hevc移植/优化-day 3：profiling
为了找出耗时的函数，需要用gprof之类的工具来分析性能/耗时，如果手动加代码测耗时部分过于繁锁。本文简述如何在linux下分析编解码器的性能。
首先修改makefile支持profiling。
修改build/linux/common/makefile.base:
##########
# do profiling : 1=yes, 0=no
##########
PROFILING?= 0
ifeq ($(PROFILING),1)
CPPFLAGS+=-pg
endif
##########
##########
# do profiling : 1=yes, 0=no
##########
PROFILING?= 0
ifeq ($(PROFILING),1)
ALL_LDFLAGS+=-pg
endif
##########修改build/linux/makefile
### do profiling : 1=yes, 0=no
PROFILING?= 1
export PROFILING重新make生成TAppEncoderStatic/TAppDecoderStatic。
接着开始测试编码：
./TAppEncoderStatic -c ../cfg/encoder_randomaccess.cfg -i ../video/RaceHorses_416x240_30.yuv -wdt 416 -hgt 240 -fr 30 --InputBitDepth=8 --OutputBitDepth=8 --InternalBitDepth=8 -b RaceHorses_416x240_30.265 -f 60
-c configuration file name
-i, --InputFile original YUV input file name
-b, --BitstreamFile bitstream output file name
-wdt, --SourceWidth Source picture width
-hgt, --SourceHeight Source picture height
-fr, --FrameRate Frame rate
--InputBitDepth bit-depth of input file
--OutputBitDepth bit-depth of output fileyuv文件可以在ftp://hvc:[email protected]/testsequences下载，注意要加-f参数，此步编码生成文件RaceHorses_416x240_30.265，用于测试解码：
./TAppDecoderStatic -d 0 -o decout.yuv -b RaceHorses_416x240_30.265
运行编码/解码程序后都会生成gmon.out，运行gprof TAppDecoderStatic输出性能分析结果，以下是较耗时的几个函数：
Each sample counts as 0.01 seconds.
% cumulative self self total
time seconds seconds calls ms/call ms/call name
12.09 0.11 0.11 26707 0.00 0.00 void TComInterpolationFilter::filter<8, false, true, false>(short const*, int, short*, int, int, int, short const*)
7.69 0.18 0.07 12952 0.01 0.01 void TComInterpolationFilter::filter<8, true, false, false>(short const*, int, short*, int, int, int, short const*)
5.49 0.23 0.05 14 3.57 3.57 TComAdaptiveLoopFilter::filterFrame(unsigned short*, unsigned short*, int, int)
4.40 0.27 0.04 641709 0.00 0.00 TDecBinCABAC::decodeBin(unsigned int&, ContextModel&)
4.40 0.31 0.04 91284 0.00 0.00 TComLoopFilter::xEdgeFilterLuma(TComDataCU*, unsigned int, unsigned int, int, int)
4.40 0.35 0.04 53182 0.00 0.00 TComDataCU::xGetColMVP(RefPicList, int, int, TComMv&, int&)
4.40 0.39 0.04 7700 0.01 0.01 TComAdaptiveLoopFilter::subfilterFrame(unsigned short*, unsigned short*, int, int, int, int, int, int)
3.30 0.42 0.03 64944 0.00 0.00 void TComInterpolationFilter::filter<4, false, true, false>(short const*, int, short*, int, int, int, short const*)vs2008下的分析办法(参考http://www.codeproject.com/KB/tips/profileWithVsPerfCmd.aspx)：
安装Visual Studio 2008 Service Pack 1 Stand-Alone Profiler.在工程设置里面的链接选项里面开启：高级->探查，即启用探查信息。
使用如下脚本进行profiling，生成的报表为csv格式。
set pt="C:\Program Files\Microsoft Visual Studio 9.0\Team Tools\Performance Tools"
%pt%\vsperfcmd /start:sample /output:TAppDecoder.vsp
%pt%\vsperfcmd /launch:J:\git\hevc\bin\vc9\Win32\Release\TAppDecoder.exe /ARGS:"-d 0 -o decout.yuv -b RaceHorses_416x240_30.265"
%pt%\vsperfcmd /shutdown
%pt%\VSPerfReport /SUMMARY:ALL TAppDecoder.vsp /OUTPUT:TAppDecoder
pauseNOTE：若使用day 2的代码(去掉异常支持)，将会出现异常terminate called after throwing an instance of 'std::ios_base::failure',what(): basic_ios::clear。本文测试时使用svn原版本代码。