Introduction to Concurrency in Programming Languages

238 Introduction to Concurrency in Programming Languages

Listing 11.2: A simple data parallel matrix multiply.

! three 2D arrays : a ,b,c

real , dime nsion (100 ,100) :: a,b , c

! loop counters

integer :: i , j

do i =1 ,100

do j =1 ,100

c(i, j ) = sum ( a(i ,:) * b (:, j ))

end do

end do

The inner loop is then easy to remove and replace with a single concise line of

code. The code for the new data parallel algorithm is shown in Listing 11.2.

What do we gain by this? The code has become simpler, as the innermost

loop is now gone and replaced by an implicit loop due to the use of array

notation. Similarly, we no longer need to explicitly zero out the element ...