profiling wierdness

[Dave Steffen]

Hi Folks,

First for the trivia: 

platform is a dual Pentium-II system running Linux (Redhat 5.1)
compiler is egcs-1.1b (Reading specs from /opt/egcs/lib/gcc-lib/i686-pc-linux-g
nu/egcs-2.91.57/specs gcc version egcs-2.91.57 19980901 (egcs-1.1 release))

Now for the problem:

	I've got some numerical code I'm trying (desperately) to speed
up. I'm compiling with 

	c++ -ansi -pg ....

and (right now) am using no optimization, so I can tell what
improvements I'm getting because of improving the algorithm.

	So I run my program:
helicon: kubo -XYHxyh -e ".1" -O "-5 5 .1" -P .66
	And I profile:
helicon: gprof kubo | c++filt > kubo.prof
	And I get
  %   cumulative   self              self     total           
 time   seconds   seconds    calls  us/call  us/call  name    
 56.86      0.29     0.29 19000000     0.02     0.02  double 
dot_product<Sparse_
vector_STLvec<double>, double>(Sparse_vector_STLvec<double> const &, double 
cons
t *)
 15.69      0.37     0.08 19001940     0.00     0.00  vector<ai<int, double>, 
__default_alloc_template<true, 0> >::end(void) const

	and so forth, for a total run time of .51 seconds. Which I
don't believe, so I run my program again, and profile again:

helicon: kubo -XYHxyh -e ".1" -O "-5 5 .1" -P .66
helicon: gprof kubo | c++filt > kubo.prof

	This time I get  %   cumulative   self              self     total           
 time   seconds   seconds    calls  us/call  us/call  name    
 66.67      0.02     0.02        9  2222.22  2222.22  C_matrix<double>
::set(double const &)
 33.33      0.03     0.01     1440     6.94     6.94  ai<int, double>::ai(int 
co nst &, double const &)

	and so forth, for for a total run time of .03 secons, which I
don't believe either. Note that the #1 user of time is a different
function. So I repeat the process; this time I get

  %   cumulative   self              self     total           
 time   seconds   seconds    calls  us/call  us/call  name    
 44.21      0.42     0.42 19000000     0.02     0.04  double 
dot_product<Sparse_
vector_STLvec<double>, double>(Sparse_vector_STLvec<double> const &, double 
const *)
 15.79      0.57     0.15 30600640     0.00     0.00  ai<int, double>
::index(void) const

	and so forth, for a total run time of .95 seconds.

Hmm...

	Explicity deleting 'gmon.out' between runs doesn't seem to
help; on three runs, gprof gives me total run times of .36, .00
(apparently nothing executed?), and .01.

	And in case you're wondering, 'time'ing the run gives results
like "0.11user 47.19system 0:47.30elapsed 99%CPU", and these numbers
are very consistent. Also, the code executes correctly and generates
identical output for all the above runs.

	Compiling with "-p" instead of "-pg" gives similar results:
.99,.93, and .01 seconds for 3 runs. Compiling with "-profile" results
in a seg fault when I execute the program (not when compiling, though;
the compilation appears to be successful.)

	SO: I'm very confused. Does anyone know what's going on? Is
there any way for me to get reasonably accurate profiling information?

Any help would be greatly appreciated. Thanks.