[Omp] Re: Omp Digest, Vol 30, Issue 3
j_larson
j_larson at mail.sisna.com
Wed Jun 14 14:18:00 PDT 2006
Hi Patricia,
FLUSH is a memory synchronization operation that affects only the thread that executes it. When a thread executes a FLUSH it updates locally kept SHARED variables to shared memory, and discards the local copies. The shared variable is now visible in shared memory to only the thread that executed the FLUSH. The visibility of these FLUSHed variables by other threads is not guaranteed.
BARRIER is a memory synchronization operation and a thread synchronization operation. Each thread that executes a BARRIER performs a FLUSH operation as described above at its own pace. In addition, each thread waits until all threads on the team have finished executing a FLUSH operation before any thread proceeds. The instant when all threads are released is called the synchronization point, and signals that all thread have completed their individual FLUSHes. All the updated shared variables are guaranteed to be visible in shared memory to all the threads on the team after the synchronization point.
Regards,
John L. Larson
Innovative Systems Lab
NCSA
---------- Original Message ----------------------------------
From: omp-request at openmp.org
Reply-To: omp at openmp.org
Date: Wed, 14 Jun 2006 12:00:05 -0700 (PDT)
>Send Omp mailing list submissions to
> omp at openmp.org
>
>To subscribe or unsubscribe via the World Wide Web, visit
> http://openmp.org/mailman/listinfo/omp
>or, via email, send a message with subject or body 'help' to
> omp-request at openmp.org
>
>You can reach the person managing the list at
> omp-owner at openmp.org
>
>When replying, please edit your Subject line so it is more specific
>than "Re: Contents of Omp digest..."
>
>
>Today's Topics:
>
> 1. flush x other synchronization operations
> (Patricia Bittencourt Sampaio)
>
>
>----------------------------------------------------------------------
>
>Message: 1
>Date: Wed, 14 Jun 2006 12:38:00 -0300 (ART)
>From: Patricia Bittencourt Sampaio <patricourt at yahoo.com.br>
>Subject: [Omp] flush x other synchronization operations
>To: Omp at openmp.org
>Message-ID: <20060614153800.32844.qmail at web50201.mail.yahoo.com>
>Content-Type: text/plain; charset=iso-8859-1
>
>
> Hello,
>
> I am a little confused about the difference
>between flush and other synchronization constructs.
>
> As OpenMP uses a Relaxed Consistency Memory Model,
>a barrier just like the end of a critical section or
>every operation that involves a page release, imposes
>a point of a consistent view of memory. In a way that
>subsequenttelly reads of a location that is shared
>among the threads will require the access to memory
>and not to the local instance of each thread.
>
> According to the behave of the flush operation,
>that is exactly what it does.
>
> So, I presume that the differences between flush
>and other synchronization operations are just twice:
>
>- Flush enforces a consistent view of memory with
>respect to a flush set (when it is specified) and not
>related to all shared variables as a barrier
>synchronization does.
>
>- In the case of a barrier, it requires that each
>thread of a team wait for all other threads to arrive
>at the barrier, while flush does not imposes this
>roll.
>
> Am I wrong about this? Do you recognize other
>differences?
>
>Best regards,
>
>
>Patricia B. Sampaio
>====================================
>Federal University of Rio de Janeiro
>
>
>
>
>_______________________________________________________
>Novidade no Yahoo! Mail: receba alertas de novas mensagens no seu celular. Registre seu aparelho agora!
>http://br.mobile.yahoo.com/mailalertas/
>
>
>
>
>------------------------------
>
>_______________________________________________
>Omp mailing list
>Omp at openmp.org
>http://openmp.org/mailman/listinfo/omp
>
>
>End of Omp Digest, Vol 30, Issue 3
>**********************************
>
_________________________________
SISNA...more service, less money.
http://www.sisna.com/exclusive/
More information about the Omp
mailing list