[x265] x265-devel Digest, Vol 41, Issue 42
Aruna Matheswaran
aruna at multicorewareinc.com
Mon Oct 31 11:40:39 CET 2016
Found following cases in encoding efficiency of no-wpp (in comparison with
wpp),
1. Drop in bitrate and raise in SSIM.
2. Drop in bitrate and drop in SSIM ( Compared visual quality of no-wpp
with wpp. no-wpp's visual quality is better than wpp in some cases and both
are of equal quality in some cases) .
3. Raise in bitrate and drop in SSIM ( Found in medium preset. No
difference in visual quality is observed).
PFA - comparison file.
On Thu, Oct 27, 2016 at 12:27 PM, <x265-devel-request at videolan.org> wrote:
> Send x265-devel mailing list submissions to
> x265-devel at videolan.org
>
> To subscribe or unsubscribe via the World Wide Web, visit
> https://mailman.videolan.org/listinfo/x265-devel
> or, via email, send a message with subject or body 'help' to
> x265-devel-request at videolan.org
>
> You can reach the person managing the list at
> x265-devel-owner at videolan.org
>
> When replying, please edit your Subject line so it is more specific
> than "Re: Contents of x265-devel digest..."
>
>
> Today's Topics:
>
> 1. Re: [PATCH] Add VBV support for No-WPP (Pradeep Ramachandran)
>
>
> ----------------------------------------------------------------------
>
> Message: 1
> Date: Thu, 27 Oct 2016 12:26:41 +0530
> From: Pradeep Ramachandran <pradeep at multicorewareinc.com>
> To: Development for x265 <x265-devel at videolan.org>
> Subject: Re: [x265] [PATCH] Add VBV support for No-WPP
> Message-ID:
> <CAKpaXXfGvcA+e6h5v0Br1sMQDrfW9e2KYZNAfxW5AN
> Ub-WAxKA at mail.gmail.com>
> Content-Type: text/plain; charset="utf-8"
>
> On Wed, Oct 26, 2016 at 11:08 AM, <aruna at multicorewareinc.com> wrote:
>
> > # HG changeset patch
> > # User Aruna Matheswaran
> > # Date 1476352364 -19800
> > # Thu Oct 13 15:22:44 2016 +0530
> > # Node ID f9e7422416c9d2d4f7b7618791a7c28592de4828
> > # Parent bc911034c2a07380630aff98fdda38038b2ae62e
> > Add VBV support for No-WPP
> >
>
> Can you please share what improvement to encoding efficiency you see on
> command-lines with VBV with and without wpp?
>
>
> > diff -r bc911034c2a0 -r f9e7422416c9 source/common/bitstream.h
> > --- a/source/common/bitstream.h Tue Oct 25 11:32:10 2016 +0530
> > +++ b/source/common/bitstream.h Thu Oct 13 15:22:44 2016 +0530
> > @@ -71,6 +71,7 @@
> > uint32_t getNumberOfWrittenBytes() const { return m_byteOccupancy; }
> > uint32_t getNumberOfWrittenBits() const { return m_byteOccupancy *
> 8
> > + m_partialByteBits; }
> > const uint8_t* getFIFO() const { return m_fifo; }
> > + void copyBits(Bitstream* stream) { m_partialByteBits =
> > stream->m_partialByteBits; m_byteOccupancy = stream->m_byteOccupancy;
> > m_partialByte = stream->m_partialByte; }
> >
> > void write(uint32_t val, uint32_t numBits);
> > void writeByte(uint32_t val);
> > diff -r bc911034c2a0 -r f9e7422416c9 source/common/framedata.h
> > --- a/source/common/framedata.h Tue Oct 25 11:32:10 2016 +0530
> > +++ b/source/common/framedata.h Thu Oct 13 15:22:44 2016 +0530
> > @@ -126,10 +126,10 @@
> > uint32_t encodedBits; /* sum of 'totalBits' of encoded CTUs */
> > uint32_t satdForVbv; /* sum of lowres (estimated) costs for
> > entire row */
> > uint32_t intraSatdForVbv; /* sum of lowres (estimated) intra
> > costs for entire row */
> > - uint32_t diagSatd;
> > - uint32_t diagIntraSatd;
> > - double diagQp;
> > - double diagQpScale;
> > + uint32_t rowSatd;
> > + uint32_t rowIntraSatd;
> > + double rowQp;
> > + double rowQpScale;
> > double sumQpRc;
> > double sumQpAq;
> > };
> > diff -r bc911034c2a0 -r f9e7422416c9 source/encoder/encoder.cpp
> > --- a/source/encoder/encoder.cpp Tue Oct 25 11:32:10 2016 +0530
> > +++ b/source/encoder/encoder.cpp Thu Oct 13 15:22:44 2016 +0530
> > @@ -149,12 +149,6 @@
> > p->bEnableWavefront = p->bDistributeModeAnalysis =
> > p->bDistributeMotionEstimation = p->lookaheadSlices = 0;
> > }
> >
> > - if (!p->bEnableWavefront && p->rc.vbvBufferSize)
> > - {
> > - x265_log(p, X265_LOG_ERROR, "VBV requires wavefront
> > parallelism\n");
> > - m_aborted = true;
> > - }
> > -
> > x265_log(p, X265_LOG_INFO, "Slices :
> > %d\n", p->maxSlices);
> >
> > char buf[128];
> > diff -r bc911034c2a0 -r f9e7422416c9 source/encoder/frameencoder.cpp
> > --- a/source/encoder/frameencoder.cpp Tue Oct 25 11:32:10 2016 +0530
> > +++ b/source/encoder/frameencoder.cpp Thu Oct 13 15:22:44 2016 +0530
> > @@ -50,6 +50,7 @@
> > m_bAllRowsStop = false;
> > m_vbvResetTriggerRow = -1;
> > m_outStreams = NULL;
> > + m_backupStreams = NULL;
> > m_substreamSizes = NULL;
> > m_nr = NULL;
> > m_tld = NULL;
> > @@ -85,6 +86,7 @@
> >
> > delete[] m_rows;
> > delete[] m_outStreams;
> > + delete[] m_backupStreams;
> > X265_FREE(m_sliceBaseRow);
> > X265_FREE(m_cuGeoms);
> > X265_FREE(m_ctuGeomMap);
> > @@ -532,6 +534,8 @@
> > if (!m_outStreams)
> > {
> > m_outStreams = new Bitstream[numSubstreams];
> > + if (!m_param->bEnableWavefront)
> > + m_backupStreams = new Bitstream[numSubstreams];
> > m_substreamSizes = X265_MALLOC(uint32_t, numSubstreams);
> > if (!m_param->bEnableSAO)
> > for (uint32_t i = 0; i < numSubstreams; i++)
> > @@ -1203,17 +1207,25 @@
> >
> > if (bIsVbv)
> > {
> > - if (!row)
> > + if (col == 0 && !m_param->bEnableWavefront)
> > + {
> > + m_backupStreams[0].copyBits(&m_outStreams[0]);
> > + curRow.bufferedEntropy.copyState(rowCoder);
> > + curRow.bufferedEntropy.loadContexts(rowCoder);
> > + }
> > + if (!row && m_vbvResetTriggerRow != intRow)
> > {
> > - curEncData.m_rowStat[row].diagQp =
> curEncData.m_avgQpRc;
> > - curEncData.m_rowStat[row].diagQpScale =
> > x265_qp2qScale(curEncData.m_avgQpRc);
> > + curEncData.m_rowStat[row].rowQp = curEncData.m_avgQpRc;
> > + curEncData.m_rowStat[row].rowQpScale =
> > x265_qp2qScale(curEncData.m_avgQpRc);
> > }
> >
> > FrameData::RCStatCU& cuStat = curEncData.m_cuStat[cuAddr];
> > - if (row >= col && row && m_vbvResetTriggerRow != intRow)
> > - cuStat.baseQp = curEncData.m_cuStat[cuAddr - numCols +
> > 1].baseQp;
> > + if (m_param->bEnableWavefront && row >= col && row &&
> > m_vbvResetTriggerRow != intRow)
> > + cuStat.baseQp = curEncData.m_cuStat[cuAddr - numCols +
> > 1].baseQp;
> > + else if (!m_param->bEnableWavefront && row &&
> > m_vbvResetTriggerRow != intRow)
> > + cuStat.baseQp = curEncData.m_rowStat[row - 1].rowQp;
> > else
> > - cuStat.baseQp = curEncData.m_rowStat[row].diagQp;
> > + cuStat.baseQp = curEncData.m_rowStat[row].rowQp;
> >
> > /* TODO: use defines from slicetype.h for lowres block size
> */
> > uint32_t block_y = (ctu->m_cuPelY >> g_maxLog2CUSize) *
> > noOfBlocks;
> > @@ -1364,21 +1376,52 @@
> > 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].rowSatd +=
> > curEncData.m_cuStat[cuAddr].vbvCost;
> > + curEncData.m_rowStat[row].rowIntraSatd +=
> > 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 end checkpoint, call vbv
> > ratecontrol.
> > +
> > + if (!m_param->bEnableWavefront && col == numCols - 1)
> > + {
> > + double qpBase = curEncData.m_cuStat[cuAddr].baseQp;
> > + int reEncode = m_top->m_rateControl->
> rowVbvRateControl(m_frame,
> > row, &m_rce, qpBase);
> > + qpBase = x265_clip3((double)m_param->rc.qpMin,
> > (double)m_param->rc.qpMax, qpBase);
> > + curEncData.m_rowStat[row].rowQp = qpBase;
> > + curEncData.m_rowStat[row].rowQpScale =
> > 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);
> > +
> > + m_vbvResetTriggerRow = row;
> > + m_outStreams[0].copyBits(&m_backupStreams[0]);
> > +
> > + rowCoder.copyState(curRow.bufferedEntropy);
> > + rowCoder.loadContexts(curRow.bufferedEntropy);
> > +
> > + curRow.completed = 0;
> > + memset(&curRow.rowStats, 0,
> sizeof(curRow.rowStats));
> > + curEncData.m_rowStat[row].numEncodedCUs = 0;
> > + curEncData.m_rowStat[row].encodedBits = 0;
> > + curEncData.m_rowStat[row].rowSatd = 0;
> > + curEncData.m_rowStat[row].rowIntraSatd = 0;
> > + curEncData.m_rowStat[row].sumQpRc = 0;
> > + curEncData.m_rowStat[row].sumQpAq = 0;
> > + }
> > + }
> > +
> > // If current block is at row diagonal checkpoint, call vbv
> > ratecontrol.
> >
> > - if (row == col && row)
> > + else if (m_param->bEnableWavefront && row == col && row)
> > {
> > double qpBase = curEncData.m_cuStat[cuAddr].baseQp;
> > - int reEncode = m_top->m_rateControl->
> > rowDiagonalVbvRateControl(m_frame, row, &m_rce, qpBase);
> > + int reEncode = m_top->m_rateControl->
> rowVbvRateControl(m_frame,
> > row, &m_rce, qpBase);
> > qpBase = x265_clip3((double)m_param->rc.qpMin,
> > (double)m_param->rc.qpMax, qpBase);
> > - curEncData.m_rowStat[row].diagQp = qpBase;
> > - curEncData.m_rowStat[row].diagQpScale =
> > x265_qp2qScale(qpBase);
> > + curEncData.m_rowStat[row].rowQp = qpBase;
> > + curEncData.m_rowStat[row].rowQpScale =
> > x265_qp2qScale(qpBase);
> >
> > if (reEncode < 0)
> > {
> > @@ -1431,8 +1474,8 @@
> > 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].rowSatd = 0;
> > + curEncData.m_rowStat[r].rowIntraSatd = 0;
> > curEncData.m_rowStat[r].sumQpRc = 0;
> > curEncData.m_rowStat[r].sumQpAq = 0;
> > }
> > diff -r bc911034c2a0 -r f9e7422416c9 source/encoder/frameencoder.h
> > --- a/source/encoder/frameencoder.h Tue Oct 25 11:32:10 2016 +0530
> > +++ b/source/encoder/frameencoder.h Thu Oct 13 15:22:44 2016 +0530
> > @@ -184,6 +184,7 @@
> > NoiseReduction* m_nr;
> > ThreadLocalData* m_tld; /* for --no-wpp */
> > Bitstream* m_outStreams;
> > + Bitstream* m_backupStreams;
> > uint32_t* m_substreamSizes;
> >
> > CUGeom* m_cuGeoms;
> > diff -r bc911034c2a0 -r f9e7422416c9 source/encoder/ratecontrol.cpp
> > --- a/source/encoder/ratecontrol.cpp Tue Oct 25 11:32:10 2016 +0530
> > +++ b/source/encoder/ratecontrol.cpp Thu Oct 13 15:22:44 2016 +0530
> > @@ -2180,7 +2180,7 @@
> > for (uint32_t row = 0; row < maxRows; row++)
> > {
> > encodedBitsSoFar += curEncData.m_rowStat[row].encodedBits;
> > - rowSatdCostSoFar = curEncData.m_rowStat[row].diagSatd;
> > + rowSatdCostSoFar = curEncData.m_rowStat[row].rowSatd;
> > uint32_t satdCostForPendingCus = curEncData.m_rowStat[row].
> satdForVbv
> > - rowSatdCostSoFar;
> > satdCostForPendingCus >>= X265_DEPTH - 8;
> > if (satdCostForPendingCus > 0)
> > @@ -2209,7 +2209,7 @@
> > }
> >
> > refRowSatdCost >>= X265_DEPTH - 8;
> > - refQScale = refEncData.m_rowStat[row].diagQpScale;
> > + refQScale = refEncData.m_rowStat[row].rowQpScale;
> > }
> >
> > if (picType == I_SLICE || qScale >= refQScale)
> > @@ -2231,7 +2231,7 @@
> > }
> > else if (picType == P_SLICE)
> > {
> > - intraCostForPendingCus = curEncData.m_rowStat[row].
> intraSatdForVbv
> > - curEncData.m_rowStat[row].diagIntraSatd;
> > + intraCostForPendingCus = curEncData.m_rowStat[row].
> intraSatdForVbv
> > - curEncData.m_rowStat[row].rowIntraSatd;
> > intraCostForPendingCus >>= X265_DEPTH - 8;
> > /* Our QP is lower than the reference! */
> > double pred_intra = predictSize(rce->rowPred[1], qScale,
> > intraCostForPendingCus);
> > @@ -2246,16 +2246,16 @@
> > return totalSatdBits + encodedBitsSoFar;
> > }
> >
> > -int RateControl::rowDiagonalVbvRateControl(Frame* curFrame, uint32_t
> > row, RateControlEntry* rce, double& qpVbv)
> > +int RateControl::rowVbvRateControl(Frame* curFrame, uint32_t row,
> > RateControlEntry* rce, double& qpVbv)
> > {
> > FrameData& curEncData = *curFrame->m_encData;
> > double qScaleVbv = x265_qp2qScale(qpVbv);
> > - uint64_t rowSatdCost = curEncData.m_rowStat[row].diagSatd;
> > + uint64_t rowSatdCost = curEncData.m_rowStat[row].rowSatd;
> > double encodedBits = curEncData.m_rowStat[row].encodedBits;
> >
> > - if (row == 1)
> > + if (m_param->bEnableWavefront && row == 1)
> > {
> > - rowSatdCost += curEncData.m_rowStat[0].diagSatd;
> > + rowSatdCost += curEncData.m_rowStat[0].rowSatd;
> > encodedBits += curEncData.m_rowStat[0].encodedBits;
> > }
> > rowSatdCost >>= X265_DEPTH - 8;
> > @@ -2263,11 +2263,11 @@
> > if (curEncData.m_slice->m_sliceType != I_SLICE)
> > {
> > Frame* refFrame = curEncData.m_slice->m_refFrameList[0][0];
> > - if (qpVbv < refFrame->m_encData->m_rowStat[row].diagQp)
> > + if (qpVbv < refFrame->m_encData->m_rowStat[row].rowQp)
> > {
> > - uint64_t intraRowSatdCost = curEncData.m_rowStat[row].
> > diagIntraSatd;
> > - if (row == 1)
> > - intraRowSatdCost += curEncData.m_rowStat[0].
> > diagIntraSatd;
> > + uint64_t intraRowSatdCost = curEncData.m_rowStat[row].
> > rowIntraSatd;
> > + if (m_param->bEnableWavefront && row == 1)
> > + intraRowSatdCost += curEncData.m_rowStat[0].
> rowIntraSatd;
> > intraRowSatdCost >>= X265_DEPTH - 8;
> > updatePredictor(rce->rowPred[1], qScaleVbv,
> > (double)intraRowSatdCost, encodedBits);
> > }
> > @@ -2328,7 +2328,7 @@
> > }
> >
> > while (qpVbv > qpMin
> > - && (qpVbv > curEncData.m_rowStat[0].diagQp ||
> > m_singleFrameVbv)
> > + && (qpVbv > curEncData.m_rowStat[0].rowQp ||
> > m_singleFrameVbv)
> > && (((accFrameBits < rce->frameSizePlanned * 0.8f &&
> qpVbv
> > <= prevRowQp)
> > || accFrameBits < (rce->bufferFill - m_bufferSize +
> > m_bufferRate) * 1.1)
> > && (!m_param->rc.bStrictCbr ? 1 : abrOvershoot < 0)))
> > @@ -2348,7 +2348,7 @@
> > accFrameBits = predictRowsSizeSum(curFrame, rce, qpVbv,
> > encodedBitsSoFar);
> > abrOvershoot = (accFrameBits + m_totalBits -
> > m_wantedBitsWindow) / totalBitsNeeded;
> > }
> > - if (qpVbv > curEncData.m_rowStat[0].diagQp &&
> > + if (qpVbv > curEncData.m_rowStat[0].rowQp &&
> > abrOvershoot < -0.1 && timeDone > 0.5 && accFrameBits <
> > rce->frameSizePlanned - rcTol)
> > {
> > qpVbv -= stepSize;
> > diff -r bc911034c2a0 -r f9e7422416c9 source/encoder/ratecontrol.h
> > --- a/source/encoder/ratecontrol.h Tue Oct 25 11:32:10 2016 +0530
> > +++ b/source/encoder/ratecontrol.h Thu Oct 13 15:22:44 2016 +0530
> > @@ -243,7 +243,7 @@
> > int rateControlStart(Frame* curFrame, RateControlEntry* rce,
> > Encoder* enc);
> > void rateControlUpdateStats(RateControlEntry* rce);
> > int rateControlEnd(Frame* curFrame, int64_t bits, RateControlEntry*
> > rce);
> > - int rowDiagonalVbvRateControl(Frame* curFrame, uint32_t row,
> > RateControlEntry* rce, double& qpVbv);
> > + int rowVbvRateControl(Frame* curFrame, uint32_t row,
> > RateControlEntry* rce, double& qpVbv);
> > int rateControlSliceType(int frameNum);
> > bool cuTreeReadFor2Pass(Frame* curFrame);
> > void hrdFullness(SEIBufferingPeriod* sei);
> > diff -r bc911034c2a0 -r f9e7422416c9 source/test/rate-control-tests.txt
> > --- a/source/test/rate-control-tests.txt Tue Oct 25 11:32:10 2016
> > +0530
> > +++ b/source/test/rate-control-tests.txt Thu Oct 13 15:22:44 2016
> > +0530
> > @@ -21,6 +21,9 @@
> > big_buck_bunny_360p24.y4m,--preset medium --bitrate 400 --vbv-bufsize
> > 600 --vbv-maxrate 600 --aud --hrd --tune fast-decode
> > sita_1920x1080_30.yuv,--preset superfast --crf 25 --vbv-bufsize 3000
> > --vbv-maxrate 4000 --vbv-bufsize 5000 --hrd --crf-max 30
> > sita_1920x1080_30.yuv,--preset superfast --bitrate 3000 --vbv-bufsize
> > 3000 --vbv-maxrate 3000 --aud --strict-cbr
> > +BasketballDrive_1920x1080_50.y4m,--preset ultrafast --bitrate 3000
> > --vbv-bufsize 3000 --vbv-maxrate 3000 --no-wpp
> > +big_buck_bunny_360p24.y4m,--preset medium --bitrate 400 --vbv-bufsize
> > 600 --vbv-maxrate 600 --no-wpp --aud --hrd --tune fast-decode
> > +sita_1920x1080_30.yuv,--preset superfast --bitrate 3000 --vbv-bufsize
> > 3000 --vbv-maxrate 3000 --aud --strict-cbr --no-wpp
> >
> >
> >
> > _______________________________________________
> > x265-devel mailing list
> > x265-devel at videolan.org
> > https://mailman.videolan.org/listinfo/x265-devel
> >
> -------------- next part --------------
> An HTML attachment was scrubbed...
> URL: <http://mailman.videolan.org/pipermail/x265-devel/
> attachments/20161027/23a697eb/attachment.html>
>
> ------------------------------
>
> Subject: Digest Footer
>
> _______________________________________________
> x265-devel mailing list
> x265-devel at videolan.org
> https://mailman.videolan.org/listinfo/x265-devel
>
>
> ------------------------------
>
> End of x265-devel Digest, Vol 41, Issue 42
> ******************************************
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.videolan.org/pipermail/x265-devel/attachments/20161031/f1c024d0/attachment-0001.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: Encoding_efficiency_VBV_with_NoWPP.xls
Type: application/vnd.ms-excel
Size: 30208 bytes
Desc: not available
URL: <http://mailman.videolan.org/pipermail/x265-devel/attachments/20161031/f1c024d0/attachment-0001.xls>
More information about the x265-devel
mailing list