--- title: 'Intel VTune Profiler' taxonomy: category: - docs external_links: process: true no_follow: true target: _blank mode: active --- [toc] ## Environment Settings{#enviroment} ### Software Version and System Requirements{#version} Version | Module File Name| System A | System B/C | System G | Cloud System | Note | ---------------| ------------------ | --------- | ----------- | --------- | ---------------- | | 2024.0 | intel-vtune/2024.0 | + | + | - | - | | 2023.2 (default) | intel-vtune/2023.2 | + | + | - | - | | 2023.1 | intel-vtune/2023.1 | + | + | - | - | | 2022.3 | intel-vtune/2022.3 | + | + | - | - | | +： Available for all users <br> \- ： Not available In the [Intel compiler](/compilers/intel) available environment, execute the module command as follows. ```nohighlight $ module load intel-vtune ``` For details on the module command, please refer to [Modules](/config/modules). ## How to Use Intel VTune Profiler{#usage} ### Intel VTune Profiler Commands{#command} Commands | Purpose ------------ | --------------------------------- vtune-gui | Acrivate the GUI version of the VTune Profiler. vtune | Activate the command line version of the VTune Profiler. ### Options{#option} #### Options of the vtune Options | Purpose -------------------------- | --------------------- -collect=_string_ | Specifies the analysis type. -app-working-dir=_string_ | Specifies the working directory. -r, -result-dir=_string_ | Specifies the directory in which to save the result. #### Main types that can be specified with -collect of vtune Types | Purpose ---------------------- | ---------------------------- threading | Displays multi-threaded parallelism. hotspots | Analyze hotspots. memory-access | Analyze memory access. ## Examples{#sample} ### How to Use in GUI{#sample_gui} 1. Compiling<br> When using the Intel VTune Profiler, compile with debugging options **-g** and the same optimization options as for the actual program execution. ```nohighlight $ icc -g -O2 test.c ``` 2. Activating VTune Profiler<br> In [MobaXterm](/login/mobaxterm) or [FastX](/login/fastx), the VTune Amplifier is activated when executing the **vtune-gui** command. Please refer to [Interactive Processing](/run/interactive) for details on the **tssrun** command. ```nohighlight $ tssrun --x11 vtune-gui ```  3. Creating projects<br> Select **New Project** in the center of the screen and the following screen will appear. Then enter the appropriate **Project name** and click **Create Project**.  4. Setting projects<br> Next, the Configure Analysis screen will appear. Specify the target program in **Application** and click **OK**. If an arguments is required, specify it here.  5. Start Analysis <br> Click Performance Snapshot, specify the target of the check, and then press **Start** with the triangle symbol.  6. Confirmation of analysis results<br> After a short wait, the results of the analysis in the VTune Profiler are displayed. In this example, we can see that the multiply1 function is using a lot of CPU time.  #### Analysis of OpenMP programs The GUI version supports analysis of OpenMP programs. When starting with the tssrun command, the --rsc option specifies the number of threads. Example：8 threads execution ```nohighlight tssrun --x11 --rsc t=8:c=8 vtune-gui ``` <!-- VTuneの起動後、「Advanced Hotspots」を選び、「Analyze OpenMP Regions」を選択して「Start」を押せば、スレッド8並列での解析が行われます。  --> #### MPI Program Analysis Please use the CUI version for an analysis of MPI program. ### How to Use in CUI{#sample_cui} 1. Compiling<br> When using Intel VTune Profiler, compile with debugging options **-g** and the same optimization options as for the actual program execution. ```nohighlight $ icc -g -O2 test.c ``` 2. Executing Check<br> A CUI version of the VTune Profiler is available with the **vtune** command. In this example, we are analyzing hotspots by specifying the directory to output the results. Please refer to [Interactive Processing](/run/interactive) for details on the **tssrun** command. ```nohighlight $ tssrun vtune -collect hotspots -r=./result.vtune ./a.out salloc: Pending job allocation 723318 salloc: job 723318 queued and waiting for resources salloc: job 723318 has been allocated resources salloc: Granted job allocation 723318 salloc: Waiting for resource configuration salloc: Nodes xb0013 are ready for job vtune: Analyzing data in the node-wide mode. The hostname (xb0013) will be added to the result path/name. vtune: Collection started. （Omitted） vtune: Executing actions 42 % Saving the resultElapsed Time: 0.075s | Application execution time is too short. Metrics data may be unreliable. | Consider reducing the sampling interval or increasing your application | execution time. | CPU Time: 0.050s Effective Time: 0s Spin Time: 0s Imbalance or Serial Spinning: 0s Lock Contention: 0s Other: 0s Overhead Time: 0.050s | A significant portion of CPU time is spent in synchronization or | threading overhead. Consider increasing task granularity or the scope | of data synchronization. | Creation: 0s Scheduling: 0s Reduction: 0s Atomics: 0s Other: 0.050s Total Thread Count: 1 Paused Time: 0s Top Hotspots Function Module CPU Time % of CPU Time(%) ----------------------------- ----------- -------- ---------------- __kmp_api_omp_get_max_threads libiomp5.so 0.050s 100.0% Effective Physical Core Utilization: 100.0% (112.000 out of 112) Effective Logical Core Utilization: 148.8% (333.230 out of 224) Collection and Platform Info Application Command Line: ../oss/openmp/omp-sample/01_hello/src/c/run.x Operating System: 4.18.0-477.15.1.el8_8.x86_64 Red Hat Enterprise Linux release 8.8 (Ootpa) Computer Name: xb0013 Result Size: 5.5 MB Collection start time: 09:39:56 21/09/2023 UTC Collection stop time: 09:39:57 21/09/2023 UTC Collector Type: Event-based counting driver,User-mode sampling and tracing CPU Name: Intel(R) Xeon(R) Processor code named Sapphirerapids Frequency: 2.000 GHz Logical CPU Count: 224 LLC size: 110.1 MB Cache Allocation Technology Level 2 capability: available Level 3 capability: available If 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 salloc: Relinquishing job allocation 723318 exit code: 0 ``` 3. Checking the results<br> Execute the **vtune-gui** command in the X (GUI) available environment, such as [MobaXterm](/login/mobaxterm) or [FastX](/login/fastx), and check the results with the GUI version of VTune Amplifier. Please refer to [Interactive Processing](/run/interactive) for details on the **tssrun** command. ```nohighlight $ tssrun --x11 vtune-gui ./result/result.vtune ```  #### Analysis of Parallel Programs{#parallel} The CUI version supports analysis of MPI and OpenMP programs. When executing VTune with tssrun, the --rsc option specifies the number of parallels. <!-- It also requires the addition of mpiexec.hydra. --> Example: Analysis with MPI4 parallel ```nohighlight $ tssrun --rsc p=4 vtune -collect hotspots -r=./result ./a.out salloc: Pending job allocation 723080 salloc: job 723080 queued and waiting for resources salloc: job 723080 has been allocated resources salloc: Granted job allocation 723080 vtune: Analyzing data in the node-wide mode. The hostname (xb0127) will be added to the result path/name. vtune: Collection started. （Omitted） vtune: ExecutinElapsed Time: 1.015sg the result CPU Time: 1.710s Effective Time: 1.710s Spin Time: 0s MPI Busy Wait Time: 0s Other: 0s Overhead Time: 0s Other: 0s Total Thread Count: 8 Paused Time: 0s Top Hotspots Function Module CPU Time % of CPU Time(%) ---------------------- -------------------- -------- ---------------- read libc.so.6 0.670s 39.2% PMPI_Init libmpi.so.12 0.112s 6.6% main allrank 0.100s 5.8% dlopen libdl.so.2 0.089s 5.2% [ld-linux-x86-64.so.2] ld-linux-x86-64.so.2 0.080s 4.7% [Others] N/A 0.658s 38.5% Effective Physical Core Utilization: 89.5% (100.261 out of 112) Effective Logical Core Utilization: 45.0% (100.843 out of 224) | The metric value is low, which may signal a poor utilization of logical | CPU cores while the utilization of physical cores is acceptable. Consider | using logical cores, which in some cases can improve processor throughput | and overall performance of multi-threaded applications. | Collection and Platform Info Application Command Line: ../lecture/20230906/mpi/allrank Operating System: 4.18.0-477.15.1.el8_8.x86_64 Red Hat Enterprise Linux release 8.8 (Ootpa) Computer Name: xb0127 Result Size: 7.2 MB Collection start time: 09:24:37 21/09/2023 UTC Collection stop time: 09:24:38 21/09/2023 UTC Collector Type: Event-based counting driver,User-mode sampling and tracing CPU Name: Intel(R) Xeon(R) Processor code named Sapphirerapids Frequency: 2.000 GHz Logical CPU Count: 224 LLC size: 110.1 MB Cache Allocation Technology Level 2 capability: available Level 3 capability: available If 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 salloc: Relinquishing job allocation 723080 exit code: 0 ``` ## Manuals{#manual} * [Intel VTune Profiler Documentation](https://www.intel.com/content/www/us/en/developer/tools/oneapi/vtune-profiler-documentation.html) * [Intel VTune Progiler User Guide](https://www.intel.com/content/www/us/en/docs/vtune-profiler/user-guide/2023-2.html)