Browse thread
[Caml-list] Profiling a function execution
[
Home
]
[ Index:
by date
|
by threads
]
[ Message by date: previous | next ] [ Message in thread: previous | next ] [ Thread: previous | next ]
[ Message by date: previous | next ] [ Message in thread: previous | next ] [ Thread: previous | next ]
| Date: | -- (:) |
| From: | Daniel_Bünzli <daniel.buenzli@e...> |
| Subject: | Re: [Caml-list] Profiling a function execution |
Le 25 nov. 03, à 19:05, Xavier Leroy a écrit :
> I haven't looked at other Unix
> kernels, but I suspect that the additional precision made possible by
> the getrusage() syscall is simply not exploited.
I have done tests using the C code appended to this email.
Indeed with the Darwin Kernel 7.0.0 (MacOS X), you don't get extra
precision with the figures returned by getrusage(). I was mistaken by
the potential of the data structure described in the man page (there's
no info about the actual granularity there).
You can remove my wish (id=1937) for getrusage bindings from your
database.
> your question begs another
> question: why would you need to do distinguish native code from
> bytecode?
I just wanted a quick way to distinguish, on output, my bytecode
profiles from native ones. This can of course be handled very easily
with a preprocessor, I was just wondering if I had missed a flag in the
style of Sys.interactive.
> We're working hard on removing the last discrepancies
> between the two compilers...
One could argue that since there exists some discrepancies such a flag
should exist. However I doubt its usefulness in common practice (which
is why I didn't file a feature wish for that one). Furthermore, if it
doesn't exist, removing the last discrepancies becomes a must...
Thanks for your answers,
Daniel
--- test.c ---
#include <limits.h>
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include <time.h>
#include <sys/types.h>
#include <sys/times.h>
#include <sys/time.h>
#include <sys/resource.h>
#define NUM_OPS ULONG_MAX / 100
void idle (void)
{
unsigned long i;
int op;
for (i = 0; i < NUM_OPS; i++) op = 1+1;
}
void diff_times (void)
{
struct tms t1, t2;
times(&t1);
idle();
times(&t2);
printf("times: %fs user %fs sys\n",
(double)(t2.tms_utime-t1.tms_utime) / CLK_TCK,
(double)(t2.tms_stime-t1.tms_stime) / CLK_TCK);
}
void diff_getrusage(void)
{
struct rusage t1, t2;
getrusage(RUSAGE_SELF, &t1);
idle();
getrusage(RUSAGE_SELF, &t2);
printf("getrusage: %fs user %fs sys\n",
(double)t2.ru_utime.tv_sec + (double)t2.ru_utime.tv_usec / 1e6 -
((double)t1.ru_utime.tv_sec + (double)t1.ru_utime.tv_usec / 1e6),
(double)t2.ru_stime.tv_sec + (double)t2.ru_stime.tv_usec / 1e6 -
((double)t1.ru_stime.tv_sec + (double)t1.ru_stime.tv_usec / 1e6));
}
int main(int argc, char **argv)
{
diff_times();
diff_getrusage();
return 0;
}
-------------------
To unsubscribe, mail caml-list-request@inria.fr Archives: http://caml.inria.fr
Bug reports: http://caml.inria.fr/bin/caml-bugs FAQ: http://caml.inria.fr/FAQ/
Beginner's list: http://groups.yahoo.com/group/ocaml_beginners