Coverage Report

Created: 2026-09-02 06:43

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
18.3k
{
66
18.3k
  const unsigned numPos = MAX_CU_SIZE >> MIN_CU_LOG2;
67
18.3k
  const int maxSizeIdx  = MAX_CU_SIZE_IDX - MIN_CU_LOG2;
68
69
  //static constexpr size_t numCu = 7921;
70
71
18.3k
  size_t numCu = 0;
72
  
73
128k
  for( int wIdx = 0; wIdx < maxSizeIdx; wIdx++ )
74
110k
  {
75
770k
    for( int hIdx = 0; hIdx < maxSizeIdx; hIdx++ )
76
660k
    {
77
21.8M
      for( unsigned y = 0; y < numPos; y++ )
78
21.1M
      {
79
697M
        for( unsigned x = 0; x < numPos; x++ )
80
676M
        {
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
676M
          if( (x + (1 << (wIdx)) <= (ctuSize >> MIN_CU_LOG2))
87
475M
              && (y + (1 << (hIdx)) <= (ctuSize >> MIN_CU_LOG2))
88
334M
              && (((x << MIN_CU_LOG2) & ((1 << (wIdx + MIN_CU_LOG2 - 1)) - 1)) == 0)
89
220M
              && (((y << MIN_CU_LOG2) & ((1 << (hIdx + MIN_CU_LOG2 - 1)) - 1)) == 0) )
90
145M
          {
91
145M
            numCu++;
92
145M
          }
93
676M
        }
94
21.1M
      }
95
660k
    }
96
110k
  }
97
98
18.3k
  m_codedCUInfoBuf = new CodedCUInfo[numCu];
99
18.3k
  CodedCUInfo* cuInfo = m_codedCUInfoBuf;
100
101
128k
  for( int wIdx = 0; wIdx < maxSizeIdx; wIdx++ )
102
110k
  {
103
770k
    for( int hIdx = 0; hIdx < maxSizeIdx; hIdx++ )
104
660k
    {
105
21.8M
      for( unsigned y = 0; y < numPos; y++ )
106
21.1M
      {
107
697M
        for( unsigned x = 0; x < numPos; x++ )
108
676M
        {
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
676M
          if(( x + (1<<(wIdx)) <= ( ctuSize >> MIN_CU_LOG2 ) )
115
475M
            && ( y + (1<<(hIdx)) <= ( ctuSize >> MIN_CU_LOG2 ) )
116
334M
            && ( ( ( x << MIN_CU_LOG2 ) & ((1 << (wIdx + MIN_CU_LOG2 - 1)) - 1) ) == 0 )
117
220M
            && ( ( ( y << MIN_CU_LOG2 ) & ((1 << (hIdx + MIN_CU_LOG2 - 1)) - 1) ) == 0 ) )
118
145M
          {
119
145M
            m_codedCUInfo[wIdx][hIdx][x][y] = cuInfo++;
120
145M
            m_codedCUInfo[wIdx][hIdx][x][y]->poc       = -1;
121
145M
            m_codedCUInfo[wIdx][hIdx][x][y]->ctuRsAddr = -1;
122
145M
          }
123
531M
          else
124
531M
          {
125
531M
            m_codedCUInfo[wIdx][hIdx][x][y] = nullptr;
126
531M
          }
127
676M
        }
128
21.1M
      }
129
660k
    }
130
110k
  }
131
18.3k
}
132
133
void CacheBlkInfoCtrl::destroy()
134
18.3k
{
135
18.3k
  delete[] m_codedCUInfoBuf;
136
18.3k
  m_codedCUInfoBuf = nullptr;
137
18.3k
}
138
139
void CacheBlkInfoCtrl::init( const Slice &slice )
140
3.59k
{
141
3.59k
  m_pcv = slice.pps->pcv;
142
3.59k
}
143
144
void CacheBlkInfoCtrl::initBlk( const UnitArea& area, int poc )
145
123k
{
146
123k
  unsigned idx1, idx2, idx3, idx4;
147
123k
  getAreaIdxNew( area.Y(), *m_pcv, idx1, idx2, idx3, idx4 );
148
149
123k
  const int ctuRsAddr = getCtuAddr( area.lumaPos(), *m_pcv );
150
123k
  CodedCUInfo* cuInfo = m_codedCUInfo[idx1][idx2][idx3][idx4];
151
152
123k
  if( cuInfo->poc != poc || cuInfo->ctuRsAddr != ctuRsAddr )
153
88.4k
  {
154
88.4k
    GCC_WARNING_DISABLE_class_memaccess
155
88.4k
    memset( cuInfo, 0, sizeof( CodedCUInfo ) );
156
88.4k
    GCC_WARNING_RESET
157
158
88.4k
    cuInfo->poc       = poc;
159
88.4k
    cuInfo->ctuRsAddr = ctuRsAddr;
160
88.4k
  }
161
123k
}
162
163
CodedCUInfo& CacheBlkInfoCtrl::getBlkInfo( const UnitArea& area )
164
203k
{
165
203k
  unsigned idx1, idx2, idx3, idx4;
166
203k
  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
203k
  return *m_codedCUInfo[idx1][idx2][idx3][idx4];
169
203k
}
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
18.3k
{
261
18.3k
  m_reuseCuResults = reuseCuResults;
262
263
18.3k
  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
18.3k
{
403
18.3k
  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
3.59k
{
419
3.59k
  if( !m_reuseCuResults ) return;
420
421
1
  bool isInitialized = m_pcv;
422
423
1
  m_pcv = slice.pps->pcv;
424
425
1
  if( isInitialized ) return;
426
427
1
  m_dummyCS.pcv = m_pcv;
428
1
}
429
430
bool BestEncInfoCache::setFromCs( const CodingStructure& cs, const EncTestMode& testMode, const Partitioner& partitioner )
431
77.9k
{
432
77.9k
  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
201k
{
461
201k
  if( !m_reuseCuResults || partitioner.treeType == TREE_C || partitioner.maxBTD <= 1 )
462
201k
  {
463
201k
    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
201k
  }
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
18.3k
{
540
18.3k
  m_pcEncCfg = &encCfg;
541
18.3k
  m_pcRdCost = pRdCost;
542
18.3k
  comprCUCtx = nullptr;
543
544
18.3k
  CacheBlkInfoCtrl::create( encCfg.m_CTUSize );
545
18.3k
  BestEncInfoCache::create( encCfg.m_reuseCuResults, encCfg.m_internChromaFormat, encCfg.m_CTUSize );
546
18.3k
}
547
548
void EncModeCtrl::destroy()
549
18.3k
{
550
18.3k
  CacheBlkInfoCtrl::destroy();
551
18.3k
  BestEncInfoCache::destroy();
552
18.3k
}
553
554
void EncModeCtrl::initCTUEncoding( const Slice &slice, int tileIdx )
555
3.59k
{
556
3.59k
  CacheBlkInfoCtrl::init( slice );
557
3.59k
  BestEncInfoCache::init( slice );
558
559
3.59k
  CHECK( !m_ComprCUCtxList.empty(), "Mode list is not empty at the beginning of a CTU" );
560
561
3.59k
  if( m_pcEncCfg->m_fastQtBtEnc )
562
3.59k
  {
563
3.59k
    m_skipThresholdE0023FastEnc = ((slice.getMinPictureDistance() <= PICTURE_DISTANCE_TH) ? FAST_SKIP_DEPTH : SKIP_DEPTH);
564
3.59k
  }
565
1
  else
566
1
  {
567
1
    m_skipThresholdE0023FastEnc = SKIP_DEPTH;
568
1
  }
569
570
3.59k
  m_tileIdx = tileIdx;
571
3.59k
}
572
573
void EncModeCtrl::initCULevel( Partitioner &partitioner, const CodingStructure& cs, int  MergeSimpleFlag)
574
123k
{
575
  // Min/max depth
576
123k
  unsigned minDepth = 0;
577
123k
  unsigned maxDepth = cs.pcv->getMaxDepth( cs.slice->sliceType, partitioner.chType );
578
123k
  if( m_pcEncCfg->m_useFastLCTU )
579
123k
  {
580
123k
    partitioner.setMaxMinDepth(minDepth, maxDepth, cs, cs.picture->useQtbttSpeedUpMode, MergeSimpleFlag);
581
123k
  }
582
583
123k
  minDepth = std::max<unsigned>( minDepth, cs.pcv->getMinDepth( cs.slice->sliceType, partitioner.chType ) );
584
123k
  maxDepth = std::min<unsigned>( maxDepth, cs.pcv->getMaxDepth( cs.slice->sliceType, partitioner.chType ) );
585
586
123k
  m_ComprCUCtxList.push_back( ComprCUCtx( cs, minDepth, maxDepth ) );
587
123k
  comprCUCtx = &m_ComprCUCtxList.back();
588
589
123k
  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
123k
  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
123k
  const bool qtBeforeBt = ( (  cuLeft  &&  cuAbove  && cuLeft ->qtDepth > partitioner.currQtDepth && cuAbove->qtDepth > partitioner.currQtDepth )
593
121k
                         || (  cuLeft  && !cuAbove  && cuLeft ->qtDepth > partitioner.currQtDepth )
594
119k
                         || ( !cuLeft  &&  cuAbove  && cuAbove->qtDepth > partitioner.currQtDepth )
595
116k
                         || ( !cuAbove && !cuLeft   && cs.area.lwidth() >= ( 32 << cs.slice->TLayer ) )
596
103k
                         || ( m_pcEncCfg->m_qtbttSpeedUp > 1 && partitioner.maxBTD < ( ( cs.slice->isIntra() && !cs.sps->IBC ) ? 3 : 2 ) ) )
597
19.7k
                         && ( cs.area.lwidth() > ( cs.pcv->getMinQtSize( *cs.slice, partitioner.chType ) << 1 ) );
598
599
  // set features
600
123k
  ComprCUCtx &cuECtx    = *comprCUCtx;
601
123k
  cuECtx.qtBeforeBt     = qtBeforeBt;
602
123k
  cuECtx.doTriHorzSplit = true;
603
123k
  cuECtx.doTriVertSplit = true;
604
123k
  cuECtx.doMoreSplits   = 3;
605
123k
  cuECtx.isReusingCu    = isReusingCuValid( cs, partitioner, cs.baseQP );
606
123k
  cuECtx.didHorzSplit   = partitioner.canSplit( CU_HORZ_SPLIT, cs );
607
123k
  cuECtx.didVertSplit   = partitioner.canSplit( CU_VERT_SPLIT, cs );
608
123k
  cuECtx.doHorChromaSplit = true;
609
123k
  cuECtx.doVerChromaSplit = true;
610
123k
  cuECtx.doQtChromaSplit  = true;
611
612
123k
  if( m_pcEncCfg->m_contentBasedFastQtbt && cs.pcv->getMaxMTTDepth(*cs.slice, partitioner.chType))
613
123k
  {
614
123k
    const CompArea& currArea = partitioner.currArea().Y();
615
123k
    int cuHeight  = currArea.height;
616
123k
    int cuWidth   = currArea.width;
617
618
123k
    const bool condIntraInter = m_pcEncCfg->m_IntraPeriod == 1 ? ( partitioner.currBtDepth == 0 ) : ( cuHeight > 32 && cuWidth > 32 );
619
620
123k
    if( cuWidth == cuHeight && condIntraInter )
621
26.1k
    {
622
26.1k
      const CPelBuf bufCurrArea = cs.getOrgBuf( partitioner.currArea().block( COMP_Y ) );
623
624
26.1k
      Intermediate_Int horVal = 0;
625
26.1k
      Intermediate_Int verVal = 0;
626
26.1k
      Intermediate_Int dupVal = 0;
627
26.1k
      Intermediate_Int dowVal = 0;
628
629
26.1k
      unsigned j, k;
630
631
2.10M
      for( k = 0; k < cuHeight - 1; k++ )
632
2.07M
      {
633
190M
        for( j = 0; j < cuWidth - 1; j++ )
634
188M
        {
635
188M
          horVal += abs( bufCurrArea.at( j + 1, k     ) - bufCurrArea.at( j, k ) );
636
188M
          verVal += abs( bufCurrArea.at( j    , k + 1 ) - bufCurrArea.at( j, k ) );
637
188M
          dowVal += abs( bufCurrArea.at( j + 1, k )     - bufCurrArea.at( j, k + 1 ) );
638
188M
          dupVal += abs( bufCurrArea.at( j + 1, k + 1 ) - bufCurrArea.at( j, k ) );
639
188M
        }
640
2.07M
      }
641
642
26.1k
      cuECtx.grad_horVal = (double)horVal;
643
26.1k
      cuECtx.grad_verVal = (double)verVal;
644
26.1k
      cuECtx.grad_dowVal = (double)dowVal;
645
26.1k
      cuECtx.grad_dupVal = (double)dupVal;
646
26.1k
    }
647
123k
  }
648
123k
}
649
650
void EncModeCtrl::finishCULevel( Partitioner &partitioner )
651
123k
{
652
123k
  m_ComprCUCtxList.pop_back();
653
123k
  comprCUCtx = m_ComprCUCtxList.size() ? &m_ComprCUCtxList.back() : nullptr;
654
123k
}
655
656
bool EncModeCtrl::trySplit( const EncTestMode& encTestmode, const CodingStructure &cs, Partitioner& partitioner, const EncTestMode& lastTestmode )
657
300k
{
658
300k
  ComprCUCtx& cuECtx = *comprCUCtx;
659
660
300k
  const PartSplit implicitSplit = partitioner.getImplicitSplit( cs );
661
300k
  const bool isBoundary         = implicitSplit != CU_DONT_SPLIT;
662
663
300k
  if( ( m_pcEncCfg->m_IntraPeriod == 1 ) && ( partitioner.chType == CH_C ) && ( !cuECtx.doQtChromaSplit ) )
664
0
  {
665
0
    cuECtx.maxDepth         = partitioner.currDepth;
666
0
  }
667
668
300k
  if( isBoundary )
669
75.3k
  {
670
75.3k
    if( encTestmode.type != ETM_SPLIT_QT )
671
29.8k
    {
672
29.8k
      return getPartSplit( encTestmode ) == implicitSplit;
673
29.8k
    }
674
45.5k
    else
675
45.5k
    {
676
45.5k
      return partitioner.canSplit( CU_QUAD_SPLIT, cs );
677
45.5k
    }
678
75.3k
  }
679
680
225k
  const Slice&           slice       = *cs.slice;
681
225k
  const uint32_t         width       = partitioner.currArea().lumaSize().width;
682
225k
  const CodingStructure *bestCS      = cuECtx.bestCS;
683
225k
  const CodingUnit      *bestCU      = cuECtx.bestCU;
684
685
225k
  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
225k
  else if( encTestmode.type == ETM_SPLIT_QT && cuECtx.maxDepth <= partitioner.currQtDepth )
691
1.22k
  {
692
    // don't check this QT depth
693
1.22k
    return false;
694
1.22k
  }
695
696
224k
  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
224k
  int skipScore = 0;
703
704
224k
  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
224k
  const PartSplit split = getPartSplit( encTestmode );
720
224k
  if( !partitioner.canSplit( split, cs ) || skipScore >= 2 || ( skipScore == 1 && m_ComprCUCtxList.size() == 2 ) )
721
57.3k
  {
722
57.3k
    if( split == CU_HORZ_SPLIT ) cuECtx.didHorzSplit = false;
723
57.3k
    if( split == CU_VERT_SPLIT ) cuECtx.didVertSplit = false;
724
57.3k
    if( split == CU_QUAD_SPLIT ) cuECtx.didQuadSplit = false;
725
57.3k
    return false;
726
57.3k
  }
727
728
166k
  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
166k
  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
166k
  if( m_pcEncCfg->m_contentBasedFastQtbt )
749
166k
  {
750
166k
    const CompArea& currArea = partitioner.currArea().Y();
751
166k
    int cuHeight  = currArea.height;
752
166k
    int cuWidth   = currArea.width;
753
754
166k
    const bool condIntraInter = m_pcEncCfg->m_IntraPeriod == 1 ? ( partitioner.currBtDepth == 0 ) : ( cuHeight > 32 && cuWidth > 32 );
755
756
166k
    if( cuWidth == cuHeight && condIntraInter && split != CU_QUAD_SPLIT )
757
8.94k
    {
758
8.94k
      const double th1 = m_pcEncCfg->m_IntraPeriod == 1 ?  1.2              :  1.0;
759
8.94k
      const double th2 = m_pcEncCfg->m_IntraPeriod == 1 ? (1.2 / sqrt( 2 )) : (1.0 / sqrt( 2 ));
760
761
8.94k
      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
8.94k
      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
8.94k
    }
770
771
166k
    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
166k
  }
779
780
166k
  if( m_pcEncCfg->m_qtbttSpeedUp > 1 )
781
166k
  {
782
166k
    const int availDepth = cs.pcv->getMaxMTTDepth( slice, partitioner.chType ) - partitioner.currMtDepth;
783
166k
    if( bestCU && bestCU->skip && availDepth <= ( 3 - m_skipThresholdE0023FastEnc ) && !isModeSplit( lastTestmode ) && split != CU_QUAD_SPLIT )
784
0
    {
785
0
      return false;
786
0
    }
787
166k
  }
788
166k
  if( bestCU && bestCU->skip && bestCU->mtDepth >= m_skipThresholdE0023FastEnc && !isModeSplit( lastTestmode ) )
789
0
  {
790
0
    return false;
791
0
  }
792
793
166k
  bool resetFeature = false;
794
795
166k
  switch( split )
796
166k
  {
797
19.4k
    case CU_QUAD_SPLIT:
798
19.4k
      {
799
19.4k
        if( !cuECtx.qtBeforeBt && bestCU )
800
17.3k
        {
801
17.3k
          unsigned maxBTD        = cs.pcv->getMaxMTTDepth( slice, partitioner.chType );
802
17.3k
          const CodingUnit *cuBR = bestCS->cus.back();
803
17.3k
          unsigned height        = partitioner.currArea().lumaSize().height;
804
805
17.3k
          if(((bestCU->btDepth == 0 && maxBTD >= ((slice.isIntra() && !slice.sps->IBC) ? 3 : 2))
806
84
            || (bestCU->btDepth == 1 && cuBR && cuBR->btDepth == 1 && maxBTD >= ((slice.isIntra() && !slice.sps->IBC) ? 4 : 3)))
807
17.2k
            && (width <= MAX_TB_SIZEY && height <= MAX_TB_SIZEY)
808
17.2k
            && cuECtx.didHorzSplit && cuECtx.didVertSplit )
809
13.3k
          {
810
13.3k
            return false;
811
13.3k
          }
812
17.3k
        }
813
6.08k
        if( bestCS )
814
6.08k
        {
815
6.08k
          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
6.08k
          int stopSplit = (m_pcEncCfg->m_FastInferMerge >> 4) && (bestCS->slice->TLayer > 4);
821
6.08k
          int limitBLsize = stopSplit ? 2048 : 1024;
822
6.08k
          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
6.08k
        }
827
6.08k
      }
828
6.08k
      break;
829
51.2k
    case CU_HORZ_SPLIT:
830
98.1k
    case CU_VERT_SPLIT:
831
98.1k
      resetFeature = true;
832
98.1k
      break;
833
27.6k
    case CU_TRIH_SPLIT:
834
27.6k
      if( cuECtx.didHorzSplit && bestCU && bestCU->btDepth == partitioner.currBtDepth && !bestCU->rootCbf )
835
22.3k
      {
836
22.3k
        return false;
837
22.3k
      }
838
839
5.27k
      if( m_pcEncCfg->m_qtbttSpeedUp > 1 )
840
5.27k
      {
841
        //unsigned maxBTD = cs.pcv->getMaxMTTDepth( slice, partitioner.chType );
842
5.27k
        if( /*maxBTD == 1 && */cuECtx.didHorzSplit && cuECtx.didVertSplit && cuECtx.bestCostHorzSplit > cuECtx.bestCostVertSplit )
843
1.50k
        {
844
1.50k
          return false;
845
1.50k
        }
846
5.27k
      }
847
848
3.76k
      if( !cuECtx.doTriHorzSplit )
849
12
      {
850
12
        return false;
851
12
      }
852
853
3.75k
      if (m_pcEncCfg->m_fastTTSplit && cuECtx.bestCostHorzSplit < MAX_DOUBLE)
854
2.76k
      {
855
2.76k
        if (cuECtx.bestCostBeforeSplit < MAX_DOUBLE && (cuECtx.bestCostHorzSplit > m_pcEncCfg->m_fastTT_th * cuECtx.bestCostBeforeSplit))
856
2.28k
        {
857
2.28k
          return false;
858
2.28k
        }
859
483
        if (cuECtx.bestCostVertSplit < MAX_DOUBLE && (cuECtx.bestCostHorzSplit > m_pcEncCfg->m_fastTT_th * cuECtx.bestCostVertSplit))
860
0
        {
861
0
          return false;
862
0
        }
863
483
      }
864
1.47k
      break;
865
21.7k
    case CU_TRIV_SPLIT:
866
21.7k
      if( cuECtx.didVertSplit && bestCU && bestCU->btDepth == partitioner.currBtDepth && !bestCU->rootCbf )
867
18.4k
      {
868
18.4k
        return false;
869
18.4k
      }
870
871
3.25k
      if( m_pcEncCfg->m_qtbttSpeedUp > 1 )
872
3.25k
      {
873
        //unsigned maxBTD = cs.pcv->getMaxMTTDepth( slice, partitioner.chType );
874
3.25k
        if( /*maxBTD == 1 && */cuECtx.didHorzSplit && cuECtx.didVertSplit && cuECtx.bestCostHorzSplit < cuECtx.bestCostVertSplit )
875
1.73k
        {
876
1.73k
          return false;
877
1.73k
        }
878
3.25k
      }
879
880
1.51k
      if( !cuECtx.doTriVertSplit )
881
43
      {
882
43
        return false;
883
43
      }
884
885
1.47k
      if (m_pcEncCfg->m_fastTTSplit && cuECtx.bestCostVertSplit < MAX_DOUBLE)
886
536
      {
887
536
        if (cuECtx.bestCostBeforeSplit < MAX_DOUBLE && (cuECtx.bestCostVertSplit > m_pcEncCfg->m_fastTT_th * cuECtx.bestCostBeforeSplit))
888
347
        {
889
347
          return false;
890
347
        }
891
189
        if (cuECtx.bestCostHorzSplit < MAX_DOUBLE && (cuECtx.bestCostVertSplit > m_pcEncCfg->m_fastTT_th * cuECtx.bestCostHorzSplit))
892
0
        {
893
0
          return false;
894
0
        }
895
189
      }
896
1.12k
      break;
897
1.12k
    default:
898
0
      THROW( "Only CU split modes are governed by the EncModeCtrl" );
899
0
      return false;
900
0
      break;
901
166k
  }
902
903
106k
  switch( split )
904
106k
  {
905
51.2k
    case CU_HORZ_SPLIT:
906
52.7k
    case CU_TRIH_SPLIT:
907
52.7k
      if( cuECtx.qtBeforeBt && cuECtx.didQuadSplit )
908
1.76k
      {
909
1.76k
        if( cuECtx.maxQtSubDepth > partitioner.currQtDepth + 1 )
910
0
        {
911
0
          if( resetFeature ) cuECtx.didHorzSplit = false;
912
0
          return false;
913
0
        }
914
1.76k
      }
915
52.7k
      break;
916
52.7k
    case CU_VERT_SPLIT:
917
47.9k
    case CU_TRIV_SPLIT:
918
47.9k
      if( cuECtx.qtBeforeBt && cuECtx.didQuadSplit )
919
1.84k
      {
920
1.84k
        if( cuECtx.maxQtSubDepth > partitioner.currQtDepth + 1 )
921
0
        {
922
0
          if( resetFeature ) cuECtx.didVertSplit = false;
923
0
          return false;
924
0
        }
925
1.84k
      }
926
47.9k
      break;
927
47.9k
    default:
928
6.08k
      break;
929
106k
  }
930
931
106k
  if( split == CU_QUAD_SPLIT )
932
6.08k
  {
933
6.08k
    cuECtx.didQuadSplit = m_pcEncCfg->m_qtbttSpeedUp <= 1 || !!cuECtx.doMoreSplits;
934
6.08k
  }
935
936
106k
  return m_pcEncCfg->m_qtbttSpeedUp <= 1 || !!cuECtx.doMoreSplits;
937
106k
}
938
939
bool EncModeCtrl::tryMode( const EncTestMode& encTestmode, const CodingStructure &cs, Partitioner& partitioner )
940
125k
{
941
125k
  CHECK( isModeSplit( encTestmode ), "wrong method");
942
943
125k
  ComprCUCtx& cuECtx = m_ComprCUCtxList.back();
944
945
125k
  if( cuECtx.minDepth > partitioner.currQtDepth && partitioner.canSplit( CU_QUAD_SPLIT, cs ) )
946
0
  {
947
    // enforce QT
948
0
    return false;
949
0
  }
950
951
125k
  const Slice&           slice        = *cs.slice;
952
125k
  const uint32_t         numComp      = getNumberValidComponents( slice.sps->chromaFormatIdc );
953
125k
  const CodedCUInfo      &relatedCU   = getBlkInfo( partitioner.currArea() );
954
125k
  const Area             &lumaArea    = partitioner.currArea().Y();
955
125k
  const CodingStructure* bestCS       = cuECtx.bestCS;
956
125k
  const EncTestMode      bestMode     = bestCS ?cuECtx.bestMode : EncTestMode();
957
958
125k
  if( encTestmode.type == ETM_INTRA )
959
77.9k
  {
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
77.9k
    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
77.9k
    if( lumaArea.width > cs.sps->getMaxTbSize() || lumaArea.height > cs.sps->getMaxTbSize() )
972
0
    {
973
0
      return false;
974
0
    }
975
976
77.9k
    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
77.9k
    if (cs.sps->IBC && !cuECtx.bestTU)
983
56.6k
    {
984
     // return true;
985
56.6k
    }
986
21.3k
    else
987
21.3k
    if( partitioner.isConsIntra() && !cuECtx.bestTU )
988
0
    {
989
      //return true;
990
0
    }
991
21.3k
    else
992
21.3k
    if ( lumaArea.width == 4 && lumaArea.height == 4 && !slice.isIntra() && !cuECtx.bestTU )
993
0
    {
994
      //return true;
995
0
    }
996
21.3k
    else
997
21.3k
    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
21.3k
    else
1007
21.3k
    if( cuECtx.bestCS && cuECtx.bestCU && cuECtx.interHad )
1008
21.3k
    {
1009
      // Get SATD threshold from best Inter-CU
1010
21.3k
      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
21.3k
    }
1017
77.9k
    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
77.9k
    if (m_pcEncCfg->m_FastIntraTools)
1026
77.9k
    {
1027
77.9k
      if (relatedCU.relatedCuIsValid)
1028
12.8k
      {
1029
12.8k
        cuECtx.relatedCuIsValid = relatedCU.relatedCuIsValid;
1030
12.8k
      }
1031
77.9k
      if (relatedCU.isIntra)
1032
21.2k
      {
1033
21.2k
        cuECtx.isIntra = relatedCU.isIntra;
1034
21.2k
      }
1035
77.9k
    }
1036
77.9k
  }
1037
47.8k
  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
47.8k
  else if (encTestmode.type == ETM_IBC || encTestmode.type == ETM_IBC_MERGE)
1064
47.8k
  {
1065
47.8k
    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
47.8k
    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
47.8k
    return slice.sps->IBC && (partitioner.currArea().lumaSize().width < 128 && partitioner.currArea().lumaSize().height < 128);
1080
47.8k
  }
1081
0
  else
1082
0
  {
1083
0
    THROW("problem");
1084
0
    return false;
1085
0
  }
1086
1087
77.9k
  STAT_COUNT_CU_MODES( partitioner.chType == CH_L, g_cuCounters1D[CU_MODES_TRIED][0][!cs.slice->isIntra() + cs.slice->depth] );
1088
77.9k
  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
77.9k
  return true;
1090
125k
}
1091
1092
1093
void EncModeCtrl::beforeSplit( Partitioner& partitioner )
1094
77.9k
{
1095
77.9k
  ComprCUCtx&           cuECtx  = m_ComprCUCtxList.back();
1096
1097
77.9k
  if( ! cuECtx.bestCS )
1098
0
    return;
1099
1100
77.9k
  CodedCUInfo    &relatedCU   = getBlkInfo( partitioner.currArea() );
1101
77.9k
  const CodingUnit&  bestCU   = *cuECtx.bestCU;
1102
1103
77.9k
  cuECtx.bestNsPredMode       = cuECtx.bestMode;
1104
77.9k
  cuECtx.bestCostBeforeSplit  = cuECtx.bestCS->cost;
1105
1106
77.9k
  setFromCs( *cuECtx.bestCS, cuECtx.bestMode, partitioner );
1107
1108
77.9k
  if( bestCU.skip )
1109
0
  {
1110
0
    cuECtx.doMoreSplits--;
1111
0
  }
1112
1113
77.9k
  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
77.9k
  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
77.9k
  else if (CU::isIBC(bestCU))
1127
0
  {
1128
0
    relatedCU.isIBC = true;
1129
0
    relatedCU.isSkip |= bestCU.skip;
1130
0
  }
1131
77.9k
  else if( CU::isIntra( bestCU ) )
1132
77.9k
  {
1133
77.9k
    relatedCU.isIntra     = true;
1134
77.9k
    if (m_pcEncCfg->m_FastIntraTools)
1135
77.9k
    {
1136
77.9k
      if ( cuECtx.intraWasTested && (!relatedCU.relatedCuIsValid || cuECtx.bestCS->cost < relatedCU.bestCost))
1137
23.9k
      {
1138
23.9k
        relatedCU.bestCost = cuECtx.bestCS->cost;
1139
23.9k
        relatedCU.relatedCuIsValid = true;
1140
23.9k
      }
1141
77.9k
    }
1142
77.9k
  }
1143
77.9k
  cuECtx.isBestNoSplitSkip = bestCU.skip;
1144
77.9k
}
1145
1146
1147
bool EncModeCtrl::useModeResult( const EncTestMode& encTestmode, CodingStructure*& tempCS, Partitioner& partitioner, const bool useEDO )
1148
163k
{
1149
163k
  ComprCUCtx& cuECtx = m_ComprCUCtxList.back();
1150
1151
1152
163k
  if(      encTestmode.type == ETM_SPLIT_BT_H )
1153
22.7k
  {
1154
22.7k
    cuECtx.bestCostHorzSplit = tempCS->cost;
1155
22.7k
  }
1156
140k
  else if( encTestmode.type == ETM_SPLIT_BT_V )
1157
22.9k
  {
1158
22.9k
    cuECtx.bestCostVertSplit = tempCS->cost;
1159
22.9k
  }
1160
117k
  else if( !isModeSplit( encTestmode ) && isModeInter( encTestmode ) && tempCS->cus.size() == 1 )
1161
0
  {
1162
0
    cuECtx.nonSkipWasTested |= !tempCS->cus.front()->skip;
1163
0
  }
1164
163k
  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
163k
  if( encTestmode.type == ETM_SPLIT_QT )
1178
17.7k
  {
1179
17.7k
    int maxQtD = 0;
1180
17.7k
    for( const auto& cu : tempCS->cus )
1181
71.1k
    {
1182
71.1k
      maxQtD = std::max<int>( maxQtD, cu->qtDepth );
1183
71.1k
    }
1184
17.7k
    cuECtx.maxQtSubDepth = maxQtD;
1185
17.7k
  }
1186
1187
163k
  int maxMtD = tempCS->pcv->getMaxMTTDepth( *tempCS->slice, partitioner.chType ) + partitioner.currImplicitBtDepth;
1188
1189
163k
  if( encTestmode.type == ETM_SPLIT_BT_H )
1190
22.7k
  {
1191
22.7k
    if( tempCS->cus.size() > 2 )
1192
901
    {
1193
901
      int h_2   = tempCS->area.blocks[partitioner.chType].height / 2;
1194
901
      int cu1_h = tempCS->cus.front()->blocks[partitioner.chType].height;
1195
901
      int cu2_h = tempCS->cus.back() ->blocks[partitioner.chType].height;
1196
1197
901
      cuECtx.doTriHorzSplit = cu1_h < h_2 || cu2_h < h_2 || partitioner.currMtDepth + 1 == maxMtD;
1198
901
    }
1199
22.7k
  }
1200
140k
  else if( encTestmode.type == ETM_SPLIT_BT_V )
1201
22.9k
  {
1202
22.9k
    if( tempCS->cus.size() > 2 )
1203
1.02k
    {
1204
1.02k
      int w_2   = tempCS->area.blocks[partitioner.chType].width / 2;
1205
1.02k
      int cu1_w = tempCS->cus.front()->blocks[partitioner.chType].width;
1206
1.02k
      int cu2_w = tempCS->cus.back() ->blocks[partitioner.chType].width;
1207
1208
1.02k
      cuECtx.doTriVertSplit = cu1_w < w_2 || cu2_w < w_2 || partitioner.currMtDepth + 1 == maxMtD;
1209
1.02k
    }
1210
22.9k
  }
1211
1212
163k
  if( encTestmode.type == ETM_SPLIT_BT_V || encTestmode.type == ETM_SPLIT_BT_H || encTestmode.type == ETM_SPLIT_QT )
1213
63.4k
  {
1214
63.4k
    bool isAllSkip = true;
1215
1216
63.4k
    for( const auto& cu : tempCS->cus )
1217
142k
    {
1218
142k
      isAllSkip &= cu->skip;
1219
142k
    }
1220
1221
63.4k
    if( isAllSkip ) cuECtx.doMoreSplits--;
1222
63.4k
    if( isAllSkip && encTestmode.type == ETM_SPLIT_QT )
1223
0
                    cuECtx.doMoreSplits--;
1224
63.4k
  }
1225
1226
  // for now just a simple decision based on RD-cost or choose tempCS if bestCS is not yet coded
1227
163k
  if( tempCS->cost != MAX_DOUBLE && ( !cuECtx.bestCS || ( ( tempCS->cost + ( useEDO ? tempCS->costDbOffset : 0 ) ) < ( cuECtx.bestCS->cost + ( useEDO ? cuECtx.bestCS->costDbOffset : 0 ) ) ) ) )
1228
147k
  {
1229
147k
    cuECtx.bestCS   = tempCS;
1230
147k
    cuECtx.bestCU   = tempCS->cus[0];
1231
147k
    cuECtx.bestTU   = cuECtx.bestCU->firstTU;
1232
147k
    cuECtx.bestMode = encTestmode;
1233
1234
147k
    return true;
1235
147k
  }
1236
15.6k
  else
1237
15.6k
  {
1238
15.6k
    return false;
1239
15.6k
  }
1240
163k
}
1241
1242
} // namespace vvenc
1243
1244
//! \}
1245