[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20200318152112.GA231037@shwetrath.localdomain>
Date: Wed, 18 Mar 2020 11:21:12 -0400
From: Vijay Thakkar <vijaythakkar@...com>
To: Arnaldo Carvalho de Melo <arnaldo.melo@...il.com>
Cc: Peter Zijlstra <peterz@...radead.org>,
Ingo Molnar <mingo@...hat.com>,
Kim Phillips <kim.phillips@....com>,
Alexander Shishkin <alexander.shishkin@...ux.intel.com>,
Jiri Olsa <jolsa@...hat.com>,
Namhyung Kim <namhyung@...nel.org>,
Martin Liška <mliska@...e.cz>,
Jon Grimm <jon.grimm@....com>, linux-kernel@...r.kernel.org,
linux-perf-users@...r.kernel.org
Subject: Re: [PATCH v5 2/3] perf vendor events amd: add Zen2 events
> floating point:
> fpu_pipe_assignment.total
> [Total number of fp uOps]
>
>
> Metric Groups:
>
I just realized that I did not add back the counters for per pipe total
assignment as I said in the commit message. Patch 3/3 adds the total
uOp assignments per pipe for Zen1 based processors. Although the PPR for
Matisse does not list these counters, I can still sample them on my
Ryzen 3900X system and they seem to report correct numbers. For example,
here is the result of running:
$> perf stat -e r100,r200,r400,r800,rf00 ls
$> Performance counter stats for 'ls':
5,047 r100:u
5,236 r200:u
31,300 r400:u
2,040 r800:u
43,623 fpu_pipe_assignment.total:u
Note that the per pipe total counters add up the overall total, which
makes me think I should submit a v6 of this patch adding the per pipe
totals for zen2 as well, especially since it is mentioned in the commit
message.
-Vijay
Powered by blists - more mailing lists