Re: [x265] [PATCH 2 of 3] rc: accumulate mv bits, coeff bits per frame
Thanks, can you merge with the latest tip?
On Fri, Jun 27, 2014 at 12:13 AM, Aarthi Priya Thirumalai
aar...@multicorewareinc.com wrote:
# HG changeset patch
# User Aarthi Thirumalaiaar...@multicorewareinc.com
# Date 1403808050 -19800
# Fri Jun 27 00:10:50 2014 +0530
# Node ID 11ddf73017d44933090a8943f4fc5098b231b56d
# Parent 1b669c33ff3a8d8f6c9bd1e18979c009baed2433
rc: accumulate mv bits, coeff bits per frame
diff -r 1b669c33ff3a -r 11ddf73017d4 source/Lib/TLibEncoder/TEncCu.cpp
--- a/source/Lib/TLibEncoder/TEncCu.cpp Wed Jun 25 22:46:45 2014 +0530
+++ b/source/Lib/TLibEncoder/TEncCu.cpp Fri Jun 27 00:10:50 2014 +0530
@@ -1365,12 +1365,14 @@
m_entropyCoder-encodePredMode(outTempCU, 0);
m_entropyCoder-encodePartSize(outTempCU, 0, depth);
m_entropyCoder-encodePredInfo(outTempCU, 0);
+outTempCU-m_mvBits = m_entropyCoder-getNumberOfWrittenBits();
// Encode Coefficients
bool bCodeDQP = getdQPFlag();
m_entropyCoder-encodeCoeff(outTempCU, 0, depth,
outTempCU-getCUSize(0), bCodeDQP);
m_rdGoOnSbacCoder-store(m_rdSbacCoders[depth][CI_TEMP_BEST]);
outTempCU-m_totalBits = m_entropyCoder-getNumberOfWrittenBits();
+outTempCU-m_coeffBits = outTempCU-m_totalBits - outTempCU-m_mvBits;
if (m_rdCost-psyRdEnabled())
{
@@ -1411,12 +1413,14 @@
m_entropyCoder-encodePredMode(outTempCU, 0);
m_entropyCoder-encodePartSize(outTempCU, 0, depth);
m_entropyCoder-encodePredInfo(outTempCU, 0);
+outTempCU-m_mvBits = m_entropyCoder-getNumberOfWrittenBits();
// Encode Coefficients
bool bCodeDQP = getdQPFlag();
m_entropyCoder-encodeCoeff(outTempCU, 0, depth,
outTempCU-getCUSize(0), bCodeDQP);
m_rdGoOnSbacCoder-store(m_rdSbacCoders[depth][CI_TEMP_BEST]);
outTempCU-m_totalBits = m_entropyCoder-getNumberOfWrittenBits();
+outTempCU-m_coeffBits = outTempCU-m_totalBits - outTempCU-m_mvBits;
if (m_rdCost-psyRdEnabled())
{
diff -r 1b669c33ff3a -r 11ddf73017d4 source/Lib/TLibEncoder/TEncSearch.cpp
--- a/source/Lib/TLibEncoder/TEncSearch.cpp Wed Jun 25 22:46:45 2014
+0530
+++ b/source/Lib/TLibEncoder/TEncSearch.cpp Fri Jun 27 00:10:50 2014 +0530
@@ -4059,6 +4059,7 @@
}
m_entropyCoder-encodeSkipFlag(cu, 0);
m_entropyCoder-encodeMergeIndex(cu, 0);
+cu-m_mvBits = m_entropyCoder-getNumberOfWrittenBits();
return m_entropyCoder-getNumberOfWrittenBits();
}
else
@@ -4073,8 +4074,11 @@
m_entropyCoder-encodePartSize(cu, 0, cu-getDepth(0));
m_entropyCoder-encodePredInfo(cu, 0);
bool bDummy = false;
+cu-m_mvBits = m_entropyCoder-getNumberOfWrittenBits();
m_entropyCoder-encodeCoeff(cu, 0, cu-getDepth(0),
cu-getCUSize(0), bDummy);
-return m_entropyCoder-getNumberOfWrittenBits();
+int totalBits = m_entropyCoder-getNumberOfWrittenBits();
+cu-m_coeffBits = totalBits - cu-m_mvBits;
+return totalBits;
}
}
diff -r 1b669c33ff3a -r 11ddf73017d4 source/encoder/compress.cpp
--- a/source/encoder/compress.cpp Wed Jun 25 22:46:45 2014 +0530
+++ b/source/encoder/compress.cpp Fri Jun 27 00:10:50 2014 +0530
@@ -63,6 +63,7 @@
m_entropyCoder-encodePredMode(cu, 0);
m_entropyCoder-encodePartSize(cu, 0, depth);
m_entropyCoder-encodePredInfo(cu, 0);
+cu-m_mvBits += m_entropyCoder-getNumberOfWrittenBits();
// Encode Coefficients
bool bCodeDQP = getdQPFlag();
@@ -71,6 +72,7 @@
m_rdGoOnSbacCoder-store(m_rdSbacCoders[depth][CI_TEMP_BEST]);
cu-m_totalBits = m_entropyCoder-getNumberOfWrittenBits();
+cu-m_coeffBits = cu-m_totalBits - cu-m_mvBits;
if (m_rdCost-psyRdEnabled())
{
int part = g_convertToBit[cu-getCUSize(0)];
diff -r 1b669c33ff3a -r 11ddf73017d4 source/encoder/frameencoder.cpp
--- a/source/encoder/frameencoder.cpp Wed Jun 25 22:46:45 2014 +0530
+++ b/source/encoder/frameencoder.cpp Fri Jun 27 00:10:50 2014 +0530
@@ -694,6 +694,11 @@
// Store probabilities of second LCU in line into buffer
if (col == 1 m_param-bEnableWavefront)
getBufferSBac(lin)-loadContexts(getSbacCoder(subStrm));
+
+// Collect Frame Stats for 2 pass
+m_frame-m_stats.mvBits += cu-m_mvBits;
+m_frame-m_stats.coeffBits += cu-m_coeffBits;
+m_frame-m_stats.miscBits += cu-m_totalBits - (cu-m_mvBits +
cu-m_coeffBits);
}
if (slice-getPPS()-getCabacInitPresentFlag())
On Thu, Jun 26, 2014 at 4:25 PM, Deepthi Nandakumar
deep...@multicorewareinc.com wrote:
Pls fix extra newlines and whitespace nits.
On Wed, Jun 25, 2014 at 10:54 PM, aar...@multicorewareinc.com wrote:
# HG changeset patch
# User Aarthi Thirumalaiaar...@multicorewareinc.com
# Date 1403716735 -19800
# Wed Jun 25 22:48:55 2014 +0530
# Node ID 0995efabd44470c1192994e1aceeb40ae606467f
# Parent e71e34d02de228eab43edf1910a71a44417d
rc:
[x265] [PATCH] psyrd: fix for inconsistent output
# HG changeset patch
# User Sumalatha Polureddysumala...@multicorewareinc.com
# Date 1403858781 -19800
# Node ID a789870889fcc9a31deff7fc6961d143b0db86c1
# Parent 1b669c33ff3a8d8f6c9bd1e18979c009baed2433
psyrd: fix for inconsistent output
maximum buffer size for zeropel is MAX_CU_SIZExMAX_CU_SIZE. since stride was
wrong,
it was accessing out of boundary memory which was different for each run, so
inconsistent output
diff -r 1b669c33ff3a -r a789870889fc source/Lib/TLibEncoder/TEncSearch.cpp
--- a/source/Lib/TLibEncoder/TEncSearch.cpp Wed Jun 25 22:46:45 2014 +0530
+++ b/source/Lib/TLibEncoder/TEncSearch.cpp Fri Jun 27 14:16:21 2014 +0530
@@ -3035,7 +3035,7 @@
{
int size = g_convertToBit[trSize];
psyEnergyY = m_rdCost-psyCost(size,
fencYuv-getLumaAddr(absPartIdx), fencYuv-getStride(),
- (pixel*)RDCost::zeroPel,
cu-getPic()-getPicYuvRec()-getStride()); // need to check whether zero
distortion is similar to psyenergy of fenc
+(pixel*)RDCost::zeroPel, trSize); // need to check whether
zero distortion is similar to psyenergy of fenc
}
int16_t *curResiY = m_qtTempShortYuv[qtLayer].getLumaAddr(absPartIdx);
X265_CHECK(m_qtTempShortYuv[qtLayer].m_width == MAX_CU_SIZE, width
not full CU\n);
___
x265-devel mailing list
x265-devel@videolan.org
https://mailman.videolan.org/listinfo/x265-devel
[x265] fix emms: move selectLambda() into xRateDistOptQuant() and issue emms before it
# HG changeset patch
# User Satoshi Nakagawa nakagawa...@oki.com
# Date 1403877807 -32400
# Fri Jun 27 23:03:27 2014 +0900
# Node ID 77f443fe169ca23969df5d5ee6968543bfa5e794
# Parent 32aa6cc3cf4d108ac92f5d29258b2c38ca888d29
fix emms: move selectLambda() into xRateDistOptQuant() and issue emms before it
diff -r 32aa6cc3cf4d -r 77f443fe169c source/Lib/TLibCommon/TComTrQuant.cpp
--- a/source/Lib/TLibCommon/TComTrQuant.cpp Thu Jun 26 17:19:08 2014 -0700
+++ b/source/Lib/TLibCommon/TComTrQuant.cpp Fri Jun 27 23:03:27 2014 +0900
@@ -508,6 +508,9 @@
uint32_t TComTrQuant::xRateDistOptQuant(TComDataCU* cu, int32_t* srcCoeff,
coeff_t* dstCoeff, uint32_t trSize,
TextType ttype, uint32_t absPartIdx,
int32_t *lastPos)
{
+x265_emms();
+selectLambda(ttype);
+
const uint32_t log2TrSize = g_convertToBit[trSize] + 2;
uint32_t absSum = 0;
int transformShift = MAX_TR_DYNAMIC_RANGE - X265_DEPTH - log2TrSize; //
Represents scaling through forward transform
diff -r 32aa6cc3cf4d -r 77f443fe169c source/Lib/TLibEncoder/TEncSearch.cpp
--- a/source/Lib/TLibEncoder/TEncSearch.cpp Thu Jun 26 17:19:08 2014 -0700
+++ b/source/Lib/TLibEncoder/TEncSearch.cpp Fri Jun 27 23:03:27 2014 +0900
@@ -428,8 +428,6 @@
int chFmt = cu-getChromaFormat();
m_trQuant-setQPforQuant(cu-getQP(0), TEXT_LUMA, QP_BD_OFFSET, 0, chFmt);
-m_trQuant-selectLambda(TEXT_LUMA);
-
absSum = m_trQuant-transformNxN(cu, residual, stride, coeff, tuSize,
TEXT_LUMA, absPartIdx, lastPos, useTransformSkip);
//--- set coded block flag ---
@@ -515,8 +513,6 @@
curChromaQpOffset = cu-getSlice()-getPPS()-getChromaCrQpOffset() +
cu-getSlice()-getSliceQpDeltaCr();
}
m_trQuant-setQPforQuant(cu-getQP(0), TEXT_CHROMA,
cu-getSlice()-getSPS()-getQpBDOffsetC(), curChromaQpOffset, chFmt);
-m_trQuant-selectLambda(ttype);
-
absSum = m_trQuant-transformNxN(cu, residual, stride, coeff, tuSize,
ttype, absPartIdx, lastPos, useTransformSkipC);
//--- set coded block flag ---
@@ -905,7 +901,6 @@
int lastPos = -1;
m_trQuant-setQPforQuant(cu-getQP(0), TEXT_LUMA, QP_BD_OFFSET, 0,
chFmt);
-m_trQuant-selectLambda(TEXT_LUMA);
absSum = m_trQuant-transformNxN(cu, residual, stride, coeff, tuSize,
TEXT_LUMA, absPartIdx, lastPos, useTransformSkip);
//--- set coded block flag ---
@@ -1421,8 +1416,6 @@
curChromaQpOffset =
cu-getSlice()-getPPS()-getChromaCrQpOffset() +
cu-getSlice()-getSliceQpDeltaCr();
}
m_trQuant-setQPforQuant(cu-getQP(0), TEXT_CHROMA,
cu-getSlice()-getSPS()-getQpBDOffsetC(), curChromaQpOffset, chFmt);
-m_trQuant-selectLambda(ttype);
-
absSum = m_trQuant-transformNxN(cu, residual, stride, coeff,
tuSize, ttype, absPartIdxC, lastPos, useTransformSkipC);
//--- set coded block flag ---
@@ -2702,13 +2695,11 @@
cu-setTransformSkipSubParts(0, TEXT_LUMA, absPartIdx, depth);
-m_trQuant-setQPforQuant(cu-getQP(0), TEXT_LUMA, QP_BD_OFFSET, 0,
chFmt);
-m_trQuant-selectLambda(TEXT_LUMA);
-
int16_t *curResiY = resiYuv-getLumaAddr(absPartIdx);
const uint32_t strideResiY = resiYuv-m_width;
const uint32_t strideResiC = resiYuv-m_cwidth;
+m_trQuant-setQPforQuant(cu-getQP(0), TEXT_LUMA, QP_BD_OFFSET, 0,
chFmt);
absSumY = m_trQuant-transformNxN(cu, curResiY, strideResiY, coeffCurY,
trSize, TEXT_LUMA, absPartIdx,
lastPosY, false, curuseRDOQ);
@@ -2746,13 +2737,11 @@
int curChromaQpOffset =
cu-getSlice()-getPPS()-getChromaCbQpOffset() +
cu-getSlice()-getSliceQpDeltaCb();
m_trQuant-setQPforQuant(cu-getQP(0), TEXT_CHROMA,
cu-getSlice()-getSPS()-getQpBDOffsetC(), curChromaQpOffset, chFmt);
-m_trQuant-selectLambda(TEXT_CHROMA_U);
absSumU = m_trQuant-transformNxN(cu, curResiU, strideResiC,
coeffCurU + subTUBufferOffset,
trSizeC, TEXT_CHROMA_U,
absPartIdxC, lastPosU, false, curuseRDOQ);
curChromaQpOffset =
cu-getSlice()-getPPS()-getChromaCrQpOffset() +
cu-getSlice()-getSliceQpDeltaCr();
m_trQuant-setQPforQuant(cu-getQP(0), TEXT_CHROMA,
cu-getSlice()-getSPS()-getQpBDOffsetC(), curChromaQpOffset, chFmt);
-m_trQuant-selectLambda(TEXT_CHROMA_V);
absSumV = m_trQuant-transformNxN(cu, curResiV, strideResiC,
coeffCurV + subTUBufferOffset,
trSizeC, TEXT_CHROMA_V,
absPartIdxC, lastPosV, false, curuseRDOQ);
@@ -2915,8 +2904,6 @@
}
m_trQuant-setQPforQuant(cu-getQP(0), TEXT_LUMA, QP_BD_OFFSET, 0,
chFmt);
-m_trQuant-selectLambda(TEXT_LUMA);
-
absSum[TEXT_LUMA][0] = m_trQuant-transformNxN(cu,
Re: [x265] [PATCH 1 of 2] improve count_nonzero by SSSE3
On 6/27/2014 4:05 PM, chen wrote: I can't understand what's your means. could you tell me more? I use some SSSE3 instruction and process 16 pixels every loop. I meant keep both sse2 and ssse3 variants. Not sure if x86inc.asm macros help with this or not. - Derek ___ x265-devel mailing list x265-devel@videolan.org https://mailman.videolan.org/listinfo/x265-devel
Re: [x265] [PATCH 1 of 2] improve count_nonzero by SSSE3
On 6/27/2014 6:08 PM, chen wrote: I use ssse3 instruction PSHUFB to replace 3 SSE2 instructions, the x86inc macro can't handle it. After patch, this function is faster ~20% and codeCoeffNxN ~7% speedup, so I don't worry about old CPU's performance. I guess SSSE3 is very prevalent nowadays -- though I am still not a fan of throwing away variants, I guess it's reasonable in this case. - Derek ___ x265-devel mailing list x265-devel@videolan.org https://mailman.videolan.org/listinfo/x265-devel
