-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
On 25.06.2012 16:59, Olivier Tardieu wrote:
> Hi
>
> This is quite tricky indeed.
> There are subtle differences between the `when' construct and your proposed
> expansion.
>
> Before I get into particulars, let me first say I don't understand the comment
> > // wait for some atomic to finish //
> so I may not fully understand the intend here.
>
> I see two main issues with the expansion.
>
> 1.
> `when' blocks until a point in the execution is reached where`G' is true.
> You proposed expansion keeps checking the condition `G'.
> So while the implementation of `when' may do the same, it can also rely
> notification mechanisms more efficient than spin loops.
>
> 2.
> The X10 scheduler is a cooperative one.
> The `when' construct suspends the execution freeing resources (think CPU or
> thread) for other activities to use.
> In the proposed loop, the activity never stops, therefore will never be
> preempted by the X10 runtime.
> While there may be a runnable activity willing to update `G', it is unlikely
> but
> entirely possible that this activity will never run, which would result in a
> deadlock.
>
> Now getting back to the comment in the code, there is no way to wait for
> another
> atomic to finish
> without again using when or atomic in a loop.
>
> I hope this helps.
>
> Olivier
>
>
> Andreas Zwinkau <zwin...@kit.edu> wrote on 06/14/2012 02:48:50 AM:
>
> > From: Andreas Zwinkau <zwin...@kit.edu>
> > To: x10-users@lists.sourceforge.net,
> > Date: 06/14/2012 02:49 AM
> > Subject: [X10-users] when semantics
> >
> The X10 spec provides two examples, how 'when' cannot be implemented
> with 'if' and 'atomic'. However, an additional loop seems to do the trick:
>
> when (G) S;
>
> ==into==>
>
> while (true) {
> atomic {
> if (G) {
> S;
> break;
> }
> }
> // wait for some atomic to finish //
> }
>
> For my simple test, it seems to work, but concurrency is tricky. ;)
So I understand that my expansion is *semantically* equivalent if we
assume at least a parallelism of two to rule out the deadlock situation
you describe.
In terms of efficiency/performance 'when' should be superior, since it
can "wait for some atomic to finish". This could be understood as an
optimization.
On the other hand, there seems to be a way to wait for another atomic to
finish with a simple call: "x10.lang.Runtime.awaitAtomic();". At least,
that seems to be what the runtime does.
- --
Andreas Zwinkau
Karlsruhe Institute of Technology (KIT)
Institut für Programmstrukturen und Datenorganisation (IPD)
Lehrstuhl Prof. Snelting
Adenauerring 20a
76131 Karlsruhe
Phone: +49 721 608 48351
Fax: +49 721 608 48457
Email: zwin...@kit.edu
Web: http://pp.info.uni-karlsruhe.de/person.php?id=107
KIT – University of the State of Baden-Wuerttemberg and
National Research Center of the Helmholtz Association
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.11 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/
iQEcBAEBAgAGBQJP6WlGAAoJEKm3M8a3pi2iYDIH/AjProM28IuqSqwYGLNgbbwr
WVwCSVqpBxQFOdaRmv/Q5hPmQPxmGwyzZNdsN+LdQTBsqWtG/AMNLt0GPD5CWRA8
aZP7UB8EiIhBYOKJ/aIoitidRf0VxKxY0VyarX8K9I1g6x6DlI59IJd+vY6xaPB7
gWecp/1Bzm2dvS+DVLR8oawH54clJ4HT9u0uaVdYHhlxLVCagfMZUUAAxTgRAJQv
tHGyXY7guKSAgxrUPZVi1UXGECbDVFkeW3agakcvhAGOrV3wflej+xTYcbZkuzGS
8ngYlL2ThIFTEZZ1YcU6KN3/jLanmRVYSif8iHdJlo6sXlU/yng+kcu5sfm2Sik=
=8IUp
-----END PGP SIGNATURE-----
------------------------------------------------------------------------------
Live Security Virtual Conference
Exclusive live event will cover all the ways today's security and
threat landscape has changed and how IT managers can respond. Discussions
will include endpoint security, mobile security and the latest in malware
threats. http://www.accelacomm.com/jaw/sfrnl04242012/114/50122263/
_______________________________________________
X10-users mailing list
X10-users@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/x10-users
