Parallel Programming Guide for HP-UX Systems

Parallel synchronization

Synchronizing code

Chapter 8 155

#pragma _CNX loop_parallel(ordered,ivar=i)

for(i=0;i<99;i++) {

#pragma _CNX ordered_section(ordgate)

for(j=0;j<m;j++) {

a[i][j] = a[i+1][j];

}

#pragma _CNX end_ordered_section

}

In this approach, each thread executes the entire j loop each time it

enters the ordered section, allowing the i loop to terminate normally

regardless of the number of threads available.

Another approach is to manually interchange the i and j loops, as shown

in the following Fortran example:

DO J = 1, M

LOCK = UNLOCK_GATE(ORDGATE)

C$DIR LOOP_PARALLEL(ORDERED)

DO I = 1, 99

C$DIR ORDERED_SECTION(ORDGATE)

A(I,J) = A(I+1,J)

C$DIR END_ORDERED_SECTION

ENDDO

Here, the I loop is parallelized on every iteration of the J loop. The

ordered section is not isolated from its parent loop, so the loop can

terminate normally. This example has added beneﬁt; elements of A are

accessed more efﬁciently.