Measure time

Measure time

[Serguei I. Ivantsov]

Hello!

Is there any function for high precision time measuring.
time() returns only in second. I need nanoseconds.

--
 Regards,
  Serguei I. Ivantsov
   GSC Game World

Re: Measure time

[Der Herr Hofrat]

> Hello!
> 
> Is there any function for high precision time measuring.
> time() returns only in second. I need nanoseconds.
>
you can directly read the TSC but that will not realy give you nanoseconds
resolution as the actual read access even on a PIII/1GHz is going to take
up to a few 100 nanoseconds, and depending on what you want to time
stamp the overall jitter of that code can easaly be in the
range of a microsecond. 

There are some hard-realtime patches to the Linux kernel that will
allow time precission of aprox. 1us (the TSC has a precission of 32ns)
but I don't think you can get below that without dedicated hardware.

for RTLinux check at ftp://ftp.rtlinux.org/pub/rtlinux/ 

hofrat

Re: Measure time

[Richard B. Johnson]

On Wed, 8 May 2002, Serguei I. Ivantsov wrote:

> Hello!
> 
> Is there any function for high precision time measuring.
> time() returns only in second. I need nanoseconds.
> 
> --
>  Regards,
>   Serguei I. Ivantsov
>    GSC Game World
> 

gettimeofday() returns seconds/microseconds. If you need nanoseconds,
you can read CPU clock cycles in Intel machines with (assembly) rdtsc
instructions and convert, based upon the CPU clock.

If you really need nanosecond resolution in code that may be
interrupted or preempted at any time, you are in a world of hurt.
I suggest you review the requirement. You may need a TMS320C30 or
similar DSP standing alone.

Cheers,
Dick Johnson

Penguin : Linux version 2.4.18 on an i686 machine (797.90 BogoMips).

                 Windows-2000/Professional isn't.

Re: Measure time

[Serguei I. Ivantsov]

Is there any function like GetTickCount() in M$ Win32 that retrieves time in
milliseconds?

--
 Regards,
  Serguei I. Ivantsov

----- Original Message -----
From: "Der Herr Hofrat" <der.herr@mail.hofr.at>
To: "Serguei I. Ivantsov" <administrator@svitonline.com>
Cc: <linux-gcc@vger.kernel.org>; <linux-kernel@vger.kernel.org>
Sent: Wednesday, May 08, 2002 3:00 PM
Subject: Re: Measure time


> > Hello!
> >
> > Is there any function for high precision time measuring.
> > time() returns only in second. I need nanoseconds.
> >
> you can directly read the TSC but that will not realy give you nanoseconds
> resolution as the actual read access even on a PIII/1GHz is going to take
> up to a few 100 nanoseconds, and depending on what you want to time
> stamp the overall jitter of that code can easaly be in the
> range of a microsecond.
>
> There are some hard-realtime patches to the Linux kernel that will
> allow time precission of aprox. 1us (the TSC has a precission of 32ns)
> but I don't think you can get below that without dedicated hardware.
>
> for RTLinux check at ftp://ftp.rtlinux.org/pub/rtlinux/
>
> hofrat

Re: Measure time

[Nate]

With you can get system time and then use the millisecond member in Win32...

<code>

SYSTEMTIME systm;

GetSystemTime(&systm);

/*
Can access members of the structure

systm.wMilliseconds
systm.wSecond
systm.wMinute
systm.wHour

... and yet more members ...
*/

</code>

~ Nate




----- Original Message -----
From: "Serguei I. Ivantsov" <admin@gsc-game.kiev.ua>
To: "Der Herr Hofrat" <der.herr@mail.hofr.at>
Cc: <linux-gcc@vger.kernel.org>; <linux-kernel@vger.kernel.org>
Sent: Wednesday, May 08, 2002 8:46 AM
Subject: Re: Measure time


> Is there any function like GetTickCount() in M$ Win32 that retrieves time
in
> milliseconds?
>
> --
>  Regards,
>   Serguei I. Ivantsov
>
> ----- Original Message -----
> From: "Der Herr Hofrat" <der.herr@mail.hofr.at>
> To: "Serguei I. Ivantsov" <administrator@svitonline.com>
> Cc: <linux-gcc@vger.kernel.org>; <linux-kernel@vger.kernel.org>
> Sent: Wednesday, May 08, 2002 3:00 PM
> Subject: Re: Measure time
>
>
> > > Hello!
> > >
> > > Is there any function for high precision time measuring.
> > > time() returns only in second. I need nanoseconds.
> > >
> > you can directly read the TSC but that will not realy give you
nanoseconds
> > resolution as the actual read access even on a PIII/1GHz is going to
take
> > up to a few 100 nanoseconds, and depending on what you want to time
> > stamp the overall jitter of that code can easaly be in the
> > range of a microsecond.
> >
> > There are some hard-realtime patches to the Linux kernel that will
> > allow time precission of aprox. 1us (the TSC has a precission of 32ns)
> > but I don't think you can get below that without dedicated hardware.
> >
> > for RTLinux check at ftp://ftp.rtlinux.org/pub/rtlinux/
> >
> > hofrat
>
> -
> To unsubscribe from this list: send the line "unsubscribe linux-gcc" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>

Re: Measure time

[Calin A. Culianu]

On Wed, 8 May 2002, Der Herr Hofrat wrote:

> > Hello!
> >
> > Is there any function for high precision time measuring.
> > time() returns only in second. I need nanoseconds.
> >
> you can directly read the TSC but that will not realy give you nanoseconds
> resolution as the actual read access even on a PIII/1GHz is going to take
> up to a few 100 nanoseconds, and depending on what you want to time
> stamp the overall jitter of that code can easaly be in the
> range of a microsecond.
>
> There are some hard-realtime patches to the Linux kernel that will
> allow time precission of aprox. 1us (the TSC has a precission of 32ns)
> but I don't think you can get below that without dedicated hardware.
>
> for RTLinux check at ftp://ftp.rtlinux.org/pub/rtlinux/

I recommend RTAI instead.  It's more feature-rich, and works with newer
kernels.. the url is http://www.aero.polimi.it/~rtai/, or
http://www.rtai.org.

-Calin

>
> hofrat
> -
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at  http://www.tux.org/lkml/
>

Re: Measure time

[george anzinger]

"Serguei I. Ivantsov" wrote:
> 
> Is there any function like GetTickCount() in M$ Win32 that retrieves time in
> milliseconds?

The kernel provides gettimeofday() which give micro seconds AND is
usually quite accurate.  On machines of about 800 MHZ and better you
can, sometimes, even get the same value on back to back readings.

The high-res-timers patch (see sig. below) implements the POSIX clocks
and timers which return values in nanoseconds, but the resolution, due
to jitter and such is still in the range of micro seconds.

If, on the other hand, you are trying to measure execution time of some
task that blocks during that time, you are in a world of hurt.  The
kernel allocates 1/HZ chunks of elapsed time to what ever task it finds
running at the 1/HZ tick.  There is NO attempt by the kernel to refine
this measurement.

George
> 
> --
>  Regards,
>   Serguei I. Ivantsov
> 
> ----- Original Message -----
> From: "Der Herr Hofrat" <der.herr@mail.hofr.at>
> To: "Serguei I. Ivantsov" <administrator@svitonline.com>
> Cc: <linux-gcc@vger.kernel.org>; <linux-kernel@vger.kernel.org>
> Sent: Wednesday, May 08, 2002 3:00 PM
> Subject: Re: Measure time
> 
> > > Hello!
> > >
> > > Is there any function for high precision time measuring.
> > > time() returns only in second. I need nanoseconds.
> > >
> > you can directly read the TSC but that will not realy give you nanoseconds
> > resolution as the actual read access even on a PIII/1GHz is going to take
> > up to a few 100 nanoseconds, and depending on what you want to time
> > stamp the overall jitter of that code can easaly be in the
> > range of a microsecond.
> >
> > There are some hard-realtime patches to the Linux kernel that will
> > allow time precission of aprox. 1us (the TSC has a precission of 32ns)
> > but I don't think you can get below that without dedicated hardware.
> >
> > for RTLinux check at ftp://ftp.rtlinux.org/pub/rtlinux/
> >
> > hofrat
> 
> -
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at  http://www.tux.org/lkml/

-- 
George Anzinger   george@mvista.com
High-res-timers:  http://sourceforge.net/projects/high-res-timers/
Real time sched:  http://sourceforge.net/projects/rtsched/
Preemption patch: http://www.kernel.org/pub/linux/kernel/people/rml

RE: Measure time

[Simon Butcher]

Hi,

ftime() will return milliseconds, but it's considered an obsolete function.
You could use gettimeofday() (as Richard Johnson suggested) to get
microseconds and divide them to get milliseconds, although I don't know how
time critical your routines are.

If you're still looking for nanoseconds, I'm told you can use
clock_gettime() but it's still quite unavailable (I've never seen it myself,
yet).. however even if it was available you possibly wouldn't get a very
high resolution from it with current systems..

 - Simon

> -----Original Message-----
> From: linux-gcc-owner@vger.kernel.org
> [mailto:linux-gcc-owner@vger.kernel.org]On Behalf Of Serguei I. Ivantsov
> Sent: Thursday, 9 May 2002 01:47
> To: Der Herr Hofrat
> Cc: linux-gcc@vger.kernel.org; linux-kernel@vger.kernel.org
> Subject: Re: Measure time
>
>
> Is there any function like GetTickCount() in M$ Win32 that
> retrieves time in
> milliseconds?
>
> --
>  Regards,
>   Serguei I. Ivantsov
>
> ----- Original Message -----
> From: "Der Herr Hofrat" <der.herr@mail.hofr.at>
> To: "Serguei I. Ivantsov" <administrator@svitonline.com>
> Cc: <linux-gcc@vger.kernel.org>; <linux-kernel@vger.kernel.org>
> Sent: Wednesday, May 08, 2002 3:00 PM
> Subject: Re: Measure time
>
>
> > > Hello!
> > >
> > > Is there any function for high precision time measuring.
> > > time() returns only in second. I need nanoseconds.
> > >
> > you can directly read the TSC but that will not realy give you
> nanoseconds
> > resolution as the actual read access even on a PIII/1GHz is
> going to take
> > up to a few 100 nanoseconds, and depending on what you want to time
> > stamp the overall jitter of that code can easaly be in the
> > range of a microsecond.
> >
> > There are some hard-realtime patches to the Linux kernel that will
> > allow time precission of aprox. 1us (the TSC has a precission of 32ns)
> > but I don't think you can get below that without dedicated hardware.
> >
> > for RTLinux check at ftp://ftp.rtlinux.org/pub/rtlinux/
> >
> > hofrat
>
> -
> To unsubscribe from this list: send the line "unsubscribe linux-gcc" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>
>

Re: Measure time

[Der Herr Hofrat]

> 
> Hi,
> 
> ftime() will return milliseconds, but it's considered an obsolete function.
> You could use gettimeofday() (as Richard Johnson suggested) to get
> microseconds and divide them to get milliseconds, although I don't know how
> time critical your routines are.
> 
> If you're still looking for nanoseconds, I'm told you can use
> clock_gettime() but it's still quite unavailable (I've never seen it myself,
> yet).. however even if it was available you possibly wouldn't get a very
> high resolution from it with current systems..
>
clock_gettime() is available in the hard realtime extensions like RTLinux .
The clock resolution is limited to 32ns though - and atleast on X86 I don't
think there is a way to get below that.

hofrat 

Re: Measure time

[george anzinger]

Der Herr Hofrat wrote:
> 
> >
> > Hi,
> >
> > ftime() will return milliseconds, but it's considered an obsolete function.
> > You could use gettimeofday() (as Richard Johnson suggested) to get
> > microseconds and divide them to get milliseconds, although I don't know how
> > time critical your routines are.
> >
> > If you're still looking for nanoseconds, I'm told you can use
> > clock_gettime() but it's still quite unavailable (I've never seen it myself,
> > yet).. however even if it was available you possibly wouldn't get a very
> > high resolution from it with current systems..
> >
> clock_gettime() is available in the hard realtime extensions like RTLinux .
> The clock resolution is limited to 32ns though - and atleast on X86 I don't
> think there is a way to get below that.

The high-res-timers patch provides clock_gettime() with resolution to
the TSC increment.  But you need to understand that this is a system
call which can take on the order of 1000 or these units.  Add in a
little cach hit/ miss and interrupt randomness and well...

Still, there it is.  Check out the web site below.

-- 
George Anzinger   george@mvista.com
High-res-timers:  http://sourceforge.net/projects/high-res-timers/
Real time sched:  http://sourceforge.net/projects/rtsched/
Preemption patch: http://www.kernel.org/pub/linux/kernel/people/rml