memset( cuInfo, 0, sizeof( CodedCUInfo ) );
156
93.8k
    GCC_WARNING_RESET
157
158
93.8k
    cuInfo->poc       = poc;
159
93.8k
    cuInfo->ctuRsAddr = ctuRsAddr;
160
93.8k
  }
161
130k
}
162
163
CodedCUInfo& CacheBlkInfoCtrl::getBlkInfo( const UnitArea& area )
164
215k
{
165
215k
  unsigned idx1, idx2, idx3, idx4;
166
215k
  getAreaIdxNew( area.Y(), *m_pcv, idx1, idx2, idx3, idx4 );
167
//  DTRACE( g_trace_ctx, D_TMP, "%d loc %d %d %d %d\n", g_trace_ctx->getChannelCounter(D_TMP), idx1, idx2, idx3, idx4);
168
215k
  return *m_codedCUInfo[idx1][idx2][idx3][idx4];
169
215k
}
170
171
void CodedCUInfo::setMv( const RefPicList refPicList, const int iRefIdx, const Mv& rMv )
172
0
{
173
0
  if( iRefIdx >= MAX_STORED_CU_INFO_REFS ) return;
174
175
0
  saveMv [refPicList][iRefIdx] = rMv;
176
0
  validMv[refPicList][iRefIdx] = true;
177
0
}
178
179
bool CodedCUInfo::getMv( const RefPicList refPicList, const int iRefIdx, Mv& rMv ) const
180
0
{
181
0
  if( iRefIdx >= MAX_STORED_CU_INFO_REFS )
182
0
  {
183
0
    rMv = saveMv[refPicList][0];
184
0
    return false;
185
0
  }
186
187
0
  rMv = saveMv[refPicList][iRefIdx];
188
0
  return validMv[refPicList][iRefIdx];
189
0
}
190
191
uint8_t CacheBlkInfoCtrl::findBestSbt( const UnitArea& area, const uint32_t curPuSse )
192
0
{
193
0
  unsigned idx1, idx2, idx3, idx4;
194
0
  getAreaIdxNew( area.Y(), *m_pcv, idx1, idx2, idx3, idx4 );
195
0
  CodedCUInfo* pSbtSave = m_codedCUInfo[idx1][idx2][idx3][idx4];
196
197
0
  for( int i = 0; i < pSbtSave->numPuInfoStored; i++ )
198
0
  {
199
0
    if( curPuSse == pSbtSave->puSse[i] )
200
0
    {
201
0
      return pSbtSave->puSbt[i];
202
0
    }
203
0
  }
204
205
0
  return MAX_UCHAR;
206
0
}
207
208
bool CacheBlkInfoCtrl::saveBestSbt( const UnitArea& area, const uint32_t curPuSse, const uint8_t curPuSbt )
209
0
{
210
0
  unsigned idx1, idx2, idx3, idx4;
211
0
  getAreaIdxNew( area.Y(), *m_pcv, idx1, idx2, idx3, idx4 );
212
0
  CodedCUInfo* pSbtSave = m_codedCUInfo[idx1][idx2][idx3][idx4];
213
214
0
  if( pSbtSave->numPuInfoStored == SBT_NUM_SL )
215
0
  {
216
0
    return false;
217
0
  }
218
219
0
  pSbtSave->puSse[pSbtSave->numPuInfoStored] = curPuSse;
220
0
  pSbtSave->puSbt[pSbtSave->numPuInfoStored] = curPuSbt;
221
0
  pSbtSave->numPuInfoStored++;
222
223
0
  return true;
224
0
}
225
226
static bool isTheSameNbHood( const CodingUnit &cu, const CodingStructure& cs, const Partitioner &partitioner, int picW, int picH )
227
0
{
228
0
  if( cu.chType != partitioner.chType )
229
0
  {
230
0
    return false;
231
0
  }
232
233
0
  const PartitioningStack &ps = partitioner.getPartStack();
234
235
0
  int i = 1;
236
237
0
  for( ; i < ps.size(); i++ )
238
0
  {
239
0
    if( ps[i].split != CU::getSplitAtDepth( cu, i - 1 ) )
240
0
    {
241
0
      break;
242
0
    }
243
0
  }
244
245
0
  const UnitArea& cmnAnc = ps[i - 1].parts[ps[i - 1].idx];
246
0
  const UnitArea cuArea  = CS::getArea( cs, cu, partitioner.chType, partitioner.treeType );
247
248
0
  for( int i = 0; i < cmnAnc.blocks.size(); i++ )
249
0
  {
250
0
    if( i < cuArea.blocks.size() && cuArea.blocks[i].valid() && cuArea.blocks[i].pos() != cmnAnc.blocks[i].pos() )
251
0
    {
252
0
      return false;
253
0
    }
254
0
  }
255
256
0
  return true;
257
0
}
258
259
void BestEncInfoCache::create( const bool reuseCuResults, const ChromaFormat chFmt, const int ctuSize )
260
19.2k
{
261
19.2k
  m_reuseCuResults = reuseCuResults;
262
263
19.2k
  if( !m_reuseCuResults ) return;
264
265
0
  const unsigned numPos = MAX_CU_SIZE >> MIN_CU_LOG2;
266
0
  const int maxSizeIdx  = MAX_CU_SIZE_IDX - MIN_CU_LOG2;
267
268
0
  static constexpr size_t yuvNom[4] = { 1, 3, 2, 3 };
269
0
  static constexpr size_t yuvDen[4] = { 0, 1, 0, 0 };
270
271
  // only true for 128x128 CTU
272
  //static constexpr size_t numCu = 7921;
273
  //static constexpr size_t numDmvrMv = 5439;
274
  //const size_t numCoeff = ( 1345600 * yuvNom[chFmt] ) >> yuvDen[chFmt];
275
276
0
  size_t numCu = 0;
277
0
  size_t numDmvrMv = 0;
278
0
  size_t numCoeff = 0;
279
  
280
0
  for( int wIdx = 0; wIdx < maxSizeIdx; wIdx++ )
281
0
  {
282
0
    for( int hIdx = 0; hIdx < maxSizeIdx; hIdx++ )
283
0
    {
284
0
      int dmvrSize = 0;
285
0
      if( hIdx >= 1 && wIdx >= 1 && (wIdx + hIdx) >= 3 )
286
0
      {
287
0
        dmvrSize = (1 << std::max( 0, (wIdx + MIN_CU_LOG2 - DMVR_SUBCU_SIZE_LOG2) )) * (1 << std::max( 0, (hIdx + MIN_CU_LOG2 - DMVR_SUBCU_SIZE_LOG2) ));
288
0
      }
289
  
290
0
      const UnitArea area( chFmt, Area( 0, 0, 1 << (wIdx + 2), 1 << (hIdx + 2) ) );
291
  
292
0
      for( unsigned x = 0; x < numPos; x++ )
293
0
      {
294
0
        for( unsigned y = 0; y < numPos; y++ )
295
0
        {
296
          // a block of width W might be offset of N * W + 1/2 W (bcs of TT), same for H
297
          // W = 1 << ( wIdx + 2 )
298
          // 1/2 W = 1 << ( wIdx + 1 )
299
          // remainder of (N+1/2)*W -> x & ( ( 1 << ( wIdx + 1 ) ) - 1 )
300
  
301
0
          if( (x + (1 << (wIdx)) <= (ctuSize >> MIN_CU_LOG2))
302
0
            && (y + (1 << (hIdx)) <= (ctuSize >> MIN_CU_LOG2))
303
0
            && (((x << MIN_CU_LOG2) & ((1 << (wIdx + MIN_CU_LOG2 - 1)) - 1)) == 0)
304
0
            && (((y << MIN_CU_LOG2) & ((1 << (hIdx + MIN_CU_LOG2 - 1)) - 1)) == 0) )
305
0
          {
306
0
            numCu++;
307
  
308
0
            numCoeff += area.Y().area();
309
  
310
  
311
            //numCu++;
312
0
            numDmvrMv += dmvrSize;
313
0
          }
314
0
        }
315
0
      }
316
0
    }
317
0
  }
318
  
319
  //std::cout << numCu << " " << numDmvrMv << " " << numCoeff << std::endl;
320
  
321
0
  numCoeff  *= yuvNom[chFmt];
322
0
  numCoeff >>= yuvDen[chFmt];
323
324
0
  m_encInfoBuf = new BestEncodingInfo[numCu];
325
0
  BestEncodingInfo* encInfo = m_encInfoBuf;
326
327
0
  m_dmvrMvBuf = new Mv[numDmvrMv];
328
0
  Mv* dmvrMv = m_dmvrMvBuf;
329
330
0
  m_pCoeff = new TCoeffSig[numCoeff];
331
0
  TCoeffSig* coeffPtr = m_pCoeff;
332
333
0
  for( int wIdx = 0; wIdx < maxSizeIdx; wIdx++ )
334
0
  {
335
0
    for( int hIdx = 0; hIdx < maxSizeIdx; hIdx++ )
336
0
    {
337
0
      int dmvrSize = 0;
338
0
      if( hIdx >= 1 && wIdx >= 1 && (wIdx + hIdx) >= 3 )
339
0
      {
340
0
        dmvrSize = (1 << std::max( 0, (wIdx + MIN_CU_LOG2 - DMVR_SUBCU_SIZE_LOG2) )) * (1 << std::max( 0, (hIdx + MIN_CU_LOG2 - DMVR_SUBCU_SIZE_LOG2) ));
341
0
      }
342
343
0
      const UnitArea area( chFmt, Area( 0, 0, 1 << (wIdx + 2), 1 << (hIdx + 2) ) );
344
345
0
      for( unsigned x = 0; x < numPos; x++ )
346
0
      {
347
0
        for( unsigned y = 0; y < numPos; y++ )
348
0
        {
349
          // a block of width W might be offset of N * W + 1/2 W (bcs of TT), same for H
350
          // W = 1 << ( wIdx + 2 )
351
          // 1/2 W = 1 << ( wIdx + 1 )
352
          // remainder of (N+1/2)*W -> x & ( ( 1 << ( wIdx + 1 ) ) - 1 )
353
354
0
          if(( x + (1<<(wIdx)) <= ( ctuSize >> MIN_CU_LOG2 ) )
355
0
            && ( y + (1<<(hIdx)) <= ( ctuSize >> MIN_CU_LOG2 ) )
356
0
            && ( ( ( x << MIN_CU_LOG2 )  & ((1 << (wIdx + MIN_CU_LOG2 - 1)) - 1) ) == 0 )
357
0
            && ( ( ( y << MIN_CU_LOG2 )  & ((1 << (hIdx + MIN_CU_LOG2 - 1)) - 1) ) == 0 ) )
358
0
          {
359
0
            m_bestEncInfo[wIdx][hIdx][x][y] = encInfo++;
360
361
0
            m_bestEncInfo[wIdx][hIdx][x][y]->cu.UnitArea::operator=( area );
362
0
            m_bestEncInfo[wIdx][hIdx][x][y]->tu.UnitArea::operator=( area );
363
364
0
            m_bestEncInfo[wIdx][hIdx][x][y]->cu.chType = CH_L;
365
0
            m_bestEncInfo[wIdx][hIdx][x][y]->cu.treeType = TREE_D;
366
0
            m_bestEncInfo[wIdx][hIdx][x][y]->cu.modeType = MODE_TYPE_ALL;
367
0
            m_bestEncInfo[wIdx][hIdx][x][y]->cu.qp = MAX_SCHAR;
368
0
            m_bestEncInfo[wIdx][hIdx][x][y]->tu.chType = CH_L;
369
370
0
            if( dmvrSize )
371
0
            {
372
0
              m_bestEncInfo[wIdx][hIdx][x][y]->cu.mvdL0SubPu = dmvrMv; 
373
0
              dmvrMv += dmvrSize;
374
0
            }
375
376
0
            TCoeffSig* coeff[MAX_NUM_TBLOCKS] = { 0, };
377
378
0
            const UnitArea& area = m_bestEncInfo[wIdx][hIdx][x][y]->tu;
379
380
0
            for( int i = 0; i < area.blocks.size(); i++ )
381
0
            {
382
0
              coeff[i] = coeffPtr; coeffPtr += area.blocks[i].area();
383
0
            }
384
385
0
            m_bestEncInfo[wIdx][hIdx][x][y]->tu.cs = &m_dummyCS;
386
0
            m_bestEncInfo[wIdx][hIdx][x][y]->tu.init( coeff );
387
388
0
            m_bestEncInfo[wIdx][hIdx][x][y]->poc      = -1;
389
0
            m_bestEncInfo[wIdx][hIdx][x][y]->testMode = EncTestMode();
390
0
          }
391
0
          else
392
0
          {
393
0
            m_bestEncInfo[wIdx][hIdx][x][y] = nullptr;
394
0
          }
395
0
        }
396
0
      }
397
0
    }
398
0
  }
399
0
}
400
401
void BestEncInfoCache::destroy()
402
19.2k
{
403
19.2k
  if( !m_reuseCuResults ) return;
404
405
0
  delete[] m_encInfoBuf;
406
0
  m_encInfoBuf = nullptr;
407
408
0
  delete[] m_pCoeff;
409
0
  m_pCoeff = nullptr;
410
411
0
  delete[] m_dmvrMvBuf;
412
0
  m_dmvrMvBuf = nullptr;
413
414
0
  m_pcv = nullptr;
415
0
}
416
417
void BestEncInfoCache::init( const Slice &slice )
418
3.84k
{
419
3.84k
  if( !m_reuseCuResults ) return;
420
421
0
  bool isInitialized = m_pcv;
422
423
0
  m_pcv = slice.pps->pcv;
424
425
0
  if( isInitialized ) return;
426
427
0
  m_dummyCS.pcv = m_pcv;
428
0
}
429
430
bool BestEncInfoCache::setFromCs( const CodingStructure& cs, const EncTestMode& testMode, const Partitioner& partitioner )
431
82.4k
{
432
82.4k
  if( !m_reuseCuResults ) return false;
433
434
0
  if( cs.cus.size() != 1 || cs.tus.size() != 1 || partitioner.maxBTD <= 1 )
435
0
  {
436
0
    return false;
437
0
  }
438
439
0
  unsigned idx1, idx2, idx3, idx4;
440
0
  getAreaIdxNew( cs.area.Y(), *m_pcv, idx1, idx2, idx3, idx4 );
441
442
0
  BestEncodingInfo& encInfo = *m_bestEncInfo[idx1][idx2][idx3][idx4];
443
444
0
  encInfo.poc            =  cs.picture->poc;
445
0
  encInfo.cu.UnitArea::operator= ( *cs.cus.front() );
446
0
  encInfo.tu.UnitArea::operator= ( *cs.tus.front() );
447
0
  encInfo.cu             = *cs.cus.front();
448
0
  for( auto &blk : cs.tus.front()->blocks )
449
0
  {
450
0
    if( blk.valid() ) encInfo.tu.copyComponentFrom( *cs.tus.front(), blk.compID );
451
0
  }
452
0
  encInfo.testMode       = testMode;
453
0
  encInfo.dist           = cs.dist;
454
0
  encInfo.costEDO        = cs.costDbOffset;
455
456
0
  return true;
457
0
}
458
459
bool BestEncInfoCache::isReusingCuValid( const CodingStructure& cs, const Partitioner& partitioner, int qp )
460
213k
{
461
213k
  if( !m_reuseCuResults || partitioner.treeType == TREE_C || partitioner.maxBTD <= 1 )
462
213k
  {
463
213k
    return false; //if save & load is allowed for chroma CUs, we should check whether luma info (pred, recon, etc) is the same, which is quite complex
464
213k
  }
465
466
0
  unsigned idx1, idx2, idx3, idx4;
467
0
  getAreaIdxNew( cs.area.Y(), *m_pcv, idx1, idx2, idx3, idx4 );
468
469
0
  BestEncodingInfo& encInfo = *m_bestEncInfo[idx1][idx2][idx3][idx4];
470
471
0
  if( encInfo.cu.treeType != partitioner.treeType || encInfo.cu.modeType != partitioner.modeType )
472
0
  {
473
0
    return false;
474
0
  }
475
0
  if( encInfo.cu.qp != qp )
476
0
    return false;
477
0
  if( cs.picture->poc != encInfo.poc 
478
0
    || CS::getArea( cs, cs.area, partitioner.chType, partitioner.treeType ) != CS::getArea( cs, encInfo.cu, partitioner.chType, partitioner.treeType ) 
479
0
    || !isTheSameNbHood( encInfo.cu, cs, partitioner, (cs.picture->Y().width), (cs.picture->Y().height))
480
0
    || CU::isIBC(encInfo.cu)
481
0
    || partitioner.currQgEnable() || cs.currQP[partitioner.chType] != encInfo.cu.qp
482
0
    )
483
0
  {
484
0
    return false;
485
0
  }
486
0
  else
487
0
  {
488
0
    return true;
489
0
  }
490
0
}
491
492
bool BestEncInfoCache::setCsFrom( CodingStructure& cs, EncTestMode& testMode, const Partitioner& partitioner ) const
493
0
{
494
0
  if( !m_reuseCuResults ) return false;
495
496
0
  unsigned idx1, idx2, idx3, idx4;
497
0
  getAreaIdxNew( cs.area.Y(), *m_pcv, idx1, idx2, idx3, idx4 );
498
499
0
  BestEncodingInfo& encInfo = *m_bestEncInfo[idx1][idx2][idx3][idx4];
500
501
0
  if( cs.picture->poc != encInfo.poc 
502
0
    || CS::getArea( cs, cs.area, partitioner.chType, partitioner.treeType ) != CS::getArea( cs, encInfo.cu, partitioner.chType, partitioner.treeType ) 
503
0
    || !isTheSameNbHood( encInfo.cu, cs, partitioner, (cs.picture->Y().width), (cs.picture->Y().height))
504
0
    || partitioner.currQgEnable() || cs.currQP[partitioner.chType] != encInfo.cu.qp
505
0
    )
506
0
  {
507
0
    return false;
508
0
  }
509
510
0
  const UnitArea ua = CS::getArea( cs, cs.area, partitioner.chType, partitioner.treeType );
511
0
  CodingUnit     &cu = cs.addCU( ua, partitioner.chType );
512
0
  cu.treeType = partitioner.treeType;
513
0
  cu.modeType = partitioner.modeType;
514
0
  cu.initPuData();
515
0
  TransformUnit  &tu = cs.addTU( ua, partitioner.chType, &cu );
516
517
0
  cu          .repositionTo( encInfo.cu );
518
0
  cu          .repositionTo( encInfo.cu );
519
0
  tu          .repositionTo( encInfo.tu );
520
521
0
  cu          = encInfo.cu;
522
0
  cu          = encInfo.cu;
523
0
  for( auto &blk : tu.blocks )
524
0
  {
525
0
    if( blk.valid() ) tu.copyComponentFrom( encInfo.tu, blk.compID );
526
0
  }
527
528
0
  testMode    = encInfo.testMode;
529
0
  cs.dist     = encInfo.dist;
530
0
  cs.costDbOffset = encInfo.costEDO;
531
0
  return true;
532
0
}
533
534
535
//////////////////////////////////////////////////////////////////////////
536
// EncModeCtrl
537
//////////////////////////////////////////////////////////////////////////
538
void EncModeCtrl::init( const VVEncCfg& encCfg, RdCost* pRdCost )
539
19.2k
{
540
19.2k
  m_pcEncCfg = &encCfg;
541
19.2k
  m_pcRdCost = pRdCost;
542
19.2k
  comprCUCtx = nullptr;
543
544
19.2k
  CacheBlkInfoCtrl::create( encCfg.m_CTUSize );
545
19.2k
  BestEncInfoCache::create( encCfg.m_reuseCuResults, encCfg.m_internChromaFormat, encCfg.m_CTUSize );
546
19.2k
}
547
548
void EncModeCtrl::destroy()
549
19.2k
{
550
19.2k
  CacheBlkInfoCtrl::destroy();
551
19.2k
  BestEncInfoCache::destroy();
552
19.2k
}
553
554
void EncModeCtrl::initCTUEncoding( const Slice &slice, int tileIdx )
555
3.84k
{
556
3.84k
  CacheBlkInfoCtrl::init( slice );
557
3.84k
  BestEncInfoCache::init( slice );
558
559
3.84k
  CHECK( !m_ComprCUCtxList.empty(), "Mode list is not empty at the beginning of a CTU" );
560
561
3.84k
  if( m_pcEncCfg->m_fastQtBtEnc )
562
3.84k
  {
563
3.84k
    m_skipThresholdE0023FastEnc = ((slice.getMinPictureDistance() <= PICTURE_DISTANCE_TH) ? FAST_SKIP_DEPTH : SKIP_DEPTH);
564
3.84k
  }
565
0
  else
566
0
  {
567
0
    m_skipThresholdE0023FastEnc = SKIP_DEPTH;
568
0
  }
569
570
3.84k
  m_tileIdx = tileIdx;
571
3.84k
}
572
573
void EncModeCtrl::initCULevel( Partitioner &partitioner, const CodingStructure& cs, int  MergeSimpleFlag)
574
130k
{
575
  // Min/max depth
576
130k
  unsigned minDepth = 0;
577
130k
  unsigned maxDepth = cs.pcv->getMaxDepth( cs.slice->sliceType, partitioner.chType );
578
130k
  if( m_pcEncCfg->m_useFastLCTU )
579
130k
  {
580
130k
    partitioner.setMaxMinDepth(minDepth, maxDepth, cs, cs.picture->useQtbttSpeedUpMode, MergeSimpleFlag);
581
130k
  }
582
583
130k
  minDepth = std::max<unsigned>( minDepth, cs.pcv->getMinDepth( cs.slice->sliceType, partitioner.chType ) );
584
130k
  maxDepth = std::min<unsigned>( maxDepth, cs.pcv->getMaxDepth( cs.slice->sliceType, partitioner.chType ) );
585
586
130k
  m_ComprCUCtxList.push_back( ComprCUCtx( cs, minDepth, maxDepth ) );
587
130k
  comprCUCtx = &m_ComprCUCtxList.back();
588
589
130k
  const CodingUnit* cuLeft  = cs.getCURestricted( cs.area.blocks[partitioner.chType].pos().offset( -1, 0 ), cs.area.blocks[partitioner.chType].pos(), cs.slice->independentSliceIdx, m_tileIdx, partitioner.chType, partitioner.treeType );
590
130k
  const CodingUnit* cuAbove = cs.getCURestricted( cs.area.blocks[partitioner.chType].pos().offset( 0, -1 ), cs.area.blocks[partitioner.chType].pos(), cs.slice->independentSliceIdx, m_tileIdx, partitioner.chType, partitioner.treeType );
591
592
130k
  const bool qtBeforeBt = ( (  cuLeft  &&  cuAbove  && cuLeft ->qtDepth > partitioner.currQtDepth && cuAbove->qtDepth > partitioner.currQtDepth )
593
128k
                         || (  cuLeft  && !cuAbove  && cuLeft ->qtDepth > partitioner.currQtDepth )
594
126k
                         || ( !cuLeft  &&  cuAbove  && cuAbove->qtDepth > partitioner.currQtDepth )
595
123k
                         || ( !cuAbove && !cuLeft   && cs.area.lwidth() >= ( 32 << cs.slice->TLayer ) )
596
110k
                         || ( m_pcEncCfg->m_qtbttSpeedUp > 1 && partitioner.maxBTD < ( ( cs.slice->isIntra() && !cs.sps->IBC ) ? 3 : 2 ) ) )
597
20.8k
                         && ( cs.area.lwidth() > ( cs.pcv->getMinQtSize( *cs.slice, partitioner.chType ) << 1 ) );
598
599
  // set features
600
130k
  ComprCUCtx &cuECtx    = *comprCUCtx;
601
130k
  cuECtx.qtBeforeBt     = qtBeforeBt;
602
130k
  cuECtx.doTriHorzSplit = true;
603
130k
  cuECtx.doTriVertSplit = true;
604
130k
  cuECtx.doMoreSplits   = 3;
605
130k
  cuECtx.isReusingCu    = isReusingCuValid( cs, partitioner, cs.baseQP );
606
130k
  cuECtx.didHorzSplit   = partitioner.canSplit( CU_HORZ_SPLIT, cs );
607
130k
  cuECtx.didVertSplit   = partitioner.canSplit( CU_VERT_SPLIT, cs );
608
130k
  cuECtx.doHorChromaSplit = true;
609
130k
  cuECtx.doVerChromaSplit = true;
610
130k
  cuECtx.doQtChromaSplit  = true;
611
612
130k
  if( m_pcEncCfg->m_contentBasedFastQtbt && cs.pcv->getMaxMTTDepth(*cs.slice, partitioner.chType))
613
130k
  {
614
130k
    const CompArea& currArea = partitioner.currArea().Y();
615
130k
    int cuHeight  = currArea.height;
616
130k
    int cuWidth   = currArea.width;
617
618
130k
    const bool condIntraInter = m_pcEncCfg->m_IntraPeriod == 1 ? ( partitioner.currBtDepth == 0 ) : ( cuHeight > 32 && cuWidth > 32 );
619
620
130k
    if( cuWidth == cuHeight && condIntraInter )
621
28.3k
    {
622
28.3k
      const CPelBuf bufCurrArea = cs.getOrgBuf( partitioner.currArea().block( COMP_Y ) );
623
624
28.3k
      Intermediate_Int horVal = 0;
625
28.3k
      Intermediate_Int verVal = 0;
626
28.3k
      Intermediate_Int dupVal = 0;
627
28.3k
      Intermediate_Int dowVal = 0;
628
629
28.3k
      unsigned j, k;
630
631
2.27M
      for( k = 0; k < cuHeight - 1; k++ )
632
2.25M
      {
633
205M
        for( j = 0; j < cuWidth - 1; j++ )
634
203M
        {
635
203M
          horVal += abs( bufCurrArea.at( j + 1, k     ) - bufCurrArea.at( j, k ) );
636
203M
          verVal += abs( bufCurrArea.at( j    , k + 1 ) - bufCurrArea.at( j, k ) );
637
203M
          dowVal += abs( bufCurrArea.at( j + 1, k )     - bufCurrArea.at( j, k + 1 ) );
638
203M
          dupVal += abs( bufCurrArea.at( j + 1, k + 1 ) - bufCurrArea.at( j, k ) );
639
203M
        }
640
2.25M
      }
641
642
28.3k
      cuECtx.grad_horVal = (double)horVal;
643
28.3k
      cuECtx.grad_verVal = (double)verVal;
644
28.3k
      cuECtx.grad_dowVal = (double)dowVal;
645
28.3k
      cuECtx.grad_dupVal = (double)dupVal;
646
28.3k
    }
647
130k
  }
648
130k
}
649
650
void EncModeCtrl::finishCULevel( Partitioner &partitioner )
651
130k
{
652
130k
  m_ComprCUCtxList.pop_back();
653
130k
  comprCUCtx = m_ComprCUCtxList.size() ? &m_ComprCUCtxList.back() : nullptr;
654
130k
}
655
656
bool EncModeCtrl::trySplit( const EncTestMode& encTestmode, const CodingStructure &cs, Partitioner& partitioner, const EncTestMode& lastTestmode )
657
317k
{
658
317k
  ComprCUCtx& cuECtx = *comprCUCtx;
659
660
317k
  const PartSplit implicitSplit = partitioner.getImplicitSplit( cs );
661
317k
  const bool isBoundary         = implicitSplit != CU_DONT_SPLIT;
662
663
317k
  if( ( m_pcEncCfg->m_IntraPeriod == 1 ) && ( partitioner.chType == CH_C ) && ( !cuECtx.doQtChromaSplit ) )
664
0
  {
665
0
    cuECtx.maxDepth         = partitioner.currDepth;
666
0
  }
667
668
317k
  if( isBoundary )
669
80.1k
  {
670
80.1k
    if( encTestmode.type != ETM_SPLIT_QT )
671
31.7k
    {
672
31.7k
      return getPartSplit( encTestmode ) == implicitSplit;
673
31.7k
    }
674
48.4k
    else
675
48.4k
    {
676
48.4k
      return partitioner.canSplit( CU_QUAD_SPLIT, cs );
677
48.4k
    }
678
80.1k
  }
679
680
237k
  const Slice&           slice       = *cs.slice;
681
237k
  const uint32_t         width       = partitioner.currArea().lumaSize().width;
682
237k
  const CodingStructure *bestCS      = cuECtx.bestCS;
683
237k
  const CodingUnit      *bestCU      = cuECtx.bestCU;
684
685
237k
  if( cuECtx.minDepth > partitioner.currQtDepth && partitioner.canSplit( CU_QUAD_SPLIT, cs ) )
686
0
  {
687
    // enforce QT
688
0
    return encTestmode.type == ETM_SPLIT_QT;
689
0
  }
690
237k
  else if( encTestmode.type == ETM_SPLIT_QT && cuECtx.maxDepth <= partitioner.currQtDepth )
691
1.38k
  {
692
    // don't check this QT depth
693
1.38k
    return false;
694
1.38k
  }
695
696
236k
  CHECK( ! isModeSplit( encTestmode ), "wrong method" );
697
698
  //////////////////////////////////////////////////////////////////////////
699
  // skip-history rule - don't split further if at least for three past levels
700
  //                     in the split tree it was found that skip is the best mode
701
  //////////////////////////////////////////////////////////////////////////
702
236k
  int skipScore = 0;
703
704
236k
  if ((!slice.isIntra() || slice.sps->IBC) && cuECtx.isBestNoSplitSkip )
705
0
  {
706
0
    for( int i = 2; i <= m_ComprCUCtxList.size(); i++ )
707
0
    {
708
0
      if( ( m_ComprCUCtxList.end() - i )->isBestNoSplitSkip )
709
0
      {
710
0
        skipScore += 1;
711
0
      }
712
0
      else
713
0
      {
714
0
        break;
715
0
      }
716
0
    }
717
0
  }
718
719
236k
  const PartSplit split = getPartSplit( encTestmode );
720
236k
  if( !partitioner.canSplit( split, cs ) || skipScore >= 2 || ( skipScore == 1 && m_ComprCUCtxList.size() == 2 ) )
721
60.2k
  {
722
60.2k
    if( split == CU_HORZ_SPLIT ) cuECtx.didHorzSplit = false;
723
60.2k
    if( split == CU_VERT_SPLIT ) cuECtx.didVertSplit = false;
724
60.2k
    if( split == CU_QUAD_SPLIT ) cuECtx.didQuadSplit = false;
725
60.2k
    return false;
726
60.2k
  }
727
728
176k
  if( isChroma( partitioner.chType ) && !cuECtx.doHorChromaSplit && ( split == CU_HORZ_SPLIT || split == CU_TRIH_SPLIT ) )
729
0
  {
730
0
    if( split == CU_HORZ_SPLIT )
731
0
    {
732
0
      cuECtx.didHorzSplit = false;
733
0
    }
734
735
0
    return false;
736
0
  }
737
738
176k
  if( isChroma( partitioner.chType ) && !cuECtx.doVerChromaSplit && ( split == CU_VERT_SPLIT || split == CU_TRIV_SPLIT ) )
739
0
  {
740
0
    if( split == CU_VERT_SPLIT )
741
0
    {
742
0
      cuECtx.didVertSplit = false;
743
0
    }
744
745
0
    return false;
746
0
  }
747
748
176k
  if( m_pcEncCfg->m_contentBasedFastQtbt )
749
176k
  {
750
176k
    const CompArea& currArea = partitioner.currArea().Y();
751
176k
    int cuHeight  = currArea.height;
752
176k
    int cuWidth   = currArea.width;
753
754
176k
    const bool condIntraInter = m_pcEncCfg->m_IntraPeriod == 1 ? ( partitioner.currBtDepth == 0 ) : ( cuHeight > 32 && cuWidth > 32 );
755
756
176k
    if( cuWidth == cuHeight && condIntraInter && split != CU_QUAD_SPLIT )
757
9.93k
    {
758
9.93k
      const double th1 = m_pcEncCfg->m_IntraPeriod == 1 ?  1.2              :  1.0;
759
9.93k
      const double th2 = m_pcEncCfg->m_IntraPeriod == 1 ? (1.2 / sqrt( 2 )) : (1.0 / sqrt( 2 ));
760
761
9.93k
      if( cuECtx.grad_horVal > th1 * cuECtx.grad_verVal && cuECtx.grad_horVal > th2 * cuECtx.grad_dowVal && cuECtx.grad_horVal > th2 * cuECtx.grad_dupVal && ( split == CU_HORZ_SPLIT || split == CU_TRIH_SPLIT ) )
762
0
      {
763
0
        return false;
764
0
      }
765
9.93k
      if( th2 * cuECtx.grad_dupVal < cuECtx.grad_verVal && th2 * cuECtx.grad_dowVal < cuECtx.grad_verVal && th1 * cuECtx.grad_horVal < cuECtx.grad_verVal && ( split == CU_VERT_SPLIT || split == CU_TRIV_SPLIT ) )
766
0
      {
767
0
        return false;
768
0
      }
769
9.93k
    }
770
771
176k
    if( m_pcEncCfg->m_IntraPeriod == 1 && cuWidth <= 32 && cuHeight <= 32 && bestCS && bestCS->tus.size() == 1 && bestCU && bestCU->depth == partitioner.currDepth && partitioner.currBtDepth > 1 && isLuma( partitioner.chType ) )
772
0
    {
773
0
      if( !bestCU->rootCbf )
774
0
      {
775
0
        return false;
776
0
      }
777
0
    }
778
176k
  }
779
780
176k
  if( m_pcEncCfg->m_qtbttSpeedUp > 1 )
781
176k
  {
782
176k
    const int availDepth = cs.pcv->getMaxMTTDepth( slice, partitioner.chType ) - partitioner.currMtDepth;
783
176k
    if( bestCU && bestCU->skip && availDepth <= ( 3 - m_skipThresholdE0023FastEnc ) && !isModeSplit( lastTestmode ) && split != CU_QUAD_SPLIT )
784
0
    {
785
0
      return false;
786
0
    }
787
176k
  }
788
176k
  if( bestCU && bestCU->skip && bestCU->mtDepth >= m_skipThresholdE0023FastEnc && !isModeSplit( lastTestmode ) )
789
0
  {
790
0
    return false;
791
0
  }
792
793
176k
  bool resetFeature = false;
794
795
176k
  switch( split )
796
176k
  {
797
20.8k
    case CU_QUAD_SPLIT:
798
20.8k
      {
799
20.8k
        if( !cuECtx.qtBeforeBt && bestCU )
800
18.6k
        {
801
18.6k
          unsigned maxBTD        = cs.pcv->getMaxMTTDepth( slice, partitioner.chType );
802
18.6k
          const CodingUnit *cuBR = bestCS->cus.back();
803
18.6k
          unsigned height        = partitioner.currArea().lumaSize().height;
804
805
18.6k
          if(((bestCU->btDepth == 0 && maxBTD >= ((slice.isIntra() && !slice.sps->IBC) ? 3 : 2))
806
91
            || (bestCU->btDepth == 1 && cuBR && cuBR->btDepth == 1 && maxBTD >= ((slice.isIntra() && !slice.sps->IBC) ? 4 : 3)))
807
18.5k
            && (width <= MAX_TB_SIZEY && height <= MAX_TB_SIZEY)
808
18.5k
            && cuECtx.didHorzSplit && cuECtx.didVertSplit )
809
14.1k
          {
810
14.1k
            return false;
811
14.1k
          }
812
18.6k
        }
813
6.68k
        if( bestCS )
814
6.68k
        {
815
6.68k
          if( m_pcEncCfg->m_useEarlyCU == 2 && bestCS->cost != MAX_DOUBLE && bestCU && bestCU->skip && cuECtx.nonSkipWasTested && bestCS->cus.size() == 1 )
816
0
          {
817
0
            return false;
818
0
          }
819
820
6.68k
          int stopSplit = (m_pcEncCfg->m_FastInferMerge >> 4) && (bestCS->slice->TLayer > 4);
821
6.68k
          int limitBLsize = stopSplit ? 2048 : 1024;
822
6.68k
          if((m_pcEncCfg->m_useEarlyCU == 1 || stopSplit) && bestCS->cost != MAX_DOUBLE && bestCU && bestCU->skip && partitioner.currArea().lumaSize().area() < limitBLsize )
823
0
          {
824
0
            return false;
825
0
          }
826
6.68k
        }
827
6.68k
      }
828
6.68k
      break;
829
54.1k
    case CU_HORZ_SPLIT:
830
103k
    case CU_VERT_SPLIT:
831
103k
      resetFeature = true;
832
103k
      break;
833
28.9k
    case CU_TRIH_SPLIT:
834
28.9k
      if( cuECtx.didHorzSplit && bestCU && bestCU->btDepth == partitioner.currBtDepth && !bestCU->rootCbf )
835
23.4k
      {
836
23.4k
        return false;
837
23.4k
      }
838
839
5.48k
      if( m_pcEncCfg->m_qtbttSpeedUp > 1 )
840
5.48k
      {
841
        //unsigned maxBTD = cs.pcv->getMaxMTTDepth( slice, partitioner.chType );
842
5.48k
        if( /*maxBTD == 1 && */cuECtx.didHorzSplit && cuECtx.didVertSplit && cuECtx.bestCostHorzSplit > cuECtx.bestCostVertSplit )
843
1.52k
        {
844
1.52k
          return false;
845
1.52k
        }
846
5.48k
      }
847
848
3.96k
      if( !cuECtx.doTriHorzSplit )
849
13
      {
850
13
        return false;
851
13
      }
852
853
3.94k
      if (m_pcEncCfg->m_fastTTSplit && cuECtx.bestCostHorzSplit < MAX_DOUBLE)
854
2.94k
      {
855
2.94k
        if (cuECtx.bestCostBeforeSplit < MAX_DOUBLE && (cuECtx.bestCostHorzSplit > m_pcEncCfg->m_fastTT_th * cuECtx.bestCostBeforeSplit))
856
2.43k
        {
857
2.43k
          return false;
858
2.43k
        }
859
509
        if (cuECtx.bestCostVertSplit < MAX_DOUBLE && (cuECtx.bestCostHorzSplit > m_pcEncCfg->m_fastTT_th * cuECtx.bestCostVertSplit))
860
0
        {
861
0
          return false;
862
0
        }
863
509
      }
864
1.51k
      break;
865
22.6k
    case CU_TRIV_SPLIT:
866
22.6k
      if( cuECtx.didVertSplit && bestCU && bestCU->btDepth == partitioner.currBtDepth && !bestCU->rootCbf )
867
19.3k
      {
868
19.3k
        return false;
869
19.3k
      }
870
871
3.37k
      if( m_pcEncCfg->m_qtbttSpeedUp > 1 )
872
3.37k
      {
873
        //unsigned maxBTD = cs.pcv->getMaxMTTDepth( slice, partitioner.chType );
874
3.37k
        if( /*maxBTD == 1 && */cuECtx.didHorzSplit && cuECtx.didVertSplit && cuECtx.bestCostHorzSplit < cuECtx.bestCostVertSplit )
875
1.86k
        {
876
1.86k
          return false;
877
1.86k
        }
878
3.37k
      }
879
880
1.51k
      if( !cuECtx.doTriVertSplit )
881
39
      {
882
39
        return false;
883
39
      }
884
885
1.47k
      if (m_pcEncCfg->m_fastTTSplit && cuECtx.bestCostVertSplit < MAX_DOUBLE)
886
520
      {
887
520
        if (cuECtx.bestCostBeforeSplit < MAX_DOUBLE && (cuECtx.bestCostVertSplit > m_pcEncCfg->m_fastTT_th * cuECtx.bestCostBeforeSplit))
888
344
        {
889
344
          return false;
890
344
        }
891
176
        if (cuECtx.bestCostHorzSplit < MAX_DOUBLE && (cuECtx.bestCostVertSplit > m_pcEncCfg->m_fastTT_th * cuECtx.bestCostHorzSplit))
892
0
        {
893
0
          return false;
894
0
        }
895
176
      }
896
1.13k
      break;
897
1.13k
    default:
898
0
      THROW( "Only CU split modes are governed by the EncModeCtrl" );
899
0
      return false;
900
0
      break;
901
176k
  }
902
903
112k
  switch( split )
904
112k
  {
905
54.1k
    case CU_HORZ_SPLIT:
906
55.6k
    case CU_TRIH_SPLIT:
907
55.6k
      if( cuECtx.qtBeforeBt && cuECtx.didQuadSplit )
908
1.86k
      {
909
1.86k
        if( cuECtx.maxQtSubDepth > partitioner.currQtDepth + 1 )
910
0
        {
911
0
          if( resetFeature ) cuECtx.didHorzSplit = false;
912
0
          return false;
913
0
        }
914
1.86k
      }
915
55.6k
      break;
916
55.6k
    case CU_VERT_SPLIT:
917
50.6k
    case CU_TRIV_SPLIT:
918
50.6k
      if( cuECtx.qtBeforeBt && cuECtx.didQuadSplit )
919
1.93k
      {
920
1.93k
        if( cuECtx.maxQtSubDepth > partitioner.currQtDepth + 1 )
921
0
        {
922
0
          if( resetFeature ) cuECtx.didVertSplit = false;
923
0
          return false;
924
0
        }
925
1.93k
      }
926
50.6k
      break;
927
50.6k
    default:
928
6.68k
      break;
929
112k
  }
930
931
112k
  if( split == CU_QUAD_SPLIT )
932
6.68k
  {
933
6.68k
    cuECtx.didQuadSplit = m_pcEncCfg->m_qtbttSpeedUp <= 1 || !!cuECtx.doMoreSplits;
934
6.68k
  }
935
936
112k
  return m_pcEncCfg->m_qtbttSpeedUp <= 1 || !!cuECtx.doMoreSplits;
937
112k
}
938
939
bool EncModeCtrl::tryMode( const EncTestMode& encTestmode, const CodingStructure &cs, Partitioner& partitioner )
940
133k
{
941
133k
  CHECK( isModeSplit( encTestmode ), "wrong method");
942
943
133k
  ComprCUCtx& cuECtx = m_ComprCUCtxList.back();
944
945
133k
  if( cuECtx.minDepth > partitioner.currQtDepth && partitioner.canSplit( CU_QUAD_SPLIT, cs ) )
946
0
  {
947
    // enforce QT
948
0
    return false;
949
0
  }
950
951
133k
  const Slice&           slice        = *cs.slice;
952
133k
  const uint32_t         numComp      = getNumberValidComponents( slice.sps->chromaFormatIdc );
953
133k
  const CodedCUInfo      &relatedCU   = getBlkInfo( partitioner.currArea() );
954
133k
  const Area             &lumaArea    = partitioner.currArea().Y();
955
133k
  const CodingStructure* bestCS       = cuECtx.bestCS;
956
133k
  const EncTestMode      bestMode     = bestCS ?cuECtx.bestMode : EncTestMode();
957
958
133k
  if( encTestmode.type == ETM_INTRA )
959
82.4k
  {
960
    // if this is removed, the IntraSearch::xIntraCodingLumaQT needs to be adapted to support Intra TU split
961
    // also isXXAvailable in IntraPrediction.cpp need to be fixed to check availability within the same CU without isDecomp
962
82.4k
    if (m_pcEncCfg->m_FastInferMerge && !slice.isIntra() && !slice.isIRAP() && !(cs.area.lwidth() == 4 && cs.area.lheight() == 4) && !partitioner.isConsIntra())
963
0
    {
964
0
      if (bestCS && (bestCS->slice->TLayer > (m_pcEncCfg->m_maxTLayer - (m_pcEncCfg->m_FastInferMerge & 7)))
965
0
        && (bestCS->bestParent != nullptr) && bestCS->bestParent->cus.size() && (bestCS->bestParent->cus[0]->skip))
966
0
      {
967
0
        return false;
968
0
      }
969
0
    }
970
971
82.4k
    if( lumaArea.width > cs.sps->getMaxTbSize() || lumaArea.height > cs.sps->getMaxTbSize() )
972
0
    {
973
0
      return false;
974
0
    }
975
976
82.4k
    if (m_pcEncCfg->m_usePbIntraFast && (!cs.slice->isIntra() || cs.slice->sps->IBC) && cuECtx.interHad == 0 && cuECtx.bestCU && !CU::isIntra(*cuECtx.bestCU))
977
0
    {
978
0
      return false;
979
0
    }
980
981
    // INTRA MODES
982
82.4k
    if (cs.sps->IBC && !cuECtx.bestTU)
983
59.7k
    {
984
     // return true;
985
59.7k
    }
986
22.7k
    else
987
22.7k
    if( partitioner.isConsIntra() && !cuECtx.bestTU )
988
0
    {
989
      //return true;
990
0
    }
991
22.7k
    else
992
22.7k
    if ( lumaArea.width == 4 && lumaArea.height == 4 && !slice.isIntra() && !cuECtx.bestTU )
993
0
    {
994
      //return true;
995
0
    }
996
22.7k
    else
997
22.7k
    if (!(slice.isIRAP() || bestMode.type == ETM_INTRA || !cuECtx.bestTU ||
998
0
      ((!m_pcEncCfg->m_bDisableIntraCUsInInterSlices) && (!relatedCU.isInter || !relatedCU.isIBC) && (
999
0
                                    ( cuECtx.bestTU->cbf[0] != 0 ) ||
1000
0
           ( ( numComp > COMP_Cb ) && cuECtx.bestTU->cbf[1] != 0 ) ||
1001
0
           ( ( numComp > COMP_Cr ) && cuECtx.bestTU->cbf[2] != 0 )  // avoid very complex intra if it is unlikely
1002
0
         ) ) ) )
1003
0
    {
1004
0
      return false;
1005
0
    }
1006
22.7k
    else
1007
22.7k
    if( cuECtx.bestCS && cuECtx.bestCU && cuECtx.interHad )
1008
22.7k
    {
1009
      // Get SATD threshold from best Inter-CU
1010
22.7k
      if( !cs.slice->isIRAP() && m_pcEncCfg->m_usePbIntraFast )
1011
0
      {
1012
0
        const DFunc dfunc = DF_HAD;
1013
0
        DistParam distParam = m_pcRdCost->setDistParam( cs.getOrgBuf( COMP_Y ), cuECtx.bestCS->getPredBuf( COMP_Y ), cs.sps->bitDepths[ CH_L ], dfunc );
1014
0
        cuECtx.interHad = distParam.distFunc( distParam );
1015
0
      }
1016
22.7k
    }
1017
82.4k
    if ((m_pcEncCfg->m_IBCFastMethod > 1) && !cs.slice->isIntra()//IBC_FAST_METHOD_NOINTRA_IBCCBF0
1018
0
      && (bestMode.type == ETM_IBC || bestMode.type == ETM_IBC_MERGE)
1019
0
      && (!cuECtx.bestCU->Y().valid() || cuECtx.bestTU->cbf[0] == 0)
1020
0
      && (!cuECtx.bestCU->Cb().valid() || cuECtx.bestTU->cbf[1] == 0)
1021
0
      && (!cuECtx.bestCU->Cr().valid() || cuECtx.bestTU->cbf[2] == 0))
1022
0
    {
1023
0
      return false;
1024
0
    }
1025
82.4k
    if (m_pcEncCfg->m_FastIntraTools)
1026
82.4k
    {
1027
82.4k
      if (relatedCU.relatedCuIsValid)
1028
13.6k
      {
1029
13.6k
        cuECtx.relatedCuIsValid = relatedCU.relatedCuIsValid;
1030
13.6k
      }
1031
82.4k
      if (relatedCU.isIntra)
1032
22.4k
      {
1033
22.4k
        cuECtx.isIntra = relatedCU.isIntra;
1034
22.4k
      }
1035
82.4k
    }
1036
82.4k
  }
1037
50.8k
  else if( isModeInter( encTestmode ) )
1038
0
  {
1039
    // INTER MODES (ME + MERGE/SKIP)
1040
0
    CHECK( slice.isIntra(), "Inter-mode should not be in the I-Slice mode list!" );
1041
1042
    // --- Check if we can quit current mode using SAVE/LOAD coding history
1043
1044
0
    if( encTestmode.type == ETM_INTER_ME )
1045
0
    {
1046
0
      if( encTestmode.opts == ETO_STANDARD )
1047
0
      {
1048
0
        if (m_pcEncCfg->m_FastInferMerge)
1049
0
        {
1050
0
          if (bestCS && (bestCS->slice->TLayer > (m_pcEncCfg->m_maxTLayer - (m_pcEncCfg->m_FastInferMerge & 7)))
1051
0
            && (bestCS->bestParent != nullptr) && bestCS->bestParent->cus.size() && (bestCS->bestParent->cus[0]->skip))
1052
0
          {
1053
0
            return false;
1054
0
          }
1055
0
        }
1056
0
        if( relatedCU.isSkip || relatedCU.isIntra )
1057
0
        {
1058
0
          return false;
1059
0
        }
1060
0
      }
1061
0
    }
1062
0
  }
1063
50.8k
  else if (encTestmode.type == ETM_IBC || encTestmode.type == ETM_IBC_MERGE)
1064
50.8k
  {
1065
50.8k
    if ((m_pcEncCfg->m_IBCFastMethod > 1) && !(slice.isIRAP() || bestMode.type == ETM_INTRA || !cuECtx.bestTU ||
1066
0
      ((!m_pcEncCfg->m_bDisableIntraCUsInInterSlices) && (!relatedCU.isInter || !relatedCU.isIBC) && (
1067
0
        (cuECtx.bestTU->cbf[0] != 0) ||
1068
0
        ((numComp > COMP_Cb) && cuECtx.bestTU->cbf[1] != 0) ||
1069
0
        ((numComp > COMP_Cr) && cuECtx.bestTU->cbf[2] != 0)  // avoid very complex intra if it is unlikely
1070
0
        ))))
1071
0
    {
1072
0
      return false;
1073
0
    }
1074
50.8k
    if ((m_pcEncCfg->m_IBCFastMethod > 3) &&(lumaArea.width == 4 && lumaArea.height == 4 && !slice.isIntra()))
1075
0
    {
1076
0
      return false;
1077
0
    }
1078
    // IBC MODES
1079
50.8k
    return slice.sps->IBC && (partitioner.currArea().lumaSize().width < 128 && partitioner.currArea().lumaSize().height < 128);
1080
50.8k
  }
1081
0
  else
1082
0
  {
1083
0
    THROW("problem");
1084
0
    return false;
1085
0
  }
1086
1087
82.4k
  STAT_COUNT_CU_MODES( partitioner.chType == CH_L, g_cuCounters1D[CU_MODES_TRIED][0][!cs.slice->isIntra() + cs.slice->depth] );
1088
82.4k
  STAT_COUNT_CU_MODES( partitioner.chType == CH_L && !cs.slice->isIntra(), g_cuCounters2D[CU_MODES_TRIED][Log2( cs.area.lheight() )][Log2( cs.area.lwidth() )] );
1089
82.4k
  return true;
1090
133k
}
1091
1092
1093
void EncModeCtrl::beforeSplit( Partitioner& partitioner )
1094
82.4k
{
1095
82.4k
  ComprCUCtx&           cuECtx  = m_ComprCUCtxList.back();
1096
1097
82.4k
  if( ! cuECtx.bestCS )
1098
0
    return;
1099
1100
82.4k
  CodedCUInfo    &relatedCU   = getBlkInfo( partitioner.currArea() );
1101
82.4k
  const CodingUnit&  bestCU   = *cuECtx.bestCU;
1102
1103
82.4k
  cuECtx.bestNsPredMode       = cuECtx.bestMode;
1104
82.4k
  cuECtx.bestCostBeforeSplit  = cuECtx.bestCS->cost;
1105
1106
82.4k
  setFromCs( *cuECtx.bestCS, cuECtx.bestMode, partitioner );
1107
1108
82.4k
  if( bestCU.skip )
1109
0
  {
1110
0
    cuECtx.doMoreSplits--;
1111
0
  }
1112
1113
82.4k
  if( partitioner.modeType == MODE_TYPE_INTRA && partitioner.chType == CH_L )
1114
0
  {
1115
0
    return; //not set best coding mode for intra coding pass
1116
0
  }
1117
1118
  // assume the non-split modes are done and set the marks for the best found mode
1119
82.4k
  if( CU::isInter( bestCU ) )
1120
0
  {
1121
0
    relatedCU.isInter     = true;
1122
0
    relatedCU.isSkip     |= bestCU.skip;
1123
0
    relatedCU.isMMVDSkip |= bestCU.mmvdSkip;
1124
0
    relatedCU.BcwIdx      = bestCU.BcwIdx;
1125
0
  }
1126
82.4k
  else if (CU::isIBC(bestCU))
1127
0
  {
1128
0
    relatedCU.isIBC = true;
1129
0
    relatedCU.isSkip |= bestCU.skip;
1130
0
  }
1131
82.4k
  else if( CU::isIntra( bestCU ) )
1132
82.4k
  {
1133
82.4k
    relatedCU.isIntra     = true;
1134
82.4k
    if (m_pcEncCfg->m_FastIntraTools)
1135
82.4k
    {
1136
82.4k
      if ( cuECtx.intraWasTested && (!relatedCU.relatedCuIsValid || cuECtx.bestCS->cost < relatedCU.bestCost))
1137
25.4k
      {
1138
25.4k
        relatedCU.bestCost = cuECtx.bestCS->cost;
1139
25.4k
        relatedCU.relatedCuIsValid = true;
1140
25.4k
      }
1141
82.4k
    }
1142
82.4k
  }
1143
82.4k
  cuECtx.isBestNoSplitSkip = bestCU.skip;
1144
82.4k
}
1145
1146
1147
bool EncModeCtrl::useModeResult( const EncTestMode& encTestmode, CodingStructure*& tempCS, Partitioner& partitioner, const bool useEDO )
1148
172k
{
1149
172k
  ComprCUCtx& cuECtx = m_ComprCUCtxList.back();
1150
1151
1152
172k
  if(      encTestmode.type == ETM_SPLIT_BT_H )
1153
24.0k
  {
1154
24.0k
    cuECtx.bestCostHorzSplit = tempCS->cost;
1155
24.0k
  }
1156
148k
  else if( encTestmode.type == ETM_SPLIT_BT_V )
1157
24.2k
  {
1158
24.2k
    cuECtx.bestCostVertSplit = tempCS->cost;
1159
24.2k
  }
1160
124k
  else if( !isModeSplit( encTestmode ) && isModeInter( encTestmode ) && tempCS->cus.size() == 1 )
1161
0
  {
1162
0
    cuECtx.nonSkipWasTested |= !tempCS->cus.front()->skip;
1163
0
  }
1164
172k
  if (m_pcEncCfg->m_AMVRspeed && encTestmode.type == ETM_INTER_ME)
1165
0
  {
1166
0
    int imvMode = (encTestmode.opts & ETO_IMV) >> ETO_IMV_SHIFT;
1167
1168
0
    if (imvMode == 0)
1169
0
    {
1170
0
      if (tempCS->cost < cuECtx.bestCostNoImv )
1171
0
      {
1172
0
        cuECtx.bestCostNoImv = tempCS->cost;
1173
0
      }
1174
0
    }
1175
0
  }
1176
1177
172k
  if( encTestmode.type == ETM_SPLIT_QT )
1178
19.0k
  {
1179
19.0k
    int maxQtD = 0;
1180
19.0k
    for( const auto& cu : tempCS->cus )
1181
75.6k
    {
1182
75.6k
      maxQtD = std::max<int>( maxQtD, cu->qtDepth );
1183
75.6k
    }
1184
19.0k
    cuECtx.maxQtSubDepth = maxQtD;
1185
19.0k
  }
1186
1187
172k
  int maxMtD = tempCS->pcv->getMaxMTTDepth( *tempCS->slice, partitioner.chType ) + partitioner.currImplicitBtDepth;
1188
1189
172k
  if( encTestmode.type == ETM_SPLIT_BT_H )
1190
24.0k
  {
1191
24.0k
    if( tempCS->cus.size() > 2 )
1192
963
    {
1193
963
      int h_2   = tempCS->area.blocks[partitioner.chType].height / 2;
1194
963
      int cu1_h = tempCS->cus.front()->blocks[partitioner.chType].height;
1195
963
      int cu2_h = tempCS->cus.back() ->blocks[partitioner.chType].height;
1196
1197
963
      cuECtx.doTriHorzSplit = cu1_h < h_2 || cu2_h < h_2 || partitioner.currMtDepth + 1 == maxMtD;
1198
963
    }
1199
24.0k
  }
1200
148k
  else if( encTestmode.type == ETM_SPLIT_BT_V )
1201
24.2k
  {
1202
24.2k
    if( tempCS->cus.size() > 2 )
1203
1.06k
    {
1204
1.06k
      int w_2   = tempCS->area.blocks[partitioner.chType].width / 2;
1205
1.06k
      int cu1_w = tempCS->cus.front()->blocks[partitioner.chType].width;
1206
1.06k
      int cu2_w = tempCS->cus.back() ->blocks[partitioner.chType].width;
1207
1208
1.06k
      cuECtx.doTriVertSplit = cu1_w < w_2 || cu2_w < w_2 || partitioner.currMtDepth + 1 == maxMtD;
1209
1.06k
    }
1210
24.2k
  }
1211
1212
172k
  if( encTestmode.type == ETM_SPLIT_BT_V || encTestmode.type == ETM_SPLIT_BT_H || encTestmode.type == ETM_SPLIT_QT )
1213
67.2k
  {
1214
67.2k
    bool isAllSkip = true;
1215
1216
67.2k
    for( const auto& cu : tempCS->cus )
1217
151k
    {
1218
151k
      isAllSkip &= cu->skip;
1219
151k
    }
1220
1221
67.2k
    if( isAllSkip ) cuECtx.doMoreSplits--;
1222
67.2k
    if( isAllSkip && encTestmode.type == ETM_SPLIT_QT )
1223
0
                    cuECtx.doMoreSplits--;
1224
67.2k
  }
1225
1226
  // for now just a simple decision based on RD-cost or choose tempCS if bestCS is not yet coded
1227
172k
  if( tempCS->cost != MAX_DOUBLE && ( !cuECtx.bestCS || ( ( tempCS->cost + ( useEDO ? tempCS->costDbOffset : 0 ) ) < ( cuECtx.bestCS->cost + ( useEDO ? cuECtx.bestCS->costDbOffset : 0 ) ) ) ) )
1228
156k
  {
1229
156k
    cuECtx.bestCS   = tempCS;
1230
156k
    cuECtx.bestCU   = tempCS->cus[0];
1231
156k
    cuECtx.bestTU   = cuECtx.bestCU->firstTU;
1232
156k
    cuECtx.bestMode = encTestmode;
1233
1234
156k
    return true;
1235
156k
  }
1236
16.4k
  else
1237
16.4k
  {
1238
16.4k
    return false;
1239
16.4k
  }
1240
172k
}
1241
1242
} // namespace vvenc
1243
1244
//! \}
1245