[prev in list] [next in list] [prev in thread] [next in thread] 

List:       oprofile-list
Subject:    Re: [PATCH 0/4] ocount: A new oprofile event counting tool
From:       Maynard Johnson <maynardj () us ! ibm ! com>
Date:       2013-07-18 19:17:22
Message-ID: 51E83F42.5000603 () us ! ibm ! com
[Download RAW message or body]

On 06/25/2013 08:56 AM, Maynard Johnson wrote:
> To:  OProfile community members
> ANNOUNCEMENT: ocount: A new oprofile event counting tool

The ocount tool has been pushed upstream and is now ready for use.  Please give it a \
try.  Let me know what you think.

For the record, the following patches that were posted to the list have been pushed:
  - Jun 25: [PATCH 1/4] Add 'check_count' parameter to parse_events function
  - Jun 25: [PATCH 2/4] Various utility routines needed by new ocount tool
  - Jun 25: [PATCH 3/4] New ocount tool and associated ocount_counter classes
  - Jun 25: [PATCH 4/4] Add manpage for new ocount tool
  - Jun 25: [PATCH] Fix problems in ocount patch set found by Coverity
  - JUl 17: [PATCH] Add ocount information to user manual
  - Jul 18: [PATCH] Post-review fixups for new ocount feature


-Maynard

> 
> This set of 4 patches introduces a new tool to the oprofile suite of tools -- \
> *ocount*. This is the implementation of the design posted on May 30 (subject "[RFC] \
> Design for new oprofile event counting tool"). 
> *Example usage/output*
> Here's the simplest case, letting ocount use the default event for the processor \
> type:  $ ocount ./my_memcpy
> 
> 	Event counts (actual) for /home/maynard/test-stuff/memcpyt:
> 		Event                   Count                    % time enabled
> 		CPU_CLK_UNHALTED        8,667,461,082            100.00
> 
> Here's an example that demonstrates counting multiple events for a multi-process \
> app, displaying the results in "brief" format, where event counts are separated by \
> process versus aggregated:  $ ocount --brief-format --separate-thread -e \
> CPU_CLK_UNHALTED,INST_RETIRED_ANY_P,LOAD_BLOCK:L1D,L2_RQSTS --process-list=32208  \
> ocount: Press Ctl-c or 'kill -SIGINT 8059' to stop counting  ^C
> 	32208,CPU_CLK_UNHALTED,16737269773,50.02
> 	32208,INST_RETIRED_ANY_P,34985891504,75.00
> 	32208,LOAD_BLOCK,119002686,74.98
> 	32208,L2_RQSTS,24188636,75.00
> 	32209,CPU_CLK_UNHALTED,8506847,16.97
> 	32209,INST_RETIRED_ANY_P,6194969,44.71
> 	32209,LOAD_BLOCK,888830,97.40
> 	32209,L2_RQSTS,185016,85.91
> 
> Note that the final value on each output line is the percent of time the event was \
> enabled. This value will be less than 100% when PMU hardware restrictions prevent \
> all specified events being counted simultaneously, thus relying on the kernel to \
> automatically multiplex the events during the run time. The counts are scaled (as \
> described in the ocount manpage), but are, necessarily, only estimates of what the \
> actual count would be had it been enabled 100% of the time. 
> 
> *Patch set description*
> 
> Patch 1/4 makes a minor change to libop/op_parse_event.c to accommodate event \
> specification differences between ocount and the profiling tools (operf and \
> opcontrol). 
> Patch 2/4 ("Various utility routines needed by new ocount tool") accounts for about \
> 1/3 of the code in this patch set, but is really just a re-packaging of several \
> utility routines that are already used by operf.  I put the utility routines into a \
> libpe_utils library.  But to keep this patch set short and sweet, I have not yet \
> changed operf to use the libpe_utils routines, so it still uses its own copies of \
> those utilities (found in libperf_events/operf_utils.cpp and \
> pe_profiling/operf.cpp).  Once the ocount patch set is committed upstream, I will \
> write another patch that removes operf's copies of those routines and change operf \
> to use the libpe_utils routines. 
> Patch 3/4 ("New ocount tool and associated ocount_counter classes") implements the \
> main executable (ocount.cpp), which parses and validates arguments, and a helper \
> class (ocount_counter.cpp), which is responsible for the interaction with the \
> perf_events kernel subsystem.  As can be seen by the size of ocount_count.cpp -- a \
> mere 615 lines -- implementing event counting was quite simple, especially compared \
> to profiling. 
> Patch 4/4 is the manpage patch for ocount. The oprofile user manual updates will \
> follow in a separate patch.  I wanted to get this implementation out there for \
> people to play with and review. 
> Please give this new tool a try-out and give me your feedback.  Note that because \
> of the new directories and files being added, you must run *./autogen.sh* before \
> the normal ./configure and make. 
> Thanks.
> 
> -Maynard
> 
> 
> ------------------------------------------------------------------------------
> This SF.net email is sponsored by Windows:
> 
> Build for Windows Store.
> 
> http://p.sf.net/sfu/windows-dev2dev
> _______________________________________________
> oprofile-list mailing list
> oprofile-list@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/oprofile-list
> 


------------------------------------------------------------------------------
See everything from the browser to the database with AppDynamics
Get end-to-end visibility with application monitoring from AppDynamics
Isolate bottlenecks and diagnose root cause in seconds.
Start your free trial of AppDynamics Pro today!
http://pubads.g.doubleclick.net/gampad/clk?id=48808831&iu=/4140/ostg.clktrk
_______________________________________________
oprofile-list mailing list
oprofile-list@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/oprofile-list


[prev in list] [next in list] [prev in thread] [next in thread]