[x265] [PATCH] analysis: Dump best MV statistics and re-use this for analysis load mode
gopu at multicorewareinc.com
gopu at multicorewareinc.com
Tue Nov 11 07:02:59 CET 2014
# HG changeset patch
# User Gopu Govindaswamy <gopu at multicorewareinc.com>
# Date 1415685422 -19800
# Tue Nov 11 11:27:02 2014 +0530
# Node ID ae9d1d48303b3e1e7fbc1ce1fe480f318654378d
# Parent 32513a4c3bd435757347e729dc14b5a1c1c6ceef
analysis: Dump best MV statistics and re-use this for analysis load mode
this patch is to fix the bug in inter information sharing when using
analysis=load|save mode, existing algorithm always dump and share the last part
best MV for each prediction, but there is multiple part's each with its own
prediction, the fix is to dump and share all part best MV's for each prediction
diff -r 32513a4c3bd4 -r ae9d1d48303b source/encoder/analysis.cpp
--- a/source/encoder/analysis.cpp Mon Nov 10 12:39:54 2014 +0900
+++ b/source/encoder/analysis.cpp Tue Nov 11 11:27:02 2014 +0530
@@ -1407,12 +1407,17 @@
if (m_param->analysisMode == X265_ANALYSIS_LOAD && m_interAnalysisData)
{
- for (int32_t i = 0; i < numPredDir; i++)
+ for (uint32_t part = 0; part < interMode.cu.getNumPartInter(); part++)
{
- interMode.bestME[i].costZero = !!m_interAnalysisData->costZero[i];
- interMode.bestME[i].mv.x = m_interAnalysisData->mvx[i];
- interMode.bestME[i].mv.y = m_interAnalysisData->mvy[i];
- interMode.bestME[i].ref = m_interAnalysisData->ref[i];
+ MotionData* bestME = interMode.bestME[part];
+ for (int32_t i = 0; i < numPredDir; i++)
+ {
+ bestME[i].costZero = !!m_interAnalysisData->costZero[i];
+ bestME[i].mv.x = m_interAnalysisData->mvx[i];
+ bestME[i].mv.y = m_interAnalysisData->mvy[i];
+ bestME[i].ref = m_interAnalysisData->ref[i];
+ }
+ m_interAnalysisData++;
}
}
if (predInterSearch(interMode, cuGeom, false, false))
@@ -1425,17 +1430,21 @@
if (m_param->analysisMode == X265_ANALYSIS_SAVE && m_interAnalysisData)
{
- for (int32_t i = 0; i < numPredDir; i++)
+ for (uint32_t part = 0; part < interMode.cu.getNumPartInter(); part++)
{
- m_interAnalysisData->costZero[i] = interMode.bestME[i].costZero;
- m_interAnalysisData->mvx[i] = interMode.bestME[i].mv.x;
- m_interAnalysisData->mvy[i] = interMode.bestME[i].mv.y;
- m_interAnalysisData->ref[i] = interMode.bestME[i].ref;
+ MotionData* bestME = interMode.bestME[part];
+ for (int32_t i = 0; i < numPredDir; i++)
+ {
+ m_interAnalysisData->costZero[i] = bestME[i].costZero;
+ m_interAnalysisData->mvx[i] = bestME[i].mv.x;
+ m_interAnalysisData->mvy[i] = bestME[i].mv.y;
+ m_interAnalysisData->ref[i] = bestME[i].ref;
+ }
+ m_interAnalysisData->zOrder = cuGeom.encodeIdx;
+ m_interAnalysisData->depth = cuGeom.depth;
+ m_interAnalysisData++;
}
- m_interAnalysisData->zOrder = cuGeom.encodeIdx;
- m_interAnalysisData->depth = cuGeom.depth;
}
- m_interAnalysisData++;
}
else
{
@@ -1453,12 +1462,17 @@
if (m_param->analysisMode == X265_ANALYSIS_LOAD && m_interAnalysisData)
{
- for (int32_t i = 0; i < numPredDir; i++)
+ for (uint32_t part = 0; part < interMode.cu.getNumPartInter(); part++)
{
- interMode.bestME[i].costZero = !!m_interAnalysisData->costZero[i];
- interMode.bestME[i].mv.x = m_interAnalysisData->mvx[i];
- interMode.bestME[i].mv.y = m_interAnalysisData->mvy[i];
- interMode.bestME[i].ref = m_interAnalysisData->ref[i];
+ MotionData* bestME = interMode.bestME[part];
+ for (int32_t i = 0; i < numPredDir; i++)
+ {
+ bestME[i].costZero = !!m_interAnalysisData->costZero[i];
+ bestME[i].mv.x = m_interAnalysisData->mvx[i];
+ bestME[i].mv.y = m_interAnalysisData->mvy[i];
+ bestME[i].ref = m_interAnalysisData->ref[i];
+ }
+ m_interAnalysisData++;
}
}
if (predInterSearch(interMode, cuGeom, bMergeOnly, true))
@@ -1467,17 +1481,21 @@
encodeResAndCalcRdInterCU(interMode, cuGeom);
if (m_param->analysisMode == X265_ANALYSIS_SAVE && m_interAnalysisData)
{
- for (int32_t i = 0; i < numPredDir; i++)
+ for (uint32_t part = 0; part < interMode.cu.getNumPartInter(); part++)
{
- m_interAnalysisData->costZero[i] = interMode.bestME[i].costZero;
- m_interAnalysisData->mvx[i] = interMode.bestME[i].mv.x;
- m_interAnalysisData->mvy[i] = interMode.bestME[i].mv.y;
- m_interAnalysisData->ref[i] = interMode.bestME[i].ref;
+ MotionData* bestME = interMode.bestME[part];
+ for (int32_t i = 0; i < numPredDir; i++)
+ {
+ m_interAnalysisData->costZero[i] = bestME[i].costZero;
+ m_interAnalysisData->mvx[i] = bestME[i].mv.x;
+ m_interAnalysisData->mvy[i] = bestME[i].mv.y;
+ m_interAnalysisData->ref[i] = bestME[i].ref;
+ }
+ m_interAnalysisData->zOrder = cuGeom.encodeIdx;
+ m_interAnalysisData->depth = cuGeom.depth;
+ m_interAnalysisData++;
}
- m_interAnalysisData->zOrder = cuGeom.encodeIdx;
- m_interAnalysisData->depth = cuGeom.depth;
}
- m_interAnalysisData++;
}
else
{
diff -r 32513a4c3bd4 -r ae9d1d48303b source/encoder/search.cpp
--- a/source/encoder/search.cpp Mon Nov 10 12:39:54 2014 +0900
+++ b/source/encoder/search.cpp Tue Nov 11 11:27:02 2014 +0530
@@ -1872,6 +1872,7 @@
uint32_t bestCost = MAX_INT;
int mvpIdx = 0;
int merange = m_param->searchRange;
+ MotionData* bestME = interMode.bestME[part];
for (int i = 0; i < AMVP_NUM_CANDS; i++)
{
MV mvCand = interMode.amvpCand[list][ref][i];
@@ -1907,15 +1908,15 @@
/* tie goes to the smallest ref ID, just like --no-pme */
ScopedLock _lock(master.m_outputLock);
- if (cost < interMode.bestME[list].cost ||
- (cost == interMode.bestME[list].cost && ref < interMode.bestME[list].ref))
+ if (cost < bestME[list].cost ||
+ (cost == bestME[list].cost && ref < bestME[list].ref))
{
- interMode.bestME[list].mv = outmv;
- interMode.bestME[list].mvp = mvp;
- interMode.bestME[list].mvpIdx = mvpIdx;
- interMode.bestME[list].ref = ref;
- interMode.bestME[list].cost = cost;
- interMode.bestME[list].bits = bits;
+ bestME[list].mv = outmv;
+ bestME[list].mvp = mvp;
+ bestME[list].mvpIdx = mvpIdx;
+ bestME[list].ref = ref;
+ bestME[list].cost = cost;
+ bestME[list].bits = bits;
}
}
@@ -1944,6 +1945,8 @@
for (int puIdx = 0; puIdx < numPart; puIdx++)
{
+ MotionData* bestME = interMode.bestME[puIdx];
+
/* sets m_puAbsPartIdx, m_puWidth, m_puHeight */
initMotionCompensation(cu, cuGeom, puIdx);
@@ -1988,17 +1991,17 @@
uint32_t bidirCost = MAX_UINT;
int bidirBits = 0;
- interMode.bestME[0].cost = MAX_UINT;
- interMode.bestME[1].cost = MAX_UINT;
+ bestME[0].cost = MAX_UINT;
+ bestME[1].cost = MAX_UINT;
getBlkBits((PartSize)cu.m_partSize[0], slice->isInterP(), puIdx, lastMode, m_listSelBits);
/* Uni-directional prediction */
- if (m_param->analysisMode == X265_ANALYSIS_LOAD && interMode.bestME[0].ref >= 0)
+ if (m_param->analysisMode == X265_ANALYSIS_LOAD && bestME[0].ref >= 0)
{
for (int l = 0; l < numPredDir; l++)
{
- int ref = interMode.bestME[l].ref;
+ int ref = bestME[l].ref;
uint32_t bits = m_listSelBits[l] + MVP_IDX_BITS;
bits += getTUBits(ref, numRefIdx[l]);
@@ -2030,13 +2033,13 @@
MV mvmin, mvmax, outmv, mvp = interMode.amvpCand[l][ref][mvpIdx];
m_me.setMVP(mvp);
- MV bmv(interMode.bestME[l].mv.x, interMode.bestME[l].mv.y);
+ MV bmv(bestME[l].mv.x, bestME[l].mv.y);
int satdCost;
- if (interMode.bestME[l].costZero)
+ if (bestME[l].costZero)
satdCost = m_me.mvcost(bmv);
else
- satdCost = interMode.bestME[l].cost;
+ satdCost = bestME[l].cost;
/* Get total cost of partition, but only include MV bit cost once */
bits += m_me.bitcost(bmv);
@@ -2045,14 +2048,14 @@
/* Refine MVP selection, updates: mvp, mvpIdx, bits, cost */
checkBestMVP(interMode.amvpCand[l][ref], outmv, mvp, mvpIdx, bits, cost);
- if (cost < interMode.bestME[l].cost)
+ if (cost < bestME[l].cost)
{
- interMode.bestME[l].mv = outmv;
- interMode.bestME[l].mvp = mvp;
- interMode.bestME[l].mvpIdx = mvpIdx;
- interMode.bestME[l].ref = ref;
- interMode.bestME[l].cost = cost;
- interMode.bestME[l].bits = bits;
+ bestME[l].mv = outmv;
+ bestME[l].mvp = mvp;
+ bestME[l].mvpIdx = mvpIdx;
+ bestME[l].ref = ref;
+ bestME[l].cost = cost;
+ bestME[l].bits = bits;
}
}
}
@@ -2148,31 +2151,31 @@
/* Refine MVP selection, updates: mvp, mvpIdx, bits, cost */
checkBestMVP(interMode.amvpCand[l][ref], outmv, mvp, mvpIdx, bits, cost);
- if (cost < interMode.bestME[l].cost)
+ if (cost < bestME[l].cost)
{
- interMode.bestME[l].mv = outmv;
- interMode.bestME[l].mvp = mvp;
- interMode.bestME[l].mvpIdx = mvpIdx;
- interMode.bestME[l].ref = ref;
- interMode.bestME[l].cost = cost;
- interMode.bestME[l].bits = bits;
+ bestME[l].mv = outmv;
+ bestME[l].mvp = mvp;
+ bestME[l].mvpIdx = mvpIdx;
+ bestME[l].ref = ref;
+ bestME[l].cost = cost;
+ bestME[l].bits = bits;
}
}
}
}
/* Bi-directional prediction */
- if (slice->isInterB() && !cu.isBipredRestriction() && interMode.bestME[0].cost != MAX_UINT && interMode.bestME[1].cost != MAX_UINT)
+ if (slice->isInterB() && !cu.isBipredRestriction() && bestME[0].cost != MAX_UINT && bestME[1].cost != MAX_UINT)
{
- bidir[0] = interMode.bestME[0];
- bidir[1] = interMode.bestME[1];
+ bidir[0] = bestME[0];
+ bidir[1] = bestME[1];
/* Generate reference subpels */
- PicYuv* refPic0 = slice->m_refPicList[0][interMode.bestME[0].ref]->m_reconPic;
- PicYuv* refPic1 = slice->m_refPicList[1][interMode.bestME[1].ref]->m_reconPic;
+ PicYuv* refPic0 = slice->m_refPicList[0][bestME[0].ref]->m_reconPic;
+ PicYuv* refPic1 = slice->m_refPicList[1][bestME[1].ref]->m_reconPic;
Yuv* bidirYuv = m_rqt[cuGeom.depth].bidirPredYuv;
- predInterLumaPixel(bidirYuv[0], *refPic0, interMode.bestME[0].mv);
- predInterLumaPixel(bidirYuv[1], *refPic1, interMode.bestME[1].mv);
+ predInterLumaPixel(bidirYuv[0], *refPic0, bestME[0].mv);
+ predInterLumaPixel(bidirYuv[1], *refPic1, bestME[1].mv);
pixel *pred0 = bidirYuv[0].getLumaAddr(m_puAbsPartIdx);
pixel *pred1 = bidirYuv[1].getLumaAddr(m_puAbsPartIdx);
@@ -2181,10 +2184,10 @@
primitives.pixelavg_pp[partEnum](tmpPredYuv.m_buf[0], tmpPredYuv.m_size, pred0, bidirYuv[0].m_size, pred1, bidirYuv[1].m_size, 32);
int satdCost = m_me.bufSATD(tmpPredYuv.m_buf[0], tmpPredYuv.m_size);
- bidirBits = interMode.bestME[0].bits + interMode.bestME[1].bits + m_listSelBits[2] - (m_listSelBits[0] + m_listSelBits[1]);
+ bidirBits = bestME[0].bits + bestME[1].bits + m_listSelBits[2] - (m_listSelBits[0] + m_listSelBits[1]);
bidirCost = satdCost + m_rdCost.getCost(bidirBits);
- bool bTryZero = interMode.bestME[0].mv.notZero() || interMode.bestME[1].mv.notZero();
+ bool bTryZero = bestME[0].mv.notZero() || bestME[1].mv.notZero();
if (bTryZero)
{
/* Do not try zero MV if unidir motion predictors are beyond
@@ -2196,32 +2199,32 @@
mvmin <<= 2;
mvmax <<= 2;
- bTryZero &= interMode.bestME[0].mvp.checkRange(mvmin, mvmax);
- bTryZero &= interMode.bestME[1].mvp.checkRange(mvmin, mvmax);
+ bTryZero &= bestME[0].mvp.checkRange(mvmin, mvmax);
+ bTryZero &= bestME[1].mvp.checkRange(mvmin, mvmax);
}
if (bTryZero)
{
/* coincident blocks of the two reference pictures */
- pixel *ref0 = m_slice->m_mref[0][interMode.bestME[0].ref].getLumaAddr(cu.m_cuAddr, cuGeom.encodeIdx + m_puAbsPartIdx);
- pixel *ref1 = m_slice->m_mref[1][interMode.bestME[1].ref].getLumaAddr(cu.m_cuAddr, cuGeom.encodeIdx + m_puAbsPartIdx);
+ pixel *ref0 = m_slice->m_mref[0][bestME[0].ref].getLumaAddr(cu.m_cuAddr, cuGeom.encodeIdx + m_puAbsPartIdx);
+ pixel *ref1 = m_slice->m_mref[1][bestME[1].ref].getLumaAddr(cu.m_cuAddr, cuGeom.encodeIdx + m_puAbsPartIdx);
intptr_t refStride = slice->m_mref[0][0].lumaStride;
primitives.pixelavg_pp[partEnum](tmpPredYuv.m_buf[0], tmpPredYuv.m_size, ref0, refStride, ref1, refStride, 32);
satdCost = m_me.bufSATD(tmpPredYuv.m_buf[0], tmpPredYuv.m_size);
- MV mvp0 = interMode.bestME[0].mvp;
- int mvpIdx0 = interMode.bestME[0].mvpIdx;
- uint32_t bits0 = interMode.bestME[0].bits - m_me.bitcost(interMode.bestME[0].mv, mvp0) + m_me.bitcost(mvzero, mvp0);
-
- MV mvp1 = interMode.bestME[1].mvp;
- int mvpIdx1 = interMode.bestME[1].mvpIdx;
- uint32_t bits1 = interMode.bestME[1].bits - m_me.bitcost(interMode.bestME[1].mv, mvp1) + m_me.bitcost(mvzero, mvp1);
+ MV mvp0 = bestME[0].mvp;
+ int mvpIdx0 = bestME[0].mvpIdx;
+ uint32_t bits0 = bestME[0].bits - m_me.bitcost(bestME[0].mv, mvp0) + m_me.bitcost(mvzero, mvp0);
+
+ MV mvp1 = bestME[1].mvp;
+ int mvpIdx1 = bestME[1].mvpIdx;
+ uint32_t bits1 = bestME[1].bits - m_me.bitcost(bestME[1].mv, mvp1) + m_me.bitcost(mvzero, mvp1);
uint32_t cost = satdCost + m_rdCost.getCost(bits0) + m_rdCost.getCost(bits1);
/* refine MVP selection for zero mv, updates: mvp, mvpidx, bits, cost */
- checkBestMVP(interMode.amvpCand[0][interMode.bestME[0].ref], mvzero, mvp0, mvpIdx0, bits0, cost);
- checkBestMVP(interMode.amvpCand[1][interMode.bestME[1].ref], mvzero, mvp1, mvpIdx1, bits1, cost);
+ checkBestMVP(interMode.amvpCand[0][bestME[0].ref], mvzero, mvp0, mvpIdx0, bits0, cost);
+ checkBestMVP(interMode.amvpCand[1][bestME[1].ref], mvzero, mvp1, mvpIdx1, bits1, cost);
if (cost < bidirCost)
{
@@ -2243,7 +2246,7 @@
}
/* select best option and store into CU */
- if (mrgCost < bidirCost && mrgCost < interMode.bestME[0].cost && mrgCost < interMode.bestME[1].cost)
+ if (mrgCost < bidirCost && mrgCost < bestME[0].cost && mrgCost < bestME[1].cost)
{
cu.m_mergeFlag[m_puAbsPartIdx] = true;
cu.m_mvpIdx[0][m_puAbsPartIdx] = merge.index; // merge candidate ID is stored in L0 MVP idx
@@ -2255,39 +2258,39 @@
totalmebits += merge.bits;
}
- else if (bidirCost < interMode.bestME[0].cost && bidirCost < interMode.bestME[1].cost)
+ else if (bidirCost < bestME[0].cost && bidirCost < bestME[1].cost)
{
lastMode = 2;
cu.m_mergeFlag[m_puAbsPartIdx] = false;
cu.setPUInterDir(3, m_puAbsPartIdx, puIdx);
cu.setPUMv(0, bidir[0].mv, m_puAbsPartIdx, puIdx);
- cu.setPURefIdx(0, interMode.bestME[0].ref, m_puAbsPartIdx, puIdx);
+ cu.setPURefIdx(0, bestME[0].ref, m_puAbsPartIdx, puIdx);
cu.m_mvd[0][m_puAbsPartIdx] = bidir[0].mv - bidir[0].mvp;
cu.m_mvpIdx[0][m_puAbsPartIdx] = bidir[0].mvpIdx;
cu.setPUMv(1, bidir[1].mv, m_puAbsPartIdx, puIdx);
- cu.setPURefIdx(1, interMode.bestME[1].ref, m_puAbsPartIdx, puIdx);
+ cu.setPURefIdx(1, bestME[1].ref, m_puAbsPartIdx, puIdx);
cu.m_mvd[1][m_puAbsPartIdx] = bidir[1].mv - bidir[1].mvp;
cu.m_mvpIdx[1][m_puAbsPartIdx] = bidir[1].mvpIdx;
totalmebits += bidirBits;
}
- else if (interMode.bestME[0].cost <= interMode.bestME[1].cost)
+ else if (bestME[0].cost <= bestME[1].cost)
{
lastMode = 0;
cu.m_mergeFlag[m_puAbsPartIdx] = false;
cu.setPUInterDir(1, m_puAbsPartIdx, puIdx);
- cu.setPUMv(0, interMode.bestME[0].mv, m_puAbsPartIdx, puIdx);
- cu.setPURefIdx(0, interMode.bestME[0].ref, m_puAbsPartIdx, puIdx);
- cu.m_mvd[0][m_puAbsPartIdx] = interMode.bestME[0].mv - interMode.bestME[0].mvp;
- cu.m_mvpIdx[0][m_puAbsPartIdx] = interMode.bestME[0].mvpIdx;
+ cu.setPUMv(0, bestME[0].mv, m_puAbsPartIdx, puIdx);
+ cu.setPURefIdx(0, bestME[0].ref, m_puAbsPartIdx, puIdx);
+ cu.m_mvd[0][m_puAbsPartIdx] = bestME[0].mv - bestME[0].mvp;
+ cu.m_mvpIdx[0][m_puAbsPartIdx] = bestME[0].mvpIdx;
cu.setPURefIdx(1, REF_NOT_VALID, m_puAbsPartIdx, puIdx);
cu.setPUMv(1, mvzero, m_puAbsPartIdx, puIdx);
- totalmebits += interMode.bestME[0].bits;
+ totalmebits += bestME[0].bits;
}
else
{
@@ -2295,15 +2298,15 @@
cu.m_mergeFlag[m_puAbsPartIdx] = false;
cu.setPUInterDir(2, m_puAbsPartIdx, puIdx);
- cu.setPUMv(1, interMode.bestME[1].mv, m_puAbsPartIdx, puIdx);
- cu.setPURefIdx(1, interMode.bestME[1].ref, m_puAbsPartIdx, puIdx);
- cu.m_mvd[1][m_puAbsPartIdx] = interMode.bestME[1].mv - interMode.bestME[1].mvp;
- cu.m_mvpIdx[1][m_puAbsPartIdx] = interMode.bestME[1].mvpIdx;
+ cu.setPUMv(1, bestME[1].mv, m_puAbsPartIdx, puIdx);
+ cu.setPURefIdx(1, bestME[1].ref, m_puAbsPartIdx, puIdx);
+ cu.m_mvd[1][m_puAbsPartIdx] = bestME[1].mv - bestME[1].mvp;
+ cu.m_mvpIdx[1][m_puAbsPartIdx] = bestME[1].mvpIdx;
cu.setPURefIdx(0, REF_NOT_VALID, m_puAbsPartIdx, puIdx);
cu.setPUMv(0, mvzero, m_puAbsPartIdx, puIdx);
- totalmebits += interMode.bestME[1].bits;
+ totalmebits += bestME[1].bits;
}
prepMotionCompensation(cu, cuGeom, puIdx);
diff -r 32513a4c3bd4 -r ae9d1d48303b source/encoder/search.h
--- a/source/encoder/search.h Mon Nov 10 12:39:54 2014 +0900
+++ b/source/encoder/search.h Tue Nov 11 11:27:02 2014 +0530
@@ -76,6 +76,8 @@
bool costZero;
};
+enum { MAX_INTER_PARTS = 2 };
+
struct Mode
{
CUData cu;
@@ -84,7 +86,7 @@
Yuv reconYuv;
Entropy contexts;
- MotionData bestME[2];
+ MotionData bestME[MAX_INTER_PARTS][2];
MV amvpCand[2][MAX_NUM_REF][AMVP_NUM_CANDS];
uint64_t rdCost; // sum of partition (psy) RD costs (sse(fenc, recon) + lambda2 * bits)
diff -r 32513a4c3bd4 -r ae9d1d48303b source/x265.h
--- a/source/x265.h Mon Nov 10 12:39:54 2014 +0900
+++ b/source/x265.h Tue Nov 11 11:27:02 2014 +0530
@@ -290,7 +290,7 @@
#define X265_ANALYSIS_OFF 0
#define X265_ANALYSIS_SAVE 1
#define X265_ANALYSIS_LOAD 2
-#define X265_MAX_PRED_MODE_PER_CU (85 * 8)
+#define X265_MAX_PRED_MODE_PER_CU (85 * 8 * 2) /* max-recursive-percu * max-predmode * numberofpart */
typedef struct
{
More information about the x265-devel
mailing list