Introduction to Intel software development tools-intel®Vtune visual Performance Analyzer

In the previous article, we introduced how to use the Intel compiler to improve program performance by improving the execution efficiency after program compilation. However, to improve the software performance, you can not only improve the compilation and Execution Code, but also analyze the program performance to identify performance bottlenecks and focus on optimization. Intel vtune visual performance analyzer is an analysis tool provided by Intel for many developers to find the performance bottleneck of hardware and software.
Statistics show that 80% of the code is executed 20% of the time when the program is running. In the Code 20%, the areas with relatively intensive activities are called hotspot. Hot Spot not only takes a lot of time, it is also often found in the following events: the cache is not in progress, the memory is missing pages, mistaken prediction branch. Such errors are often hidden and hard to be found. However, as long as you can find and optimize these hot spot, you can get twice the result with half the effort. Vtune mainly uses the following visual analysis solutions to help software and hardware developers find hot spot.
1. Sampling:
Display the instruction address histogram of the program execution in a graphical manner to help identify the performance bottleneck in the code. After sampling data is collected, you can view the data by process, thread, module, function, or instruction address. Sampling requires very low performance overhead and no code modification. With the Sample chart, you can easily find out which codes are in active-intensive areas, which are divided by modules. The Code module corresponding to the longest purple-red area represents hotspot, optimization is required during development.
2. Call diagram:
The call diagram contains the following information:
Number of calls of the U function and the number of calls of its function
The time that u spends on every function or method.
The time it takes for the U function to block or wait.
Key Path of U's calling hierarchy
The time consumed by u is greater than N % of the total time, where n is specified by the user.
3. counter monitor:
The "counter monitor" allows you to view application performance in real time. Monitor any of the more than 200 available operating system counters. You can create custom performance monitors to monitor software and hardware performance.
The Count information includes the redirection network error rate, memory usage, context switching rate, and CPU time.
After reading the above three visual analysis solutions, you must have your own ideas on how to use vtune to find performance bottlenecks. In addition, vtune also provides some other practical functions, such:
Multithreading support
You can view the sample data or call graphs of multiple threads at the same time, or view the data of a specific thread or processor separately on the monitor.
The Remote Data Collection user can configure, start, and stop the evaluation data collection Session of the remote machine on the host, and then import the data to the vtune on the host system.
This feature supports Linux and Windows.
Intel optimization Assistant
It can explain the sample data and counter monitor data, automatically determine the performance bottleneck in the code, and then provide in-depth insights and adjustment suggestions. It also provides adjustment suggestions based on C, C ++, Fortran, Java, or assembly language.
Vtune supports Windows and Linux. Supports Fortran, C #, C/C ++, and Java programming, and is fully integrated with mainstream development environments, including
Microsoft's Visual Studio, Intel C/C ++, a variety of Fo RT Ra n compilers, Compaq visualfortran, Borland's Delphi, C ++ builder, and IBM Visual Age.