[deb_x265.git] / source / encoder / frameencoder.cpp

/*****************************************************************************
 * Copyright (C) 2013 x265 project
 *
 * Authors: Chung Shin Yee <shinyee@multicorewareinc.com>
 *          Min Chen <chenm003@163.com>
 *          Steve Borho <steve@borho.org>
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation; either version 2 of the License, or
 * (at your option) any later version.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with this program; if not, write to the Free Software
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02111, USA.
 *
 * This program is also available under a commercial proprietary license.
 * For more information, contact us at license @ x265.com.
 *****************************************************************************/

#include "common.h"
#include "frame.h"
#include "framedata.h"
#include "wavefront.h"
#include "param.h"

#include "encoder.h"
#include "frameencoder.h"
#include "common.h"
#include "slicetype.h"
#include "nal.h"

namespace x265 {
void weightAnalyse(Slice& slice, Frame& frame, x265_param& param);

FrameEncoder::FrameEncoder()
    : WaveFront(NULL)
    , m_threadActive(true)
{
    m_totalTime = 0;
    m_frameEncoderID = 0;
    m_bAllRowsStop = false;
    m_vbvResetTriggerRow = -1;
    m_outStreams = NULL;
    m_substreamSizes = NULL;
    m_nr = NULL;
    m_tld = NULL;
    m_rows = NULL;
    m_top = NULL;
    m_param = NULL;
    m_frame = NULL;
    m_cuGeoms = NULL;
    m_ctuGeomMap = NULL;
    memset(&m_frameStats, 0, sizeof(m_frameStats));
    memset(&m_rce, 0, sizeof(RateControlEntry));
}

void FrameEncoder::destroy()
{
    if (m_pool)
        JobProvider::flush();  // ensure no worker threads are using this frame

    m_threadActive = false;
    m_enable.trigger();

    delete[] m_rows;
    delete[] m_outStreams;
    X265_FREE(m_cuGeoms);
    X265_FREE(m_ctuGeomMap);
    X265_FREE(m_substreamSizes);
    X265_FREE(m_nr);

    m_frameFilter.destroy();

    if (m_param->bEmitHRDSEI || !!m_param->interlaceMode)
    {
        delete m_rce.picTimingSEI;
        delete m_rce.hrdTiming;
    }

    // wait for worker thread to exit
    stop();
}

bool FrameEncoder::init(Encoder *top, int numRows, int numCols, int id)
{
    m_top = top;
    m_param = top->m_param;
    m_numRows = numRows;
    m_numCols = numCols;
    m_filterRowDelay = (m_param->bEnableSAO && m_param->bSaoNonDeblocked) ?
                        2 : (m_param->bEnableSAO || m_param->bEnableLoopFilter ? 1 : 0);
    m_filterRowDelayCus = m_filterRowDelay * numCols;
    m_frameEncoderID = id;
    m_rows = new CTURow[m_numRows];
    bool ok = !!m_numRows;

    int range  = m_param->searchRange; /* fpel search */
        range += 1;                    /* diamond search range check lag */
        range += 2;                    /* subpel refine */
        range += NTAPS_LUMA / 2;       /* subpel filter half-length */
    m_refLagRows = 1 + ((range + g_maxCUSize - 1) / g_maxCUSize);

    // NOTE: 2 times of numRows because both Encoder and Filter in same queue
    if (!WaveFront::init(m_numRows * 2))
    {
        x265_log(m_param, X265_LOG_ERROR, "unable to initialize wavefront queue\n");
        m_pool = NULL;
    }

    m_frameFilter.init(top, this, numRows);

    // initialize HRD parameters of SPS
    if (m_param->bEmitHRDSEI || !!m_param->interlaceMode)
    {
        m_rce.picTimingSEI = new SEIPictureTiming;
        m_rce.hrdTiming = new HRDTiming;

        ok &= m_rce.picTimingSEI && m_rce.hrdTiming;
    }

    if (m_param->noiseReductionIntra || m_param->noiseReductionInter)
        m_nr = X265_MALLOC(NoiseReduction, 1);
    if (m_nr)
        memset(m_nr, 0, sizeof(NoiseReduction));
    else
        m_param->noiseReductionIntra = m_param->noiseReductionInter = 0;

    start();
    return ok;
}

/* Generate a complete list of unique geom sets for the current picture dimensions */
bool FrameEncoder::initializeGeoms()
{
    /* Geoms only vary between CTUs in the presence of picture edges */
    int maxCUSize = m_param->maxCUSize;
    int heightRem = m_param->sourceHeight & (maxCUSize - 1);
    int widthRem = m_param->sourceWidth & (maxCUSize - 1);
    int allocGeoms = 1; // body
    if (heightRem && widthRem)
        allocGeoms = 4; // body, right, bottom, corner
    else if (heightRem || widthRem)
        allocGeoms = 2; // body, right or bottom

    m_ctuGeomMap = X265_MALLOC(uint32_t, m_numRows * m_numCols);
    m_cuGeoms = X265_MALLOC(CUGeom, allocGeoms * CUGeom::MAX_GEOMS);
    if (!m_cuGeoms || !m_ctuGeomMap)
        return false;

    // body
    CUData::calcCTUGeoms(maxCUSize, maxCUSize, maxCUSize, m_cuGeoms);
    memset(m_ctuGeomMap, 0, sizeof(uint32_t) * m_numRows * m_numCols);
    if (allocGeoms == 1)
        return true;

    int countGeoms = 1;
    if (widthRem)
    {
        // right
        CUData::calcCTUGeoms(widthRem, maxCUSize, maxCUSize, m_cuGeoms + countGeoms * CUGeom::MAX_GEOMS);
        for (int i = 0; i < m_numRows; i++)
        {
            uint32_t ctuAddr = m_numCols * (i + 1) - 1;
            m_ctuGeomMap[ctuAddr] = countGeoms * CUGeom::MAX_GEOMS;
        }
        countGeoms++;
    }
    if (heightRem)
    {
        // bottom
        CUData::calcCTUGeoms(maxCUSize, heightRem, maxCUSize, m_cuGeoms + countGeoms * CUGeom::MAX_GEOMS);
        for (uint32_t i = 0; i < m_numCols; i++)
        {
            uint32_t ctuAddr = m_numCols * (m_numRows - 1) + i;
            m_ctuGeomMap[ctuAddr] = countGeoms * CUGeom::MAX_GEOMS;
        }
        countGeoms++;

        if (widthRem)
        {
            // corner
            CUData::calcCTUGeoms(widthRem, heightRem, maxCUSize, m_cuGeoms + countGeoms * CUGeom::MAX_GEOMS);

            uint32_t ctuAddr = m_numCols * m_numRows - 1;
            m_ctuGeomMap[ctuAddr] = countGeoms * CUGeom::MAX_GEOMS;
            countGeoms++;
        }
        X265_CHECK(countGeoms == allocGeoms, "geometry match check failure\n");
    }

    return true;
}

bool FrameEncoder::startCompressFrame(Frame* curFrame)
{
    m_frame = curFrame;
    curFrame->m_encData->m_frameEncoderID = m_frameEncoderID; // Each Frame knows the ID of the FrameEncoder encoding it
    curFrame->m_encData->m_slice->m_mref = m_mref;

    if (!m_cuGeoms)
    {
        if (!initializeGeoms())
            return false;
    }

    m_enable.trigger();
    return true;
}

void FrameEncoder::threadMain()
{
    // worker thread routine for FrameEncoder
    do
    {
        m_enable.wait(); // Encoder::encode() triggers this event
        if (m_threadActive)
        {
            compressFrame();
            m_done.trigger(); // FrameEncoder::getEncodedPicture() blocks for this event
        }
    }
    while (m_threadActive);
}

void FrameEncoder::compressFrame()
{
    //ProfileScopeEvent(frameThread);
    int64_t startCompressTime = x265_mdate();
    Slice* slice = m_frame->m_encData->m_slice;

    /* 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) */
    if (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);

    // Weighted Prediction parameters estimation.
    bool bUseWeightP = slice->m_sliceType == P_SLICE && slice->m_pps->bUseWeightPred;
    bool bUseWeightB = slice->m_sliceType == B_SLICE && slice->m_pps->bUseWeightedBiPred;
    if (bUseWeightP || bUseWeightB)
        weightAnalyse(*slice, *m_frame, *m_param);
    else
        slice->disableWeights();

    // Generate motion references
    int numPredDir = slice->isInterP() ? 1 : slice->isInterB() ? 2 : 0;
    for (int l = 0; l < numPredDir; l++)
    {
        for (int ref = 0; ref < slice->m_numRefIdx[l]; ref++)
        {
            WeightParam *w = NULL;
            if ((bUseWeightP || bUseWeightB) && slice->m_weightPredTable[l][ref][0].bPresentFlag)
                w = slice->m_weightPredTable[l][ref];
            m_mref[l][ref].init(slice->m_refPicList[l][ref]->m_reconPic, w, *m_param);
        }
    }

    /* 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);
    m_rce.newQp = qp;

    /* Clip slice QP to 0-51 spec range before encoding */
    slice->m_sliceQp = Clip3(-QP_BD_OFFSET, QP_MAX_SPEC, qp);

    m_initSliceContext.resetEntropy(*slice);

    m_frameFilter.start(m_frame, m_initSliceContext, qp);

    // reset entropy coders
    m_entropyCoder.load(m_initSliceContext);
    for (int i = 0; i < m_numRows; i++)
        m_rows[i].init(m_initSliceContext);

    uint32_t numSubstreams = m_param->bEnableWavefront ? slice->m_sps->numCuInHeight : 1;
    if (!m_outStreams)
    {
        m_outStreams = new Bitstream[numSubstreams];
        m_substreamSizes = X265_MALLOC(uint32_t, numSubstreams);
        if (!m_param->bEnableSAO)
            for (uint32_t i = 0; i < numSubstreams; i++)
                m_rows[i].rowGoOnCoder.setBitstream(&m_outStreams[i]);
    }
    else
        for (uint32_t i = 0; i < numSubstreams; i++)
            m_outStreams[i].resetBits();

    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 fixed
            bpSei->m_auCpbRemovalDelayDelta = 1;
            bpSei->m_cpbDelayOffset = 0;
            bpSei->m_dpbDelayOffset = 0;

            // hrdFullness() calculates the initial CPB removal delay and offset
            m_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;
        }

        // The recovery point SEI message assists a decoder in determining when the decoding
        // process will produce acceptable pictures for display after the decoder initiates
        // random access. The m_recoveryPocCnt is in units of POC(picture order count) which
        // means pictures encoded after the CRA but precede it in display order(leading) are
        // implicitly discarded after a random access seek regardless of the value of
        // m_recoveryPocCnt. Our encoder does not use references prior to the most recent CRA,
        // so all pictures following the CRA in POC order are guaranteed to be displayable,
        // so m_recoveryPocCnt is always 0.
        SEIRecoveryPoint sei_recovery_point;
        sei_recovery_point.m_recoveryPocCnt = 0;
        sei_recovery_point.m_exactMatchingFlag = true;
        sei_recovery_point.m_brokenLinkFlag = false;

        m_bs.resetBits();
        sei_recovery_point.write(m_bs, *slice->m_sps);
        m_bs.writeByteAlignment();

        m_nalList.serialize(NAL_UNIT_PREFIX_SEI, m_bs);
    }

    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 */;
            else
                sei->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 message
            sei->m_auCpbRemovalDelay = X265_MIN(X265_MAX(1, m_rce.encodeOrder - m_top->m_lastBPSEI), (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);
    }

    // Analyze CTU rows, most of the hard work is done here
    // frame is compressed in a wave-front pattern if WPP is enabled. Loop filter runs as a
    // wave-front behind the CU compression and reconstruction
    compressCTURows();

    if (m_param->rc.bStatWrite)
    {
        int totalI = 0, totalP = 0, totalSkip = 0;

        // accumulate intra,inter,skip cu count per frame for 2 pass
        for (int i = 0; i < m_numRows; i++)
        {
            m_frameStats.mvBits    += m_rows[i].rowStats.mvBits;
            m_frameStats.coeffBits += m_rows[i].rowStats.coeffBits;
            m_frameStats.miscBits  += m_rows[i].rowStats.miscBits;
            totalI                 += m_rows[i].rowStats.iCuCnt;
            totalP                 += m_rows[i].rowStats.pCuCnt;
            totalSkip              += m_rows[i].rowStats.skipCuCnt;
        }
        int totalCuCount = totalI + totalP + totalSkip;
        m_frameStats.percentIntra = (double)totalI / totalCuCount;
        m_frameStats.percentInter = (double)totalP / totalCuCount;
        m_frameStats.percentSkip  = (double)totalSkip / totalCuCount;
    }

    m_bs.resetBits();
    m_entropyCoder.load(m_initSliceContext);
    m_entropyCoder.setBitstream(&m_bs);
    m_entropyCoder.codeSliceHeader(*slice, *m_frame->m_encData);

    // finish encode of each CTU row, only required when SAO is enabled
    if (m_param->bEnableSAO)
        encodeSlice();

    // serialize each row, record final lengths in slice header
    uint32_t maxStreamSize = m_nalList.serializeSubstreams(m_substreamSizes, numSubstreams, m_outStreams);

    // complete the slice header by writing WPP row-starts
    m_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 SEI
        if (type != NAL_UNIT_PREFIX_SEI && type != NAL_UNIT_SUFFIX_SEI)
        {
            bytes += m_nalList.m_nal[i].sizeBytes;
            // and exclude start code prefix
            bytes -= (!i || type == NAL_UNIT_SPS || type == NAL_UNIT_PPS) ? 4 : 3;
        }
    }
    m_accessUnitBits = bytes << 3;

    m_elapsedCompressTime = (double)(x265_mdate() - startCompressTime) / 1000000;
    /* rateControlEnd may also block for earlier frames to call rateControlUpdateStats */
    if (m_top->m_rateControl->rateControlEnd(m_frame, m_accessUnitBits, &m_rce, &m_frameStats) < 0)
        m_top->m_aborted = true;

    /* Accumulate NR statistics from all worker threads */
    if (m_nr)
    {
        for (int i = 0; i < m_top->m_numThreadLocalData; i++)
        {
            NoiseReduction* nr = &m_top->m_threadLocalData[i].analysis.m_quant.m_frameNr[m_frameEncoderID];
            for (int cat = 0; cat < MAX_NUM_TR_CATEGORIES; cat++)
            {
                for(int coeff = 0; coeff < MAX_NUM_TR_COEFFS; coeff++)
                    m_nr->residualSum[cat][coeff] += nr->residualSum[cat][coeff];
            
                m_nr->count[cat] += nr->count[cat];
            }
        }
    }

    noiseReductionUpdate();

    /* Copy updated NR coefficients back to all worker threads */
    if (m_nr)
    {
        for (int i = 0; i < m_top->m_numThreadLocalData; i++)
        {
            NoiseReduction* nr = &m_top->m_threadLocalData[i].analysis.m_quant.m_frameNr[m_frameEncoderID];
            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);
            memset(nr->residualSum, 0, sizeof(uint32_t) * MAX_NUM_TR_CATEGORIES * MAX_NUM_TR_COEFFS);
        }
    }

    // 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++)
        {
            Frame *refpic = slice->m_refPicList[l][ref];
            ATOMIC_DEC(&refpic->m_countRefEncoders);
        }
    }
}

void FrameEncoder::encodeSlice()
{
    Slice* slice = m_frame->m_encData->m_slice;
    const uint32_t widthInLCUs = slice->m_sps->numCuInWidth;
    const uint32_t lastCUAddr = (slice->m_endCUAddr + NUM_CU_PARTITIONS - 1) / NUM_CU_PARTITIONS;
    const uint32_t numSubstreams = m_param->bEnableWavefront ? slice->m_sps->numCuInHeight : 1;

    SAOParam* saoParam = slice->m_sps->bUseSAO ? m_frame->m_encData->m_saoParam : NULL;
    for (uint32_t cuAddr = 0; cuAddr < lastCUAddr; cuAddr++)
    {
        uint32_t col = cuAddr % widthInLCUs;
        uint32_t lin = cuAddr / widthInLCUs;
        uint32_t subStrm = lin % numSubstreams;
        CUData* ctu = m_frame->m_encData->getPicCTU(cuAddr);

        m_entropyCoder.setBitstream(&m_outStreams[subStrm]);

        // Synchronize cabac probabilities with upper-right CTU if it's available and we're at the start of a line.
        if (m_param->bEnableWavefront && !col && lin)
        {
            m_entropyCoder.copyState(m_initSliceContext);
            m_entropyCoder.loadContexts(m_rows[lin - 1].bufferedEntropy);
        }

        if (saoParam)
        {
            if (saoParam->bSaoFlag[0] || saoParam->bSaoFlag[1])
            {
                int mergeLeft = col && saoParam->ctuParam[0][cuAddr].mergeMode == SAO_MERGE_LEFT;
                int mergeUp = lin && saoParam->ctuParam[0][cuAddr].mergeMode == SAO_MERGE_UP;
                if (col)
                    m_entropyCoder.codeSaoMerge(mergeLeft);
                if (lin && !mergeLeft)
                    m_entropyCoder.codeSaoMerge(mergeUp);
                if (!mergeLeft && !mergeUp)
                {
                    if (saoParam->bSaoFlag[0])
                        m_entropyCoder.codeSaoOffset(saoParam->ctuParam[0][cuAddr], 0);
                    if (saoParam->bSaoFlag[1])
                    {
                        m_entropyCoder.codeSaoOffset(saoParam->ctuParam[1][cuAddr], 1);
                        m_entropyCoder.codeSaoOffset(saoParam->ctuParam[2][cuAddr], 2);
                    }
                }
            }
            else
            {
                for (int i = 0; i < 3; i++)
                    saoParam->ctuParam[i][cuAddr].reset();
            }
        }

        // final coding (bitstream generation) for this CU
        m_entropyCoder.encodeCTU(*ctu, m_cuGeoms[m_ctuGeomMap[cuAddr]]);

        if (m_param->bEnableWavefront)
        {
            if (col == 1)
                // Store probabilities of second CTU in line into buffer
                m_rows[lin].bufferedEntropy.loadContexts(m_entropyCoder);

            if (col == widthInLCUs - 1)
                m_entropyCoder.finishSlice();
        }
    }
    if (!m_param->bEnableWavefront)
        m_entropyCoder.finishSlice();
}

void FrameEncoder::compressCTURows()
{
    Slice* slice = m_frame->m_encData->m_slice;

    m_bAllRowsStop = false;
    m_vbvResetTriggerRow = -1;

    m_SSDY = m_SSDU = m_SSDV = 0;
    m_ssim = 0;
    m_ssimCnt = 0;
    memset(&m_frameStats, 0, sizeof(m_frameStats));

    bool bUseWeightP = slice->m_pps->bUseWeightPred && slice->m_sliceType == P_SLICE;
    bool bUseWeightB = slice->m_pps->bUseWeightedBiPred && slice->m_sliceType == B_SLICE;
    int numPredDir = slice->isInterP() ? 1 : slice->isInterB() ? 2 : 0;

    m_rows[0].active = true;
    if (m_pool && m_param->bEnableWavefront)
    {
        WaveFront::clearEnabledRowMask();
        WaveFront::enqueue();

        for (int row = 0; row < m_numRows; row++)
        {
            // block until all reference frames have reconstructed the rows we need
            for (int l = 0; l < numPredDir; l++)
            {
                for (int ref = 0; ref < slice->m_numRefIdx[l]; ref++)
                {
                    Frame *refpic = slice->m_refPicList[l][ref];

                    int 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);
                }
            }

            enableRowEncoder(row);
            if (row == 0)
                enqueueRowEncoder(0);
            else
                m_pool->pokeIdleThread();
        }

        m_completionEvent.wait();

        WaveFront::dequeue();
    }
    else
    {
        for (int i = 0; i < this->m_numRows + m_filterRowDelay; i++)
        {
            // Encode
            if (i < m_numRows)
            {
                // block until all reference frames have reconstructed the rows we need
                for (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];

                        int 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);
                    }
                }

                processRowEncoder(i, *m_tld);
            }

            // Filter
            if (i >= m_filterRowDelay)
                m_frameFilter.processRow(i - m_filterRowDelay);
        }
    }
    m_frameTime = (double)m_totalTime / 1000000;
    m_totalTime = 0;
}

void FrameEncoder::processRow(int row, int threadId)
{
    const int realRow = row >> 1;
    const int typeNum = row & 1;

    ThreadLocalData& tld = threadId >= 0 ? m_top->m_threadLocalData[threadId] : *m_tld;
    
    if (!typeNum)
        processRowEncoder(realRow, tld);
    else
    {
        m_frameFilter.processRow(realRow);

        // NOTE: Active next row
        if (realRow != m_numRows - 1)
            enqueueRowFilter(realRow + 1);
        else
            m_completionEvent.trigger();
    }
}

// Called by worker threads
void FrameEncoder::processRowEncoder(int row, ThreadLocalData& tld)
{
    CTURow& curRow = m_rows[row];

    {
        ScopedLock self(curRow.lock);
        if (!curRow.active)
            /* VBV restart is in progress, exit out */
            return;
        if (curRow.busy)
        {
            /* On multi-socket Windows servers, we have seen problems with
             * ATOMIC_CAS which resulted in multiple worker threads processing
             * the same CU row, which often resulted in bad pointer accesses. We
             * believe the problem is fixed, but are leaving this check in place
             * to prevent crashes in case it is not */
            x265_log(m_param, X265_LOG_WARNING,
                     "internal error - simultaneous row access detected. Please report HW to x265-devel@videolan.org\n");
            return;
        }
        curRow.busy = true;
    }

    /* When WPP is enabled, every row has its own row coder instance. Otherwise
     * they share row 0 */
    Entropy& rowCoder = m_param->bEnableWavefront ? m_rows[row].rowGoOnCoder : m_rows[0].rowGoOnCoder;
    FrameData& curEncData = *m_frame->m_encData;
    Slice *slice = curEncData.m_slice;

    int64_t startTime = x265_mdate();
    const uint32_t numCols = m_numCols;
    const uint32_t lineStartCUAddr = row * numCols;
    bool bIsVbv = m_param->rc.vbvBufferSize > 0 && m_param->rc.vbvMaxBitrate > 0;

    while (curRow.completed < numCols)
    {
        ProfileScopeEvent(encodeCTU);

        int col = curRow.completed;
        const uint32_t cuAddr = lineStartCUAddr + col;
        CUData* ctu = curEncData.getPicCTU(cuAddr);
        ctu->initCTU(*m_frame, cuAddr, slice->m_sliceQp);

        if (bIsVbv)
        {
            if (!row)
            {
                curEncData.m_rowStat[row].diagQp = curEncData.m_avgQpRc;
                curEncData.m_rowStat[row].diagQpScale = x265_qp2qScale(curEncData.m_avgQpRc);
            }

            if (row >= col && row && m_vbvResetTriggerRow != row)
                curEncData.m_cuStat[cuAddr].baseQp = curEncData.m_cuStat[cuAddr - numCols + 1].baseQp;
            else
                curEncData.m_cuStat[cuAddr].baseQp = curEncData.m_rowStat[row].diagQp;
        }
        else
            curEncData.m_cuStat[cuAddr].baseQp = curEncData.m_avgQpRc;

        if (m_param->rc.aqMode || bIsVbv)
        {
            int qp = calcQpForCu(cuAddr, curEncData.m_cuStat[cuAddr].baseQp);
            tld.analysis.setQP(*slice, qp);
            qp = Clip3(QP_MIN, QP_MAX_SPEC, qp);
            ctu->setQPSubParts((int8_t)qp, 0, 0);
            curEncData.m_rowStat[row].sumQpAq += qp;
        }
        else
            tld.analysis.setQP(*slice, slice->m_sliceQp);

        if (m_param->bEnableWavefront && !col && row)
        {
            // Load SBAC coder context from previous row and initialize row state.
            rowCoder.copyState(m_initSliceContext);
            rowCoder.loadContexts(m_rows[row - 1].bufferedEntropy);
        }

        // Does all the CU analysis, returns best top level mode decision
        Mode& best = tld.analysis.compressCTU(*ctu, *m_frame, m_cuGeoms[m_ctuGeomMap[cuAddr]], rowCoder);

        /* advance top-level row coder to include the context of this CTU.
         * if SAO is disabled, rowCoder writes the final CTU bitstream */
        rowCoder.encodeCTU(*ctu, m_cuGeoms[m_ctuGeomMap[cuAddr]]);

        if (m_param->bEnableWavefront && col == 1)
            // Save CABAC state for next row
            curRow.bufferedEntropy.loadContexts(rowCoder);

        // Completed CU processing
        curRow.completed++;

        if (m_param->bLogCuStats || m_param->rc.bStatWrite)
            collectCTUStatistics(*ctu);

        // copy no. of intra, inter Cu cnt per row into frame stats for 2 pass
        if (m_param->rc.bStatWrite)
        {
            curRow.rowStats.mvBits += best.mvBits;
            curRow.rowStats.coeffBits += best.coeffBits;
            curRow.rowStats.miscBits += best.totalBits - (best.mvBits + best.coeffBits);
            StatisticLog* log = &m_sliceTypeLog[slice->m_sliceType];

            for (uint32_t depth = 0; depth <= g_maxCUDepth; depth++)
            {
                /* 1 << shift == number of 8x8 blocks at current depth */
                int shift = 2 * (g_maxCUDepth - depth);
                curRow.rowStats.iCuCnt += log->qTreeIntraCnt[depth] << shift;
                curRow.rowStats.pCuCnt += log->qTreeInterCnt[depth] << shift;
                curRow.rowStats.skipCuCnt += log->qTreeSkipCnt[depth] << shift;

                // clear the row cu data from thread local object
                log->qTreeIntraCnt[depth] = log->qTreeInterCnt[depth] = log->qTreeSkipCnt[depth] = 0;
            }
        }

        curEncData.m_cuStat[cuAddr].totalBits = best.totalBits;
        x265_emms();

        if (bIsVbv)
        {
            // Update encoded bits, satdCost, baseQP for each CU
            curEncData.m_rowStat[row].diagSatd      += curEncData.m_cuStat[cuAddr].vbvCost;
            curEncData.m_rowStat[row].diagIntraSatd += curEncData.m_cuStat[cuAddr].intraVbvCost;
            curEncData.m_rowStat[row].encodedBits   += curEncData.m_cuStat[cuAddr].totalBits;
            curEncData.m_rowStat[row].sumQpRc       += curEncData.m_cuStat[cuAddr].baseQp;
            curEncData.m_rowStat[row].numEncodedCUs = cuAddr;

            // If current block is at row diagonal checkpoint, call vbv ratecontrol.

            if (row == col && row)
            {
                double qpBase = curEncData.m_cuStat[cuAddr].baseQp;
                int reEncode = m_top->m_rateControl->rowDiagonalVbvRateControl(m_frame, row, &m_rce, qpBase);
                qpBase = Clip3((double)QP_MIN, (double)QP_MAX_MAX, qpBase);
                curEncData.m_rowStat[row].diagQp = qpBase;
                curEncData.m_rowStat[row].diagQpScale =  x265_qp2qScale(qpBase);

                if (reEncode < 0)
                {
                    x265_log(m_param, X265_LOG_DEBUG, "POC %d row %d - encode restart required for VBV, to %.2f from %.2f\n",
                             m_frame->m_poc, row, qpBase, curEncData.m_cuStat[cuAddr].baseQp);

                    // prevent the WaveFront::findJob() method from providing new jobs
                    m_vbvResetTriggerRow = row;
                    m_bAllRowsStop = true;

                    for (int r = m_numRows - 1; r >= row; r--)
                    {
                        CTURow& stopRow = m_rows[r];

                        if (r != row)
                        {
                            /* if row was active (ready to be run) clear active bit and bitmap bit for this row */
                            stopRow.lock.acquire();
                            while (stopRow.active)
                            {
                                if (dequeueRow(r * 2))
                                    stopRow.active = false;
                                else
                                {
                                    /* we must release the row lock to allow the thread to exit */
                                    stopRow.lock.release();
                                    GIVE_UP_TIME();
                                    stopRow.lock.acquire();
                                }
                            }
                            stopRow.lock.release();

                            bool bRowBusy = true;
                            do
                            {
                                stopRow.lock.acquire();
                                bRowBusy = stopRow.busy;
                                stopRow.lock.release();

                                if (bRowBusy)
                                {
                                    GIVE_UP_TIME();
                                }
                            }
                            while (bRowBusy);
                        }

                        m_outStreams[r].resetBits();
                        stopRow.completed = 0;
                        memset(&stopRow.rowStats, 0, sizeof(stopRow.rowStats));
                        curEncData.m_rowStat[r].numEncodedCUs = 0;
                        curEncData.m_rowStat[r].encodedBits = 0;
                        curEncData.m_rowStat[r].diagSatd = 0;
                        curEncData.m_rowStat[r].diagIntraSatd = 0;
                        curEncData.m_rowStat[r].sumQpRc = 0;
                        curEncData.m_rowStat[r].sumQpAq = 0;
                    }

                    m_bAllRowsStop = false;
                }
            }
        }

        // NOTE: do CU level Filter
        if (m_param->bEnableSAO && m_param->bSaoNonDeblocked)
            // SAO parameter estimation using non-deblocked pixels for CTU bottom and right boundary areas
            m_frameFilter.m_sao.calcSaoStatsCu_BeforeDblk(m_frame, col, row);

        // NOTE: active next row
        if (curRow.completed >= 2 && row < m_numRows - 1)
        {
            ScopedLock below(m_rows[row + 1].lock);
            if (m_rows[row + 1].active == false &&
                m_rows[row + 1].completed + 2 <= curRow.completed &&
                (!m_bAllRowsStop || row + 1 < m_vbvResetTriggerRow))
            {
                m_rows[row + 1].active = true;
                enqueueRowEncoder(row + 1);
            }
        }

        ScopedLock self(curRow.lock);
        if ((m_bAllRowsStop && row > m_vbvResetTriggerRow) ||
            (row > 0 && curRow.completed < numCols - 1 && m_rows[row - 1].completed < m_rows[row].completed + 2))
        {
            curRow.active = false;
            curRow.busy = false;
            m_totalTime += x265_mdate() - startTime;
            return;
        }
    }

    /* *this row of CTUs has been encoded* */

    /* flush row bitstream (if WPP and no SAO) or flush frame if no WPP and no SAO */
    if (!m_param->bEnableSAO && (m_param->bEnableWavefront || row == m_numRows - 1))
        rowCoder.finishSlice();

    /* If encoding with ABR, update update bits and complexity in rate control
     * after a number of rows so the next frame's rateControlStart has more
     * accurate data for estimation. At the start of the encode we update stats
     * after half the frame is encoded, but after this initial period we update
     * after refLagRows (the number of rows reference frames must have completed
     * before referencees may begin encoding) */
    int rowCount = 0;
    if (m_param->rc.rateControlMode == X265_RC_ABR)
    {
        if ((uint32_t)m_rce.encodeOrder <= 2 * (m_param->fpsNum / m_param->fpsDenom))
            rowCount = X265_MIN((m_numRows + 1) / 2, m_numRows - 1);
        else
            rowCount = X265_MIN(m_refLagRows, m_numRows - 1);
    }
    if (row == rowCount)
    {
        m_rce.rowTotalBits = 0;
        if (bIsVbv)
            for (int i = 0; i < rowCount; i++)
                m_rce.rowTotalBits += curEncData.m_rowStat[i].encodedBits;
        else
            for (uint32_t cuAddr = 0; cuAddr < rowCount * numCols; cuAddr++)
                m_rce.rowTotalBits += curEncData.m_cuStat[cuAddr].totalBits;

        m_top->m_rateControl->rateControlUpdateStats(&m_rce);
    }

    if (m_param->bEnableWavefront)
    {
        /* trigger row-wise loop filters */
        if (row >= m_filterRowDelay)
        {
            enableRowFilter(row - m_filterRowDelay);

            /* NOTE: Activate filter if first row (row 0) */
            if (row == m_filterRowDelay)
                enqueueRowFilter(0);
        }
        if (row == m_numRows - 1)
        {
            for (int i = m_numRows - m_filterRowDelay; i < m_numRows; i++)
                enableRowFilter(i);
        }
    }

    m_totalTime += x265_mdate() - startTime;
    curRow.busy = false;
}

void FrameEncoder::collectCTUStatistics(CUData& ctu)
{
    StatisticLog* log = &m_sliceTypeLog[ctu.m_slice->m_sliceType];

    if (ctu.m_slice->m_sliceType == I_SLICE)
    {
        uint32_t depth = 0;
        for (uint32_t absPartIdx = 0; absPartIdx < ctu.m_numPartitions; absPartIdx += ctu.m_numPartitions >> (depth * 2))
        {
            depth = ctu.m_cuDepth[absPartIdx];

            log->totalCu++;
            log->cntIntra[depth]++;
            log->qTreeIntraCnt[depth]++;

            if (ctu.m_predMode[absPartIdx] == MODE_NONE)
            {
                log->totalCu--;
                log->cntIntra[depth]--;
                log->qTreeIntraCnt[depth]--;
            }
            else if (ctu.m_partSize[absPartIdx] != SIZE_2Nx2N)
            {
                /* TODO: log intra modes at absPartIdx +0 to +3 */
                X265_CHECK(depth == g_maxCUDepth, "Intra NxN found at improbable depth\n");
                log->cntIntraNxN++;
                log->cntIntra[depth]--;
            }
            else if (ctu.m_lumaIntraDir[absPartIdx] > 1)
                log->cuIntraDistribution[depth][ANGULAR_MODE_ID]++;
            else
                log->cuIntraDistribution[depth][ctu.m_lumaIntraDir[absPartIdx]]++;
        }
    }
    else
    {
        uint32_t depth = 0;
        for (uint32_t absPartIdx = 0; absPartIdx < ctu.m_numPartitions; absPartIdx += ctu.m_numPartitions >> (depth * 2))
        {
            depth = ctu.m_cuDepth[absPartIdx];

            log->totalCu++;
            log->cntTotalCu[depth]++;

            if (ctu.m_predMode[absPartIdx] == MODE_NONE)
            {
                log->totalCu--;
                log->cntTotalCu[depth]--;
            }
            else if (ctu.isSkipped(absPartIdx))
            {
                log->totalCu--;
                log->cntSkipCu[depth]++;
                log->qTreeSkipCnt[depth]++;
            }
            else if (ctu.isInter(absPartIdx))
            {
                log->cntInter[depth]++;
                log->qTreeInterCnt[depth]++;

                if (ctu.m_partSize[absPartIdx] < AMP_ID)
                    log->cuInterDistribution[depth][ctu.m_partSize[absPartIdx]]++;
                else
                    log->cuInterDistribution[depth][AMP_ID]++;
            }
            else if (ctu.isIntra(absPartIdx))
            {
                log->cntIntra[depth]++;
                log->qTreeIntraCnt[depth]++;

                if (ctu.m_partSize[absPartIdx] != SIZE_2Nx2N)
                {
                    X265_CHECK(depth == g_maxCUDepth, "Intra NxN found at improbable depth\n");
                    log->cntIntraNxN++;
                    /* TODO: log intra modes at absPartIdx +0 to +3 */
                }
                else if (ctu.m_lumaIntraDir[absPartIdx] > 1)
                    log->cuIntraDistribution[depth][ANGULAR_MODE_ID]++;
                else
                    log->cuIntraDistribution[depth][ctu.m_lumaIntraDir[absPartIdx]]++;
            }
        }
    }
}

/* DCT-domain noise reduction / adaptive deadzone from libavcodec */
void FrameEncoder::noiseReductionUpdate()
{
    if (!m_nr)
        return;

    static const uint32_t maxBlocksPerTrSize[4] = {1 << 18, 1 << 16, 1 << 14, 1 << 12};

    for (int cat = 0; cat < MAX_NUM_TR_CATEGORIES; cat++)
    {
        int trSize = cat & 3;
        int coefCount = 1 << ((trSize + 2) * 2);

        if (m_nr->count[cat] > maxBlocksPerTrSize[trSize])
        {
            for (int i = 0; i < coefCount; i++)
                m_nr->residualSum[cat][i] >>= 1;
            m_nr->count[cat] >>= 1;
        }

        int nrStrength = cat < 8 ? m_param->noiseReductionIntra : m_param->noiseReductionInter;
        uint64_t scaledCount = (uint64_t)nrStrength * m_nr->count[cat];

        for (int i = 0; i < coefCount; i++)
        {
            uint64_t value = scaledCount + m_nr->residualSum[cat][i] / 2;
            uint64_t denom = m_nr->residualSum[cat][i] + 1;
            m_nr->offsetDenoise[cat][i] = (uint16_t)(value / denom);
        }

        // Don't denoise DC coefficients
        m_nr->offsetDenoise[cat][0] = 0;
    }
}

int FrameEncoder::calcQpForCu(uint32_t ctuAddr, double baseQp)
{
    x265_emms();
    double qp = baseQp;

    FrameData& curEncData = *m_frame->m_encData;
    /* clear cuCostsForVbv from when vbv row reset was triggered */
    bool bIsVbv = m_param->rc.vbvBufferSize > 0 && m_param->rc.vbvMaxBitrate > 0;
    if (bIsVbv)
    {
        curEncData.m_cuStat[ctuAddr].vbvCost = 0;
        curEncData.m_cuStat[ctuAddr].intraVbvCost = 0;
    }

    /* Derive qpOffet for each CU by averaging offsets for all 16x16 blocks in the cu. */
    double qp_offset = 0;
    uint32_t maxBlockCols = (m_frame->m_fencPic->m_picWidth + (16 - 1)) / 16;
    uint32_t maxBlockRows = (m_frame->m_fencPic->m_picHeight + (16 - 1)) / 16;
    uint32_t noOfBlocks = g_maxCUSize / 16;
    uint32_t block_y = (ctuAddr / curEncData.m_slice->m_sps->numCuInWidth) * noOfBlocks;
    uint32_t block_x = (ctuAddr * noOfBlocks) - block_y * curEncData.m_slice->m_sps->numCuInWidth;

    /* Use cuTree offsets if cuTree enabled and frame is referenced, else use AQ offsets */
    bool isReferenced = IS_REFERENCED(m_frame);
    double *qpoffs = (isReferenced && m_param->rc.cuTree) ? m_frame->m_lowres.qpCuTreeOffset : m_frame->m_lowres.qpAqOffset;

    uint32_t cnt = 0, idx = 0;
    for (uint32_t h = 0; h < noOfBlocks && block_y < maxBlockRows; h++, block_y++)
    {
        for (uint32_t w = 0; w < noOfBlocks && (block_x + w) < maxBlockCols; w++)
        {
            idx = block_x + w + (block_y * maxBlockCols);
            if (m_param->rc.aqMode)
                qp_offset += qpoffs[idx];
            if (bIsVbv)
            {
                curEncData.m_cuStat[ctuAddr].vbvCost += m_frame->m_lowres.lowresCostForRc[idx] & LOWRES_COST_MASK;
                curEncData.m_cuStat[ctuAddr].intraVbvCost += m_frame->m_lowres.intraCost[idx];
            }
            cnt++;
        }
    }

    qp_offset /= cnt;
    qp += qp_offset;

    return Clip3(QP_MIN, QP_MAX_MAX, (int)(qp + 0.5));
}

Frame *FrameEncoder::getEncodedPicture(NALList& output)
{
    if (m_frame)
    {
        /* block here until worker thread completes */
        m_done.wait();

        Frame *ret = m_frame;
        m_frame = NULL;
        output.takeContents(m_nalList);
        return ret;
    }

    return NULL;
}
}