VTune Amplifier Hotspot_Analysis reports "Failed to finalize results"

[Petr_K_]

Hi,

we are trying to evaluate the Intel VTune Amplifier XE as possible standard performance analyzer tool for our development team

the build we are trying to use is

(build 206420) Copyright 2009-2011 Intel Corporation. All rights reserved.

The problem:

and the Highly Accurate CPU data mode does not seem to work. Everytime the analysis runs it returns with the error message: "Failed to finalize results".

The Analyzer is running as Administrator, the Collect Highly Accurate CPU data checkbox is checked.

I tried eeverything - rebooting the machine,restarting both the analyzer process and the VTune process.

Command line:

amplxe-cl -collect hotspots -knob accurate-cpu-time-detection=true -no-follow-child -target-duration-type=short -no-allow-multiple-runs -no-analyze-system -data-limit=100 -slow-frames-threshold=40 -fast-frames-threshold=100 --target-process star-ccm+.exe

The database has been cleared, elapsed time is 0.062 seconds.
Raw data has been loaded to the database, elapsed time is 0.004 seconds.
Finalizing the result took 0.069 seconds.

The lightweight analysis seems to work, it just does not provide anything the free CodeAnalyst does not already give us.

Windows 7 / 64 bit / Intel Xeon X5570 2.93 GHz - 2 Processors workstation
24 GB of memory

The application is compiled with Intel C++ 11.1.065

[Rob5]

Petr,

Is the behavior specific to the application you are analyzing or does the behavior also occur when performing hotspots analysis on the Tachyon sample application that is included with Intel VTune Amplifier XE?

- Rob

[Peter_W_Intel]

Was it possible that star-ccm+.exe worked on system mode? Hotspots (or concurrency, or locksandwaits) collector works on user-mode data collection. That means, user mode data collector can only capture the performance data of user applications.

You can try to attachto the process of amplxe-gui.exe (tool itself), it works. But you will fail to attach the process of explorer.exe which is the system application.

Youmay use HW event based sampling collector (including lightweight-hotspots) to profile system applications.

Regards, Peter

[Petr_K_]

not specific - tachyon behaves the same way - I made some progress though - it seems that when the application is launched from inside the vtune it works, but when the vtune attaches to the running process either by pid or by application name it does not - the collection seems to run but comes back empty

given our application startup complexity it is more practical to attach than to launch from vtune (MPI HPC app)

is this behavior intended - I could not find anything related in the documentation

Thanks

Petr

[Petr_K_]

no, it is a just a regular application - no special privileges, running as a regular currently logged in user

[Peter_W_Intel]

If this is a regular application, but it may depend on user's environment variables, paths?

If using VTune to launch application succeeded, I suspect that we have torun applicationand run VTune in same environment. I mean:
1. Start application in console (cmd), without using VTune
2. Use hotspots analysis, attach to target process - in same console. What happens?

Regards, Peter

[Rob5]

Petr,

Thank you for the additional information. The article Is 'Collect Highly Accurate CPU Time' Option Required? may be of interest. As you mentioned, to collect highly accurate CPU time, run with administrative privileges.

Additionally, the on-line user guide provides information about Collecting Highly Accurate CPU Time Datahere.

- Rob

[Rob5]

Petr,

Thank you for the feedback. I will research and make an enhancement request for additional detail to be added about this to the command-line portion of the on-line users guide.

- Rob