[x265] [PATCH] analysis: dump the best depth and re-use it for analysis-mode=load
gopu at multicorewareinc.com
gopu at multicorewareinc.com
Thu Jan 8 13:05:44 CET 2015
# HG changeset patch
# User Gopu Govindaswamy <gopu at multicorewareinc.com>
# Date 1420713484 -19800
# Thu Jan 08 16:08:04 2015 +0530
# Node ID f307fc02faf6111642737516cf34d3dd5f3b2820
# Parent 24da7b0accdafb1f9e7e54182c114cb264d09bab
analysis: dump the best depth and re-use it for analysis-mode=load
For inter frame currently sharing the best ref and in addition to that
share the depth and re-use it for analysis mode=load, the best depth can be
shared only if Predmode is MODE_SKIP, otherwise ignored. currently this changes
works only rdlevel > 4
diff -r 24da7b0accda -r f307fc02faf6 source/common/common.h
--- a/source/common/common.h Wed Dec 24 10:34:59 2014 +0530
+++ b/source/common/common.h Thu Jan 08 16:08:04 2015 +0530
@@ -353,10 +353,12 @@
}
};
-/* Stores inter (motion estimation) analysis data for a single frame */
+/* Stores inter analysis data for a single frame */
struct analysis_inter_data
{
int32_t* ref;
+ uint8_t* depth;
+ uint8_t* modes;
};
/* Stores intra analysis data for a single frame. This struct needs better packing */
diff -r 24da7b0accda -r f307fc02faf6 source/encoder/analysis.cpp
--- a/source/encoder/analysis.cpp Wed Dec 24 10:34:59 2014 +0530
+++ b/source/encoder/analysis.cpp Thu Jan 08 16:08:04 2015 +0530
@@ -143,9 +143,9 @@
}
}
+ uint32_t zOrder = 0;
if (m_slice->m_sliceType == I_SLICE)
{
- uint32_t zOrder = 0;
compressIntraCU(ctu, cuGeom, zOrder);
if (m_param->analysisMode == X265_ANALYSIS_SAVE && m_frame->m_analysisData.intraData)
{
@@ -173,7 +173,15 @@
else if (m_param->rdLevel <= 4)
compressInterCU_rd0_4(ctu, cuGeom);
else
- compressInterCU_rd5_6(ctu, cuGeom);
+ {
+ compressInterCU_rd5_6(ctu, cuGeom, zOrder);
+ if (m_param->analysisMode == X265_ANALYSIS_SAVE && m_frame->m_analysisData.interData)
+ {
+ CUData *bestCU = &m_modeDepth[0].bestMode->cu;
+ memcpy(&m_reuseInterDataCTU->depth[ctu.m_cuAddr * numPartition], bestCU->m_cuDepth, sizeof(uint8_t) * numPartition);
+ memcpy(&m_reuseInterDataCTU->modes[ctu.m_cuAddr * numPartition], bestCU->m_predMode, sizeof(uint8_t) * numPartition);
+ }
+ }
}
return *m_modeDepth[0].bestMode;
@@ -1037,7 +1045,7 @@
md.bestMode->reconYuv.copyToPicYuv(*m_frame->m_reconPic, cuAddr, cuGeom.encodeIdx);
}
-void Analysis::compressInterCU_rd5_6(const CUData& parentCTU, const CUGeom& cuGeom)
+void Analysis::compressInterCU_rd5_6(const CUData& parentCTU, const CUGeom& cuGeom, uint32_t &zOrder)
{
uint32_t depth = cuGeom.depth;
ModeDepth& md = m_modeDepth[depth];
@@ -1046,6 +1054,45 @@
bool mightSplit = !(cuGeom.flags & CUGeom::LEAF);
bool mightNotSplit = !(cuGeom.flags & CUGeom::SPLIT_MANDATORY);
+ if (m_param->analysisMode == X265_ANALYSIS_LOAD)
+ {
+ uint8_t* reuseDepth = &m_reuseInterDataCTU->depth[parentCTU.m_cuAddr * parentCTU.m_numPartitions];
+ uint8_t* reuseModes = &m_reuseInterDataCTU->modes[parentCTU.m_cuAddr * parentCTU.m_numPartitions];
+ if (mightNotSplit && depth == reuseDepth[zOrder] && zOrder == cuGeom.encodeIdx && reuseModes[zOrder] == MODE_SKIP)
+ {
+ md.pred[PRED_SKIP].cu.initSubCU(parentCTU, cuGeom);
+ md.pred[PRED_MERGE].cu.initSubCU(parentCTU, cuGeom);
+ checkMerge2Nx2N_rd5_6(md.pred[PRED_SKIP], md.pred[PRED_MERGE], cuGeom);
+
+ if ((m_slice->m_sliceType != B_SLICE || m_param->bIntraInBFrames) &&
+ (!m_param->bEnableCbfFastMode || md.bestMode->cu.getQtRootCbf(0)))
+ {
+ md.pred[PRED_INTRA].cu.initSubCU(parentCTU, cuGeom);
+ checkIntra(md.pred[PRED_INTRA], cuGeom, SIZE_2Nx2N, NULL);
+ checkBestMode(md.pred[PRED_INTRA], depth);
+
+ if (depth == g_maxCUDepth && cuGeom.log2CUSize > m_slice->m_sps->quadtreeTULog2MinSize)
+ {
+ md.pred[PRED_INTRA_NxN].cu.initSubCU(parentCTU, cuGeom);
+ checkIntra(md.pred[PRED_INTRA_NxN], cuGeom, SIZE_NxN, &reuseModes[zOrder]);
+ checkBestMode(md.pred[PRED_INTRA_NxN], depth);
+ }
+ }
+
+ if (m_bTryLossless)
+ tryLossless(cuGeom);
+
+ if (mightSplit)
+ addSplitFlagCost(*md.bestMode, cuGeom.depth);
+
+ // increment zOrder offset to point to next best depth in sharedDepth buffer
+ zOrder += g_depthInc[g_maxCUDepth - 1][reuseDepth[zOrder]];
+
+ mightSplit = false;
+ mightNotSplit = false;
+ }
+ }
+
if (mightNotSplit)
{
md.pred[PRED_SKIP].cu.initSubCU(parentCTU, cuGeom);
@@ -1177,7 +1224,7 @@
{
m_modeDepth[0].fencYuv.copyPartToYuv(nd.fencYuv, childGeom.encodeIdx);
m_rqt[nextDepth].cur.load(*nextContext);
- compressInterCU_rd5_6(parentCTU, childGeom);
+ compressInterCU_rd5_6(parentCTU, childGeom, zOrder);
// Save best CU and pred data for this sub CU
splitCU->copyPartFrom(nd.bestMode->cu, childGeom, subPartIdx);
@@ -1186,7 +1233,10 @@
nextContext = &nd.bestMode->contexts;
}
else
+ {
splitCU->setEmptyPart(childGeom, subPartIdx);
+ zOrder += g_depthInc[g_maxCUDepth - 1][nextDepth];
+ }
}
nextContext->store(splitPred->contexts);
if (mightNotSplit)
diff -r 24da7b0accda -r f307fc02faf6 source/encoder/analysis.h
--- a/source/encoder/analysis.h Wed Dec 24 10:34:59 2014 +0530
+++ b/source/encoder/analysis.h Thu Jan 08 16:08:04 2015 +0530
@@ -101,7 +101,7 @@
/* full analysis for a P or B slice CU */
void compressInterCU_dist(const CUData& parentCTU, const CUGeom& cuGeom);
void compressInterCU_rd0_4(const CUData& parentCTU, const CUGeom& cuGeom);
- void compressInterCU_rd5_6(const CUData& parentCTU, const CUGeom& cuGeom);
+ void compressInterCU_rd5_6(const CUData& parentCTU, const CUGeom& cuGeom, uint32_t &zOrder);
/* measure merge and skip */
void checkMerge2Nx2N_rd0_4(Mode& skip, Mode& merge, const CUGeom& cuGeom);
diff -r 24da7b0accda -r f307fc02faf6 source/encoder/encoder.cpp
--- a/source/encoder/encoder.cpp Wed Dec 24 10:34:59 2014 +0530
+++ b/source/encoder/encoder.cpp Thu Jan 08 16:08:04 2015 +0530
@@ -1578,6 +1578,8 @@
analysis_inter_data *interData = (analysis_inter_data*)analysis->interData;
CHECKED_MALLOC_ZERO(interData, analysis_inter_data, 1);
CHECKED_MALLOC(interData->ref, int32_t, analysis->numCUsInFrame * X265_MAX_PRED_MODE_PER_CTU * 2);
+ CHECKED_MALLOC(interData->depth, uint8_t, analysis->numPartitions * analysis->numCUsInFrame);
+ CHECKED_MALLOC(interData->modes, uint8_t, analysis->numPartitions * analysis->numCUsInFrame);
analysis->interData = interData;
}
return;
@@ -1599,6 +1601,8 @@
else
{
X265_FREE(((analysis_inter_data*)analysis->interData)->ref);
+ X265_FREE(((analysis_inter_data*)analysis->interData)->depth);
+ X265_FREE(((analysis_inter_data*)analysis->interData)->modes);
X265_FREE(analysis->interData);
}
}
@@ -1662,12 +1666,16 @@
else if (analysis->sliceType == X265_TYPE_P)
{
X265_FREAD(((analysis_inter_data *)analysis->interData)->ref, sizeof(int32_t), analysis->numCUsInFrame * X265_MAX_PRED_MODE_PER_CTU, m_analysisFile);
+ X265_FREAD(((analysis_inter_data *)analysis->interData)->depth, sizeof(uint8_t), analysis->numCUsInFrame * analysis->numPartitions, m_analysisFile);
+ X265_FREAD(((analysis_inter_data *)analysis->interData)->modes, sizeof(uint8_t), analysis->numCUsInFrame * analysis->numPartitions, m_analysisFile);
consumedBytes += frameRecordSize;
totalConsumedBytes = consumedBytes;
}
else
{
X265_FREAD(((analysis_inter_data *)analysis->interData)->ref, sizeof(int32_t), analysis->numCUsInFrame * X265_MAX_PRED_MODE_PER_CTU * 2, m_analysisFile);
+ X265_FREAD(((analysis_inter_data *)analysis->interData)->depth, sizeof(uint8_t), analysis->numCUsInFrame * analysis->numPartitions, m_analysisFile);
+ X265_FREAD(((analysis_inter_data *)analysis->interData)->modes, sizeof(uint8_t), analysis->numCUsInFrame * analysis->numPartitions, m_analysisFile);
consumedBytes += frameRecordSize;
}
#undef X265_FREAD
@@ -1691,9 +1699,15 @@
if (analysis->sliceType == X265_TYPE_IDR || analysis->sliceType == X265_TYPE_I)
analysis->frameRecordSize += sizeof(uint8_t) * analysis->numCUsInFrame * analysis->numPartitions * 3;
else if (analysis->sliceType == X265_TYPE_P)
+ {
analysis->frameRecordSize += sizeof(int32_t) * analysis->numCUsInFrame * X265_MAX_PRED_MODE_PER_CTU;
+ analysis->frameRecordSize += sizeof(uint8_t) * analysis->numCUsInFrame * analysis->numPartitions * 2;
+ }
else
+ {
analysis->frameRecordSize += sizeof(int32_t) * analysis->numCUsInFrame * X265_MAX_PRED_MODE_PER_CTU * 2;
+ analysis->frameRecordSize += sizeof(uint8_t) * analysis->numCUsInFrame * analysis->numPartitions * 2;
+ }
X265_FWRITE(&analysis->frameRecordSize, sizeof(uint32_t), 1, m_analysisFile);
X265_FWRITE(&analysis->poc, sizeof(int), 1, m_analysisFile);
@@ -1709,11 +1723,15 @@
}
else if (analysis->sliceType == X265_TYPE_P)
{
- X265_FWRITE(((analysis_inter_data *)analysis->interData)->ref, sizeof(int32_t), analysis->numCUsInFrame * X265_MAX_PRED_MODE_PER_CTU, m_analysisFile);
+ X265_FWRITE(((analysis_inter_data*)analysis->interData)->ref, sizeof(int32_t), analysis->numCUsInFrame * X265_MAX_PRED_MODE_PER_CTU, m_analysisFile);
+ X265_FWRITE(((analysis_inter_data*)analysis->interData)->depth, sizeof(uint8_t), analysis->numCUsInFrame * analysis->numPartitions, m_analysisFile);
+ X265_FWRITE(((analysis_inter_data*)analysis->interData)->modes, sizeof(uint8_t), analysis->numCUsInFrame * analysis->numPartitions, m_analysisFile);
}
else
{
- X265_FWRITE(((analysis_inter_data *)analysis->interData)->ref, sizeof(int32_t), analysis->numCUsInFrame * X265_MAX_PRED_MODE_PER_CTU * 2, m_analysisFile);
+ X265_FWRITE(((analysis_inter_data*)analysis->interData)->ref, sizeof(int32_t), analysis->numCUsInFrame * X265_MAX_PRED_MODE_PER_CTU * 2, m_analysisFile);
+ X265_FWRITE(((analysis_inter_data*)analysis->interData)->depth, sizeof(uint8_t), analysis->numCUsInFrame * analysis->numPartitions, m_analysisFile);
+ X265_FWRITE(((analysis_inter_data*)analysis->interData)->modes, sizeof(uint8_t), analysis->numCUsInFrame * analysis->numPartitions, m_analysisFile);
}
#undef X265_FWRITE
}
More information about the x265-devel
mailing list