gprof question

gprof question

[Jean-Pierre Rosen]

I have a multi-tasks, IO intensive program that I ran under gprof to identify bottlenecks (GNAT 3.13p under Win95).
Although the (real) execution time is about 3mn 30s, gprof shows only 6 samples (at 0.01s. period!).
My explanation is that interrupts (and therefore gprof sampling) are suspended during IO suspension, and that all time spent in IO
procedures escapes analysis. Is this correct ? It would be terribly annoying in my case...

--
---------------------------------------------------------
           J-P. Rosen (rosen@adalog.fr)
Visit Adalog's web site at http://www.adalog.fr

Re: gprof question

[Ted Dennison]

In article <9hpdkf$nhb$1@s1.read.news.oleane.net>, Jean-Pierre Rosen says...
>
>I have a multi-tasks, IO intensive program that I ran under gprof to identify bottlenecks (GNAT 3.13p under Win95).
>Although the (real) execution time is about 3mn 30s, gprof shows only 6 samples (at 0.01s. period!).
>My explanation is that interrupts (and therefore gprof sampling) are suspended during IO suspension, and that all time spent in IO
>procedures escapes analysis. Is this correct ? It would be terribly annoying in my case...


I tried at one point to use gprof on WinNT, but was unable to get any useful
results. If you have better luck, I'd love to hear how you did it.

---
T.E.D.    homepage   - http://www.telepath.com/dennison/Ted/TED.html
          home email - mailto:dennison@telepath.com

Re: gprof question

[Jean-Pierre Rosen]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: text/plain, Size: 578 bytes --]


"Ted Dennison" <dennison@telepath.com> a �crit dans le message news: jI_%6.4357$Kf3.34110@www.newsranger.com...
> I tried at one point to use gprof on WinNT, but was unable to get any useful
> results. If you have better luck, I'd love to hear how you did it.
>
I forgot to mention that I tried it also under Linux, with the same problem. So I think it is a real issue with gprof, not something
specific to Win-whatever.

--
---------------------------------------------------------
           J-P. Rosen (rosen@adalog.fr)
Visit Adalog's web site at http://www.adalog.fr

Re: gprof question

[Ted Dennison]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: text/plain, Size: 896 bytes --]

In article <9hq5q2$85f$1@s1.read.news.oleane.net>, Jean-Pierre Rosen says...
>
>
>"Ted Dennison" <dennison@telepath.com> a �crit dans le message news: jI_%6.4357$Kf3.34110@www.newsranger.com...
>> I tried at one point to use gprof on WinNT, but was unable to get any useful
>> results. If you have better luck, I'd love to hear how you did it.
>>
>I forgot to mention that I tried it also under Linux, with the same problem. 
>So I think it is a real issue with gprof, not something
>specific to Win-whatever.

Well, your problem sounds suspciously close to what I was getting.

I've never tried gprof anywhere else, so its quite possible that my problems
were "pilot error". But I have used several other profilers, so I *ought* to
have been able to figure it out.

---
T.E.D.    homepage   - http://www.telepath.com/dennison/Ted/TED.html
          home email - mailto:dennison@telepath.com

Re: gprof question

[Jean-Pierre Rosen]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: text/plain, Size: 683 bytes --]


"Ted Dennison" <dennison@telepath.com> a �crit dans le message news: j2207.4674$Kf3.38403@www.newsranger.com...
> I've never tried gprof anywhere else, so its quite possible that my problems
> were "pilot error". But I have used several other profilers, so I *ought* to
> have been able to figure it out.
>
I browsed the (former) DejaNews, and this seems to confirm that gprof measures *CPU time* only.
That can be certainly useful for some programs, but not mine. Anyone knows about a profiler that cares about *real* time ?

--
---------------------------------------------------------
           J-P. Rosen (rosen@adalog.fr)
Visit Adalog's web site at http://www.adalog.fr

Re: gprof question

[Ted Dennison]

In article <9hrttg$q32$1@s1.read.news.oleane.net>, Jean-Pierre Rosen says...
>I browsed the (former) DejaNews, and this seems to confirm that gprof measures 
>*CPU time* only. That can be certainly useful for some programs, but not mine.
>Anyone knows about a profiler that cares about *real* time ?

For Win9x? Most PC developers use NT (even when their target is 9x-only) for a
reason. Win9x just doesn't support software development very well. It may turn
out that your only recourse for "wall time" is to do it yourself iteratively
with Ada.Calendar.

---
T.E.D.    homepage   - http://www.telepath.com/dennison/Ted/TED.html
          home email - mailto:dennison@telepath.com

Re: gprof question

[Larry Kilgallen]

In article <9hpdkf$nhb$1@s1.read.news.oleane.net>, "Jean-Pierre Rosen" <rosen@adalog.fr> writes:
> I have a multi-tasks, IO intensive program that I ran under gprof to identify bottlenecks (GNAT 3.13p under Win95).
> Although the (real) execution time is about 3mn 30s, gprof shows only 6 samples (at 0.01s. period!).
> My explanation is that interrupts (and therefore gprof sampling) are suspended during IO suspension, and that all time spent in IO
> procedures escapes analysis. Is this correct ? It would be terribly annoying in my case...

You might consider whether the operating system in question has the
ability to wait for IO in the mode of the caller.  If it does not,
there is no way a user mode interrupt could come while you are
waiting for IO.

But I am not sure I know what "interrupts" are for a user mode
program on Windows or Linux.  On VMS the corresponding thing
might be "AST's" which certainly work when waiting for IO
(unless GNAT does something quite wrong).  Perhaps trying
it on VMS would get you "close-enough" profile results to
the real answers for Windows.