[x265] [PATCH] analysis: avoid redundant rect/amp mode analysis based on split block rdCost and mvCost for rd-0/4
Ashok Kumar Mishra
ashok at multicorewareinc.com
Thu Oct 15 17:06:07 CEST 2015
Below are the performance testing on Haswell with and without limiting
rect/amp analysis mode in slow preset.
*Before*
D:\ashok>x265_b.exe --input
\\HEVC-TEST-2\testsequences\ducks_take_off_1080p50.y4m --preset slow
--hash=1 --no-info --psnr --ssim -o test_b.hevc --bitrate 6000
encoded 500 frames in 157.90s (3.17 fps), 6060.98 kb/s, Avg QP:40.41,
Global PSNR: 29.485, SSIM Mean Y: 0.7780801 ( 6.538 dB)
*After*
D:\ashok>x265.exe --input
\\HEVC-TEST-2\testsequences\ducks_take_off_1080p50.y4m --preset slow
--hash=1 --no-info --psnr --ssim -o test.hevc --bitrate 6000
--limit-rect-amp 1
encoded 500 frames in 148.53s (3.37 fps), 6062.77 kb/s, Avg QP:40.43,
Global PSNR: 29.487, SSIM Mean Y: 0.7780540 ( 6.538 dB)
D:\ashok>x265.exe --input
\\HEVC-TEST-2\testsequences\ducks_take_off_1080p50.y4m --preset slow
--hash=1 --no-info --psnr --ssim -o test.hevc --bitrate 6000 --limit-refs 1
encoded 500 frames in 136.84s (3.65 fps), 6061.15 kb/s, Avg QP:40.42,
Global PSNR: 29.480, SSIM Mean Y: 0.7778692 ( 6.534 dB)
D:\ashok>x265.exe --input
\\HEVC-TEST-2\testsequences\ducks_take_off_1080p50.y4m --preset slow
--hash=1 --no-info --psnr --ssim -o test.hevc --bitrate 6000 --limit-refs 1
--limit-rect-amp 1
encoded 500 frames in 133.06s (3.76 fps), 6062.52 kb/s, Avg QP:40.43,
Global PSNR: 29.481, SSIM Mean Y: 0.7779036 ( 6.535 dB)
On Thu, Oct 15, 2015 at 8:31 PM, <ashok at multicorewareinc.com> wrote:
> # HG changeset patch
> # User Ashok Kumar Mishra<ashok at multicorewareinc.com>
> # Date 1444897694 -19800
> # Thu Oct 15 13:58:14 2015 +0530
> # Node ID 65d7c1f5baf5fa619d773fcc2e1361d46f6df7f1
> # Parent f3963e7e75b8dcb599250c082357e08fd32191a5
> analysis: avoid redundant rect/amp mode analysis based on split block
> rdCost and mvCost for rd-0/4
>
> diff -r f3963e7e75b8 -r 65d7c1f5baf5 source/encoder/analysis.cpp
> --- a/source/encoder/analysis.cpp Wed Oct 14 17:44:33 2015 +0530
> +++ b/source/encoder/analysis.cpp Thu Oct 15 13:58:14 2015 +0530
> @@ -809,7 +809,7 @@
> return refMask;
> }
>
> -uint32_t Analysis::compressInterCU_rd0_4(const CUData& parentCTU, const
> CUGeom& cuGeom, int32_t qp)
> +SplitData Analysis::compressInterCU_rd0_4(const CUData& parentCTU, const
> CUGeom& cuGeom, int32_t qp)
> {
> uint32_t depth = cuGeom.depth;
> uint32_t cuAddr = parentCTU.m_cuAddr;
> @@ -823,7 +823,13 @@
> uint32_t minDepth = topSkipMinDepth(parentCTU, cuGeom);
> bool earlyskip = false;
> bool splitIntra = true;
> - uint32_t splitRefs[4] = { 0, 0, 0, 0 };
> +
> + SplitData splitData[4];
> + splitData[0].initSplitCUData();
> + splitData[1].initSplitCUData();
> + splitData[2].initSplitCUData();
> + splitData[3].initSplitCUData();
> +
> /* Step 1. Evaluate Merge/Skip candidates for likely early-outs */
> if (mightNotSplit && depth >= minDepth)
> {
> @@ -869,7 +875,7 @@
> if (m_slice->m_pps->bUseDQP && nextDepth <=
> m_slice->m_pps->maxCuDQPDepth)
> nextQP = setLambdaFromQP(parentCTU,
> calculateQpforCuSize(parentCTU, childGeom));
>
> - splitRefs[subPartIdx] = compressInterCU_rd0_4(parentCTU,
> childGeom, nextQP);
> + splitData[subPartIdx] = compressInterCU_rd0_4(parentCTU,
> childGeom, nextQP);
>
> // Save best CU and pred data for this sub CU
> splitIntra |= nd.bestMode->cu.isIntra(0);
> @@ -899,7 +905,7 @@
> /* Split CUs
> * 0 1
> * 2 3 */
> - uint32_t allSplitRefs = splitRefs[0] | splitRefs[1] | splitRefs[2] |
> splitRefs[3];
> + uint32_t allSplitRefs = splitData[0].splitRefs |
> splitData[1].splitRefs | splitData[2].splitRefs | splitData[3].splitRefs;
> /* Step 3. Evaluate ME (2Nx2N, rect, amp) and intra modes at current
> depth */
> if (mightNotSplit && depth >= minDepth)
> {
> @@ -917,7 +923,7 @@
> {
> CUData& cu = md.pred[PRED_2Nx2N].cu;
> uint32_t refMask = cu.getBestRefIdx(0);
> - allSplitRefs = splitRefs[0] = splitRefs[1] = splitRefs[2]
> = splitRefs[3] = refMask;
> + allSplitRefs = splitData[0].splitRefs =
> splitData[1].splitRefs = splitData[2].splitRefs = splitData[3].splitRefs =
> refMask;
> }
>
> if (m_slice->m_sliceType == B_SLICE)
> @@ -929,23 +935,82 @@
> Mode *bestInter = &md.pred[PRED_2Nx2N];
> if (m_param->bEnableRectInter)
> {
> - refMasks[0] = splitRefs[0] | splitRefs[2]; /* left */
> - refMasks[1] = splitRefs[1] | splitRefs[3]; /* right */
> - md.pred[PRED_Nx2N].cu.initSubCU(parentCTU, cuGeom, qp);
> - checkInter_rd0_4(md.pred[PRED_Nx2N], cuGeom, SIZE_Nx2N,
> refMasks);
> - if (md.pred[PRED_Nx2N].sa8dCost < bestInter->sa8dCost)
> - bestInter = &md.pred[PRED_Nx2N];
> -
> - refMasks[0] = splitRefs[0] | splitRefs[1]; /* top */
> - refMasks[1] = splitRefs[2] | splitRefs[3]; /* bot */
> - md.pred[PRED_2NxN].cu.initSubCU(parentCTU, cuGeom, qp);
> - checkInter_rd0_4(md.pred[PRED_2NxN], cuGeom, SIZE_2NxN,
> refMasks);
> - if (md.pred[PRED_2NxN].sa8dCost < bestInter->sa8dCost)
> - bestInter = &md.pred[PRED_2NxN];
> + uint64_t splitCost = splitData[0].rdCost +
> splitData[1].rdCost + splitData[2].rdCost + splitData[3].rdCost;
> + ModeDepth& md = m_modeDepth[depth];
> + uint32_t threshold_2NxN, threshold_Nx2N;
> +
> + if (m_slice->m_sliceType == P_SLICE)
> + {
> + threshold_2NxN = splitData[0].mvCost[0] +
> splitData[1].mvCost[0];
> + threshold_Nx2N = splitData[0].mvCost[0] +
> splitData[2].mvCost[0];
> + }
> + else
> + {
> + threshold_2NxN = (splitData[0].mvCost[0] +
> splitData[1].mvCost[0]
> + + splitData[0].mvCost[1] +
> splitData[1].mvCost[1] + 1) >> 1;
> + threshold_Nx2N = (splitData[0].mvCost[0] +
> splitData[2].mvCost[0]
> + + splitData[0].mvCost[1] +
> splitData[2].mvCost[1] + 1) >> 1;
> + }
> +
> + int try_2NxN_first = threshold_2NxN < threshold_Nx2N;
> + if (try_2NxN_first && splitCost < md.bestMode->rdCost +
> threshold_2NxN)
> + {
> + refMasks[0] = splitData[0].splitRefs |
> splitData[1].splitRefs; /* top */
> + refMasks[1] = splitData[2].splitRefs |
> splitData[3].splitRefs; /* bot */
> + md.pred[PRED_2NxN].cu.initSubCU(parentCTU, cuGeom,
> qp);
> + checkInter_rd0_4(md.pred[PRED_2NxN], cuGeom,
> SIZE_2NxN, refMasks);
> + if (md.pred[PRED_2NxN].sa8dCost < bestInter->sa8dCost)
> + bestInter = &md.pred[PRED_2NxN];
> + }
> +
> + if (splitCost < md.bestMode->rdCost + threshold_Nx2N)
> + {
> + refMasks[0] = splitData[0].splitRefs |
> splitData[2].splitRefs; /* left */
> + refMasks[1] = splitData[1].splitRefs |
> splitData[3].splitRefs; /* right */
> + md.pred[PRED_Nx2N].cu.initSubCU(parentCTU, cuGeom,
> qp);
> + checkInter_rd0_4(md.pred[PRED_Nx2N], cuGeom,
> SIZE_Nx2N, refMasks);
> + if (md.pred[PRED_Nx2N].sa8dCost < bestInter->sa8dCost)
> + bestInter = &md.pred[PRED_Nx2N];
> + }
> +
> + if (!try_2NxN_first && splitCost < md.bestMode->rdCost +
> threshold_2NxN)
> + {
> + refMasks[0] = splitData[0].splitRefs |
> splitData[1].splitRefs; /* top */
> + refMasks[1] = splitData[2].splitRefs |
> splitData[3].splitRefs; /* bot */
> + md.pred[PRED_2NxN].cu.initSubCU(parentCTU, cuGeom,
> qp);
> + checkInter_rd0_4(md.pred[PRED_2NxN], cuGeom,
> SIZE_2NxN, refMasks);
> + if (md.pred[PRED_2NxN].sa8dCost < bestInter->sa8dCost)
> + bestInter = &md.pred[PRED_2NxN];
> + }
> }
>
> if (m_slice->m_sps->maxAMPDepth > depth)
> {
> + uint64_t splitCost = splitData[0].rdCost +
> splitData[1].rdCost + splitData[2].rdCost + splitData[3].rdCost;
> + ModeDepth& md = m_modeDepth[depth];
> + uint32_t threshold_2NxnU, threshold_2NxnD,
> threshold_nLx2N, threshold_nRx2N;
> +
> + if (m_slice->m_sliceType == P_SLICE)
> + {
> + threshold_2NxnU = splitData[0].mvCost[0] +
> splitData[1].mvCost[0];
> + threshold_2NxnD = splitData[2].mvCost[0] +
> splitData[3].mvCost[0];
> +
> + threshold_nLx2N = splitData[0].mvCost[0] +
> splitData[2].mvCost[0];
> + threshold_nRx2N = splitData[1].mvCost[0] +
> splitData[3].mvCost[0];
> + }
> + else
> + {
> + threshold_2NxnU = (splitData[0].mvCost[0] +
> splitData[1].mvCost[0]
> + + splitData[0].mvCost[1] +
> splitData[1].mvCost[1] + 1) >> 1;
> + threshold_2NxnD = (splitData[2].mvCost[0] +
> splitData[3].mvCost[0]
> + + splitData[2].mvCost[1] +
> splitData[3].mvCost[1] + 1) >> 1;
> +
> + threshold_nLx2N = (splitData[0].mvCost[0] +
> splitData[2].mvCost[0]
> + + splitData[0].mvCost[1] +
> splitData[2].mvCost[1] + 1) >> 1;
> + threshold_nRx2N = (splitData[1].mvCost[0] +
> splitData[3].mvCost[0]
> + + splitData[1].mvCost[1] +
> splitData[3].mvCost[1] + 1) >> 1;
> + }
> +
> bool bHor = false, bVer = false;
> if (bestInter->cu.m_partSize[0] == SIZE_2NxN)
> bHor = true;
> @@ -960,35 +1025,69 @@
>
> if (bHor)
> {
> - refMasks[0] = splitRefs[0] | splitRefs[1]; /* 25% top
> */
> - refMasks[1] = allSplitRefs; /* 75% bot
> */
> - md.pred[PRED_2NxnU].cu.initSubCU(parentCTU, cuGeom,
> qp);
> - checkInter_rd0_4(md.pred[PRED_2NxnU], cuGeom,
> SIZE_2NxnU, refMasks);
> - if (md.pred[PRED_2NxnU].sa8dCost <
> bestInter->sa8dCost)
> - bestInter = &md.pred[PRED_2NxnU];
> -
> - refMasks[0] = allSplitRefs; /* 75% top
> */
> - refMasks[1] = splitRefs[2] | splitRefs[3]; /* 25% bot
> */
> - md.pred[PRED_2NxnD].cu.initSubCU(parentCTU, cuGeom,
> qp);
> - checkInter_rd0_4(md.pred[PRED_2NxnD], cuGeom,
> SIZE_2NxnD, refMasks);
> - if (md.pred[PRED_2NxnD].sa8dCost <
> bestInter->sa8dCost)
> - bestInter = &md.pred[PRED_2NxnD];
> + int try_2NxnD_first = threshold_2NxnD <
> threshold_2NxnU;
> + if (try_2NxnD_first && splitCost <
> md.bestMode->rdCost + threshold_2NxnD)
> + {
> + refMasks[0] = allSplitRefs;
> /* 75% top */
> + refMasks[1] = splitData[2].splitRefs |
> splitData[3].splitRefs; /* 25% bot */
> + md.pred[PRED_2NxnD].cu.initSubCU(parentCTU,
> cuGeom, qp);
> + checkInter_rd0_4(md.pred[PRED_2NxnD], cuGeom,
> SIZE_2NxnD, refMasks);
> + if (md.pred[PRED_2NxnD].sa8dCost <
> bestInter->sa8dCost)
> + bestInter = &md.pred[PRED_2NxnD];
> + }
> +
> + if (splitCost < md.bestMode->rdCost + threshold_2NxnU)
> + {
> + refMasks[0] = splitData[0].splitRefs |
> splitData[1].splitRefs; /* 25% top */
> + refMasks[1] = allSplitRefs;
> /* 75% bot */
> + md.pred[PRED_2NxnU].cu.initSubCU(parentCTU,
> cuGeom, qp);
> + checkInter_rd0_4(md.pred[PRED_2NxnU], cuGeom,
> SIZE_2NxnU, refMasks);
> + if (md.pred[PRED_2NxnU].sa8dCost <
> bestInter->sa8dCost)
> + bestInter = &md.pred[PRED_2NxnU];
> + }
> +
> + if (!try_2NxnD_first && splitCost <
> md.bestMode->rdCost + threshold_2NxnD)
> + {
> + refMasks[0] = allSplitRefs;
> /* 75% top */
> + refMasks[1] = splitData[2].splitRefs |
> splitData[3].splitRefs; /* 25% bot */
> + md.pred[PRED_2NxnD].cu.initSubCU(parentCTU,
> cuGeom, qp);
> + checkInter_rd0_4(md.pred[PRED_2NxnD], cuGeom,
> SIZE_2NxnD, refMasks);
> + if (md.pred[PRED_2NxnD].sa8dCost <
> bestInter->sa8dCost)
> + bestInter = &md.pred[PRED_2NxnD];
> + }
> }
> if (bVer)
> {
> - refMasks[0] = splitRefs[0] | splitRefs[2]; /* 25%
> left */
> - refMasks[1] = allSplitRefs; /* 75%
> right */
> - md.pred[PRED_nLx2N].cu.initSubCU(parentCTU, cuGeom,
> qp);
> - checkInter_rd0_4(md.pred[PRED_nLx2N], cuGeom,
> SIZE_nLx2N, refMasks);
> - if (md.pred[PRED_nLx2N].sa8dCost <
> bestInter->sa8dCost)
> - bestInter = &md.pred[PRED_nLx2N];
> -
> - refMasks[0] = allSplitRefs; /* 75%
> left */
> - refMasks[1] = splitRefs[1] | splitRefs[3]; /* 25%
> right */
> - md.pred[PRED_nRx2N].cu.initSubCU(parentCTU, cuGeom,
> qp);
> - checkInter_rd0_4(md.pred[PRED_nRx2N], cuGeom,
> SIZE_nRx2N, refMasks);
> - if (md.pred[PRED_nRx2N].sa8dCost <
> bestInter->sa8dCost)
> - bestInter = &md.pred[PRED_nRx2N];
> + int try_nRx2N_first = threshold_nRx2N <
> threshold_nLx2N;
> + if (try_nRx2N_first && splitCost <
> md.bestMode->rdCost + threshold_nRx2N)
> + {
> + refMasks[0] = allSplitRefs;
> /* 75% left */
> + refMasks[1] = splitData[1].splitRefs |
> splitData[3].splitRefs; /* 25% right */
> + md.pred[PRED_nRx2N].cu.initSubCU(parentCTU,
> cuGeom, qp);
> + checkInter_rd0_4(md.pred[PRED_nRx2N], cuGeom,
> SIZE_nRx2N, refMasks);
> + if (md.pred[PRED_nRx2N].sa8dCost <
> bestInter->sa8dCost)
> + bestInter = &md.pred[PRED_nRx2N];
> + }
> +
> + if (splitCost < md.bestMode->rdCost + threshold_nLx2N)
> + {
> + refMasks[0] = splitData[0].splitRefs |
> splitData[2].splitRefs; /* 25% left */
> + refMasks[1] = allSplitRefs;
> /* 75% right */
> + md.pred[PRED_nLx2N].cu.initSubCU(parentCTU,
> cuGeom, qp);
> + checkInter_rd0_4(md.pred[PRED_nLx2N], cuGeom,
> SIZE_nLx2N, refMasks);
> + if (md.pred[PRED_nLx2N].sa8dCost <
> bestInter->sa8dCost)
> + bestInter = &md.pred[PRED_nLx2N];
> + }
> +
> + if (!try_nRx2N_first && splitCost <
> md.bestMode->rdCost + threshold_nRx2N)
> + {
> + refMasks[0] = allSplitRefs;
> /* 75% left */
> + refMasks[1] = splitData[1].splitRefs |
> splitData[3].splitRefs; /* 25% right */
> + md.pred[PRED_nRx2N].cu.initSubCU(parentCTU,
> cuGeom, qp);
> + checkInter_rd0_4(md.pred[PRED_nRx2N], cuGeom,
> SIZE_nRx2N, refMasks);
> + if (md.pred[PRED_nRx2N].sa8dCost <
> bestInter->sa8dCost)
> + bestInter = &md.pred[PRED_nRx2N];
> + }
> }
> }
> bool bTryIntra = m_slice->m_sliceType != B_SLICE ||
> m_param->bIntraInBFrames;
> @@ -1139,19 +1238,32 @@
> }
>
> /* determine which motion references the parent CU should search */
> - uint32_t refMask;
> - if (!(m_param->limitReferences & X265_REF_LIMIT_DEPTH))
> - refMask = 0;
> - else if (md.bestMode == &md.pred[PRED_SPLIT])
> - refMask = allSplitRefs;
> - else
> - {
> - /* use best merge/inter mode, in case of intra use 2Nx2N inter
> references */
> - CUData& cu = md.bestMode->cu.isIntra(0) ? md.pred[PRED_2Nx2N].cu
> : md.bestMode->cu;
> - uint32_t numPU = cu.getNumPartInter(0);
> - refMask = 0;
> - for (uint32_t puIdx = 0, subPartIdx = 0; puIdx < numPU; puIdx++,
> subPartIdx += cu.getPUOffset(puIdx, 0))
> - refMask |= cu.getBestRefIdx(subPartIdx);
> + SplitData splitCUData;
> + if (!(m_param->limitReferences & X265_REF_LIMIT_DEPTH))
> + splitCUData.splitRefs = 0;
> + else if (md.bestMode == &md.pred[PRED_SPLIT])
> + splitCUData.splitRefs = allSplitRefs;
> + else
> + {
> + /* use best merge/inter mode, in case of intra use 2Nx2N inter
> references */
> + CUData& cu = md.bestMode->cu.isIntra(0) ? md.pred[PRED_2Nx2N].cu
> : md.bestMode->cu;
> + uint32_t numPU = cu.getNumPartInter(0);
> + splitCUData.splitRefs = 0;
> + for (uint32_t puIdx = 0, subPartIdx = 0; puIdx < numPU; puIdx++,
> subPartIdx += cu.getPUOffset(puIdx, 0))
> + splitCUData.splitRefs |= cu.getBestRefIdx(subPartIdx);
> + }
> +
> + if (!m_param->limitRectAmp)
> + {
> + splitCUData.mvCost[0] = 0; // L0
> + splitCUData.mvCost[1] = 0; // L1
> + splitCUData.rdCost = 0;
> + }
> + else
> + {
> + splitCUData.mvCost[0] = md.pred[PRED_2Nx2N].bestME[0][0].mvCost;
> // L0
> + splitCUData.mvCost[1] = md.pred[PRED_2Nx2N].bestME[0][1].mvCost;
> // L1
> + splitCUData.rdCost = md.bestMode->rdCost;
> }
>
> if (mightNotSplit)
> @@ -1169,7 +1281,7 @@
> if (m_param->rdLevel)
> md.bestMode->reconYuv.copyToPicYuv(reconPic, cuAddr,
> cuGeom.absPartIdx);
>
> - return refMask;
> + return splitCUData;
> }
>
> SplitData Analysis::compressInterCU_rd5_6(const CUData& parentCTU, const
> CUGeom& cuGeom, uint32_t &zOrder, int32_t qp)
> diff -r f3963e7e75b8 -r 65d7c1f5baf5 source/encoder/analysis.h
> --- a/source/encoder/analysis.h Wed Oct 14 17:44:33 2015 +0530
> +++ b/source/encoder/analysis.h Thu Oct 15 13:58:14 2015 +0530
> @@ -131,7 +131,7 @@
>
> /* full analysis for a P or B slice CU */
> uint32_t compressInterCU_dist(const CUData& parentCTU, const CUGeom&
> cuGeom, int32_t qp);
> - uint32_t compressInterCU_rd0_4(const CUData& parentCTU, const CUGeom&
> cuGeom, int32_t qp);
> + SplitData compressInterCU_rd0_4(const CUData& parentCTU, const
> CUGeom& cuGeom, int32_t qp);
> SplitData compressInterCU_rd5_6(const CUData& parentCTU, const
> CUGeom& cuGeom, uint32_t &zOrder, int32_t qp);
>
> /* measure merge and skip */
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.videolan.org/pipermail/x265-devel/attachments/20151015/cd3ff5ca/attachment-0001.html>
More information about the x265-devel
mailing list