一、Intel Vtune的安装
在前面初步认识了一下几个性能的测试工具,本篇重点介绍一下Intel VTune Profiler,VTune是一个强大的性能分析工具,它属于Intel oneAPI工具包中工具的一种。VTune的安装只介绍在Linux平台下的场景(Windows安装相对简单)。
1、两种安装方法
第一种方法:
打开网址https://www.intel.com/content/www/us/en/developer/tools/oneapi/toolkits.html#base-kit。找到合适的版本安装,不过这个安装是一个全版本的安装,可能会安装很多其它工具。
第二种方法:
使用命令安装:
sudo yum install intel-oneapi-vtune # CentOS
sudo apt install intel-oneapi-vtune # Ubuntu/Debian
或使用离线安装包:
wget https://registrationcenter-download.intel.com/akdlm/IRC_NAS/6bfca885-4156-491e-849b-1cd7da9cc760/intel-oneapi-base-toolkit-2025.1.1.36_offline.sh
sudo sh ./intel-oneapi-base-toolkit-2025.1.1.36_offline.sh -a --silent --cli --eula accept
2、配置
主要是设置环境变量和权限
//配置环境变量
source /opt/intel/oneapi/vtune/latest/env/vars.sh # oneAPI 版本
source /opt/intel/vtune_amplifier/amplxe-vars.sh # 老版本
//设置用户权限
sudo usermod -aG vtune $USER
3、验证
可以运行相关脚本:
bash /opt/intel/oneapi/vtune/latest/bin64/vtune-self-checker.sh
二、主要功能
VTune有三种使用方式,一个是使用GUI的方式,另外一种是命令方式,还有一种是远程应用的方式。
1、使用GUI方式
使用下面的命令启动:
vtune-gui
然后就可以使用UI进行处理相关的测试。首先创建一个新的测试项目,选择相应的类型;其次配置测试程序的路径或进程ID;最后启动分析并查看结果,如对火焰图或调用栈等进行分析。
如下图:
2、命令方式
执行下面命令:
vtune -collect hotspots -r ./result_dir -- ./your_application #运行
amplxe-cl -report hotspots -r ./result_dir -format text -report-output ./report.txt #生成报告
hotspots为分析的类型(其它还有locksandwaits等),生成的报告支持txt,csv及html。
3、远程应用方式
远程方式也比较简单,它也分成有无界面的操作方式,其实这个和VTune本身没有什么太大关系。可以使用远程桌面的一些工具(如向日葵,VNC Server等)或在一些开发IDE上使用插件(如VSCODE的oneAPI插件)。这里就不再展开分析说明。
三、例程应用
下面看一个例程分析,对OneAPI自带的矩阵测试程序进行:
vtune -collect hotspots -r ~/result -- ./matrix
vtune: Warning: Microarchitecture performance insights will not be available. Make sure the sampling driver is installed and enabled on your system.
vtune: Collection started. To stop the collection, either press CTRL-C or enter from another console window: vtune -r /home/fpc/result -command stop.
Addr of buf1 = 0x7f339f7b5010
Offs of buf1 = 0x7f339f7b5180
Addr of buf2 = 0x7f339d7b4010
Offs of buf2 = 0x7f339d7b41c0
Addr of buf3 = 0x7f339b7b3010
Offs of buf3 = 0x7f339b7b3100
Addr of buf4 = 0x7f33997b2010
Offs of buf4 = 0x7f33997b2140
Threads #: 16 Pthreads
Matrix size: 2048
Using multiply kernel: multiply1
Execution time = 3.516 seconds
vtune: Collection stopped.
vtune: Using result path `/home/fpc/result'
vtune: Executing actions 20 % Resolving information for `libtpsstool.so'
vtune: Warning: Cannot locate debugging information for file `/opt/intel/oneapi/vtune/2024.0/lib64/libtpsstool.so'.
vtune: Executing actions 75 % Generating a report Elapsed Time: 3.535sCPU Time: 46.929sEffective Time: 46.929sSpin Time: 0sOverhead Time: 0sTotal Thread Count: 17Paused Time: 0sTop Hotspots
Function Module CPU Time % of CPU Time(%)
--------- --------- -------- ----------------
multiply1 matrix 46.909s 100.0%
init_arr matrix 0.010s 0.0%
__GI_ libc.so.6 0.010s 0.0%
Collection and Platform InfoApplication Command Line: ./matrixOperating System: 5.19.0-50-generic DISTRIB_ID=Kylin DISTRIB_RELEASE=V10 DISTRIB_CODENAME=kylin DISTRIB_DESCRIPTION="Kylin V10 SP1" DISTRIB_KYLIN_RELEASE=V10 DISTRIB_VERSION_TYPE=enterprise DISTRIB_VERSION_MODE=normalComputer Name: fjfResult Size: 4.5 MBCollection start time: 10:55:09 12/05/2025 UTCCollection stop time: 10:55:13 12/05/2025 UTCCollector Type: User-mode sampling and tracingCPUName: Intel(R) microarchitecture code named Alderlake-SFrequency: 2.112 GHzLogical CPU Count: 20Cache Allocation TechnologyLevel 2 capability: not detectedLevel 3 capability: not detectedIf you want to skip descriptions of detected performance issues in the report,
enter: vtune -report summary -report-knob show-issues=false -r <my_result_dir>.
Alternatively, you may view the report in the csv format: vtune -report
<report_name> -format=csv.
vtune: Executing actions 100 % done
同时,会在指定的目录下,生成一个文件夹,内部有不少的供分析的相关文件,此处为/home/user/result
如果在执行命令时出现:
vtune: Error: Cannot start data collection because the scope of ptrace system call is limited. To enable profiling, please set /proc/sys/kernel/yama/ptrace_scope to 0. To make this change permanent, set kernel.yama.ptrace_scope to 0 in /etc/sysctl.d/10-ptrace.conf and reboot the machine.
vtune: Warning: Microarchitecture performance insights will not be available. Make sure the sampling driver is installed and enabled on your system
可执行命令:
echo 0 | sudo tee /proc/sys/kernel/yama/ptrace_scope
或
sudo sysctl -w kernel.yama.ptrace_scope=0
四、总结
会熟练的使用各种测试工具,是对程序进行优化的前提。特别是在一些性能需求相当关键的场景下,不借助工具会导致性能优化的效率急剧降低。磨刀不误砍柴工,与诸君共勉!