[x265] [PATCH] SAO: removed unnecessary copy of sao type and offsets in offset rdo calculation
ashok at multicorewareinc.com
ashok at multicorewareinc.com
Mon Jan 18 16:51:10 CET 2016
# HG changeset patch
# User Ashok Kumar Mishra<ashok at multicorewareinc.com>
# Date 1453132257 -19800
# Mon Jan 18 21:20:57 2016 +0530
# Node ID add820b8cde0fc82a643fd32717def5a3e2f3888
# Parent e7768afed62fad78a193ea0d54137c4f5d8d2db3
SAO: removed unnecessary copy of sao type and offsets in offset rdo calculation
diff -r e7768afed62f -r add820b8cde0 source/encoder/entropy.cpp
--- a/source/encoder/entropy.cpp Mon Jan 18 21:20:27 2016 +0530
+++ b/source/encoder/entropy.cpp Mon Jan 18 21:20:57 2016 +0530
@@ -1026,6 +1026,44 @@
}
}
+void Entropy::codeSaoOffsetEO(int *offset, int typeIdx, int plane)
+{
+ if (plane != 2)
+ {
+ encodeBin(1, m_contextState[OFF_SAO_TYPE_IDX_CTX]);
+ encodeBinEP(1);
+ }
+
+ enum { OFFSET_THRESH = 1 << X265_MIN(X265_DEPTH - 5, 5) };
+
+ codeSaoMaxUvlc(offset[0], OFFSET_THRESH - 1);
+ codeSaoMaxUvlc(offset[1], OFFSET_THRESH - 1);
+ codeSaoMaxUvlc(-offset[2], OFFSET_THRESH - 1);
+ codeSaoMaxUvlc(-offset[3], OFFSET_THRESH - 1);
+ if (plane != 2)
+ encodeBinsEP((uint32_t)(typeIdx), 2);
+}
+
+void Entropy::codeSaoOffsetBO(int *offset, int bandPos, int plane)
+{
+ if (plane != 2)
+ {
+ encodeBin(1, m_contextState[OFF_SAO_TYPE_IDX_CTX]);
+ encodeBinEP(0);
+ }
+
+ enum { OFFSET_THRESH = 1 << X265_MIN(X265_DEPTH - 5, 5) };
+
+ for (int i = 0; i < SAO_BO_LEN; i++)
+ codeSaoMaxUvlc(abs(offset[i]), OFFSET_THRESH - 1);
+
+ for (int i = 0; i < SAO_BO_LEN; i++)
+ if (offset[i] != 0)
+ encodeBinEP(offset[i] < 0);
+
+ encodeBinsEP(bandPos, 5);
+}
+
/** initialize context model with respect to QP and initialization value */
uint8_t sbacInit(int qp, int initValue)
{
diff -r e7768afed62f -r add820b8cde0 source/encoder/entropy.h
--- a/source/encoder/entropy.h Mon Jan 18 21:20:27 2016 +0530
+++ b/source/encoder/entropy.h Mon Jan 18 21:20:57 2016 +0530
@@ -169,6 +169,7 @@
void codeCoeffNxN(const CUData& cu, const coeff_t* coef, uint32_t absPartIdx, uint32_t log2TrSize, TextType ttype);
inline void codeSaoMerge(uint32_t code) { encodeBin(code, m_contextState[OFF_SAO_MERGE_FLAG_CTX]); }
+ inline void codeSaoType(uint32_t code) { encodeBin(code, m_contextState[OFF_SAO_TYPE_IDX_CTX]); }
inline void codeMVPIdx(uint32_t symbol) { encodeBin(symbol, m_contextState[OFF_MVP_IDX_CTX]); }
inline void codeMergeFlag(const CUData& cu, uint32_t absPartIdx) { encodeBin(cu.m_mergeFlag[absPartIdx], m_contextState[OFF_MERGE_FLAG_EXT_CTX]); }
inline void codeSkipFlag(const CUData& cu, uint32_t absPartIdx) { encodeBin(cu.isSkipped(absPartIdx), m_contextState[OFF_SKIP_FLAG_CTX + cu.getCtxSkipFlag(absPartIdx)]); }
@@ -182,6 +183,8 @@
inline void codeTransformSkipFlags(uint32_t transformSkip, TextType ttype) { encodeBin(transformSkip, m_contextState[OFF_TRANSFORMSKIP_FLAG_CTX + (ttype ? NUM_TRANSFORMSKIP_FLAG_CTX : 0)]); }
void codeDeltaQP(const CUData& cu, uint32_t absPartIdx);
void codeSaoOffset(const SaoCtuParam& ctuParam, int plane);
+ void codeSaoOffsetEO(int *offset, int typeIdx, int plane);
+ void codeSaoOffsetBO(int *offset, int bandPos, int plane);
/* RDO functions */
void estBit(EstBitsSbac& estBitsSbac, uint32_t log2TrSize, bool bIsLuma) const;
diff -r e7768afed62f -r add820b8cde0 source/encoder/sao.cpp
--- a/source/encoder/sao.cpp Mon Jan 18 21:20:27 2016 +0530
+++ b/source/encoder/sao.cpp Mon Jan 18 21:20:57 2016 +0530
@@ -1598,17 +1598,17 @@
void SAO::saoLumaComponentParamDist(SAOParam* saoParam, int addr, double* mergeDist)
{
int64_t bestDist = 0;
+ int bestTypeIdx = -1;
SaoCtuParam* lclCtuParam = &saoParam->ctuParam[0][addr];
- double bestRDCostTableBo = MAX_DOUBLE;
- int bestClassTableBo = 0;
int currentDistortionTableBo[MAX_NUM_SAO_CLASS];
double currentRdCostTableBo[MAX_NUM_SAO_CLASS];
m_entropyCoder.load(m_rdContexts.temp);
m_entropyCoder.resetBits();
- m_entropyCoder.codeSaoOffset(*lclCtuParam, 0);
+ m_entropyCoder.codeSaoType(0);
+
double dCostPartBest = m_entropyCoder.getNumberOfWrittenBits() * m_lumaLambda;
//EO distortion calculation
@@ -1633,16 +1633,9 @@
estDist += estSaoDist(count, (int)offsetOut << SAO_BIT_INC, offsetOrg);
}
- SaoCtuParam ctuParamRdo;
- ctuParamRdo.mergeMode = SAO_MERGE_NONE;
- ctuParamRdo.typeIdx = typeIdx;
- ctuParamRdo.bandPos = 0;
- for (int classIdx = 0; classIdx < SAO_NUM_OFFSET; classIdx++)
- ctuParamRdo.offset[classIdx] = (int)m_offset[0][typeIdx][classIdx + 1];
-
m_entropyCoder.load(m_rdContexts.temp);
m_entropyCoder.resetBits();
- m_entropyCoder.codeSaoOffset(ctuParamRdo, 0);
+ m_entropyCoder.codeSaoOffsetEO(m_offset[0][typeIdx] + 1, typeIdx, 0);
uint32_t estRate = m_entropyCoder.getNumberOfWrittenBits();
double cost = (double)estDist + m_lumaLambda * (double)estRate;
@@ -1650,12 +1643,21 @@
if (cost < dCostPartBest)
{
dCostPartBest = cost;
- copySaoUnit(lclCtuParam, &ctuParamRdo);
bestDist = estDist;
+ bestTypeIdx = typeIdx;
}
}
- //BO distortion calculation
+ if (bestTypeIdx != -1)
+ {
+ lclCtuParam->mergeMode = SAO_MERGE_NONE;
+ lclCtuParam->typeIdx = bestTypeIdx;
+ lclCtuParam->bandPos = 0;
+ for (int classIdx = 0; classIdx < SAO_NUM_OFFSET; classIdx++)
+ lclCtuParam->offset[classIdx] = (int)m_offset[0][bestTypeIdx][classIdx + 1];
+ }
+
+ //BO RDO
int64_t estDist = 0;
for (int classIdx = 1; classIdx < SAO_NUM_BO_CLASSES + 1; classIdx++)
{
@@ -1678,6 +1680,9 @@
}
// Estimate Best Position
+ double bestRDCostTableBo = MAX_DOUBLE;
+ int bestClassTableBo = 0;
+
for (int i = 0; i < SAO_NUM_BO_CLASSES - SAO_BO_LEN + 1; i++)
{
double currentRDCost = 0.0;
@@ -1695,16 +1700,9 @@
for (int classIdx = bestClassTableBo; classIdx < bestClassTableBo + SAO_BO_LEN; classIdx++)
estDist += currentDistortionTableBo[classIdx];
- SaoCtuParam ctuParamRdo;
- ctuParamRdo.mergeMode = SAO_MERGE_NONE;
- ctuParamRdo.typeIdx = SAO_BO;
- ctuParamRdo.bandPos = bestClassTableBo;
- for (int classIdx = 0; classIdx < SAO_NUM_OFFSET; classIdx++)
- ctuParamRdo.offset[classIdx] = (int)m_offset[0][SAO_BO][classIdx + ctuParamRdo.bandPos + 1];
-
m_entropyCoder.load(m_rdContexts.temp);
m_entropyCoder.resetBits();
- m_entropyCoder.codeSaoOffset(ctuParamRdo, 0);
+ m_entropyCoder.codeSaoOffsetBO(m_offset[0][SAO_BO] + (bestClassTableBo + 1), bestClassTableBo, 0);
uint32_t estRate = m_entropyCoder.getNumberOfWrittenBits();
double cost = (double)estDist + m_lumaLambda * (double)estRate;
@@ -1712,8 +1710,13 @@
if (cost < dCostPartBest)
{
dCostPartBest = cost;
- copySaoUnit(lclCtuParam, &ctuParamRdo);
bestDist = estDist;
+
+ lclCtuParam->mergeMode = SAO_MERGE_NONE;
+ lclCtuParam->typeIdx = SAO_BO;
+ lclCtuParam->bandPos = bestClassTableBo;
+ for (int classIdx = 0; classIdx < SAO_NUM_OFFSET; classIdx++)
+ lclCtuParam->offset[classIdx] = (int)m_offset[0][SAO_BO][classIdx + bestClassTableBo + 1];
}
mergeDist[0] = ((double)bestDist / m_lumaLambda);
@@ -1725,21 +1728,21 @@
void SAO::saoChromaComponentParamDist(SAOParam* saoParam, int addr, double* mergeDist)
{
int64_t bestDist = 0;
+ int bestTypeIdx = -1;
SaoCtuParam* lclCtuParam[2] = { &saoParam->ctuParam[1][addr], &saoParam->ctuParam[2][addr] };
double currentRdCostTableBo[MAX_NUM_SAO_CLASS];
+ int currentDistortionTableBo[MAX_NUM_SAO_CLASS];
int bestClassTableBo[2] = { 0, 0 };
- int currentDistortionTableBo[MAX_NUM_SAO_CLASS];
m_entropyCoder.load(m_rdContexts.temp);
m_entropyCoder.resetBits();
- m_entropyCoder.codeSaoOffset(*lclCtuParam[0], 1);
- m_entropyCoder.codeSaoOffset(*lclCtuParam[1], 2);
+ m_entropyCoder.codeSaoType(0);
double costPartBest = m_entropyCoder.getNumberOfWrittenBits() * m_chromaLambda;
- //EO distortion calculation
+ //EO RDO
for (int typeIdx = 0; typeIdx < MAX_NUM_SAO_TYPE - 1; typeIdx++)
{
int64_t estDist[2] = {0, 0};
@@ -1760,24 +1763,17 @@
offsetOrg = 0;
offsetOut = 0;
}
- if (typeIdx != SAO_BO)
- estDist[compIdx - 1] += estSaoDist(count, (int)offsetOut << SAO_BIT_INC, offsetOrg);
+
+ estDist[compIdx - 1] += estSaoDist(count, (int)offsetOut << SAO_BIT_INC, offsetOrg);
}
}
m_entropyCoder.load(m_rdContexts.temp);
m_entropyCoder.resetBits();
- SaoCtuParam ctuParamRdo[2];
for (int compIdx = 0; compIdx < 2; compIdx++)
{
- ctuParamRdo[compIdx].mergeMode = SAO_MERGE_NONE;
- ctuParamRdo[compIdx].typeIdx = typeIdx;
- ctuParamRdo[compIdx].bandPos = 0;
- for (int classIdx = 0; classIdx < SAO_NUM_OFFSET; classIdx++)
- ctuParamRdo[compIdx].offset[classIdx] = (int)m_offset[compIdx + 1][typeIdx][classIdx + 1];
-
- m_entropyCoder.codeSaoOffset(ctuParamRdo[compIdx], compIdx + 1);
+ m_entropyCoder.codeSaoOffsetEO(m_offset[compIdx + 1][typeIdx] + 1, typeIdx, compIdx + 1);
}
uint32_t estRate = m_entropyCoder.getNumberOfWrittenBits();
@@ -1786,13 +1782,24 @@
if (cost < costPartBest)
{
costPartBest = cost;
- copySaoUnit(lclCtuParam[0], &ctuParamRdo[0]);
- copySaoUnit(lclCtuParam[1], &ctuParamRdo[1]);
bestDist = (estDist[0] + estDist[1]);
+ bestTypeIdx = typeIdx;
}
}
- // BO distortion calculation
+ if (bestTypeIdx != -1)
+ {
+ for (int compIdx = 0; compIdx < 2; compIdx++)
+ {
+ lclCtuParam[compIdx]->mergeMode = SAO_MERGE_NONE;
+ lclCtuParam[compIdx]->typeIdx = bestTypeIdx;
+ lclCtuParam[compIdx]->bandPos = 0;
+ for (int classIdx = 0; classIdx < SAO_NUM_OFFSET; classIdx++)
+ lclCtuParam[compIdx]->offset[classIdx] = (int)m_offset[compIdx + 1][bestTypeIdx][classIdx + 1];
+ }
+ }
+
+ // BO RDO
int64_t estDist[2];
// Estimate Best Position
@@ -1841,16 +1848,9 @@
m_entropyCoder.load(m_rdContexts.temp);
m_entropyCoder.resetBits();
- SaoCtuParam ctuParamRdo[2];
for (int compIdx = 0; compIdx < 2; compIdx++)
{
- ctuParamRdo[compIdx].mergeMode = SAO_MERGE_NONE;
- ctuParamRdo[compIdx].typeIdx = SAO_BO;
- ctuParamRdo[compIdx].bandPos = bestClassTableBo[compIdx];
- for (int classIdx = 0; classIdx < SAO_NUM_OFFSET; classIdx++)
- ctuParamRdo[compIdx].offset[classIdx] = (int)m_offset[compIdx + 1][SAO_BO][classIdx + ctuParamRdo[compIdx].bandPos + 1];
-
- m_entropyCoder.codeSaoOffset(ctuParamRdo[compIdx], compIdx + 1);
+ m_entropyCoder.codeSaoOffsetBO(m_offset[compIdx + 1][SAO_BO] + (bestClassTableBo[compIdx] + 1), bestClassTableBo[compIdx], compIdx + 1);
}
uint32_t estRate = m_entropyCoder.getNumberOfWrittenBits();
@@ -1859,9 +1859,16 @@
if (cost < costPartBest)
{
costPartBest = cost;
- copySaoUnit(lclCtuParam[0], &ctuParamRdo[0]);
- copySaoUnit(lclCtuParam[1], &ctuParamRdo[1]);
bestDist = (estDist[0] + estDist[1]);
+
+ for (int compIdx = 0; compIdx < 2; compIdx++)
+ {
+ lclCtuParam[compIdx]->mergeMode = SAO_MERGE_NONE;
+ lclCtuParam[compIdx]->typeIdx = SAO_BO;
+ lclCtuParam[compIdx]->bandPos = bestClassTableBo[compIdx];
+ for (int classIdx = 0; classIdx < SAO_NUM_OFFSET; classIdx++)
+ lclCtuParam[compIdx]->offset[classIdx] = (int)m_offset[compIdx + 1][SAO_BO][classIdx + bestClassTableBo[compIdx] + 1];
+ }
}
mergeDist[0] += ((double)bestDist / m_chromaLambda);
diff -r e7768afed62f -r add820b8cde0 source/encoder/sao.h
--- a/source/encoder/sao.h Mon Jan 18 21:20:27 2016 +0530
+++ b/source/encoder/sao.h Mon Jan 18 21:20:57 2016 +0530
@@ -150,7 +150,7 @@
void rdoSaoUnitRow(SAOParam* saoParam, int idxY);
void rdoSaoUnitCu(SAOParam* saoParam, int rowBaseAddr, int idxX, int addr);
- void saoStatisticInitOffset(int plane);
+ void saoStatsInitialOffset(int plane);
friend class FrameFilter;
};
More information about the x265-devel
mailing list