Context navigation

Back to Ticket #266

Ticket #266: EncSlice.cpp

File EncSlice.cpp, 82.1 KB (added by seuhong, 7 years ago)

Line
1	/* The copyright in this software is being made available under the BSD
2	* License, included below. This software may be subject to other third party
3	* and contributor rights, including patent rights, and no such rights are
4	* granted under this license.
5	*
6	* Copyright (c) 2010-2019, ITU/ISO/IEC
7	* All rights reserved.
8	*
9	* Redistribution and use in source and binary forms, with or without
10	* modification, are permitted provided that the following conditions are met:
11	*
12	* * Redistributions of source code must retain the above copyright notice,
13	* this list of conditions and the following disclaimer.
14	* * Redistributions in binary form must reproduce the above copyright notice,
15	* this list of conditions and the following disclaimer in the documentation
16	* and/or other materials provided with the distribution.
17	* * Neither the name of the ITU/ISO/IEC nor the names of its contributors may
18	* be used to endorse or promote products derived from this software without
19	* specific prior written permission.
20	*
21	* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
22	* AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23	* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24	* ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS
25	* BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
26	* CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
27	* SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
28	* INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
29	* CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
30	* ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
31	* THE POSSIBILITY OF SUCH DAMAGE.
32	*/
33
34	/** \file EncSlice.cpp
35	\brief slice encoder class
36	*/
37
38	#include "EncSlice.h"
39
40	#include "EncLib.h"
41	#include "CommonLib/UnitTools.h"
42	#include "CommonLib/Picture.h"
43	#if K0149_BLOCK_STATISTICS
44	#include "CommonLib/dtrace_blockstatistics.h"
45	#endif
46
47	#if ENABLE_WPP_PARALLELISM
48	#include <mutex>
49	extern recursive_mutex g_cache_mutex;
50	#endif
51
52	#include <math.h>
53
54	//! \ingroup EncoderLib
55	//! \{
56
57	// ====================================================================================================================
58	// Constructor / destructor / create / destroy
59	// ====================================================================================================================
60
61	EncSlice::EncSlice()
62	: m_encCABACTableIdx(I_SLICE)
63	#if ENABLE_QPA
64	, m_adaptedLumaQP(-1)
65	#endif
66	{
67	}
68
69	EncSlice::~EncSlice()
70	{
71	destroy();
72	}
73
74	void EncSlice::create( int iWidth, int iHeight, ChromaFormat chromaFormat, uint32_t iMaxCUWidth, uint32_t iMaxCUHeight, uint8_t uhTotalDepth )
75	{
76	}
77
78	void EncSlice::destroy()
79	{
80	// free lambda and QP arrays
81	m_vdRdPicLambda.clear();
82	m_vdRdPicQp.clear();
83	m_viRdPicQp.clear();
84	}
85
86	void EncSlice::init( EncLib* pcEncLib, const SPS& sps )
87	{
88	m_pcCfg = pcEncLib;
89	m_pcLib = pcEncLib;
90	m_pcListPic = pcEncLib->getListPic();
91
92	m_pcGOPEncoder = pcEncLib->getGOPEncoder();
93	m_pcCuEncoder = pcEncLib->getCuEncoder();
94	m_pcInterSearch = pcEncLib->getInterSearch();
95	m_CABACWriter = pcEncLib->getCABACEncoder()->getCABACWriter (&sps);
96	m_CABACEstimator = pcEncLib->getCABACEncoder()->getCABACEstimator(&sps);
97	m_pcTrQuant = pcEncLib->getTrQuant();
98	m_pcRdCost = pcEncLib->getRdCost();
99
100	// create lambda and QP arrays
101	m_vdRdPicLambda.resize(m_pcCfg->getDeltaQpRD() * 2 + 1 );
102	m_vdRdPicQp.resize( m_pcCfg->getDeltaQpRD() * 2 + 1 );
103	m_viRdPicQp.resize( m_pcCfg->getDeltaQpRD() * 2 + 1 );
104	m_pcRateCtrl = pcEncLib->getRateCtrl();
105	}
106
107	void
108	EncSlice::setUpLambda( Slice* slice, const double dLambda, int iQP)
109	{
110	// store lambda
111	m_pcRdCost ->setLambda( dLambda, slice->getSPS()->getBitDepths() );
112
113	// for RDO
114	// in RdCost there is only one lambda because the luma and chroma bits are not separated, instead we weight the distortion of chroma.
115	double dLambdas[MAX_NUM_COMPONENT] = { dLambda };
116	for( uint32_t compIdx = 1; compIdx < MAX_NUM_COMPONENT; compIdx++ )
117	{
118	const ComponentID compID = ComponentID( compIdx );
119	int chromaQPOffset = slice->getPPS()->getQpOffset( compID ) + slice->getSliceChromaQpDelta( compID );
120	int qpc = ( iQP + chromaQPOffset < 0 ) ? iQP : getScaledChromaQP( iQP + chromaQPOffset, m_pcCfg->getChromaFormatIdc() );
121	double tmpWeight = pow( 2.0, ( iQP - qpc ) / 3.0 ); // takes into account of the chroma qp mapping and chroma qp Offset
122	#if JVET_N0193_LFNST
123	if( m_pcCfg->getDepQuantEnabledFlag() && !( m_pcCfg->getLFNST() ) )
124	#else
125	if( m_pcCfg->getDepQuantEnabledFlag() )
126	#endif
127	{
128	tmpWeight *= ( m_pcCfg->getGOPSize() >= 8 ? pow( 2.0, 0.1/3.0 ) : pow( 2.0, 0.2/3.0 ) ); // increase chroma weight for dependent quantization (in order to reduce bit rate shift from chroma to luma)
129	}
130	m_pcRdCost->setDistortionWeight( compID, tmpWeight );
131	#if ENABLE_WPP_PARALLELISM
132	for( int jId = 1; jId < ( m_pcLib->getNumWppThreads() + m_pcLib->getNumWppExtraLines() ); jId++ )
133	{
134	m_pcLib->getRdCost( slice->getPic()->scheduler.getWppDataId( jId ) )->setDistortionWeight( compID, tmpWeight );
135	}
136	#endif
137	dLambdas[compIdx] = dLambda / tmpWeight;
138	}
139
140	#if RDOQ_CHROMA_LAMBDA
141	// for RDOQ
142	m_pcTrQuant->setLambdas( dLambdas );
143	#else
144	m_pcTrQuant->setLambda( dLambda );
145	#endif
146
147	// for SAO
148	slice->setLambdas( dLambdas );
149	}
150
151	#if ENABLE_QPA
152
153	static inline int apprI3Log2 (const double d) // rounded 3*log2(d)
154	{
155	return d < 1.5e-13 ? -128 : int (floor (3.0 * log (d) / log (2.0) + 0.5));
156	}
157
158	static inline int lumaDQPOffset (const uint32_t avgLumaValue, const int bitDepth)
159	{
160	return (1 - int ((3 * uint64_t (avgLumaValue * avgLumaValue)) >> uint64_t (2 * bitDepth - 1)));
161	}
162
163	static void filterAndCalculateAverageEnergies (const Pel* pSrc, const int iSrcStride,
164	double &hpEner, const int iHeight, const int iWidth,
165	const uint32_t uBitDepth /* luma bit-depth (4-16) */)
166	{
167	uint64_t saAct = 0;
168
169	// skip first row as there may be a black border frame
170	pSrc += iSrcStride;
171	// center rows
172	for (int y = 1; y < iHeight - 1; y++)
173	{
174	// skip column as there may be a black border frame
175
176	for (int x = 1; x < iWidth - 1; x++) // and columns
177	{
178	const int f = 12 * (int)pSrc[x ] - 2 * ((int)pSrc[x-1] + (int)pSrc[x+1] + (int)pSrc[x -iSrcStride] + (int)pSrc[x +iSrcStride])
179	- (int)pSrc[x-1-iSrcStride] - (int)pSrc[x+1-iSrcStride] - (int)pSrc[x-1+iSrcStride] - (int)pSrc[x+1+iSrcStride];
180	saAct += abs (f);
181	}
182	// skip column as there may be a black border frame
183	pSrc += iSrcStride;
184	}
185	// skip last row as there may be a black border frame
186
187	hpEner = double(saAct) / double((iWidth - 2) * (iHeight - 2));
188
189	// lower limit, compensate for highpass amplification
190	if (hpEner < double(1 << (uBitDepth - 4))) hpEner = double(1 << (uBitDepth - 4));
191	}
192
193	#ifndef GLOBAL_AVERAGING
194	#define GLOBAL_AVERAGING 1 // "global" averaging of a_k across a set instead of one picture
195	#endif
196
197	#if GLOBAL_AVERAGING
198	static double getAveragePictureEnergy (const CPelBuf picOrig, const uint32_t uBitDepth)
199	{
200	const double hpEnerPic = 16.0 * sqrt ((3840.0 * 2160.0) / double(picOrig.width * picOrig.height)) * double(1 << uBitDepth);
201
202	return sqrt (hpEnerPic); // square-root of a_pic value
203	}
204	#endif
205
206	static int getGlaringColorQPOffset (Picture* const pcPic, const int ctuAddr, const uint32_t startAddr, const uint32_t boundingAddr,
207	const int bitDepth, uint32_t &avgLumaValue)
208	{
209	const PreCalcValues& pcv = *pcPic->cs->pcv;
210	const ChromaFormat chrFmt = pcPic->chromaFormat;
211	const uint32_t chrWidth = pcv.maxCUWidth >> getChannelTypeScaleX (CH_C, chrFmt);
212	const uint32_t chrHeight = pcv.maxCUHeight >> getChannelTypeScaleY (CH_C, chrFmt);
213	const int midLevel = 1 << (bitDepth - 1);
214	int chrValue = MAX_INT;
215	avgLumaValue = (startAddr < boundingAddr) ? 0 : (uint32_t)pcPic->getOrigBuf().Y().computeAvg();
216
217	if (ctuAddr >= 0) // luma
218	{
219	avgLumaValue = (uint32_t)pcPic->m_iOffsetCtu[ctuAddr];
220	}
221	else if (startAddr < boundingAddr)
222	{
223	for (uint32_t ctuTsAddr = startAddr; ctuTsAddr < boundingAddr; ctuTsAddr++)
224	{
225	const uint32_t ctuRsAddr = pcPic->tileMap->getCtuTsToRsAddrMap (ctuTsAddr);
226
227	avgLumaValue += pcPic->m_iOffsetCtu[ctuRsAddr];
228	}
229	avgLumaValue = (avgLumaValue + ((boundingAddr - startAddr) >> 1)) / (boundingAddr - startAddr);
230	}
231
232	for (uint32_t comp = COMPONENT_Cb; comp < MAX_NUM_COMPONENT; comp++)
233	{
234	const ComponentID compID = (ComponentID)comp;
235	int avgCompValue;
236
237	if (ctuAddr >= 0) // chroma
238	{
239	const CompArea chrArea = clipArea (CompArea (compID, chrFmt, Area ((ctuAddr % pcv.widthInCtus) * chrWidth, (ctuAddr / pcv.widthInCtus) * chrHeight, chrWidth, chrHeight)), pcPic->block (compID));
240
241	avgCompValue = pcPic->getOrigBuf (chrArea).computeAvg();
242	}
243	else avgCompValue = pcPic->getOrigBuf (pcPic->block (compID)).computeAvg();
244
245	if (chrValue > avgCompValue) chrValue = avgCompValue; // minimum of the DC offsets
246	}
247	CHECK (chrValue < 0, "DC offset cannot be negative!");
248
249	chrValue = (int)avgLumaValue - chrValue;
250
251	if (chrValue > midLevel) return apprI3Log2 (double (chrValue * chrValue) / double (midLevel * midLevel));
252
253	return 0;
254	}
255
256	static int applyQPAdaptationChroma (Picture* const pcPic, Slice* const pcSlice, EncCfg* const pcEncCfg, const int sliceQP)
257	{
258	const int bitDepth = pcSlice->getSPS()->getBitDepth (CHANNEL_TYPE_LUMA); // overall image bit-depth
259	double hpEner[MAX_NUM_COMPONENT] = {0.0, 0.0, 0.0};
260	int optSliceChromaQpOffset[2] = {0, 0};
261	int savedLumaQP = -1;
262	uint32_t meanLuma = MAX_UINT;
263
264	for (uint32_t comp = 0; comp < getNumberValidComponents (pcPic->chromaFormat); comp++)
265	{
266	const ComponentID compID = (ComponentID)comp;
267	const CPelBuf picOrig = pcPic->getOrigBuf (pcPic->block (compID));
268
269	filterAndCalculateAverageEnergies (picOrig.buf, picOrig.stride, hpEner[comp],
270	picOrig.height, picOrig.width, bitDepth - (isChroma (compID) ? 1 : 0));
271	if (isChroma (compID))
272	{
273	const int adaptChromaQPOffset = 2.0 * hpEner[comp] <= hpEner[0] ? 0 : apprI3Log2 (2.0 * hpEner[comp] / hpEner[0]);
274
275	if (savedLumaQP < 0)
276	{
277	#if GLOBAL_AVERAGING
278	int averageAdaptedLumaQP = Clip3 (0, MAX_QP, sliceQP + apprI3Log2 (hpEner[0] / getAveragePictureEnergy (pcPic->getOrigBuf().Y(), bitDepth)));
279	#else
280	int averageAdaptedLumaQP = Clip3 (0, MAX_QP, sliceQP); // mean slice QP
281	#endif
282
283	averageAdaptedLumaQP += getGlaringColorQPOffset (pcPic, -1 /ctuRsAddr/, 0 /startAddr/, 0 /boundingAddr/, bitDepth, meanLuma);
284
285	if (averageAdaptedLumaQP > MAX_QP
286	#if SHARP_LUMA_DELTA_QP
287	&& (pcEncCfg->getLumaLevelToDeltaQPMapping().mode != LUMALVL_TO_DQP_NUM_MODES)
288	#endif
289	) averageAdaptedLumaQP = MAX_QP;
290	#if SHARP_LUMA_DELTA_QP
291
292	// change mean picture QP index based on picture's average luma value (Sharp)
293	if (pcEncCfg->getLumaLevelToDeltaQPMapping().mode == LUMALVL_TO_DQP_NUM_MODES)
294	{
295	if (meanLuma == MAX_UINT) meanLuma = pcPic->getOrigBuf().Y().computeAvg();
296
297	averageAdaptedLumaQP = Clip3 (0, MAX_QP, averageAdaptedLumaQP + lumaDQPOffset (meanLuma, bitDepth));
298	}
299	#endif
300
301	savedLumaQP = averageAdaptedLumaQP;
302	} // savedLumaQP < 0
303
304	const int lumaChromaMappingDQP = savedLumaQP - getScaledChromaQP (savedLumaQP, pcEncCfg->getChromaFormatIdc());
305
306	optSliceChromaQpOffset[comp-1] = std::min (3 + lumaChromaMappingDQP, adaptChromaQPOffset + lumaChromaMappingDQP);
307	}
308	}
309
310	pcEncCfg->setSliceChromaOffsetQpIntraOrPeriodic (pcEncCfg->getSliceChromaOffsetQpPeriodicity(), optSliceChromaQpOffset);
311
312	return savedLumaQP;
313	}
314
315	#endif // ENABLE_QPA
316
317	/**
318	- non-referenced frame marking
319	- QP computation based on temporal structure
320	- lambda computation based on QP
321	- set temporal layer ID and the parameter sets
322	.
323	\param pcPic picture class
324	\param pocLast POC of last picture
325	\param pocCurr current POC
326	\param iNumPicRcvd number of received pictures
327	\param iGOPid POC offset for hierarchical structure
328	\param rpcSlice slice header class
329	\param isField true for field coding
330	*/
331	void EncSlice::initEncSlice(Picture* pcPic, const int pocLast, const int pocCurr, const int iGOPid, Slice*& rpcSlice, const bool isField
332	, bool isEncodeLtRef
333	)
334	{
335	double dQP;
336	double dLambda;
337
338	rpcSlice = pcPic->slices[0];
339	rpcSlice->setSliceBits(0);
340	rpcSlice->setPic( pcPic );
341	rpcSlice->initSlice();
342	int multipleFactor = m_pcCfg->getUseCompositeRef() ? 2 : 1;
343	if (m_pcCfg->getUseCompositeRef() && isEncodeLtRef)
344	{
345	rpcSlice->setPicOutputFlag(false);
346	}
347	else
348	{
349	rpcSlice->setPicOutputFlag(true);
350	}
351	rpcSlice->setPOC( pocCurr );
352	rpcSlice->setDepQuantEnabledFlag( m_pcCfg->getDepQuantEnabledFlag() );
353	#if HEVC_USE_SIGN_HIDING
354	rpcSlice->setSignDataHidingEnabledFlag( m_pcCfg->getSignDataHidingEnabledFlag() );
355	#endif
356
357	#if SHARP_LUMA_DELTA_QP
358	pcPic->fieldPic = isField;
359	m_gopID = iGOPid;
360	#endif
361
362	// depth computation based on GOP size
363	int depth;
364	{
365	int poc = rpcSlice->getPOC();
366	if(isField)
367	{
368	poc = (poc/2) % (m_pcCfg->getGOPSize()/2);
369	}
370	else
371	{
372	poc = poc % (m_pcCfg->getGOPSize() * multipleFactor);
373	}
374
375	if ( poc == 0 )
376	{
377	depth = 0;
378	}
379	else
380	{
381	int step = m_pcCfg->getGOPSize() * multipleFactor;
382	depth = 0;
383	for( int i=step>>1; i>=1; i>>=1 )
384	{
385	for (int j = i; j<(m_pcCfg->getGOPSize() * multipleFactor); j += step)
386	{
387	if ( j == poc )
388	{
389	i=0;
390	break;
391	}
392	}
393	step >>= 1;
394	depth++;
395	}
396	}
397
398	if(m_pcCfg->getHarmonizeGopFirstFieldCoupleEnabled() && poc != 0)
399	{
400	if (isField && ((rpcSlice->getPOC() % 2) == 1))
401	{
402	depth++;
403	}
404	}
405	}
406
407	// slice type
408	SliceType eSliceType;
409
410	eSliceType=B_SLICE;
411	if(!(isField && pocLast == 1) \|\| !m_pcCfg->getEfficientFieldIRAPEnabled())
412	{
413	if(m_pcCfg->getDecodingRefreshType() == 3)
414	{
415	eSliceType = (pocLast == 0 \|\| pocCurr % (m_pcCfg->getIntraPeriod() * multipleFactor) == 0 \|\| m_pcGOPEncoder->getGOPSize() == 0) ? I_SLICE : eSliceType;
416	}
417	else
418	{
419	eSliceType = (pocLast == 0 \|\| (pocCurr - (isField ? 1 : 0)) % (m_pcCfg->getIntraPeriod() * multipleFactor) == 0 \|\| m_pcGOPEncoder->getGOPSize() == 0) ? I_SLICE : eSliceType;
420	}
421	}
422
423	rpcSlice->setSliceType ( eSliceType );
424
425	// ------------------------------------------------------------------------------------------------------------------
426	// Non-referenced frame marking
427	// ------------------------------------------------------------------------------------------------------------------
428
429	#if !JVET_M0101_HLS
430	if(pocLast == 0)
431	{
432	rpcSlice->setTemporalLayerNonReferenceFlag(false);
433	}
434	else
435	{
436	rpcSlice->setTemporalLayerNonReferenceFlag(!m_pcCfg->getGOPEntry(iGOPid).m_refPic);
437	}
438	#endif
439	pcPic->referenced = true;
440
441	// ------------------------------------------------------------------------------------------------------------------
442	// QP setting
443	// ------------------------------------------------------------------------------------------------------------------
444
445	#if X0038_LAMBDA_FROM_QP_CAPABILITY
446	dQP = m_pcCfg->getQPForPicture(iGOPid, rpcSlice);
447	#else
448	dQP = m_pcCfg->getBaseQP();
449	if(eSliceType!=I_SLICE)
450	{
451	#if SHARP_LUMA_DELTA_QP
452	if (!(( m_pcCfg->getMaxDeltaQP() == 0) && (!m_pcCfg->getLumaLevelToDeltaQPMapping().isEnabled()) && (dQP == -rpcSlice->getSPS()->getQpBDOffset(CHANNEL_TYPE_LUMA) ) && (rpcSlice->getPPS()->getTransquantBypassEnabledFlag())))
453	#else
454	if (!(( m_pcCfg->getMaxDeltaQP() == 0 ) && (dQP == -rpcSlice->getSPS()->getQpBDOffset(CHANNEL_TYPE_LUMA) ) && (rpcSlice->getPPS()->getTransquantBypassEnabledFlag())))
455	#endif
456	{
457	dQP += m_pcCfg->getGOPEntry(iGOPid).m_QPOffset;
458	}
459	}
460
461	// modify QP
462	const int* pdQPs = m_pcCfg->getdQPs();
463	if ( pdQPs )
464	{
465	dQP += pdQPs[ rpcSlice->getPOC() ];
466	}
467
468	if (m_pcCfg->getCostMode()==COST_LOSSLESS_CODING)
469	{
470	dQP=LOSSLESS_AND_MIXED_LOSSLESS_RD_COST_TEST_QP;
471	m_pcCfg->setDeltaQpRD(0);
472	}
473	#endif
474
475	// ------------------------------------------------------------------------------------------------------------------
476	// Lambda computation
477	// ------------------------------------------------------------------------------------------------------------------
478
479	#if X0038_LAMBDA_FROM_QP_CAPABILITY
480	const int temporalId=m_pcCfg->getGOPEntry(iGOPid).m_temporalId;
481	#if !SHARP_LUMA_DELTA_QP
482	const std::vector<double> &intraLambdaModifiers=m_pcCfg->getIntraLambdaModifier();
483	#endif
484	#endif
485	int iQP;
486	double dOrigQP = dQP;
487
488	// pre-compute lambda and QP values for all possible QP candidates
489	for ( int iDQpIdx = 0; iDQpIdx < 2 * m_pcCfg->getDeltaQpRD() + 1; iDQpIdx++ )
490	{
491	// compute QP value
492	dQP = dOrigQP + ((iDQpIdx+1)>>1)*(iDQpIdx%2 ? -1 : 1);
493	#if SHARP_LUMA_DELTA_QP
494	dLambda = calculateLambda(rpcSlice, iGOPid, depth, dQP, dQP, iQP );
495	#else
496	// compute lambda value
497	int NumberBFrames = ( m_pcCfg->getGOPSize() - 1 );
498	int SHIFT_QP = 12;
499
500	int bitdepth_luma_qp_scale =
501	6
502	* (rpcSlice->getSPS()->getBitDepth(CHANNEL_TYPE_LUMA) - 8
503	- DISTORTION_PRECISION_ADJUSTMENT(rpcSlice->getSPS()->getBitDepth(CHANNEL_TYPE_LUMA)));
504	double qp_temp = (double) dQP + bitdepth_luma_qp_scale - SHIFT_QP;
505	#if FULL_NBIT
506	double qp_temp_orig = (double) dQP - SHIFT_QP;
507	#endif
508	// Case #1: I or P-slices (key-frame)
509	double dQPFactor = m_pcCfg->getGOPEntry(iGOPid).m_QPFactor;
510	if ( eSliceType==I_SLICE )
511	{
512	if (m_pcCfg->getIntraQpFactor()>=0.0 && m_pcCfg->getGOPEntry(iGOPid).m_sliceType != I_SLICE)
513	{
514	dQPFactor=m_pcCfg->getIntraQpFactor();
515	}
516	else
517	{
518	#if X0038_LAMBDA_FROM_QP_CAPABILITY
519	if(m_pcCfg->getLambdaFromQPEnable())
520	{
521	dQPFactor=0.57;
522	}
523	else
524	{
525	#endif
526	double dLambda_scale = 1.0 - Clip3( 0.0, 0.5, 0.05*(double)(isField ? NumberBFrames/2 : NumberBFrames) );
527
528	dQPFactor=0.57*dLambda_scale;
529	#if X0038_LAMBDA_FROM_QP_CAPABILITY
530	}
531	#endif
532	}
533	}
534	#if X0038_LAMBDA_FROM_QP_CAPABILITY
535	else if( m_pcCfg->getLambdaFromQPEnable() )
536	{
537	dQPFactor=0.57;
538	}
539	#endif
540
541	dLambda = dQPFactor*pow( 2.0, qp_temp/3.0 );
542
543	#if X0038_LAMBDA_FROM_QP_CAPABILITY
544	if(!m_pcCfg->getLambdaFromQPEnable() && depth>0)
545	#else
546	if ( depth>0 )
547	#endif
548	{
549	#if FULL_NBIT
550	dLambda *= Clip3( 2.00, 4.00, (qp_temp_orig / 6.0) ); // (j == B_SLICE && p_cur_frm->layer != 0 )
551	#else
552	dLambda *= Clip3( 2.00, 4.00, (qp_temp / 6.0) ); // (j == B_SLICE && p_cur_frm->layer != 0 )
553	#endif
554	}
555
556	// if hadamard is used in ME process
557	if ( !m_pcCfg->getUseHADME() && rpcSlice->getSliceType( ) != I_SLICE )
558	{
559	dLambda *= 0.95;
560	}
561
562	#if X0038_LAMBDA_FROM_QP_CAPABILITY
563	double lambdaModifier;
564	if( rpcSlice->getSliceType( ) != I_SLICE \|\| intraLambdaModifiers.empty())
565	{
566	lambdaModifier = m_pcCfg->getLambdaModifier( temporalId );
567	}
568	else
569	{
570	lambdaModifier = intraLambdaModifiers[ (temporalId < intraLambdaModifiers.size()) ? temporalId : (intraLambdaModifiers.size()-1) ];
571	}
572	dLambda *= lambdaModifier;
573	#endif
574
575	iQP = Clip3( -rpcSlice->getSPS()->getQpBDOffset( CHANNEL_TYPE_LUMA ), MAX_QP, (int) floor( dQP + 0.5 ) );
576	#endif
577
578	m_vdRdPicLambda[iDQpIdx] = dLambda;
579	m_vdRdPicQp [iDQpIdx] = dQP;
580	m_viRdPicQp [iDQpIdx] = iQP;
581	}
582
583	// obtain dQP = 0 case
584	dLambda = m_vdRdPicLambda[0];
585	dQP = m_vdRdPicQp [0];
586	iQP = m_viRdPicQp [0];
587
588	#if !X0038_LAMBDA_FROM_QP_CAPABILITY
589	const int temporalId=m_pcCfg->getGOPEntry(iGOPid).m_temporalId;
590	const std::vector<double> &intraLambdaModifiers=m_pcCfg->getIntraLambdaModifier();
591	#endif
592
593	#if W0038_CQP_ADJ
594	#if ENABLE_QPA
595	m_adaptedLumaQP = -1;
596
597	if ((m_pcCfg->getUsePerceptQPA() \|\| m_pcCfg->getSliceChromaOffsetQpPeriodicity() > 0) && !m_pcCfg->getUseRateCtrl() && rpcSlice->getPPS()->getSliceChromaQpFlag() &&
598	(rpcSlice->isIntra() \|\| (m_pcCfg->getSliceChromaOffsetQpPeriodicity() > 0 && (rpcSlice->getPOC() % m_pcCfg->getSliceChromaOffsetQpPeriodicity()) == 0)))
599	{
600	m_adaptedLumaQP = applyQPAdaptationChroma (pcPic, rpcSlice, m_pcCfg, iQP);
601	}
602	#endif
603	if(rpcSlice->getPPS()->getSliceChromaQpFlag())
604	{
605	const bool bUseIntraOrPeriodicOffset = (rpcSlice->isIntra() && !rpcSlice->getSPS()->getIBCFlag()) \|\| (m_pcCfg->getSliceChromaOffsetQpPeriodicity() > 0 && (rpcSlice->getPOC() % m_pcCfg->getSliceChromaOffsetQpPeriodicity()) == 0);
606	int cbQP = bUseIntraOrPeriodicOffset ? m_pcCfg->getSliceChromaOffsetQpIntraOrPeriodic(false) : m_pcCfg->getGOPEntry(iGOPid).m_CbQPoffset;
607	int crQP = bUseIntraOrPeriodicOffset ? m_pcCfg->getSliceChromaOffsetQpIntraOrPeriodic(true) : m_pcCfg->getGOPEntry(iGOPid).m_CrQPoffset;
608
609	cbQP = Clip3( -12, 12, cbQP + rpcSlice->getPPS()->getQpOffset(COMPONENT_Cb) ) - rpcSlice->getPPS()->getQpOffset(COMPONENT_Cb);
610	crQP = Clip3( -12, 12, crQP + rpcSlice->getPPS()->getQpOffset(COMPONENT_Cr) ) - rpcSlice->getPPS()->getQpOffset(COMPONENT_Cr);
611	rpcSlice->setSliceChromaQpDelta(COMPONENT_Cb, Clip3( -12, 12, cbQP));
612	CHECK(!(rpcSlice->getSliceChromaQpDelta(COMPONENT_Cb)+rpcSlice->getPPS()->getQpOffset(COMPONENT_Cb)<=12 && rpcSlice->getSliceChromaQpDelta(COMPONENT_Cb)+rpcSlice->getPPS()->getQpOffset(COMPONENT_Cb)>=-12), "Unspecified error");
613	rpcSlice->setSliceChromaQpDelta(COMPONENT_Cr, Clip3( -12, 12, crQP));
614	CHECK(!(rpcSlice->getSliceChromaQpDelta(COMPONENT_Cr)+rpcSlice->getPPS()->getQpOffset(COMPONENT_Cr)<=12 && rpcSlice->getSliceChromaQpDelta(COMPONENT_Cr)+rpcSlice->getPPS()->getQpOffset(COMPONENT_Cr)>=-12), "Unspecified error");
615	}
616	else
617	{
618	rpcSlice->setSliceChromaQpDelta( COMPONENT_Cb, 0 );
619	rpcSlice->setSliceChromaQpDelta( COMPONENT_Cr, 0 );
620	#if JVET_N0054_JOINT_CHROMA
621	rpcSlice->setSliceChromaQpDelta( JOINT_CbCr, 0 );
622	#endif
623	}
624	#endif
625
626	#if !X0038_LAMBDA_FROM_QP_CAPABILITY
627	double lambdaModifier;
628	if( rpcSlice->getSliceType( ) != I_SLICE \|\| intraLambdaModifiers.empty())
629	{
630	lambdaModifier = m_pcCfg->getLambdaModifier( temporalId );
631	}
632	else
633	{
634	lambdaModifier = intraLambdaModifiers[ (temporalId < intraLambdaModifiers.size()) ? temporalId : (intraLambdaModifiers.size()-1) ];
635	}
636
637	dLambda *= lambdaModifier;
638	#endif
639
640	setUpLambda(rpcSlice, dLambda, iQP);
641
642	#if WCG_EXT
643	// cost = Distortion + Lambda*R,
644	// when QP is adjusted by luma, distortion is changed, so we have to adjust lambda to match the distortion, then the cost function becomes
645	// costA = Distortion + AdjustedLambda * R -- currently, costA is still used when calculating intermediate cost of using SAD, HAD, resisual etc.
646	// an alternative way is to weight the distortion to before the luma QP adjustment, then the cost function becomes
647	// costB = weightedDistortion + Lambda * R -- currently, costB is used to calculat final cost, and when DF_FUNC is DF_DEFAULT
648	m_pcRdCost->saveUnadjustedLambda();
649	#endif
650
651	if (m_pcCfg->getFastMEForGenBLowDelayEnabled())
652	{
653	// restore original slice type
654
655	if(!(isField && pocLast == 1) \|\| !m_pcCfg->getEfficientFieldIRAPEnabled())
656	{
657	if(m_pcCfg->getDecodingRefreshType() == 3)
658	{
659	eSliceType = (pocLast == 0 \|\| (pocCurr) % (m_pcCfg->getIntraPeriod() * multipleFactor) == 0 \|\| m_pcGOPEncoder->getGOPSize() == 0) ? I_SLICE : eSliceType;
660	}
661	else
662	{
663	eSliceType = (pocLast == 0 \|\| (pocCurr - (isField ? 1 : 0)) % (m_pcCfg->getIntraPeriod() * multipleFactor) == 0 \|\| m_pcGOPEncoder->getGOPSize() == 0) ? I_SLICE : eSliceType;
664	}
665	}
666
667	rpcSlice->setSliceType ( eSliceType );
668	}
669
670	if (m_pcCfg->getUseRecalculateQPAccordingToLambda())
671	{
672	dQP = xGetQPValueAccordingToLambda( dLambda );
673	iQP = Clip3( -rpcSlice->getSPS()->getQpBDOffset( CHANNEL_TYPE_LUMA ), MAX_QP, (int) floor( dQP + 0.5 ) );
674	}
675
676	rpcSlice->setSliceQp ( iQP );
677	rpcSlice->setSliceQpDelta ( 0 );
678	#if !W0038_CQP_ADJ
679	rpcSlice->setSliceChromaQpDelta( COMPONENT_Cb, 0 );
680	rpcSlice->setSliceChromaQpDelta( COMPONENT_Cr, 0 );
681	#if JVET_N0054_JOINT_CHROMA
682	rpcSlice->setSliceChromaQpDelta( JOINT_CbCr, 0 );
683	#endif
684	#endif
685	rpcSlice->setUseChromaQpAdj( rpcSlice->getPPS()->getPpsRangeExtension().getChromaQpOffsetListEnabledFlag() );
686	rpcSlice->setNumRefIdx(REF_PIC_LIST_0,m_pcCfg->getGOPEntry(iGOPid).m_numRefPicsActive);
687	rpcSlice->setNumRefIdx(REF_PIC_LIST_1,m_pcCfg->getGOPEntry(iGOPid).m_numRefPicsActive);
688
689	if ( m_pcCfg->getDeblockingFilterMetric() )
690	{
691	rpcSlice->setDeblockingFilterOverrideFlag(true);
692	rpcSlice->setDeblockingFilterDisable(false);
693	rpcSlice->setDeblockingFilterBetaOffsetDiv2( 0 );
694	rpcSlice->setDeblockingFilterTcOffsetDiv2( 0 );
695	}
696	else if (rpcSlice->getPPS()->getDeblockingFilterControlPresentFlag())
697	{
698	rpcSlice->setDeblockingFilterOverrideFlag( rpcSlice->getPPS()->getDeblockingFilterOverrideEnabledFlag() );
699	rpcSlice->setDeblockingFilterDisable( rpcSlice->getPPS()->getPPSDeblockingFilterDisabledFlag() );
700	if ( !rpcSlice->getDeblockingFilterDisable())
701	{
702	if ( rpcSlice->getDeblockingFilterOverrideFlag() && eSliceType!=I_SLICE)
703	{
704	rpcSlice->setDeblockingFilterBetaOffsetDiv2( m_pcCfg->getGOPEntry(iGOPid).m_betaOffsetDiv2 + m_pcCfg->getLoopFilterBetaOffset() );
705	rpcSlice->setDeblockingFilterTcOffsetDiv2( m_pcCfg->getGOPEntry(iGOPid).m_tcOffsetDiv2 + m_pcCfg->getLoopFilterTcOffset() );
706	}
707	else
708	{
709	rpcSlice->setDeblockingFilterBetaOffsetDiv2( m_pcCfg->getLoopFilterBetaOffset() );
710	rpcSlice->setDeblockingFilterTcOffsetDiv2( m_pcCfg->getLoopFilterTcOffset() );
711	}
712	}
713	}
714	else
715	{
716	rpcSlice->setDeblockingFilterOverrideFlag( false );
717	rpcSlice->setDeblockingFilterDisable( false );
718	rpcSlice->setDeblockingFilterBetaOffsetDiv2( 0 );
719	rpcSlice->setDeblockingFilterTcOffsetDiv2( 0 );
720	}
721
722	rpcSlice->setDepth ( depth );
723
724	pcPic->layer = temporalId;
725	if(eSliceType==I_SLICE)
726	{
727	pcPic->layer = 0;
728	}
729	rpcSlice->setTLayer( pcPic->layer );
730
731	rpcSlice->setSliceMode ( m_pcCfg->getSliceMode() );
732	rpcSlice->setSliceArgument ( m_pcCfg->getSliceArgument() );
733	#if HEVC_DEPENDENT_SLICES
734	rpcSlice->setSliceSegmentMode ( m_pcCfg->getSliceSegmentMode() );
735	rpcSlice->setSliceSegmentArgument ( m_pcCfg->getSliceSegmentArgument() );
736	#endif
737	rpcSlice->setMaxNumMergeCand ( m_pcCfg->getMaxNumMergeCand() );
738	rpcSlice->setMaxNumAffineMergeCand( m_pcCfg->getMaxNumAffineMergeCand() );
739	#if JVET_N0400_SIGNAL_TRIANGLE_CAND_NUM
740	rpcSlice->setMaxNumTriangleCand ( m_pcCfg->getMaxNumTriangleCand() );
741	#endif
742	rpcSlice->setSplitConsOverrideFlag(false);
743	rpcSlice->setMinQTSize( rpcSlice->getSPS()->getMinQTSize(eSliceType));
744	rpcSlice->setMaxBTDepth( rpcSlice->isIntra() ? rpcSlice->getSPS()->getMaxBTDepthI() : rpcSlice->getSPS()->getMaxBTDepth() );
745	rpcSlice->setMaxBTSize( rpcSlice->isIntra() ? rpcSlice->getSPS()->getMaxBTSizeI() : rpcSlice->getSPS()->getMaxBTSize() );
746	rpcSlice->setMaxTTSize( rpcSlice->isIntra() ? rpcSlice->getSPS()->getMaxTTSizeI() : rpcSlice->getSPS()->getMaxTTSize() );
747	if ( eSliceType == I_SLICE && rpcSlice->getSPS()->getUseDualITree() )
748	{
749	rpcSlice->setMinQTSizeIChroma( rpcSlice->getSPS()->getMinQTSize(eSliceType, CHANNEL_TYPE_CHROMA) );
750	rpcSlice->setMaxBTDepthIChroma( rpcSlice->getSPS()->getMaxBTDepthIChroma() );
751	rpcSlice->setMaxBTSizeIChroma( rpcSlice->getSPS()->getMaxBTSizeIChroma() );
752	rpcSlice->setMaxTTSizeIChroma( rpcSlice->getSPS()->getMaxTTSizeIChroma() );
753	}
754	#if JVET_N0329_IBC_SEARCH_IMP
755	rpcSlice->setDisableSATDForRD(false);
756	#endif
757	}
758
759
760	#if SHARP_LUMA_DELTA_QP
761	double EncSlice::calculateLambda( const Slice* slice,
762	const int GOPid, // entry in the GOP table
763	const int depth, // slice GOP hierarchical depth.
764	const double refQP, // initial slice-level QP
765	const double dQP, // initial double-precision QP
766	int &iQP ) // returned integer QP.
767	{
768	enum SliceType eSliceType = slice->getSliceType();
769	const bool isField = slice->getPic()->fieldPic;
770	const int NumberBFrames = ( m_pcCfg->getGOPSize() - 1 );
771	const int SHIFT_QP = 12;
772	#if X0038_LAMBDA_FROM_QP_CAPABILITY
773	const int temporalId=m_pcCfg->getGOPEntry(GOPid).m_temporalId;
774	const std::vector<double> &intraLambdaModifiers=m_pcCfg->getIntraLambdaModifier();
775	#endif
776
777	int bitdepth_luma_qp_scale = 6
778	* (slice->getSPS()->getBitDepth(CHANNEL_TYPE_LUMA) - 8
779	- DISTORTION_PRECISION_ADJUSTMENT(slice->getSPS()->getBitDepth(CHANNEL_TYPE_LUMA)));
780	double qp_temp = dQP + bitdepth_luma_qp_scale - SHIFT_QP;
781	// Case #1: I or P-slices (key-frame)
782	double dQPFactor = m_pcCfg->getGOPEntry(GOPid).m_QPFactor;
783	if ( eSliceType==I_SLICE )
784	{
785	if (m_pcCfg->getIntraQpFactor()>=0.0 && m_pcCfg->getGOPEntry(GOPid).m_sliceType != I_SLICE)
786	{
787	dQPFactor=m_pcCfg->getIntraQpFactor();
788	}
789	else
790	{
791	#if X0038_LAMBDA_FROM_QP_CAPABILITY
792	if(m_pcCfg->getLambdaFromQPEnable())
793	{
794	dQPFactor=0.57;
795	}
796	else
797	{
798	#endif
799	double dLambda_scale = 1.0 - Clip3( 0.0, 0.5, 0.05*(double)(isField ? NumberBFrames/2 : NumberBFrames) );
800	dQPFactor=0.57*dLambda_scale;
801	#if X0038_LAMBDA_FROM_QP_CAPABILITY
802	}
803	#endif
804	}
805	}
806	#if X0038_LAMBDA_FROM_QP_CAPABILITY
807	else if( m_pcCfg->getLambdaFromQPEnable() )
808	{
809	dQPFactor=0.57;
810	}
811	#endif
812
813	double dLambda = dQPFactor*pow( 2.0, qp_temp/3.0 );
814
815	#if X0038_LAMBDA_FROM_QP_CAPABILITY
816	if( !(m_pcCfg->getLambdaFromQPEnable()) && depth>0 )
817	#else
818	if ( depth>0 )
819	#endif
820	{
821	double qp_temp_ref = refQP + bitdepth_luma_qp_scale - SHIFT_QP;
822	dLambda *= Clip3(2.00, 4.00, (qp_temp_ref / 6.0)); // (j == B_SLICE && p_cur_frm->layer != 0 )
823	}
824
825	// if hadamard is used in ME process
826	if ( !m_pcCfg->getUseHADME() && slice->getSliceType( ) != I_SLICE )
827	{
828	dLambda *= 0.95;
829	}
830
831	#if X0038_LAMBDA_FROM_QP_CAPABILITY
832	double lambdaModifier;
833	if( eSliceType != I_SLICE \|\| intraLambdaModifiers.empty())
834	{
835	lambdaModifier = m_pcCfg->getLambdaModifier( temporalId );
836	}
837	else
838	{
839	lambdaModifier = intraLambdaModifiers[ (temporalId < intraLambdaModifiers.size()) ? temporalId : (intraLambdaModifiers.size()-1) ];
840	}
841	dLambda *= lambdaModifier;
842	#endif
843
844	iQP = Clip3( -slice->getSPS()->getQpBDOffset( CHANNEL_TYPE_LUMA ), MAX_QP, (int) floor( dQP + 0.5 ) );
845
846	if( m_pcCfg->getDepQuantEnabledFlag() )
847	{
848	dLambda *= pow( 2.0, 0.25/3.0 ); // slight lambda adjustment for dependent quantization (due to different slope of quantizer)
849	}
850
851	// NOTE: the lambda modifiers that are sometimes applied later might be best always applied in here.
852	return dLambda;
853	}
854	#endif
855
856	void EncSlice::resetQP( Picture* pic, int sliceQP, double lambda )
857	{
858	Slice* slice = pic->slices[0];
859
860	// store lambda
861	slice->setSliceQp( sliceQP );
862	setUpLambda(slice, lambda, sliceQP);
863	}
864
865	#if ENABLE_QPA
866	static bool applyQPAdaptation (Picture* const pcPic, Slice* const pcSlice, const PreCalcValues& pcv,
867	const uint32_t startAddr, const uint32_t boundingAddr, const bool useSharpLumaDQP,
868	const bool useFrameWiseQPA, const int previouslyAdaptedLumaQP = -1)
869	{
870	const int bitDepth = pcSlice->getSPS()->getBitDepth (CHANNEL_TYPE_LUMA);
871	const int iQPIndex = pcSlice->getSliceQp(); // initial QP index for current slice, used in following loops
872	const TileMap& tileMap = *pcPic->tileMap;
873	bool sliceQPModified = false;
874	uint32_t meanLuma = MAX_UINT;
875	double hpEnerAvg = 0.0;
876
877	#if GLOBAL_AVERAGING
878	if (!useFrameWiseQPA \|\| previouslyAdaptedLumaQP < 0) // mean visual activity value and luma value in each CTU
879	#endif
880	{
881	for (uint32_t ctuTsAddr = startAddr; ctuTsAddr < boundingAddr; ctuTsAddr++)
882	{
883	const uint32_t ctuRsAddr = tileMap.getCtuTsToRsAddrMap (ctuTsAddr);
884	const Position pos ((ctuRsAddr % pcv.widthInCtus) * pcv.maxCUWidth, (ctuRsAddr / pcv.widthInCtus) * pcv.maxCUHeight);
885	const CompArea ctuArea = clipArea (CompArea (COMPONENT_Y, pcPic->chromaFormat, Area (pos.x, pos.y, pcv.maxCUWidth, pcv.maxCUHeight)), pcPic->Y());
886	const CompArea fltArea = clipArea (CompArea (COMPONENT_Y, pcPic->chromaFormat, Area (pos.x > 0 ? pos.x - 1 : 0, pos.y > 0 ? pos.y - 1 : 0, pcv.maxCUWidth + (pos.x > 0 ? 2 : 1), pcv.maxCUHeight + (pos.y > 0 ? 2 : 1))), pcPic->Y());
887	const CPelBuf picOrig = pcPic->getOrigBuf (fltArea);
888	double hpEner = 0.0;
889
890	filterAndCalculateAverageEnergies (picOrig.buf, picOrig.stride, hpEner,
891	picOrig.height, picOrig.width, bitDepth);
892	hpEnerAvg += hpEner;
893	pcPic->m_uEnerHpCtu[ctuRsAddr] = hpEner;
894	pcPic->m_iOffsetCtu[ctuRsAddr] = pcPic->getOrigBuf (ctuArea).computeAvg();
895	}
896
897	hpEnerAvg /= double (boundingAddr - startAddr);
898	}
899	#if GLOBAL_AVERAGING
900	const double hpEnerPic = 1.0 / getAveragePictureEnergy (pcPic->getOrigBuf().Y(), bitDepth); // inverse, speed
901	#else
902	const double hpEnerPic = 1.0 / hpEnerAvg; // speedup: multiply instead of divide in loop below; 1.0 for tuning
903	#endif
904
905	if (useFrameWiseQPA \|\| (iQPIndex >= MAX_QP))
906	{
907	int iQPFixed = (previouslyAdaptedLumaQP < 0) ? Clip3 (0, MAX_QP, iQPIndex + apprI3Log2 (hpEnerAvg * hpEnerPic)) : previouslyAdaptedLumaQP;
908
909	if (isChromaEnabled (pcPic->chromaFormat) && (iQPIndex < MAX_QP) && (previouslyAdaptedLumaQP < 0))
910	{
911	iQPFixed += getGlaringColorQPOffset (pcPic, -1 /ctuRsAddr/, startAddr, boundingAddr, bitDepth, meanLuma);
912
913	if (iQPFixed > MAX_QP
914	#if SHARP_LUMA_DELTA_QP
915	&& !useSharpLumaDQP
916	#endif
917	) iQPFixed = MAX_QP;
918	}
919	#if SHARP_LUMA_DELTA_QP
920
921	// change new fixed QP based on average CTU luma value (Sharp)
922	if (useSharpLumaDQP && (iQPIndex < MAX_QP) && (previouslyAdaptedLumaQP < 0))
923	{
924	if (meanLuma == MAX_UINT) // collect picture mean luma value
925	{
926	meanLuma = 0;
927
928	for (uint32_t ctuTsAddr = startAddr; ctuTsAddr < boundingAddr; ctuTsAddr++)
929	{
930	const uint32_t ctuRsAddr = tileMap.getCtuTsToRsAddrMap (ctuTsAddr);
931
932	meanLuma += pcPic->m_iOffsetCtu[ctuRsAddr]; // CTU mean
933	}
934	meanLuma = (meanLuma + ((boundingAddr - startAddr) >> 1)) / (boundingAddr - startAddr);
935	}
936	iQPFixed = Clip3 (0, MAX_QP, iQPFixed + lumaDQPOffset (meanLuma, bitDepth));
937	}
938	#endif
939
940	if (iQPIndex >= MAX_QP) iQPFixed = MAX_QP;
941	else
942	if (iQPFixed != iQPIndex)
943	{
944	const double* oldLambdas = pcSlice->getLambdas();
945	const double corrFactor = pow (2.0, double(iQPFixed - iQPIndex) / 3.0);
946	const double newLambdas[MAX_NUM_COMPONENT] = {oldLambdas[0] * corrFactor, oldLambdas[1] * corrFactor, oldLambdas[2] * corrFactor};
947
948	CHECK (iQPIndex != pcSlice->getSliceQpBase(), "Invalid slice QP!");
949	pcSlice->setLambdas (newLambdas);
950	pcSlice->setSliceQp (iQPFixed); // update the slice/base QPs
951	pcSlice->setSliceQpBase (iQPFixed);
952
953	sliceQPModified = true;
954	}
955
956	for (uint32_t ctuTsAddr = startAddr; ctuTsAddr < boundingAddr; ctuTsAddr++)
957	{
958	const uint32_t ctuRsAddr = tileMap.getCtuTsToRsAddrMap (ctuTsAddr);
959
960	pcPic->m_iOffsetCtu[ctuRsAddr] = (Pel)iQPFixed; // fixed QPs
961	}
962	}
963	else // CTU-wise QPA
964	{
965	for (uint32_t ctuTsAddr = startAddr; ctuTsAddr < boundingAddr; ctuTsAddr++)
966	{
967	const uint32_t ctuRsAddr = tileMap.getCtuTsToRsAddrMap (ctuTsAddr);
968
969	int iQPAdapt = Clip3 (0, MAX_QP, iQPIndex + apprI3Log2 (pcPic->m_uEnerHpCtu[ctuRsAddr] * hpEnerPic));
970
971	if (pcv.widthInCtus > 1) // try to enforce CTU SNR greater than zero dB
972	{
973	meanLuma = (uint32_t)pcPic->m_iOffsetCtu[ctuRsAddr];
974
975	if (isChromaEnabled (pcPic->chromaFormat))
976	{
977	iQPAdapt += getGlaringColorQPOffset (pcPic, (int)ctuRsAddr, startAddr, boundingAddr, bitDepth, meanLuma);
978
979	if (iQPAdapt > MAX_QP
980	#if SHARP_LUMA_DELTA_QP
981	&& !useSharpLumaDQP
982	#endif
983	) iQPAdapt = MAX_QP;
984	CHECK (meanLuma != (uint32_t)pcPic->m_iOffsetCtu[ctuRsAddr], "luma DC offsets don't match");
985	}
986	#if SHARP_LUMA_DELTA_QP
987
988	// change adaptive QP based on mean CTU luma value (Sharp)
989	if (useSharpLumaDQP)
990	{
991	#if ENABLE_QPA_SUB_CTU
992	pcPic->m_uEnerHpCtu[ctuRsAddr] = (double)meanLuma; // for sub-CTU QPA
993	#endif
994	iQPAdapt = Clip3 (0, MAX_QP, iQPAdapt + lumaDQPOffset (meanLuma, bitDepth));
995	}
996
997	#endif
998	#if JVET_N0246_MODIFIED_QUANTSCALES
999	const uint32_t uRefScale = g_invQuantScales[0][iQPAdapt % 6] << ((iQPAdapt / 6) + bitDepth - 4);
1000	#else
1001	const uint32_t uRefScale = g_invQuantScales[iQPAdapt % 6] << ((iQPAdapt / 6) + bitDepth - 4);
1002	#endif
1003	const CompArea subArea = clipArea (CompArea (COMPONENT_Y, pcPic->chromaFormat, Area ((ctuRsAddr % pcv.widthInCtus) * pcv.maxCUWidth, (ctuRsAddr / pcv.widthInCtus) * pcv.maxCUHeight, pcv.maxCUWidth, pcv.maxCUHeight)), pcPic->Y());
1004	const Pel* pSrc = pcPic->getOrigBuf (subArea).buf;
1005	const SizeType iSrcStride = pcPic->getOrigBuf (subArea).stride;
1006	const SizeType iSrcHeight = pcPic->getOrigBuf (subArea).height;
1007	const SizeType iSrcWidth = pcPic->getOrigBuf (subArea).width;
1008	uint32_t uAbsDCless = 0;
1009
1010	// compute sum of absolute DC-less (high-pass) luma values
1011	for (SizeType h = 0; h < iSrcHeight; h++)
1012	{
1013	for (SizeType w = 0; w < iSrcWidth; w++)
1014	{
1015	uAbsDCless += (uint32_t)abs (pSrc[w] - (Pel)meanLuma);
1016	}
1017	pSrc += iSrcStride;
1018	}
1019
1020	if (iSrcHeight >= 64 \|\| iSrcWidth >= 64) // normalization
1021	{
1022	const uint64_t blockSize = uint64_t(iSrcWidth * iSrcHeight);
1023
1024	uAbsDCless = uint32_t((uint64_t(uAbsDCless) * 64*64 + (blockSize >> 1)) / blockSize);
1025	}
1026
1027	if (uAbsDCless < 6464) uAbsDCless = 6464; // limit to 1
1028
1029	// reduce QP index if CTU would be fully quantized to zero
1030	if (uAbsDCless < uRefScale)
1031	{
1032	const int limit = std::min (0, ((iQPIndex + 4) >> 3) - 6);
1033	const int redVal = std::max (limit, apprI3Log2 ((double)uAbsDCless / (double)uRefScale));
1034
1035	iQPAdapt = std::max (0, iQPAdapt + redVal);
1036	}
1037	}
1038
1039	pcPic->m_iOffsetCtu[ctuRsAddr] = (Pel)iQPAdapt; // adapted QPs
1040
1041	#if ENABLE_QPA_SUB_CTU
1042	if (pcv.widthInCtus > 1 && pcSlice->getPPS()->getCuQpDeltaSubdiv() == 0) // reduce local DQP rate peaks
1043	#elif ENABLE_QPA_SUB_CTU
1044	if (pcv.widthInCtus > 1 && pcSlice->getPPS()->getMaxCuDQPDepth() == 0) // reduce local DQP rate peaks
1045	#else
1046	if (pcv.widthInCtus > 1) // try to reduce local bitrate peaks via minimum smoothing of the adapted QPs
1047	#endif
1048	{
1049	iQPAdapt = ctuRsAddr % pcv.widthInCtus; // horizontal offset
1050	if (iQPAdapt == 0)
1051	{
1052	iQPAdapt = (ctuRsAddr > 1) ? pcPic->m_iOffsetCtu[ctuRsAddr - 2] : 0;
1053	}
1054	else // iQPAdapt >= 1
1055	{
1056	iQPAdapt = (iQPAdapt > 1) ? std::min (pcPic->m_iOffsetCtu[ctuRsAddr - 2], pcPic->m_iOffsetCtu[ctuRsAddr]) : pcPic->m_iOffsetCtu[ctuRsAddr];
1057	}
1058	if (ctuRsAddr > pcv.widthInCtus)
1059	{
1060	iQPAdapt = std::min (iQPAdapt, (int)pcPic->m_iOffsetCtu[ctuRsAddr - 1 - pcv.widthInCtus]);
1061	}
1062	if ((ctuRsAddr > 0) && (pcPic->m_iOffsetCtu[ctuRsAddr - 1] < (Pel)iQPAdapt))
1063	{
1064	pcPic->m_iOffsetCtu[ctuRsAddr - 1] = (Pel)iQPAdapt;
1065	}
1066	if ((ctuTsAddr == boundingAddr - 1) && (ctuRsAddr > pcv.widthInCtus)) // last CTU in the given slice
1067	{
1068	iQPAdapt = std::min (pcPic->m_iOffsetCtu[ctuRsAddr - 1], pcPic->m_iOffsetCtu[ctuRsAddr - pcv.widthInCtus]);
1069	if (pcPic->m_iOffsetCtu[ctuRsAddr] < (Pel)iQPAdapt)
1070	{
1071	pcPic->m_iOffsetCtu[ctuRsAddr] = (Pel)iQPAdapt;
1072	}
1073	}
1074	}
1075	} // end iteration over all CTUs in current slice
1076	}
1077
1078	return sliceQPModified;
1079	}
1080
1081	#if ENABLE_QPA_SUB_CTU
1082	static int applyQPAdaptationSubCtu (CodingStructure &cs, const UnitArea ctuArea, const uint32_t ctuAddr, const bool useSharpLumaDQP)
1083	{
1084	const PreCalcValues &pcv = *cs.pcv;
1085	const Picture *pcPic = cs.picture;
1086	const int bitDepth = cs.slice->getSPS()->getBitDepth (CHANNEL_TYPE_LUMA); // overall image bit-depth
1087	const int adaptedCtuQP = pcPic ? pcPic->m_iOffsetCtu[ctuAddr] : cs.slice->getSliceQpBase();
1088
1089	if (!pcPic \|\| cs.pps->getCuQpDeltaSubdiv() == 0) return adaptedCtuQP;
1090
1091	for (unsigned addr = 0; addr < cs.picture->m_subCtuQP.size(); addr++)
1092	{
1093	cs.picture->m_subCtuQP[addr] = (int8_t)adaptedCtuQP;
1094	}
1095	if (cs.slice->getSliceQp() < MAX_QP && pcv.widthInCtus > 1)
1096	{
1097	#if SHARP_LUMA_DELTA_QP
1098	const int lumaCtuDQP = useSharpLumaDQP ? lumaDQPOffset ((uint32_t)pcPic->m_uEnerHpCtu[ctuAddr], bitDepth) : 0;
1099	#endif
1100	#if MAX_TB_SIZE_SIGNALLING
1101	const unsigned mts = std::min (cs.sps->getMaxTbSize(), pcv.maxCUWidth);
1102	#else
1103	const unsigned mts = std::min<uint32_t> (MAX_TB_SIZEY, pcv.maxCUWidth);
1104	#endif
1105	const unsigned mtsLog2 = (unsigned)g_aucLog2[mts];
1106	const unsigned stride = pcv.maxCUWidth >> mtsLog2;
1107	unsigned numAct = 0; // number of block activities
1108	double sumAct = 0.0; // sum of all block activities
1109	double subAct[16]; // individual block activities
1110	#if SHARP_LUMA_DELTA_QP
1111	uint32_t subMLV[16]; // individual mean luma values
1112	#endif
1113
1114	CHECK (mts * 4 < pcv.maxCUWidth \|\| mts * 4 < pcv.maxCUHeight, "max. transform size is too small for given CTU size");
1115
1116	for (unsigned h = 0; h < (pcv.maxCUHeight >> mtsLog2); h++)
1117	{
1118	for (unsigned w = 0; w < stride; w++)
1119	{
1120	const unsigned addr = w + h * stride;
1121	const PosType x = ctuArea.lx() + w * mts;
1122	const PosType y = ctuArea.ly() + h * mts;
1123	const CompArea fltArea = clipArea (CompArea (COMPONENT_Y, pcPic->chromaFormat, Area (x > 0 ? x - 1 : 0, y > 0 ? y - 1 : 0, mts + (x > 0 ? 2 : 1), mts + (y > 0 ? 2 : 1))), pcPic->Y());
1124	const CPelBuf picOrig = pcPic->getOrigBuf (fltArea);
1125
1126	if (x >= pcPic->lwidth() \|\| y >= pcPic->lheight())
1127	{
1128	continue;
1129	}
1130	filterAndCalculateAverageEnergies (picOrig.buf, picOrig.stride, subAct[addr],
1131	picOrig.height, picOrig.width, bitDepth);
1132	numAct++;
1133	sumAct += subAct[addr];
1134	#if SHARP_LUMA_DELTA_QP
1135
1136	if (useSharpLumaDQP)
1137	{
1138	const CompArea subArea = clipArea (CompArea (COMPONENT_Y, pcPic->chromaFormat, Area (x, y, mts, mts)), pcPic->Y());
1139
1140	subMLV[addr] = pcPic->getOrigBuf (subArea).computeAvg();
1141	}
1142	#endif
1143	}
1144	}
1145	if (sumAct <= 0.0) return adaptedCtuQP;
1146
1147	sumAct = double(numAct) / sumAct; // 1.0 / (average CTU activity)
1148
1149	for (unsigned h = 0; h < (pcv.maxCUHeight >> mtsLog2); h++)
1150	{
1151	for (unsigned w = 0; w < stride; w++)
1152	{
1153	const unsigned addr = w + h * stride;
1154
1155	if (ctuArea.lx() + w * mts >= pcPic->lwidth() \|\| ctuArea.ly() + h * mts >= pcPic->lheight())
1156	{
1157	continue;
1158	}
1159	cs.picture->m_subCtuQP[addr] = (int8_t)Clip3 (0, MAX_QP, adaptedCtuQP + apprI3Log2 (subAct[addr] * sumAct));
1160	#if SHARP_LUMA_DELTA_QP
1161
1162	// change adapted QP based on mean sub-CTU luma value (Sharp)
1163	if (useSharpLumaDQP)
1164	{
1165	cs.picture->m_subCtuQP[addr] = (int8_t)Clip3 (0, MAX_QP, (int)cs.picture->m_subCtuQP[addr] - lumaCtuDQP + lumaDQPOffset (subMLV[addr], bitDepth));
1166	}
1167	#endif
1168	}
1169	}
1170	}
1171
1172	return adaptedCtuQP;
1173	}
1174	#endif // ENABLE_QPA_SUB_CTU
1175	#endif // ENABLE_QPA
1176
1177	// ====================================================================================================================
1178	// Public member functions
1179	// ====================================================================================================================
1180
1181	//! set adaptive search range based on poc difference
1182	void EncSlice::setSearchRange( Slice* pcSlice )
1183	{
1184	int iCurrPOC = pcSlice->getPOC();
1185	int iRefPOC;
1186	int iGOPSize = m_pcCfg->getGOPSize();
1187	int iOffset = (iGOPSize >> 1);
1188	int iMaxSR = m_pcCfg->getSearchRange();
1189	int iNumPredDir = pcSlice->isInterP() ? 1 : 2;
1190
1191	for (int iDir = 0; iDir < iNumPredDir; iDir++)
1192	{
1193	RefPicList e = ( iDir ? REF_PIC_LIST_1 : REF_PIC_LIST_0 );
1194	for (int iRefIdx = 0; iRefIdx < pcSlice->getNumRefIdx(e); iRefIdx++)
1195	{
1196	iRefPOC = pcSlice->getRefPic(e, iRefIdx)->getPOC();
1197	int newSearchRange = Clip3(m_pcCfg->getMinSearchWindow(), iMaxSR, (iMaxSRADAPT_SR_SCALEabs(iCurrPOC - iRefPOC)+iOffset)/iGOPSize);
1198	m_pcInterSearch->setAdaptiveSearchRange(iDir, iRefIdx, newSearchRange);
1199	#if ENABLE_WPP_PARALLELISM
1200	for( int jId = 1; jId < m_pcLib->getNumCuEncStacks(); jId++ )
1201	{
1202	m_pcLib->getInterSearch( jId )->setAdaptiveSearchRange( iDir, iRefIdx, newSearchRange );
1203	}
1204	#endif
1205	}
1206	}
1207	}
1208
1209	/**
1210	Multi-loop slice encoding for different slice QP
1211
1212	\param pcPic picture class
1213	*/
1214	void EncSlice::precompressSlice( Picture* pcPic )
1215	{
1216	// if deltaQP RD is not used, simply return
1217	if ( m_pcCfg->getDeltaQpRD() == 0 )
1218	{
1219	return;
1220	}
1221
1222	if ( m_pcCfg->getUseRateCtrl() )
1223	{
1224	THROW("\nMultiple QP optimization is not allowed when rate control is enabled." );
1225	}
1226
1227	Slice* pcSlice = pcPic->slices[getSliceSegmentIdx()];
1228
1229	#if HEVC_DEPENDENT_SLICES
1230	if (pcSlice->getDependentSliceSegmentFlag())
1231	{
1232	// if this is a dependent slice segment, then it was optimised
1233	// when analysing the entire slice.
1234	return;
1235	}
1236	#endif
1237
1238	if (pcSlice->getSliceMode()==FIXED_NUMBER_OF_BYTES)
1239	{
1240	// TODO: investigate use of average cost per CTU so that this Slice Mode can be used.
1241	THROW( "Unable to optimise Slice-level QP if Slice Mode is set to FIXED_NUMBER_OF_BYTES\n" );
1242	}
1243
1244	double dPicRdCostBest = MAX_DOUBLE;
1245	uint32_t uiQpIdxBest = 0;
1246
1247	double dFrameLambda;
1248	int SHIFT_QP = 12
1249	+ 6
1250	* (pcSlice->getSPS()->getBitDepth(CHANNEL_TYPE_LUMA) - 8
1251	- DISTORTION_PRECISION_ADJUSTMENT(pcSlice->getSPS()->getBitDepth(CHANNEL_TYPE_LUMA)));
1252
1253	// set frame lambda
1254	if (m_pcCfg->getGOPSize() > 1)
1255	{
1256	dFrameLambda = 0.68 * pow (2, (m_viRdPicQp[0] - SHIFT_QP) / 3.0) * (pcSlice->isInterB()? 2 : 1);
1257	}
1258	else
1259	{
1260	dFrameLambda = 0.68 * pow (2, (m_viRdPicQp[0] - SHIFT_QP) / 3.0);
1261	}
1262
1263	// for each QP candidate
1264	for ( uint32_t uiQpIdx = 0; uiQpIdx < 2 * m_pcCfg->getDeltaQpRD() + 1; uiQpIdx++ )
1265	{
1266	pcSlice ->setSliceQp ( m_viRdPicQp [uiQpIdx] );
1267	setUpLambda(pcSlice, m_vdRdPicLambda[uiQpIdx], m_viRdPicQp [uiQpIdx]);
1268
1269	// try compress
1270	compressSlice ( pcPic, true, m_pcCfg->getFastDeltaQp());
1271
1272	uint64_t uiPicDist = m_uiPicDist; // Distortion, as calculated by compressSlice.
1273	// NOTE: This distortion is the chroma-weighted SSE distortion for the slice.
1274	// Previously a standard SSE distortion was calculated (for the entire frame).
1275	// Which is correct?
1276	#if W0038_DB_OPT
1277	// TODO: Update loop filter, SAO and distortion calculation to work on one slice only.
1278	// uiPicDist = m_pcGOPEncoder->preLoopFilterPicAndCalcDist( pcPic );
1279	#endif
1280	// compute RD cost and choose the best
1281	double dPicRdCost = double( uiPicDist ) + dFrameLambda * double( m_uiPicTotalBits );
1282
1283	if ( dPicRdCost < dPicRdCostBest )
1284	{
1285	uiQpIdxBest = uiQpIdx;
1286	dPicRdCostBest = dPicRdCost;
1287	}
1288	}
1289
1290	// set best values
1291	pcSlice ->setSliceQp ( m_viRdPicQp [uiQpIdxBest] );
1292	setUpLambda(pcSlice, m_vdRdPicLambda[uiQpIdxBest], m_viRdPicQp [uiQpIdxBest]);
1293	}
1294
1295	void EncSlice::calCostSliceI(Picture* pcPic) // TODO: this only analyses the first slice segment. What about the others?
1296	{
1297	double iSumHadSlice = 0;
1298	Slice * const pcSlice = pcPic->slices[getSliceSegmentIdx()];
1299	const TileMap &tileMap = *pcPic->tileMap;
1300	const PreCalcValues& pcv = *pcPic->cs->pcv;
1301	const SPS &sps = *(pcSlice->getSPS());
1302	const int shift = sps.getBitDepth(CHANNEL_TYPE_LUMA)-8;
1303	const int offset = (shift>0)?(1<<(shift-1)):0;
1304
1305	#if HEVC_DEPENDENT_SLICES
1306	pcSlice->setSliceSegmentBits(0);
1307	#endif
1308
1309	uint32_t startCtuTsAddr, boundingCtuTsAddr;
1310	xDetermineStartAndBoundingCtuTsAddr ( startCtuTsAddr, boundingCtuTsAddr, pcPic );
1311
1312	for( uint32_t ctuTsAddr = startCtuTsAddr, ctuRsAddr = tileMap.getCtuTsToRsAddrMap( startCtuTsAddr);
1313	ctuTsAddr < boundingCtuTsAddr;
1314	ctuRsAddr = tileMap.getCtuTsToRsAddrMap(++ctuTsAddr) )
1315	{
1316	Position pos( (ctuRsAddr % pcv.widthInCtus) * pcv.maxCUWidth, (ctuRsAddr / pcv.widthInCtus) * pcv.maxCUHeight);
1317
1318	const int height = std::min( pcv.maxCUHeight, pcv.lumaHeight - pos.y );
1319	const int width = std::min( pcv.maxCUWidth, pcv.lumaWidth - pos.x );
1320	const CompArea blk( COMPONENT_Y, pcv.chrFormat, pos, Size( width, height));
1321	int iSumHad = m_pcCuEncoder->updateCtuDataISlice( pcPic->getOrigBuf( blk ) );
1322
1323	(m_pcRateCtrl->getRCPic()->getLCU(ctuRsAddr)).m_costIntra=(iSumHad+offset)>>shift;
1324	iSumHadSlice += (m_pcRateCtrl->getRCPic()->getLCU(ctuRsAddr)).m_costIntra;
1325
1326	}
1327	m_pcRateCtrl->getRCPic()->setTotalIntraCost(iSumHadSlice);
1328	}
1329
1330	/** \param pcPic picture class
1331	*/
1332	void EncSlice::compressSlice( Picture* pcPic, const bool bCompressEntireSlice, const bool bFastDeltaQP )
1333	{
1334	// if bCompressEntireSlice is true, then the entire slice (not slice segment) is compressed,
1335	// effectively disabling the slice-segment-mode.
1336
1337	Slice* const pcSlice = pcPic->slices[getSliceSegmentIdx()];
1338	uint32_t startCtuTsAddr;
1339	uint32_t boundingCtuTsAddr;
1340
1341	#if HEVC_DEPENDENT_SLICES
1342	pcSlice->setSliceSegmentBits(0);
1343	#endif
1344	xDetermineStartAndBoundingCtuTsAddr ( startCtuTsAddr, boundingCtuTsAddr, pcPic );
1345	if (bCompressEntireSlice)
1346	{
1347	boundingCtuTsAddr = pcSlice->getSliceCurEndCtuTsAddr();
1348	#if HEVC_DEPENDENT_SLICES
1349	pcSlice->setSliceSegmentCurEndCtuTsAddr(boundingCtuTsAddr);
1350	#endif
1351	}
1352
1353	// initialize cost values - these are used by precompressSlice (they should be parameters).
1354	m_uiPicTotalBits = 0;
1355	m_uiPicDist = 0;
1356
1357	pcSlice->setSliceQpBase( pcSlice->getSliceQp() );
1358
1359	m_CABACEstimator->initCtxModels( *pcSlice );
1360
1361	#if ENABLE_SPLIT_PARALLELISM \|\| ENABLE_WPP_PARALLELISM
1362	for( int jId = 1; jId < m_pcLib->getNumCuEncStacks(); jId++ )
1363	{
1364	CABACWriter* cw = m_pcLib->getCABACEncoder( jId )->getCABACEstimator( pcSlice->getSPS() );
1365	cw->initCtxModels( *pcSlice );
1366	}
1367
1368	#endif
1369	m_pcCuEncoder->getModeCtrl()->setFastDeltaQp(bFastDeltaQP);
1370
1371
1372	//------------------------------------------------------------------------------
1373	// Weighted Prediction parameters estimation.
1374	//------------------------------------------------------------------------------
1375	// calculate AC/DC values for current picture
1376	if( pcSlice->getPPS()->getUseWP() \|\| pcSlice->getPPS()->getWPBiPred() )
1377	{
1378	xCalcACDCParamSlice(pcSlice);
1379	}
1380
1381	const bool bWp_explicit = (pcSlice->getSliceType()==P_SLICE && pcSlice->getPPS()->getUseWP()) \|\| (pcSlice->getSliceType()==B_SLICE && pcSlice->getPPS()->getWPBiPred());
1382
1383	if ( bWp_explicit )
1384	{
1385	//------------------------------------------------------------------------------
1386	// Weighted Prediction implemented at Slice level. SliceMode=2 is not supported yet.
1387	//------------------------------------------------------------------------------
1388	#if HEVC_DEPENDENT_SLICES
1389	if ( pcSlice->getSliceMode()==FIXED_NUMBER_OF_BYTES \|\| pcSlice->getSliceSegmentMode()==FIXED_NUMBER_OF_BYTES )
1390	#else
1391	if(pcSlice->getSliceMode() == FIXED_NUMBER_OF_BYTES)
1392	#endif
1393	{
1394	EXIT("Weighted Prediction is not yet supported with slice mode determined by max number of bins.");
1395	}
1396
1397	xEstimateWPParamSlice( pcSlice, m_pcCfg->getWeightedPredictionMethod() );
1398	pcSlice->initWpScaling(pcSlice->getSPS());
1399
1400	// check WP on/off
1401	xCheckWPEnable( pcSlice );
1402	}
1403
1404
1405	#if HEVC_DEPENDENT_SLICES
1406	// Adjust initial state if this is the start of a dependent slice.
1407	{
1408	const TileMap& tileMap = *pcPic->tileMap;
1409	const uint32_t ctuRsAddr = tileMap.getCtuTsToRsAddrMap( startCtuTsAddr);
1410	const uint32_t currentTileIdx = tileMap.getTileIdxMap(ctuRsAddr);
1411	const Tile& currentTile = tileMap.tiles[currentTileIdx];
1412	const uint32_t firstCtuRsAddrOfTile = currentTile.getFirstCtuRsAddr();
1413	if( pcSlice->getDependentSliceSegmentFlag() && ctuRsAddr != firstCtuRsAddrOfTile )
1414	{
1415	// This will only occur if dependent slice-segments (m_entropyCodingSyncContextState=true) are being used.
1416	if( currentTile.getTileWidthInCtus() >= 2 \|\| !m_pcCfg->getEntropyCodingSyncEnabledFlag() )
1417	{
1418	m_CABACEstimator->getCtx() = m_lastSliceSegmentEndContextState;
1419	m_CABACEstimator->start();
1420	}
1421	}
1422	}
1423	#endif
1424
1425	#if HEVC_DEPENDENT_SLICES
1426	if( !pcSlice->getDependentSliceSegmentFlag() )
1427	{
1428	#endif
1429	pcPic->m_prevQP[0] = pcPic->m_prevQP[1] = pcSlice->getSliceQp();
1430	#if HEVC_DEPENDENT_SLICES
1431	}
1432	#endif
1433
1434	CHECK( pcPic->m_prevQP[0] == std::numeric_limits<int>::max(), "Invalid previous QP" );
1435
1436	CodingStructure& cs = *pcPic->cs;
1437	cs.slice = pcSlice;
1438	cs.pcv = pcSlice->getPPS()->pcv;
1439	cs.fracBits = 0;
1440
1441	if( startCtuTsAddr == 0 && ( pcSlice->getPOC() != m_pcCfg->getSwitchPOC() \|\| -1 == m_pcCfg->getDebugCTU() ) )
1442	{
1443	cs.initStructData (pcSlice->getSliceQp(), pcSlice->getPPS()->getTransquantBypassEnabledFlag());
1444	}
1445
1446	#if ENABLE_QPA
1447	if (m_pcCfg->getUsePerceptQPA() && !m_pcCfg->getUseRateCtrl() && (boundingCtuTsAddr > startCtuTsAddr))
1448	{
1449	if (applyQPAdaptation (pcPic, pcSlice, *cs.pcv, startCtuTsAddr, boundingCtuTsAddr, m_pcCfg->getLumaLevelToDeltaQPMapping().mode == LUMALVL_TO_DQP_NUM_MODES,
1450	(m_pcCfg->getBaseQP() >= 38) \|\| (m_pcCfg->getSourceWidth() <= 512 && m_pcCfg->getSourceHeight() <= 320), m_adaptedLumaQP))
1451	{
1452	m_CABACEstimator->initCtxModels (*pcSlice);
1453	#if ENABLE_SPLIT_PARALLELISM \|\| ENABLE_WPP_PARALLELISM
1454	for (int jId = 1; jId < m_pcLib->getNumCuEncStacks(); jId++)
1455	{
1456	CABACWriter* cw = m_pcLib->getCABACEncoder (jId)->getCABACEstimator (pcSlice->getSPS());
1457	cw->initCtxModels (*pcSlice);
1458	}
1459	#endif
1460	#if HEVC_DEPENDENT_SLICES
1461	if (!pcSlice->getDependentSliceSegmentFlag())
1462	{
1463	#endif
1464	pcPic->m_prevQP[0] = pcPic->m_prevQP[1] = pcSlice->getSliceQp();
1465	#if HEVC_DEPENDENT_SLICES
1466	}
1467	#endif
1468	if (startCtuTsAddr == 0)
1469	{
1470	cs.currQP[0] = cs.currQP[1] = pcSlice->getSliceQp(); // cf code above
1471	}
1472	}
1473	}
1474	#endif // ENABLE_QPA
1475
1476	#if ENABLE_WPP_PARALLELISM
1477	bool bUseThreads = m_pcCfg->getNumWppThreads() > 1;
1478	if( bUseThreads )
1479	{
1480	CHECK( startCtuTsAddr != 0 \|\| boundingCtuTsAddr != pcPic->cs->pcv->sizeInCtus, "not intended" );
1481
1482	pcPic->cs->allocateVectorsAtPicLevel();
1483
1484	omp_set_num_threads( m_pcCfg->getNumWppThreads() + m_pcCfg->getNumWppExtraLines() );
1485
1486	#pragma omp parallel for schedule(static,1) if(bUseThreads)
1487	for( int ctuTsAddr = startCtuTsAddr; ctuTsAddr < boundingCtuTsAddr; ctuTsAddr += widthInCtus )
1488	{
1489	// wpp thread start
1490	pcPic->scheduler.setWppThreadId();
1491	#if ENABLE_SPLIT_PARALLELISM
1492	pcPic->scheduler.setSplitThreadId( 0 );
1493	#endif
1494	encodeCtus( pcPic, bCompressEntireSlice, bFastDeltaQP, ctuTsAddr, ctuTsAddr + widthInCtus, m_pcLib );
1495	// wpp thread stop
1496	}
1497	}
1498	else
1499	#endif
1500	#if K0149_BLOCK_STATISTICS
1501	const SPS *sps = pcSlice->getSPS();
1502	CHECK(sps == 0, "No SPS present");
1503	writeBlockStatisticsHeader(sps);
1504	#endif
1505	m_pcInterSearch->resetAffineMVList();
1506	encodeCtus( pcPic, bCompressEntireSlice, bFastDeltaQP, startCtuTsAddr, boundingCtuTsAddr, m_pcLib );
1507
1508	#if HEVC_DEPENDENT_SLICES
1509	// store context state at the end of this slice-segment, in case the next slice is a dependent slice and continues using the CABAC contexts.
1510	if( pcSlice->getPPS()->getDependentSliceSegmentsEnabledFlag() )
1511	{
1512	m_lastSliceSegmentEndContextState = m_CABACEstimator->getCtx();//ctx end of dep.slice
1513	}
1514	#endif
1515
1516	}
1517
1518	void EncSlice::checkDisFracMmvd( Picture* pcPic, uint32_t startCtuTsAddr, uint32_t boundingCtuTsAddr )
1519	{
1520	CodingStructure& cs = *pcPic->cs;
1521	Slice* pcSlice = cs.slice;
1522	const PreCalcValues& pcv = *cs.pcv;
1523	const uint32_t widthInCtus = pcv.widthInCtus;
1524	const TileMap& tileMap = *pcPic->tileMap;
1525	const uint32_t hashThreshold = 20;
1526	uint32_t totalCtu = 0;
1527	uint32_t hashRatio = 0;
1528
1529	if ( !pcSlice->getSPS()->getFpelMmvdEnabledFlag() )
1530	{
1531	return;
1532	}
1533
1534	for ( uint32_t ctuTsAddr = startCtuTsAddr; ctuTsAddr < boundingCtuTsAddr; ctuTsAddr++ )
1535	{
1536	const uint32_t ctuRsAddr = tileMap.getCtuTsToRsAddrMap( ctuTsAddr );
1537	const uint32_t ctuXPosInCtus = ctuRsAddr % widthInCtus;
1538	const uint32_t ctuYPosInCtus = ctuRsAddr / widthInCtus;
1539
1540	const Position pos ( ctuXPosInCtus * pcv.maxCUWidth, ctuYPosInCtus * pcv.maxCUHeight );
1541	const UnitArea ctuArea( cs.area.chromaFormat, Area( pos.x, pos.y, pcv.maxCUWidth, pcv.maxCUHeight ) );
1542
1543	hashRatio += m_pcCuEncoder->getIbcHashMap().getHashHitRatio( ctuArea.Y() );
1544	totalCtu++;
1545	}
1546
1547	if ( hashRatio > totalCtu * hashThreshold )
1548	{
1549	pcSlice->setDisFracMMVD( true );
1550	}
1551	if (!pcSlice->getDisFracMMVD()) {
1552	bool useIntegerMVD = (pcPic->lwidth()pcPic->lheight() > 1920 1080);
1553	pcSlice->setDisFracMMVD( useIntegerMVD );
1554	}
1555	}
1556
1557	void EncSlice::encodeCtus( Picture* pcPic, const bool bCompressEntireSlice, const bool bFastDeltaQP, uint32_t startCtuTsAddr, uint32_t boundingCtuTsAddr, EncLib* pEncLib )
1558	{
1559	CodingStructure& cs = *pcPic->cs;
1560	Slice* pcSlice = cs.slice;
1561	const PreCalcValues& pcv = *cs.pcv;
1562	const uint32_t widthInCtus = pcv.widthInCtus;
1563	const TileMap& tileMap = *pcPic->tileMap;
1564	#if ENABLE_QPA
1565	const int iQPIndex = pcSlice->getSliceQpBase();
1566	#endif
1567
1568	#if ENABLE_WPP_PARALLELISM
1569	const int dataId = pcPic->scheduler.getWppDataId();
1570	#elif ENABLE_SPLIT_PARALLELISM
1571	const int dataId = 0;
1572	#endif
1573	CABACWriter* pCABACWriter = pEncLib->getCABACEncoder( PARL_PARAM0( dataId ) )->getCABACEstimator( pcSlice->getSPS() );
1574	TrQuant* pTrQuant = pEncLib->getTrQuant( PARL_PARAM0( dataId ) );
1575	RdCost* pRdCost = pEncLib->getRdCost( PARL_PARAM0( dataId ) );
1576	EncCfg* pCfg = pEncLib;
1577	RateCtrl* pRateCtrl = pEncLib->getRateCtrl();
1578	#if ENABLE_WPP_PARALLELISM
1579	// first version dont use ctx from above
1580	pCABACWriter->initCtxModels( *pcSlice );
1581	#endif
1582	#if RDOQ_CHROMA_LAMBDA
1583	pTrQuant ->setLambdas( pcSlice->getLambdas() );
1584	#else
1585	pTrQuant ->setLambda ( pcSlice->getLambdas()[0] );
1586	#endif
1587	pRdCost ->setLambda ( pcSlice->getLambdas()[0], pcSlice->getSPS()->getBitDepths() );
1588
1589	int prevQP[2];
1590	int currQP[2];
1591	prevQP[0] = prevQP[1] = pcSlice->getSliceQp();
1592	currQP[0] = currQP[1] = pcSlice->getSliceQp();
1593
1594	#if HEVC_DEPENDENT_SLICES
1595	if( !pcSlice->getDependentSliceSegmentFlag() )
1596	{
1597	#endif
1598	prevQP[0] = prevQP[1] = pcSlice->getSliceQp();
1599	#if HEVC_DEPENDENT_SLICES
1600	}
1601	#endif
1602	if ( pcSlice->getSPS()->getFpelMmvdEnabledFlag() \|\|
1603	(pcSlice->getSPS()->getIBCFlag() && m_pcCuEncoder->getEncCfg()->getIBCHashSearch()))
1604	{
1605	#if JVET_N0329_IBC_SEARCH_IMP
1606	m_pcCuEncoder->getIbcHashMap().rebuildPicHashMap(cs.picture->getTrueOrigBuf());
1607	if (m_pcCfg->getIntraPeriod() != -1)
1608	{
1609	int hashBlkHitPerc = m_pcCuEncoder->getIbcHashMap().calHashBlkMatchPerc(cs.area.Y());
1610	cs.slice->setDisableSATDForRD(hashBlkHitPerc > 59);
1611	}
1612	#else
1613	if (pcSlice->getSPS()->getUseReshaper() && m_pcLib->getReshaper()->getCTUFlag() && pcSlice->getSPS()->getIBCFlag())
1614	cs.picture->getOrigBuf(COMPONENT_Y).rspSignal(m_pcLib->getReshaper()->getFwdLUT());
1615	m_pcCuEncoder->getIbcHashMap().rebuildPicHashMap( cs.picture->getOrigBuf() );
1616	if (pcSlice->getSPS()->getUseReshaper() && m_pcLib->getReshaper()->getCTUFlag() && pcSlice->getSPS()->getIBCFlag())
1617	cs.picture->getOrigBuf().copyFrom(cs.picture->getTrueOrigBuf());
1618	#endif
1619	}
1620	checkDisFracMmvd( pcPic, startCtuTsAddr, boundingCtuTsAddr );
1621	// for every CTU in the slice segment (may terminate sooner if there is a byte limit on the slice-segment)
1622	for( uint32_t ctuTsAddr = startCtuTsAddr; ctuTsAddr < boundingCtuTsAddr; ctuTsAddr++ )
1623	{
1624	const int32_t ctuRsAddr = tileMap.getCtuTsToRsAddrMap( ctuTsAddr );
1625
1626	// update CABAC state
1627	const uint32_t firstCtuRsAddrOfTile = tileMap.tiles[tileMap.getTileIdxMap(ctuRsAddr)].getFirstCtuRsAddr();
1628	const uint32_t tileXPosInCtus = firstCtuRsAddrOfTile % widthInCtus;
1629	const uint32_t ctuXPosInCtus = ctuRsAddr % widthInCtus;
1630	const uint32_t ctuYPosInCtus = ctuRsAddr / widthInCtus;
1631
1632	const Position pos (ctuXPosInCtus * pcv.maxCUWidth, ctuYPosInCtus * pcv.maxCUHeight);
1633	const UnitArea ctuArea( cs.area.chromaFormat, Area( pos.x, pos.y, pcv.maxCUWidth, pcv.maxCUHeight ) );
1634	DTRACE_UPDATE( g_trace_ctx, std::make_pair( "ctu", ctuRsAddr ) );
1635
1636	if( pCfg->getSwitchPOC() != pcPic->poc \|\| -1 == pCfg->getDebugCTU() )
1637	#if FIXBUG_RESET_LUT
1638	if ((cs.slice->getSliceType() != I_SLICE \|\| cs.sps->getIBCFlag()) && ctuXPosInCtus == tileXPosInCtus)
1639	#else
1640	if ((cs.slice->getSliceType() != I_SLICE \|\| cs.sps->getIBCFlag()) && ctuXPosInCtus == 0)
1641	#endif
1642	{
1643	cs.motionLut.lut.resize(0);
1644	cs.motionLut.lutIbc.resize(0);
1645	#if !JVET_N0266_SMALL_BLOCKS
1646	cs.motionLut.lutShare.resize(0);
1647	#endif
1648	cs.motionLut.lutShareIbc.resize(0);
1649	}
1650
1651	#if ENABLE_WPP_PARALLELISM
1652	pcPic->scheduler.wait( ctuXPosInCtus, ctuYPosInCtus );
1653	#endif
1654
1655	if (ctuRsAddr == firstCtuRsAddrOfTile)
1656	{
1657	pCABACWriter->initCtxModels( *pcSlice );
1658	prevQP[0] = prevQP[1] = pcSlice->getSliceQp();
1659	}
1660	else if (ctuXPosInCtus == tileXPosInCtus && pEncLib->getEntropyCodingSyncEnabledFlag())
1661	{
1662	// reset and then update contexts to the state at the end of the top-right CTU (if within current slice and tile).
1663	pCABACWriter->initCtxModels( *pcSlice );
1664	if( cs.getCURestricted( pos.offset(pcv.maxCUWidth, -1), pcSlice->getIndependentSliceIdx(), tileMap.getTileIdxMap( pos ), CH_L ) )
1665	{
1666	// Top-right is available, we use it.
1667	pCABACWriter->getCtx() = pEncLib->m_entropyCodingSyncContextState;
1668	}
1669	prevQP[0] = prevQP[1] = pcSlice->getSliceQp();
1670	}
1671
1672	#if ENABLE_WPP_PARALLELISM
1673	if( ctuXPosInCtus == 0 && ctuYPosInCtus > 0 && widthInCtus > 1 && ( pEncLib->getNumWppThreads() > 1 \|\| pEncLib->getEnsureWppBitEqual() ) )
1674	{
1675	pCABACWriter->getCtx() = pEncLib->m_entropyCodingSyncContextStateVec[ctuYPosInCtus-1]; // last line
1676	}
1677	#else
1678	#endif
1679
1680	#if RDOQ_CHROMA_LAMBDA && ENABLE_QPA && !ENABLE_QPA_SUB_CTU
1681	double oldLambdaArray[MAX_NUM_COMPONENT] = {0.0};
1682	#endif
1683	const double oldLambda = pRdCost->getLambda();
1684	if ( pCfg->getUseRateCtrl() )
1685	{
1686	int estQP = pcSlice->getSliceQp();
1687	double estLambda = -1.0;
1688	double bpp = -1.0;
1689
1690	if( ( pcPic->slices[0]->isIRAP() && pCfg->getForceIntraQP() ) \|\| !pCfg->getLCULevelRC() )
1691	{
1692	estQP = pcSlice->getSliceQp();
1693	}
1694	else
1695	{
1696	bpp = pRateCtrl->getRCPic()->getLCUTargetBpp(pcSlice->isIRAP());
1697	if ( pcPic->slices[0]->isIRAP())
1698	{
1699	estLambda = pRateCtrl->getRCPic()->getLCUEstLambdaAndQP(bpp, pcSlice->getSliceQp(), &estQP);
1700	}
1701	else
1702	{
1703	estLambda = pRateCtrl->getRCPic()->getLCUEstLambda( bpp );
1704	estQP = pRateCtrl->getRCPic()->getLCUEstQP ( estLambda, pcSlice->getSliceQp() );
1705	}
1706
1707	estQP = Clip3( -pcSlice->getSPS()->getQpBDOffset(CHANNEL_TYPE_LUMA), MAX_QP, estQP );
1708
1709	pRdCost->setLambda(estLambda, pcSlice->getSPS()->getBitDepths());
1710
1711	#if RDOQ_CHROMA_LAMBDA
1712	// set lambda for RDOQ
1713	const double chromaLambda = estLambda / pRdCost->getChromaWeight();
1714	const double lambdaArray[MAX_NUM_COMPONENT] = { estLambda, chromaLambda, chromaLambda };
1715	pTrQuant->setLambdas( lambdaArray );
1716	#else
1717	pTrQuant->setLambda( estLambda );
1718	#endif
1719	}
1720
1721	pRateCtrl->setRCQP( estQP );
1722	}
1723	#if ENABLE_QPA
1724	else if (pCfg->getUsePerceptQPA() && pcSlice->getPPS()->getUseDQP())
1725	{
1726	#if ENABLE_QPA_SUB_CTU
1727	const int adaptedQP = applyQPAdaptationSubCtu (cs, ctuArea, ctuRsAddr, m_pcCfg->getLumaLevelToDeltaQPMapping().mode == LUMALVL_TO_DQP_NUM_MODES);
1728	#else
1729	const int adaptedQP = pcPic->m_iOffsetCtu[ctuRsAddr];
1730	#endif
1731	const double newLambda = pcSlice->getLambdas()[0] * pow (2.0, double (adaptedQP - iQPIndex) / 3.0);
1732	pcPic->m_uEnerHpCtu[ctuRsAddr] = newLambda; // for ALF and SAO
1733	#if !ENABLE_QPA_SUB_CTU
1734	#if RDOQ_CHROMA_LAMBDA
1735	pTrQuant->getLambdas (oldLambdaArray); // save the old lambdas
1736	const double chromaLambda = newLambda / pRdCost->getChromaWeight();
1737	const double lambdaArray[MAX_NUM_COMPONENT] = {newLambda, chromaLambda, chromaLambda};
1738	pTrQuant->setLambdas (lambdaArray);
1739	#else
1740	pTrQuant->setLambda (newLambda);
1741	#endif
1742	pRdCost->setLambda (newLambda, pcSlice->getSPS()->getBitDepths());
1743	#endif
1744	currQP[0] = currQP[1] = adaptedQP;
1745	}
1746	#endif
1747
1748	bool updateGbiCodingOrder = cs.slice->getSliceType() == B_SLICE && ctuTsAddr == startCtuTsAddr;
1749	if( updateGbiCodingOrder )
1750	{
1751	resetGbiCodingOrder(false, cs);
1752	m_pcInterSearch->initWeightIdxBits();
1753	}
1754	if (pcSlice->getSPS()->getUseReshaper())
1755	{
1756	m_pcCuEncoder->setDecCuReshaperInEncCU(m_pcLib->getReshaper(), pcSlice->getSPS()->getChromaFormatIdc());
1757
1758	#if ENABLE_SPLIT_PARALLELISM \|\| ENABLE_WPP_PARALLELISM
1759	for (int jId = 1; jId < m_pcLib->getNumCuEncStacks(); jId++)
1760	{
1761	m_pcLib->getCuEncoder(jId)->setDecCuReshaperInEncCU(m_pcLib->getReshaper(jId), pcSlice->getSPS()->getChromaFormatIdc());
1762	}
1763	#endif
1764	}
1765	if( !cs.slice->isIntra() && pCfg->getMCTSEncConstraint() )
1766	{
1767	pcPic->mctsInfo.init( &cs, ctuRsAddr );
1768	}
1769
1770	if (pCfg->getSwitchPOC() != pcPic->poc \|\| ctuRsAddr >= pCfg->getDebugCTU())
1771	#if ENABLE_WPP_PARALLELISM
1772	pEncLib->getCuEncoder( dataId )->compressCtu( cs, ctuArea, ctuRsAddr, prevQP, currQP );
1773	#else
1774	m_pcCuEncoder->compressCtu( cs, ctuArea, ctuRsAddr, prevQP, currQP );
1775	#endif
1776
1777	#if K0149_BLOCK_STATISTICS
1778	getAndStoreBlockStatistics(cs, ctuArea);
1779	#endif
1780
1781	pCABACWriter->resetBits();
1782	pCABACWriter->coding_tree_unit( cs, ctuArea, prevQP, ctuRsAddr, true );
1783	const int numberOfWrittenBits = int( pCABACWriter->getEstFracBits() >> SCALE_BITS );
1784
1785	// Calculate if this CTU puts us over slice bit size.
1786	// cannot terminate if current slice/slice-segment would be 0 Ctu in size,
1787	const uint32_t validEndOfSliceCtuTsAddr = ctuTsAddr + (ctuTsAddr == startCtuTsAddr ? 1 : 0);
1788	// Set slice end parameter
1789	if(pcSlice->getSliceMode()==FIXED_NUMBER_OF_BYTES && pcSlice->getSliceBits()+numberOfWrittenBits > (pcSlice->getSliceArgument()<<3))
1790	{
1791	#if HEVC_DEPENDENT_SLICES
1792	pcSlice->setSliceSegmentCurEndCtuTsAddr(validEndOfSliceCtuTsAddr);
1793	#endif
1794	pcSlice->setSliceCurEndCtuTsAddr(validEndOfSliceCtuTsAddr);
1795	boundingCtuTsAddr=validEndOfSliceCtuTsAddr;
1796	}
1797	#if HEVC_DEPENDENT_SLICES
1798	else if((!bCompressEntireSlice) && pcSlice->getSliceSegmentMode()==FIXED_NUMBER_OF_BYTES && pcSlice->getSliceSegmentBits()+numberOfWrittenBits > (pcSlice->getSliceSegmentArgument()<<3))
1799	{
1800	pcSlice->setSliceSegmentCurEndCtuTsAddr(validEndOfSliceCtuTsAddr);
1801	boundingCtuTsAddr=validEndOfSliceCtuTsAddr;
1802	}
1803	#endif
1804	if (boundingCtuTsAddr <= ctuTsAddr)
1805	{
1806	break;
1807	}
1808
1809	#if ENABLE_WPP_PARALLELISM \|\| ENABLE_SPLIT_PARALLELISM
1810	#pragma omp critical
1811	#endif
1812	pcSlice->setSliceBits( ( uint32_t ) ( pcSlice->getSliceBits() + numberOfWrittenBits ) );
1813	#if ENABLE_WPP_PARALLELISM \|\| ENABLE_SPLIT_PARALLELISM
1814	#pragma omp critical
1815	#endif
1816	#if HEVC_DEPENDENT_SLICES
1817	pcSlice->setSliceSegmentBits( pcSlice->getSliceSegmentBits() + numberOfWrittenBits );
1818	#endif
1819
1820	// Store probabilities of second CTU in line into buffer - used only if wavefront-parallel-processing is enabled.
1821	if( ctuXPosInCtus == tileXPosInCtus + 1 && pEncLib->getEntropyCodingSyncEnabledFlag() )
1822	{
1823	pEncLib->m_entropyCodingSyncContextState = pCABACWriter->getCtx();
1824	}
1825	#if ENABLE_WPP_PARALLELISM
1826	if( ctuXPosInCtus == 1 && ( pEncLib->getNumWppThreads() > 1 \|\| pEncLib->getEnsureWppBitEqual() ) )
1827	{
1828	pEncLib->m_entropyCodingSyncContextStateVec[ctuYPosInCtus] = pCABACWriter->getCtx();
1829	}
1830	#endif
1831
1832	#if !ENABLE_WPP_PARALLELISM
1833	int actualBits = int(cs.fracBits >> SCALE_BITS);
1834	actualBits -= (int)m_uiPicTotalBits;
1835	#endif
1836	if ( pCfg->getUseRateCtrl() )
1837	{
1838	#if ENABLE_WPP_PARALLELISM
1839	int actualBits = int( cs.fracBits >> SCALE_BITS );
1840	actualBits -= (int)m_uiPicTotalBits;
1841	#endif
1842	int actualQP = g_RCInvalidQPValue;
1843	double actualLambda = pRdCost->getLambda();
1844	int numberOfEffectivePixels = 0;
1845
1846	int numberOfSkipPixel = 0;
1847	for (auto &cu : cs.traverseCUs(ctuArea, CH_L))
1848	{
1849	numberOfSkipPixel += cu.skip*cu.lumaSize().area();
1850	}
1851
1852	for( auto &cu : cs.traverseCUs( ctuArea, CH_L ) )
1853	{
1854	if( !cu.skip \|\| cu.rootCbf )
1855	{
1856	numberOfEffectivePixels += cu.lumaSize().area();
1857	break;
1858	}
1859	}
1860	double skipRatio = (double)numberOfSkipPixel / ctuArea.lumaSize().area();
1861	CodingUnit* cu = cs.getCU( ctuArea.lumaPos(), CH_L );
1862
1863	if ( numberOfEffectivePixels == 0 )
1864	{
1865	actualQP = g_RCInvalidQPValue;
1866	}
1867	else
1868	{
1869	actualQP = cu->qp;
1870	}
1871	pRdCost->setLambda(oldLambda, pcSlice->getSPS()->getBitDepths());
1872	pRateCtrl->getRCPic()->updateAfterCTU(pRateCtrl->getRCPic()->getLCUCoded(), actualBits, actualQP, actualLambda, skipRatio,
1873	pcSlice->isIRAP() ? 0 : pCfg->getLCULevelRC());
1874	}
1875	#if ENABLE_QPA && !ENABLE_QPA_SUB_CTU
1876	else if (pCfg->getUsePerceptQPA() && pcSlice->getPPS()->getUseDQP())
1877	{
1878	#if RDOQ_CHROMA_LAMBDA
1879	pTrQuant->setLambdas (oldLambdaArray);
1880	#else
1881	pTrQuant->setLambda (oldLambda);
1882	#endif
1883	pRdCost->setLambda (oldLambda, pcSlice->getSPS()->getBitDepths());
1884	}
1885	#endif
1886
1887	#if !ENABLE_WPP_PARALLELISM
1888	m_uiPicTotalBits += actualBits;
1889	m_uiPicDist = cs.dist;
1890	#endif
1891	#if ENABLE_WPP_PARALLELISM
1892	pcPic->scheduler.setReady( ctuXPosInCtus, ctuYPosInCtus );
1893	#endif
1894	}
1895
1896	// this is wpp exclusive section
1897
1898	// m_uiPicTotalBits += actualBits;
1899	// m_uiPicDist = cs.dist;
1900
1901	}
1902
1903	void EncSlice::encodeSlice ( Picture* pcPic, OutputBitstream* pcSubstreams, uint32_t &numBinsCoded )
1904	{
1905
1906	Slice *const pcSlice = pcPic->slices[getSliceSegmentIdx()];
1907	const TileMap& tileMap = *pcPic->tileMap;
1908	#if HEVC_DEPENDENT_SLICES
1909	const uint32_t startCtuTsAddr = pcSlice->getSliceSegmentCurStartCtuTsAddr();
1910	const uint32_t boundingCtuTsAddr = pcSlice->getSliceSegmentCurEndCtuTsAddr();
1911	const bool depSliceSegmentsEnabled = pcSlice->getPPS()->getDependentSliceSegmentsEnabledFlag();
1912	#else
1913	const uint32_t startCtuTsAddr = pcSlice->getSliceCurStartCtuTsAddr();
1914	const uint32_t boundingCtuTsAddr = pcSlice->getSliceCurEndCtuTsAddr();
1915	#endif
1916	const bool wavefrontsEnabled = pcSlice->getPPS()->getEntropyCodingSyncEnabledFlag();
1917
1918
1919	// setup coding structure
1920	CodingStructure& cs = *pcPic->cs;
1921	cs.slice = pcSlice;
1922	// initialise entropy coder for the slice
1923	m_CABACWriter->initCtxModels( *pcSlice );
1924
1925	DTRACE( g_trace_ctx, D_HEADER, "=========== POC: %d ===========\n", pcSlice->getPOC() );
1926
1927	#if HEVC_DEPENDENT_SLICES
1928	if (depSliceSegmentsEnabled)
1929	{
1930	// modify initial contexts with previous slice segment if this is a dependent slice.
1931	const uint32_t ctuRsAddr = tileMap.getCtuTsToRsAddrMap( startCtuTsAddr );
1932	const uint32_t currentTileIdx = tileMap.getTileIdxMap(ctuRsAddr);
1933	const Tile& currentTile = tileMap.tiles[currentTileIdx];
1934	const uint32_t firstCtuRsAddrOfTile = currentTile.getFirstCtuRsAddr();
1935
1936	if( pcSlice->getDependentSliceSegmentFlag() && ctuRsAddr != firstCtuRsAddrOfTile )
1937	{
1938	if( currentTile.getTileWidthInCtus() >= 2 \|\| !wavefrontsEnabled )
1939	{
1940	m_CABACWriter->getCtx() = m_lastSliceSegmentEndContextState;
1941	}
1942	}
1943	}
1944
1945	if( !pcSlice->getDependentSliceSegmentFlag() )
1946	{
1947	#endif
1948	pcPic->m_prevQP[0] = pcPic->m_prevQP[1] = pcSlice->getSliceQp();
1949	#if HEVC_DEPENDENT_SLICES
1950	}
1951	#endif
1952
1953	const PreCalcValues& pcv = *cs.pcv;
1954	const uint32_t widthInCtus = pcv.widthInCtus;
1955
1956	// for every CTU in the slice segment...
1957
1958	for( uint32_t ctuTsAddr = startCtuTsAddr; ctuTsAddr < boundingCtuTsAddr; ctuTsAddr++ )
1959	{
1960	const uint32_t ctuRsAddr = tileMap.getCtuTsToRsAddrMap(ctuTsAddr);
1961	const Tile& currentTile = tileMap.tiles[tileMap.getTileIdxMap(ctuRsAddr)];
1962	const uint32_t firstCtuRsAddrOfTile = currentTile.getFirstCtuRsAddr();
1963	const uint32_t tileXPosInCtus = firstCtuRsAddrOfTile % widthInCtus;
1964	const uint32_t tileYPosInCtus = firstCtuRsAddrOfTile / widthInCtus;
1965	const uint32_t ctuXPosInCtus = ctuRsAddr % widthInCtus;
1966	const uint32_t ctuYPosInCtus = ctuRsAddr / widthInCtus;
1967	const uint32_t uiSubStrm = tileMap.getSubstreamForCtuAddr(ctuRsAddr, true, pcSlice);
1968
1969	DTRACE_UPDATE( g_trace_ctx, std::make_pair( "ctu", ctuRsAddr ) );
1970
1971	const Position pos (ctuXPosInCtus * pcv.maxCUWidth, ctuYPosInCtus * pcv.maxCUHeight);
1972	const UnitArea ctuArea (cs.area.chromaFormat, Area(pos.x, pos.y, pcv.maxCUWidth, pcv.maxCUHeight));
1973	m_CABACWriter->initBitstream( &pcSubstreams[uiSubStrm] );
1974
1975	// set up CABAC contexts' state for this CTU
1976	if (ctuRsAddr == firstCtuRsAddrOfTile)
1977	{
1978	if (ctuTsAddr != startCtuTsAddr) // if it is the first CTU, then the entropy coder has already been reset
1979	{
1980	m_CABACWriter->initCtxModels( *pcSlice );
1981	}
1982	}
1983	else if (ctuXPosInCtus == tileXPosInCtus && wavefrontsEnabled)
1984	{
1985	// Synchronize cabac probabilities with upper-right CTU if it's available and at the start of a line.
1986	if (ctuTsAddr != startCtuTsAddr) // if it is the first CTU, then the entropy coder has already been reset
1987	{
1988	m_CABACWriter->initCtxModels( *pcSlice );
1989	}
1990	if( cs.getCURestricted( pos.offset( pcv.maxCUWidth, -1 ), pcSlice->getIndependentSliceIdx(), tileMap.getTileIdxMap( pos ), CH_L ) )
1991	{
1992	// Top-right is available, so use it.
1993	m_CABACWriter->getCtx() = m_entropyCodingSyncContextState;
1994	}
1995	}
1996
1997	bool updateGbiCodingOrder = cs.slice->getSliceType() == B_SLICE && ctuTsAddr == startCtuTsAddr;
1998	if( updateGbiCodingOrder )
1999	{
2000	resetGbiCodingOrder(false, cs);
2001	}
2002
2003	m_CABACWriter->coding_tree_unit( cs, ctuArea, pcPic->m_prevQP, ctuRsAddr );
2004
2005	// store probabilities of second CTU in line into buffer
2006	if( ctuXPosInCtus == tileXPosInCtus + 1 && wavefrontsEnabled )
2007	{
2008	m_entropyCodingSyncContextState = m_CABACWriter->getCtx();
2009	}
2010
2011	// terminate the sub-stream, if required (end of slice-segment, end of tile, end of wavefront-CTU-row):
2012	if( ctuTsAddr + 1 == boundingCtuTsAddr \|\|
2013	( ctuXPosInCtus + 1 == tileXPosInCtus + currentTile.getTileWidthInCtus () &&
2014	( ctuYPosInCtus + 1 == tileYPosInCtus + currentTile.getTileHeightInCtus() \|\| wavefrontsEnabled )
2015	)
2016	)
2017	{
2018	m_CABACWriter->end_of_slice();
2019
2020	// Byte-alignment in slice_data() when new tile
2021	pcSubstreams[uiSubStrm].writeByteAlignment();
2022
2023	// write sub-stream size
2024	if( ctuTsAddr + 1 != boundingCtuTsAddr )
2025	{
2026	pcSlice->addSubstreamSize( (pcSubstreams[uiSubStrm].getNumberOfWrittenBits() >> 3) + pcSubstreams[uiSubStrm].countStartCodeEmulations() );
2027	}
2028	}
2029	} // CTU-loop
2030
2031	#if HEVC_DEPENDENT_SLICES
2032	if( depSliceSegmentsEnabled )
2033	{
2034	m_lastSliceSegmentEndContextState = m_CABACWriter->getCtx();//ctx end of dep.slice
2035	}
2036	#endif
2037
2038	#if HEVC_DEPENDENT_SLICES
2039	if (pcSlice->getPPS()->getCabacInitPresentFlag() && !pcSlice->getPPS()->getDependentSliceSegmentsEnabledFlag())
2040	#else
2041	if(pcSlice->getPPS()->getCabacInitPresentFlag())
2042	#endif
2043	{
2044	m_encCABACTableIdx = m_CABACWriter->getCtxInitId( *pcSlice );
2045	}
2046	else
2047	{
2048	m_encCABACTableIdx = pcSlice->getSliceType();
2049	}
2050	numBinsCoded = m_CABACWriter->getNumBins();
2051
2052	}
2053
2054	void EncSlice::calculateBoundingCtuTsAddrForSlice(uint32_t &startCtuTSAddrSlice, uint32_t &boundingCtuTSAddrSlice, bool &haveReachedTileBoundary,
2055	Picture* pcPic, const int sliceMode, const int sliceArgument)
2056	{
2057	Slice* pcSlice = pcPic->slices[getSliceSegmentIdx()];
2058	const TileMap& tileMap = *( pcPic->tileMap );
2059	const PPS &pps = *( pcSlice->getPPS() );
2060	const uint32_t numberOfCtusInFrame = pcPic->cs->pcv->sizeInCtus;
2061	boundingCtuTSAddrSlice=0;
2062	haveReachedTileBoundary=false;
2063
2064	switch (sliceMode)
2065	{
2066	case FIXED_NUMBER_OF_CTU:
2067	{
2068	uint32_t ctuAddrIncrement = sliceArgument;
2069	boundingCtuTSAddrSlice = ((startCtuTSAddrSlice + ctuAddrIncrement) < numberOfCtusInFrame) ? (startCtuTSAddrSlice + ctuAddrIncrement) : numberOfCtusInFrame;
2070	}
2071	break;
2072	case FIXED_NUMBER_OF_BYTES:
2073	boundingCtuTSAddrSlice = numberOfCtusInFrame; // This will be adjusted later if required.
2074	break;
2075	case FIXED_NUMBER_OF_TILES:
2076	{
2077	const uint32_t tileIdx = tileMap.getTileIdxMap( tileMap.getCtuTsToRsAddrMap(startCtuTSAddrSlice) );
2078	const uint32_t tileTotalCount = (pps.getNumTileColumnsMinus1()+1) * (pps.getNumTileRowsMinus1()+1);
2079	uint32_t ctuAddrIncrement = 0;
2080
2081	for(uint32_t tileIdxIncrement = 0; tileIdxIncrement < sliceArgument; tileIdxIncrement++)
2082	{
2083	if((tileIdx + tileIdxIncrement) < tileTotalCount)
2084	{
2085	uint32_t tileWidthInCtus = tileMap.tiles[tileIdx + tileIdxIncrement].getTileWidthInCtus();
2086	uint32_t tileHeightInCtus = tileMap.tiles[tileIdx + tileIdxIncrement].getTileHeightInCtus();
2087	ctuAddrIncrement += (tileWidthInCtus * tileHeightInCtus);
2088	}
2089	}
2090
2091	boundingCtuTSAddrSlice = ((startCtuTSAddrSlice + ctuAddrIncrement) < numberOfCtusInFrame) ? (startCtuTSAddrSlice + ctuAddrIncrement) : numberOfCtusInFrame;
2092	}
2093	break;
2094	default:
2095	boundingCtuTSAddrSlice = numberOfCtusInFrame;
2096	break;
2097	}
2098
2099	// Adjust for tiles and wavefronts.
2100	const bool wavefrontsAreEnabled = pps.getEntropyCodingSyncEnabledFlag();
2101
2102	if ((sliceMode == FIXED_NUMBER_OF_CTU \|\| sliceMode == FIXED_NUMBER_OF_BYTES) &&
2103	(pps.getNumTileRowsMinus1() > 0 \|\| pps.getNumTileColumnsMinus1() > 0))
2104	{
2105	const uint32_t ctuRsAddr = tileMap.getCtuTsToRsAddrMap(startCtuTSAddrSlice);
2106	const uint32_t startTileIdx = tileMap.getTileIdxMap(ctuRsAddr);
2107	const Tile& startingTile = tileMap.tiles[startTileIdx];
2108	const uint32_t tileStartTsAddr = tileMap.getCtuRsToTsAddrMap(startingTile.getFirstCtuRsAddr());
2109	const uint32_t tileStartWidth = startingTile.getTileWidthInCtus();
2110	const uint32_t tileStartHeight = startingTile.getTileHeightInCtus();
2111	const uint32_t tileLastTsAddr_excl = tileStartTsAddr + tileStartWidth*tileStartHeight;
2112	const uint32_t tileBoundingCtuTsAddrSlice = tileLastTsAddr_excl;
2113	const uint32_t ctuColumnOfStartingTile = ((startCtuTSAddrSlice-tileStartTsAddr)%tileStartWidth);
2114	if (wavefrontsAreEnabled && ctuColumnOfStartingTile!=0)
2115	{
2116	// WPP: if a slice does not start at the beginning of a CTB row, it must end within the same CTB row
2117	const uint32_t numberOfCTUsToEndOfRow = tileStartWidth - ctuColumnOfStartingTile;
2118	const uint32_t wavefrontTileBoundingCtuAddrSlice = startCtuTSAddrSlice + numberOfCTUsToEndOfRow;
2119	if (wavefrontTileBoundingCtuAddrSlice < boundingCtuTSAddrSlice)
2120	{
2121	boundingCtuTSAddrSlice = wavefrontTileBoundingCtuAddrSlice;
2122	}
2123	}
2124
2125	if (tileBoundingCtuTsAddrSlice < boundingCtuTSAddrSlice)
2126	{
2127	boundingCtuTSAddrSlice = tileBoundingCtuTsAddrSlice;
2128	haveReachedTileBoundary = true;
2129	}
2130	}
2131	else if ((sliceMode == FIXED_NUMBER_OF_CTU \|\| sliceMode == FIXED_NUMBER_OF_BYTES) && wavefrontsAreEnabled && ((startCtuTSAddrSlice % pcPic->cs->pcv->widthInCtus) != 0))
2132	{
2133	// Adjust for wavefronts (no tiles).
2134	// WPP: if a slice does not start at the beginning of a CTB row, it must end within the same CTB row
2135	boundingCtuTSAddrSlice = std::min(boundingCtuTSAddrSlice, startCtuTSAddrSlice - (startCtuTSAddrSlice % pcPic->cs->pcv->widthInCtus) + (pcPic->cs->pcv->widthInCtus));
2136	}
2137	}
2138
2139	/** Determines the starting and bounding CTU address of current slice / dependent slice
2140	* \param [out] startCtuTsAddr
2141	* \param [out] boundingCtuTsAddr
2142	* \param [in] pcPic
2143
2144	* Updates startCtuTsAddr, boundingCtuTsAddr with appropriate CTU address
2145	*/
2146	void EncSlice::xDetermineStartAndBoundingCtuTsAddr ( uint32_t& startCtuTsAddr, uint32_t& boundingCtuTsAddr, Picture* pcPic )
2147	{
2148	Slice* pcSlice = pcPic->slices[getSliceSegmentIdx()];
2149
2150	// Non-dependent slice
2151	uint32_t startCtuTsAddrSlice = pcSlice->getSliceCurStartCtuTsAddr();
2152	bool haveReachedTileBoundarySlice = false;
2153	uint32_t boundingCtuTsAddrSlice;
2154	calculateBoundingCtuTsAddrForSlice(startCtuTsAddrSlice, boundingCtuTsAddrSlice, haveReachedTileBoundarySlice, pcPic,
2155	m_pcCfg->getSliceMode(), m_pcCfg->getSliceArgument());
2156	pcSlice->setSliceCurEndCtuTsAddr( boundingCtuTsAddrSlice );
2157	pcSlice->setSliceCurStartCtuTsAddr( startCtuTsAddrSlice );
2158
2159	#if HEVC_DEPENDENT_SLICES
2160	// Dependent slice
2161	uint32_t startCtuTsAddrSliceSegment = pcSlice->getSliceSegmentCurStartCtuTsAddr();
2162	bool haveReachedTileBoundarySliceSegment = false;
2163	uint32_t boundingCtuTsAddrSliceSegment;
2164	calculateBoundingCtuTsAddrForSlice(startCtuTsAddrSliceSegment, boundingCtuTsAddrSliceSegment, haveReachedTileBoundarySliceSegment, pcPic,
2165	m_pcCfg->getSliceSegmentMode(), m_pcCfg->getSliceSegmentArgument());
2166	if (boundingCtuTsAddrSliceSegment>boundingCtuTsAddrSlice)
2167	{
2168	boundingCtuTsAddrSliceSegment = boundingCtuTsAddrSlice;
2169	}
2170	pcSlice->setSliceSegmentCurEndCtuTsAddr( boundingCtuTsAddrSliceSegment );
2171	pcSlice->setSliceSegmentCurStartCtuTsAddr(startCtuTsAddrSliceSegment);
2172
2173	// Make a joint decision based on reconstruction and dependent slice bounds
2174	startCtuTsAddr = std::max(startCtuTsAddrSlice, startCtuTsAddrSliceSegment);
2175	boundingCtuTsAddr = boundingCtuTsAddrSliceSegment;
2176	#else
2177	startCtuTsAddr = startCtuTsAddrSlice;
2178	boundingCtuTsAddr = boundingCtuTsAddrSlice;
2179	#endif
2180	}
2181
2182	double EncSlice::xGetQPValueAccordingToLambda ( double lambda )
2183	{
2184	return 4.2005*log(lambda) + 13.7122;
2185	}
2186
2187	//! \}

Download in other formats:

Original format

JVET VVC

Context navigation

Ticket #266: EncSlice.cpp

Download in other formats: