Re: [bgl-discuss] PAPI counters

[Haihang You <you@xxxxxxxxxx>]

You can not count PAPI_FMA_INS, PAPI_FML_INS, or PAPI_FAD_INS together. 
Any two of them will cause conflict if I remember correctly.

Haihang

On Mon, 10 Apr 2006, Jeff Nucciarone wrote:

> Alan Gray wrote:
> > Hi Jeff
> >
> > At least the following counters are available:
> >
> > PAPI_L3_TCM  -  Level 3 total cache misses
> > PAPI_L3_LDM  -  Level 3 load misses
> > PAPI_L3_STM  -  Level 3 store misses
> > PAPI_FMA_INS  -  Floating Multiply Add instructions completed
> > PAPI_TOT_CYC  -  Total cycles
> > PAPI_L2_DCH  -  L2 data cache hits
> > PAPI_L2_DCA  -  L2 data cache accesses
> > PAPI_L3_TCH  -  L3 total cache hits
> > PAPI_FML_INS  -  Floating Multiply instructions
> > PAPI_FAD_INS  -  Floating Add instructions
> >
> > In addition, there are another 5 derived events unique to the BG/L 
> > architecture:
> >
> > PAPI_BGL_OED     - FPU Oedipus operations
> > PAPI_BGL_TS_32B  - 32B torus packets sent in any direction
> > PAPI_BGL_TS_FULL - UPC cycles any link has no tokens
> > PAPI_BGL_TR_DPKT - No data packets sent on any tree channel
> > PAPI_BGL_TR_FULL - UPC cycles any tree receiver is full
>
> I resubmitted my application with the following:
>
> > COUNTER1=GET_TIME_OF_DAY:COUNTER2=PAPI_TOT_CYC:COUNTER3=PAPI_FMA_INS:
> > COUNTER4=PAPI_L3_TCM:COUNTER5=PAPI_FP_STALL:COUNTER6=PAPI_FML_INS:
> > COUNTER7=PAPI_FAD_INS:COUNTER8=PAPI_BGL_OED
>
>
> Unfortunatly when it ran I received the following messages in the output file 
> and unfortunately no counter info beyond MULTI__GET_TIME_OF_DAY:
>
> Error adding Papi events!
> The events you have chosed conflict.
> This could be a limit on either the number of events
> allowed by this hardware, or the combination of events chosen.
> The papi layer calls are being disabled!
> Deleting papiMCL in position: 1
> Setting papi flags in counterUsed array to false ...
> counterUsed[{1,2,3,5,6,7}] is now false ...
> Finished disabling papi layer calls!
>
> Obviously there is some conflict in my choices.  If I were a betting person 
> I'd guess it was PAPI_BGL_OED.
>
> Thoughts? Advice?
>
> thanks,
>
> --Jeff
>
>
> --
> Jeff Nucciarone     nucci@xxxxxxx          http://www.personal.psu.edu/nucci
>  Senior Research Programmer, High Performance Computing Group, ITS/ASET
>  The Pennsylvania State University
>                               "Don't just do it........  do it right."

*************************************************************************
Haihang You, Research Associate 
Innovative Computing Laboratory               E-mail: you@xxxxxxxxxx
Department of Computer Science                Phone : (865)974-5433(O)
University of Tennessee                       Fax   : (865)974-8296
Suite 203, 1122 Volunteer Blvd 
Knoxville TN 37996-3450                       http://www.cs.utk.edu/~you
*************************************************************************

- --------------------------------------------------------------------
To add or remove yourself from this mailing list, use the 'notifyme'
command on any BGL machine. To remove: notifyme -n, to add: notifyme -y.