[Omp] Newbye question about workshare

[Breshears, Clay]

It seems to me that there is an obvious data race, not on the loop
index, but on the dx, dy, and dz variables (vectors?).  Can you make
these private?

						clay 

-----Original Message-----
From: omp-bounces at openmp.org [mailto:omp-bounces at openmp.org] On Behalf
Of Tim Prince
Sent: Wednesday, January 31, 2007 7:32 AM
To: l.gontrani at caspur.it
Cc: omp at openmp.org
Subject: Re: [Omp] Newbye question about workshare

l.gontrani at caspur.it wrote:
> Hi,
> I am experiencing a problem with parallizing the following f90 code (a

> pairwise distance matrix construction): 
> 
> DO i=1,int(.5*natoms)-1
>   dx=Coord(:,1)-CSHIFT(Coord(:,1),SHIFT=i)
>   dy=Coord(:,2)-CSHIFT(Coord(:,2),SHIFT=i)
>   dx=Coord(:,3)-CSHIFT(Coord(:,3),SHIFT=i)
>   dx=dx*dx
>   dy=dy*dy
>   dz=dz*dz
>   dist(natoms*(i-1)+1:natoms*li)=DSQRT(dx+dy+dz)
> ENDDO 
> 
> I was wondering if I can use the WORKSHARE construct in this case..
> I tried to use $OMP DO,private(i), but it didn't work (of course)
because of 
> the race condition for i 
Did you test it without OpenMP?  It's difficult to believe the array 
sizes match in the last assignment, or that DSQRT is appropriate.  How 
could there be a race condition on the DO index?
_______________________________________________
Omp mailing list
Omp at openmp.org
http://openmp.org/mailman/listinfo/omp