Analyzers
Talk to fellow users of Intel Analyzer tools (Intel VTune™ Profiler, Intel Advisor)

Zero CPU time when collecting stacks and context switches

Øyvind_J_
Beginner
333 Views

Hi,

I have trouble with the Advanced Hotspots analysis on Win XP 64-bit using Visual studio 2010 with Intel(R) Visual Fortran Composer XE 2013 Update 3. I use VTune Amplifier XE 2013, Update 15 (build 328102).

I am running the parallel nqueens_fortran example. The Basic Hotspots analysis works as expected, and also the Advanced Hotspots analysis when I select only the "Hotspots" radio button.  However, if I select "Hotspots, stacks and context switches" (and leave all other settings as they were), the problem shows up: The application runs fine, and completes in about 6 seconds. Still, VTunes reports the CPU time as zero, both in the summary, and for all functions in the top-down and bottom-up views.

I have tried lowering the sampling interval all the way down to 0.1 ms, and I have tried to run the case through the VTune GUI as administrator. Also, I tried to upgrade from Update 14 to 15, but the problem persists. Below, I have pasted the content of the summary tab and the command line copied from the analysis type tab.

Does anybody have suggestions for what might be the cause of these problems?

Cheers,

Øyvind

 

From the Summary tab:

Elapsed Time:    5.454s
  CPU Time:    0s
  Instructions Retired:    92,025,599,526
  CPI Rate:    0.562
  Wait Rate:    0.006
  CPU Frequency Ratio:    0.000
  Paused Time:    0s
  Wait Time:    0.000s
  Inactive Time:    8.672s
  Overhead Time:    0s
  Spin Time:    0s

[The CPU usage histogram shows only one column at 0 CPU.]

Collection and Platform Info
  Application Command Line:    D:\tmp\nqueens_fortran\nqueens_fortran\vf9\nqueens_parallel\Release\nqueens_parallel.exe 14
  Operating System:    Windows XP Professional x64 Edition
  Computer Name:    XXX.YYY.com
  Result Size:    10 MB
  CPU
    Name:    45nm Intel(R) Core(TM) microarchitecture
    Frequency:    2.8 GHz
    Logical CPU Count:    8

The command line is:

"C:\Program Files (x86)\Intel\VTune Amplifier XE 2013\bin32\amplxe-cl" -collect-with runsa -knob enable-stack-collection=true -knob event-config=CPU_CLK_UNHALTED.CORE:sa=2830000,CPU_CLK_UNHALTED.REF:sa=2830000,INST_RETIRED.ANY:sa=2830000 -app-working-dir D:\tmp\nqueens_fortran\nqueens_fortran\vf9\nqueens_parallel\Release -- D:\tmp\nqueens_fortran\nqueens_fortran\vf9\nqueens_parallel\Release\nqueens_parallel.exe 14

0 Kudos
1 Solution
Kirill_R_Intel
Employee
333 Views

You seem to have old CPU based on "Penryn" micro architecture. These CPUs have problem with PMU event "CPU_CLK_UNHALTED.REF" - in your result with stacks it is zero (you can check by switching viewpoint to "Hardware Event Counts"). This event is used to calculate CPU time. Collection without stack uses a bit different technique to count the event, so it works. This issue can be solved only by upgrading to another CPU.

Regards,
Kirill

 

View solution in original post

0 Kudos
4 Replies
Kirill_R_Intel
Employee
333 Views

Øyvind, can you attach the result to the thread? And please provide exact version of your OS, that shows up when you run "msinfo32" command?

0 Kudos
Øyvind_J_
Beginner
333 Views

Thanks,

I attached the result folder for a run with stack information (the problematic one), and also the result folder for a run without stack information for reference.

msinfo32 System summary:

OS Name    Microsoft(R) Windows(R) XP Professional x64 Edition
Version    5.2.3790 Service Pack 2 Build 3790
Other OS Description     Not Available
OS Manufacturer    Microsoft Corporation
System Name    xxxxx
System Manufacturer    Hewlett-Packard
System Model    HP xw8600 Workstation
System Type    x64-based PC
Processor    EM64T Family 6 Model 23 Stepping 6 GenuineIntel ~2833 Mhz
Processor    EM64T Family 6 Model 23 Stepping 6 GenuineIntel ~2833 Mhz
Processor    EM64T Family 6 Model 23 Stepping 6 GenuineIntel ~2833 Mhz
Processor    EM64T Family 6 Model 23 Stepping 6 GenuineIntel ~2833 Mhz
Processor    EM64T Family 6 Model 23 Stepping 6 GenuineIntel ~2833 Mhz
Processor    EM64T Family 6 Model 23 Stepping 6 GenuineIntel ~2833 Mhz
Processor    EM64T Family 6 Model 23 Stepping 6 GenuineIntel ~2833 Mhz
Processor    EM64T Family 6 Model 23 Stepping 6 GenuineIntel ~2833 Mhz
BIOS Version/Date    Hewlett-Packard 786F5 v01.35, 28.04.2009
SMBIOS Version    2.5
Windows Directory    C:\WINDOWS
System Directory    C:\WINDOWS\system32
Boot Device    \Device\HarddiskVolume1
Locale    United States
Hardware Abstraction Layer    Version = "5.2.3790.3959 (srv03_sp2_rtm.070216-1710)"
User Name   xxx\yyy
Time Zone    W. Europe Standard Time
Total Physical Memory    12 287,05 MB
Available Physical Memory    8,40 GB
Total Virtual Memory    13,50 GB
Available Virtual Memory    10,25 GB
Page File Space    2,00 GB
Page File    C:\pagefile.sys

 

0 Kudos
Kirill_R_Intel
Employee
334 Views

You seem to have old CPU based on "Penryn" micro architecture. These CPUs have problem with PMU event "CPU_CLK_UNHALTED.REF" - in your result with stacks it is zero (you can check by switching viewpoint to "Hardware Event Counts"). This event is used to calculate CPU time. Collection without stack uses a bit different technique to count the event, so it works. This issue can be solved only by upgrading to another CPU.

Regards,
Kirill

 

0 Kudos
Øyvind_J_
Beginner
333 Views

Kirill, thanks for the explanation!

Øyvind

0 Kudos
Reply