Hi Nikhil --
As you correctly guess, we do have routines to do chunking of iteration
spaces across tasks, and I've been thinking more and more that we should
expose them as helper functions/iterators with nice interfaces (and
documentation) for the end-user.
What we have right now (with the warning that it's not well-documented,
packaged, or set up for the end-user) is in modules/dists/DSIUtil.chpl,
for what it's worth.
Even if a coforall turns out to be sufficient in your case, I do think
that providing the user some way to get at this type of information within
a forall loop will be helpful for the intermediate programmer... (so I
think it's something we should continue to work on trying to design and
support).
-Brad
On Tue, 22 Mar 2016, Nikhil Padmanabhan wrote:
> Brad, Vassily --
>
> Thanks for the responses....
>
> Brad is correct in that what I was trying to do was to see if a forall
> could do something like :
>
> coforall iwork in WorkerList {
> // Do some initialization that is local to each iwork
> var MyD = splitDomain(BigD, iwork, WorkerList);
> for ii in MyD {
> //
> }
> // Finish up, again local to each iwork
> }
>
> I guess this is my mental model for what the forall loop actually
> translates to (with some internal WorkerList), but thinking about it some
> more, I can see that this is probably too limiting.
>
> Looking at the code that I have above, it actually is quite straightforward
> (and not really worth abstracting further). The only step that requires
> some thought is how to split the domain. This is easy for a simple domain
> in 1D, but it might be useful to have a simple version that handles an
> arbitrary domain (with the obvious comment that a user might want to
> rewrite this for a specific application).
>
> I assume something like this already exists inside Chapel -- is there some
> way for the user to access this?
>
> Thanks!
> -- Nikhil
>
>
>
> ---------------------------------
> Nikhil Padmanabhan
> [email protected]
>
> On Tue, Mar 22, 2016 at 12:45 AM, Brad Chamberlain <[email protected]> wrote:
>
>>
>> I don't think that provides what Nikhil's asking for, though.
>> Specifically, I think what he wants is:
>>
>> forall i in D do
>> ... some way of referring directly/implicitly the ID from
>> 0..#numTasks for this particular task (similar to
>> 'here' in locale space, but in the task space created
>> by the loop) where numTasks is the number of tasks
>> created by the forall loop...
>>
>> Of course, you'd also want/need a way to query what that number is. And to
>> make it extra challenging from a design perspective, that number could
>> potentially vary as the forall loop runs...
>>
>> FWIW, this is something I find myself wanting from time to time too, but
>> haven't had that "aha!" moment as to how we should provide it. Well,
>> except for the long-term plan to support task teams and to have a way of
>> getting at the task team(s) that implement parallel constructs like forall
>> loops (or coforall loops) or in which case it might be something as simple
>> as:
>>
>> forall i in D do
>> (myTID, numTasks) = thisTeam.getMembership()
>>
>> -Brad
>>
>>
>>
>> On Mon, 21 Mar 2016, Vassily Litvinov wrote:
>>
>> Nikhil,
>>>
>>> Try using "in" as a forall intent. That should give you task-private
>>> variables in a forall loop.
>>>
>>> Vassily
>>>
>>> On 03/21/16 17:04, Nikhil Padmanabhan wrote:
>>>
>>>> Dear all,
>>>>
>>>> For the code executing inside a forall loop, is there any way in which it
>>>> can determine what its "thread" id is?
>>>>
>>>> I'm imagining a case here where I need to call a function that takes in
>>>> some state variable (eg. like a random number generator) that it might
>>>> modify. The simplest way I can think of doing this in parallel would be
>>>> to
>>>> have Nthread copies of the state variable, and then every thread just
>>>> modifies its own copy.
>>>>
>>>> I can certainly do this using a coforall loop; I was just curious if
>>>> there
>>>> was a way to do this even in forall loops....
>>>>
>>>> Thanks!
>>>> -- Nikhil
>>>> ---------------------------------
>>>> Nikhil Padmanabhan
>>>> [email protected]
>>>>
>>>>
>>>>
>>>>
>>>> ------------------------------------------------------------------------------
>>>> Transform Data into Opportunity.
>>>> Accelerate data analysis in your applications with
>>>> Intel Data Analytics Acceleration Library.
>>>> Click to learn more.
>>>> http://pubads.g.doubleclick.net/gampad/clk?id=278785351&iu=/4140
>>>>
>>>>
>>>>
>>>> _______________________________________________
>>>> Chapel-users mailing list
>>>> [email protected]
>>>> https://lists.sourceforge.net/lists/listinfo/chapel-users
>>>>
>>>>
>>>
>>> ------------------------------------------------------------------------------
>>> Transform Data into Opportunity.
>>> Accelerate data analysis in your applications with
>>> Intel Data Analytics Acceleration Library.
>>> Click to learn more.
>>> http://pubads.g.doubleclick.net/gampad/clk?id=278785351&iu=/4140
>>> _______________________________________________
>>> Chapel-users mailing list
>>> [email protected]
>>> https://lists.sourceforge.net/lists/listinfo/chapel-users
>>>
>>>
>
------------------------------------------------------------------------------
Transform Data into Opportunity.
Accelerate data analysis in your applications with
Intel Data Analytics Acceleration Library.
Click to learn more.
http://pubads.g.doubleclick.net/gampad/clk?id=278785351&iu=/4140
_______________________________________________
Chapel-users mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/chapel-users
