As inter cu use 4 sub-cu analysis data to get the thresholds for rect and amp
pu, if the sub-cu skip some or one part(s) of 1/4 cu, rect and amp cu cannot
get part of sub-cu 2Nx2N mv cost, and this will break the thresholds
calculation for rect and amp pu in current depth.
Do you have any other good solutions for this problem?
Thanks!
---Original---
From: "Pradeep Ramachandran"<[email protected]>
Date: 2017/8/18 18:49:51
To: "Development for x265"<[email protected]>;
Subject: Re: [x265] [PATCH] intra: skip RD analysis when sum of
subCUsplitcostbigger than non-split cost
Pushed to default branch. I agree that this looks like a bitexact change, and
gives an nice perf boost.
Can this also be extended to inter analysis as the same logic should work there
too, and we don't have an early out there?
Thanks,
Pradeep.
On Sat, Aug 12, 2017 at 11:13 PM, Tom Vaughan
<[email protected]> wrote:
Thanks for this additional explanation, and thanks again for your contribution!
From: x265-devel [mailto:[email protected]] On Behalf Of Ximing
Cheng
Sent: Friday, August 11, 2017 12:32 PM
To: Ximing Cheng
Subject: Re: [x265] [PATCH] intra: skip RD analysis when sum of sub
CUsplitcostbigger than non-split cost
In fact, this skip is not a fast skip algorithm.
As the sum of split cost is larger than none split CU's best cost (both rdcost
of sub-cu and none split CU are without split flag cost), which means splitting
into 4 parts at this depth of cu is a worse case compared with none split CU.
So that, the remain N * 1/4 parts of CU analysis is useless.
....................
. A . B .
. . .
....................
. C . D .
. . .
.................... (A B C D is the 4 parts of a CU)
If sum of sub CU split cost(A_Cost + B_Cost) larger than non-split
cost(NSCost), assume NSCost < A_Cost + B_Cost, the remain parts (C, D)
continue to analysis rd.
C_Cost + D_Cost >= 0 --->
NSCost < A_Cost + B_Cost + C_Cost + D_Cost ---> (likely that)
NSCost + splitCost(splitflag = 0) < A_Cost + B_Cost + C_Cost + D_Cost +
splitCost(splitflag = 1) ---> choose none split
So, C and D rd analysis can be skipped.
So in my test cases, the MD5 checksum of the output bitstream is the same with
the original after this skip.
------------------ Original ------------------
From: "Ximing Cheng";<[email protected]>;
Send time: Friday, Aug 4, 2017 1:56 AM
To: "x265-devel"<[email protected]>;
Subject: [x265] [PATCH] intra: skip RD analysis when sum of sub
CUsplitcostbigger than non-split cost
# HG changeset patch
# User Ximing Cheng <[email protected]>
# Date 1501782508 -28800
# Fri Aug 04 01:48:28 2017 +0800
# Node ID 5943a1f73d5814a3a723f814a4dd0635b1fe2b35
# Parent d11482e5fedbcdaf62ee3c6872f43827d99ad181
intra: skip RD analysis when sum of sub CUsplitcost bigger than non-split cost
diff -r d11482e5fedb -r 5943a1f73d58 source/CMakeLists.txt
--- a/source/CMakeLists.txt Mon Jul 24 11:15:38 2017 +0530
+++ b/source/CMakeLists.txt Fri Aug 04 01:48:28 2017 +0800
@@ -29,7 +29,7 @@
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 131)
+set(X265_BUILD 132)
configure_file("${PROJECT_SOURCE_DIR}/x265.def.in"
"${PROJECT_BINARY_DIR}/x265.def")
configure_file("${PROJECT_SOURCE_DIR}/x265_config.h.in"
diff -r d11482e5fedb -r 5943a1f73d58 source/common/param.cpp
--- a/source/common/param.cpp Mon Jul 24 11:15:38 2017 +0530
+++ b/source/common/param.cpp Fri Aug 04 01:48:28 2017 +0800
@@ -157,6 +157,7 @@
param->bEnableConstrainedIntra = 0;
param->bEnableStrongIntraSmoothing = 1;
param->bEnableFastIntra = 0;
+ param->bEnableSplitRdSkip = 0;
/* Inter Coding tools */
param->searchMethod = X265_HEX_SEARCH;
@@ -975,6 +976,7 @@
OPT("refine-inter")p->interRefine = atobool(value);
OPT("refine-mv")p->mvRefine = atobool(value);
OPT("force-flush")p->forceFlush = atoi(value);
+ OPT("splitrd-skip") p->bEnableSplitRdSkip = atobool(value);
else
return X265_PARAM_BAD_NAME;
}
@@ -1431,6 +1433,7 @@
TOOLOPT(param->bEnableRdRefine, "rd-refine");
TOOLOPT(param->bEnableEarlySkip, "early-skip");
TOOLOPT(param->bEnableRecursionSkip, "rskip");
+ TOOLOPT(param->bEnableSplitRdSkip, "splitrd-skip");
TOOLVAL(param->noiseReductionIntra, "nr-intra=%d");
TOOLVAL(param->noiseReductionInter, "nr-inter=%d");
TOOLOPT(param->bEnableTSkipFast, "tskip-fast");
@@ -1560,6 +1563,7 @@
BOOL(p->bEnableTSkipFast, "tskip-fast");
BOOL(p->bCULossless, "cu-lossless");
BOOL(p->bIntraInBFrames, "b-intra");
+ BOOL(p->bEnableSplitRdSkip, "splitrd-skip");
s += sprintf(s, " rdpenalty=%d", p->rdPenalty);
s += sprintf(s, " psy-rd=%.2f", p->psyRd);
s += sprintf(s, " psy-rdoq=%.2f", p->psyRdoq);
diff -r d11482e5fedb -r 5943a1f73d58 source/encoder/analysis.cpp
--- a/source/encoder/analysis.cpp Mon Jul 24 11:15:38 2017 +0530
+++ b/source/encoder/analysis.cpp Fri Aug 04 01:48:28 2017 +0800
@@ -485,7 +485,7 @@
md.bestMode->reconYuv.copyToPicYuv(*m_frame->m_reconPic,
parentCTU.m_cuAddr, cuGeom.absPartIdx);
}
-void Analysis::compressIntraCU(const CUData& parentCTU, const CUGeom& cuGeom,
int32_t qp)
+uint64_t Analysis::compressIntraCU(const CUData& parentCTU, const CUGeom&
cuGeom, int32_t qp)
{
uint32_t depth = cuGeom.depth;
ModeDepth& md = m_modeDepth[depth];
@@ -560,6 +560,8 @@
invalidateContexts(nextDepth);
Entropy* nextContext = &m_rqt[depth].cur;
int32_t nextQP = qp;
+ uint64_t curCost = 0;
+ int skipSplitCheck = 0;
for (uint32_t subPartIdx = 0; subPartIdx < 4; subPartIdx++)
{
@@ -572,7 +574,17 @@
if (m_slice->m_pps->bUseDQP && nextDepth <=
m_slice->m_pps->maxCuDQPDepth)
nextQP = setLambdaFromQP(parentCTU,
calculateQpforCuSize(parentCTU, childGeom));
- compressIntraCU(parentCTU, childGeom, nextQP);
+ if (m_param->bEnableSplitRdSkip)
+ {
+ curCost += compressIntraCU(parentCTU, childGeom, nextQP);
+ if (m_modeDepth[depth].bestMode && curCost >
m_modeDepth[depth].bestMode->rdCost)
+ {
+ skipSplitCheck = 1;
+ break;
+ }
+ }
+ else
+ compressIntraCU(parentCTU, childGeom, nextQP);
// Save best CU and pred data for this sub CU
splitCU->copyPartFrom(nd.bestMode->cu, childGeom, subPartIdx);
@@ -590,14 +602,17 @@
memset(parentCTU.m_cuDepth + childGeom.absPartIdx, 0,
childGeom.numPartitions);
}
}
- nextContext->store(splitPred->contexts);
- if (mightNotSplit)
- addSplitFlagCost(*splitPred, cuGeom.depth);
- else
- updateModeCost(*splitPred);
-
- checkDQPForSplitPred(*splitPred, cuGeom);
- checkBestMode(*splitPred, depth);
+ if (!skipSplitCheck)
+ {
+ nextContext->store(splitPred->contexts);
+ if (mightNotSplit)
+ addSplitFlagCost(*splitPred, cuGeom.depth);
+ else
+ updateModeCost(*splitPred);
+
+ checkDQPForSplitPred(*splitPred, cuGeom);
+ checkBestMode(*splitPred, depth);
+ }
}
if (m_param->bEnableRdRefine && depth <= m_slice->m_pps->maxCuDQPDepth)
@@ -620,6 +635,8 @@
md.bestMode->cu.copyToPic(depth);
if (md.bestMode != &md.pred[PRED_SPLIT])
md.bestMode->reconYuv.copyToPicYuv(*m_frame->m_reconPic,
parentCTU.m_cuAddr, cuGeom.absPartIdx);
+
+ return md.bestMode->rdCost;
}
void Analysis::PMODE::processTasks(int workerThreadId)
diff -r d11482e5fedb -r 5943a1f73d58 source/encoder/analysis.h
--- a/source/encoder/analysis.h Mon Jul 24 11:15:38 2017 +0530
+++ b/source/encoder/analysis.h Fri Aug 04 01:48:28 2017 +0800
@@ -145,7 +145,7 @@
void qprdRefine(const CUData& parentCTU, const CUGeom& cuGeom, int32_t qp,
int32_t lqp);
/* full analysis for an I-slice CU */
- void compressIntraCU(const CUData& parentCTU, const CUGeom& cuGeom,
int32_t qp);
+ uint64_t compressIntraCU(const CUData& parentCTU, const CUGeom& cuGeom,
int32_t qp);
/* full analysis for a P or B slice CU */
uint32_t compressInterCU_dist(const CUData& parentCTU, const CUGeom&
cuGeom, int32_t qp);
diff -r d11482e5fedb -r 5943a1f73d58 source/x265.h
--- a/source/x265.h Mon Jul 24 11:15:38 2017 +0530
+++ b/source/x265.h Fri Aug 04 01:48:28 2017 +0800
@@ -1482,6 +1482,9 @@
/* Force flushing the frames from encoder */
int forceFlush;
+
+ /* Enable skipping split RD analysis when sum of split CU rdCost larger
than none split CU rdCost for Intra CU */
+ int bEnableSplitRdSkip;
} x265_param;
/* x265_param_alloc:
diff -r d11482e5fedb -r 5943a1f73d58 source/x265cli.h
--- a/source/x265cli.h Mon Jul 24 11:15:38 2017 +0530
+++ b/source/x265cli.h Fri Aug 04 01:48:28 2017 +0800
@@ -281,6 +281,8 @@
{ "refine-mv", no_argument, NULL, 0 },
{ "no-refine-mv", no_argument, NULL, 0 },
{ "force-flush", required_argument, NULL, 0 },
+ { "splitrd-skip", no_argument, NULL, 0 },
+ { "no-splitrd-skip", no_argument, NULL, 0 },
{ 0, 0, 0, 0 },
{ 0, 0, 0, 0 },
{ 0, 0, 0, 0 },
@@ -375,6 +377,7 @@
H0(" --[no-]early-skip Enable early SKIP detection. Default
%s\n", OPT(param->bEnableEarlySkip));
H0(" --[no-]rskip Enable early exit from recursion.
Default %s\n", OPT(param->bEnableRecursionSkip));
H1(" --[no-]tskip-fast Enable fast intra transform skipping.
Default %s\n", OPT(param->bEnableTSkipFast));
+ H1(" --[no-]splitrd-skip Enable skipping split RD analysis
when sum of split CU rdCost larger than none split CU rdCost for Intra CU.
Default %s\n", OPT(param->bEnableSplitRdSkip));
H1(" --nr-intra <integer> An integer value in range of 0 to
2000, which denotes strength of noise reduction in intra CUs. Default 0\n");
H1(" --nr-inter <integer> An integer value in range of 0 to
2000, which denotes strength of noise reduction in inter CUs. Default 0\n");
H0(" --ctu-info <integer> Enable receiving ctu information
asynchronously and determine reaction to the CTU information (0, 1, 2, 4, 6)
Default 0\n"
_______________________________________________
x265-devel mailing list
[email protected]
https://mailman.videolan.org/listinfo/x265-devel
_______________________________________________
x265-devel mailing list
[email protected]
https://mailman.videolan.org/listinfo/x265-devel_______________________________________________
x265-devel mailing list
[email protected]
https://mailman.videolan.org/listinfo/x265-devel
