Window: Google Trace View

To access this window: in the Trace Analyzer window, expand a Google trace and open one of the converted cuts.

The Google trace opens in one of the supported browsers.

You can open the collected traces in Google Chrome*, Opera*, or Yandex* browsers. The icon of the supported browser that you have installed appears next to the trace name.

In the Google trace view, you can review concurrent execution of all your application threads, as well as correlate CPU and GPU activity:

Google trace toolbar. View the title of the trace cut and use navigation and display controls:

  • View Options drop-down menu lists Google trace display options:
    • Flow events - visualize dependencies between CPU and GPU tasks. If this option is enabled, an arrow connects each GPU task with the exact CPU task on which it was scheduled.
    • Highlight VSync - display VSync data. With this option enabled, the boundaries between violet and white stripes represent VSync events:
  • The search field enables you to search for specific tasks in the captured trace:

    Using the / buttons, you can navigate between the found tasks.

  • Help button displays information on navigation controls.

Timeline toolbar. Correlate all the displayed events with the timeline. The range of the timeline corresponds to the length of the trace cut. The default trace cut length is 5s.

Depending on the zoom-in level, the time intervals are displayed in seconds, milliseconds, microseconds, on nanoseconds.

Thread timeline. Analyze thread length, execution order, and distribution between different CPUs.

The blocks of the same color represent the same thread. If you zoom into the trace, you can see the thread ID on each block.

GPU Frames. Visualize GPU frame execution over time. If the Flow events option is enabled, an arrow connects each GPU task with the exact CPU task on which it was scheduled.

Click the frame object to view a summary on frame duration, arguments, and related events in the Statistics pane below.

File operations overview. Click the colored circle to review the details of the file operation in the Statistics pane below. For your convenience, the displayed file path is inverted to show the filename first.

Metrics lanes. Review the collected data for the metrics selected in the profile for this analysis or from the System Analyzer window.

CPU Frame. Analyze the CPU frame execution over time. If you add any markers to the source code, your Google trace displays the marked code blocks with the user-defined label. This enables you to analyze the execution order and performance of specific parts of your application.

In the image above, you can see two CPU Frame blocks. One was calculated by Intel GPA at trace capture time, the other is user-defined. To understand the origin of the block, you can click the object and review the details in the Statistics pane below. For each object, the source domain is listed in the Category section.

 

Main Thread. Explore your application performance for graphics-related low-level API calls and user tasks created with the Intel® ITT API.

The context switch bar above the task visualization indicates the execution status. The green color of the bar indicates that the corresponding tasks were actively executing on the CPU. You can click a section of the bar to view detailed information on the execution status.

The vertically aligned colored blocks represent parent-child relations between tasks. If you click a task, you can review the task details in the Statistics pane below:

  • Title - the name of the task.
  • Category - source domain.
  • Start - the time passed from the beginning of the trace cut before the task execution started.
  • Wall Duration - the total execution time of the selected task, including child processes, if any.
  • CPU Duration - the time during which the task was actively executing on the CPU.
  • Self Time - the task execution time excluding child task execution.
  • CPU Self Time - the task execution time on the CPU, excluding child task execution.

 

CPU Queue. Review the GPU queue generation on the CPU side. If the Flow events option is enabled, an arrow connects each GPU task with the exact CPU task on which it was scheduled.

 

Statistics pane. When you click an object on the timeline, you can see the detailed information about the object in this pane.

 

For more complete information about compiler optimizations, see our Optimization Notice.