Re: [Gomp-discuss] Decision about semaphores ..

[Lars Segerlund]

 I agree with you, however every parallel section in openMP exept those 
specificly told not to syncronize, and also all barriers, single and 
some other directives need's an efficient join/exclusion mechanism.

 If there is any place to optimize this will be it :-) ..

 No arguing ... same thing different words

 / Lars Segerlund.


Scott Robert Ladd wrote:
> Perhaps you misunderstood me... all parallel programs have fork/joins, and
> of course we need to implement them. I was just saying that a well-written
> parallel program *minimizes* the number of fork/joins.
>
> ..Scott
>
>
> > -----Original Message-----
> > From: address@hidden
> > [mailto:address@hidden Behalf
> > Of Lars Segerlund
> > Sent: Wednesday, February 19, 2003 09:19
> > To: address@hidden
> > Subject: Re: [Gomp-discuss] Decision about semaphores ..
> >
> >
> >
> >  It's not an option if we want to have fork/join's or not, they are
> > explicitly specified in openMP. A semaphore is as good a join as youre
> > going to get, ( compared to multiple waits ).
> >
> >  But Ok, I get it POSIX.
> >
> >  / Seger.
> >
> > Scott Robert Ladd wrote:
> >
> > > Biagio Lucini wrote:
> > >
> > >
> > > > My vote is for posix, for the reason you explain.
> > > >
> > > > In a well-structured parallel program there must be as few as possible
> > > > fork/join. In an hyperideal case, you should just fork close to the
> > > > beginning and join right before the end. So I'm not concerned about this
> > > > overhead: part of it is down to the programer, the rest should be
> > > > negligible anyway.
> > >
> > >
> > > I agree with Biago. A parallel program that contains too many
> >
> > fork/joins is
> >
> > > unlikely to perform effectively. In this case, I think
> >
> > portability (posix)
> >
> > > is most important.
> > >
> > > ..Scott
> > >
> > >
> > >
> > > _______________________________________________
> > > Gomp-discuss mailing list
> > > address@hidden
> > > http://mail.nongnu.org/mailman/listinfo/gomp-discuss
> >
> >
> >
> > _______________________________________________
> > Gomp-discuss mailing list
> > address@hidden
> > http://mail.nongnu.org/mailman/listinfo/gomp-discuss
>
>
>
>
> _______________________________________________
> Gomp-discuss mailing list
> address@hidden
> http://mail.nongnu.org/mailman/listinfo/gomp-discuss