Intel® VTune™ Profiler supports collecting performance data on QNX* target systems.
Data collection is possible via command line interface from a host system running Windows* or Linux* to the target QNX system. The collected traces are transferred to the host system via ethernet and stored for review. After collection, the performance results can be imported and viewed in the Intel VTune Profiler user interface.
The target collector can be integrated into the target QNX image during the image build process and requires only 1 MB of space on the target file system. Because the traces are transferred to the host system, collection can be done on target systems with limited storage capacity or with read-only file systems.
Host System: Linux* or Windows* system with QNX BSP and VTune Profiler installed
Target System: Supported processor with QNX7 operating with instrumental kernel, connected to the host system via ethernet. Supported processors include Intel® Pentium®, Intel® Celeron®, or Intel Atom® processors formerly code named Apollo Lake or Intel Atom® processors formerly code named Denverton.
Turn off firewall restrictions for network connections between the host system and target system
Set up Your System
Complete the following steps on your host and target system to install collectors and enable performance analysis using Intel VTune Profiler:
Ensure that the host system is connected to the target QNX system via ethernet and log in to the target QNX system using a command window.
Make the <install-dir>/target/qnx_x86_64/bin64/sep file on the host system available on the target QNX system by copying, mounting a network share, or integrating it into the target image.
On the host system, launch the VTune Profiler user interface, click New Project, specify a project name, and click Create Project.
Click Configure Analysis, select local host in the WHERE pane, and click Search Binaries.
In the Binary/Symbol Search window, browse to the location of the kernel and application target modules on the host system, and click OK.
Analysis is run using collectors previously installed on the target QNX system and a command invoked on the host Windows or Linux system. All result files are saved to the host system.
On the target QNX system, run the following command: <sep-dir>/sep
Where <sep-dir> is the location where the sep file was copied. The target collector loads and waits for the host system to connect.
On the host system, run one of the following analysis commands.
Hotspots with call stacks: <install-dir>/bin64/sep -start -d <duration> -target-ip <target-ip-address> -target-port 9321 -lbr call_stack -out <filename>.tb7
/opt/intel/vtune_profiler/bin64/sep -start -d 60 -target-ip 12.345.67.89 -target-port 9321 -lbr call_stack -out hotspots_callstacks.tb7NOTE:
Call stacks are hardware based and limited to a depth of 16 frames. Due to hardware limitations, the depth of the captured call stack can be less than 16 frames.
Custom CPU events: <install-dir>/bin64/sep.exe -start -d <duration> -target-ip <target-ip-address> -target-port 9321 -ec "<event-list>" -out <filename>.tb7
/opt/intel/vtune_profiler/bin64/sep.exe -start -d 60 -target-ip 12.345.67.89 -target-port 9321 -ec "MEM_LOAD_UOPS_RETIRED.DRAM_HIT,MEM_LOAD_UOPS_RETIRED.HITM,MEM_LOAD_UOPS_RETIRED.L2_HIT" -out custom.tb7
See the Sampling Enabling Product User's Guide for more information.
After collection begins, run the application on the target QNX system or ensure that it is already running. The analysis collects system-wide data. Collection stops automatically when the specified duration is complete.
After collection is complete, stop the application on the target QNX system if it is not already finished.
View and Interpret Results
After collection is complete, the *.tb7 result file is available on the host system.
On the host system, import the *.tb7 file into the previously created project.
Switch to the Hotspots viewpoint and review the performance data collected.
If you collected hotspots data, begin with the Summary window in the Hotspots viewpoint. The Top Hotspots list shows the top 5 functions that occupied the most CPU time. Double-click a function to be taken to the Bottom-up window where you can see aggregated performance data and a timeline showing activity over the entire collection. For more information, see Hotspots View.
If you collected CPU event data, begin with the Microarchitecture Exploration viewpoint. For more information, see Microarchitecture Exploration View.