Date: Wed, 23 Oct 1996 15:10:00 -0700 (PDT) From: "Steven G. Kargl" <sgk@troutmask.apl.washington.edu> To: terry@lambert.org (Terry Lambert) Cc: freebsd-current@freebsd.org Subject: Re: Is profiling code broken? Message-ID: <199610232210.PAA07504@troutmask.apl.washington.edu> In-Reply-To: <199610232122.OAA10485@phaeton.artisoft.com> from Terry Lambert at "Oct 23, 96 02:22:19 pm"
next in thread | previous in thread | raw e-mail | index | archive | help
According to Terry Lambert:
> > program t
> > implicit none
> > double precision x, f
> > integer i, j
> > external f
> >
> > do 1 i =1, 1000
> > do 2 j = 1, 1000
> > x = f(dble(i))
> > 2 continue
> > 1 continue
> > end
>
> I would expect the second do loop to be removed by a decent optimizer;
> ij is not a dependent variable. Maybe the back end changed to 2.7.2 or
> the default optimization flags have changed in the compilation script?
Okay, it was a simple example program. On my code with 5k lines that
can take an hour or more to run, I still get no profiling information
other than the number of times a function (subroutine) has been called.
Also, in the profile output included in the original message, you'll
see that the f() function was call 1000000 time as expected.
[Terry's remaining analysis deleted]
>
> > Additionally, I noticed that the profiled libraries built during a
> > make world use a -p flag instead of -pg. Why? We do not currently
> > have prof(1) in the source tree.
>
> g is debugging information; you don't need it for profiling?
>
man gcc
...
-pg Generate extra code to write profile information
suitable for the analysis program gprof.
...
If you compile with -p alone, gcc (or ld) tries to link in /usr/lib/mrt0.o
which does not exist on my FreeBSD system.
--
Steve
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199610232210.PAA07504>