[x265] [PATCH 1 of 2 EXPERIMENTAL] slice: prep work for per-numa recon picture copies
Deepthi Nandakumar
deepthi at multicorewareinc.com
Wed Aug 5 11:54:28 CEST 2015
Steve,
allocRecon is now being called unconditionally, so a recon frame gets
allocated for every single frame, as opposed to an available frame being
chosen from the DPB freeList?
On Tue, Aug 4, 2015 at 9:03 AM, Steve Borho <steve at borho.org> wrote:
> # HG changeset patch
> # User Steve Borho <steve at borho.org>
> # Date 1438642558 18000
> # Mon Aug 03 17:55:58 2015 -0500
> # Node ID 02e84edaa14399a3a68ade8617c63422f51a305b
> # Parent 1f161d9c6e35e32998d38ebf5b6dec96f1ef43e2
> slice: prep work for per-numa recon picture copies
>
> Rename slice.m_refPicList to m_refFrameList since it is an array of Frame
> pointers, and make a new slice.m_refReconPicList array which points
> directly
> to the motion reference PicYuv buffers (bypassing the Frame structure)
>
> The reconstructed pictures are now allocated by the frame encoder worker
> thread
> making them socket-local. The per-node structures devolve to a single
> pointer
> de-reference when NUMA support is not compiled in, minimizing the impact
> when
> the feature is disabled.
>
> The shared offset buffers were moved from the top-level encoder into the
> SPS
> structure so the FrameData functions could use them directly (avoiding a
> major
> layering violations). The offset buffers are computed based on SPS values,
> so it
> seems minimally ugly to keep them there.
>
> Later commits will make the PicYuv buffers per-NUMA node and copy
> reconstructed
> pixels between nodes on demand.
>
> diff -r 1f161d9c6e35 -r 02e84edaa143 source/common/cudata.cpp
> --- a/source/common/cudata.cpp Mon Aug 03 14:56:21 2015 -0500
> +++ b/source/common/cudata.cpp Mon Aug 03 17:55:58 2015 -0500
> @@ -1676,7 +1676,7 @@
> if (tempRefIdx != -1)
> {
> uint32_t cuAddr = neighbours[MD_COLLOCATED].cuAddr[picList];
> - const Frame* colPic =
> m_slice->m_refPicList[m_slice->isInterB() &&
> !m_slice->m_colFromL0Flag][m_slice->m_colRefIdx];
> + const Frame* colPic =
> m_slice->m_refFrameList[m_slice->isInterB() &&
> !m_slice->m_colFromL0Flag][m_slice->m_colRefIdx];
> const CUData* colCU = colPic->m_encData->getPicCTU(cuAddr);
>
> // Scale the vector
> @@ -1857,7 +1857,7 @@
>
> bool CUData::getColMVP(MV& outMV, int& outRefIdx, int picList, int
> cuAddr, int partUnitIdx) const
> {
> - const Frame* colPic = m_slice->m_refPicList[m_slice->isInterB() &&
> !m_slice->m_colFromL0Flag][m_slice->m_colRefIdx];
> + const Frame* colPic = m_slice->m_refFrameList[m_slice->isInterB() &&
> !m_slice->m_colFromL0Flag][m_slice->m_colRefIdx];
> const CUData* colCU = colPic->m_encData->getPicCTU(cuAddr);
>
> uint32_t absPartAddr = partUnitIdx & TMVP_UNIT_MASK;
> @@ -1892,7 +1892,7 @@
> // Cache the collocated MV.
> bool CUData::getCollocatedMV(int cuAddr, int partUnitIdx,
> InterNeighbourMV *neighbour) const
> {
> - const Frame* colPic = m_slice->m_refPicList[m_slice->isInterB() &&
> !m_slice->m_colFromL0Flag][m_slice->m_colRefIdx];
> + const Frame* colPic = m_slice->m_refFrameList[m_slice->isInterB() &&
> !m_slice->m_colFromL0Flag][m_slice->m_colRefIdx];
> const CUData* colCU = colPic->m_encData->getPicCTU(cuAddr);
>
> uint32_t absPartAddr = partUnitIdx & TMVP_UNIT_MASK;
> diff -r 1f161d9c6e35 -r 02e84edaa143 source/common/deblock.cpp
> --- a/source/common/deblock.cpp Mon Aug 03 14:56:21 2015 -0500
> +++ b/source/common/deblock.cpp Mon Aug 03 17:55:58 2015 -0500
> @@ -209,8 +209,8 @@
> const Slice* const sliceQ = cuQ->m_slice;
> const Slice* const sliceP = cuP->m_slice;
>
> - const Frame* refP0 = sliceP->getRefPic(0, cuP->m_refIdx[0][partP]);
> - const Frame* refQ0 = sliceQ->getRefPic(0, cuQ->m_refIdx[0][partQ]);
> + const Frame* refP0 =
> sliceP->m_refFrameList[0][cuP->m_refIdx[0][partP]];
> + const Frame* refQ0 =
> sliceQ->m_refFrameList[0][cuQ->m_refIdx[0][partQ]];
> const MV& mvP0 = refP0 ? cuP->m_mv[0][partP] : zeroMv;
> const MV& mvQ0 = refQ0 ? cuQ->m_mv[0][partQ] : zeroMv;
>
> @@ -221,8 +221,8 @@
> }
>
> // (sliceQ->isInterB() || sliceP->isInterB())
> - const Frame* refP1 = sliceP->getRefPic(1, cuP->m_refIdx[1][partP]);
> - const Frame* refQ1 = sliceQ->getRefPic(1, cuQ->m_refIdx[1][partQ]);
> + const Frame* refP1 =
> sliceP->m_refFrameList[1][cuP->m_refIdx[1][partP]];
> + const Frame* refQ1 =
> sliceQ->m_refFrameList[1][cuQ->m_refIdx[1][partQ]];
> const MV& mvP1 = refP1 ? cuP->m_mv[1][partP] : zeroMv;
> const MV& mvQ1 = refQ1 ? cuQ->m_mv[1][partQ] : zeroMv;
>
> @@ -366,7 +366,7 @@
>
> void Deblock::edgeFilterLuma(const CUData* cuQ, uint32_t absPartIdx,
> uint32_t depth, int32_t dir, int32_t edge, const uint8_t blockStrength[])
> {
> - PicYuv* reconPic = cuQ->m_encData->m_reconPic;
> + PicYuv* reconPic = cuQ->m_encData->getOutputRecon();
> pixel* src = reconPic->getLumaAddr(cuQ->m_cuAddr, absPartIdx);
> intptr_t stride = reconPic->m_stride;
> const PPS* pps = cuQ->m_slice->m_pps;
> @@ -474,7 +474,7 @@
> : ((g_zscanToPelY[absPartIdx] + edge * UNIT_SIZE) >>
> cuQ->m_vChromaShift)) % DEBLOCK_SMALLEST_BLOCK == 0,
> "invalid edge\n");
>
> - PicYuv* reconPic = cuQ->m_encData->m_reconPic;
> + PicYuv* reconPic = cuQ->m_encData->getOutputRecon();
> intptr_t stride = reconPic->m_strideC;
> intptr_t srcOffset = reconPic->getChromaAddrOffset(cuQ->m_cuAddr,
> absPartIdx);
> bool bCheckNoFilter = pps->bTransquantBypassEnabled;
> diff -r 1f161d9c6e35 -r 02e84edaa143 source/common/frame.cpp
> --- a/source/common/frame.cpp Mon Aug 03 14:56:21 2015 -0500
> +++ b/source/common/frame.cpp Mon Aug 03 17:55:58 2015 -0500
> @@ -35,7 +35,6 @@
> m_reconRowCount.set(0);
> m_countRefEncoders = 0;
> m_encData = NULL;
> - m_reconPic = NULL;
> m_next = NULL;
> m_prev = NULL;
> m_param = NULL;
> @@ -54,26 +53,13 @@
> bool Frame::allocEncodeData(x265_param *param, const SPS& sps)
> {
> m_encData = new FrameData;
> - m_reconPic = new PicYuv;
> - m_encData->m_reconPic = m_reconPic;
> - bool ok = m_encData->create(param, sps) &&
> m_reconPic->create(param->sourceWidth, param->sourceHeight,
> param->internalCsp);
> - if (ok)
> - {
> - /* initialize right border of m_reconpicYuv as SAO may read
> beyond the
> - * end of the picture accessing uninitialized pixels */
> - int maxHeight = sps.numCuInHeight * g_maxCUSize;
> - memset(m_reconPic->m_picOrg[0], 0, sizeof(pixel) *
> m_reconPic->m_stride * maxHeight);
> - memset(m_reconPic->m_picOrg[1], 0, sizeof(pixel) *
> m_reconPic->m_strideC * (maxHeight >> m_reconPic->m_vChromaShift));
> - memset(m_reconPic->m_picOrg[2], 0, sizeof(pixel) *
> m_reconPic->m_strideC * (maxHeight >> m_reconPic->m_vChromaShift));
> - }
> - return ok;
> + return m_encData->create(*param, sps);
> }
>
> /* prepare to re-use a FrameData instance to encode a new picture */
> void Frame::reinit(const SPS& sps)
> {
> m_bChromaExtended = false;
> - m_reconPic = m_encData->m_reconPic;
> m_encData->reinit(sps);
> }
>
> @@ -93,12 +79,5 @@
> m_fencPic = NULL;
> }
>
> - if (m_reconPic)
> - {
> - m_reconPic->destroy();
> - delete m_reconPic;
> - m_reconPic = NULL;
> - }
> -
> m_lowres.destroy();
> }
> diff -r 1f161d9c6e35 -r 02e84edaa143 source/common/frame.h
> --- a/source/common/frame.h Mon Aug 03 14:56:21 2015 -0500
> +++ b/source/common/frame.h Mon Aug 03 17:55:58 2015 -0500
> @@ -41,10 +41,9 @@
> {
> public:
>
> - /* These two items will be NULL until the Frame begins to be encoded,
> at which point
> - * it will be assigned a FrameData instance, which comes with a
> reconstructed image PicYuv */
> + /* will be NULL until the Frame begins to be encoded, at which point
> it will
> + * be assigned a FrameData instance */
> FrameData* m_encData;
> - PicYuv* m_reconPic;
>
> /* Data associated with x265_picture */
> PicYuv* m_fencPic;
> diff -r 1f161d9c6e35 -r 02e84edaa143 source/common/framedata.cpp
> --- a/source/common/framedata.cpp Mon Aug 03 14:56:21 2015 -0500
> +++ b/source/common/framedata.cpp Mon Aug 03 17:55:58 2015 -0500
> @@ -22,24 +22,39 @@
>
> *****************************************************************************/
>
> #include "framedata.h"
> +#include "threadpool.h"
> #include "picyuv.h"
>
> using namespace X265_NS;
>
> +PerNodeRecon::~PerNodeRecon()
> +{
> + if (reconPic)
> + {
> + reconPic->destroy();
> + delete reconPic;
> + }
> +}
> +
> FrameData::FrameData()
> {
> memset(this, 0, sizeof(*this));
> }
>
> -bool FrameData::create(x265_param *param, const SPS& sps)
> +bool FrameData::create(const x265_param& param, const SPS& sps)
> {
> - m_param = param;
> + m_param = ¶m;
> m_slice = new Slice;
> m_picCTU = new CUData[sps.numCUsInFrame];
> +#if (defined(_WIN32_WINNT) && _WIN32_WINNT >= _WIN32_WINNT_WIN7) ||
> HAVE_LIBNUMA
> + m_nodes = new PerNodeRecon[ThreadPool::getNumaNodeCount()];
> +#else
> + m_nodes = new PerNodeRecon[1];
> +#endif
>
> - m_cuMemPool.create(0, param->internalCsp, sps.numCUsInFrame);
> + m_cuMemPool.create(0, param.internalCsp, sps.numCUsInFrame);
> for (uint32_t ctuAddr = 0; ctuAddr < sps.numCUsInFrame; ctuAddr++)
> - m_picCTU[ctuAddr].initialize(m_cuMemPool, 0, param->internalCsp,
> ctuAddr);
> + m_picCTU[ctuAddr].initialize(m_cuMemPool, 0, param.internalCsp,
> ctuAddr);
>
> CHECKED_MALLOC(m_cuStat, RCStatCU, sps.numCUsInFrame);
> CHECKED_MALLOC(m_rowStat, RCStatRow, sps.numCuInHeight);
> @@ -54,11 +69,51 @@
> {
> memset(m_cuStat, 0, sps.numCUsInFrame * sizeof(*m_cuStat));
> memset(m_rowStat, 0, sps.numCuInHeight * sizeof(*m_rowStat));
> +#if (defined(_WIN32_WINNT) && _WIN32_WINNT >= _WIN32_WINNT_WIN7) ||
> HAVE_LIBNUMA
> + for (int i = 0; i < ThreadPool::getNumaNodeCount(); i++)
> + m_nodes[i].rows = 0;
> +#else
> + m_nodes->rows = 0;
> +#endif
> +}
> +
> +bool FrameData::allocRecon(const SPS& sps, int node)
> +{
> + if (m_nodes[node].reconPic)
> + return true;
> +
> + ScopedLock s(m_nodes[node].copyLock);
> +
> + if (m_nodes[node].reconPic)
> + return true;
> +
> + PicYuv* reconPic = new PicYuv;
> + m_nodes[node].reconPic = reconPic;
> +
> + int maxHeight = sps.numCuInHeight * g_maxCUSize;
> + if (reconPic->create(m_param->sourceWidth, m_param->sourceHeight,
> m_param->internalCsp))
> + {
> + /* initialize right border of recon PicYuv as SAO may read beyond
> the end
> + * of the picture accessing uninitialized pixels */
> + memset(reconPic->m_picOrg[0], 0, sizeof(pixel) *
> reconPic->m_stride * maxHeight);
> + memset(reconPic->m_picOrg[1], 0, sizeof(pixel) *
> reconPic->m_strideC * (maxHeight >> reconPic->m_vChromaShift));
> + memset(reconPic->m_picOrg[2], 0, sizeof(pixel) *
> reconPic->m_strideC * (maxHeight >> reconPic->m_vChromaShift));
> +
> + /* use pre-calculated cu/pu offsets cached in the SPS structure */
> + reconPic->m_cuOffsetC = sps.cuOffsetC;
> + reconPic->m_cuOffsetY = sps.cuOffsetY;
> + reconPic->m_buOffsetC = sps.buOffsetC;
> + reconPic->m_buOffsetY = sps.buOffsetY;
> + return true;
> + }
> +
> + return false;
> }
>
> void FrameData::destroy()
> {
> delete [] m_picCTU;
> + delete [] m_nodes;
> delete m_slice;
> delete m_saoParam;
>
> diff -r 1f161d9c6e35 -r 02e84edaa143 source/common/framedata.h
> --- a/source/common/framedata.h Mon Aug 03 14:56:21 2015 -0500
> +++ b/source/common/framedata.h Mon Aug 03 17:55:58 2015 -0500
> @@ -27,6 +27,7 @@
> #include "common.h"
> #include "slice.h"
> #include "cudata.h"
> +#include "threading.h"
>
> namespace X265_NS {
> // private namespace
> @@ -83,6 +84,21 @@
> }
> };
>
> +struct PerNodeRecon
> +{
> + PicYuv* reconPic;
> + Lock copyLock;
> + int rows;
> +
> + PerNodeRecon()
> + {
> + rows = 0;
> + reconPic = NULL;
> + }
> +
> + ~PerNodeRecon();
> +};
> +
> /* Per-frame data that is used during encodes and referenced while the
> picture
> * is available for reference. A FrameData instance is attached to a
> Frame as it
> * comes out of the lookahead. Frames which are not being encoded do not
> have a
> @@ -93,13 +109,14 @@
> class FrameData
> {
> public:
> + PerNodeRecon* m_nodes;
> + int m_ownerNode; /* NUMA node of writing frame
> encoder */
>
> Slice* m_slice;
> SAOParam* m_saoParam;
> - x265_param* m_param;
> + const x265_param* m_param;
>
> FrameData* m_freeListNext;
> - PicYuv* m_reconPic;
> bool m_bHasReferences; /* used during DPB/RPS updates */
> int m_frameEncoderID; /* the ID of the FrameEncoder
> encoding this frame */
> JobProvider* m_jobProvider;
> @@ -142,11 +159,23 @@
>
> FrameData();
>
> - bool create(x265_param *param, const SPS& sps);
> + bool create(const x265_param& param, const SPS& sps);
> + bool allocRecon(const SPS& sps, int node);
> void reinit(const SPS& sps);
> void destroy();
>
> - CUData* getPicCTU(uint32_t ctuAddr) { return &m_picCTU[ctuAddr]; }
> + inline CUData* getPicCTU(uint32_t ctuAddr) { return
> &m_picCTU[ctuAddr]; }
> +
> +#if (defined(_WIN32_WINNT) && _WIN32_WINNT >= _WIN32_WINNT_WIN7) ||
> HAVE_LIBNUMA
> + /* this function should only be called by worker threads which are
> encoding this particular
> + * frame. workers which are using this frame as a motion reference
> should use their slice's
> + * m_refReconPicList[][] array instead. Its pointers will always
> reference a buffer which
> + * was allocated on their local socket */
> + /* TODO: X265_CHECK that current node == m_ownerNode */
> + inline PicYuv* getOutputRecon() { return
> m_nodes[m_ownerNode].reconPic; }
> +#else
> + inline PicYuv* getOutputRecon() { return m_nodes->reconPic; }
> +#endif
> };
> }
>
> diff -r 1f161d9c6e35 -r 02e84edaa143 source/common/predict.cpp
> --- a/source/common/predict.cpp Mon Aug 03 14:56:21 2015 -0500
> +++ b/source/common/predict.cpp Mon Aug 03 17:55:58 2015 -0500
> @@ -109,18 +109,18 @@
> ShortYuv& shortYuv = m_predShortYuv[0];
>
> if (bLuma)
> - predInterLumaShort(pu, shortYuv,
> *cu.m_slice->m_refPicList[0][refIdx0]->m_reconPic, mv0);
> + predInterLumaShort(pu, shortYuv,
> *cu.m_slice->m_refReconPicList[0][refIdx0], mv0);
> if (bChroma)
> - predInterChromaShort(pu, shortYuv,
> *cu.m_slice->m_refPicList[0][refIdx0]->m_reconPic, mv0);
> + predInterChromaShort(pu, shortYuv,
> *cu.m_slice->m_refReconPicList[0][refIdx0], mv0);
>
> addWeightUni(pu, predYuv, shortYuv, wv0, bLuma, bChroma);
> }
> else
> {
> if (bLuma)
> - predInterLumaPixel(pu, predYuv,
> *cu.m_slice->m_refPicList[0][refIdx0]->m_reconPic, mv0);
> + predInterLumaPixel(pu, predYuv,
> *cu.m_slice->m_refReconPicList[0][refIdx0], mv0);
> if (bChroma)
> - predInterChromaPixel(pu, predYuv,
> *cu.m_slice->m_refPicList[0][refIdx0]->m_reconPic, mv0);
> + predInterChromaPixel(pu, predYuv,
> *cu.m_slice->m_refReconPicList[0][refIdx0], mv0);
> }
> }
> else
> @@ -179,13 +179,13 @@
>
> if (bLuma)
> {
> - predInterLumaShort(pu, m_predShortYuv[0],
> *cu.m_slice->m_refPicList[0][refIdx0]->m_reconPic, mv0);
> - predInterLumaShort(pu, m_predShortYuv[1],
> *cu.m_slice->m_refPicList[1][refIdx1]->m_reconPic, mv1);
> + predInterLumaShort(pu, m_predShortYuv[0],
> *cu.m_slice->m_refReconPicList[0][refIdx0], mv0);
> + predInterLumaShort(pu, m_predShortYuv[1],
> *cu.m_slice->m_refReconPicList[1][refIdx1], mv1);
> }
> if (bChroma)
> {
> - predInterChromaShort(pu, m_predShortYuv[0],
> *cu.m_slice->m_refPicList[0][refIdx0]->m_reconPic, mv0);
> - predInterChromaShort(pu, m_predShortYuv[1],
> *cu.m_slice->m_refPicList[1][refIdx1]->m_reconPic, mv1);
> + predInterChromaShort(pu, m_predShortYuv[0],
> *cu.m_slice->m_refReconPicList[0][refIdx0], mv0);
> + predInterChromaShort(pu, m_predShortYuv[1],
> *cu.m_slice->m_refReconPicList[1][refIdx1], mv1);
> }
>
> if (pwp0 && pwp1 && (pwp0->bPresentFlag ||
> pwp1->bPresentFlag))
> @@ -203,18 +203,18 @@
> ShortYuv& shortYuv = m_predShortYuv[0];
>
> if (bLuma)
> - predInterLumaShort(pu, shortYuv,
> *cu.m_slice->m_refPicList[0][refIdx0]->m_reconPic, mv0);
> + predInterLumaShort(pu, shortYuv,
> *cu.m_slice->m_refReconPicList[0][refIdx0], mv0);
> if (bChroma)
> - predInterChromaShort(pu, shortYuv,
> *cu.m_slice->m_refPicList[0][refIdx0]->m_reconPic, mv0);
> + predInterChromaShort(pu, shortYuv,
> *cu.m_slice->m_refReconPicList[0][refIdx0], mv0);
>
> addWeightUni(pu, predYuv, shortYuv, wv0, bLuma, bChroma);
> }
> else
> {
> if (bLuma)
> - predInterLumaPixel(pu, predYuv,
> *cu.m_slice->m_refPicList[0][refIdx0]->m_reconPic, mv0);
> + predInterLumaPixel(pu, predYuv,
> *cu.m_slice->m_refReconPicList[0][refIdx0], mv0);
> if (bChroma)
> - predInterChromaPixel(pu, predYuv,
> *cu.m_slice->m_refPicList[0][refIdx0]->m_reconPic, mv0);
> + predInterChromaPixel(pu, predYuv,
> *cu.m_slice->m_refReconPicList[0][refIdx0], mv0);
> }
> }
> else
> @@ -230,18 +230,18 @@
> ShortYuv& shortYuv = m_predShortYuv[0];
>
> if (bLuma)
> - predInterLumaShort(pu, shortYuv,
> *cu.m_slice->m_refPicList[1][refIdx1]->m_reconPic, mv1);
> + predInterLumaShort(pu, shortYuv,
> *cu.m_slice->m_refReconPicList[1][refIdx1], mv1);
> if (bChroma)
> - predInterChromaShort(pu, shortYuv,
> *cu.m_slice->m_refPicList[1][refIdx1]->m_reconPic, mv1);
> + predInterChromaShort(pu, shortYuv,
> *cu.m_slice->m_refReconPicList[1][refIdx1], mv1);
>
> addWeightUni(pu, predYuv, shortYuv, wv0, bLuma, bChroma);
> }
> else
> {
> if (bLuma)
> - predInterLumaPixel(pu, predYuv,
> *cu.m_slice->m_refPicList[1][refIdx1]->m_reconPic, mv1);
> + predInterLumaPixel(pu, predYuv,
> *cu.m_slice->m_refReconPicList[1][refIdx1], mv1);
> if (bChroma)
> - predInterChromaPixel(pu, predYuv,
> *cu.m_slice->m_refPicList[1][refIdx1]->m_reconPic, mv1);
> + predInterChromaPixel(pu, predYuv,
> *cu.m_slice->m_refReconPicList[1][refIdx1], mv1);
> }
> }
> }
> @@ -600,8 +600,9 @@
> int tuSize = 1 << intraNeighbors.log2TrSize;
> int tuSize2 = tuSize << 1;
>
> - pixel* adiOrigin = cu.m_encData->m_reconPic->getLumaAddr(cu.m_cuAddr,
> cuGeom.absPartIdx + puAbsPartIdx);
> - intptr_t picStride = cu.m_encData->m_reconPic->m_stride;
> + PicYuv* reconPic = cu.m_encData->getOutputRecon();
> + pixel* adiOrigin = reconPic->getLumaAddr(cu.m_cuAddr,
> cuGeom.absPartIdx + puAbsPartIdx);
> + intptr_t picStride = reconPic->m_stride;
>
> fillReferenceSamples(adiOrigin, picStride, intraNeighbors,
> intraNeighbourBuf[0]);
>
> @@ -648,8 +649,9 @@
>
> void Predict::initAdiPatternChroma(const CUData& cu, const CUGeom&
> cuGeom, uint32_t puAbsPartIdx, const IntraNeighbors& intraNeighbors,
> uint32_t chromaId)
> {
> - const pixel* adiOrigin =
> cu.m_encData->m_reconPic->getChromaAddr(chromaId, cu.m_cuAddr,
> cuGeom.absPartIdx + puAbsPartIdx);
> - intptr_t picStride = cu.m_encData->m_reconPic->m_strideC;
> + PicYuv* reconPic = cu.m_encData->getOutputRecon();
> + const pixel* adiOrigin = reconPic->getChromaAddr(chromaId,
> cu.m_cuAddr, cuGeom.absPartIdx + puAbsPartIdx);
> + intptr_t picStride = reconPic->m_strideC;
>
> fillReferenceSamples(adiOrigin, picStride, intraNeighbors,
> intraNeighbourBuf[0]);
>
> diff -r 1f161d9c6e35 -r 02e84edaa143 source/common/slice.cpp
> --- a/source/common/slice.cpp Mon Aug 03 14:56:21 2015 -0500
> +++ b/source/common/slice.cpp Mon Aug 03 17:55:58 2015 -0500
> @@ -33,7 +33,9 @@
> {
> if (m_sliceType == I_SLICE)
> {
> - memset(m_refPicList, 0, sizeof(m_refPicList));
> + memset(m_refFrameList, 0, sizeof(m_refFrameList));
> + memset(m_refReconPicList, 0, sizeof(m_refReconPicList));
> + memset(m_refPOCList, 0, sizeof(m_refPOCList));
> m_numRefIdx[1] = m_numRefIdx[0] = 0;
> return;
> }
> @@ -106,13 +108,13 @@
> {
> cIdx = rIdx % numPocTotalCurr;
> X265_CHECK(cIdx >= 0 && cIdx < numPocTotalCurr, "RPS index check
> fail\n");
> - m_refPicList[0][rIdx] = rpsCurrList0[cIdx];
> + m_refFrameList[0][rIdx] = rpsCurrList0[cIdx];
> }
>
> if (m_sliceType != B_SLICE)
> {
> m_numRefIdx[1] = 0;
> - memset(m_refPicList[1], 0, sizeof(m_refPicList[1]));
> + memset(m_refFrameList[1], 0, sizeof(m_refFrameList[1]));
> }
> else
> {
> @@ -120,13 +122,13 @@
> {
> cIdx = rIdx % numPocTotalCurr;
> X265_CHECK(cIdx >= 0 && cIdx < numPocTotalCurr, "RPS index
> check fail\n");
> - m_refPicList[1][rIdx] = rpsCurrList1[cIdx];
> + m_refFrameList[1][rIdx] = rpsCurrList1[cIdx];
> }
> }
>
> for (int dir = 0; dir < 2; dir++)
> for (int numRefIdx = 0; numRefIdx < m_numRefIdx[dir]; numRefIdx++)
> - m_refPOCList[dir][numRefIdx] =
> m_refPicList[dir][numRefIdx]->m_poc;
> + m_refPOCList[dir][numRefIdx] =
> m_refFrameList[dir][numRefIdx]->m_poc;
> }
>
> void Slice::disableWeights()
> diff -r 1f161d9c6e35 -r 02e84edaa143 source/common/slice.h
> --- a/source/common/slice.h Mon Aug 03 14:56:21 2015 -0500
> +++ b/source/common/slice.h Mon Aug 03 17:55:58 2015 -0500
> @@ -31,6 +31,7 @@
>
> class Frame;
> class PicList;
> +class PicYuv;
> class MotionReference;
>
> enum SliceType
> @@ -209,6 +210,13 @@
>
> struct SPS
> {
> + /* cached PicYuv offset arrays, shared by all instances of
> + * PicYuv created by this encoder */
> + intptr_t* cuOffsetY;
> + intptr_t* cuOffsetC;
> + intptr_t* buOffsetY;
> + intptr_t* buOffsetC;
> +
> int chromaFormatIdc; // use param
> uint32_t picWidthInLumaSamples; // use param
> uint32_t picHeightInLumaSamples; // use param
> @@ -242,6 +250,11 @@
>
> Window conformanceWindow;
> VUI vuiParameters;
> +
> + SPS()
> + {
> + memset(this, 0, sizeof(*this));
> + }
> };
>
> struct PPS
> @@ -321,7 +334,8 @@
> uint32_t m_colRefIdx; // never modified
>
> int m_numRefIdx[2];
> - Frame* m_refPicList[2][MAX_NUM_REF + 1];
> + Frame* m_refFrameList[2][MAX_NUM_REF + 1];
> + PicYuv* m_refReconPicList[2][MAX_NUM_REF + 1];
> int m_refPOCList[2][MAX_NUM_REF + 1];
>
> uint32_t m_maxNumMergeCand; // use param
> @@ -332,14 +346,9 @@
> m_lastIDR = 0;
> m_sLFaseFlag = true;
> m_numRefIdx[0] = m_numRefIdx[1] = 0;
> - for (int i = 0; i < MAX_NUM_REF; i++)
> - {
> - m_refPicList[0][i] = NULL;
> - m_refPicList[1][i] = NULL;
> - m_refPOCList[0][i] = 0;
> - m_refPOCList[1][i] = 0;
> - }
> -
> + memset(m_refFrameList, 0, sizeof(m_refFrameList));
> + memset(m_refReconPicList, 0, sizeof(m_refReconPicList));
> + memset(m_refPOCList, 0, sizeof(m_refPOCList));
> disableWeights();
> }
>
> @@ -347,8 +356,6 @@
>
> void setRefPicList(PicList& picList);
>
> - const Frame* getRefPic(int list, int refIdx) const { return refIdx >=
> 0 ? m_refPicList[list][refIdx] : NULL; }
> -
> bool getRapPicFlag() const
> {
> return m_nalUnitType == NAL_UNIT_CODED_SLICE_IDR_W_RADL
> diff -r 1f161d9c6e35 -r 02e84edaa143 source/encoder/analysis.cpp
> --- a/source/encoder/analysis.cpp Mon Aug 03 14:56:21 2015 -0500
> +++ b/source/encoder/analysis.cpp Mon Aug 03 17:55:58 2015 -0500
> @@ -173,8 +173,8 @@
> if (!m_param->rdLevel)
> {
> /* In RD Level 0/1, copy source pixels into the reconstructed
> block so
> - * they are available for intra predictions */
> - m_modeDepth[0].fencYuv.copyToPicYuv(*m_frame->m_reconPic,
> ctu.m_cuAddr, 0);
> + * they are available for intra predictions */
> +
> m_modeDepth[0].fencYuv.copyToPicYuv(*m_frame->m_encData->getOutputRecon(),
> ctu.m_cuAddr, 0);
>
> compressInterCU_rd0_4(ctu, cuGeom, qp);
>
> @@ -337,7 +337,7 @@
> /* Copy best data to encData CTU and recon */
> md.bestMode->cu.copyToPic(depth);
> if (md.bestMode != &md.pred[PRED_SPLIT])
> - md.bestMode->reconYuv.copyToPicYuv(*m_frame->m_reconPic,
> parentCTU.m_cuAddr, cuGeom.absPartIdx);
> +
> md.bestMode->reconYuv.copyToPicYuv(*m_frame->m_encData->getOutputRecon(),
> parentCTU.m_cuAddr, cuGeom.absPartIdx);
> }
>
> void Analysis::PMODE::processTasks(int workerThreadId)
> @@ -747,7 +747,7 @@
> /* Copy best data to encData CTU and recon */
> md.bestMode->cu.copyToPic(depth);
> if (md.bestMode != &md.pred[PRED_SPLIT])
> - md.bestMode->reconYuv.copyToPicYuv(*m_frame->m_reconPic, cuAddr,
> cuGeom.absPartIdx);
> +
> md.bestMode->reconYuv.copyToPicYuv(*m_frame->m_encData->getOutputRecon(),
> cuAddr, cuGeom.absPartIdx);
> }
>
> uint32_t Analysis::compressInterCU_rd0_4(const CUData& parentCTU, const
> CUGeom& cuGeom, int32_t qp)
> @@ -757,6 +757,8 @@
> ModeDepth& md = m_modeDepth[depth];
> md.bestMode = NULL;
>
> + PicYuv& reconPic = *m_frame->m_encData->getOutputRecon();
> +
> bool mightSplit = !(cuGeom.flags & CUGeom::LEAF);
> bool mightNotSplit = !(cuGeom.flags & CUGeom::SPLIT_MANDATORY);
> uint32_t minDepth = topSkipMinDepth(parentCTU, cuGeom);
> @@ -1051,7 +1053,7 @@
> residualTransformQuantIntra(*md.bestMode, cuGeom,
> 0, 0, tuDepthRange);
> getBestIntraModeChroma(*md.bestMode, cuGeom);
> residualQTIntraChroma(*md.bestMode, cuGeom, 0, 0);
> -
> md.bestMode->reconYuv.copyFromPicYuv(*m_frame->m_reconPic, cu.m_cuAddr,
> cuGeom.absPartIdx); // TODO:
> + md.bestMode->reconYuv.copyFromPicYuv(reconPic,
> cu.m_cuAddr, cuGeom.absPartIdx); // TODO:
> }
> }
> }
> @@ -1107,7 +1109,7 @@
> X265_CHECK(md.bestMode->ok(), "best mode is not ok");
> md.bestMode->cu.copyToPic(depth);
> if (m_param->rdLevel)
> - md.bestMode->reconYuv.copyToPicYuv(*m_frame->m_reconPic, cuAddr,
> cuGeom.absPartIdx);
> + md.bestMode->reconYuv.copyToPicYuv(reconPic, cuAddr,
> cuGeom.absPartIdx);
>
> return refMask;
> }
> @@ -1356,7 +1358,7 @@
> /* Copy best data to encData CTU and recon */
> X265_CHECK(md.bestMode->ok(), "best mode is not ok");
> md.bestMode->cu.copyToPic(depth);
> - md.bestMode->reconYuv.copyToPicYuv(*m_frame->m_reconPic,
> parentCTU.m_cuAddr, cuGeom.absPartIdx);
> +
> md.bestMode->reconYuv.copyToPicYuv(*m_frame->m_encData->getOutputRecon(),
> parentCTU.m_cuAddr, cuGeom.absPartIdx);
>
> return refMask;
> }
> @@ -1851,6 +1853,8 @@
>
> cu.copyFromPic(ctu, cuGeom);
>
> + PicYuv& reconPic = *m_frame->m_encData->getOutputRecon();
> +
> Yuv& fencYuv = m_modeDepth[cuGeom.depth].fencYuv;
> if (cuGeom.depth)
> m_modeDepth[0].fencYuv.copyPartToYuv(fencYuv, absPartIdx);
> @@ -1906,7 +1910,6 @@
> /* residualTransformQuantInter() wrote transformed residual back
> into
> * resiYuv. Generate the recon pixels by adding it to the
> prediction */
>
> - PicYuv& reconPic = *m_frame->m_reconPic;
> if (cu.m_cbf[0][0])
> primitives.cu[sizeIdx].add_ps(reconPic.getLumaAddr(cu.m_cuAddr,
> absPartIdx), reconPic.m_stride,
> predY, resiYuv.m_buf[0],
> predYuv.m_size, resiYuv.m_size);
> @@ -1969,7 +1972,7 @@
> if (m_slice->m_numRefIdx[0])
> {
> numRefs++;
> - const CUData& cu =
> *m_slice->m_refPicList[0][0]->m_encData->getPicCTU(parentCTU.m_cuAddr);
> + const CUData& cu =
> *m_slice->m_refFrameList[0][0]->m_encData->getPicCTU(parentCTU.m_cuAddr);
> previousQP = cu.m_qp[0];
> if (!cu.m_cuDepth[cuGeom.absPartIdx])
> return 0;
> @@ -1983,7 +1986,7 @@
> if (m_slice->m_numRefIdx[1])
> {
> numRefs++;
> - const CUData& cu =
> *m_slice->m_refPicList[1][0]->m_encData->getPicCTU(parentCTU.m_cuAddr);
> + const CUData& cu =
> *m_slice->m_refFrameList[1][0]->m_encData->getPicCTU(parentCTU.m_cuAddr);
> if (!cu.m_cuDepth[cuGeom.absPartIdx])
> return 0;
> for (uint32_t i = 0; i < cuGeom.numPartitions; i += 4)
> diff -r 1f161d9c6e35 -r 02e84edaa143 source/encoder/dpb.cpp
> --- a/source/encoder/dpb.cpp Mon Aug 03 14:56:21 2015 -0500
> +++ b/source/encoder/dpb.cpp Mon Aug 03 17:55:58 2015 -0500
> @@ -47,16 +47,12 @@
> delete curFrame;
> }
>
> - while (m_picSymFreeList)
> + while (m_frameDataFreeList)
> {
> - FrameData* next = m_picSymFreeList->m_freeListNext;
> - m_picSymFreeList->destroy();
> -
> - m_picSymFreeList->m_reconPic->destroy();
> - delete m_picSymFreeList->m_reconPic;
> -
> - delete m_picSymFreeList;
> - m_picSymFreeList = next;
> + FrameData* next = m_frameDataFreeList->m_freeListNext;
> + m_frameDataFreeList->destroy();
> + delete m_frameDataFreeList;
> + m_frameDataFreeList = next;
> }
> }
>
> @@ -79,10 +75,9 @@
> iterFrame = m_picList.first();
>
> m_freeList.pushBack(*curFrame);
> - curFrame->m_encData->m_freeListNext = m_picSymFreeList;
> - m_picSymFreeList = curFrame->m_encData;
> + curFrame->m_encData->m_freeListNext = m_frameDataFreeList;
> + m_frameDataFreeList = curFrame->m_encData;
> curFrame->m_encData = NULL;
> - curFrame->m_reconPic = NULL;
> }
> }
> }
> @@ -171,7 +166,7 @@
> {
> for (int ref = 0; ref < slice->m_numRefIdx[l]; ref++)
> {
> - Frame *refpic = slice->m_refPicList[l][ref];
> + Frame *refpic = slice->m_refFrameList[l][ref];
> ATOMIC_INC(&refpic->m_countRefEncoders);
> }
> }
> diff -r 1f161d9c6e35 -r 02e84edaa143 source/encoder/dpb.h
> --- a/source/encoder/dpb.h Mon Aug 03 14:56:21 2015 -0500
> +++ b/source/encoder/dpb.h Mon Aug 03 17:55:58 2015 -0500
> @@ -46,14 +46,14 @@
> bool m_bTemporalSublayer;
> PicList m_picList;
> PicList m_freeList;
> - FrameData* m_picSymFreeList;
> + FrameData* m_frameDataFreeList;
>
> DPB(x265_param *param)
> {
> m_lastIDR = 0;
> m_pocCRA = 0;
> m_bRefreshPending = false;
> - m_picSymFreeList = NULL;
> + m_frameDataFreeList = NULL;
> m_maxRefL0 = param->maxNumReferences;
> m_maxRefL1 = param->bBPyramid ? 2 : 1;
> m_bOpenGOP = param->bOpenGOP;
> diff -r 1f161d9c6e35 -r 02e84edaa143 source/encoder/encoder.cpp
> --- a/source/encoder/encoder.cpp Mon Aug 03 14:56:21 2015 -0500
> +++ b/source/encoder/encoder.cpp Mon Aug 03 17:55:58 2015 -0500
> @@ -66,10 +66,6 @@
> m_outputCount = 0;
> m_param = NULL;
> m_latestParam = NULL;
> - m_cuOffsetY = NULL;
> - m_cuOffsetC = NULL;
> - m_buOffsetY = NULL;
> - m_buOffsetC = NULL;
> m_threadPool = NULL;
> m_analysisFile = NULL;
> for (int i = 0; i < X265_MAX_FRAME_THREADS; i++)
> @@ -318,10 +314,10 @@
> delete m_rateControl;
> }
>
> - X265_FREE(m_cuOffsetY);
> - X265_FREE(m_cuOffsetC);
> - X265_FREE(m_buOffsetY);
> - X265_FREE(m_buOffsetC);
> + X265_FREE(m_sps.cuOffsetY);
> + X265_FREE(m_sps.cuOffsetC);
> + X265_FREE(m_sps.buOffsetY);
> + X265_FREE(m_sps.buOffsetC);
>
> if (m_analysisFile)
> fclose(m_analysisFile);
> @@ -416,12 +412,12 @@
> /* the first PicYuv created is asked to generate the CU
> and block unit offset
> * arrays which are then shared with all subsequent
> PicYuv (orig and recon)
> * allocated by this top level encoder */
> - if (m_cuOffsetY)
> + if (m_sps.cuOffsetY)
> {
> - inFrame->m_fencPic->m_cuOffsetC = m_cuOffsetC;
> - inFrame->m_fencPic->m_cuOffsetY = m_cuOffsetY;
> - inFrame->m_fencPic->m_buOffsetC = m_buOffsetC;
> - inFrame->m_fencPic->m_buOffsetY = m_buOffsetY;
> + inFrame->m_fencPic->m_cuOffsetC = m_sps.cuOffsetC;
> + inFrame->m_fencPic->m_cuOffsetY = m_sps.cuOffsetY;
> + inFrame->m_fencPic->m_buOffsetC = m_sps.buOffsetC;
> + inFrame->m_fencPic->m_buOffsetY = m_sps.buOffsetY;
> }
> else
> {
> @@ -435,10 +431,10 @@
> }
> else
> {
> - m_cuOffsetC = inFrame->m_fencPic->m_cuOffsetC;
> - m_cuOffsetY = inFrame->m_fencPic->m_cuOffsetY;
> - m_buOffsetC = inFrame->m_fencPic->m_buOffsetC;
> - m_buOffsetY = inFrame->m_fencPic->m_buOffsetY;
> + m_sps.cuOffsetC = inFrame->m_fencPic->m_cuOffsetC;
> + m_sps.cuOffsetY = inFrame->m_fencPic->m_cuOffsetY;
> + m_sps.buOffsetC = inFrame->m_fencPic->m_buOffsetC;
> + m_sps.buOffsetY = inFrame->m_fencPic->m_buOffsetY;
> }
> }
> }
> @@ -538,7 +534,7 @@
>
> if (pic_out)
> {
> - PicYuv *recpic = outFrame->m_reconPic;
> + PicYuv *recpic = outFrame->m_encData->getOutputRecon();
> pic_out->poc = slice->m_poc;
> pic_out->bitDepth = X265_DEPTH;
> pic_out->userData = outFrame->m_userData;
> @@ -633,10 +629,10 @@
> if (frameEnc && !pass)
> {
> /* give this frame a FrameData instance before encoding */
> - if (m_dpb->m_picSymFreeList)
> + if (m_dpb->m_frameDataFreeList)
> {
> - frameEnc->m_encData = m_dpb->m_picSymFreeList;
> - m_dpb->m_picSymFreeList =
> m_dpb->m_picSymFreeList->m_freeListNext;
> + frameEnc->m_encData = m_dpb->m_frameDataFreeList;
> + m_dpb->m_frameDataFreeList =
> m_dpb->m_frameDataFreeList->m_freeListNext;
> frameEnc->reinit(m_sps);
> }
> else
> @@ -647,10 +643,6 @@
> slice->m_pps = &m_pps;
> slice->m_maxNumMergeCand = m_param->maxNumMergeCand;
> slice->m_endCUAddr =
> slice->realEndAddress(m_sps.numCUsInFrame * NUM_4x4_PARTITIONS);
> - frameEnc->m_reconPic->m_cuOffsetC = m_cuOffsetC;
> - frameEnc->m_reconPic->m_cuOffsetY = m_cuOffsetY;
> - frameEnc->m_reconPic->m_buOffsetC = m_buOffsetC;
> - frameEnc->m_reconPic->m_buOffsetY = m_buOffsetY;
> }
>
> curEncoder->m_rce.encodeOrder = m_encodedFrameNum++;
> @@ -1054,7 +1046,7 @@
>
> void Encoder::finishFrameStats(Frame* curFrame, FrameEncoder *curEncoder,
> uint64_t bits, x265_frame_stats* frameStats)
> {
> - PicYuv* reconPic = curFrame->m_reconPic;
> + PicYuv *reconPic = curFrame->m_encData->getOutputRecon();
>
> //===== calculate PSNR =====
> int width = reconPic->m_picWidth -
> m_sps.conformanceWindow.rightOffset;
> diff -r 1f161d9c6e35 -r 02e84edaa143 source/encoder/encoder.h
> --- a/source/encoder/encoder.h Mon Aug 03 14:56:21 2015 -0500
> +++ b/source/encoder/encoder.h Mon Aug 03 17:55:58 2015 -0500
> @@ -93,13 +93,6 @@
> int m_numPools;
> int m_curEncoder;
>
> - /* cached PicYuv offset arrays, shared by all instances of
> - * PicYuv created by this encoder */
> - intptr_t* m_cuOffsetY;
> - intptr_t* m_cuOffsetC;
> - intptr_t* m_buOffsetY;
> - intptr_t* m_buOffsetC;
> -
> /* Collect statistics globally */
> EncStats m_analyzeAll;
> EncStats m_analyzeI;
> diff -r 1f161d9c6e35 -r 02e84edaa143 source/encoder/frameencoder.cpp
> --- a/source/encoder/frameencoder.cpp Mon Aug 03 14:56:21 2015 -0500
> +++ b/source/encoder/frameencoder.cpp Mon Aug 03 17:55:58 2015 -0500
> @@ -329,6 +329,16 @@
> if (m_frame->m_lowres.bKeyframe && m_param->bRepeatHeaders)
> m_top->getStreamHeaders(m_nalList, m_entropyCoder, m_bs);
>
> +#if (defined(_WIN32_WINNT) && _WIN32_WINNT >= _WIN32_WINNT_WIN7) ||
> HAVE_LIBNUMA
> + int numaNode = 0; // m_pool ? m_pool->m_numaNode : 0;
> +#else
> + int numaNode = 0;
> +#endif
> +
> + /* Claim this frame as being encoded by this NUMA node */
> + m_frame->m_encData->allocRecon(m_top->m_sps, numaNode); /* TODO: bail
> if failure */
> + m_frame->m_encData->m_ownerNode = numaNode;
> +
> // 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;
> @@ -357,7 +367,9 @@
> 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);
> +
> slice->m_refFrameList[l][ref]->m_encData->allocRecon(m_top->m_sps,
> numaNode); /* TODO: bail if failure */
> + slice->m_refReconPicList[l][ref] =
> slice->m_refFrameList[l][ref]->m_encData->m_nodes[numaNode].reconPic;
> + m_mref[l][ref].init(slice->m_refReconPicList[l][ref], w,
> *m_param);
> }
> }
>
> @@ -477,7 +489,7 @@
> /* 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
> + * RateControlEnd here, after the slice contexts are initialized. For
> the rest - ABR
> * and VBV, unlock only after rateControlUpdateStats of this frame is
> called */
> if (m_param->rc.rateControlMode != X265_RC_ABR &&
> !m_top->m_rateControl->m_isVbv)
> {
> @@ -501,12 +513,14 @@
> {
> for (int ref = 0; ref < slice->m_numRefIdx[l]; ref++)
> {
> - Frame *refpic = slice->m_refPicList[l][ref];
> + Frame *refpic = slice->m_refFrameList[l][ref];
>
> uint32_t reconRowCount =
> refpic->m_reconRowCount.get();
> while ((reconRowCount != m_numRows) && (reconRowCount
> < row + m_refLagRows))
> reconRowCount =
> refpic->m_reconRowCount.waitForChange(reconRowCount);
>
> + /* TODO: if refpic->m_encData->m_ownerNode !=
> numaNode, copy rows */
> +
> if ((bUseWeightP || bUseWeightB) &&
> m_mref[l][ref].isWeighted)
> m_mref[l][ref].applyWeight(row + m_refLagRows,
> m_numRows);
> }
> @@ -540,12 +554,14 @@
> int list = l;
> for (int ref = 0; ref < slice->m_numRefIdx[list];
> ref++)
> {
> - Frame *refpic = slice->m_refPicList[list][ref];
> + Frame *refpic = slice->m_refFrameList[list][ref];
>
> uint32_t reconRowCount =
> refpic->m_reconRowCount.get();
> while ((reconRowCount != m_numRows) &&
> (reconRowCount < i + m_refLagRows))
> reconRowCount =
> refpic->m_reconRowCount.waitForChange(reconRowCount);
>
> + /* TODO: if refpic->m_encData->m_ownerNode !=
> numaNode, copy rows */
> +
> if ((bUseWeightP || bUseWeightB) &&
> m_mref[l][ref].isWeighted)
> m_mref[list][ref].applyWeight(i +
> m_refLagRows, m_numRows);
> }
> @@ -697,7 +713,7 @@
> {
> for (int ref = 0; ref < slice->m_numRefIdx[l]; ref++)
> {
> - Frame *refpic = slice->m_refPicList[l][ref];
> + Frame *refpic = slice->m_refFrameList[l][ref];
> ATOMIC_DEC(&refpic->m_countRefEncoders);
> }
> }
> diff -r 1f161d9c6e35 -r 02e84edaa143 source/encoder/framefilter.cpp
> --- a/source/encoder/framefilter.cpp Mon Aug 03 14:56:21 2015 -0500
> +++ b/source/encoder/framefilter.cpp Mon Aug 03 17:55:58 2015 -0500
> @@ -160,7 +160,7 @@
>
> void FrameFilter::processRowPost(int row)
> {
> - PicYuv *reconPic = m_frame->m_reconPic;
> + PicYuv *reconPic = m_frame->m_encData->getOutputRecon();
> const uint32_t numCols =
> m_frame->m_encData->m_slice->m_sps->numCuInWidth;
> const uint32_t lineStartCUAddr = row * numCols;
> const int realH = getCUHeight(row);
> @@ -233,10 +233,10 @@
> }
> if (m_param->bEnableSsim && m_ssimBuf)
> {
> - pixel *rec = m_frame->m_reconPic->m_picOrg[0];
> + pixel *rec = reconPic->m_picOrg[0];
> pixel *fenc = m_frame->m_fencPic->m_picOrg[0];
> - intptr_t stride1 = m_frame->m_fencPic->m_stride;
> - intptr_t stride2 = m_frame->m_reconPic->m_stride;
> + intptr_t stride1 = reconPic->m_stride;
> + intptr_t stride2 = m_frame->m_fencPic->m_stride;
> uint32_t bEnd = ((row + 1) == (this->m_numRows - 1));
> uint32_t bStart = (row == 0);
> uint32_t minPixY = row * g_maxCUSize - 4 * !bStart;
> @@ -407,7 +407,7 @@
> int size = cu->m_log2CUSize[absPartIdx] - 2;
> uint32_t cuAddr = cu->m_cuAddr;
>
> - PicYuv* reconPic = frame.m_reconPic;
> + PicYuv* reconPic = frame.m_encData->getOutputRecon();
> PicYuv* fencPic = frame.m_fencPic;
>
> pixel* dst = reconPic->getLumaAddr(cuAddr, absPartIdx);
> diff -r 1f161d9c6e35 -r 02e84edaa143 source/encoder/ratecontrol.cpp
> --- a/source/encoder/ratecontrol.cpp Mon Aug 03 14:56:21 2015 -0500
> +++ b/source/encoder/ratecontrol.cpp Mon Aug 03 17:55:58 2015 -0500
> @@ -1345,10 +1345,10 @@
> {
> /* B-frames don't have independent rate control, but rather get
> the
> * average QP of the two adjacent P-frames + an offset */
> - Slice* prevRefSlice =
> m_curSlice->m_refPicList[0][0]->m_encData->m_slice;
> - Slice* nextRefSlice =
> m_curSlice->m_refPicList[1][0]->m_encData->m_slice;
> - double q0 = m_curSlice->m_refPicList[0][0]->m_encData->m_avgQpRc;
> - double q1 = m_curSlice->m_refPicList[1][0]->m_encData->m_avgQpRc;
> + Slice* prevRefSlice =
> m_curSlice->m_refFrameList[0][0]->m_encData->m_slice;
> + Slice* nextRefSlice =
> m_curSlice->m_refFrameList[1][0]->m_encData->m_slice;
> + double q0 =
> m_curSlice->m_refFrameList[0][0]->m_encData->m_avgQpRc;
> + double q1 =
> m_curSlice->m_refFrameList[1][0]->m_encData->m_avgQpRc;
> bool i0 = prevRefSlice->m_sliceType == I_SLICE;
> bool i1 = nextRefSlice->m_sliceType == I_SLICE;
> int dt0 = abs(m_curSlice->m_poc - prevRefSlice->m_poc);
> @@ -1364,9 +1364,9 @@
> q0 = q1;
> }
> }
> - if (prevRefSlice->m_sliceType == B_SLICE &&
> IS_REFERENCED(m_curSlice->m_refPicList[0][0]))
> + if (prevRefSlice->m_sliceType == B_SLICE &&
> IS_REFERENCED(m_curSlice->m_refFrameList[0][0]))
> q0 -= m_pbOffset / 2;
> - if (nextRefSlice->m_sliceType == B_SLICE &&
> IS_REFERENCED(m_curSlice->m_refPicList[1][0]))
> + if (nextRefSlice->m_sliceType == B_SLICE &&
> IS_REFERENCED(m_curSlice->m_refFrameList[1][0]))
> q1 -= m_pbOffset / 2;
> if (i0 && i1)
> q = (q0 + q1) / 2 + m_ipOffset;
> @@ -1483,7 +1483,7 @@
> * Then bias the quant up or down if total size so far was
> far from
> * the target.
> * Result: Depending on the value of rate_tolerance, there is
> a
> - * tradeoff between quality and bitrate precision. But at
> large
> + * trade-off between quality and bitrate precision. But at
> large
> * tolerances, the bit distribution approaches that of 2pass.
> */
>
> double overflow = 1;
> @@ -1832,7 +1832,7 @@
> double qScale = x265_qp2qScale(qpVbv);
> FrameData& curEncData = *curFrame->m_encData;
> int picType = curEncData.m_slice->m_sliceType;
> - Frame* refFrame = curEncData.m_slice->m_refPicList[0][0];
> + Frame* refFrame = curEncData.m_slice->m_refFrameList[0][0];
>
> uint32_t maxRows = curEncData.m_slice->m_sps->numCuInHeight;
> uint32_t maxCols = curEncData.m_slice->m_sps->numCuInWidth;
> @@ -1921,7 +1921,7 @@
> updatePredictor(rce->rowPred[0], qScaleVbv, (double)rowSatdCost,
> encodedBits);
> if (curEncData.m_slice->m_sliceType == P_SLICE)
> {
> - Frame* refFrame = curEncData.m_slice->m_refPicList[0][0];
> + Frame* refFrame = curEncData.m_slice->m_refFrameList[0][0];
> if (qpVbv < refFrame->m_encData->m_rowStat[row].diagQp)
> {
> uint64_t intraRowSatdCost =
> curEncData.m_rowStat[row].diagIntraSatd;
> diff -r 1f161d9c6e35 -r 02e84edaa143 source/encoder/sao.cpp
> --- a/source/encoder/sao.cpp Mon Aug 03 14:56:21 2015 -0500
> +++ b/source/encoder/sao.cpp Mon Aug 03 17:55:58 2015 -0500
> @@ -243,12 +243,13 @@
> void SAO::processSaoCu(int addr, int typeIdx, int plane)
> {
> int x, y;
> - const CUData* cu = m_frame->m_encData->getPicCTU(addr);
> - pixel* rec = m_frame->m_reconPic->getPlaneAddr(plane, addr);
> - intptr_t stride = plane ? m_frame->m_reconPic->m_strideC :
> m_frame->m_reconPic->m_stride;
> + PicYuv* reconPic = m_frame->m_encData->getOutputRecon();
> + pixel* rec = reconPic->getPlaneAddr(plane, addr);
> + intptr_t stride = plane ? reconPic->m_strideC : reconPic->m_stride;
> uint32_t picWidth = m_param->sourceWidth;
> uint32_t picHeight = m_param->sourceHeight;
> - int ctuWidth = g_maxCUSize;
> + const CUData* cu = m_frame->m_encData->getPicCTU(addr);
> + int ctuWidth = g_maxCUSize;
> int ctuHeight = g_maxCUSize;
> uint32_t lpelx = cu->m_cuPelX;
> uint32_t tpely = cu->m_cuPelY;
> @@ -572,7 +573,8 @@
> /* Process SAO all units */
> void SAO::processSaoUnitRow(SaoCtuParam* ctuParam, int idxY, int plane)
> {
> - intptr_t stride = plane ? m_frame->m_reconPic->m_strideC :
> m_frame->m_reconPic->m_stride;
> + PicYuv* reconPic = m_frame->m_encData->getOutputRecon();
> + intptr_t stride = plane ? reconPic->m_strideC : reconPic->m_stride;
> uint32_t picWidth = m_param->sourceWidth;
> int ctuWidth = g_maxCUSize;
> int ctuHeight = g_maxCUSize;
> @@ -585,12 +587,12 @@
>
> if (!idxY)
> {
> - pixel* rec = m_frame->m_reconPic->m_picOrg[plane];
> + pixel* rec = reconPic->m_picOrg[plane];
> memcpy(m_tmpU1[plane], rec, sizeof(pixel) * picWidth);
> }
>
> int addr = idxY * m_numCuInWidth;
> - pixel* rec = plane ? m_frame->m_reconPic->getChromaAddr(plane, addr)
> : m_frame->m_reconPic->getLumaAddr(addr);
> + pixel* rec = plane ? reconPic->getChromaAddr(plane, addr) :
> reconPic->getLumaAddr(addr);
>
> for (int i = 0; i < ctuHeight + 1; i++)
> {
> @@ -635,7 +637,7 @@
> }
> else if (idxX != (m_numCuInWidth - 1))
> {
> - rec = plane ? m_frame->m_reconPic->getChromaAddr(plane, addr)
> : m_frame->m_reconPic->getLumaAddr(addr);
> + rec = plane ? reconPic->getChromaAddr(plane, addr) :
> reconPic->getLumaAddr(addr);
>
> for (int i = 0; i < ctuHeight + 1; i++)
> {
> @@ -671,12 +673,13 @@
> /* Calculate SAO statistics for current CTU without non-crossing slice */
> void SAO::calcSaoStatsCu(int addr, int plane)
> {
> + const PicYuv* reconPic = m_frame->m_encData->getOutputRecon();
> const CUData* cu = m_frame->m_encData->getPicCTU(addr);
> const pixel* fenc0 = m_frame->m_fencPic->getPlaneAddr(plane, addr);
> - const pixel* rec0 = m_frame->m_reconPic->getPlaneAddr(plane, addr);
> + const pixel* rec0 = reconPic->getPlaneAddr(plane, addr);
> const pixel* fenc;
> const pixel* rec;
> - intptr_t stride = plane ? m_frame->m_reconPic->m_strideC :
> m_frame->m_reconPic->m_stride;
> + intptr_t stride = plane ? reconPic->m_strideC : reconPic->m_stride;
> uint32_t picWidth = m_param->sourceWidth;
> uint32_t picHeight = m_param->sourceHeight;
> int ctuWidth = g_maxCUSize;
> @@ -825,9 +828,10 @@
>
> int x, y;
> const CUData* cu = frame->m_encData->getPicCTU(addr);
> + const PicYuv* reconPic = m_frame->m_encData->getOutputRecon();
> const pixel* fenc;
> const pixel* rec;
> - intptr_t stride = m_frame->m_reconPic->m_stride;
> + intptr_t stride = reconPic->m_stride;
> uint32_t picWidth = m_param->sourceWidth;
> uint32_t picHeight = m_param->sourceHeight;
> int ctuWidth = g_maxCUSize;
> @@ -861,7 +865,7 @@
> {
> if (plane == 1)
> {
> - stride = frame->m_reconPic->m_strideC;
> + stride = reconPic->m_strideC;
> picWidth >>= m_hChromaShift;
> picHeight >>= m_vChromaShift;
> ctuWidth >>= m_hChromaShift;
> @@ -881,7 +885,7 @@
> count = m_countPreDblk[addr][plane][SAO_BO];
>
> const pixel* fenc0 = m_frame->m_fencPic->getPlaneAddr(plane,
> addr);
> - const pixel* rec0 = m_frame->m_reconPic->getPlaneAddr(plane,
> addr);
> + const pixel* rec0 = reconPic->getPlaneAddr(plane, addr);
> fenc = fenc0;
> rec = rec0;
>
> diff -r 1f161d9c6e35 -r 02e84edaa143 source/encoder/search.cpp
> --- a/source/encoder/search.cpp Mon Aug 03 14:56:21 2015 -0500
> +++ b/source/encoder/search.cpp Mon Aug 03 17:55:58 2015 -0500
> @@ -446,8 +446,9 @@
> }
>
> // set reconstruction for next intra prediction blocks if full TU
> prediction won
> - pixel* picReconY = m_frame->m_reconPic->getLumaAddr(cu.m_cuAddr,
> cuGeom.absPartIdx + absPartIdx);
> - intptr_t picStride = m_frame->m_reconPic->m_stride;
> + PicYuv* reconPic = m_frame->m_encData->getOutputRecon();
> + pixel* picReconY = reconPic->getLumaAddr(cu.m_cuAddr,
> cuGeom.absPartIdx + absPartIdx);
> + intptr_t picStride = reconPic->m_stride;
> primitives.cu[sizeIdx].copy_pp(picReconY, picStride, reconQt,
> reconQtStride);
>
> outCost.rdcost += fullCost.rdcost;
> @@ -611,8 +612,9 @@
> }
>
> // set reconstruction for next intra prediction blocks
> - pixel* picReconY = m_frame->m_reconPic->getLumaAddr(cu.m_cuAddr,
> cuGeom.absPartIdx + absPartIdx);
> - intptr_t picStride = m_frame->m_reconPic->m_stride;
> + PicYuv* reconPic = m_frame->m_encData->getOutputRecon();
> + pixel* picReconY = reconPic->getLumaAddr(cu.m_cuAddr,
> cuGeom.absPartIdx + absPartIdx);
> + intptr_t picStride = reconPic->m_stride;
> primitives.cu[sizeIdx].copy_pp(picReconY, picStride, reconQt,
> reconQtStride);
>
> outCost.rdcost += fullCost.rdcost;
> @@ -661,8 +663,9 @@
> uint32_t sizeIdx = log2TrSize - 2;
> primitives.cu[sizeIdx].calcresidual(fenc, pred, residual,
> stride);
>
> - pixel* picReconY =
> m_frame->m_reconPic->getLumaAddr(cu.m_cuAddr, cuGeom.absPartIdx +
> absPartIdx);
> - intptr_t picStride = m_frame->m_reconPic->m_stride;
> + PicYuv* reconPic = m_frame->m_encData->getOutputRecon();
> + pixel* picReconY = reconPic->getLumaAddr(cu.m_cuAddr,
> cuGeom.absPartIdx + absPartIdx);
> + intptr_t picStride = reconPic->m_stride;
>
> uint32_t numSig = m_quant.transformNxN(cu, fenc, stride,
> residual, stride, coeffY, log2TrSize, TEXT_LUMA, absPartIdx, false);
> if (numSig)
> @@ -821,8 +824,9 @@
> coeff_t* coeffC = m_rqt[qtLayer].coeffRQT[chromaId] +
> coeffOffsetC;
> pixel* reconQt =
> m_rqt[qtLayer].reconQtYuv.getChromaAddr(chromaId, absPartIdxC);
> uint32_t reconQtStride = m_rqt[qtLayer].reconQtYuv.m_csize;
> - pixel* picReconC =
> m_frame->m_reconPic->getChromaAddr(chromaId, cu.m_cuAddr, cuGeom.absPartIdx
> + absPartIdxC);
> - intptr_t picStride = m_frame->m_reconPic->m_strideC;
> + PicYuv* reconPic = m_frame->m_encData->getOutputRecon();
> + pixel* picReconC = reconPic->getChromaAddr(chromaId,
> cu.m_cuAddr, cuGeom.absPartIdx + absPartIdxC);
> + intptr_t picStride = reconPic->m_strideC;
>
> uint32_t chromaPredMode = cu.m_chromaIntraDir[absPartIdxC];
> if (chromaPredMode == DM_CHROMA_IDX)
> @@ -998,8 +1002,9 @@
> cu.setCbfPartRange(bCbf << tuDepth, ttype, absPartIdxC,
> tuIterator.absPartIdxStep);
> cu.setTransformSkipPartRange(bTSkip, ttype, absPartIdxC,
> tuIterator.absPartIdxStep);
>
> - pixel* reconPicC =
> m_frame->m_reconPic->getChromaAddr(chromaId, cu.m_cuAddr, cuGeom.absPartIdx
> + absPartIdxC);
> - intptr_t picStride = m_frame->m_reconPic->m_strideC;
> + PicYuv* reconPic = m_frame->m_encData->getOutputRecon();
> + pixel* reconPicC = reconPic->getChromaAddr(chromaId,
> cu.m_cuAddr, cuGeom.absPartIdx + absPartIdxC);
> + intptr_t picStride = reconPic->m_strideC;
> primitives.cu[sizeIdxC].copy_pp(reconPicC, picStride,
> reconQt, reconQtStride);
>
> outDist += bDist;
> @@ -1108,8 +1113,9 @@
> int16_t* residual = resiYuv.getChromaAddr(chromaId,
> absPartIdxC);
> uint32_t coeffOffsetC = absPartIdxC << (LOG2_UNIT_SIZE * 2 -
> (m_hChromaShift + m_vChromaShift));
> coeff_t* coeffC = cu.m_trCoeff[ttype] + coeffOffsetC;
> - pixel* picReconC =
> m_frame->m_reconPic->getChromaAddr(chromaId, cu.m_cuAddr, cuGeom.absPartIdx
> + absPartIdxC);
> - intptr_t picStride = m_frame->m_reconPic->m_strideC;
> + PicYuv* reconPic = m_frame->m_encData->getOutputRecon();
> + pixel* picReconC = reconPic->getChromaAddr(chromaId,
> cu.m_cuAddr, cuGeom.absPartIdx + absPartIdxC);
> + intptr_t picStride = reconPic->m_strideC;
>
> uint32_t chromaPredMode = cu.m_chromaIntraDir[absPartIdxC];
> if (chromaPredMode == DM_CHROMA_IDX)
> @@ -1591,10 +1597,11 @@
> * output recon picture, so it cannot proceed in parallel
> with anything else when doing INTRA_NXN. Also
> * it is not updating m_rdContexts[depth].cur for the later
> PUs which I suspect is slightly wrong. I think
> * that the contexts should be tracked through each PU */
> - pixel* dst =
> m_frame->m_reconPic->getLumaAddr(cu.m_cuAddr, cuGeom.absPartIdx +
> absPartIdx);
> - uint32_t dststride = m_frame->m_reconPic->m_stride;
> - const pixel* src = reconYuv->getLumaAddr(absPartIdx);
> - uint32_t srcstride = reconYuv->m_size;
> + PicYuv* reconPic = m_frame->m_encData->getOutputRecon();
> + pixel* dst = reconPic->getLumaAddr(cu.m_cuAddr,
> cuGeom.absPartIdx + absPartIdx);
> + uint32_t dststride = reconPic->m_stride;
> + const pixel* src = reconYuv->getLumaAddr(absPartIdx);
> + uint32_t srcstride = reconYuv->m_size;
> primitives.cu[log2TrSize - 2].copy_pp(dst, dststride, src,
> srcstride);
> }
> }
> @@ -1757,15 +1764,16 @@
> if (!tuIterator.isLastSection())
> {
> uint32_t zorder = cuGeom.absPartIdx + absPartIdxC;
> - uint32_t dststride = m_frame->m_reconPic->m_strideC;
> + PicYuv* reconPic = m_frame->m_encData->getOutputRecon();
> + uint32_t dststride = reconPic->m_strideC;
> const pixel* src;
> pixel* dst;
>
> - dst = m_frame->m_reconPic->getCbAddr(cu.m_cuAddr, zorder);
> + dst = reconPic->getCbAddr(cu.m_cuAddr, zorder);
> src = reconYuv.getCbAddr(absPartIdxC);
> primitives.chroma[m_csp].cu[size].copy_pp(dst, dststride,
> src, reconYuv.m_csize);
>
> - dst = m_frame->m_reconPic->getCrAddr(cu.m_cuAddr, zorder);
> + dst = reconPic->getCrAddr(cu.m_cuAddr, zorder);
> src = reconYuv.getCrAddr(absPartIdxC);
> primitives.chroma[m_csp].cu[size].copy_pp(dst, dststride,
> src, reconYuv.m_csize);
> }
> @@ -1866,7 +1874,7 @@
> /* find the lowres motion vector from lookahead in middle of current PU */
> MV Search::getLowresMV(const CUData& cu, const PredictionUnit& pu, int
> list, int ref)
> {
> - int diffPoc = abs(m_slice->m_poc -
> m_slice->m_refPicList[list][ref]->m_poc);
> + int diffPoc = abs(m_slice->m_poc - m_slice->m_refPOCList[list][ref]);
> if (diffPoc > m_param->bframes + 1)
> /* poc difference is out of range for lookahead */
> return 0;
> @@ -1906,7 +1914,7 @@
> else
> {
> cu.clipMv(mvCand);
> - predInterLumaPixel(pu, tmpPredYuv,
> *m_slice->m_refPicList[list][ref]->m_reconPic, mvCand);
> + predInterLumaPixel(pu, tmpPredYuv,
> *m_slice->m_refReconPicList[list][ref], mvCand);
> costs[i] =
> m_me.bufSAD(tmpPredYuv.getLumaAddr(pu.puAbsPartIdx), tmpPredYuv.m_size);
> }
> }
> @@ -2197,8 +2205,8 @@
> }
> else
> {
> - PicYuv* refPic0 =
> slice->m_refPicList[0][bestME[0].ref]->m_reconPic;
> - PicYuv* refPic1 =
> slice->m_refPicList[1][bestME[1].ref]->m_reconPic;
> + PicYuv* refPic0 =
> slice->m_refReconPicList[0][bestME[0].ref];
> + PicYuv* refPic1 =
> slice->m_refReconPicList[1][bestME[1].ref];
> Yuv* bidirYuv = m_rqt[cuGeom.depth].bidirPredYuv;
>
> /* Generate reference subpels */
> diff -r 1f161d9c6e35 -r 02e84edaa143 source/encoder/slicetype.cpp
> --- a/source/encoder/slicetype.cpp Mon Aug 03 14:56:21 2015 -0500
> +++ b/source/encoder/slicetype.cpp Mon Aug 03 17:55:58 2015 -0500
> @@ -714,16 +714,16 @@
>
> case P_SLICE:
> b = p1 = poc - l0poc;
> - frames[p0] = &slice->m_refPicList[0][0]->m_lowres;
> + frames[p0] = &slice->m_refFrameList[0][0]->m_lowres;
> frames[b] = &curFrame->m_lowres;
> break;
>
> case B_SLICE:
> b = poc - l0poc;
> p1 = b + l1poc - poc;
> - frames[p0] = &slice->m_refPicList[0][0]->m_lowres;
> + frames[p0] = &slice->m_refFrameList[0][0]->m_lowres;
> frames[b] = &curFrame->m_lowres;
> - frames[p1] = &slice->m_refPicList[1][0]->m_lowres;
> + frames[p1] = &slice->m_refFrameList[1][0]->m_lowres;
> break;
>
> default:
> diff -r 1f161d9c6e35 -r 02e84edaa143 source/encoder/weightPrediction.cpp
> --- a/source/encoder/weightPrediction.cpp Mon Aug 03 14:56:21 2015
> -0500
> +++ b/source/encoder/weightPrediction.cpp Mon Aug 03 17:55:58 2015
> -0500
> @@ -259,7 +259,7 @@
> for (int list = 0; list < cache.numPredDir; list++)
> {
> WeightParam *weights = wp[list][0];
> - Frame *refFrame = slice.m_refPicList[list][0];
> + Frame *refFrame = slice.m_refFrameList[list][0];
> Lowres& refLowres = refFrame->m_lowres;
> int diffPoc = abs(curPoc - refFrame->m_poc);
>
> _______________________________________________
> 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/20150805/233ab089/attachment-0001.html>
More information about the x265-devel
mailing list