[Qemu-devel] Patch: cpu-all.h for better cycle counter

[Qemu-devel] Patch: cpu-all.h for better cycle counter

[Todd T. Fries]

This is relative to the 20070319 snapshot.

--- cpu-all.h.orig	Fri Mar 16 18:58:11 2007
+++ cpu-all.h	Tue Mar 20 21:14:10 2007
@@ -1012,13 +1012,22 @@ static inline int64_t cpu_get_real_ticks
 #endif
 }
 #else
-/* The host CPU doesn't have an easily accessible cycle counter.
-   Just return a monotonically increasing vlue.  This will be totally wrong,
-   but hopefully better than nothing.  */
+# warning non-optimized CPU
+#include <sys/time.h>
+#include <time.h>
+
 static inline int64_t cpu_get_real_ticks (void)
 {
-    static int64_t ticks = 0;
-    return ticks++;
+	struct timeval tv;
+	static int64_t i = 0;
+	int64_t j;
+        
+	gettimeofday(&tv, NULL);
+	do {
+	   j = (tv.tv_sec * (uint64_t) 1000000) + tv.tv_usec;
+	} while (i == j);
+	i = j;
+	return j;
 }
 #endif
 
-- 
Todd Fries .. todd@fries.net

 _____________________________________________
|                                             \  1.636.410.0632 (voice)
| Free Daemon Consulting, LLC                 \  1.405.227.9094 (voice)
| http://FreeDaemonConsulting.com             \  1.866.792.3418 (FAX)
| "..in support of free software solutions."  \          250797 (FWD)
|                                             \
 \\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\
                                                 
              37E7 D3EB 74D0 8D66 A68D  B866 0326 204E 3F42 004A
                        http://todd.fries.net/pgp.txt

Re: [Qemu-devel] Patch: cpu-all.h for better cycle counter

[andrzej zaborowski]

Hi,

On 21/03/07, Todd T. Fries <todd@fries.net> wrote:
> This is relative to the 20070319 snapshot.
>
> --- cpu-all.h.orig      Fri Mar 16 18:58:11 2007
> +++ cpu-all.h   Tue Mar 20 21:14:10 2007
> @@ -1012,13 +1012,22 @@ static inline int64_t cpu_get_real_ticks
>  #endif
>  }
>  #else
> -/* The host CPU doesn't have an easily accessible cycle counter.
> -   Just return a monotonically increasing vlue.  This will be totally wrong,
> -   but hopefully better than nothing.  */
> +# warning non-optimized CPU
> +#include <sys/time.h>
> +#include <time.h>
> +
>  static inline int64_t cpu_get_real_ticks (void)
>  {
> -    static int64_t ticks = 0;
> -    return ticks++;
> +       struct timeval tv;
> +       static int64_t i = 0;
> +       int64_t j;
> +
> +       gettimeofday(&tv, NULL);
> +       do {
> +          j = (tv.tv_sec * (uint64_t) 1000000) + tv.tv_usec;
> +       } while (i == j);
> +       i = j;
> +       return j;

Isn't this an infinite loop? gettimeofday() was left out of the loop.

How about "return j + (ticks++)" instead of the loop? If I understand
correctly it may slow things down to below 1Hz.

>  }
>  #endif
>
> --
> Todd Fries .. todd@fries.net
>
>  _____________________________________________
> |                                             \  1.636.410.0632 (voice)
> | Free Daemon Consulting, LLC                 \  1.405.227.9094 (voice)
> | http://FreeDaemonConsulting.com             \  1.866.792.3418 (FAX)
> | "..in support of free software solutions."  \          250797 (FWD)
> |                                             \
>  \\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\
>
>               37E7 D3EB 74D0 8D66 A68D  B866 0326 204E 3F42 004A
>                         http://todd.fries.net/pgp.txt
>
>
>
>

Regards,
Andrew

Re: [Qemu-devel] Patch: cpu-all.h for better cycle counter

[andrzej zaborowski]

On 26/03/07, andrzej zaborowski <balrog@zabor.org> wrote:
> Hi,
>
> On 21/03/07, Todd T. Fries <todd@fries.net> wrote:
> > This is relative to the 20070319 snapshot.
> >
> > --- cpu-all.h.orig      Fri Mar 16 18:58:11 2007
> > +++ cpu-all.h   Tue Mar 20 21:14:10 2007
> > @@ -1012,13 +1012,22 @@ static inline int64_t cpu_get_real_ticks
> >  #endif
> >  }
> >  #else
> > -/* The host CPU doesn't have an easily accessible cycle counter.
> > -   Just return a monotonically increasing vlue.  This will be totally wrong,
> > -   but hopefully better than nothing.  */
> > +# warning non-optimized CPU
> > +#include <sys/time.h>
> > +#include <time.h>
> > +
> >  static inline int64_t cpu_get_real_ticks (void)
> >  {
> > -    static int64_t ticks = 0;
> > -    return ticks++;
> > +       struct timeval tv;
> > +       static int64_t i = 0;
> > +       int64_t j;
> > +
> > +       gettimeofday(&tv, NULL);
> > +       do {
> > +          j = (tv.tv_sec * (uint64_t) 1000000) + tv.tv_usec;
> > +       } while (i == j);
> > +       i = j;
> > +       return j;
>
> Isn't this an infinite loop? gettimeofday() was left out of the loop.
>
> How about "return j + (ticks++)" instead of the loop? If I understand
> correctly it may slow things down to below 1Hz.

(I wanted to say MHz)

>
> >  }
> >  #endif
> >
> > --
> > Todd Fries .. todd@fries.net
> >
> >  _____________________________________________
> > |                                             \  1.636.410.0632 (voice)
> > | Free Daemon Consulting, LLC                 \  1.405.227.9094 (voice)
> > | http://FreeDaemonConsulting.com             \  1.866.792.3418 (FAX)
> > | "..in support of free software solutions."  \          250797 (FWD)
> > |                                             \
> >  \\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\
> >
> >               37E7 D3EB 74D0 8D66 A68D  B866 0326 204E 3F42 004A
> >                         http://todd.fries.net/pgp.txt
> >
> >
> >
> >
>
> Regards,
> Andrew
>

Re: [Qemu-devel] Patch: cpu-all.h for better cycle counter

[Marc Lörner]

On Monday 26 March 2007 00:08, andrzej zaborowski wrote:
> On 26/03/07, andrzej zaborowski <balrog@zabor.org> wrote:
<snip>
> > > +# warning non-optimized CPU
> > > +#include <sys/time.h>
> > > +#include <time.h>
> > > +
> > >  static inline int64_t cpu_get_real_ticks (void)
> > >  {
> > > -    static int64_t ticks = 0;
> > > -    return ticks++;
> > > +       struct timeval tv;
> > > +       static int64_t i = 0;
> > > +       int64_t j;
> > > +
> > > +       gettimeofday(&tv, NULL);
> > > +       do {
> > > +          j = (tv.tv_sec * (uint64_t) 1000000) + tv.tv_usec;
> > > +       } while (i == j);
> > > +       i = j;
> > > +       return j;
> >
> > Isn't this an infinite loop? gettimeofday() was left out of the loop.
> >
> > How about "return j + (ticks++)" instead of the loop? If I understand
> > correctly it may slow things down to below 1Hz.
>
> (I wanted to say MHz)

I dont think so, in the loop "j" is set and the while-condition is "j==i",
so unless "(tv.tv_sec * (uint64_t) 1000000) + tv.tv_usec" always computes the 
same value the do-while block gets only executed once.
>
<snip>

-- 
Freundliche Gruesse,
Marc Lörner

Tel.: +49 (0)9103-715-30046
Fax.: +49 (0)9103-715-271
Internet: http://www.hob.de
E-Mail: marc.loerner@hob.de

HOB GmbH & Co. KG
Schwadermühlstraße 3
90556 Cadolzburg

Geschaeftsfuehrer Klaus Brandstaetter - Franz Wiedenmann

AG Fuerth  HRA 5180
Steuer-Nr. 218/163/00107
USt-ID-Nr. DE 132747002

Komplementaerin HOB electronic BeteiligungsGmbH
AG Fuerth HRB 3416

Re: [Qemu-devel] Patch: cpu-all.h for better cycle counter

[andrzej zaborowski]

On 26/03/07, Marc Lörner <marc.loerner@hob.de> wrote:
> On Monday 26 March 2007 00:08, andrzej zaborowski wrote:
> > On 26/03/07, andrzej zaborowski <balrog@zabor.org> wrote:
> <snip>
> > > > +# warning non-optimized CPU
> > > > +#include <sys/time.h>
> > > > +#include <time.h>
> > > > +
> > > >  static inline int64_t cpu_get_real_ticks (void)
> > > >  {
> > > > -    static int64_t ticks = 0;
> > > > -    return ticks++;
> > > > +       struct timeval tv;
> > > > +       static int64_t i = 0;
> > > > +       int64_t j;
> > > > +
> > > > +       gettimeofday(&tv, NULL);
> > > > +       do {
> > > > +          j = (tv.tv_sec * (uint64_t) 1000000) + tv.tv_usec;
> > > > +       } while (i == j);
> > > > +       i = j;
> > > > +       return j;
> > >
> > > Isn't this an infinite loop? gettimeofday() was left out of the loop.
> > >
> > > How about "return j + (ticks++)" instead of the loop? If I understand
> > > correctly it may slow things down to below 1Hz.
> >
> > (I wanted to say MHz)
>
> I dont think so, in the loop "j" is set and the while-condition is "j==i",
> so unless "(tv.tv_sec * (uint64_t) 1000000) + tv.tv_usec" always computes the
> same value the do-while block gets only executed once.

Well, it does always compute the same value - inside the loop. Doesn't
it? And across one microsecond also the same value in all calls to
cpu_get_real_ticks.

<snip>

Regards

Re: [Qemu-devel] Patch: cpu-all.h for better cycle counter

[Marc Lörner]

On Monday 26 March 2007 22:37, andrzej zaborowski wrote:
> On 26/03/07, Marc Lörner <marc.loerner@hob.de> wrote:
> > On Monday 26 March 2007 00:08, andrzej zaborowski wrote:
> > > On 26/03/07, andrzej zaborowski <balrog@zabor.org> wrote:
> >
> > <snip>
> >
> > > > > +# warning non-optimized CPU
> > > > > +#include <sys/time.h>
> > > > > +#include <time.h>
> > > > > +
> > > > >  static inline int64_t cpu_get_real_ticks (void)
> > > > >  {
> > > > > -    static int64_t ticks = 0;
> > > > > -    return ticks++;
> > > > > +       struct timeval tv;
> > > > > +       static int64_t i = 0;
> > > > > +       int64_t j;
> > > > > +
> > > > > +       gettimeofday(&tv, NULL);
> > > > > +       do {
> > > > > +          j = (tv.tv_sec * (uint64_t) 1000000) + tv.tv_usec;
> > > > > +       } while (i == j);
> > > > > +       i = j;
> > > > > +       return j;
> > > >
> > > > Isn't this an infinite loop? gettimeofday() was left out of the loop.
> > > >
> > > > How about "return j + (ticks++)" instead of the loop? If I understand
> > > > correctly it may slow things down to below 1Hz.
> > >
> > > (I wanted to say MHz)
> >
> > I dont think so, in the loop "j" is set and the while-condition is
> > "j==i", so unless "(tv.tv_sec * (uint64_t) 1000000) + tv.tv_usec" always
> > computes the same value the do-while block gets only executed once.
>
> Well, it does always compute the same value - inside the loop. Doesn't
> it? And across one microsecond also the same value in all calls to
> cpu_get_real_ticks.
>
> <snip>
>
> Regards

okay, sometimes one doesn't see the wood for the trees.
The over and over computation of the same value doesn't make sense, 
so the call of gettimeofday should be in the loop (i think).
-- 
Freundliche Gruesse,
Marc Lörner

Tel.: +49 (0)9103-715-30046
Fax.: +49 (0)9103-715-271
Internet: http://www.hob.de
E-Mail: marc.loerner@hob.de

HOB GmbH & Co. KG
Schwadermühlstraße 3
90556 Cadolzburg

Geschaeftsfuehrer Klaus Brandstaetter - Franz Wiedenmann

AG Fuerth  HRA 5180
Steuer-Nr. 218/163/00107
USt-ID-Nr. DE 132747002

Komplementaerin HOB electronic BeteiligungsGmbH
AG Fuerth HRB 3416