Very Computer

by **Mark R. Ludw** » Fri, 20 Mar 1992 12:33:27

I have the desire to determine where an application is spending its
CPU time. The standard tools have helped me find the routines, but
what I really need to know is which statements are executing the most.

I have the same application on the following platforms. While I'd
prefer a tool which works on more than one, I'd be satisfied if it
worked on any of the first three.

HP 9000/720; HP-UX A.08.07 A
IBM RISC System/6000; AIX 3.2
Apollo DN10000; Domain/OS 10.3
Personal Iris; IRIX 3.3.1
Sun-4/260; SunOS 4.1.1b
VAX 3600; Ultrix 4.2

Thanks in advance for any tips/pointers. I'll summarize on request.$$
--

"Seize the day next week at the latest." -- Boynton

by **M.C.Whead** » Wed, 25 Mar 1992 23:58:00

Quote:>I have the desire to determine where an application is spending its
>CPU time. The standard tools have helped me find the routines, but
>what I really need to know is which statements are executing the most.

We market a tool called ``rtp'' which shows you where an application
is spending its time on a per-routine and per-line basis - it does
this in real time, whilst the target program is running. It runs
under X on VAXstations (ultrix) and under X or SunView on the Sun3 or
Sun4. Email me your surface mail address if you would like more details.
--
__
/ ) _/_
/--< . . __ /
/___/_(_/_/ _<__ Mark Wheadon.

by **Van Snyd** » Sun, 05 Apr 1992 03:16:44

Quote:>My post expressed the desire to determine where an application is
>spending its CPU time. The standard tools have helped me find the
>routines, but what I really need to know is which statements are
>executing the most.

...
On my PC, Lahey Computer Systems is my favorite Fortran vendor. They
built a profiler from their de*. Their de* has hooks for
each line, entry and exit from a subprogram, conditions, ..., just
like xdb (or dbx, depending on vendor). If the underlying machine has
a sufficiently accurate clock, one can build a profiler from a
de* by using the same machanism as "break on line", "break on
entry" and "break on exit." I don't understand why these things
aren't more widely available.
--

by **Dionys** » Sat, 11 Apr 1992 21:14:42

>My post expressed the desire to determine where an application is
>spending its CPU time.
>----------------------------------------------------------------------

> The VAX is not amongst your top 3, but Ultrix has a
>good tool called pixie.
>[[ Peter and I went back and forth a couple times before I realized that
> this tool, pixie, turns out to be only on RISC/Ultrix. It is
> apparently not on the VAX architecture. What added to the
> confusion is that DEC's numbering scheme is not by itself unique.
> Peter referred to his DEC 3100 which is decidedly different from
> our VAX 3100. ]]

Ok, I did this. I have a program the does a 'system()' call, works
with out pixie.

I run pixie on the executable, produce exectuable.pixie, and
executable.Addrs. Everything appears ok up to here.

Then I run the pixie executable.

Everytime I does a 'system()' call, the pixie executable get a SEGV
_dwmult (it does not give a file or line number).

Is there *any* way to get pixie to do debugging on this? Am I doing
something wrong?

The machine is a DECstation 5000/something, running 4.2.

thanks
tim

Very Computer

Profiling C and Fortran at the statement level

Profiling C and Fortran at the statement level

Profiling C and Fortran at the statement level

Profiling C and Fortran at the statement level

Profiling C and Fortran at the statement level