[x265] [PATCH 1 of 3] De-couple analysis reuse levels of --analysis-save and --analysis-load
Aruna Matheswaran
aruna at multicorewareinc.com
Tue Jan 7 09:54:58 CET 2020
# HG changeset patch
# User Aruna Matheswaran <aruna at multicorewareinc.com>
# Date 1576499107 -19800
# Mon Dec 16 17:55:07 2019 +0530
# Node ID 4d870dd7c314334b27413de922d08d9c01d5860b
# Parent 52135ffd9bcdd32b944311e9e66c473894c0a2bd
De-couple analysis reuse levels of --analysis-save and --analysis-load.
Currently, analysis reuse level of the load encode has to be the same as that of
the reuse level of the save encode. This commit de-couples the reuse level
constraints and let the load encode re-use a subset of the available analysis
info.
Two CLI and x265_param options are introduced to get the desired reuse level
during save and load.
The existing option --analysis-reuse-level is deprecated.
diff -r 52135ffd9bcd -r 4d870dd7c314 doc/reST/cli.rst
--- a/doc/reST/cli.rst Mon Dec 23 14:40:32 2019 +0530
+++ b/doc/reST/cli.rst Mon Dec 16 17:55:07 2019 +0530
@@ -918,12 +918,14 @@
Specify a filename for `multi-pass-opt-analysis` and `multi-pass-opt-distortion`.
If no filename is specified, x265_analysis.dat is used.
-.. option:: --analysis-reuse-level <1..10>
-
- Amount of information stored/reused in :option:`--analysis-reuse-mode` is distributed across levels.
+.. option:: --analysis-save-reuse-level <1..10>, --analysis-load-reuse-level <1..10>
+
+ :option:`--analysis-save-reuse-level` denotes the amount of information stored during :option:`--analysis-save` and
+ :option:`--analysis-load-reuse-level` denotes the amount of information reused during :option:`--analysis-load`.
Higher the value, higher the information stored/reused, faster the encode. Default 5.
- Note that --analysis-reuse-level must be paired with analysis-reuse-mode.
+ Note that :option:`--analysis-save-reuse-level` and :option:`--analysis-load-reuse-level` must be paired
+ with :option:`--analysis-save` and :option:`--analysis-load` respectively.
+--------------+------------------------------------------+
| Level | Description |
diff -r 52135ffd9bcd -r 4d870dd7c314 source/CMakeLists.txt
--- a/source/CMakeLists.txt Mon Dec 23 14:40:32 2019 +0530
+++ b/source/CMakeLists.txt Mon Dec 16 17:55:07 2019 +0530
@@ -29,7 +29,7 @@
option(STATIC_LINK_CRT "Statically link C runtime for release builds" OFF)
mark_as_advanced(FPROFILE_USE FPROFILE_GENERATE NATIVE_BUILD)
# X265_BUILD must be incremented each time the public API is changed
-set(X265_BUILD 184)
+set(X265_BUILD 185)
configure_file("${PROJECT_SOURCE_DIR}/x265.def.in"
"${PROJECT_BINARY_DIR}/x265.def")
configure_file("${PROJECT_SOURCE_DIR}/x265_config.h.in"
diff -r 52135ffd9bcd -r 4d870dd7c314 source/common/param.cpp
--- a/source/common/param.cpp Mon Dec 23 14:40:32 2019 +0530
+++ b/source/common/param.cpp Mon Dec 16 17:55:07 2019 +0530
@@ -319,7 +319,9 @@
param->bOptCUDeltaQP = 0;
param->bAQMotion = 0;
param->bHDROpt = 0;
- param->analysisReuseLevel = 5;
+ param->analysisReuseLevel = 0; /*DEPRECATED*/
+ param->analysisSaveReuseLevel = 0;
+ param->analysisLoadReuseLevel = 0;
param->toneMapFile = NULL;
param->bDhdr10opt = 0;
param->dolbyProfile = 0;
@@ -1219,7 +1221,14 @@
OPT("multi-pass-opt-distortion") p->analysisMultiPassDistortion = atobool(value);
OPT("aq-motion") p->bAQMotion = atobool(value);
OPT("dynamic-rd") p->dynamicRd = atof(value);
- OPT("analysis-reuse-level") p->analysisReuseLevel = atoi(value);
+ OPT("analysis-reuse-level")
+ {
+ p->analysisReuseLevel = atoi(value);
+ p->analysisSaveReuseLevel = atoi(value);
+ p->analysisLoadReuseLevel = atoi(value);
+ }
+ OPT("analysis-save-reuse-level") p->analysisSaveReuseLevel = atoi(value);
+ OPT("analysis-load-reuse-level") p->analysisLoadReuseLevel = atoi(value);
OPT("ssim-rd")
{
int bval = atobool(value);
@@ -1260,7 +1269,7 @@
}
else if (strcmp(strdup(value), "off") == 0)
{
- p->bAnalysisType = NO_INFO;
+ p->bAnalysisType = DEFAULT;
}
else
{
@@ -1696,8 +1705,10 @@
"Constant QP is incompatible with 2pass");
CHECK(param->rc.bStrictCbr && (param->rc.bitrate <= 0 || param->rc.vbvBufferSize <=0),
"Strict-cbr cannot be applied without specifying target bitrate or vbv bufsize");
- CHECK((param->analysisSave || param->analysisLoad) && (param->analysisReuseLevel < 1 || param->analysisReuseLevel > 10),
- "Invalid analysis refine level. Value must be between 1 and 10 (inclusive)");
+ CHECK(param->analysisSave && (param->analysisSaveReuseLevel < 1 || param->analysisSaveReuseLevel > 10),
+ "Invalid analysis save refine level. Value must be between 1 and 10 (inclusive)");
+ CHECK(param->analysisLoad && (param->analysisLoadReuseLevel < 1 || param->analysisLoadReuseLevel > 10),
+ "Invalid analysis load refine level. Value must be between 1 and 10 (inclusive)");
CHECK(param->analysisLoad && (param->mvRefine < 1 || param->mvRefine > 3),
"Invalid mv refinement level. Value must be between 1 and 3 (inclusive)");
CHECK(param->scaleFactor > 2, "Invalid scale-factor. Supports factor <= 2");
@@ -2160,6 +2171,8 @@
if (p->analysisLoad)
s += sprintf(s, " analysis-load");
s += sprintf(s, " analysis-reuse-level=%d", p->analysisReuseLevel);
+ s += sprintf(s, " analysis-save-reuse-level=%d", p->analysisSaveReuseLevel);
+ s += sprintf(s, " analysis-load-reuse-level=%d", p->analysisLoadReuseLevel);
s += sprintf(s, " scale-factor=%d", p->scaleFactor);
s += sprintf(s, " refine-intra=%d", p->intraRefine);
s += sprintf(s, " refine-inter=%d", p->interRefine);
@@ -2477,6 +2490,8 @@
dst->bEmitHRDSEI = src->bEmitHRDSEI;
dst->bHDROpt = src->bHDROpt;
dst->analysisReuseLevel = src->analysisReuseLevel;
+ dst->analysisSaveReuseLevel = src->analysisSaveReuseLevel;
+ dst->analysisLoadReuseLevel = src->analysisLoadReuseLevel;
dst->bLimitSAO = src->bLimitSAO;
if (src->toneMapFile) dst->toneMapFile = strdup(src->toneMapFile);
else dst->toneMapFile = NULL;
diff -r 52135ffd9bcd -r 4d870dd7c314 source/encoder/analysis.cpp
--- a/source/encoder/analysis.cpp Mon Dec 23 14:40:32 2019 +0530
+++ b/source/encoder/analysis.cpp Mon Dec 16 17:55:07 2019 +0530
@@ -202,14 +202,17 @@
m_reuseDepth = &m_reuseInterDataCTU->depth[ctu.m_cuAddr * ctu.m_numPartitions];
}
- if ((m_param->analysisSave || m_param->analysisLoad) && m_slice->m_sliceType != I_SLICE && m_param->analysisReuseLevel > 1 && m_param->analysisReuseLevel < 10)
+ int reuseLevel = X265_MAX(m_param->analysisSaveReuseLevel, m_param->analysisLoadReuseLevel);
+ if ((m_param->analysisSave || m_param->analysisLoad) && m_slice->m_sliceType != I_SLICE && reuseLevel > 1 && reuseLevel < 10)
{
int numPredDir = m_slice->isInterP() ? 1 : 2;
m_reuseInterDataCTU = m_frame->m_analysisData.interData;
- m_reuseRef = &m_reuseInterDataCTU->ref [ctu.m_cuAddr * X265_MAX_PRED_MODE_PER_CTU * numPredDir];
+ if (((m_param->analysisSaveReuseLevel > 1) && (m_param->analysisSaveReuseLevel < 7)) ||
+ ((m_param->analysisLoadReuseLevel > 1) && (m_param->analysisLoadReuseLevel < 7)))
+ m_reuseRef = &m_reuseInterDataCTU->ref[ctu.m_cuAddr * X265_MAX_PRED_MODE_PER_CTU * numPredDir];
m_reuseDepth = &m_reuseInterDataCTU->depth[ctu.m_cuAddr * ctu.m_numPartitions];
m_reuseModes = &m_reuseInterDataCTU->modes[ctu.m_cuAddr * ctu.m_numPartitions];
- if (m_param->analysisReuseLevel > 4)
+ if (reuseLevel > 4)
{
m_reusePartSize = &m_reuseInterDataCTU->partSize[ctu.m_cuAddr * ctu.m_numPartitions];
m_reuseMergeFlag = &m_reuseInterDataCTU->mergeFlag[ctu.m_cuAddr * ctu.m_numPartitions];
@@ -223,7 +226,7 @@
if (m_slice->m_sliceType == I_SLICE)
{
x265_analysis_intra_data* intraDataCTU = m_frame->m_analysisData.intraData;
- if (m_param->analysisLoad && m_param->analysisReuseLevel > 1)
+ if (m_param->analysisLoadReuseLevel > 1)
{
memcpy(ctu.m_cuDepth, &intraDataCTU->depth[ctu.m_cuAddr * numPartition], sizeof(uint8_t) * numPartition);
memcpy(ctu.m_lumaIntraDir, &intraDataCTU->modes[ctu.m_cuAddr * numPartition], sizeof(uint8_t) * numPartition);
@@ -234,9 +237,9 @@
}
else
{
- bool bCopyAnalysis = ((m_param->analysisLoad && m_param->analysisReuseLevel == 10) || (m_param->bAnalysisType == AVC_INFO && m_param->analysisReuseLevel >= 7 && ctu.m_numPartitions <= 16));
- bool BCompressInterCUrd0_4 = (m_param->bAnalysisType == AVC_INFO && m_param->analysisReuseLevel >= 7 && m_param->rdLevel <= 4);
- bool BCompressInterCUrd5_6 = (m_param->bAnalysisType == AVC_INFO && m_param->analysisReuseLevel >= 7 && m_param->rdLevel >= 5 && m_param->rdLevel <= 6);
+ bool bCopyAnalysis = ((m_param->analysisLoadReuseLevel == 10) || (m_param->bAnalysisType == AVC_INFO && m_param->analysisLoadReuseLevel >= 7 && ctu.m_numPartitions <= 16));
+ bool BCompressInterCUrd0_4 = (m_param->bAnalysisType == AVC_INFO && m_param->analysisLoadReuseLevel >= 7 && m_param->rdLevel <= 4);
+ bool BCompressInterCUrd5_6 = (m_param->bAnalysisType == AVC_INFO && m_param->analysisLoadReuseLevel >= 7 && m_param->rdLevel >= 5 && m_param->rdLevel <= 6);
bCopyAnalysis = bCopyAnalysis || BCompressInterCUrd0_4 || BCompressInterCUrd5_6;
if (bCopyAnalysis)
@@ -275,8 +278,8 @@
/* generate residual for entire CTU at once and copy to reconPic */
encodeResidue(ctu, cuGeom);
}
- else if ((m_param->analysisLoad && m_param->analysisReuseLevel == 10 && (!(m_param->bAnalysisType == HEVC_INFO) || m_slice->m_sliceType != P_SLICE)) ||
- ((m_param->bAnalysisType == AVC_INFO) && m_param->analysisReuseLevel >= 7 && ctu.m_numPartitions <= 16))
+ else if ((m_param->analysisLoadReuseLevel == 10 && (!(m_param->bAnalysisType == HEVC_INFO) || m_slice->m_sliceType != P_SLICE)) ||
+ ((m_param->bAnalysisType == AVC_INFO) && m_param->analysisLoadReuseLevel >= 7 && ctu.m_numPartitions <= 16))
{
x265_analysis_inter_data* interDataCTU = m_frame->m_analysisData.interData;
int posCTU = ctu.m_cuAddr * numPartition;
@@ -456,7 +459,7 @@
int bestCUQP = qp;
int lambdaQP = lqp;
bool doQPRefine = (bDecidedDepth && depth <= m_slice->m_pps->maxCuDQPDepth) || (!bDecidedDepth && depth == m_slice->m_pps->maxCuDQPDepth);
- if (m_param->analysisReuseLevel >= 7)
+ if (m_param->analysisLoadReuseLevel >= 7)
doQPRefine = false;
if (doQPRefine)
{
@@ -1164,7 +1167,7 @@
SplitData splitCUData;
bool bHEVCBlockAnalysis = (m_param->bAnalysisType == AVC_INFO && cuGeom.numPartitions > 16);
- bool bRefineAVCAnalysis = (m_param->analysisReuseLevel == 7 && (m_modeFlag[0] || m_modeFlag[1]));
+ bool bRefineAVCAnalysis = (m_param->analysisLoadReuseLevel == 7 && (m_modeFlag[0] || m_modeFlag[1]));
bool bNooffloading = !(m_param->bAnalysisType == AVC_INFO);
if (bHEVCBlockAnalysis || bRefineAVCAnalysis || bNooffloading)
@@ -1259,7 +1262,7 @@
mightSplit &= !bDecidedDepth;
}
}
- if ((m_param->analysisLoad && m_param->analysisReuseLevel > 1 && m_param->analysisReuseLevel != 10))
+ if ((m_param->analysisLoadReuseLevel > 1 && m_param->analysisLoadReuseLevel != 10))
{
if (mightNotSplit && depth == m_reuseDepth[cuGeom.absPartIdx])
{
@@ -1273,7 +1276,7 @@
if (m_param->rdLevel)
skipModes = m_param->bEnableEarlySkip && md.bestMode;
}
- if (m_param->analysisReuseLevel > 4 && m_reusePartSize[cuGeom.absPartIdx] == SIZE_2Nx2N)
+ if (m_param->analysisLoadReuseLevel > 4 && m_reusePartSize[cuGeom.absPartIdx] == SIZE_2Nx2N)
{
if (m_reuseModes[cuGeom.absPartIdx] != MODE_INTRA && m_reuseModes[cuGeom.absPartIdx] != 4)
{
@@ -1300,7 +1303,8 @@
}
}
/* Step 1. Evaluate Merge/Skip candidates for likely early-outs, if skip mode was not set above */
- if ((mightNotSplit && depth >= minDepth && !md.bestMode && !bCtuInfoCheck) || (m_param->bAnalysisType == AVC_INFO && m_param->analysisReuseLevel == 7 && (m_modeFlag[0] || m_modeFlag[1]))) /* TODO: Re-evaluate if analysis load/save still works */
+ if ((mightNotSplit && depth >= minDepth && !md.bestMode && !bCtuInfoCheck) || (m_param->bAnalysisType == AVC_INFO && m_param->analysisLoadReuseLevel == 7 && (m_modeFlag[0] || m_modeFlag[1])))
+ /* TODO: Re-evaluate if analysis load/save still works */
{
/* Compute Merge Cost */
md.pred[PRED_MERGE].cu.initSubCU(parentCTU, cuGeom, qp);
@@ -1310,7 +1314,7 @@
skipModes = (m_param->bEnableEarlySkip || m_refineLevel == 2)
&& md.bestMode && md.bestMode->cu.isSkipped(0); // TODO: sa8d threshold per depth
}
- if (md.bestMode && m_param->bEnableRecursionSkip && !bCtuInfoCheck && !(m_param->bAnalysisType == AVC_INFO && m_param->analysisReuseLevel == 7 && (m_modeFlag[0] || m_modeFlag[1])))
+ if (md.bestMode && m_param->bEnableRecursionSkip && !bCtuInfoCheck && !(m_param->bAnalysisType == AVC_INFO && m_param->analysisLoadReuseLevel == 7 && (m_modeFlag[0] || m_modeFlag[1])))
{
skipRecursion = md.bestMode->cu.isSkipped(0);
if (mightSplit && depth >= minDepth && !skipRecursion)
@@ -1321,7 +1325,7 @@
skipRecursion = complexityCheckCU(*md.bestMode);
}
}
- if (m_param->bAnalysisType == AVC_INFO && md.bestMode && cuGeom.numPartitions <= 16 && m_param->analysisReuseLevel == 7)
+ if (m_param->bAnalysisType == AVC_INFO && md.bestMode && cuGeom.numPartitions <= 16 && m_param->analysisLoadReuseLevel == 7)
skipRecursion = true;
/* Step 2. Evaluate each of the 4 split sub-blocks in series */
if (mightSplit && !skipRecursion)
@@ -1378,7 +1382,7 @@
splitPred->sa8dCost = m_rdCost.calcRdSADCost((uint32_t)splitPred->distortion, splitPred->sa8dBits);
}
/* If analysis mode is simple do not Evaluate other modes */
- if (m_param->bAnalysisType == AVC_INFO && m_param->analysisReuseLevel == 7)
+ if (m_param->bAnalysisType == AVC_INFO && m_param->analysisLoadReuseLevel == 7)
{
if (m_slice->m_sliceType == P_SLICE)
{
@@ -1857,7 +1861,7 @@
SplitData splitCUData;
bool bHEVCBlockAnalysis = (m_param->bAnalysisType == AVC_INFO && cuGeom.numPartitions > 16);
- bool bRefineAVCAnalysis = (m_param->analysisReuseLevel == 7 && (m_modeFlag[0] || m_modeFlag[1]));
+ bool bRefineAVCAnalysis = (m_param->analysisLoadReuseLevel == 7 && (m_modeFlag[0] || m_modeFlag[1]));
bool bNooffloading = !(m_param->bAnalysisType == AVC_INFO);
if (bHEVCBlockAnalysis || bRefineAVCAnalysis || bNooffloading)
@@ -1953,7 +1957,7 @@
mightSplit &= !bDecidedDepth;
}
}
- if (m_param->analysisLoad && m_param->analysisReuseLevel > 1 && m_param->analysisReuseLevel != 10)
+ if (m_param->analysisLoadReuseLevel > 1 && m_param->analysisLoadReuseLevel != 10)
{
if (mightNotSplit && depth == m_reuseDepth[cuGeom.absPartIdx])
{
@@ -1971,7 +1975,7 @@
if (m_param->bEnableRecursionSkip && depth && m_modeDepth[depth - 1].bestMode)
skipRecursion = md.bestMode && !md.bestMode->cu.getQtRootCbf(0);
}
- if (m_param->analysisReuseLevel > 4 && m_reusePartSize[cuGeom.absPartIdx] == SIZE_2Nx2N)
+ if (m_param->analysisLoadReuseLevel > 4 && m_reusePartSize[cuGeom.absPartIdx] == SIZE_2Nx2N)
skipRectAmp = true && !!md.bestMode;
}
}
@@ -1999,7 +2003,7 @@
}
/* Step 1. Evaluate Merge/Skip candidates for likely early-outs */
if ((mightNotSplit && !md.bestMode && !bCtuInfoCheck) ||
- (m_param->bAnalysisType == AVC_INFO && m_param->analysisReuseLevel == 7 && (m_modeFlag[0] || m_modeFlag[1])))
+ (m_param->bAnalysisType == AVC_INFO && m_param->analysisLoadReuseLevel == 7 && (m_modeFlag[0] || m_modeFlag[1])))
{
md.pred[PRED_SKIP].cu.initSubCU(parentCTU, cuGeom, qp);
md.pred[PRED_MERGE].cu.initSubCU(parentCTU, cuGeom, qp);
@@ -2014,7 +2018,7 @@
if (m_param->bEnableRecursionSkip && depth && m_modeDepth[depth - 1].bestMode)
skipRecursion = md.bestMode && !md.bestMode->cu.getQtRootCbf(0);
}
- if (m_param->bAnalysisType == AVC_INFO && md.bestMode && cuGeom.numPartitions <= 16 && m_param->analysisReuseLevel == 7)
+ if (m_param->bAnalysisType == AVC_INFO && md.bestMode && cuGeom.numPartitions <= 16 && m_param->analysisLoadReuseLevel == 7)
skipRecursion = true;
// estimate split cost
/* Step 2. Evaluate each of the 4 split sub-blocks in series */
@@ -2068,7 +2072,7 @@
checkDQPForSplitPred(*splitPred, cuGeom);
}
/* If analysis mode is simple do not Evaluate other modes */
- if (m_param->bAnalysisType == AVC_INFO && m_param->analysisReuseLevel == 7)
+ if (m_param->bAnalysisType == AVC_INFO && m_param->analysisLoadReuseLevel == 7)
{
if (m_slice->m_sliceType == P_SLICE)
{
@@ -2461,7 +2465,7 @@
for (uint32_t part = 0; part < numPU; part++)
{
PredictionUnit pu(mode.cu, cuGeom, part);
- if ((m_param->analysisLoad && m_param->analysisReuseLevel == 10) || (m_param->bAnalysisType == AVC_INFO && m_param->analysisReuseLevel >= 7))
+ if (m_param->analysisLoadReuseLevel == 10 || (m_param->bAnalysisType == AVC_INFO && m_param->analysisLoadReuseLevel >= 7))
{
x265_analysis_inter_data* interDataCTU = m_frame->m_analysisData.interData;
int cuIdx = (mode.cu.m_cuAddr * parentCTU.m_numPartitions) + cuGeom.absPartIdx;
@@ -2552,7 +2556,7 @@
checkDQPForSplitPred(*md.bestMode, cuGeom);
}
- if (m_param->bAnalysisType == AVC_INFO && m_param->analysisReuseLevel == 7)
+ if (m_param->bAnalysisType == AVC_INFO && m_param->analysisLoadReuseLevel == 7)
{
for (int list = 0; list < m_slice->isInterB() + 1; list++)
{
@@ -2607,7 +2611,7 @@
if (m_slice->m_pps->bUseDQP && nextDepth <= m_slice->m_pps->maxCuDQPDepth)
nextQP = setLambdaFromQP(parentCTU, calculateQpforCuSize(parentCTU, childGeom));
- int lamdaQP = (m_param->analysisReuseLevel >= 7) ? nextQP : lqp;
+ int lamdaQP = (m_param->analysisLoadReuseLevel >= 7) ? nextQP : lqp;
if (split)
m_param->rdLevel > 4 ? compressInterCU_rd5_6(parentCTU, childGeom, nextQP) : compressInterCU_rd0_4(parentCTU, childGeom, nextQP);
@@ -3013,7 +3017,7 @@
interMode.cu.setPredModeSubParts(MODE_INTER);
int numPredDir = m_slice->isInterP() ? 1 : 2;
- if (m_param->analysisLoad && m_reuseInterDataCTU && m_param->analysisReuseLevel > 1 && m_param->analysisReuseLevel != 10)
+ if (m_param->analysisLoadReuseLevel > 1 && m_param->analysisLoadReuseLevel != 10 && m_reuseInterDataCTU)
{
int refOffset = cuGeom.geomRecurId * 16 * numPredDir + partSize * numPredDir * 2;
int index = 0;
@@ -3056,7 +3060,7 @@
}
interMode.sa8dCost = m_rdCost.calcRdSADCost((uint32_t)interMode.distortion, interMode.sa8dBits);
- if (m_param->analysisSave && m_reuseInterDataCTU && m_param->analysisReuseLevel > 1)
+ if (m_param->analysisSaveReuseLevel > 1 && m_reuseInterDataCTU)
{
int refOffset = cuGeom.geomRecurId * 16 * numPredDir + partSize * numPredDir * 2;
int index = 0;
@@ -3078,7 +3082,7 @@
interMode.cu.setPredModeSubParts(MODE_INTER);
int numPredDir = m_slice->isInterP() ? 1 : 2;
- if (m_param->analysisLoad && m_reuseInterDataCTU && m_param->analysisReuseLevel > 1 && m_param->analysisReuseLevel != 10)
+ if (m_param->analysisLoadReuseLevel > 1 && m_param->analysisLoadReuseLevel != 10 && m_reuseInterDataCTU)
{
int refOffset = cuGeom.geomRecurId * 16 * numPredDir + partSize * numPredDir * 2;
int index = 0;
@@ -3113,7 +3117,7 @@
/* predInterSearch sets interMode.sa8dBits, but this is ignored */
encodeResAndCalcRdInterCU(interMode, cuGeom);
- if (m_param->analysisSave && m_reuseInterDataCTU && m_param->analysisReuseLevel > 1)
+ if (m_param->analysisSaveReuseLevel > 1 && m_reuseInterDataCTU)
{
int refOffset = cuGeom.geomRecurId * 16 * numPredDir + partSize * numPredDir * 2;
int index = 0;
@@ -3612,7 +3616,7 @@
qp += distortionData->offset[ctu.m_cuAddr];
}
- if (m_param->analysisLoad && m_param->analysisReuseLevel == 10 && m_param->rc.cuTree)
+ if (m_param->analysisLoadReuseLevel == 10 && m_param->rc.cuTree)
{
int cuIdx = (ctu.m_cuAddr * ctu.m_numPartitions) + cuGeom.absPartIdx;
if (ctu.m_slice->m_sliceType == I_SLICE)
diff -r 52135ffd9bcd -r 4d870dd7c314 source/encoder/api.cpp
--- a/source/encoder/api.cpp Mon Dec 23 14:40:32 2019 +0530
+++ b/source/encoder/api.cpp Mon Dec 16 17:55:07 2019 +0530
@@ -771,6 +771,12 @@
int numDir = 2; //irrespective of P or B slices set direction as 2
uint32_t numPlanes = param->internalCsp == X265_CSP_I400 ? 1 : 3;
+ int maxReuseLevel = X265_MAX(param->analysisSaveReuseLevel, param->analysisLoadReuseLevel);
+ int minReuseLevel = (param->analysisSaveReuseLevel && param->analysisLoadReuseLevel) ?
+ X265_MIN(param->analysisSaveReuseLevel, param->analysisLoadReuseLevel) : maxReuseLevel;
+
+ bool isMultiPassOpt = param->analysisMultiPassRefine || param->analysisMultiPassDistortion;
+
#if X265_DEPTH < 10 && (LINKED_10BIT || LINKED_12BIT)
uint32_t numCUs_sse_t = param->internalBitDepth > 8 ? analysis->numCUsInFrame << 1 : analysis->numCUsInFrame;
#elif X265_DEPTH >= 10 && LINKED_8BIT
@@ -778,7 +784,7 @@
#else
uint32_t numCUs_sse_t = analysis->numCUsInFrame;
#endif
- if (param->analysisMultiPassRefine || param->analysisMultiPassDistortion || param->ctuDistortionRefine)
+ if (isMultiPassOpt || param->ctuDistortionRefine)
{
//Allocate memory for distortionData pointer
CHECKED_MALLOC_ZERO(distortionData, x265_analysis_distortion_data, 1);
@@ -792,7 +798,7 @@
analysis->distortionData = distortionData;
}
- if (param->bDisableLookahead && isVbv)
+ if (!isMultiPassOpt && param->bDisableLookahead && isVbv)
{
CHECKED_MALLOC_ZERO(analysis->lookahead.intraSatdForVbv, uint32_t, analysis->numCuInHeight);
CHECKED_MALLOC_ZERO(analysis->lookahead.satdForVbv, uint32_t, analysis->numCuInHeight);
@@ -801,20 +807,23 @@
}
//Allocate memory for weightParam pointer
- if (!(param->bAnalysisType == AVC_INFO))
+ if (!isMultiPassOpt && !(param->bAnalysisType == AVC_INFO))
CHECKED_MALLOC_ZERO(analysis->wt, x265_weight_param, numPlanes * numDir);
- if (param->analysisReuseLevel < 2)
+ if (maxReuseLevel < 2)
return;
//Allocate memory for intraData pointer
CHECKED_MALLOC_ZERO(intraData, x265_analysis_intra_data, 1);
CHECKED_MALLOC(intraData->depth, uint8_t, analysis->numPartitions * analysis->numCUsInFrame);
- CHECKED_MALLOC_ZERO(intraData->modes, uint8_t, analysis->numPartitions * analysis->numCUsInFrame);
- CHECKED_MALLOC_ZERO(intraData->partSizes, char, analysis->numPartitions * analysis->numCUsInFrame);
- CHECKED_MALLOC_ZERO(intraData->chromaModes, uint8_t, analysis->numPartitions * analysis->numCUsInFrame);
- if (param->rc.cuTree)
- CHECKED_MALLOC_ZERO(intraData->cuQPOff, int8_t, analysis->numPartitions * analysis->numCUsInFrame);
+ if (!isMultiPassOpt)
+ {
+ CHECKED_MALLOC_ZERO(intraData->modes, uint8_t, analysis->numPartitions * analysis->numCUsInFrame);
+ CHECKED_MALLOC_ZERO(intraData->partSizes, char, analysis->numPartitions * analysis->numCUsInFrame);
+ CHECKED_MALLOC_ZERO(intraData->chromaModes, uint8_t, analysis->numPartitions * analysis->numCUsInFrame);
+ if (param->rc.cuTree)
+ CHECKED_MALLOC_ZERO(intraData->cuQPOff, int8_t, analysis->numPartitions * analysis->numCUsInFrame);
+ }
analysis->intraData = intraData;
//Allocate memory for interData pointer based on ReuseLevels
@@ -822,19 +831,19 @@
CHECKED_MALLOC(interData->depth, uint8_t, analysis->numPartitions * analysis->numCUsInFrame);
CHECKED_MALLOC_ZERO(interData->modes, uint8_t, analysis->numPartitions * analysis->numCUsInFrame);
- if (param->rc.cuTree)
+ if (param->rc.cuTree && !isMultiPassOpt)
CHECKED_MALLOC_ZERO(interData->cuQPOff, int8_t, analysis->numPartitions * analysis->numCUsInFrame);
CHECKED_MALLOC_ZERO(interData->mvpIdx[0], uint8_t, analysis->numPartitions * analysis->numCUsInFrame);
CHECKED_MALLOC_ZERO(interData->mvpIdx[1], uint8_t, analysis->numPartitions * analysis->numCUsInFrame);
CHECKED_MALLOC_ZERO(interData->mv[0], x265_analysis_MV, analysis->numPartitions * analysis->numCUsInFrame);
CHECKED_MALLOC_ZERO(interData->mv[1], x265_analysis_MV, analysis->numPartitions * analysis->numCUsInFrame);
- if (param->analysisReuseLevel > 4)
+ if (maxReuseLevel > 4)
{
CHECKED_MALLOC_ZERO(interData->partSize, uint8_t, analysis->numPartitions * analysis->numCUsInFrame);
CHECKED_MALLOC_ZERO(interData->mergeFlag, uint8_t, analysis->numPartitions * analysis->numCUsInFrame);
}
- if (param->analysisReuseLevel >= 7)
+ if (maxReuseLevel >= 7)
{
CHECKED_MALLOC_ZERO(interData->interDir, uint8_t, analysis->numPartitions * analysis->numCUsInFrame);
CHECKED_MALLOC_ZERO(interData->sadCost, int64_t, analysis->numPartitions * analysis->numCUsInFrame);
@@ -844,14 +853,13 @@
CHECKED_MALLOC_ZERO(analysis->modeFlag[dir], uint8_t, analysis->numPartitions * analysis->numCUsInFrame);
}
}
- else
+ if ((minReuseLevel >= 2) && (minReuseLevel <= 6))
{
- if (param->analysisMultiPassRefine || param->analysisMultiPassDistortion){
- CHECKED_MALLOC_ZERO(interData->ref, int32_t, 2 * analysis->numPartitions * analysis->numCUsInFrame);
- }
- else
- CHECKED_MALLOC_ZERO(interData->ref, int32_t, analysis->numCUsInFrame * X265_MAX_PRED_MODE_PER_CTU * numDir);
+ CHECKED_MALLOC_ZERO(interData->ref, int32_t, analysis->numCUsInFrame * X265_MAX_PRED_MODE_PER_CTU * numDir);
}
+ if (isMultiPassOpt)
+ CHECKED_MALLOC_ZERO(interData->ref, int32_t, 2 * analysis->numPartitions * analysis->numCUsInFrame);
+
analysis->interData = interData;
return;
@@ -862,10 +870,15 @@
void x265_free_analysis_data(x265_param *param, x265_analysis_data* analysis)
{
+ int maxReuseLevel = X265_MAX(param->analysisSaveReuseLevel, param->analysisLoadReuseLevel);
+ int minReuseLevel = (param->analysisSaveReuseLevel && param->analysisLoadReuseLevel) ?
+ X265_MIN(param->analysisSaveReuseLevel, param->analysisLoadReuseLevel) : maxReuseLevel;
+
bool isVbv = param->rc.vbvMaxBitrate > 0 && param->rc.vbvBufferSize > 0;
+ bool isMultiPassOpt = param->analysisMultiPassRefine || param->analysisMultiPassDistortion;
//Free memory for Lookahead pointers
- if (param->bDisableLookahead && isVbv)
+ if (!isMultiPassOpt && param->bDisableLookahead && isVbv)
{
X265_FREE(analysis->lookahead.satdForVbv);
X265_FREE(analysis->lookahead.intraSatdForVbv);
@@ -887,21 +900,24 @@
}
/* Early exit freeing weights alone if level is 1 (when there is no analysis inter/intra) */
- if (analysis->wt && !(param->bAnalysisType == AVC_INFO))
+ if (!isMultiPassOpt && analysis->wt && !(param->bAnalysisType == AVC_INFO))
X265_FREE(analysis->wt);
- if (param->analysisReuseLevel < 2)
+ if (maxReuseLevel < 2)
return;
//Free memory for intraData pointers
if (analysis->intraData)
{
X265_FREE((analysis->intraData)->depth);
- X265_FREE((analysis->intraData)->modes);
- X265_FREE((analysis->intraData)->partSizes);
- X265_FREE((analysis->intraData)->chromaModes);
- if (param->rc.cuTree)
- X265_FREE((analysis->intraData)->cuQPOff);
+ if (!isMultiPassOpt)
+ {
+ X265_FREE((analysis->intraData)->modes);
+ X265_FREE((analysis->intraData)->partSizes);
+ X265_FREE((analysis->intraData)->chromaModes);
+ if (param->rc.cuTree)
+ X265_FREE((analysis->intraData)->cuQPOff);
+ }
X265_FREE(analysis->intraData);
analysis->intraData = NULL;
}
@@ -911,19 +927,19 @@
{
X265_FREE((analysis->interData)->depth);
X265_FREE((analysis->interData)->modes);
- if (param->rc.cuTree)
+ if (!isMultiPassOpt && param->rc.cuTree)
X265_FREE((analysis->interData)->cuQPOff);
X265_FREE((analysis->interData)->mvpIdx[0]);
X265_FREE((analysis->interData)->mvpIdx[1]);
X265_FREE((analysis->interData)->mv[0]);
X265_FREE((analysis->interData)->mv[1]);
- if (param->analysisReuseLevel > 4)
+ if (maxReuseLevel > 4)
{
X265_FREE((analysis->interData)->mergeFlag);
X265_FREE((analysis->interData)->partSize);
}
- if (param->analysisReuseLevel >= 7)
+ if (maxReuseLevel >= 7)
{
int numDir = 2;
X265_FREE((analysis->interData)->interDir);
@@ -932,13 +948,13 @@
{
X265_FREE((analysis->interData)->refIdx[dir]);
if (analysis->modeFlag[dir] != NULL)
- {
+ {
X265_FREE(analysis->modeFlag[dir]);
analysis->modeFlag[dir] = NULL;
}
}
}
- else
+ if (((minReuseLevel >= 2) && (minReuseLevel <= 6)) || isMultiPassOpt)
X265_FREE((analysis->interData)->ref);
X265_FREE(analysis->interData);
analysis->interData = NULL;
diff -r 52135ffd9bcd -r 4d870dd7c314 source/encoder/encoder.cpp
--- a/source/encoder/encoder.cpp Mon Dec 23 14:40:32 2019 +0530
+++ b/source/encoder/encoder.cpp Mon Dec 16 17:55:07 2019 +0530
@@ -650,7 +650,7 @@
if (analysis_data->sliceType == X265_TYPE_IDR || analysis_data->sliceType == X265_TYPE_I)
{
curFrame->m_analysisData.sliceType = X265_TYPE_I;
- if (m_param->analysisReuseLevel < 7)
+ if (m_param->analysisLoadReuseLevel < 7)
return -1;
curFrame->m_analysisData.numPartitions = m_param->num4x4Partitions;
int num16x16inCUWidth = m_param->maxCUSize >> 4;
@@ -680,7 +680,7 @@
else
{
uint32_t numDir = analysis_data->sliceType == X265_TYPE_P ? 1 : 2;
- if (m_param->analysisReuseLevel < 7)
+ if (m_param->analysisLoadReuseLevel < 7)
return -1;
curFrame->m_analysisData.numPartitions = m_param->num4x4Partitions;
int num16x16inCUWidth = m_param->maxCUSize >> 4;
@@ -722,7 +722,7 @@
(interData)->mvpIdx[k][cuPos + cuOffset] = (srcInterData)->mvpIdx[k][(mbIndex * 16) + cuOffset];
(interData)->refIdx[k][cuPos + cuOffset] = (srcInterData)->refIdx[k][(mbIndex * 16) + cuOffset];
memcpy(&(interData)->mv[k][cuPos + cuOffset], &(srcInterData)->mv[k][(mbIndex * 16) + cuOffset], sizeof(MV));
- if (m_param->analysisReuseLevel == 7 && numPU == PU_2Nx2N &&
+ if (m_param->analysisLoadReuseLevel == 7 && numPU == PU_2Nx2N &&
((interData)->depth[cuPos + cuOffset] == (m_param->maxCUSize >> 5)))
{
int mv_x = (interData)->mv[k][cuPos + cuOffset].x;
@@ -756,7 +756,7 @@
if (analysis_data->sliceType == X265_TYPE_IDR || analysis_data->sliceType == X265_TYPE_I)
{
curFrame->m_analysisData.sliceType = X265_TYPE_I;
- if (m_param->analysisReuseLevel < 2)
+ if (m_param->analysisLoadReuseLevel < 2)
return -1;
curFrame->m_analysisData.numPartitions = m_param->num4x4Partitions;
@@ -777,7 +777,7 @@
else
{
uint32_t numDir = analysis_data->sliceType == X265_TYPE_P ? 1 : 2;
- if (m_param->analysisReuseLevel < 2)
+ if (m_param->analysisLoadReuseLevel < 2)
return -1;
curFrame->m_analysisData.numPartitions = m_param->num4x4Partitions;
@@ -790,7 +790,7 @@
memset(&(currInterData)->depth[count], (interData)->depth[d], bytes);
memset(&(currInterData)->modes[count], (interData)->modes[d], bytes);
memcpy(&(currInterData)->sadCost[count], &(analysis_data->interData)->sadCost[d], bytes);
- if (m_param->analysisReuseLevel > 4)
+ if (m_param->analysisLoadReuseLevel > 4)
{
memset(&(currInterData)->partSize[count], (interData)->partSize[d], bytes);
int numPU = nbPartsTable[(interData)->partSize[d]];
@@ -798,7 +798,7 @@
{
if (pu) d++;
(currInterData)->mergeFlag[count + pu] = (interData)->mergeFlag[d];
- if (m_param->analysisReuseLevel >= 7)
+ if (m_param->analysisLoadReuseLevel >= 7)
{
(currInterData)->interDir[count + pu] = (interData)->interDir[d];
for (uint32_t i = 0; i < numDir; i++)
@@ -806,7 +806,7 @@
(currInterData)->mvpIdx[i][count + pu] = (interData)->mvpIdx[i][d];
(currInterData)->refIdx[i][count + pu] = (interData)->refIdx[i][d];
memcpy(&(currInterData)->mv[i][count + pu], &(interData)->mv[i][d], sizeof(MV));
- if (m_param->analysisReuseLevel == 7 && numPU == PU_2Nx2N && m_param->num4x4Partitions <= 16)
+ if (m_param->analysisLoadReuseLevel == 7 && numPU == PU_2Nx2N && m_param->num4x4Partitions <= 16)
{
int mv_x = (currInterData)->mv[i][count + pu].x;
int mv_y = (currInterData)->mv[i][count + pu].y;
@@ -3577,10 +3577,46 @@
p->rc.rfConstantMin = 0;
}
- if (!(p->bAnalysisType == HEVC_INFO) && (p->analysisLoad || p->analysisSave) && p->rc.cuTree && p->analysisReuseLevel < 10)
- {
- x265_log(p, X265_LOG_WARNING, "cu-tree works only with analysis reuse level 10, Disabling cu-tree\n");
- p->rc.cuTree = 0;
+ if (p->analysisSaveReuseLevel && !p->analysisSave)
+ {
+ x265_log(p, X265_LOG_WARNING, "analysis-save-reuse-level can be set only when analysis-save is enabled."
+ " Resetting analysis-save-reuse-level to 0.\n");
+ p->analysisSaveReuseLevel = 0;
+ }
+
+ if (p->analysisLoadReuseLevel && !p->analysisLoad)
+ {
+ x265_log(p, X265_LOG_WARNING, "analysis-load-reuse-level can be set only when analysis-load is enabled."
+ " Resetting analysis-load-reuse-level to 0.\n");
+ p->analysisLoadReuseLevel = 0;
+ }
+
+ if (p->analysisSave && !p->analysisSaveReuseLevel)
+ {
+ x265_log(p, X265_LOG_WARNING, "analysis-save-reuse-level must be set when analysis-save is enabled."
+ " Setting analysis-save-reuse-level to 5.\n");
+ p->analysisSaveReuseLevel = 5;
+ }
+
+ if (p->analysisLoad && !p->analysisLoadReuseLevel)
+ {
+ x265_log(p, X265_LOG_WARNING, "analysis-load-reuse-level must be set when analysis-load is enabled."
+ " Setting analysis-load-reuse-level to 5.\n");
+ p->analysisLoadReuseLevel = 5;
+ }
+
+ if ((p->bAnalysisType == DEFAULT) && p->rc.cuTree)
+ {
+ if (p->analysisSaveReuseLevel && p->analysisSaveReuseLevel < 10)
+ {
+ x265_log(p, X265_LOG_WARNING, "cu-tree works only with analysis-save-reuse-level 10, Disabling cu-tree\n");
+ p->rc.cuTree = 0;
+ }
+ if (p->analysisLoadReuseLevel && p->analysisLoadReuseLevel < 10)
+ {
+ x265_log(p, X265_LOG_WARNING, "cu-tree works only with analysis-load-reuse-level 10, Disabling cu-tree\n");
+ p->rc.cuTree = 0;
+ }
}
if ((p->analysisLoad || p->analysisSave) && (p->bDistributeModeAnalysis || p->bDistributeMotionEstimation))
@@ -3602,45 +3638,37 @@
{
p->scaleFactor = 0;
}
- else if ((!p->analysisLoad && !p->analysisSave) || (p->analysisReuseLevel > 6 && p->analysisReuseLevel != 10))
+ else if ((p->analysisSaveReuseLevel > 6 && p->analysisSaveReuseLevel != 10) || (p->analysisLoadReuseLevel > 6 && p->analysisLoadReuseLevel != 10))
{
- x265_log(p, X265_LOG_WARNING, "Input scaling works with analysis load/save and analysis-reuse-level 1-6 and 10. Disabling scale-factor.\n");
+ x265_log(p, X265_LOG_WARNING, "Input scaling works with analysis-save/load and analysis-save/load-reuse-level 1-6 and 10. Disabling scale-factor.\n");
p->scaleFactor = 0;
}
}
- if (p->intraRefine)
- {
- if (!p->analysisLoad || p->analysisReuseLevel < 10)
- {
- x265_log(p, X265_LOG_WARNING, "Intra refinement requires analysis load, analysis-reuse-level 10. Disabling intra refine.\n");
- p->intraRefine = 0;
- }
- }
-
- if (p->interRefine)
- {
- if (!p->analysisLoad || p->analysisReuseLevel < 10)
- {
- x265_log(p, X265_LOG_WARNING, "Inter refinement requires analysis load, analysis-reuse-level 10. Disabling inter refine.\n");
- p->interRefine = 0;
- }
- }
-
- if (p->bDynamicRefine)
- {
- if (!p->analysisLoad || p->analysisReuseLevel < 10)
- {
- x265_log(p, X265_LOG_WARNING, "Dynamic refinement requires analysis load, analysis-reuse-level 10. Disabling dynamic refine.\n");
- p->bDynamicRefine = 0;
- }
+ if (p->intraRefine && p->analysisLoadReuseLevel && p->analysisLoadReuseLevel < 10)
+ {
+ x265_log(p, X265_LOG_WARNING, "Intra refinement requires analysis load, analysis-load-reuse-level 10. Disabling intra refine.\n");
+ p->intraRefine = 0;
+ }
+
+ if (p->interRefine && p->analysisLoadReuseLevel && p->analysisLoadReuseLevel < 10)
+ {
+ x265_log(p, X265_LOG_WARNING, "Inter refinement requires analysis load, analysis-load-reuse-level 10. Disabling inter refine.\n");
+ p->interRefine = 0;
+ }
+
+ if (p->bDynamicRefine && p->analysisLoadReuseLevel && p->analysisLoadReuseLevel < 10)
+ {
+ x265_log(p, X265_LOG_WARNING, "Dynamic refinement requires analysis load, analysis-load-reuse-level 10. Disabling dynamic refine.\n");
+ p->bDynamicRefine = 0;
+
if (p->interRefine)
{
x265_log(p, X265_LOG_WARNING, "Inter refine cannot be used with dynamic refine. Disabling refine-inter.\n");
p->interRefine = 0;
}
}
- if (p->scaleFactor && p->analysisLoad && !p->interRefine && !p->bDynamicRefine && p->analysisReuseLevel == 10)
+ if (p->scaleFactor && !p->interRefine && !p->bDynamicRefine && p->analysisLoadReuseLevel == 10)
{
x265_log(p, X265_LOG_WARNING, "Inter refinement 0 is not supported with scaling and analysis-reuse-level=10. Enabling refine-inter 1.\n");
p->interRefine = 1;
@@ -4205,7 +4233,7 @@
{
if (m_param->bAnalysisType == HEVC_INFO)
return;
- if (m_param->analysisReuseLevel < 2)
+ if (m_param->analysisLoadReuseLevel < 2)
return;
uint8_t *tempBuf = NULL, *depthBuf = NULL, *modeBuf = NULL, *partSizes = NULL;
@@ -4265,7 +4293,7 @@
uint32_t numDir = analysis->sliceType == X265_TYPE_P ? 1 : 2;
uint32_t numPlanes = m_param->internalCsp == X265_CSP_I400 ? 1 : 3;
X265_FREAD((WeightParam*)analysis->wt, sizeof(WeightParam), numPlanes * numDir, m_analysisFileIn, (picIn->analysisData.wt));
- if (m_param->analysisReuseLevel < 2)
+ if (m_param->analysisLoadReuseLevel < 2)
return;
uint8_t *tempBuf = NULL, *depthBuf = NULL, *modeBuf = NULL, *partSize = NULL, *mergeFlag = NULL;
@@ -4274,9 +4302,9 @@
int8_t* refIdx[2];
int8_t* cuQPBuf = NULL;
- int numBuf = m_param->analysisReuseLevel > 4 ? 4 : 2;
+ int numBuf = m_param->analysisLoadReuseLevel > 4 ? 4 : 2;
bool bIntraInInter = false;
- if (m_param->analysisReuseLevel == 10)
+ if (m_param->analysisLoadReuseLevel == 10)
{
numBuf++;
bIntraInInter = (analysis->sliceType == X265_TYPE_P || m_param->bIntraInBFrames);
@@ -4299,14 +4327,14 @@
X265_FREAD(modeBuf, sizeof(uint8_t), depthBytes, m_analysisFileIn, interPic->modes);
if (m_param->rc.cuTree) { X265_FREAD(cuQPBuf, sizeof(int8_t), depthBytes, m_analysisFileIn, interPic->cuQPOff); }
- if (m_param->analysisReuseLevel > 4)
+ if (m_param->analysisLoadReuseLevel > 4)
{
partSize = modeBuf + depthBytes;
mergeFlag = partSize + depthBytes;
X265_FREAD(partSize, sizeof(uint8_t), depthBytes, m_analysisFileIn, interPic->partSize);
X265_FREAD(mergeFlag, sizeof(uint8_t), depthBytes, m_analysisFileIn, interPic->mergeFlag);
- if (m_param->analysisReuseLevel == 10)
+ if (m_param->analysisLoadReuseLevel == 10)
{
interDir = mergeFlag + depthBytes;
X265_FREAD(interDir, sizeof(uint8_t), depthBytes, m_analysisFileIn, interPic->interDir);
@@ -4337,7 +4365,7 @@
memset(&(analysis->interData)->modes[count], modeBuf[d], bytes);
if (m_param->rc.cuTree)
memset(&(analysis->interData)->cuQPOff[count], cuQPBuf[d], bytes);
- if (m_param->analysisReuseLevel > 4)
+ if (m_param->analysisLoadReuseLevel > 4)
{
if (m_param->scaleFactor && modeBuf[d] == MODE_INTRA && partSize[d] == SIZE_NxN)
partSize[d] = SIZE_2Nx2N;
@@ -4347,7 +4375,7 @@
{
if (pu) d++;
(analysis->interData)->mergeFlag[count + pu] = mergeFlag[d];
- if (m_param->analysisReuseLevel == 10)
+ if (m_param->analysisLoadReuseLevel == 10)
{
(analysis->interData)->interDir[count + pu] = interDir[d];
for (uint32_t i = 0; i < numDir; i++)
@@ -4363,7 +4391,7 @@
}
}
}
- if (m_param->analysisReuseLevel == 10 && bIntraInInter)
+ if (m_param->analysisLoadReuseLevel == 10 && bIntraInInter)
memset(&(analysis->intraData)->chromaModes[count], chromaDir[d], bytes);
}
count += bytes;
@@ -4373,7 +4401,7 @@
X265_FREE(cuQPBuf);
X265_FREE(tempBuf);
}
- if (m_param->analysisReuseLevel == 10)
+ if (m_param->analysisLoadReuseLevel == 10)
{
if (m_param->bAnalysisType != HEVC_INFO)
{
@@ -4556,7 +4584,7 @@
if (analysis->sliceType == X265_TYPE_IDR || analysis->sliceType == X265_TYPE_I)
{
- if (m_param->analysisReuseLevel < 2)
+ if (m_param->analysisLoadReuseLevel < 2)
return;
uint8_t *tempBuf = NULL, *depthBuf = NULL, *modeBuf = NULL, *partSizes = NULL;
@@ -4625,7 +4653,7 @@
uint32_t numDir = analysis->sliceType == X265_TYPE_P ? 1 : 2;
uint32_t numPlanes = m_param->internalCsp == X265_CSP_I400 ? 1 : 3;
X265_FREAD((WeightParam*)analysis->wt, sizeof(WeightParam), numPlanes * numDir, m_analysisFileIn, (picIn->analysisData.wt));
- if (m_param->analysisReuseLevel < 2)
+ if (m_param->analysisLoadReuseLevel < 2)
return;
uint8_t *tempBuf = NULL, *depthBuf = NULL, *modeBuf = NULL, *partSize = NULL, *mergeFlag = NULL;
@@ -4634,9 +4662,9 @@
int8_t* refIdx[2];
int8_t* cuQPBuf = NULL;
- int numBuf = m_param->analysisReuseLevel > 4 ? 4 : 2;
+ int numBuf = m_param->analysisLoadReuseLevel > 4 ? 4 : 2;
bool bIntraInInter = false;
- if (m_param->analysisReuseLevel == 10)
+ if (m_param->analysisLoadReuseLevel == 10)
{
numBuf++;
bIntraInInter = (analysis->sliceType == X265_TYPE_P || m_param->bIntraInBFrames);
@@ -4652,13 +4680,13 @@
X265_FREAD(depthBuf, sizeof(uint8_t), depthBytes, m_analysisFileIn, interPic->depth);
X265_FREAD(modeBuf, sizeof(uint8_t), depthBytes, m_analysisFileIn, interPic->modes);
if (m_param->rc.cuTree) { X265_FREAD(cuQPBuf, sizeof(int8_t), depthBytes, m_analysisFileIn, interPic->cuQPOff); }
- if (m_param->analysisReuseLevel > 4)
+ if (m_param->analysisLoadReuseLevel > 4)
{
partSize = modeBuf + depthBytes;
mergeFlag = partSize + depthBytes;
X265_FREAD(partSize, sizeof(uint8_t), depthBytes, m_analysisFileIn, interPic->partSize);
X265_FREAD(mergeFlag, sizeof(uint8_t), depthBytes, m_analysisFileIn, interPic->mergeFlag);
- if (m_param->analysisReuseLevel == 10)
+ if (m_param->analysisLoadReuseLevel == 10)
{
interDir = mergeFlag + depthBytes;
X265_FREAD(interDir, sizeof(uint8_t), depthBytes, m_analysisFileIn, interPic->interDir);
@@ -4702,10 +4730,10 @@
memset(&(analysis->interData)->modes[count], modeBuf[d], bytes);
if (m_param->rc.cuTree)
memset(&(analysis->interData)->cuQPOff[count], cuQPBuf[d], bytes);
- if (m_param->analysisReuseLevel == 10 && bIntraInInter)
+ if (m_param->analysisLoadReuseLevel == 10 && bIntraInInter)
memset(&(analysis->intraData)->chromaModes[count], chromaDir[d], bytes);
- if (m_param->analysisReuseLevel > 4)
+ if (m_param->analysisLoadReuseLevel > 4)
{
puOrientation puOrient;
puOrient.init();
@@ -4731,7 +4759,7 @@
d++;
(analysis->interData)->mergeFlag[count + pu] = mergeFlag[d];
- if (m_param->analysisReuseLevel == 10)
+ if (m_param->analysisLoadReuseLevel == 10)
{
(analysis->interData)->interDir[count + pu] = interDir[d];
MV mvCopy[2];
@@ -4766,7 +4794,7 @@
X265_FREE(cuQPBuf);
X265_FREE(tempBuf);
- if (m_param->analysisReuseLevel == 10)
+ if (m_param->analysisLoadReuseLevel == 10)
{
for (uint32_t i = 0; i < numDir; i++)
{
@@ -4855,7 +4883,6 @@
X265_PARAM_VALIDATE(saveParam->intraRefresh, sizeof(int), 1, &m_param->bIntraRefresh, intra-refresh);
X265_PARAM_VALIDATE(saveParam->maxNumReferences, sizeof(int), 1, &m_param->maxNumReferences, ref);
- X265_PARAM_VALIDATE(saveParam->analysisReuseLevel, sizeof(int), 1, &m_param->analysisReuseLevel, analysis-reuse-level);
X265_PARAM_VALIDATE(saveParam->keyframeMax, sizeof(int), 1, &m_param->keyframeMax, keyint);
X265_PARAM_VALIDATE(saveParam->keyframeMin, sizeof(int), 1, &m_param->keyframeMin, min-keyint);
X265_PARAM_VALIDATE(saveParam->openGOP, sizeof(int), 1, &m_param->bOpenGOP, open-gop);
@@ -4871,6 +4898,7 @@
int sourceHeight, sourceWidth;
if (writeFlag)
{
+ X265_PARAM_VALIDATE(saveParam->analysisReuseLevel, sizeof(int), 1, &m_param->analysisSaveReuseLevel, analysis - save - reuse - level);
sourceHeight = m_param->sourceHeight - m_conformanceWindow.bottomOffset;
sourceWidth = m_param->sourceWidth - m_conformanceWindow.rightOffset;
X265_PARAM_VALIDATE(saveParam->sourceWidth, sizeof(int), 1, &sourceWidth, res-width);
@@ -4880,10 +4908,35 @@
else
{
fileOffset = m_analysisFileIn;
+
+ int saveLevel = 0;
+ bool isIncompatibleReuseLevel = false;
+ int loadLevel = m_param->analysisLoadReuseLevel;
+
+ X265_FREAD(&saveLevel, sizeof(int), 1, m_analysisFileIn, &(saveParam->analysisReuseLevel));
+
+ if (loadLevel == 10 && saveLevel != 10)
+ isIncompatibleReuseLevel = true;
+ else if (((loadLevel >= 7) && (loadLevel <= 9)) && ((saveLevel < 7) || (saveLevel > 9)))
+ isIncompatibleReuseLevel = true;
+ else if ((loadLevel == 5 || loadLevel == 6) && ((saveLevel != 5) && (saveLevel != 6)))
+ isIncompatibleReuseLevel = true;
+ else if ((loadLevel >= 2 && loadLevel <= 4) && (saveLevel < 2 || saveLevel > 6))
+ isIncompatibleReuseLevel = true;
+ else if (!saveLevel)
+ isIncompatibleReuseLevel = true;
+
+ if (isIncompatibleReuseLevel)
+ {
+ x265_log(NULL, X265_LOG_ERROR, "Error reading analysis data. Incompatible reuse-levels.\n");
+ m_aborted = true;
+ return -1;
+ }
+
bool error = false;
int curSourceHeight = m_param->sourceHeight - m_conformanceWindow.bottomOffset;
int curSourceWidth = m_param->sourceWidth - m_conformanceWindow.rightOffset;
-
+
X265_FREAD(&sourceWidth, sizeof(int), 1, m_analysisFileIn, &(saveParam->sourceWidth));
X265_FREAD(&sourceHeight, sizeof(int), 1, m_analysisFileIn, &(saveParam->sourceHeight));
X265_FREAD(&readValue, sizeof(int), 1, m_analysisFileIn, &(saveParam->maxCUSize));
@@ -5205,7 +5258,7 @@
analysis->frameRecordSize += analysis->numCUsInFrame * sizeof(sse_t);
}
- if (m_param->analysisReuseLevel > 1)
+ if (m_param->analysisSaveReuseLevel > 1)
{
if (analysis->sliceType == X265_TYPE_IDR || analysis->sliceType == X265_TYPE_I)
@@ -5258,14 +5311,14 @@
interDataCTU->depth[depthBytes] = depth;
predMode = ctu->m_predMode[absPartIdx];
- if (m_param->analysisReuseLevel != 10 && ctu->m_refIdx[1][absPartIdx] != -1)
+ if (m_param->analysisSaveReuseLevel != 10 && ctu->m_refIdx[1][absPartIdx] != -1)
predMode = 4; // used as indicator if the block is coded as bidir
interDataCTU->modes[depthBytes] = predMode;
if (m_param->rc.cuTree)
interDataCTU->cuQPOff[depthBytes] = (int8_t)(ctu->m_qpAnalysis[absPartIdx] - baseQP);
- if (m_param->analysisReuseLevel > 4)
+ if (m_param->analysisSaveReuseLevel > 4)
{
partSize = ctu->m_partSize[absPartIdx];
interDataCTU->partSize[depthBytes] = partSize;
@@ -5278,7 +5331,7 @@
if (puIdx) depthBytes++;
interDataCTU->mergeFlag[depthBytes] = ctu->m_mergeFlag[puabsPartIdx];
- if (m_param->analysisReuseLevel == 10)
+ if (m_param->analysisSaveReuseLevel == 10)
{
interDataCTU->interDir[depthBytes] = ctu->m_interDir[puabsPartIdx];
for (uint32_t dir = 0; dir < numDir; dir++)
@@ -5289,12 +5342,12 @@
}
}
}
- if (m_param->analysisReuseLevel == 10 && bIntraInInter)
+ if (m_param->analysisSaveReuseLevel == 10 && bIntraInInter)
intraDataCTU->chromaModes[depthBytes] = ctu->m_chromaIntraDir[absPartIdx];
}
absPartIdx += ctu->m_numPartitions >> (depth * 2);
}
- if (m_param->analysisReuseLevel == 10 && bIntraInInter)
+ if (m_param->analysisSaveReuseLevel == 10 && bIntraInInter)
memcpy(&intraDataCTU->modes[ctu->m_cuAddr * ctu->m_numPartitions], ctu->m_lumaIntraDir, sizeof(uint8_t)* ctu->m_numPartitions);
}
}
@@ -5309,10 +5362,10 @@
analysis->frameRecordSize += depthBytes * 2;
if (m_param->rc.cuTree)
analysis->frameRecordSize += (sizeof(int8_t) * depthBytes);
- if (m_param->analysisReuseLevel > 4)
+ if (m_param->analysisSaveReuseLevel > 4)
analysis->frameRecordSize += (depthBytes * 2);
- if (m_param->analysisReuseLevel == 10)
+ if (m_param->analysisSaveReuseLevel == 10)
{
/* Add Size of interDir, mvpIdx, refIdx, mv, luma and chroma modes */
analysis->frameRecordSize += depthBytes;
@@ -5344,7 +5397,7 @@
if (analysis->sliceType > X265_TYPE_I)
X265_FWRITE((WeightParam*)analysis->wt, sizeof(WeightParam), numPlanes * numDir, m_analysisFileOut);
- if (m_param->analysisReuseLevel < 2)
+ if (m_param->analysisSaveReuseLevel < 2)
return;
if (analysis->sliceType == X265_TYPE_IDR || analysis->sliceType == X265_TYPE_I)
@@ -5362,11 +5415,11 @@
X265_FWRITE((analysis->interData)->modes, sizeof(uint8_t), depthBytes, m_analysisFileOut);
if (m_param->rc.cuTree)
X265_FWRITE((analysis->interData)->cuQPOff, sizeof(int8_t), depthBytes, m_analysisFileOut);
- if (m_param->analysisReuseLevel > 4)
+ if (m_param->analysisSaveReuseLevel > 4)
{
X265_FWRITE((analysis->interData)->partSize, sizeof(uint8_t), depthBytes, m_analysisFileOut);
X265_FWRITE((analysis->interData)->mergeFlag, sizeof(uint8_t), depthBytes, m_analysisFileOut);
- if (m_param->analysisReuseLevel == 10)
+ if (m_param->analysisSaveReuseLevel == 10)
{
X265_FWRITE((analysis->interData)->interDir, sizeof(uint8_t), depthBytes, m_analysisFileOut);
if (bIntraInInter) X265_FWRITE((analysis->intraData)->chromaModes, sizeof(uint8_t), depthBytes, m_analysisFileOut);
@@ -5380,7 +5433,7 @@
X265_FWRITE((analysis->intraData)->modes, sizeof(uint8_t), analysis->numCUsInFrame * analysis->numPartitions, m_analysisFileOut);
}
}
- if (m_param->analysisReuseLevel != 10)
+ if (m_param->analysisSaveReuseLevel != 10)
X265_FWRITE((analysis->interData)->ref, sizeof(int32_t), analysis->numCUsInFrame * X265_MAX_PRED_MODE_PER_CTU * numDir, m_analysisFileOut);
}
diff -r 52135ffd9bcd -r 4d870dd7c314 source/encoder/search.cpp
--- a/source/encoder/search.cpp Mon Dec 23 14:40:32 2019 +0530
+++ b/source/encoder/search.cpp Mon Dec 16 17:55:07 2019 +0530
@@ -2208,7 +2208,7 @@
x265_analysis_inter_data* interDataCTU = NULL;
int cuIdx;
cuIdx = (interMode.cu.m_cuAddr * m_param->num4x4Partitions) + cuGeom.absPartIdx;
- if (m_param->analysisReuseLevel == 10 && m_param->interRefine > 1)
+ if (m_param->analysisLoadReuseLevel == 10 && m_param->interRefine > 1)
{
interDataCTU = m_frame->m_analysisData.interData;
if ((cu.m_predMode[pu.puAbsPartIdx] == interDataCTU->modes[cuIdx + pu.puAbsPartIdx])
@@ -2227,7 +2227,7 @@
cu.getNeighbourMV(puIdx, pu.puAbsPartIdx, interMode.interNeighbours);
/* Uni-directional prediction */
- if ((m_param->analysisLoad && m_param->analysisReuseLevel > 1 && m_param->analysisReuseLevel != 10)
+ if ((m_param->analysisLoadReuseLevel > 1 && m_param->analysisLoadReuseLevel != 10)
|| (m_param->analysisMultiPassRefine && m_param->rc.bStatRead) || (m_param->bAnalysisType == AVC_INFO) || (useAsMVP))
{
for (int list = 0; list < numPredDir; list++)
diff -r 52135ffd9bcd -r 4d870dd7c314 source/x265.h
--- a/source/x265.h Mon Dec 23 14:40:32 2019 +0530
+++ b/source/x265.h Mon Dec 16 17:55:07 2019 +0530
@@ -320,7 +320,7 @@
typedef enum
{
- NO_INFO = 0,
+ DEFAULT = 0,
AVC_INFO = 1,
HEVC_INFO = 2,
}AnalysisRefineType;
@@ -1656,7 +1656,7 @@
/* A value between 1 and 10 (both inclusive) determines the level of
* information stored/reused in analysis save/load. Higher the refine
- * level higher the information stored/reused. Default is 5 */
+ * level higher the information stored/reused. Default is 5. Now deprecated. */
int analysisReuseLevel;
/* Limit Sample Adaptive Offset filter computation by early terminating SAO
@@ -1859,6 +1859,16 @@
/* Enable HME search ranges for L0, L1 and L2 respectively. */
int hmeRange[3];
+
+ /* A value between 1 and 10 (both inclusive) determines the level of
+ * analysis information stored in analysis-save. Higher the refine level higher
+ * the information stored. Default is 5 */
+ int analysisSaveReuseLevel;
+
+ /* A value between 1 and 10 (both inclusive) determines the level of
+ * analysis information reused in analysis-load. Higher the refine level higher
+ * the information reused. Default is 5 */
+ int analysisLoadReuseLevel;
} x265_param;
/* x265_param_alloc:
diff -r 52135ffd9bcd -r 4d870dd7c314 source/x265cli.h
--- a/source/x265cli.h Mon Dec 23 14:40:32 2019 +0530
+++ b/source/x265cli.h Mon Dec 16 17:55:07 2019 +0530
@@ -275,7 +275,9 @@
{ "no-multi-pass-opt-rps", no_argument, NULL, 0 },
{ "analysis-reuse-mode", required_argument, NULL, 0 }, /* DEPRECATED */
{ "analysis-reuse-file", required_argument, NULL, 0 },
- { "analysis-reuse-level", required_argument, NULL, 0 },
+ { "analysis-reuse-level", required_argument, NULL, 0 }, /* DEPRECATED */
+ { "analysis-save-reuse-level", required_argument, NULL, 0 },
+ { "analysis-load-reuse-level", required_argument, NULL, 0 },
{ "analysis-save", required_argument, NULL, 0 },
{ "analysis-load", required_argument, NULL, 0 },
{ "scale-factor", required_argument, NULL, 0 },
@@ -549,7 +551,9 @@
H0(" --analysis-save <filename> Dump analysis info into the specified file. Default Disabled\n");
H0(" --analysis-load <filename> Load analysis buffers from the file specified. Default Disabled\n");
H0(" --analysis-reuse-file <filename> Specify file name used for either dumping or reading analysis data. Deault x265_analysis.dat\n");
- H0(" --analysis-reuse-level <1..10> Level of analysis reuse indicates amount of info stored/reused in save/load mode, 1:least..10:most. Default %d\n", param->analysisReuseLevel);
+ H0(" --analysis-reuse-level <1..10> Level of analysis reuse indicates amount of info stored/reused in save/load mode, 1:least..10:most. Now deprecated. Default %d\n", param->analysisReuseLevel);
+ H0(" --analysis-save-reuse-level <1..10> Indicates the amount of analysis info stored in save mode, 1:least..10:most. Default %d\n", param->analysisSaveReuseLevel);
+ H0(" --analysis-load-reuse-level <1..10> Indicates the amount of analysis info reused in load mode, 1:least..10:most. Default %d\n", param->analysisLoadReuseLevel);
H0(" --refine-analysis-type <string> Reuse anlaysis information received through API call. Supported options are avc and hevc. Default disabled - %d\n", param->bAnalysisType);
H0(" --scale-factor <int> Specify factor by which input video is scaled down for analysis save mode. Default %d\n", param->scaleFactor);
H0(" --refine-intra <0..4> Enable intra refinement for encode that uses analysis-load.\n"
-------------- next part --------------
A non-text attachment was scrubbed...
Name: x265_push-1.patch
Type: text/x-patch
Size: 61211 bytes
Desc: not available
URL: <http://mailman.videolan.org/pipermail/x265-devel/attachments/20200107/f70deb9a/attachment-0001.bin>
More information about the x265-devel
mailing list