Subj : Re: pthread_cond to create sema?
To   : comp.programming.threads
From : Giancarlo Niccolai
Date : Thu Jan 20 2005 10:57 am

Paolo wrote:

> Giancarlo Niccolai <searchforme@ingoogle.com> wrote:
>> Paolo wrote:
> 
>>> 
>>> Are these P() and V() correct if i want to implement
>>> semaphores p() and v() functions?
>>> 
>>> messagecounter is a global int variable,
>>> semaforo is a mutex used only here
>>> sema is a cond variable
>>> 
>>> 
>>> 
>>> #define P() {\
>>> pthread_mutex_lock(&semaforo);\
>>> while ( TW_messagecounter <= 0) {\
>>>  pthread_cond_wait(&sema, &semaforo);\
>>> }\
>>> messagecounter--;\
>>> pthread_mutex_unlock(&semaforo);\
>>> }
>>> 
>>> #define V() {\
>>> pthread_mutex_lock(&semaforo);\
>>> messagecounter++;\
>>> pthread_cond_signal(&sema);\
>>> pthread_mutex_unlock(&semaforo);\
>>> }
>>> 
>>> Thank you!
>>> Paolo
> 
>> Yes, but you can make it better by checking if there are waiting agents
>> before signaling and pushing the cleanup function before waiting (and
>> popping it after waiting).
> 
> Do you think i could get a speed boost avoiding a signal for each V() with
> a new variable and a check on it?

It depends on lower level implementation; also "a boost" is probably too
much. A "little better" maybe is nearer to what you can get.

> 
>>Minimal cleanup is the pthread_unlock_mutex, so
>> that if your wait is killed by another thread, the mutex gets unlocked
>> before the agent is definitely stopped.
> 
> How can i unlock the mutex if i'm killed, before i have to exit?
> with a signal handler?


search google for man pthread_cleanup_push.

Gian.

.