[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [PATCH v3 1/3] scripts/performance: Add topN_perf.py script
|
From: |
Ahmed Karaman |
|
Subject: |
Re: [PATCH v3 1/3] scripts/performance: Add topN_perf.py script |
|
Date: |
Fri, 26 Jun 2020 16:18:35 +0200 |
On Thu, Jun 25, 2020 at 11:45 AM Aleksandar Markovic
<aleksandar.qemu.devel@gmail.com> wrote:
>
> сре, 24. јун 2020. у 17:32 Ahmed Karaman
> <ahmedkhaledkaraman@gmail.com> је написао/ла:
> >
> > Syntax:
> > topN_perf.py [-h] [-n] <number of displayed top functions> -- \
> > <qemu executable> [<qemu executable options>] \
> > <target executable> [<target execurable options>]
> >
> > [-h] - Print the script arguments help message.
> > [-n] - Specify the number of top functions to print.
> > - If this flag is not specified, the tool defaults to 25.
> >
> > Example of usage:
> > topN_perf.py -n 20 -- qemu-arm coulomb_double-arm
> >
> > Example Output:
> > No. Percentage Name Caller
> > ---- ---------- ------------------------- -------------------------
> > 1 16.25% float64_mul qemu-x86_64
> > 2 12.01% float64_sub qemu-x86_64
> > 3 11.99% float64_add qemu-x86_64
> > 4 5.69% helper_mulsd qemu-x86_64
> > 5 4.68% helper_addsd qemu-x86_64
> > 6 4.43% helper_lookup_tb_ptr qemu-x86_64
> > 7 4.28% helper_subsd qemu-x86_64
> > 8 2.71% f64_compare qemu-x86_64
> > 9 2.71% helper_ucomisd qemu-x86_64
> > 10 1.04% helper_pand_xmm qemu-x86_64
> > 11 0.71% float64_div qemu-x86_64
> > 12 0.63% helper_pxor_xmm qemu-x86_64
> > 13 0.50% 0x00007f7b7004ef95 [JIT] tid 491
> > 14 0.50% 0x00007f7b70044e83 [JIT] tid 491
> > 15 0.36% helper_por_xmm qemu-x86_64
> > 16 0.32% helper_cc_compute_all qemu-x86_64
> > 17 0.30% 0x00007f7b700433f0 [JIT] tid 491
> > 18 0.30% float64_compare_quiet qemu-x86_64
> > 19 0.27% soft_f64_addsub qemu-x86_64
> > 20 0.26% round_to_int qemu-x86_64
> >
> > Signed-off-by: Ahmed Karaman <ahmedkhaledkaraman@gmail.com>
> > ---
> > scripts/performance/topN_perf.py | 142 +++++++++++++++++++++++++++++++
> > 1 file changed, 142 insertions(+)
> > create mode 100755 scripts/performance/topN_perf.py
> >
> > diff --git a/scripts/performance/topN_perf.py
> > b/scripts/performance/topN_perf.py
> > new file mode 100755
> > index 0000000000..d2b939c375
> > --- /dev/null
> > +++ b/scripts/performance/topN_perf.py
> > @@ -0,0 +1,142 @@
> > +#!/usr/bin/env python3
> > +
> > +# Print the top N most executed functions in QEMU using perf.
> > +# Syntax:
> > +# topN_perf.py [-h] [-n] <number of displayed top functions> -- \
> > +# <qemu executable> [<qemu executable options>] \
> > +# <target executable> [<target execurable options>]
> > +#
> > +# [-h] - Print the script arguments help message.
> > +# [-n] - Specify the number of top functions to print.
> > +# - If this flag is not specified, the tool defaults to 25.
> > +#
> > +# Example of usage:
> > +# topN_perf.py -n 20 -- qemu-arm coulomb_double-arm
> > +#
> > +# This file is a part of the project "TCG Continuous Benchmarking".
> > +#
> > +# Copyright (C) 2020 Ahmed Karaman <ahmedkhaledkaraman@gmail.com>
> > +# Copyright (C) 2020 Aleksandar Markovic
> > <aleksandar.qemu.devel@gmail.com>
> > +#
> > +# This program is free software: you can redistribute it and/or modify
> > +# it under the terms of the GNU General Public License as published by
> > +# the Free Software Foundation, either version 2 of the License, or
> > +# (at your option) any later version.
> > +#
> > +# This program is distributed in the hope that it will be useful,
> > +# but WITHOUT ANY WARRANTY; without even the implied warranty of
> > +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
> > +# GNU General Public License for more details.
> > +#
> > +# You should have received a copy of the GNU General Public License
> > +# along with this program. If not, see <https://www.gnu.org/licenses/>.
> > +
> > +import argparse
> > +import os
> > +import subprocess
> > +import sys
> > +
> > +
> > +# Parse the command line arguments
> > +parser = argparse.ArgumentParser(
> > + usage='topN_perf.py [-h] [-n] <number of displayed top functions > --
> > '
> > + '<qemu executable> [<qemu executable options>] '
> > + '<target executable> [<target executable options>]')
> > +
> > +parser.add_argument('-n', dest='top', type=int, default=25,
> > + help='Specify the number of top functions to print.')
> > +
> > +parser.add_argument('command', type=str, nargs='+', help=argparse.SUPPRESS)
> > +
> > +args = parser.parse_args()
> > +
> > +# Extract the needed variables from the args
> > +command = args.command
> > +top = args.top
> > +
> > +# Insure that perf is installed
> > +check_perf = subprocess.run(["which", "perf"], stdout=subprocess.DEVNULL)
> > +if check_perf.returncode:
> > + sys.exit("Please install perf before running the script!")
>
> I would rename "chech_perf" to "check_perf_presence". It is more
> specific, clearer.
>
> > +
> > +# Insure user has previllage to run perf
> > +check_perf_executability = subprocess.run(["perf", "stat", "ls", "/"],
> > + stdout=subprocess.DEVNULL,
> > stderr=subprocess.DEVNULL)
> > +if check_perf_executability.returncode:
> > + sys.exit(
> > +"""
> > +Error:
> > +You may not have permission to collect stats.
> > +
> > +Consider tweaking /proc/sys/kernel/perf_event_paranoid,
> > +which controls use of the performance events system by
> > +unprivileged users (without CAP_SYS_ADMIN).
> > +
> > + -1: Allow use of (almost) all events by all users
> > + Ignore mlock limit after perf_event_mlock_kb without CAP_IPC_LOCK
> > + 0: Disallow ftrace function tracepoint by users without CAP_SYS_ADMIN
> > + Disallow raw tracepoint access by users without CAP_SYS_ADMIN
> > + 1: Disallow CPU event access by users without CAP_SYS_ADMIN
> > + 2: Disallow kernel profiling by users without CAP_SYS_ADMIN
> > +
> > +To make this setting permanent, edit /etc/sysctl.conf too, e.g.:
> > + kernel.perf_event_paranoid = -1
> > +"""
> > +)
>
> Very good.
>
> > +
> > +# Run perf record
> > +perf_record = subprocess.run((["perf", "record"] + command),
> > + stdout=subprocess.DEVNULL,
> > stderr=subprocess.PIPE)
> > +if perf_record.returncode:
> > + os.unlink('perf.data')
> > + sys.exit(perf_record.stderr.decode("utf-8"))
>
> Here, the file "perf.data" will be created in the current working
> directory. If one existed prior to script execution, it will be
> overwritten.
>
> I think such "corruption" of current working directory is not optimal.
> It would be better that the script doesn't touch current working
> directory at all (perhaps user wants to keep perf.data he obtained
> from some experiment in the past.
>
> Therefore, I think it would be better if you specify output of "perf
> report" to be "/tmp/perf.data", not "perf.data", which is the default.
> There is an option of "perf record" to specify the output file:
>
> -o, --output=
> Output file name.
>
> > +
> > +# Save perf report output to perf_report.out
> > +with open("perf_report.out", "w") as output:
> > + perf_report = subprocess.run(
> > + ["perf", "report", "--stdio"], stdout=output,
> > stderr=subprocess.PIPE)
> > + if perf_report.returncode:
> > + os.unlink('perf.data')
> > + output.close()
> > + os.unlink('perf_report.out')
> > + sys.exit(perf_report.stderr.decode("utf-8"))
>
> For similar reasons described above, input file should be
> "/tmp/perf.data". Option of "perf report" for input file:
>
> -i, --input=
> Input file name.
>
> Output file should be "/tmp/perf_report.out", not "perf_report.out".
>
> > +
> > +# Read the reported data to functions[]
> > +functions = []
> > +with open("perf_report.out", "r") as data:
>
> "/tmp/perf_report.out"
>
> > + # Only read lines that are not comments (comments start with #)
> > + # Only read lines that are not empty
> > + functions = [line for line in data.readlines() if line and line[0]
> > + != '#' and line[0] != "\n"]
> > +
> > +# Limit the number of top functions to "top"
> > +number_of_top_functions = top if len(functions) > top else len(functions)
> > +
> > +# Store the data of the top functions in top_functions[]
> > +top_functions = functions[:number_of_top_functions]
> > +
> > +# Print table header
> > +print('{:>4} {:>10} {:<30} {}\n{} {} {} {}'.format('No.',
> > + 'Percentage',
> > + 'Name',
>
> 'Function Name' would be more ergonomic here.
>
> > + 'Caller',
>
> Please replace 'Caller' with 'Invoked by'. 'Caller' implies a function
> that directly calls the function in question. 'Invoked by' avoids such
> confusion, and it just feels more appropriate here.
>
> > + '-' * 4,
> > + '-' * 10,
> > + '-' * 30,
> > + '-' * 25))
> > +
> > +
> > +# Print top N functions
> > +for (index, function) in enumerate(top_functions, start=1):
> > + function_data = function.split()
> > + function_percentage = function_data[0]
> > + function_name = function_data[-1]
> > + function_caller = ' '.join(function_data[2:-2])
>
> function_invoker
>
> > + print('{:>4} {:>10} {:<30} {}'.format(index,
> > + function_percentage,
> > + function_name,
> > + function_caller))
>
> function_invoker
>
> > +
> > +# Remove intermediate files
> > +os.unlink('perf.data')
> > +os.unlink('perf_report.out')
>
> os.unlink('/tmp/perf.data')
> os.unlink('/tmp/perf_report.out')
>
>
> > --
> > 2.17.1
> >
Thanks Mr. Aleksandar. These are really valid points. I'll add these
updates in v4 of this series.
Best regards,
Ahmed Karaman