Counting native events

Vincent_B_1 — Mon, 29 Jun 2015 15:08:33 GMT

Hi,

I try to count some performance events of a part of an application written in C.

So far, I have used PAPI to count events. It works fine for preset events. However, when I profile native events, all of them turn out to be translated into the same Event Code : 0x40000022 (an output of papi_avail is below). It makes no sense, but no error occurs when I profile them. What could be wrong ? How could I debug this ?

Also, I've got a list (from the perfmon directory) of events that can be counted for the architecture I work with (Ivy bridge) and the adresses and values of the associated registers. Is there an alternative to PAPI I could use to count a particular event of that list, over a certain part of the program, knowing the name of the event and the registers informations ? Would it work to simply write and read to the corresponding registers manually (for example using pcm-msr) ?

Thanks in advance for your help,

Vincent

$ papi_avail -e LLC_MISSES

Available events and hardware information.

--------------------------------------------------------------------------------

PAPI Version : 5.3.0.0

Vendor string and code : GenuineIntel (1)

Model string and code : Intel(R) Xeon(R) CPU E5-4610 v2 @ 2.30GHz (62)

CPU Revision : 4.000000

CPUID Info : Family: 6 Model: 62 Stepping: 4

CPU Max Megahertz : 2294

CPU Min Megahertz : 2294

Hdw Threads per core : 1

Cores per Socket : 8

Sockets : 4

NUMA Nodes : 4

CPUs per Node : 8

Total CPUs : 32

Running in a VM : no

Number Hardware Counters : 11

Max Multiplex Counters : 64

--------------------------------------------------------------------------------

Event name: LLC_MISSES

Event Code: 0x40000022

Number of Register Values: 0

Description: |Alias for LAST_LEVEL_CACHE_MISSES|

Unit Masks:

Mask Info: |:e=0|edge level (may require counter-mask >= 1)|

Mask Info: |:i=0|invert|

Mask Info: |:c=0|counter-mask in range [0-255]|

Mask Info: |:t=0|measure any thread|

Mask Info: |:u=0|monitor at user level|

Mask Info: |:k=0|monitor at kernel level|

-------------------------------------------------------------------------

avail.c PASSED

Vincent,

Thomas_W_Intel — Mon, 29 Jun 2015 15:38:37 GMT

Vincent,

Unfortunately, I'm can't help you with PAPI but you might want to have a look at Intel Performance Counter Monitor. It can be used to count L3 misses on some systems.

Kind regards

Thomas

topic Counting native events in Software Tuning, Performance Optimization & Platform Monitoring

Counting native events

Vincent,