Parallel Programming Guide for HP-UX Systems

Parallel synchronization

Synchronizing code

Chapter 8 149

C$DIR NEXT_TASK

CONTRIB2 = 0.0

DO I = 1, N

CONTRIB2 = CONTRIB2 + FUNC2(J)

ENDDO

C$DIR CRITICAL_SECTION (SUM_GATE)

GLOBAL_SUM = GLOBAL_SUM + CONTRIB2

C$DIR END_CRITICAL_SECTION

C$DIR END_TASKS

LOCK = FREE_GATE(SUM_GATE)

Here, both parallel tasks must access the shared GLOBAL_SUM variable.

To ensure that GLOBAL_SUM is updated by only one task at a time, it is

placed in a critical section. The critical sections both reference the

SUM_GATE variable. This variable is unlocked on entry into the parallel

code (gates are always unlocked when they are allocated).

When one task reaches the critical section, the CRITICAL_SECTION

directive automatically locks SUM_GATE. The END_CRITICAL_SECTION

directive unlocks SUM_GATE on exit from the section. Because access to

both critical sections is controlled by a single gate, the sections must

execute one at a time.

Example 8-5 Gated critical sections

Gated critical sections are also useful in loops containing multiple

critical sections when there are dependences between the critical

sections. If no dependences exist between the sections, gates are not

needed. The compiler automatically supplies a unique gate for every

critical section lacking a gate.

The C example below uses gates so that threads do not update at the

same time, within a critical section:

static far_shared float absum;

static gate_t gate1;

int adjb[...];

lock = alloc_gate(&gate1);