[x265-commits] [x265] Improved sao implementation by limiting sao types
Ashok Kumar Mishra
ashok at multicorewareinc.com
Tue Apr 11 01:03:03 CEST 2017
details: http://hg.videolan.org/x265/rev/c7b7c736696f
branches:
changeset: 11735:c7b7c736696f
user: Ashok Kumar Mishra <ashok at multicorewareinc.com>
date: Mon Apr 10 11:42:20 2017 +0530
description:
Improved sao implementation by limiting sao types
diffstat:
doc/reST/cli.rst | 6 +-
source/CMakeLists.txt | 2 +-
source/common/param.cpp | 6 +-
source/encoder/encoder.cpp | 2 +-
source/encoder/sao.cpp | 217 ++++++++++++++++++++++++++-----------------
source/encoder/sao.h | 4 +-
source/test/smoke-tests.txt | 2 +-
source/x265.h | 27 ++---
source/x265cli.h | 3 +
9 files changed, 155 insertions(+), 114 deletions(-)
diffs (truncated from 524 to 300 lines):
diff -r 3c26e29120aa -r c7b7c736696f doc/reST/cli.rst
--- a/doc/reST/cli.rst Wed Feb 22 15:24:25 2017 +0530
+++ b/doc/reST/cli.rst Mon Apr 10 11:42:20 2017 +0530
@@ -1689,10 +1689,14 @@ Loop filters
When enabled, non-deblocked pixels are used for SAO analysis. When
disabled, SAO analysis skips the right/bottom boundary areas.
Default disabled
+.. option:: --limit-sao, --no-limit-sao
+ Limit SAO filter computation by early terminating SAO process based
+ on inter prediction mode, CTU spatial-domain correlations, and relations
+ between luma and chroma.
+ Default disabled
VUI (Video Usability Information) options
=========================================
-
x265 emits a VUI with only the timing info by default. If the SAR is
specified (or read from a Y4M header) it is also included. All other
VUI fields must be manually specified.
diff -r 3c26e29120aa -r c7b7c736696f source/CMakeLists.txt
--- a/source/CMakeLists.txt Wed Feb 22 15:24:25 2017 +0530
+++ b/source/CMakeLists.txt Mon Apr 10 11:42:20 2017 +0530
@@ -29,7 +29,7 @@ option(NATIVE_BUILD "Target the build CP
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 113)
+set(X265_BUILD 114)
configure_file("${PROJECT_SOURCE_DIR}/x265.def.in"
"${PROJECT_BINARY_DIR}/x265.def")
configure_file("${PROJECT_SOURCE_DIR}/x265_config.h.in"
diff -r 3c26e29120aa -r c7b7c736696f source/common/param.cpp
--- a/source/common/param.cpp Wed Feb 22 15:24:25 2017 +0530
+++ b/source/common/param.cpp Mon Apr 10 11:42:20 2017 +0530
@@ -187,7 +187,7 @@ void x265_param_default(x265_param* para
/* SAO Loop Filter */
param->bEnableSAO = 1;
param->bSaoNonDeblocked = 0;
-
+ param->bLimitSAO = 0;
/* Coding Quality */
param->cbQpOffset = 0;
param->crQpOffset = 0;
@@ -272,9 +272,7 @@ void x265_param_default(x265_param* para
param->bAQMotion = 0;
param->bHDROpt = 0;
param->analysisRefineLevel = 5;
-
}
-
int x265_param_default_preset(x265_param* param, const char* preset, const char* tune)
{
#if EXPORT_C_API
@@ -949,6 +947,7 @@ int x265_param_parse(x265_param* p, cons
}
OPT("hdr") p->bEmitHDRSEI = atobool(value);
OPT("hdr-opt") p->bHDROpt = atobool(value);
+ OPT("limit-sao") p->bLimitSAO = atobool(value);
else
return X265_PARAM_BAD_NAME;
}
@@ -1658,6 +1657,7 @@ char *x265_param2string(x265_param* p, i
BOOL(p->bEmitHDRSEI, "hdr");
BOOL(p->bHDROpt, "hdr-opt");
s += sprintf(s, " refine-level=%d", p->analysisRefineLevel);
+ BOOL(p->bLimitSAO, "limit-sao");
#undef BOOL
return buf;
}
diff -r 3c26e29120aa -r c7b7c736696f source/encoder/encoder.cpp
--- a/source/encoder/encoder.cpp Wed Feb 22 15:24:25 2017 +0530
+++ b/source/encoder/encoder.cpp Mon Apr 10 11:42:20 2017 +0530
@@ -2140,7 +2140,7 @@ void Encoder::configure(x265_param *p)
/* some options make no sense if others are disabled */
p->bSaoNonDeblocked &= p->bEnableSAO;
p->bEnableTSkipFast &= p->bEnableTransformSkip;
-
+ p->bLimitSAO &= p->bEnableSAO;
/* initialize the conformance window */
m_conformanceWindow.bEnabled = false;
m_conformanceWindow.rightOffset = 0;
diff -r 3c26e29120aa -r c7b7c736696f source/encoder/sao.cpp
--- a/source/encoder/sao.cpp Wed Feb 22 15:24:25 2017 +0530
+++ b/source/encoder/sao.cpp Mon Apr 10 11:42:20 2017 +0530
@@ -734,6 +734,7 @@ void SAO::generateChromaOffsets(SaoCtuPa
/* Calculate SAO statistics for current CTU without non-crossing slice */
void SAO::calcSaoStatsCTU(int addr, int plane)
{
+ Slice* slice = m_frame->m_encData->m_slice;
const PicYuv* reconPic = m_frame->m_reconPic;
const CUData* cu = m_frame->m_encData->getPicCTU(addr);
const pixel* fenc0 = m_frame->m_fencPic->getPlaneAddr(plane, addr);
@@ -857,60 +858,60 @@ void SAO::calcSaoStatsCTU(int addr, int
primitives.saoCuStatsE1(diff + startY * MAX_CU_SIZE, rec0 + startY * stride, stride, upBuff1, endX, endY - startY, m_offsetOrg[plane][SAO_EO_1], m_count[plane][SAO_EO_1]);
}
+ if (!m_param->bLimitSAO || ((slice->m_sliceType == P_SLICE && !cu->isSkipped(0)) ||
+ (slice->m_sliceType != B_SLICE)))
+ {
+ // SAO_EO_2: // dir: 135
+ {
+ if (m_param->bSaoNonDeblocked)
+ {
+ skipB = 4;
+ skipR = 5;
+ }
- // SAO_EO_2: // dir: 135
- {
- if (m_param->bSaoNonDeblocked)
+ fenc = fenc0;
+ rec = rec0;
+
+ startX = !lpelx;
+ endX = (rpelx == picWidth) ? ctuWidth - 1 : ctuWidth - skipR + plane_offset;
+
+ startY = bAboveUnavail;
+ endY = (bpely == picHeight) ? ctuHeight - 1 : ctuHeight - skipB + plane_offset;
+ if (startY)
+ {
+ fenc += stride;
+ rec += stride;
+ }
+
+ primitives.sign(upBuff1, &rec[startX], &rec[startX - stride - 1], (endX - startX));
+
+ primitives.saoCuStatsE2(diff + startX + startY * MAX_CU_SIZE, rec0 + startX + startY * stride, stride, upBuff1, upBufft, endX - startX, endY - startY, m_offsetOrg[plane][SAO_EO_2], m_count[plane][SAO_EO_2]);
+ }
+ // SAO_EO_3: // dir: 45
{
- skipB = 4;
- skipR = 5;
+ if (m_param->bSaoNonDeblocked)
+ {
+ skipB = 4;
+ skipR = 5;
+ }
+ fenc = fenc0;
+ rec = rec0;
+ startX = !lpelx;
+ endX = (rpelx == picWidth) ? ctuWidth - 1 : ctuWidth - skipR + plane_offset;
+
+ startY = bAboveUnavail;
+ endY = (bpely == picHeight) ? ctuHeight - 1 : ctuHeight - skipB + plane_offset;
+
+ if (startY)
+ {
+ fenc += stride;
+ rec += stride;
+ }
+
+ primitives.sign(upBuff1, &rec[startX - 1], &rec[startX - 1 - stride + 1], (endX - startX + 1));
+
+ primitives.saoCuStatsE3(diff + startX + startY * MAX_CU_SIZE, rec0 + startX + startY * stride, stride, upBuff1 + 1, endX - startX, endY - startY, m_offsetOrg[plane][SAO_EO_3], m_count[plane][SAO_EO_3]);
}
-
- fenc = fenc0;
- rec = rec0;
-
- startX = !lpelx;
- endX = (rpelx == picWidth) ? ctuWidth - 1 : ctuWidth - skipR + plane_offset;
-
- startY = bAboveUnavail;
- endY = (bpely == picHeight) ? ctuHeight - 1 : ctuHeight - skipB + plane_offset;
- if (startY)
- {
- fenc += stride;
- rec += stride;
- }
-
- primitives.sign(upBuff1, &rec[startX], &rec[startX - stride - 1], (endX - startX));
-
- primitives.saoCuStatsE2(diff + startX + startY * MAX_CU_SIZE, rec0 + startX + startY * stride, stride, upBuff1, upBufft, endX - startX, endY - startY, m_offsetOrg[plane][SAO_EO_2], m_count[plane][SAO_EO_2]);
- }
-
- // SAO_EO_3: // dir: 45
- {
- if (m_param->bSaoNonDeblocked)
- {
- skipB = 4;
- skipR = 5;
- }
-
- fenc = fenc0;
- rec = rec0;
-
- startX = !lpelx;
- endX = (rpelx == picWidth) ? ctuWidth - 1 : ctuWidth - skipR + plane_offset;
-
- startY = bAboveUnavail;
- endY = (bpely == picHeight) ? ctuHeight - 1 : ctuHeight - skipB + plane_offset;
-
- if (startY)
- {
- fenc += stride;
- rec += stride;
- }
-
- primitives.sign(upBuff1, &rec[startX - 1], &rec[startX - 1 - stride + 1], (endX - startX + 1));
-
- primitives.saoCuStatsE3(diff + startX + startY * MAX_CU_SIZE, rec0 + startX + startY * stride, stride, upBuff1 + 1, endX - startX, endY - startY, m_offsetOrg[plane][SAO_EO_3], m_count[plane][SAO_EO_3]);
}
}
}
@@ -1224,10 +1225,8 @@ void SAO::rdoSaoUnitRowEnd(const SAOPara
void SAO::rdoSaoUnitCu(SAOParam* saoParam, int rowBaseAddr, int idxX, int addr)
{
Slice* slice = m_frame->m_encData->m_slice;
-// int qp = slice->m_sliceQp;
const CUData* cu = m_frame->m_encData->getPicCTU(addr);
int qp = cu->m_qp[0];
-
int64_t lambda[2] = { 0 };
int qpCb = qp + slice->m_pps->chromaQpOffset[0] + slice->m_chromaQpOffset[0];
@@ -1262,18 +1261,6 @@ void SAO::rdoSaoUnitCu(SAOParam* saoPara
for (int i = 0; i < planes; i++)
saoParam->ctuParam[i][addr].reset();
-
- if (saoParam->bSaoFlag[0])
- calcSaoStatsCTU(addr, 0);
-
- if (saoParam->bSaoFlag[1])
- {
- calcSaoStatsCTU(addr, 1);
- calcSaoStatsCTU(addr, 2);
- }
-
- saoStatsInitialOffset(planes);
-
// SAO distortion calculation
m_entropyCoder.load(m_rdContexts.cur);
m_entropyCoder.resetBits();
@@ -1282,15 +1269,44 @@ void SAO::rdoSaoUnitCu(SAOParam* saoPara
if (allowMerge[1])
m_entropyCoder.codeSaoMerge(0);
m_entropyCoder.store(m_rdContexts.temp);
+ memset(m_offset, 0, sizeof(m_offset));
+ int64_t bestCost = 0;
+ int64_t rateDist = 0;
+
+ bool bAboveLeftAvail = true;
+ for (int mergeIdx = 0; mergeIdx < 2; ++mergeIdx)
+ {
+ if (!allowMerge[mergeIdx])
+ continue;
+
+ SaoCtuParam* mergeSrcParam = &(saoParam->ctuParam[0][addrMerge[mergeIdx]]);
+ bAboveLeftAvail = bAboveLeftAvail && (mergeSrcParam->typeIdx == -1);
+ }
+ // Don't apply sao if ctu is skipped or ajacent ctus are sao off
+ bool bSaoOff = (slice->m_sliceType == B_SLICE) && (cu->isSkipped(0) || bAboveLeftAvail);
// Estimate distortion and cost of new SAO params
- int64_t bestCost = 0;
- int64_t rateDist = 0;
- // Estimate distortion and cost of new SAO params
- saoLumaComponentParamDist(saoParam, addr, rateDist, lambda, bestCost);
- if (chroma)
- saoChromaComponentParamDist(saoParam, addr, rateDist, lambda, bestCost);
+ if (saoParam->bSaoFlag[0])
+ {
+ if (!m_param->bLimitSAO || !bSaoOff)
+ {
+ calcSaoStatsCTU(addr, 0);
+ saoStatsInitialOffset(addr, 0);
+ saoLumaComponentParamDist(saoParam, addr, rateDist, lambda, bestCost);
+ }
+ }
+ SaoCtuParam* lclCtuParam = &saoParam->ctuParam[0][addr];
+ if (saoParam->bSaoFlag[1])
+ {
+ if (!m_param->bLimitSAO || ((lclCtuParam->typeIdx != -1) && !bSaoOff))
+ {
+ calcSaoStatsCTU(addr, 1);
+ calcSaoStatsCTU(addr, 2);
+ saoStatsInitialOffset(addr, 1);
+ saoChromaComponentParamDist(saoParam, addr, rateDist, lambda, bestCost);
+ }
+ }
if (saoParam->bSaoFlag[0] || saoParam->bSaoFlag[1])
{
// Cost of merge left or Up
@@ -1357,17 +1373,27 @@ void SAO::rdoSaoUnitCu(SAOParam* saoPara
}
}
-
// Rounds the division of initial offsets by the number of samples in
// each of the statistics table entries.
-void SAO::saoStatsInitialOffset(int planes)
+void SAO::saoStatsInitialOffset(int addr, int planes)
{
- memset(m_offset, 0, sizeof(m_offset));
+ Slice* slice = m_frame->m_encData->m_slice;
+ const CUData* cu = m_frame->m_encData->getPicCTU(addr);
+ int maxSaoType;
More information about the x265-commits
mailing list