From: Thomas Koenig <tkoenig@netcologne.de>
To: Nicolas Koenig <koenigni@student.ethz.ch>,
Jerry DeLisle <jvdelisle@charter.net>
Cc: GCC-Fortran-ML <fortran@gcc.gnu.org>
Subject: Re: [Patch, fortran] PR35339 Optimize implied do loops in io statements
Date: Mon, 29 May 2017 06:40:00 -0000 [thread overview]
Message-ID: <f68088b9-2eb9-7a14-2984-77a6b7fd9835@netcologne.de> (raw)
In-Reply-To: <4bf3850e-393c-6ef8-3895-4a17931c1b9b@student.ethz.ch>
Hi Nicolas,
> gcc@dcm-linux:~/pr/35339> gfortran -O3 perf_test.f90
> gcc@dcm-linux:~/pr/35339> time ./a.out
>
> real 0m1.879s
> user 0m1.838s
> sys 0m0.040s
> gcc@dcm-linux:~/pr/35339> gfortran-7 -O3 perf_test.f90
> gcc@dcm-linux:~/pr/35339> time ./a.out
>
> real 0m1.929s
> user 0m1.901s
> sys 0m0.028s
For formatted I/O, the advantage will not be large because
of the CPU time needed for converting numbers to decimal.
People who care about speed in I/O use unformatted I/O,
and there is a big advantage to using an array slice:
$ cat tst.f90
module x
implicit none
contains
double precision function v2s(v)
integer, dimension(8), intent(in) :: v
v2s = v(8)*1d-3+v(7)+v(6)*60.d0+v(5)*3600.d0
end function v2s
end module x
program main
use x
implicit none
integer, parameter :: n=10**7
real, dimension(n) :: a
real:: t1, t2
integer :: i
integer, dimension(8) :: v1, v2
call random_number(a)
open(unit=10,file="test.dat",form="unformatted")
call cpu_time(t1)
call date_and_time(values=v1)
write (10) (a(i),i=1,n)
call date_and_time(values=v2)
call cpu_time(t2)
print *,"CPU time implied DO loop = ", t2-t1
print *,"Wall time = ", v2s(v2)-v2s(v1)
call cpu_time(t1)
call date_and_time(values=v1)
write (10) a(1:i:1)
call cpu_time(t2)
call date_and_time(values=v2)
print *,"CPU time Array= ", t2-t1
print *,"Wall time = ", v2s(v2)-v2s(v1)
close(10,status="delete")
end program main
$ gfortran -O3 tst.f90
$ ./a.out
CPU time implied DO loop = 0.374823004
Wall time = 0.50600000000122236
CPU time Array= 6.18300140E-02
Wall time = 0.10599999999976717
$
WARNING: multiple messages have this Message-ID
From: Thomas Koenig <tkoenig@netcologne.de>
To: Nicolas Koenig <koenigni@student.ethz.ch>,
Jerry DeLisle <jvdelisle@charter.net>
Cc: GCC-Fortran-ML <fortran@gcc.gnu.org>
Subject: Re: [Patch, fortran] PR35339 Optimize implied do loops in io statements
Date: Mon, 29 May 2017 14:06:00 -0000 [thread overview]
Message-ID: <f68088b9-2eb9-7a14-2984-77a6b7fd9835@netcologne.de> (raw)
Message-ID: <20170529140600.e6VIdTP2MoY52YR_Xf2-_Cu5UGyNWo8NqqI5029ZUhk@z> (raw)
In-Reply-To: <4bf3850e-393c-6ef8-3895-4a17931c1b9b@student.ethz.ch>
Hi Nicolas,
> gcc@dcm-linux:~/pr/35339> gfortran -O3 perf_test.f90
> gcc@dcm-linux:~/pr/35339> time ./a.out
>
> real 0m1.879s
> user 0m1.838s
> sys 0m0.040s
> gcc@dcm-linux:~/pr/35339> gfortran-7 -O3 perf_test.f90
> gcc@dcm-linux:~/pr/35339> time ./a.out
>
> real 0m1.929s
> user 0m1.901s
> sys 0m0.028s
For formatted I/O, the advantage will not be large because
of the CPU time needed for converting numbers to decimal.
People who care about speed in I/O use unformatted I/O,
and there is a big advantage to using an array slice:
$ cat tst.f90
module x
implicit none
contains
double precision function v2s(v)
integer, dimension(8), intent(in) :: v
v2s = v(8)*1d-3+v(7)+v(6)*60.d0+v(5)*3600.d0
end function v2s
end module x
program main
use x
implicit none
integer, parameter :: n=10**7
real, dimension(n) :: a
real:: t1, t2
integer :: i
integer, dimension(8) :: v1, v2
call random_number(a)
open(unit=10,file="test.dat",form="unformatted")
call cpu_time(t1)
call date_and_time(values=v1)
write (10) (a(i),i=1,n)
call date_and_time(values=v2)
call cpu_time(t2)
print *,"CPU time implied DO loop = ", t2-t1
print *,"Wall time = ", v2s(v2)-v2s(v1)
call cpu_time(t1)
call date_and_time(values=v1)
write (10) a(1:i:1)
call cpu_time(t2)
call date_and_time(values=v2)
print *,"CPU time Array= ", t2-t1
print *,"Wall time = ", v2s(v2)-v2s(v1)
close(10,status="delete")
end program main
$ gfortran -O3 tst.f90
$ ./a.out
CPU time implied DO loop = 0.374823004
Wall time = 0.50600000000122236
CPU time Array= 6.18300140E-02
Wall time = 0.10599999999976717
$
next prev parent reply other threads:[~2017-05-29 6:40 UTC|newest]
Thread overview: 23+ messages / expand[flat|nested] mbox.gz Atom feed top
2017-05-27 19:49 Nicolas Koenig
2017-05-28 22:07 ` Jerry DeLisle
2017-05-28 23:32 ` Nicolas Koenig
2017-05-29 6:40 ` Thomas Koenig [this message]
2017-05-29 14:06 ` Thomas Koenig
2017-05-29 15:32 ` Jerry DeLisle
2017-05-29 15:49 Dominique d'Humières
2017-05-29 16:24 ` Nicolas Koenig
2017-05-29 16:51 ` Dominique d'Humières
2017-05-31 15:40 ` Bernhard Reutner-Fischer
2017-05-31 15:49 ` Dominique d'Humières
2017-05-31 19:03 ` Dominique d'Humières
2017-05-31 23:15 ` Nicolas Koenig
2017-06-01 9:31 ` Dominique d'Humières
2017-06-01 14:19 ` Dominique d'Humières
2017-06-01 14:37 ` Dominique d'Humières
2017-06-01 18:26 ` Nicolas Koenig
2017-06-03 13:48 ` Nicolas Koenig
2017-06-03 16:25 ` Jerry DeLisle
2017-06-03 18:12 ` Bernhard Reutner-Fischer
2017-06-05 20:39 ` Nicolas Koenig
2017-06-06 11:05 ` Markus Trippelsdorf
2017-06-07 15:13 ` Renlin Li
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=f68088b9-2eb9-7a14-2984-77a6b7fd9835@netcologne.de \
--to=tkoenig@netcologne.de \
--cc=fortran@gcc.gnu.org \
--cc=jvdelisle@charter.net \
--cc=koenigni@student.ethz.ch \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).