Subject: gprof userlevel profiling on strongarm?
To: None <port-arm32@netbsd.org>
From: Jonathan Stone <jonathan@narrowbandaudio.com>
List: port-arm32
Date: 12/23/2000 01:21:34
Does profiling work on arm32?

I'm trying to profile a local app.  According to gprof, the gmon.out
file only has ticks for one-quarter to one-half of the run-time
reported by time(1), even though I've built all my own .po files
with -pg and linked against profiling libraries.

On some profiling runs, some functions (which I know are heavy
time-sinks, from instrumenting them with gettimeofday() before
and after each call) show up with zero ticks.

Last, mcount() and _mcount() seem to account for up to one-half
the observed ticks.

Building the same app on an i386 (with -pg, and linking with profiling
libs) seems to work fine, so I'm fairly sure I'm doing the right things.

I have very dim memories of a 16-bit counter for pc-sample regions in
old 4BSD gprof; could that be a contributing factor?



any pointers appreciated
--Jonathan