Thank you, Vass for the response.
ok so there is race condition, I agree.
for distributed computation
I actually do
var ad = A.domain.dims();
var bd = B.domain.dims();
var cd = C.domain.dims();
var (arow, acol) = ad;
var (brow, bcol) = bd;
var (crow, ccol) = cd;
var lock:sync bool;
forall i in arow do
forall j in bcol do
forall k in acol do
lock = true;
C(i,j) += A(i,k)*B(k,j);
var unlock = lock;
Will this do the distributed computation?
To avoid race condition I was planning to use sync variable.
On Sat, May 3, 2014 at 9:21 AM, Vassily Litvinov <[email protected]> wrote:
> Hi Bibek,
>
> * In this loop from your example:
>
>
> forall k in acol do
> C(i,j) += A(i,k)*B(k,j);
>
> all iterations may execute in parallel, i.e. all read and write into
> C(i,j) at the same time. Races can arise easily.
>
> BTW in your context, given that this loop is nested in two other parallel
> loops, it may increase performance if you make it a serial loop ('for'). In
> such case there should be no races.
>
> * Declaring var A,B,C:[D] int; ensures distributed storage. This does not
> by itself guarantee distributed computation. Whether and how the
> computation is distributed is determined by what your forall iterates over.
> E.g.
>
>
> forall i in arow do
>
> If 'arow' is a range or a domain that is not distributed, all iterations
> of the loop will be started on the current locale.
>
> In contrast, e.g. this loop:
>
> forall (i,j) in D do
>
> will start a particular iteration (i,j) on the same node as the one that
> hosts A(i,j), B(i,j), C(i,j).
>
> Vass
>
>
> On Sat, 3 May 2014, Bibek Ghimire wrote:
>
> Hi there, I have written a simple distributed matrix
>> multiplication code in chapel.
>> The main part of the code is
>>
>> var D: domain(2) dmapped Block(boundingBox={1..n,1..n})= {1..n,1..n}
>>
>> which declares domain mapping of type Block and this Block type helps to
>> distributes data
>> into multiple nodes. So now I create an Array for them to distributed
>> along the nodes.
>>
>> var A,B,C:[D] int;
>>
>> //var c:[n, n] int ; q1) What if I create c this way, will it reside on
>> locale 0 and all result come to node 0?
>> Then I do the multiplication here
>>
>> forall i in arow do
>>
>> forall j in bcol do
>> forall k in acol do C(i,j) += A(i,k)*B(k,j);
>> I thought there will not be any race condition here because we are
>> reading array A, B and writing into C's individual array position. q2) Is
>> it ok to do so?
>>
>> I was looking into the BlockDist.chpl to figure out how its internal
>> works but little confused.
>>
>> Thank you,
>>
>> Bibek
>>
>>
>>
>>
------------------------------------------------------------------------------
"Accelerate Dev Cycles with Automated Cross-Browser Testing - For FREE
Instantly run your Selenium tests across 300+ browser/OS combos. Get
unparalleled scalability from the best Selenium testing platform available.
Simple to use. Nothing to install. Get started now for free."
http://p.sf.net/sfu/SauceLabs
_______________________________________________
Chapel-users mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/chapel-users
