Community
cancel
Showing results for 
Search instead for 
Did you mean: 
Prasanna_B_
Beginner
141 Views

amplxe-cl inclusive and exclusive HW event data

Jump to solution

Dear Vtune experts,

In the vtune GUI documentation, I found that Top-Down Tree window displays inclusive and exclusive performance data from the perspective of the function call stacks during execution. I would like to obtain the inclusive and exclusive HW event data via command line. Currently, I use the following command and it generates report on exclusive data but I want to generate inclusive data on the HW counters.

HW_COUNTERS="CPU_CLK_UNHALTED,INSTRUCTIONS_EXECUTED,VPU_INSTRUCTIONS_EXECUTED,VPU_ELEMENTS_ACTIVE,EXEC_STAGE_CYCLES,DATA_READ_OR_WRITE,DATA_READ_MISS_OR_WRITE_MISS,L2_DATA_READ_MISS_CACHE_FILL,L2_DATA_WRITE_MISS_CACHE_FILL,L2_DATA_READ_MISS_MEM_FILL,L2_DATA_WRITE_MISS_MEM_FILL,L2_VICTIM_REQ_WITH_DATA,HWP_L2MISS,L1_DATA_HIT_INFLIGHT_PF1,SNP_HITM_L2,DATA_PAGE_WALK,LONG_DATA_PAGE_WALK"

export I_MPI_FABRICS=shm
export I_MPI_PIN_DOMAIN=omp
export I_MPI_MIC=1
export OMP_NUM_THREADS=120
export I_MPI_DEBUG=5

amplxe-cl -collect-with runsa-knc -knob event-config=$HW_COUNTERS --target-duration-type=veryshort --search-dir all:rp=./ mpirun -f mpi_hosts -np 1 bash -c "ulimit -s unlimited && ./app.exe" > res-01rpn-120tpr.log

amplxe-cl -report hotspots -r r000runsa_knc -format text -group-by function -csv-delimiter comma -report-output result.csv

Thanks a lot

 

0 Kudos
1 Solution
Peter_W_Intel
Employee
141 Views

However in amplxe-cl, when I generate the report, I only get Self values for HW event

Current product U16 has no "Total by HW event type", see release notes.

VTune™ Amplifier XE data collection on Intel® Xeon Phi™ coprocessor (codename: Knights Corner) currently is limited to hardware event-based sampling data collected from target units (200179057)

  • No information on function call stacks is recorded during collection. However, you may mistake partial call chains appearing in result Groups for real call stack information. These partial chains are the result of inline function information in debug symbol tables and can be ignored.

View solution in original post

6 Replies
Peter_W_Intel
Employee
141 Views

If you want to see HW event data, please use -

amplxe-cl -report hw-events -r r000runsa_knc -format text -group-by function -csv-delimiter comma -report-output result.csv

Prasanna_B_
Beginner
141 Views

Thanks Peter.

Sorry, I forgot to include the last line of my script:

amplxe-cl -report hw-events -group-by function -r r000runsa_knc -format csv -csv-delimiter comma -report-output hw_results.csv

However, with this command, I get exclusive HW event data for each function. I would like to generate inclusive HW event. Is there a flag that I have to add/enable/disable during data collection/report generation?

Thanks for your help.

 


 

Peter_W_Intel
Employee
141 Views

I don't know which inclusive event count cannot be displayed, if you use "hw-events" instead of "hotspots"...in report.

What is VTune version you are using? Latest product is U16. Can you please tell me which event data cannot be displayed. Thank you.

Regards, Peter 

Prasanna_B_
Beginner
141 Views

Peter,

I have U 15.

In the vtune GUI documentation i saw the following:

Top-down Tree window displays hotspot functions in the call tree, performance metrics for a function only (Self value) and for a function and its children together (Total value).

However in amplxe-cl, when I generate the report, I only get Self values for HW event

For example, VPU_ELEMENTS_ACTIVE:Self,Hardware Event Count:CPU_CLK_UNHALTED:Self,..etc. There is no "Total".

Do you think its a version issue?

Thanks

Peter_W_Intel
Employee
142 Views

However in amplxe-cl, when I generate the report, I only get Self values for HW event

Current product U16 has no "Total by HW event type", see release notes.

VTune™ Amplifier XE data collection on Intel® Xeon Phi™ coprocessor (codename: Knights Corner) currently is limited to hardware event-based sampling data collected from target units (200179057)

  • No information on function call stacks is recorded during collection. However, you may mistake partial call chains appearing in result Groups for real call stack information. These partial chains are the result of inline function information in debug symbol tables and can be ignored.

View solution in original post

Prasanna_B_
Beginner
141 Views

Peter,

Thanks for clarification.

Reply