In the previous article, we introduced how to use the Intel compiler to enhance the program performance by improving program compilation. However, improving software performance can not only start from increasing compilation execution code, more time you need to analyze program performance, and find out the performance bottleneck. Intel VTune Visual Performance Analyzer is an analytical tool for Intel for many developers to find hardware and software performance bottlenecks.
Statistics show that the program is executing 20% of code in operation 80% of the time. In this 20% code, the relative intensive area of the event is called HotSpot. Hot Spot not only consumes a lot of time, but also frequently discovered in the following events: cache, memory lattice, misunderstand branch. This type of error is often very concealed and it is difficult to find. But as long as you can find and optimize these HOT SPOT, you can achieve the effect of halving. VTune mainly helps hardware and software developers looking for hotspot by a series of visual analysis programs.
1. Sampling:
The instruction address histogram executed by the program is displayed to help determine the performance bottleneck in the code. After the sample data is collected, you can view it by process, thread, module, function, or command address. Sampling only extremely low performance overhead and does not need to modify the code. By the sample map, it is convenient to understand which code is in an active dense area, and the code module corresponding to the module is divided into module. The code module corresponding to the longest purple area is a hotspot, which requires focusing in development. Optimization.
2. Call diagram:
The call map contains the following information:
U function called the number of times and the function called it
u Time to spend on each function or method
u Function costs in blocking or waiting for time
u Customer path to the hierarchy
u Time consumption accounts for a function of N% over a total time, where N is specified by the user.
3. Counter monitor:
"Counter Monitor" can view the performance of the application in real time. You can monitor any one of more than 200 available operating system counters. Users can create custom performance monitors to monitor soft and hardware performance.
Count information includes: redirect network error rate, memory saving, context switching rate, CPU time, etc.
I have seen three visual analysis programs, you must find how to use VTune to find the performance bottleneck has your own ideas. In addition, VTune also provides some other practical functions, as follows:
Multi-threaded support
You can view the sampling data of multiple threads at the same time or "call diagram" or separately view data for specific threads or processors in the display.
Remote data acquisition users can configure, start, and stop the evaluation data acquisition session of the remote machine, and then import these data into the host system VTune.
This feature supports Linux and Windows.
Intel optimization assistant
It can explain the sampling data and counter monitor data, and automatically determine the performance bottleneck in the code, and then put forward in-depth insights and adjustments. It can also provide adjustment suggestions based on C, C , Fortran, Java or assembly language.
VTUNE supports Windows and Linux. Support Fortran, C #, C / C and Java programming, and is fully integrated with the current mainstream development environment, including
Microsoft's Visual Studio, Intel C / C , a variety of FO RT RA N compiler, Compaq Visualfortran, Borland's Delphi, C Builder, and IBM Visual Age, etc.
