RE: Does clock() work?
On 08 January 2008 17:57, Norton Allen wrote: Is this a known problem? Do others get this result, or do I have a corrupted library? Reproduces here. I'll take a look into it later tonight if nobody beats me to it. cheers, DaveK -- Can't think of a witty .sigline today -- Unsubscribe info: http://cygwin.com/ml/#unsubscribe-simple Problem reports: http://cygwin.com/problems.html Documentation: http://cygwin.com/docs.html FAQ: http://cygwin.com/faq/
Re: Does clock() work?
Mr Webber wrote: CLOCKS_PER_SEC is a machine dependent macro, but not so machine dependent to recognize that my 32-bit windows box has dual processors. Not useful for benchmarking, is it. It's not quite clear to me why multiple processors would affect the interpretation of CLOCKS_PER_SEC, or why such a simple model would not work in a single-threaded app for basic benchmarking. I'm not talking about a utility to launch commercial apps (which might be multithreaded, etc.), just: * record the current time * do something single-threaded * record the current time and calculate elapsed time clock is not the way to go. It is a crude estimation of processor time. On regular UNIX times(2) is the function to use -- cygwin does not seem to have it. Any other suggestions for timing resolution better than one second on cygwin? -Norton -- Unsubscribe info: http://cygwin.com/ml/#unsubscribe-simple Problem reports: http://cygwin.com/problems.html Documentation: http://cygwin.com/docs.html FAQ: http://cygwin.com/faq/
Re: Does clock() work?
On 2008-01-08 17:57Z, Norton Allen wrote:
[snip code that times this inner loop:]
for ( i = 0; i 8; i++ ) {
sleep(1);
cur_time = clock();
printf( clock() = %ld\n, cur_time );
I would expect the clock() values to increase by approximately 1000 on
each iteration. (Yes, the sleep() seems to be working, as the lines come
out at about 1 Hz.)
I get similar results with the same program. According to
C99 7.23.2.1/2,
The clock function determines the processor time used.
so I'd guess that sleep() is consuming only wall-clock time.
Here's your program with an inner loop that consumes cycles:
/tmp[0]$cat clock_test.c
#include time.h
#include unistd.h
#include stdio.h
#include limits.h
int main( int argc, char **argv ) {
clock_t cur_time, cps = CLOCKS_PER_SEC;
int i, j;
volatile unsigned int v;
printf( CLOCKS_PER_SEC = %ld\n, cps );
for ( i = 0; i 8; i++ ) {
for ( j = 0; j INT_MAX / 10; j++ ) {
v++;
}
cur_time = clock();
printf( clock() = %ld\n, cur_time );
}
return 0;
}
/tmp[0]$gcc -o clock_test.exe clock_test.c
/tmp[0]$./clock_test
CLOCKS_PER_SEC = 1000
clock() = 437
clock() = 859
clock() = 1265
clock() = 1687
clock() = 2093
clock() = 2515
clock() = 2937
clock() = 3343
--
Unsubscribe info: http://cygwin.com/ml/#unsubscribe-simple
Problem reports: http://cygwin.com/problems.html
Documentation: http://cygwin.com/docs.html
FAQ: http://cygwin.com/faq/
RE: Does clock() work?
CLOCKS_PER_SEC is a machine dependent macro, but not so machine dependent to
recognize that my 32-bit windows box has dual processors. Not useful for
benchmarking, is it.
clock is not the way to go. It is a crude estimation of processor time. On
regular UNIX times(2) is the function to use -- cygwin does not seem to have
it.
I don't know if this helps, but:
My cygwin reports
$ ./chktime
CLOCKS_PER_SEC = 1000
clock() = 31
clock() = 31
clock() = 31
clock() = 31
clock() = 31
clock() = 31
clock() = 31
clock() = 31
My 64-bit Fedora/Linux reports
# ./chktime
CLOCKS_PER_SEC = 100
clock() = 0
clock() = 0
clock() = 0
clock() = 0
clock() = 0
clock() = 0
clock() = 0
clock() = 0
-Original Message-
From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of
Norton Allen
Sent: Tuesday, January 08, 2008 12:57 PM
To: cygwin@cygwin.com
Subject: Does clock() work?
I am trying to write a benchmark application, and figured I'd use
clock() for sub-second resolution timing, but I got non-sensical results. I
check the cygwin archives, but the only mention I saw was that clock()
didn't work on Win98. Here's my test code, chktime.c:
#include time.h
#include unistd.h
#include stdio.h
int main( int argc, char **argv ) {
clock_t cur_time, cps = CLOCKS_PER_SEC;
int i;
printf( CLOCKS_PER_SEC = %ld\n, cps );
for ( i = 0; i 8; i++ ) {
sleep(1);
cur_time = clock();
printf( clock() = %ld\n, cur_time );
}
return 0;
}
and here's the output I get:
Cygwin ./chktime
CLOCKS_PER_SEC = 1000
clock() = 171
clock() = 171
clock() = 171
clock() = 171
clock() = 171
clock() = 171
clock() = 171
clock() = 171
Cygwin
I would expect the clock() values to increase by approximately 1000 on each
iteration. (Yes, the sleep() seems to be working, as the lines come out at
about 1 Hz.)
Is this a known problem? Do others get this result, or do I have a corrupted
library?
-Norton Allen
--
Unsubscribe info: http://cygwin.com/ml/#unsubscribe-simple
Problem reports: http://cygwin.com/problems.html
Documentation: http://cygwin.com/docs.html
FAQ: http://cygwin.com/faq/
--
Unsubscribe info: http://cygwin.com/ml/#unsubscribe-simple
Problem reports: http://cygwin.com/problems.html
Documentation: http://cygwin.com/docs.html
FAQ: http://cygwin.com/faq/
Re: Does clock() work?
Greg Chicares wrote: I get similar results with the same program. According to C99 7.23.2.1/2, The clock function determines the processor time used. so I'd guess that sleep() is consuming only wall-clock time. Ah, good catch. Thank you. Now, I am definitely interested in wall clock elapsed time. Is there anything available that will give me real time at resolution greater than one second? -Norton -- Unsubscribe info: http://cygwin.com/ml/#unsubscribe-simple Problem reports: http://cygwin.com/problems.html Documentation: http://cygwin.com/docs.html FAQ: http://cygwin.com/faq/
Re: Does clock() work?
On 2008-01-08 19:15Z, Norton Allen wrote:
Now, I am definitely interested in wall clock elapsed time. Is there
anything available that will give me real time at resolution greater
than one second?
/tmp[0]$cat clock_test.c
#include time.h
#include unistd.h
#include stdio.h
#include sys/time.h // gettimeofday()
int main( int argc, char **argv ) {
clock_t cur_time, cps = CLOCKS_PER_SEC;
int i;
struct timeval x;
printf( CLOCKS_PER_SEC = %ld\n, cps );
for ( i = 0; i 8; i++ ) {
sleep(1);
cur_time = clock();
printf( clock() = %ld\n, cur_time );
gettimeofday(x, 0);
printf( gettimeofday() = %ld\n, 100 * x.tv_sec + x.tv_usec );
}
return 0;
}
/tmp[0]$gcc -o clock_test.exe clock_test.c
/tmp[0]$./clock_test
CLOCKS_PER_SEC = 1000
clock() = 30
gettimeofday() = 210033718
clock() = 30
gettimeofday() = 211033718
clock() = 30
gettimeofday() = 212033718
clock() = 30
gettimeofday() = 213033718
clock() = 30
gettimeofday() = 214033718
clock() = 30
gettimeofday() = 215033718
clock() = 30
gettimeofday() = 216033718
clock() = 30
gettimeofday() = 217033718
--
Unsubscribe info: http://cygwin.com/ml/#unsubscribe-simple
Problem reports: http://cygwin.com/problems.html
Documentation: http://cygwin.com/docs.html
FAQ: http://cygwin.com/faq/
Re: Does clock() work?
Norton Allen wrote: Greg Chicares wrote: I get similar results with the same program. According to C99 7.23.2.1/2, The clock function determines the processor time used. so I'd guess that sleep() is consuming only wall-clock time. Ah, good catch. Thank you. Now, I am definitely interested in wall clock elapsed time. Is there anything available that will give me real time at resolution greater than one second? I found gettimeofday() that seems to do the trick. -N -- Unsubscribe info: http://cygwin.com/ml/#unsubscribe-simple Problem reports: http://cygwin.com/problems.html Documentation: http://cygwin.com/docs.html FAQ: http://cygwin.com/faq/
RE: Does clock() work?
Took another look and found that times(2), though not documented, is available in Cygwin (as a macro in sys/times.h). Try it. You should be able to get real granular time with it, since it also returns a clock_t, without massaging the data returned with any magic CONSTANTS that vary from mach to mach, skewing the results. -Original Message- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of Norton Allen Sent: Tuesday, January 08, 2008 1:40 PM To: Mr Webber; cygwin@cygwin.com Subject: Re: Does clock() work? Mr Webber wrote: CLOCKS_PER_SEC is a machine dependent macro, but not so machine dependent to recognize that my 32-bit windows box has dual processors. Not useful for benchmarking, is it. It's not quite clear to me why multiple processors would affect the interpretation of CLOCKS_PER_SEC, or why such a simple model would not work in a single-threaded app for basic benchmarking. I'm not talking about a utility to launch commercial apps (which might be multithreaded, etc.), just: * record the current time * do something single-threaded * record the current time and calculate elapsed time clock is not the way to go. It is a crude estimation of processor time. On regular UNIX times(2) is the function to use -- cygwin does not seem to have it. Any other suggestions for timing resolution better than one second on cygwin? -Norton -- Unsubscribe info: http://cygwin.com/ml/#unsubscribe-simple Problem reports: http://cygwin.com/problems.html Documentation: http://cygwin.com/docs.html FAQ: http://cygwin.com/faq/
RE: Does clock() work?
On 08 January 2008 19:35, Greg Chicares wrote: [ TEST CASE ] /tmp[0]$./clock_test CLOCKS_PER_SEC = 1000 clock() = 30 gettimeofday() = 210033718 clock() = 30 gettimeofday() = 211033718 clock() = 30 gettimeofday() = 212033718 clock() = 30 gettimeofday() = 213033718 clock() = 30 gettimeofday() = 214033718 clock() = 30 gettimeofday() = 215033718 clock() = 30 gettimeofday() = 216033718 clock() = 30 gettimeofday() = 217033718 clock_setres might help with that. (Disclaimer: Unverified statement). cheers, DaveK -- Can't think of a witty .sigline today -- Unsubscribe info: http://cygwin.com/ml/#unsubscribe-simple Problem reports: http://cygwin.com/problems.html Documentation: http://cygwin.com/docs.html FAQ: http://cygwin.com/faq/
