Parallel Programming in Fortran 95 using OpenMP - People

50 3. PRIVATE, SHARED & Co.
DYNAMIC : in the previous scheduling method the problem of non homogeneous iterations
has been pointed out. In such a situation it turns out to be difficult to forecast
the optimal distribution of the iterations among the threads. A solution would be
to assign the different blocks of work in a dynamic way: as one thread finishes its
piece of work, it gets a new one. This idea is precisely what lies behind the present
scheduling method.
When SCHEDULE(DYNAMIC,chunk) is specified, the iteration space is divided into
pieces of work with a size equal to chunk. If this optional parameter is not given,
then a size equal to one iteration is considered. Thereafter, each thread gets one of
these pieces of work. When they have finished with their task, they get assigned a
new one until no pieces of work are left.
The best way to understand the working principle of this scheduling method is by
means of an example. In the top part of figure 3.7 12 different pieces of work are
shown. The length of each of these blocks represents the amount of computational
time required to compute them. These pieces of work are going to be distributed
over a team of three threads. The resulting execution schema is shown in the bottom
part of figure 3.7. Clearly to see is that each thread gets a new piece or work when
it finishes the previous one.
GUIDED : the previous dynamic method enhances the capabilities and performance with
respect to the static method, but it adds an overhead in computational cost due to
the handling and distributing of the different iteration packages. This overhead is
reduced as the size of the pieces of work increases, but then a larger non-equilibrium
between the different threads is to expect.
Another dynamic approach to this problem could be to have pieces of work with
decreasing sizes, so that their associated work is smaller and smaller as they are
assigned to the different threads. The decreasing law is of exponential nature so
that the following pieces of work have half the number of iterations as the previous
ones. This is what the present scheduling method does.
The optional parameter chunk specifies the smallest number of iterations grouped
into one piece of work. As before, the last piece of work may have a smaller number
of iterations than specified in chunk in order to cover completely the iteration space.
The number of pieces of work with equal size and the largest size in use are OpenMPimplementation
dependent.
In figure 3.8 an example of the resulting pieces of work is shown. The associated
do-loop has the following form:
!$OMP DO SCHEDULE(GUIDED,256)
do i = 1, 10230
...
enddo
!$OMP END DO