[comp.sys.acorn] pmp - Poor Mans Profiler (UNOFFICAL version :-))

[Graham Toal]

Archive-name: pmp/gthack

[Serendipitously, our mailer broke when I tried to post this earlier, so
I've had a spare 10 minutes in which to 'tidy' the code up a little :-)]

This was Dan Bernstein's Poor Mans Profiler -- a utility to find out how often
particular lines of your source have been executed.  My apologies to Dan for
beating it to death.  It now works only on ANSI-compatible systems, like
the Acorn Archimedes for which this posting is intended. (comp.sys.acorn)
[We've been banned from posting these nasty long sources on c.s.a until we
can forge enough votes for a comp.sources.acorn :-(]

%------------------------------------------- h.pmplib
#ifndef PMPLIB_H
#define PMPLIB_H 1
#include <stdio.h>
struct _p {long s; long l; char *f; struct _p *p;};
extern struct _p *_pc; FILE *_pm(struct _p *t, FILE *fi);
#define _pr(_ps) (_ps.p = _pc, _pc = &_ps)
#define P do {static struct _p p = {0,__LINE__,__FILE__,0};\
if (!((p.s)++)) _pr(p);} while(0)
#define pmp(fn) fclose(_pm(_pc, fopen(fn,"w")))
#endif
%------------------------------------------- c.pmplib
#include "pmplib.h"
struct _p *_pc = 0;FILE *_pm(struct _p *t, FILE *fi) {return((t && fi) ?
 _pm(t->p, fi), fprintf(fi,"%s, %ld: %ld\n",t->f,t->l,t->s), fi : fi);}
%------------------------------------------- c.test
#include "pmplib.h"

int main()
{
 int i;
 int j;

 /* P: produce profiling information. */
 /* pmp("test-pmp"): save pmp's results in test-pmp. */

 for (i = 0;i < 1000;i++)
  {
   P;
   for (j = 0;j < 1000;j++)
    {
     P;
     if (i % 2)
       P;
    }
  }
 pmp("test-pmp");
}
%------------------------------------------- sample-out
c.test, 19: 1000
c.test, 22: 1000000
c.test, 24: 500000
%------------------------------------------- README
This is pmp, the poor man's profiler.

pmp gives you statement counts.

pmp has one big advantage over every other profiling system I've seen:
it doesn't force optimization off. You can place profiling marks
anywhere in your code; you only hurt optimization where you put the
marks. So if you're sick of having your program slowed down dramatically
just because the profiler sticks its nose into your inner loops, pmp
should be a welcome relief.

Of course, pmp is also the smallest profiler known to man. [*now* :-)]

Other than that, it's pretty poor.

This was once pmp version 1.0, 9/13/90, by Daniel J. Bernstein.
pmp is public domain.

This version is gtoal at ed.ac.uk's hack of the alt.sources posting --
the timing stuff (which used SIGVTALARM) has been removed, and I've
(ahem) shortened the code somewhat :-)

To use pmp, just cc test.c pmplib.c
If you have an editor which lets you step through error listings (Emacs? Twin?)
in synch with your source, use it to step through the test-pmp output file,
otherwise merge the profile info with the source by hand (or hack up a prog
to do it).


Good luck!
PS Dan - I removed a few lines to make it print the profile in forward
(= reverse to you) order to support the editor-stepping feature... :-)
%------------------------------------------- END
-- 
(* Posted from tharr.uucp - Public Access Unix - +44 (234) 261804 *)