pthread_cond_timedwait

pthread_cond_timedwait

[Mikael.Ambrus]

	[Ambrus Mikael]  Dear pthreads colleagues, 

	I'm writing a program that uses pthread_cond_timedwait. In the book
that I'm using (Pthreads Programming by Nicols, Buttlar & Proulux Farell )
it says that this function should suspend the thread until some other thread
calls  pthread_cond_signal, pthread_cond_broadcast OR the system timer is
greater than or equal to the third argument (abstime). 

	Since clock_gettime is not implemented and I cant find some other
function that tells me the system time, I've tried to implement my own
version of clock_gettime by using the ansi function clock(). This function
returns the number of ticks that has elapsed since the program was started.

	But it seams that this is not the same absolute time that
pthread_cond_timedwait requires since the thread won't awaken.

	Is there another way to acquire the relevant system time?

	Another thing that has bothered me is that I recently downloaded the
latest snip of phtreads and now my read/write locks won't work. Since the
precompiled lib isn't recognised by the linker (neither with gcc v20.1 nor
MSVC 6.0) for some reason, I built the libs using the buildlib.bat provided.
Here's a snip of the code that fails:

	int pthread_rdwr_wunlock_np (
	   pthread_rdwr_t *rdwrp
	){
	   assert( pthread_mutex_lock(&(rdwrp->mutex)) == 0 );
	   if (rdwrp->writers_writing == 0) {
	      assert( pthread_mutex_unlock(&(rdwrp->mutex)) == 0 );
	      return(-1);            
	   }else{
	      rdwrp->writers_writing = 0;
	      assert(pthread_cond_broadcast(&(rdwrp->lock_free)) == 0);

	   };

	   assert( pthread_mutex_unlock(&(rdwrp->mutex)) == 0 );
	   return(0);
	};

	pthread_cond_broadcast returns EINVAL. Does anyone recognise this
problem?

	Wkr
	Michael Ambrus
	Siemens Elema
>  

Re: pthread_cond_timedwait

[Scott Lightner]

_ftime() will give you the current time with milliseconds.

I believe struct timespec is in nanoseconds so you will lose some precision.

Scott
----- Original Message -----
From: <Mikael.Ambrus@elema.siemens.se>
To: <pthreads-win32@sourceware.cygnus.com>
Sent: Thursday, September 09, 1999 11:47 AM
Subject: pthread_cond_timedwait


> [Ambrus Mikael]  Dear pthreads colleagues,
>
> I'm writing a program that uses pthread_cond_timedwait. In the book
> that I'm using (Pthreads Programming by Nicols, Buttlar & Proulux Farell )
> it says that this function should suspend the thread until some other
thread
> calls  pthread_cond_signal, pthread_cond_broadcast OR the system timer is
> greater than or equal to the third argument (abstime).
>
> Since clock_gettime is not implemented and I cant find some other
> function that tells me the system time, I've tried to implement my own
> version of clock_gettime by using the ansi function clock(). This function
> returns the number of ticks that has elapsed since the program was
started.
>
> But it seams that this is not the same absolute time that
> pthread_cond_timedwait requires since the thread won't awaken.
>
> Is there another way to acquire the relevant system time?
>
> Another thing that has bothered me is that I recently downloaded the
> latest snip of phtreads and now my read/write locks won't work. Since the
> precompiled lib isn't recognised by the linker (neither with gcc v20.1 nor
> MSVC 6.0) for some reason, I built the libs using the buildlib.bat
provided.
> Here's a snip of the code that fails:
>
> int pthread_rdwr_wunlock_np (
>    pthread_rdwr_t *rdwrp
> ){
>    assert( pthread_mutex_lock(&(rdwrp->mutex)) == 0 );
>    if (rdwrp->writers_writing == 0) {
>       assert( pthread_mutex_unlock(&(rdwrp->mutex)) == 0 );
>       return(-1);
>    }else{
>       rdwrp->writers_writing = 0;
>       assert(pthread_cond_broadcast(&(rdwrp->lock_free)) == 0);
>
>    };
>
>    assert( pthread_mutex_unlock(&(rdwrp->mutex)) == 0 );
>    return(0);
> };
>
> pthread_cond_broadcast returns EINVAL. Does anyone recognise this
> problem?
>
> Wkr
> Michael Ambrus
> Siemens Elema
> >
>

Re: [pthread-win32] pthread_cond_timedwait

[Tristan Savatier]

use _ftime().

For systems where _ftime() is not available, you can use
GetSystemTime and convert it to a FILETIME with SystemTimeToFileTime.
Note: _ftime is NOT available on WinCE, so I had to do that
for the WinCE port of posix_thread...

The problem is that the posix standard uses system time defined
as number of microseconds since jan 1, 1970, whereas the windows
FILETIME is defined as the number of 100 nanosecs since jan 1,
1601!

I have written two small routines to convert between those
two:

#ifndef USE_FTIME

/* time between jan 1, 1601 and jan 1, 1970 in units of 100 nanoseconds
*/
#define TIMESPEC_TO_FILETIME_OFFSET (((LONGLONG)27111902 << 32) +
(LONGLONG)3577643008)

static void
timespec_to_filetime(const struct timespec *ts, FILETIME *ft)
     /*
      *
-------------------------------------------------------------------
	  * converts struct timespec
	  * where the time is expressed in seconds and nanoseconds from Jan 1,
1970.
	  * into FILETIME (as set by GetSystemTimeAsFileTime), where the time
is
	  * expressed in 100 nanoseconds from Jan 1, 1601,
      *
-------------------------------------------------------------------
	  */
{
	*(LONGLONG *)ft = ts->tv_sec * 10000000 + (ts->tv_nsec + 50) / 100 +
TIMESPEC_TO_FILETIME_OFFSET;
}

static void
filetime_to_timespec(const FILETIME *ft, struct timespec *ts)
     /*
      *
-------------------------------------------------------------------
	  * converts FILETIME (as set by GetSystemTimeAsFileTime), where the
time is
	  * expressed in 100 nanoseconds from Jan 1, 1601,
	  * into struct timespec
	  * where the time is expressed in seconds and nanoseconds from Jan 1,
1970.
      *
-------------------------------------------------------------------
	  */
{
	ts->tv_sec = (int)((*(LONGLONG *)ft - TIMESPEC_TO_FILETIME_OFFSET) /
10000000);
	ts->tv_nsec = (int)((*(LONGLONG *)ft - TIMESPEC_TO_FILETIME_OFFSET -
((LONGLONG)ts->tv_sec * (LONGLONG)10000000)) * 100);
}

#endif /* not USE_FTIME */

---------------------------

To get the current system time:

	  _ftime(&currSysTime);

or

	  {
	    FILETIME ft;
	    SYSTEMTIME st;

	    GetSystemTime(&st);
            SystemTimeToFileTime(&st, &ft);
	    /* GetSystemTimeAsFileTime(&ft); would be faster, but it does not
exist on WinCE */

	    filetime_to_timespec(&ft, &currSysTime);
	  }

--------------------------

I hope this helps!

-t

Mikael.Ambrus@elema.siemens.se wrote:
> 
>         [Ambrus Mikael]  Dear pthreads colleagues,
> 
>         I'm writing a program that uses pthread_cond_timedwait. In the book
> that I'm using (Pthreads Programming by Nicols, Buttlar & Proulux Farell )
> it says that this function should suspend the thread until some other thread
> calls  pthread_cond_signal, pthread_cond_broadcast OR the system timer is
> greater than or equal to the third argument (abstime).
> 
>         Since clock_gettime is not implemented and I cant find some other
> function that tells me the system time, I've tried to implement my own
> version of clock_gettime by using the ansi function clock(). This function
> returns the number of ticks that has elapsed since the program was started.
> 
>         But it seams that this is not the same absolute time that
> pthread_cond_timedwait requires since the thread won't awaken.
> 
>         Is there another way to acquire the relevant system time?
> 
>         Another thing that has bothered me is that I recently downloaded the
> latest snip of phtreads and now my read/write locks won't work. Since the
> precompiled lib isn't recognised by the linker (neither with gcc v20.1 nor
> MSVC 6.0) for some reason, I built the libs using the buildlib.bat provided.
> Here's a snip of the code that fails:
> 
>         int pthread_rdwr_wunlock_np (
>            pthread_rdwr_t *rdwrp
>         ){
>            assert( pthread_mutex_lock(&(rdwrp->mutex)) == 0 );
>            if (rdwrp->writers_writing == 0) {
>               assert( pthread_mutex_unlock(&(rdwrp->mutex)) == 0 );
>               return(-1);
>            }else{
>               rdwrp->writers_writing = 0;
>               assert(pthread_cond_broadcast(&(rdwrp->lock_free)) == 0);
> 
>            };
> 
>            assert( pthread_mutex_unlock(&(rdwrp->mutex)) == 0 );
>            return(0);
>         };
> 
>         pthread_cond_broadcast returns EINVAL. Does anyone recognise this
> problem?
> 
>         Wkr
>         Michael Ambrus
>         Siemens Elema
> >

-- 
Regards, -- Tristan Savatier (President, MpegTV LLC)

MpegTV: http://www.mpegtv.com - Xaudio: http://www.xaudio.com
MPEG.ORG: http://www.mpeg.org - Tel: (415) 864 6466

Re: pthread_cond_timedwait

[Ross Johnson]

tests\stress1.c uses the following routine. The args are: a pointer to a
timespec instance and; the number of millisecs into the future. It
returns the first arg so you can call it directly from within
pthread_cond_wait's argument list:

struct timespec *
millisecondsFromNow (struct timespec * time, int millisecs)
{
  struct _timeb currSysTime;
  int64_t nanosecs, secs;
  const int64_t NANOSEC_PER_MILLISEC = 1000000;
  const int64_t NANOSEC_PER_SEC = 1000000000;

  /* get current system time and add millisecs */
  _ftime(&currSysTime);

  nanosecs = ((int64_t) (millisecs + currSysTime.millitm)) * NANOSEC_PER_MILLISEC;
  if (nanosecs >= NANOSEC_PER_SEC)
    {
      secs = currSysTime.time + 1;
      nanosecs %= NANOSEC_PER_SEC;
    }
  else
    {
      secs = currSysTime.time;
    }

  time->tv_nsec = (long)nanosecs;
  time->tv_sec = (long)secs;

  return time;
}

Ross

On Fri, 2005-06-10 at 21:23 +0100, Steve Croall (TIBCO) wrote: 
> Hi,
> 
> The timeout parameter you pass to pthread_cond_timedwait() is a struct 
> timespec type, which is defined in pthread.h, if HAVE_STRUCT_TIMESPEC is 
> not defined at build time.
> 
> struct timespec {
>          long tv_sec;
>          long tv_nsec;
> };
> 
> Just set the tv_nsec to the amount of nanoseconds required.  The pthread 
> source on Windows uses the following calculation to convert the above 
> structure to milliseconds:
> 
>    tmpAbsMilliseconds =  (int64_t)abstime->tv_sec * MILLISEC_PER_SEC;
>    tmpAbsMilliseconds += ((int64_t)abstime->tv_nsec + 
> (NANOSEC_PER_MILLISEC/2)) / NANOSEC_PER_MILLISEC;
> 
> Using the following defines:
> 
>    const int64_t NANOSEC_PER_MILLISEC = 1000000;
>    const int64_t MILLISEC_PER_SEC = 1000;
> 
> Steve.
> 
> Allan Comar wrote:
> > Hi all, I am needing something that I couldn't find in nowhere else, I am trying to use pthread_cond_timedwait but I want that the time that I need to wait is setted im milliseconds, I already could do in seconds and i am having a real hard time trying it in milliseconds. Any Ideas ? I am using MSVC6.0 with pthreads.
> > 
> 

Re: pthread_cond_timedwait

[Steve Croall (TIBCO)]

Hi,

The timeout parameter you pass to pthread_cond_timedwait() is a struct 
timespec type, which is defined in pthread.h, if HAVE_STRUCT_TIMESPEC is 
not defined at build time.

struct timespec {
         long tv_sec;
         long tv_nsec;
};

Just set the tv_nsec to the amount of nanoseconds required.  The pthread 
source on Windows uses the following calculation to convert the above 
structure to milliseconds:

   tmpAbsMilliseconds =  (int64_t)abstime->tv_sec * MILLISEC_PER_SEC;
   tmpAbsMilliseconds += ((int64_t)abstime->tv_nsec + 
(NANOSEC_PER_MILLISEC/2)) / NANOSEC_PER_MILLISEC;

Using the following defines:

   const int64_t NANOSEC_PER_MILLISEC = 1000000;
   const int64_t MILLISEC_PER_SEC = 1000;

Steve.

Allan Comar wrote:
> Hi all, I am needing something that I couldn't find in nowhere else, I am trying to use pthread_cond_timedwait but I want that the time that I need to wait is setted im milliseconds, I already could do in seconds and i am having a real hard time trying it in milliseconds. Any Ideas ? I am using MSVC6.0 with pthreads.
> 

-- 

J. Senior Software Engineer, Tibco Software Ltd.
T. +44 (0) 1792 360773
M. +44 (0) 7788 971394
E. scroall@tibco.com
W. www.tibco.com

pthread_cond_timedwait

[Allan Comar]

Hi all, I am needing something that I couldn't find in nowhere else, I am trying to use pthread_cond_timedwait but I want that the time that I need to wait is setted im milliseconds, I already could do in seconds and i am having a real hard time trying it in milliseconds. Any Ideas ? I am using MSVC6.0 with pthreads.

RE: pthread_cond_timedwait

[Bossom, John]

I just saw a potential bug in your code, below:

assert( pthread_mutex_lock(&(rdwrp->mutex)) == 0 );

I recommend that you do not call functions within an 'assert'
unless you don't mind the compiler optimizing out the 'assert'
statement! (i.e. your lock will not happen)

Quote from MS Dev Studio, 5.0 for 'assert':
"The ANSI assert macro is typically used to identify logic errors during
program development, by implementing the expression argument to evaluate to
false only when the program is operating incorrectly. After debugging is
complete, assertion checking can be turned off without modifying the source
file by defining the identifier NDEBUG. NDEBUG can be defined with a /D
command-line option or with a #define directive. If NDEBUG is defined with
#define, the directive must appear before ASSERT.H is included."



I recommend you using:

	int	lockResult;

	lockResult = pthread_mutex_lock(&(rdwrp->mutex));
	assert( lockResult == 0 );


-----Original Message-----
From: Mikael.Ambrus@elema.siemens.se
[ mailto:Mikael.Ambrus@elema.siemens.se ]
Sent: Thursday, September 09, 1999 11:47 AM
To: pthreads-win32@sourceware.cygnus.com
Subject: pthread_cond_timedwait


	[Ambrus Mikael]  Dear pthreads colleagues, 

	I'm writing a program that uses pthread_cond_timedwait. In the book
that I'm using (Pthreads Programming by Nicols, Buttlar & Proulux Farell )
it says that this function should suspend the thread until some other thread
calls  pthread_cond_signal, pthread_cond_broadcast OR the system timer is
greater than or equal to the third argument (abstime). 

	Since clock_gettime is not implemented and I cant find some other
function that tells me the system time, I've tried to implement my own
version of clock_gettime by using the ansi function clock(). This function
returns the number of ticks that has elapsed since the program was started.

	But it seams that this is not the same absolute time that
pthread_cond_timedwait requires since the thread won't awaken.

	Is there another way to acquire the relevant system time?

	Another thing that has bothered me is that I recently downloaded the
latest snip of phtreads and now my read/write locks won't work. Since the
precompiled lib isn't recognised by the linker (neither with gcc v20.1 nor
MSVC 6.0) for some reason, I built the libs using the buildlib.bat provided.
Here's a snip of the code that fails:

	int pthread_rdwr_wunlock_np (
	   pthread_rdwr_t *rdwrp
	){
	   assert( pthread_mutex_lock(&(rdwrp->mutex)) == 0 );
	   if (rdwrp->writers_writing == 0) {
	      assert( pthread_mutex_unlock(&(rdwrp->mutex)) == 0 );
	      return(-1);            
	   }else{
	      rdwrp->writers_writing = 0;
	      assert(pthread_cond_broadcast(&(rdwrp->lock_free)) == 0);

	   };

	   assert( pthread_mutex_unlock(&(rdwrp->mutex)) == 0 );
	   return(0);
	};

	pthread_cond_broadcast returns EINVAL. Does anyone recognise this
problem?

	Wkr
	Michael Ambrus
	Siemens Elema
>  

RE: pthread_cond_timedwait

[Medina, Aurelio]

To all,

Speaking of read/write locks, does anyone know if the POSIX read/write lock
routines (e.g. pthread_rwlock_init()) that I submitted will eventually
become part of the Pthreads-Win32 library?  Seems like someone is already
working on another implementation.  The routines that I submitted were
introduced in UNIX98 and are currently a X/Open standard.  These routines
are portable and are currently available on HP-UX 11.0.

Thanks,
Aurelio Medina

	-----Original Message-----
	From:	Mikael.Ambrus@elema.siemens.se
[SMTP:Mikael.Ambrus@elema.siemens.se]
	Sent:	Thursday, September 09, 1999 10:47 AM
	To:	pthreads-win32@sourceware.cygnus.com
	Subject:	pthread_cond_timedwait

		[Ambrus Mikael]  Dear pthreads colleagues, 

		I'm writing a program that uses pthread_cond_timedwait. In
the book
	that I'm using (Pthreads Programming by Nicols, Buttlar & Proulux
Farell )
	it says that this function should suspend the thread until some
other thread
	calls  pthread_cond_signal, pthread_cond_broadcast OR the system
timer is
	greater than or equal to the third argument (abstime). 

		Since clock_gettime is not implemented and I cant find some
other
	function that tells me the system time, I've tried to implement my
own
	version of clock_gettime by using the ansi function clock(). This
function
	returns the number of ticks that has elapsed since the program was
started.

		But it seams that this is not the same absolute time that
	pthread_cond_timedwait requires since the thread won't awaken.

		Is there another way to acquire the relevant system time?

		Another thing that has bothered me is that I recently
downloaded the
	latest snip of phtreads and now my read/write locks won't work.
Since the
	precompiled lib isn't recognised by the linker (neither with gcc
v20.1 nor
	MSVC 6.0) for some reason, I built the libs using the buildlib.bat
provided.
	Here's a snip of the code that fails:

		int pthread_rdwr_wunlock_np (
		   pthread_rdwr_t *rdwrp
		){
		   assert( pthread_mutex_lock(&(rdwrp->mutex)) == 0 );
		   if (rdwrp->writers_writing == 0) {
		      assert( pthread_mutex_unlock(&(rdwrp->mutex)) == 0 );
		      return(-1);            
		   }else{
		      rdwrp->writers_writing = 0;
		      assert(pthread_cond_broadcast(&(rdwrp->lock_free)) ==
0);

		   };

		   assert( pthread_mutex_unlock(&(rdwrp->mutex)) == 0 );
		   return(0);
		};

		pthread_cond_broadcast returns EINVAL. Does anyone recognise
this
	problem?

		Wkr
		Michael Ambrus
		Siemens Elema
	>