Analyzers
Community support for Analyzers (Intel VTune™ Profiler, Intel Advisor, Intel Inspector)
4963 Discussions

How to collect the detailed analysis of advanced-hotspots results

Ayam
Beginner
486 Views

Hello,

I am running the application for the advance hotspot analysis using intel vtune. On GUI, this analysis gives me information like Function/Call Stack, CPU Time by Utilization and CPI Rate. I want to get the exact information but using command-line. If I use command

/opt/intel/vtune_amplifier_xe/bin64/amplxe-cl -report hotspots -r r006ah/

the information I get are the columns of Function, Module and CPU Time:Self but CPI Rate column is missing. Can you please give me any idea how to extract CPI Rate using commandline.

Appreciate your help.

Regards,

0 Kudos
1 Solution
Peter_W_Intel
Employee
486 Views

No. There is no call stack info if your don't add "-knob collection-detail=stack-sampling", top-down report only displays function's self-time without callers' info.

You still need to use hw-events report to calculate CPI value, top-down report don't provide instructions retired counts.

 

View solution in original post

0 Kudos
4 Replies
Peter_W_Intel
Employee
486 Views

There is no direct CPI value in report, if you do "amplxe-cl -R hw-events -r r00?ah", You can calculate CPI for specific hot function by using formula, CPU_CLK_UNHALTED.THREAD / INST_RETIRED.ANY 

 

0 Kudos
Ayam
Beginner
486 Views

Thank you Peter Wang, that will work too.

Appreciate your help.

 

0 Kudos
Ayam
Beginner
486 Views

One more question, why total number of function stack using command "/opt/intel/vtune_amplifier_xe/bin64/amplxe-cl -report top-down -r r003ah/" are different then the function calculated using  /opt/intel/vtune_amplifier_xe/bin64/amplxe-cl -report hw-events -r r003ah/.

Can I make them consistent? What I mean is; the function that is taking more time i need to get the CPI rate of that function.

0 Kudos
Peter_W_Intel
Employee
487 Views

No. There is no call stack info if your don't add "-knob collection-detail=stack-sampling", top-down report only displays function's self-time without callers' info.

You still need to use hw-events report to calculate CPI value, top-down report don't provide instructions retired counts.

 

0 Kudos
Reply