Hi,
Yes there is a major difference between unhalted_reference_cycles in fixed
and generic counters. This is a bug on all Intel processors. Or you can call
this a feature!
Turns out when measured in generic counters it measures at bus clock speed
whereas in the fixed counter it counts at the max clock speed of the CPU.
That is why on some processors I have called it :BUS.
On Dec 18, 2009 6:11 PM, "Min Lim" <my...@renci.org> wrote:
Hello,
I found that there exists inconsistency between two different versions
of the same event : fixed counter and generic counter. As I know, the
following 3 events are using fixed counters on Nehalem:
INSTRUCTIONS_RETIRED,UNHALTED_CORE_CYCLES,UNHALTED_REFERENCE_CYCLES
And, their generic counter versions are named respectively as:
INST_RETIRED:ANY_P,CPU_CLK_UNHALTED:THREAD_P,CPU_CLK_UNHALTED:REF_P
Am I right? If so, their numbers should be same. But they are
significantly different in my tests. Here is the output:
[my...@bill stats]$ pfmon -u -k
-eINSTRUCTIONS_RETIRED,INST_RETIRED:ANY_P,UNHALTED_CORE_CYCLES,CPU_CLK_UNHALTED:THREAD_P,UNHALTED_REFERENCE_CYCLES,CPU_CLK_UNHALTED:REF_P
--system-wide --aggregate --session-timeout=1 --cpu-list=0,1,2,3,4,5,6,7
<session to end in 1 seconds>
3712730498 INSTRUCTIONS_RETIRED
3712731630 INST_RETIRED:ANY_P
807137982 UNHALTED_CORE_CYCLES
2954627584 CPU_CLK_UNHALTED:THREAD_P
807126428 UNHALTED_REFERENCE_CYCLES
134300459 CPU_CLK_UNHALTED:REF_P
The result shows that the numbers of the events (UNHALTED_CORE_CYCLES
and UNHALTED_REFERENCE_CYCLES) are significantly different by their
versions. When counting the fixed events via MSR interface directly,
UNHALTED_CORE_CYCLES and UNHALTED_REFERENCE_CYCLES are close to
CPU_CLK_UNHALTED:THREAD_P.
Do you know where the inconsistency comes from or am I missing something?
Thanks
------------------------------------------------------------------------------
This SF.Net email is sponsored by the Verizon Developer Community
Take advantage of Verizon's best-in-class app development support
A streamlined, 14 day to market process makes app distribution fast and easy
Join now and get one step closer to millions of Verizon customers
http://p.sf.net/sfu/verizon-dev2dev
_______________________________________________
perfmon2-devel mailing list
perfmon2-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/perfmon2-devel
------------------------------------------------------------------------------
This SF.Net email is sponsored by the Verizon Developer Community
Take advantage of Verizon's best-in-class app development support
A streamlined, 14 day to market process makes app distribution fast and easy
Join now and get one step closer to millions of Verizon customers
http://p.sf.net/sfu/verizon-dev2dev
_______________________________________________
perfmon2-devel mailing list
perfmon2-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/perfmon2-devel
