[x265] [PATCH] make cu level stats logging run time configurable
Steve Borho
steve at borho.org
Sat Jun 21 00:30:24 CEST 2014
On Fri, Jun 20, 2014 at 2:05 PM, <aarthi at multicorewareinc.com> wrote:
> # HG changeset patch
> # User Aarthi Thirumalai<aarthi at multicorewareinc.com>
> # Date 1403263555 -19800
> # Fri Jun 20 16:55:55 2014 +0530
> # Node ID 71a8e0d9f931dfccf77db25e575841fdfe6baba4
> # Parent c2ebebb66bf4fd92c83288e821e98a24a4ba2610
> make cu level stats logging run time configurable
I've queued this with a few changes:
* changed --log-stats to --cu-stats to prevent --log from being an
ambiguous completion of --log-level
* bumped X265_BUILD
* added more detail to the commit message about *why* this was necessary
> diff -r c2ebebb66bf4 -r 71a8e0d9f931 source/CMakeLists.txt
> --- a/source/CMakeLists.txt Fri Jun 20 12:29:37 2014 +0530
> +++ b/source/CMakeLists.txt Fri Jun 20 16:55:55 2014 +0530
> @@ -183,12 +183,7 @@
> endif()
> endif(WARNINGS_AS_ERRORS)
>
> -option(LOG_CU_STATISTICS "Log Mode Decision Statistics at the CU Level" OFF)
> -if(LOG_CU_STATISTICS)
> - add_definitions(-DLOG_CU_STATISTICS=1)
> -else(LOG_CU_STATISTICS)
> - add_definitions(-DLOG_CU_STATISTICS=0)
> -endif(LOG_CU_STATISTICS)
> +
>
> option(ENABLE_PPA "Enable PPA profiling instrumentation" OFF)
> if(ENABLE_PPA)
> diff -r c2ebebb66bf4 -r 71a8e0d9f931 source/Lib/TLibEncoder/TEncCu.cpp
> --- a/source/Lib/TLibEncoder/TEncCu.cpp Fri Jun 20 12:29:37 2014 +0530
> +++ b/source/Lib/TLibEncoder/TEncCu.cpp Fri Jun 20 16:55:55 2014 +0530
> @@ -317,36 +317,35 @@
> m_tempCU[0]->initCU(cu->getPic(), cu->getAddr());
>
> // analysis of CU
> -#if LOG_CU_STATISTICS
> int numPartition = cu->getTotalNumPart();
> -#endif
>
> if (m_bestCU[0]->getSlice()->getSliceType() == I_SLICE)
> {
> xCompressIntraCU(m_bestCU[0], m_tempCU[0], 0, false);
> -#if LOG_CU_STATISTICS
> - int i = 0, part;
> - do
> + if (m_param->bLogCuStats || m_param->rc.bStatWrite)
> {
> - m_log->totalCu++;
> - part = cu->getDepth(i);
> - int next = numPartition >> (part * 2);
> - if (part == g_maxCUDepth - 1 && cu->getPartitionSize(i) != SIZE_2Nx2N)
> + int i = 0, part;
> + do
> {
> - m_log->cntIntraNxN++;
> + m_log->totalCu++;
> + part = cu->getDepth(i);
> + int next = numPartition >> (part * 2);
> + if (part == g_maxCUDepth - 1 && cu->getPartitionSize(i) != SIZE_2Nx2N)
> + {
> + m_log->cntIntraNxN++;
> + }
> + else
> + {
> + m_log->cntIntra[part]++;
> + if (cu->getLumaIntraDir()[i] > 1)
> + m_log->cuIntraDistribution[part][ANGULAR_MODE_ID]++;
> + else
> + m_log->cuIntraDistribution[part][cu->getLumaIntraDir()[i]]++;
> + }
> + i += next;
> }
> - else
> - {
> - m_log->cntIntra[part]++;
> - if (cu->getLumaIntraDir()[i] > 1)
> - m_log->cuIntraDistribution[part][ANGULAR_MODE_ID]++;
> - else
> - m_log->cuIntraDistribution[part][cu->getLumaIntraDir()[i]]++;
> - }
> - i += next;
> + while (i < numPartition);
> }
> - while (i < numPartition);
> -#endif // if LOG_CU_STATISTICS
> }
> else
> {
> @@ -360,48 +359,49 @@
> }
> else
> xCompressCU(m_bestCU[0], m_tempCU[0], 0, false);
> -#if LOG_CU_STATISTICS
> - int i = 0, part;
> - do
> + if (m_param->bLogCuStats || m_param->rc.bStatWrite)
> {
> - part = cu->getDepth(i);
> - m_log->cntTotalCu[part]++;
> - int next = numPartition >> (part * 2);
> - if (cu->isSkipped(i))
> + int i = 0, part;
> + do
> {
> - m_log->cntSkipCu[part]++;
> - }
> - else
> - {
> - m_log->totalCu++;
> - if (cu->getPredictionMode(0) == MODE_INTER)
> + part = cu->getDepth(i);
> + m_log->cntTotalCu[part]++;
> + int next = numPartition >> (part * 2);
> + if (cu->isSkipped(i))
> {
> - m_log->cntInter[part]++;
> - if (cu->getPartitionSize(0) < AMP_ID)
> - m_log->cuInterDistribution[part][cu->getPartitionSize(0)]++;
> - else
> - m_log->cuInterDistribution[part][AMP_ID]++;
> + m_log->cntSkipCu[part]++;
> }
> - else if (cu->getPredictionMode(0) == MODE_INTRA)
> + else
> {
> - if (part == g_maxCUDepth - 1 && cu->getPartitionSize(0) == SIZE_NxN)
> + m_log->totalCu++;
> + if (cu->getPredictionMode(0) == MODE_INTER)
> {
> - m_log->cntIntraNxN++;
> + m_log->cntInter[part]++;
> + if (cu->getPartitionSize(0) < AMP_ID)
> + m_log->cuInterDistribution[part][cu->getPartitionSize(0)]++;
> + else
> + m_log->cuInterDistribution[part][AMP_ID]++;
> }
> - else
> + else if (cu->getPredictionMode(0) == MODE_INTRA)
> {
> - m_log->cntIntra[part]++;
> - if (cu->getLumaIntraDir()[0] > 1)
> - m_log->cuIntraDistribution[part][ANGULAR_MODE_ID]++;
> + if (part == g_maxCUDepth - 1 && cu->getPartitionSize(0) == SIZE_NxN)
> + {
> + m_log->cntIntraNxN++;
> + }
> else
> - m_log->cuIntraDistribution[part][cu->getLumaIntraDir()[0]]++;
> + {
> + m_log->cntIntra[part]++;
> + if (cu->getLumaIntraDir()[0] > 1)
> + m_log->cuIntraDistribution[part][ANGULAR_MODE_ID]++;
> + else
> + m_log->cuIntraDistribution[part][cu->getLumaIntraDir()[0]]++;
> + }
> }
> }
> + i = i + next;
> }
> - i = i + next;
> + while (i < numPartition);
> }
> - while (i < numPartition);
> -#endif // if LOG_CU_STATISTICS
> }
> }
>
> diff -r c2ebebb66bf4 -r 71a8e0d9f931 source/Lib/TLibEncoder/TEncCu.h
> --- a/source/Lib/TLibEncoder/TEncCu.h Fri Jun 20 12:29:37 2014 +0530
> +++ b/source/Lib/TLibEncoder/TEncCu.h Fri Jun 20 16:55:55 2014 +0530
> @@ -131,10 +131,9 @@
>
> public:
>
> -#if LOG_CU_STATISTICS
> StatisticLog m_sliceTypeLog[3];
> StatisticLog* m_log;
> -#endif
> +
> TEncCu();
>
> void init(Encoder* top);
> diff -r c2ebebb66bf4 -r 71a8e0d9f931 source/common/param.cpp
> --- a/source/common/param.cpp Fri Jun 20 12:29:37 2014 +0530
> +++ b/source/common/param.cpp Fri Jun 20 16:55:55 2014 +0530
> @@ -103,6 +103,7 @@
> param->frameNumThreads = 0;
> param->poolNumThreads = 0;
> param->csvfn = NULL;
> + param->bLogCuStats = 0;
>
> /* Source specifications */
> param->internalBitDepth = x265_max_bit_depth;
> @@ -559,6 +560,7 @@
> p->logLevel = parseName(value, logLevelNames, bError) - 1;
> }
> }
> + OPT("log-stats") p->bLogCuStats = atobool(value);
> OPT("repeat-headers") p->bRepeatHeaders = atobool(value);
> OPT("wpp") p->bEnableWavefront = atobool(value);
> OPT("ctu") p->maxCUSize = (uint32_t)atoi(value);
> diff -r c2ebebb66bf4 -r 71a8e0d9f931 source/encoder/encoder.cpp
> --- a/source/encoder/encoder.cpp Fri Jun 20 12:29:37 2014 +0530
> +++ b/source/encoder/encoder.cpp Fri Jun 20 16:55:55 2014 +0530
> @@ -522,154 +522,154 @@
>
> void Encoder::printSummary()
> {
> -#if LOG_CU_STATISTICS
> - for (int sliceType = 2; sliceType >= 0; sliceType--)
> + if (m_param->bLogCuStats)
> {
> - if (sliceType == P_SLICE && !m_analyzeP.m_numPics)
> - continue;
> - if (sliceType == B_SLICE && !m_analyzeB.m_numPics)
> - continue;
> + for (int sliceType = 2; sliceType >= 0; sliceType--)
> + {
> + if (sliceType == P_SLICE && !m_analyzeP.m_numPics)
> + continue;
> + if (sliceType == B_SLICE && !m_analyzeB.m_numPics)
> + continue;
>
> - StatisticLog finalLog;
> - for (int depth = 0; depth < (int)g_maxCUDepth; depth++)
> - {
> - for (int j = 0; j < m_totalFrameThreads; j++)
> + StatisticLog finalLog;
> + for (int depth = 0; depth < (int)g_maxCUDepth; depth++)
> {
> - for (int row = 0; row < m_frameEncoder[0].m_numRows; row++)
> + for (int j = 0; j < m_totalFrameThreads; j++)
> {
> - StatisticLog& enclog = m_frameEncoder[j].m_rows[row].m_cuCoder.m_sliceTypeLog[sliceType];
> - if (depth == 0)
> - finalLog.totalCu += enclog.totalCu;
> - finalLog.cntIntra[depth] += enclog.cntIntra[depth];
> - for (int m = 0; m < INTER_MODES; m++)
> + for (int row = 0; row < m_frameEncoder[0].m_numRows; row++)
> {
> - if (m < INTRA_MODES)
> + StatisticLog& enclog = m_frameEncoder[j].m_rows[row].m_cuCoder.m_sliceTypeLog[sliceType];
> + if (depth == 0)
> + finalLog.totalCu += enclog.totalCu;
> + finalLog.cntIntra[depth] += enclog.cntIntra[depth];
> + for (int m = 0; m < INTER_MODES; m++)
> {
> - finalLog.cuIntraDistribution[depth][m] += enclog.cuIntraDistribution[depth][m];
> + if (m < INTRA_MODES)
> + {
> + finalLog.cuIntraDistribution[depth][m] += enclog.cuIntraDistribution[depth][m];
> + }
> + finalLog.cuInterDistribution[depth][m] += enclog.cuInterDistribution[depth][m];
> }
> - finalLog.cuInterDistribution[depth][m] += enclog.cuInterDistribution[depth][m];
> - }
>
> - if (depth == (int)g_maxCUDepth - 1)
> - finalLog.cntIntraNxN += enclog.cntIntraNxN;
> - if (sliceType != I_SLICE)
> - {
> - finalLog.cntTotalCu[depth] += enclog.cntTotalCu[depth];
> - finalLog.cntInter[depth] += enclog.cntInter[depth];
> - finalLog.cntSkipCu[depth] += enclog.cntSkipCu[depth];
> + if (depth == (int)g_maxCUDepth - 1)
> + finalLog.cntIntraNxN += enclog.cntIntraNxN;
> + if (sliceType != I_SLICE)
> + {
> + finalLog.cntTotalCu[depth] += enclog.cntTotalCu[depth];
> + finalLog.cntInter[depth] += enclog.cntInter[depth];
> + finalLog.cntSkipCu[depth] += enclog.cntSkipCu[depth];
> + }
> }
> }
> - }
>
> - uint64_t cntInter, cntSkipCu, cntIntra = 0, cntIntraNxN = 0, encCu = 0;
> - uint64_t cuInterDistribution[INTER_MODES], cuIntraDistribution[INTRA_MODES];
> + uint64_t cntInter, cntSkipCu, cntIntra = 0, cntIntraNxN = 0, encCu = 0;
> + uint64_t cuInterDistribution[INTER_MODES], cuIntraDistribution[INTRA_MODES];
>
> - // check for 0/0, if true assign 0 else calculate percentage
> - for (int n = 0; n < INTER_MODES; n++)
> - {
> - if (finalLog.cntInter[depth] == 0)
> - cuInterDistribution[n] = 0;
> + // check for 0/0, if true assign 0 else calculate percentage
> + for (int n = 0; n < INTER_MODES; n++)
> + {
> + if (finalLog.cntInter[depth] == 0)
> + cuInterDistribution[n] = 0;
> + else
> + cuInterDistribution[n] = (finalLog.cuInterDistribution[depth][n] * 100) / finalLog.cntInter[depth];
> + if (n < INTRA_MODES)
> + {
> + if (finalLog.cntIntra[depth] == 0)
> + {
> + cntIntraNxN = 0;
> + cuIntraDistribution[n] = 0;
> + }
> + else
> + {
> + cntIntraNxN = (finalLog.cntIntraNxN * 100) / finalLog.cntIntra[depth];
> + cuIntraDistribution[n] = (finalLog.cuIntraDistribution[depth][n] * 100) / finalLog.cntIntra[depth];
> + }
> + }
> + }
> +
> + if (finalLog.totalCu == 0)
> + {
> + encCu = 0;
> + }
> else
> - cuInterDistribution[n] = (finalLog.cuInterDistribution[depth][n] * 100) / finalLog.cntInter[depth];
> - if (n < INTRA_MODES)
> {
> - if (finalLog.cntIntra[depth] == 0)
> + if (sliceType == I_SLICE)
> {
> - cntIntraNxN = 0;
> - cuIntraDistribution[n] = 0;
> + cntIntra = (finalLog.cntIntra[depth] * 100) / finalLog.totalCu;
> + cntIntraNxN = (finalLog.cntIntraNxN * 100) / finalLog.totalCu;
> }
> else
> {
> - cntIntraNxN = (finalLog.cntIntraNxN * 100) / finalLog.cntIntra[depth];
> - cuIntraDistribution[n] = (finalLog.cuIntraDistribution[depth][n] * 100) / finalLog.cntIntra[depth];
> + encCu = ((finalLog.cntIntra[depth] + finalLog.cntInter[depth]) * 100) / finalLog.totalCu;
> }
> }
> - }
> -
> - if (finalLog.totalCu == 0)
> - {
> - encCu = 0;
> - }
> - else
> - {
> if (sliceType == I_SLICE)
> {
> - cntIntra = (finalLog.cntIntra[depth] * 100) / finalLog.totalCu;
> - cntIntraNxN = (finalLog.cntIntraNxN * 100) / finalLog.totalCu;
> - }
> - else
> - {
> - encCu = ((finalLog.cntIntra[depth] + finalLog.cntInter[depth]) * 100) / finalLog.totalCu;
> - }
> - }
> - if (sliceType == I_SLICE)
> - {
> - cntInter = 0;
> - cntSkipCu = 0;
> - }
> - else
> - {
> - if (finalLog.cntTotalCu[depth] == 0)
> - {
> cntInter = 0;
> - cntIntra = 0;
> cntSkipCu = 0;
> }
> else
> {
> - cntInter = (finalLog.cntInter[depth] * 100) / finalLog.cntTotalCu[depth];
> - cntIntra = (finalLog.cntIntra[depth] * 100) / finalLog.cntTotalCu[depth];
> - cntSkipCu = (finalLog.cntSkipCu[depth] * 100) / finalLog.cntTotalCu[depth];
> + if (finalLog.cntTotalCu[depth] == 0)
> + {
> + cntInter = 0;
> + cntIntra = 0;
> + cntSkipCu = 0;
> + }
> + else
> + {
> + cntInter = (finalLog.cntInter[depth] * 100) / finalLog.cntTotalCu[depth];
> + cntIntra = (finalLog.cntIntra[depth] * 100) / finalLog.cntTotalCu[depth];
> + cntSkipCu = (finalLog.cntSkipCu[depth] * 100) / finalLog.cntTotalCu[depth];
> + }
> }
> - }
> - // print statistics
> - int cuSize = g_maxCUSize >> depth;
> - char stats[256] = { 0 };
> - int len = 0;
> - if (sliceType != I_SLICE)
> - {
> - len += sprintf(stats + len, "EncCU "X265_LL "%% Merge "X265_LL "%%", encCu, cntSkipCu);
> - }
> - if (cntInter)
> - {
> - len += sprintf(stats + len, " Inter "X265_LL "%%", cntInter);
> - if (m_param->bEnableAMP)
> - len += sprintf(stats + len, "(%dx%d "X265_LL "%% %dx%d "X265_LL "%% %dx%d "X265_LL "%% AMP "X265_LL "%%)",
> - cuSize, cuSize, cuInterDistribution[0],
> - cuSize / 2, cuSize, cuInterDistribution[2],
> - cuSize, cuSize / 2, cuInterDistribution[1],
> - cuInterDistribution[3]);
> - else if (m_param->bEnableRectInter)
> - len += sprintf(stats + len, "(%dx%d "X265_LL "%% %dx%d "X265_LL "%% %dx%d "X265_LL "%%)",
> - cuSize, cuSize, cuInterDistribution[0],
> - cuSize / 2, cuSize, cuInterDistribution[2],
> - cuSize, cuSize / 2, cuInterDistribution[1]);
> - }
> - if (cntIntra)
> - {
> - len += sprintf(stats + len, " Intra "X265_LL "%%(DC "X265_LL "%% P "X265_LL "%% Ang "X265_LL "%%",
> - cntIntra, cuIntraDistribution[0],
> - cuIntraDistribution[1], cuIntraDistribution[2]);
> + // print statistics
> + int cuSize = g_maxCUSize >> depth;
> + char stats[256] = { 0 };
> + int len = 0;
> if (sliceType != I_SLICE)
> {
> - if (depth == (int)g_maxCUDepth - 1)
> - len += sprintf(stats + len, " %dx%d "X265_LL "%%", cuSize / 2, cuSize / 2, cntIntraNxN);
> + len += sprintf(stats + len, "EncCU "X265_LL "%% Merge "X265_LL "%%", encCu, cntSkipCu);
> }
> + if (cntInter)
> + {
> + len += sprintf(stats + len, " Inter "X265_LL "%%", cntInter);
> + if (m_param->bEnableAMP)
> + len += sprintf(stats + len, "(%dx%d "X265_LL "%% %dx%d "X265_LL "%% %dx%d "X265_LL "%% AMP "X265_LL "%%)",
> + cuSize, cuSize, cuInterDistribution[0],
> + cuSize / 2, cuSize, cuInterDistribution[2],
> + cuSize, cuSize / 2, cuInterDistribution[1],
> + cuInterDistribution[3]);
> + else if (m_param->bEnableRectInter)
> + len += sprintf(stats + len, "(%dx%d "X265_LL "%% %dx%d "X265_LL "%% %dx%d "X265_LL "%%)",
> + cuSize, cuSize, cuInterDistribution[0],
> + cuSize / 2, cuSize, cuInterDistribution[2],
> + cuSize, cuSize / 2, cuInterDistribution[1]);
> + }
> + if (cntIntra)
> + {
> + len += sprintf(stats + len, " Intra "X265_LL "%%(DC "X265_LL "%% P "X265_LL "%% Ang "X265_LL "%%",
> + cntIntra, cuIntraDistribution[0],
> + cuIntraDistribution[1], cuIntraDistribution[2]);
> + if (sliceType != I_SLICE)
> + {
> + if (depth == (int)g_maxCUDepth - 1)
> + len += sprintf(stats + len, " %dx%d "X265_LL "%%", cuSize / 2, cuSize / 2, cntIntraNxN);
> + }
>
> - len += sprintf(stats + len, ")");
> - if (sliceType == I_SLICE)
> - {
> - if (depth == (int)g_maxCUDepth - 1)
> - len += sprintf(stats + len, " %dx%d: "X265_LL "%%", cuSize / 2, cuSize / 2, cntIntraNxN);
> + len += sprintf(stats + len, ")");
> + if (sliceType == I_SLICE)
> + {
> + if (depth == (int)g_maxCUDepth - 1)
> + len += sprintf(stats + len, " %dx%d: "X265_LL "%%", cuSize / 2, cuSize / 2, cntIntraNxN);
> + }
> }
> + const char slicechars[] = "BPI";
> + if (stats[0])
> + x265_log(m_param, X265_LOG_INFO, "%c%-2d: %s\n", slicechars[sliceType], cuSize, stats);
> }
> - const char slicechars[] = "BPI";
> - if (stats[0])
> - x265_log(m_param, X265_LOG_INFO, "%c%-2d: %s\n", slicechars[sliceType], cuSize, stats);
> }
> }
> -
> -#endif // if LOG_CU_STATISTICS
> if (m_param->logLevel >= X265_LOG_INFO)
> {
> char buffer[200];
> diff -r c2ebebb66bf4 -r 71a8e0d9f931 source/encoder/frameencoder.cpp
> --- a/source/encoder/frameencoder.cpp Fri Jun 20 12:29:37 2014 +0530
> +++ b/source/encoder/frameencoder.cpp Fri Jun 20 16:55:55 2014 +0530
> @@ -259,13 +259,16 @@
>
> slice->setScalingList(m_top->getScalingList());
> slice->getScalingList()->setUseTransformSkip(m_pps.getUseTransformSkip());
> -#if LOG_CU_STATISTICS
> - for (int i = 0; i < m_numRows; i++)
> +
> + /* We also need analysis of cu distribution across various modes in multipass encodes */
> + if (m_param->bLogCuStats || m_param->rc.bStatWrite)
> {
> - m_rows[i].m_cuCoder.m_log = &m_rows[i].m_cuCoder.m_sliceTypeLog[sliceType];
> + for (int i = 0; i < m_numRows; i++)
> + {
> + m_rows[i].m_cuCoder.m_log = &m_rows[i].m_cuCoder.m_sliceTypeLog[sliceType];
> + }
> }
>
> -#endif
> if (slice->getPPS()->getDeblockingFilterControlPresentFlag())
> {
> slice->getPPS()->setDeblockingFilterOverrideEnabledFlag(!m_top->m_loopFilterOffsetInPPS);
> diff -r c2ebebb66bf4 -r 71a8e0d9f931 source/x265.cpp
> --- a/source/x265.cpp Fri Jun 20 12:29:37 2014 +0530
> +++ b/source/x265.cpp Fri Jun 20 16:55:55 2014 +0530
> @@ -71,6 +71,8 @@
> { "log-level", required_argument, NULL, 0 },
> { "level", required_argument, NULL, 0 },
> { "csv", required_argument, NULL, 0 },
> + { "no-log-stats", no_argument, NULL, 0 },
> + { "log-stats", no_argument, NULL, 0 },
> { "y4m", no_argument, NULL, 0 },
> { "no-progress", no_argument, NULL, 0 },
> { "output", required_argument, NULL, 'o' },
> @@ -315,6 +317,7 @@
> H0(" --log-level <string> Logging level: none error warning info debug full. Default %s\n", logLevelNames[param->logLevel + 1]);
> H0(" --csv <filename> Comma separated log file, log level >= 3 frame log, else one line per run\n");
> H0(" --no-progress Disable CLI progress reports\n");
> + H0(" --[no-]log-stats Enable logging stats about distribution of cu across all modes. Default %s\n",OPT(param->bLogCuStats));
> H0("-o/--output <filename> Bitstream output file name\n");
> H0("\nInput Options:\n");
> H0(" --input <filename> Raw YUV or Y4M input file name. `-` for stdin\n");
> diff -r c2ebebb66bf4 -r 71a8e0d9f931 source/x265.h
> --- a/source/x265.h Fri Jun 20 12:29:37 2014 +0530
> +++ b/source/x265.h Fri Jun 20 16:55:55 2014 +0530
> @@ -674,6 +674,9 @@
> * chosen, the cu-transquant-bypass flag is set for that CU. */
> int bCULossless;
>
> + /* Enable analysis and logging distribution of Cus encoded across various modes during mode decision.*/
> + int bLogCuStats;
> +
> /*== Rate Control ==*/
>
> struct
> _______________________________________________
> x265-devel mailing list
> x265-devel at videolan.org
> https://mailman.videolan.org/listinfo/x265-devel
--
Steve Borho
More information about the x265-devel
mailing list