[x265-commits] [x265] Generating sum & ssd values for weightp decision in looka...
Shazeb Nawaz Khan
shazeb at multicorewareinc.com
Mon Nov 4 07:21:21 CET 2013
details: http://hg.videolan.org/x265/rev/2ab39c2dd50f
branches:
changeset: 4822:2ab39c2dd50f
user: Shazeb Nawaz Khan <shazeb at multicorewareinc.com>
date: Mon Nov 04 11:40:49 2013 +0530
description:
Generating sum & ssd values for weightp decision in lookahead
diffstat:
source/common/lowres.h | 8 ++++--
source/encoder/encoder.cpp | 2 +-
source/encoder/ratecontrol.cpp | 46 ++++++++++++++++++++++++++---------------
3 files changed, 35 insertions(+), 21 deletions(-)
diffs (133 lines):
diff -r 8621008756ba -r 2ab39c2dd50f source/common/lowres.h
--- a/source/common/lowres.h Sat Nov 02 10:19:22 2013 -0500
+++ b/source/common/lowres.h Mon Nov 04 11:40:49 2013 +0530
@@ -42,6 +42,8 @@ struct Lowres : public ReferencePlanes
int frameNum; // Presentation frame number
int sliceType; // Slice type decided by lookahead
int leadingBframes; // number of leading B frames for P or I
+ uint64_t m_wp_ssd[3]; // This is different than m_SSDY, this is sum(pixel^2) - sum(pixel)^2 for entire frame
+ uint64_t m_wp_sum[3];
bool bIntraCalculated;
bool bScenecut; // Set to false if the frame cannot possibly be part of a real scenecut.
@@ -51,12 +53,12 @@ struct Lowres : public ReferencePlanes
/* lookahead output data */
int costEst[X265_BFRAME_MAX + 2][X265_BFRAME_MAX + 2];
int costEstAq[X265_BFRAME_MAX + 2][X265_BFRAME_MAX + 2];
- int32_t *rowSatds[X265_BFRAME_MAX + 2][X265_BFRAME_MAX + 2];
+ int32_t* rowSatds[X265_BFRAME_MAX + 2][X265_BFRAME_MAX + 2];
int intraMbs[X265_BFRAME_MAX + 2];
- int32_t *intraCost;
+ int32_t* intraCost;
int satdCost;
uint16_t(*lowresCosts[X265_BFRAME_MAX + 2][X265_BFRAME_MAX + 2]);
- int32_t *lowresMvCosts[2][X265_BFRAME_MAX + 1];
+ int32_t *lowresMvCosts[2][X265_BFRAME_MAX + 1];
MV *lowresMvs[2][X265_BFRAME_MAX + 1];
void create(TComPic *pic, int bframes, int32_t *aqMode);
diff -r 8621008756ba -r 2ab39c2dd50f source/encoder/encoder.cpp
--- a/source/encoder/encoder.cpp Sat Nov 02 10:19:22 2013 -0500
+++ b/source/encoder/encoder.cpp Mon Nov 04 11:40:49 2013 +0530
@@ -217,7 +217,7 @@ int Encoder::encode(bool flush, const x2
// Encoder holds a reference count until collecting stats
ATOMIC_INC(&pic->m_countRefEncoders);
- if (param.rc.aqMode)
+ if (param.rc.aqMode || param.bEnableWeightedPred)
m_rateControl->calcAdaptiveQuantFrame(pic);
m_lookahead->addPicture(pic, pic_in->sliceType);
}
diff -r 8621008756ba -r 2ab39c2dd50f source/encoder/ratecontrol.cpp
--- a/source/encoder/ratecontrol.cpp Sat Nov 02 10:19:22 2013 -0500
+++ b/source/encoder/ratecontrol.cpp Mon Nov 04 11:40:49 2013 +0530
@@ -51,25 +51,26 @@ static inline double qp2qScale(double qp
}
/* Compute variance to derive AC energy of each block */
-static inline uint32_t acEnergyVar(uint64_t sum_ssd, int shift)
+static inline uint32_t acEnergyVar(TComPic *pic, uint64_t sum_ssd, int shift, int i)
{
uint32_t sum = (uint32_t)sum_ssd;
uint32_t ssd = (uint32_t)(sum_ssd >> 32);
-
+ pic->m_lowres.m_wp_sum[i] += sum;
+ pic->m_lowres.m_wp_ssd[i] += ssd;
return ssd - ((uint64_t)sum * sum >> shift);
}
/* Find the energy of each block in Y/Cb/Cr plane */
-static inline uint32_t acEnergyPlane(pixel* src, int srcStride, int bChroma)
+static inline uint32_t acEnergyPlane(TComPic *pic, pixel* src, int srcStride, int bChroma)
{
if (bChroma)
{
ALIGN_VAR_8(pixel, pix[8 * 8]);
primitives.blockcpy_pp(8, 8, pix, 8, src, srcStride);
- return acEnergyVar(primitives.var[LUMA_8x8](pix, 8), 6);
+ return acEnergyVar(pic, primitives.var[LUMA_8x8](pix, 8), 6, bChroma);
}
else
- return acEnergyVar(primitives.var[LUMA_16x16](src, srcStride), 8);
+ return acEnergyVar(pic, primitives.var[LUMA_16x16](src, srcStride), 8, bChroma);
}
/* Find the total AC energy of each block in all planes */
@@ -81,9 +82,9 @@ double RateControl::acEnergyCu(TComPic*
uint32_t blockOffsetChroma = (block_x >> 1) + ((block_y >> 1) * cStride);
uint32_t var;
- var = acEnergyPlane(pic->getPicYuvOrg()->getLumaAddr() + blockOffsetLuma, stride, 0);
- var += acEnergyPlane(pic->getPicYuvOrg()->getCbAddr() + blockOffsetChroma, cStride, 1);
- var += acEnergyPlane(pic->getPicYuvOrg()->getCrAddr() + blockOffsetChroma, cStride, 1);
+ var = acEnergyPlane(pic, pic->getPicYuvOrg()->getLumaAddr() + blockOffsetLuma, stride, 0);
+ var += acEnergyPlane(pic, pic->getPicYuvOrg()->getCbAddr() + blockOffsetChroma, cStride, 1);
+ var += acEnergyPlane(pic, pic->getPicYuvOrg()->getCrAddr() + blockOffsetChroma, cStride, 2);
x265_emms();
double strength = cfg->param.rc.aqStrength * 1.0397f;
return strength * (X265_LOG2(X265_MAX(var, 1)) - 14.427f);
@@ -92,24 +93,35 @@ double RateControl::acEnergyCu(TComPic*
void RateControl::calcAdaptiveQuantFrame(TComPic *pic)
{
/* Actual adaptive quantization */
- if (cfg->param.rc.aqMode)
+ int maxCol = pic->getPicYuvOrg()->getWidth();
+ int maxRow = pic->getPicYuvOrg()->getHeight();
+
+ /* Calculate Qp offset for each 16x16 block in the frame */
+ int block_xy = 0;
+ int block_x = 0, block_y = 0;
+ for (block_y = 0; block_y < maxRow; block_y += 16)
{
- int maxCol = pic->getPicYuvOrg()->getWidth();
- int maxRow = pic->getPicYuvOrg()->getHeight();
-
- /* Calculate Qp offset for each 16x16 block in the frame */
- int block_xy = 0;
- for (int block_y = 0; block_y < maxRow; block_y += 16)
+ for (block_x = 0; block_x < maxCol; block_x += 16)
{
- for (int block_x = 0; block_x < maxCol; block_x += 16)
+ double qp_adj = acEnergyCu(pic, block_x, block_y);
+ if (cfg->param.rc.aqMode)
{
- double qp_adj = acEnergyCu(pic, block_x, block_y);
pic->m_lowres.m_qpAqOffset[block_xy] = qp_adj;
pic->m_lowres.m_invQscaleFactor[block_xy] = x265_exp2fix8(qp_adj);
block_xy++;
}
}
}
+ if (cfg->param.bEnableWeightedPred)
+ {
+ for(int i=0; i < 3; i++)
+ {
+ UInt64 sum, ssd;
+ sum = pic->m_lowres.m_wp_sum[i];
+ ssd = pic->m_lowres.m_wp_ssd[i];
+ pic->m_lowres.m_wp_ssd[i] = ssd - (sum*sum + (block_x * block_y) / 2 ) / (block_x * block_y);
+ }
+ }
}
RateControl::RateControl(TEncCfg * _cfg)
More information about the x265-commits
mailing list