Operation and Use, Volume 2

Using the Profile Visualization Tool

The PVT is a post-mortem analysis tool. It is designed to process profile data files generated by the PCT used in application profiling. For more information on the PCT, refer to Using the Performance Collection Tool. After processing profile data, you can view the results in the PVT's graphical user interface display. You can also generate report and summary files. The PVT provides a command-line interface to process individual profile files directly into a summary file without initializing the graphic display. The command-line interface also enables you to generate textual profile reports. This section begins with a discussion of the PVT's graphical user interface, followed by a description of the command-line interface.

Using the Profile Visualization Tool's Graphical User Interface

The PVT provides a graphical user interface that enables you to process profile data files and view the results. The options available in the graphical user interface correspond to the commands available in the PVT's command-line interface. For more information on the command-line interface, refer to Using the Profile Visualization Tool's Command Line Interface.

Profile Visualization Tool (Graphical User Interface) Overview

The PVT's graphical user interface allows you to process and view profile data. You can load one or more files for processing and view the results in a variety of ways. After initializing the graphical user interface, you can choose the appropriate options:

The following sections describe the graphical user interface in greater detail.

Starting the Profile Visualization Tool

You can start the PVT in either graphical-user-interface (GUI) mode or command-line mode. For instructions on starting the PVT in command-line mode, refer to Using the Profile Visualization Tool's Command Line Interface. To start the PVT in graphical-user-interface mode:

Enter the pvt command at the AIX command prompt.

$ pvt

Doing this starts the PVT in graphical-user-interface mode and opens its first window - the Main Display.

To start the PVT in graphical-user-interface mode with input profile data loaded and showing in the Main Display window, enter:

$ pvt one_or_more_file_names

The following figure shows an example of the Main Display window with input profile data loaded.

View figure.

The Main Display window shows a hierarchical list of all the functions being profiled. The window is divided into two panes, the left one for viewing source code structure and the right one for viewing profile data. Each pane has a corresponding menu: the Source View menu and the Data View menu. Both the Source View and Data View menus are grayed out if no input file is loaded. The two panes share the same vertical scroll bar and are scrolled together. You can resize the panes horizontally to change their relative proportion in the Main Display window.

The source code structure pane uses ASCII text to show the identifier of each displayed object. The profile data pane represents a selected profile data field, which uses a bar chart to show the profile data associated with each object. The data value is displayed in front of the bar. When you select an object in the source code structure pane, an object menu opens that provides some actions associated with the selected object. You left-click to select an object, and right-click to bring up the selected object's object menu. When you select an object, the Object menu in the Main Display window will become available also, providing the same functions as the popup object menu.

If you load a summary profile file to start the GUI, process objects are labeled as summary process object in order to distinguish them from the process objects available in an individual profile file. Each function object has a set of statistics records associated with each profile data field.

Following are explanations of the Source View and Data View menus.

Viewing Source Code Structure

The Source View is a drop-down menu with two options: a Thread-Centric View and a Function-Centric View. The same options are available under the View drop-down menu in the Main Display window. See Viewing program variables for more information. If the input file you are loading to start the GUI is a summary file, there will be no thread information in the file. The structure displayed will be the same no matter which view is used.

Viewing Selected Profile Data

The Data View is a drop-down menu that enables you to change the type of data to be shown in the Main Display window. The Data View menu options include the following categories:

• Function Call Count
• Wall Clock Time
• Resource Usage
• Hardware Counters.

You will find similar options available in the View drop-down menu. When a particular data type is unavailable in any of the input data files, its corresponding menu option in the View menu is grayed out. The Data View drop-down menu only shows the options that have corresponding values in the input data files. When a set of files is loaded, Function Call Count is the default field in the Data View menu.

Accessing the Profile Visualization Tool's online help system

The PVT's graphical user interface has been designed to be intuitive and easy to use. However, if you do have any trouble, you can refer to the PVT's online help system. To access the tool's online help, select Help > Contents off the main window's menu bar. Many dialogs of the tool also provide Help buttons or menu items for starting the help system.

View figure.

If you open the help from one of the PVT's dialogs, a help topic describing that dialog is displayed. If you open the help from the main window, a task overview topic is displayed.

The PVT help contains topics for each of the major tasks you can perform with the PVT. The left hand pane of the window enables you to navigate the help system to display the needed help topic in the right hand pane. There are three ways to navigate the help system -- using the contents tab, using the index tab, or using the search tab:

View figure.

• the contents tab is displayed by default. Simply click on any entry in the contents tab to display the help topic.
• the index tab shows an index of the entire help system. Simply click on any entry in the index to display its associated help topic. To search the index, type a string in the Find field and press <enter>. The first index entry containing the string is highlighted. Press <enter> again to search for the next occurrence of the string in the index.
• the search tab enables you to search the help for all occurrences of a text string. Simply type the string in the Find field and press <enter>. A list of all help topics containing the string is displayed. The topics are listed in descending order according to the number of occurrences of the string. The help topic with the most occurrences of the string is displayed by default.

Using the Profile Visualization Tool's Command Line Interface

The PVT provides a command-line interface that enables you to process profile files directly without initializing the graphical user interface. The subcommands available in the command-line interface correspond to the options available in the graphical user interface. For more information on the graphical user interface, refer to Using the Profile Visualization Tool's Graphical User Interface.

Profile Visualization Tool (Command Line Interface) Overview

The PVT's command-line interface allows you to process profile data directly without using the graphical user interface. After initializing the command-line interface, you can enter the appropriate subcommands that enable you to:

• Load files for processing
• Create a summary file of all the loaded data
• Generate textual reports of profile data
• Export profile data to a file.

The following sections describe the command-line interface in greater detail.

Starting the Profile Visualization Tool in Command-Line Mode

To start the PVT in command-line mode, enter:

pvt -c

Doing this starts a command-line session without associated profile data. To start a command-line session with associated profile data, enter:

pvt -c one_or_more_file_names

Once you start a command-line session, the command line prompt changes to pvt> and remains this way until you enter the exit command to end the command-line session.

The following sections describe the command-line mode subcommands.

Loading Files

You can load a set of profile data files into the session with the load command. Enter:

load one_or_more_file_names

If a set of data already exists, then the existing data is discarded and the newly loaded data becomes the current data to be used in future actions.

Creating a Summary File

You can create a summary file of all the loaded data with the sum command. Enter:

sum summary_file_name

The merged summary data is written to the file that you specify in the command, with a suffix of .cdf being appended to the specified file name.

Generating Reports

You can generate textual reports of profile data using the report command. You can specify several different options with the report command, depending on what type of output you want. To show a list of available report types, enter:

report list

The result will look something like:

• [0] call_count: function call count report
• [1] wclock: wall clock timer report
• [2] ru_cpu: CPU usage reports
• [3] ru_mem: memory usage report
• [4] ru_paging: paging activities reports
• [5] ru_cswitch: context switch activities reports
• [6] pmc_cycle: instructions per cycle hardware counter reports
• [7] pmc_fpu: floating-point hardware counter reports
• [8] pmc_fxu: fixed-point hardware counter reports
• [9] pmc_branch: branch hardware counter reports
• [10] pmc_lsu: load and store hardware counter reports
• [11] pmc_cache: cache hardware counter reports
• [12] pmc_misc: miscellaneous hardware counter reports

To generate all the available reports to a file, enter:

report output_file_name

To generate reports by report name to a file, enter:

report "one_or_more_report_names" output_file_name

For example:

report "wclock,ru_cpu" output

To generate reports by report id to a file, enter:

report "one_or_more_report_ids" output_file_name

For example:

report "1,2" output

The report names or report ids in double quotes must be separated by a comma, with no blank space in between. No matter how many reports are selected in one report command, all the reports are output to a single file specified in the report command.

Exporting Files

You can export profile data to a specified file using the export command. Enter:

export output_file_name

A suffix .txt will be appended to the specified file name.

The currently loaded profile data is written to the user-specified file in plain text format, so the data can be loaded easily into a spreadsheet tool like Lotus^(R) 1-2-3^(R). The data that is loaded into the tool can be grouped into the following types of records:

• Profile-session record associated with each process (that is, profile session)
• Individual function or thread records
• Function statistics records.

Exiting the Profile Visualization Tool

You can end a command-line session with the exit command. Enter:

exit