x265探索与研究（九）：compressFrame()函数

compressFrame()函数是一个功能繁杂且分析难度较大的函数，主要包括时间戳的初始化工作、access unit的设计、加权预测技术、运动参考帧的估计、当前Slice的QP值确定、熵编码相关信息配置、并行计算与否及其空间的申请、SEI相关配置、线程控制、CTU分析、Multi-pass Encoding、滤波与去噪处理等等，其中最重要的就是调用了encodeSlice()函数。

compressFrame()函数中调用的主要函数如下图所示：

下面给出compressFrame()函数的代码分析：

/*=============================================================*/
/*====== Analysed by: RuiDong Fang ====== Csdn Blog:     http://blog.csdn.net/frd2009041510 ====== Date:       2016.04.18====== Funtion：  compressFrame()函数,编码一帧。*/
/*=============================================================*/
void FrameEncoder::compressFrame()
{///1、初始化一些变量ProfileScopeEvent(frameThread);    //帧线程的档次范围m_startCompressTime = x265_mdate();  //编码的开始时间戳(timestamp时间戳 when frame encoder is given a frame)m_totalActiveWorkerCount = 0;  //统计m_activeWorkerCount的和，即经过CTU压缩后统计，进入帧前初始化为0（sum of m_activeWorkerCount sampled at end of each CTU）m_activeWorkerCountSamples = 0;  //当前帧已经编码分析完毕的CTU个数（count of times m_activeWorkerCount was sampled (think vbv restarts)）m_totalWorkerElapsedTime = 0;    //初始化所有CTU编码滤波占用的时间（total elapsed time spent by worker threads processing CTUs）m_totalNoWorkerTime = 0;  //初始化当前帧编码占用的时间（total elapsed time without any active worker threads）m_countRowBlocks = 0;   //正在运行的CTU行因为上一行没有完成完毕而强制退出的个数，在帧编码前初始化为0（count of workers forced to abandon a row because of top dependency）m_allRowsAvailableTime = 0;  //初始化当前帧所有CTU行准备好的时间点（timestamp when all reference dependencies are resolved）m_stallStartTime = 0;   //初始化正在进行编码的rows个数为0时的时间点（timestamp when worker count becomes 0）m_completionCount = 0;m_bAllRowsStop = false;   //是否将所有CTU的编码停止,在每帧进入前初始化为false,在CTU编码决策中需要重新编码时将置为truem_vbvResetTriggerRow = -1;  //需要重新编码的CTU行号,每帧开始编码前初始化为-1m_SSDY = m_SSDU = m_SSDV = 0;m_ssim = 0;m_ssimCnt = 0;memset(&(m_frame->m_encData->m_frameStats), 0, sizeof(m_frame->m_encData->m_frameStats));    //将当前帧的统计信息初始化为0///2、存取单元access unit/* Emit access unit delimiter unless this is the first frame and the user is* not repeating headers (since AUD is supposed to be the first NAL in the access* unit) *///当是第一帧时，需要给出一个定界符Slice* slice = m_frame->m_encData->m_slice;   //获取当前sliceif (m_param->bEnableAccessUnitDelimiters && (m_frame->m_poc || m_param->bRepeatHeaders)){m_bs.resetBits();m_entropyCoder.setBitstream(&m_bs);m_entropyCoder.codeAUD(*slice);m_bs.writeByteAlignment();m_nalList.serialize(NAL_UNIT_ACCESS_UNIT_DELIMITER, m_bs);}if (m_frame->m_lowres.bKeyframe && m_param->bRepeatHeaders)m_top->getStreamHeaders(m_nalList, m_entropyCoder, m_bs);  //====================get stream headers///3、加权预测// Weighted Prediction parameters estimation.bool bUseWeightP = slice->m_sliceType == P_SLICE && slice->m_pps->bUseWeightPred; //P帧加权预测标志位bool bUseWeightB = slice->m_sliceType == B_SLICE && slice->m_pps->bUseWeightedBiPred;    //B帧加权预测标志位//加权预测的参数估计if (bUseWeightP || bUseWeightB)   //使能加权预测{
#if DETAILED_CU_STATSm_cuStats.countWeightAnalyze++;ScopedElapsedTime time(m_cuStats.weightAnalyzeTime);
#endifWeightAnalysis wa(*this); //用于多线程加权分析if (m_pool && wa.tryBondPeers(*this, 1)) //从当前job中拥有核并且sleep状态的核可以触发多线程，如果没有可用核则在当前线程中完成进入else/* use an idle worker for weight analysis */wa.waitForExit();   //一直等待到任务全部完成，这里等待的是核释放，内核释放了任务也就完成了elseweightAnalyse(*slice, *m_frame, *m_param);    //====================加权预测分析(每个list的第一帧分析加权与否，其它不加权)}elseslice->disableWeights(); //不使能加权预测///4、运动参考，配置参考帧信息// Generate motion referencesint numPredDir = slice->isInterP() ? 1 : slice->isInterB() ? 2 : 0;    //预测方向的数量，若为P帧则数量为1，若为B帧则数量为2，否则为0，即获取当前有几个listfor (int l = 0; l < numPredDir; l++)   //根据预测方向的数量遍历list的个数{for (int ref = 0; ref < slice->m_numRefIdx[l]; ref++) //遍历当前list的所有参考帧{WeightParam *w = NULL;if ((bUseWeightP || bUseWeightB) && slice->m_weightPredTable[l][ref][0].bPresentFlag)    //如果是P帧或B帧，并且bPresentFlag为1w = slice->m_weightPredTable[l][ref]; //获取加权参数m_mref[l][ref].init(slice->m_refPicList[l][ref]->m_reconPic, w, *m_param);    //获取参考帧信息，申请加权帧内存}}///5、根据码率控制确定QP的大小/* Get the QP for this frame from rate control. This call may block until* frames ahead of it in encode order have called rateControlEnd() */int qp = m_top->m_rateControl->rateControlStart(m_frame, &m_rce, m_top);    //====================rateControlStart()函数，即码率控制开始m_rce.newQp = qp; //获取当前估计的量化参数/* Clip slice QP to 0-51 spec range before encoding */slice->m_sliceQp = x265_clip3(-QP_BD_OFFSET, QP_MAX_SPEC, qp);   //编码前，根据Offset确定/修正当前Slice的具体QP，取值必须处在0~51内///6、熵编码前的一些准备m_initSliceContext.resetEntropy(*slice); //====================重置熵编码相关信息m_frameFilter.start(m_frame, m_initSliceContext, qp);    //====================滤波，在后面的调用中若bEnableSAO使能，则SAO/* ensure all rows are blocked prior to initializing row CTU counters */WaveFront::clearEnabledRowMask();   //将当前WPPmap全部初始化为不可执行/* reset entropy coders */m_entropyCoder.load(m_initSliceContext);for (uint32_t i = 0; i < m_numRows; i++)m_rows[i].init(m_initSliceContext);///7、并行与否，申请空间uint32_t numSubstreams = m_param->bEnableWavefront ? slice->m_sps->numCuInHeight : 1;    //一帧并行的流数（CTU行数）//根据并行的流数申请相关空间if (!m_outStreams) {m_outStreams = new Bitstream[numSubstreams];  //申请空间m_substreamSizes = X265_MALLOC(uint32_t, numSubstreams); //申请空间if (!m_param->bEnableSAO)  //若没有使能SAOfor (uint32_t i = 0; i < numSubstreams; i++)m_rows[i].rowGoOnCoder.setBitstream(&m_outStreams[i]);}elsefor (uint32_t i = 0; i < numSubstreams; i++)m_outStreams[i].resetBits();///8、SEI（Supplemental Enhancement Information）相关配置int prevBPSEI = m_rce.encodeOrder ? m_top->m_lastBPSEI : 0;if (m_frame->m_lowres.bKeyframe){if (m_param->bEmitHRDSEI){SEIBufferingPeriod* bpSei = &m_top->m_rateControl->m_bufPeriodSEI;// since the temporal layer HRD is not ready, we assumed it is fixedbpSei->m_auCpbRemovalDelayDelta = 1;bpSei->m_cpbDelayOffset = 0;bpSei->m_dpbDelayOffset = 0;// hrdFullness() calculates the initial CPB removal delay and offsetm_top->m_rateControl->hrdFullness(bpSei);m_bs.resetBits();bpSei->write(m_bs, *slice->m_sps);m_bs.writeByteAlignment();m_nalList.serialize(NAL_UNIT_PREFIX_SEI, m_bs);m_top->m_lastBPSEI = m_rce.encodeOrder;}}if (m_param->bEmitHRDSEI || !!m_param->interlaceMode){SEIPictureTiming *sei = m_rce.picTimingSEI;const VUI *vui = &slice->m_sps->vuiParameters;const HRDInfo *hrd = &vui->hrdParameters;int poc = slice->m_poc;if (vui->frameFieldInfoPresentFlag){if (m_param->interlaceMode == 2)sei->m_picStruct = (poc & 1) ? 1 /* top */ : 2 /* bottom */;else if (m_param->interlaceMode == 1)sei->m_picStruct = (poc & 1) ? 2 /* bottom */ : 1 /* top */;elsesei->m_picStruct = 0;sei->m_sourceScanType = 0;sei->m_duplicateFlag = false;}if (vui->hrdParametersPresentFlag){// The m_aucpbremoval delay specifies how many clock ticks the// access unit associated with the picture timing SEI message has to// wait after removal of the access unit with the most recent// buffering period SEI messagesei->m_auCpbRemovalDelay = X265_MIN(X265_MAX(1, m_rce.encodeOrder - prevBPSEI), (1 << hrd->cpbRemovalDelayLength));sei->m_picDpbOutputDelay = slice->m_sps->numReorderPics + poc - m_rce.encodeOrder;}m_bs.resetBits();sei->write(m_bs, *slice->m_sps);m_bs.writeByteAlignment();m_nalList.serialize(NAL_UNIT_PREFIX_SEI, m_bs);}///9、配合RC的线程控制/* CQP and CRF (without capped VBV) doesn't use mid-frame statistics to * tune RateControl parameters for other frames.* Hence, for these modes, update m_startEndOrder and unlock RC for previous threads waiting in* RateControlEnd here, after the slicecontexts are initialized. For the rest - ABR* and VBV, unlock only after rateControlUpdateStats of this frame is called *///因为m_startEndOrder在rateControlUpdateStats中只对ABR或者VBV模式更新,在此更新为了配合RC的线程控制if (m_param->rc.rateControlMode != X265_RC_ABR && !m_top->m_rateControl->m_isVbv){m_top->m_rateControl->m_startEndOrder.incr(); //更新计数if (m_rce.encodeOrder < m_param->frameNumThreads - 1)   //刚启动时多更新一次m_top->m_rateControl->m_startEndOrder.incr(); // faked rateControlEnd calls for negative frames}//================================================================================================================大部分的计算处理都在此处之下///10、CTU分析（分两种情况：WPP使能、WPP不使能）/* Analyze CTU rows, most of the hard work is done here.  Frame is* compressed in a wave-front pattern if WPP is enabled. Row based loop* filters runs behind the CTU compression and reconstruction */m_rows[0].active = true;   //触发第一个CTU行if (m_param->bEnableWavefront)    //如果WPP使能{for (uint32_t row = 0; row < m_numRows; row++)  //遍历所有CTU行{// block until all reference frames have reconstructed the rows we needfor (int l = 0; l < numPredDir; l++)    //当前list的个数{for (int ref = 0; ref < slice->m_numRefIdx[l]; ref++)  //当前list 中ref的个数{Frame *refpic = slice->m_refPicList[l][ref];   //获取参考帧uint32_t reconRowCount = refpic->m_reconRowCount.get();while ((reconRowCount != m_numRows) && (reconRowCount < row + m_refLagRows))reconRowCount = refpic->m_reconRowCount.waitForChange(reconRowCount);if ((bUseWeightP || bUseWeightB) && m_mref[l][ref].isWeighted)m_mref[l][ref].applyWeight(row + m_refLagRows, m_numRows);}}//当前外部参考块（如参考帧对应的参考块）准备完毕,将当前row对应位置的map置为1,标记可以执行enableRowEncoder(row); //===========================/* clear external dependency for this row */if (!row) //如果是第一行{m_row0WaitTime = x265_mdate();    //获取当前帧开始编码的的时间点,用于计算当前帧的编码时间//当前内部参考块准备完毕,将当前row对应位置的map置为1,标记可以执行enqueueRowEncoder(0); //===========================/* clear internal dependency, start wavefront */}tryWakeOne(); //CTU行准备好并触发wpp， 在findjob中运行}m_allRowsAvailableTime = x265_mdate(); //当前帧所有CTU行准备好的时间点tryWakeOne(); //多触发一次/* ensure one thread is active or help-wanted flag is set prior to blocking */static const int block_ms = 250;  //超时时间//每250ms触发一次,保证全部CTU行都能够执行（如果m_completionEvent在某一位置触发，则会造成不超时，循环退出）while (m_completionEvent.timedWait(block_ms))tryWakeOne(); //触发}else   //如果WPP不使能{for (uint32_t i = 0; i < m_numRows + m_filterRowDelay; i++){// compressif (i < m_numRows){// block until all reference frames have reconstructed the rows we needfor (int l = 0; l < numPredDir; l++){int list = l;for (int ref = 0; ref < slice->m_numRefIdx[list]; ref++){Frame *refpic = slice->m_refPicList[list][ref];    //获取参考帧uint32_t reconRowCount = refpic->m_reconRowCount.get();while ((reconRowCount != m_numRows) && (reconRowCount < i + m_refLagRows))reconRowCount = refpic->m_reconRowCount.waitForChange(reconRowCount);if ((bUseWeightP || bUseWeightB) && m_mref[l][ref].isWeighted)m_mref[list][ref].applyWeight(i + m_refLagRows, m_numRows);}}if (!i)m_row0WaitTime = x265_mdate();else if (i == m_numRows - 1)m_allRowsAvailableTime = x265_mdate();processRowEncoder(i, m_tld[m_localTldIdx]);    //===========================Called by worker threads}// filterif (i >= m_filterRowDelay)m_frameFilter.processRow(i - m_filterRowDelay); //===========================滤波每一行(Row based loop filters)}}///11、Multi-pass encodingif (m_param->rc.bStatWrite)  //Multi-pass encoding(Enable writing the stats in a multi-pass encode to the stat output file){int totalI = 0, totalP = 0, totalSkip = 0;// accumulate intra,inter,skip cu count per frame for 2 passfor (uint32_t i = 0; i < m_numRows; i++){m_frame->m_encData->m_frameStats.mvBits    += m_rows[i].rowStats.mvBits;m_frame->m_encData->m_frameStats.coeffBits += m_rows[i].rowStats.coeffBits;m_frame->m_encData->m_frameStats.miscBits  += m_rows[i].rowStats.miscBits;totalI                                     += m_rows[i].rowStats.intra8x8Cnt;totalP                                     += m_rows[i].rowStats.inter8x8Cnt;totalSkip                                  += m_rows[i].rowStats.skip8x8Cnt;}int totalCuCount = totalI + totalP + totalSkip; //总CU数目m_frame->m_encData->m_frameStats.percent8x8Intra = (double)totalI / totalCuCount;m_frame->m_encData->m_frameStats.percent8x8Inter = (double)totalP / totalCuCount;m_frame->m_encData->m_frameStats.percent8x8Skip  = (double)totalSkip / totalCuCount;}for (uint32_t i = 0; i < m_numRows; i++){m_frame->m_encData->m_frameStats.cntIntraNxN      += m_rows[i].rowStats.cntIntraNxN;m_frame->m_encData->m_frameStats.totalCu          += m_rows[i].rowStats.totalCu;m_frame->m_encData->m_frameStats.totalCtu         += m_rows[i].rowStats.totalCtu;m_frame->m_encData->m_frameStats.lumaDistortion   += m_rows[i].rowStats.lumaDistortion;m_frame->m_encData->m_frameStats.chromaDistortion += m_rows[i].rowStats.chromaDistortion;m_frame->m_encData->m_frameStats.psyEnergy        += m_rows[i].rowStats.psyEnergy;m_frame->m_encData->m_frameStats.lumaLevel        += m_rows[i].rowStats.lumaLevel;if (m_rows[i].rowStats.maxLumaLevel > m_frame->m_encData->m_frameStats.maxLumaLevel)m_frame->m_encData->m_frameStats.maxLumaLevel = m_rows[i].rowStats.maxLumaLevel;for (uint32_t depth = 0; depth <= g_maxCUDepth; depth++){m_frame->m_encData->m_frameStats.cntSkipCu[depth] += m_rows[i].rowStats.cntSkipCu[depth];m_frame->m_encData->m_frameStats.cntMergeCu[depth] += m_rows[i].rowStats.cntMergeCu[depth];for (int m = 0; m < INTER_MODES; m++)m_frame->m_encData->m_frameStats.cuInterDistribution[depth][m] += m_rows[i].rowStats.cuInterDistribution[depth][m];for (int n = 0; n < INTRA_MODES; n++)m_frame->m_encData->m_frameStats.cuIntraDistribution[depth][n] += m_rows[i].rowStats.cuIntraDistribution[depth][n];}}m_frame->m_encData->m_frameStats.avgLumaDistortion   = (double)(m_frame->m_encData->m_frameStats.lumaDistortion) / m_frame->m_encData->m_frameStats.totalCtu;m_frame->m_encData->m_frameStats.avgChromaDistortion = (double)(m_frame->m_encData->m_frameStats.chromaDistortion) / m_frame->m_encData->m_frameStats.totalCtu;m_frame->m_encData->m_frameStats.avgPsyEnergy        = (double)(m_frame->m_encData->m_frameStats.psyEnergy) / m_frame->m_encData->m_frameStats.totalCtu;m_frame->m_encData->m_frameStats.avgLumaLevel        = m_frame->m_encData->m_frameStats.lumaLevel / m_frame->m_encData->m_frameStats.totalCtu;m_frame->m_encData->m_frameStats.percentIntraNxN     = (double)(m_frame->m_encData->m_frameStats.cntIntraNxN * 100) / m_frame->m_encData->m_frameStats.totalCu;for (uint32_t depth = 0; depth <= g_maxCUDepth; depth++){m_frame->m_encData->m_frameStats.percentSkipCu[depth]  = (double)(m_frame->m_encData->m_frameStats.cntSkipCu[depth] * 100) / m_frame->m_encData->m_frameStats.totalCu;m_frame->m_encData->m_frameStats.percentMergeCu[depth] = (double)(m_frame->m_encData->m_frameStats.cntMergeCu[depth] * 100) / m_frame->m_encData->m_frameStats.totalCu;for (int n = 0; n < INTRA_MODES; n++)m_frame->m_encData->m_frameStats.percentIntraDistribution[depth][n] = (double)(m_frame->m_encData->m_frameStats.cuIntraDistribution[depth][n] * 100) / m_frame->m_encData->m_frameStats.totalCu;uint64_t cuInterRectCnt = 0; // sum of Nx2N, 2NxN countscuInterRectCnt += m_frame->m_encData->m_frameStats.cuInterDistribution[depth][1] + m_frame->m_encData->m_frameStats.cuInterDistribution[depth][2];m_frame->m_encData->m_frameStats.percentInterDistribution[depth][0] = (double)(m_frame->m_encData->m_frameStats.cuInterDistribution[depth][0] * 100) / m_frame->m_encData->m_frameStats.totalCu;m_frame->m_encData->m_frameStats.percentInterDistribution[depth][1] = (double)(cuInterRectCnt * 100) / m_frame->m_encData->m_frameStats.totalCu;m_frame->m_encData->m_frameStats.percentInterDistribution[depth][2] = (double)(m_frame->m_encData->m_frameStats.cuInterDistribution[depth][3] * 100) / m_frame->m_encData->m_frameStats.totalCu;}m_bs.resetBits();m_entropyCoder.load(m_initSliceContext);m_entropyCoder.setBitstream(&m_bs);m_entropyCoder.codeSliceHeader(*slice, *m_frame->m_encData); //===========================codeSliceHeader()函数///12、SAO，并调用encodeSlice()// finish encode of each CTU row, only required when SAO is enabledif (m_param->bEnableSAO)encodeSlice();    //===========================调用encodeSlice()// serialize each row, record final lengths in slice headeruint32_t maxStreamSize = m_nalList.serializeSubstreams(m_substreamSizes, numSubstreams, m_outStreams);// complete the slice header by writing WPP row-startsm_entropyCoder.setBitstream(&m_bs);if (slice->m_pps->bEntropyCodingSyncEnabled)m_entropyCoder.codeSliceHeaderWPPEntryPoints(*slice, m_substreamSizes, maxStreamSize);m_bs.writeByteAlignment();m_nalList.serialize(slice->m_nalUnitType, m_bs);if (m_param->decodedPictureHashSEI){if (m_param->decodedPictureHashSEI == 1){m_seiReconPictureDigest.m_method = SEIDecodedPictureHash::MD5;for (int i = 0; i < 3; i++)MD5Final(&m_state[i], m_seiReconPictureDigest.m_digest[i]);}else if (m_param->decodedPictureHashSEI == 2){m_seiReconPictureDigest.m_method = SEIDecodedPictureHash::CRC;for (int i = 0; i < 3; i++)crcFinish(m_crc[i], m_seiReconPictureDigest.m_digest[i]);}else if (m_param->decodedPictureHashSEI == 3){m_seiReconPictureDigest.m_method = SEIDecodedPictureHash::CHECKSUM;for (int i = 0; i < 3; i++)checksumFinish(m_checksum[i], m_seiReconPictureDigest.m_digest[i]);}m_bs.resetBits();m_seiReconPictureDigest.write(m_bs, *slice->m_sps);m_bs.writeByteAlignment();m_nalList.serialize(NAL_UNIT_SUFFIX_SEI, m_bs);}uint64_t bytes = 0;for (uint32_t i = 0; i < m_nalList.m_numNal; i++){int type = m_nalList.m_nal[i].type;// exclude SEIif (type != NAL_UNIT_PREFIX_SEI && type != NAL_UNIT_SUFFIX_SEI){bytes += m_nalList.m_nal[i].sizeBytes;// and exclude start code prefixbytes -= (!i || type == NAL_UNIT_SPS || type == NAL_UNIT_PPS) ? 4 : 3;}}m_accessUnitBits = bytes << 3;m_endCompressTime = x265_mdate();  //timestamp after all CTUs are compressed
//  printf("编码时间(Compressinging Time): %d ms\n",m_endCompressTime-m_startCompressTime);   //added by Fred/* rateControlEnd may also block for earlier frames to call rateControlUpdateStats */if (m_top->m_rateControl->rateControlEnd(m_frame, m_accessUnitBits, &m_rce) < 0)m_top->m_aborted = true;   //一般不进入,错误返回/* Decrement referenced frame reference counts, allow them to be recycled */for (int l = 0; l < numPredDir; l++){for (int ref = 0; ref < slice->m_numRefIdx[l]; ref++)   //遍历每个List的参考帧{Frame *refpic = slice->m_refPicList[l][ref]; //获取参考帧地址ATOMIC_DEC(&refpic->m_countRefEncoders);    //被参考计数减一}}///13、去噪处理NR updates（Noise Reduction）int numTLD;   //TLD=Thread Local Data当前并行线程个数if (m_pool)numTLD = m_param->bEnableWavefront ? m_pool->m_numWorkers : m_pool->m_numWorkers + m_pool->m_numProviders; //应用WPP，为当前机器的核数，否则为当前核数加上当前的线程个数elsenumTLD = 1; //不应用多线程，并行个数为1if (m_nr) //如果noise reduction(更新去噪偏移值){/* Accumulate NR statistics from all worker threads *///加速所有线程的NR统计for (int i = 0; i < numTLD; i++)  //遍历当前帧编码应用所有线程：累加当前帧各个系数的统计数字{//获取当前线程对应当前帧的去噪类；i 确定当前帧的所有线程，如每个WPP行  m_jpId 确定当前帧NoiseReduction* nr = &m_tld[i].analysis.m_quant.m_frameNr[m_jpId]; for (int cat = 0; cat < MAX_NUM_TR_CATEGORIES; cat++) //遍历所有TU类别{for (int coeff = 0; coeff < MAX_NUM_TR_COEFFS; coeff++)m_nr->residualSum[cat][coeff] += nr->residualSum[cat][coeff];   //累加当前帧所有TU对应位置的系数绝对值和m_nr->count[cat] += nr->count[cat];   //累加当前帧的TU计数}}noiseReductionUpdate();   //===========================DCT-domain noise reduction / adaptive deadzone from libavcodec/* Copy updated NR coefficients back to all worker threads *///拷贝NR系数至所有的工作线程for (int i = 0; i < numTLD; i++)   //遍历当前帧编码应用所有线程{//获取当前线程对应当前帧的去噪类；i 确定当前帧的所有线程，如每个WPP行  m_jpId 确定当前帧NoiseReduction* nr = &m_tld[i].analysis.m_quant.m_frameNr[m_jpId];//将当前更新的去噪偏移值拷贝到各个线程中的去噪偏移中去memcpy(nr->offsetDenoise, m_nr->offsetDenoise, sizeof(uint16_t) * MAX_NUM_TR_CATEGORIES * MAX_NUM_TR_COEFFS);memset(nr->count, 0, sizeof(uint32_t) * MAX_NUM_TR_CATEGORIES);  //初始为0memset(nr->residualSum, 0, sizeof(uint32_t) * MAX_NUM_TR_CATEGORIES * MAX_NUM_TR_COEFFS);  //初始为0}}#if DETAILED_CU_STATS   //统计数据/* Accumulate CU statistics from each worker thread, we could report* per-frame stats here, but currently we do not. */for (int i = 0; i < numTLD; i++)m_cuStats.accumulate(m_tld[i].analysis.m_stats[m_jpId]);
#endifm_endFrameTime = x265_mdate();   //帧编码的结束时间(timestamp after RCEnd, NR updates, etc)}

x265探索与研究（九）：compressFrame()函数相关推荐

x265探索与研究（一）：x265下载安装与配置
x265下载安装与配置研究了这么久的HEVC Test Model(HM),相信大家对x265开源代码的实现与框架早就充满了好奇,接下来的日子,我将把自己入手学习与探索"x265开源代码的 ...
x265探索与研究（三）：如何播放*.265格式的视频或解码视频流
如何播放*.265格式的视频或解码视频流如下图,在得到.265格式的视频或视频流后应如何播放265格式的视频呢?本博文总结出5种播放265格式视频或视频流的基本方法. 方式一:Elecard HEV ...
x265探索与研究（五）：如何用VS调试x265？
如何用VS调试x265? 1.设置cli为启动项用VS打开工程项目,如下图: 右击cli,设置为启动项,如下图: 2.配置路径和命令行参数右击cli选择Properties,如下图所示进入如下界 ...
会计专业应用计算机操作,基于中职会计专业《计算机应用基础》课程的教学探索与研究.doc...
基于中职会计专业<计算机应用基础>课程的教学探索与研究基于中职会计专业<计算机应用基础>课程的教学探索与研究谭仔群桂林市财贸管理干部中等专业学校 X 关注成功! 加关注后 ...
基于wasm的探索与研究（一）
WebAssembly初探本次分享的文章是基于WebAssembly的探索与研究.最近需要做一个与加密相关的项目,想将后端的加密方案直接放到前端使用,好处是加密方案代码只用维护一套,且后端方案更贴近 ...
DL之DNN：基于Tensorflow框架对神经网络算法进行参数初始化的常用九大函数及其使用案例
DL之DNN:基于Tensorflow框架对神经网络算法进行参数初始化的常用九大函数及其使用案例目录基于Tensorflow框架对神经网络算法进行初始化的常用函数及其使用案例 1.初始化的常用函数
深度学习（31）随机梯度下降九: Himmelblau函数优化实战
深度学习(31)随机梯度下降九: Himmelblau函数优化实战 1. Himmelblau函数 2. 函数优化实战 1. Himmelblau函数 Himmelblau函数是用来测试后话算法的常用 ...
计算机课平时成绩重要吗,离散数学课程平时成绩评定方法的探索与研究
林静史册摘要:离散数学是计算机科学与技术专业重要的核心基础课程.目前,不少大学离散数学课程的最终成绩是依据平时成绩和期终考试成绩两个部分综合评定的.为了提高课程评价的公正性和提升教学效果,文章對该 ...
java写三次函数导数,用导数研究三次函数
<用导数研究三次函数>由会员分享,可在线阅读,更多相关<用导数研究三次函数(14页珍藏版)>请在人人文库网上搜索. 1.用导数研究三次函数1. 知识点解析1.定义:定义1.形如 ...

x265探索与研究（九）：compressFrame()函数

x265探索与研究（九）：compressFrame()函数

x265探索与研究（九）：compressFrame()函数相关推荐

最新文章

热门文章