Parallel Programming Guide for HP-UX Systems

Parallel synchronization

Synchronizing code

Chapter 8148

Synchronizing code

Code containing dependences are parallelized by synchronizing the way

the parallel tasks access the dependence. This is done manually using

the gates, barriers and synchronization functions discussed earlier in

this chapter, or semiautomatically using critical and ordered sections,

described in the following sections.

Using critical sections

The critical_section example shown below isolates a single critical

section in a loop, so that the critical_section directive does not

require a gate. In this case, the critical section directives automate

allocation, locking, unlocking and deallocation of the needed gate.

Multiple dependences and dependences in manually-deﬁned parallel

tasks are handled when user-deﬁned gates are used with the directives.

Example 8-4 critical_section

The following Fortran example, however, uses the manual methods of

code synchronization:

REAL GLOBAL_SUM

C$DIR FAR_SHARED(GLOBAL_SUM)

C$DIR GATE(SUM_GATE)

LOCK = ALLOC_GATE(SUM_GATE)

C$DIR BEGIN_TASKS

CONTRIB1 = 0.0

DO J = 1, M

CONTRIB1 = CONTRIB1 + FUNC1(J)

ENDDO

C$DIR CRITICAL_SECTION (SUM_GATE)

GLOBAL_SUM = GLOBAL_SUM + CONTRIB1

C$DIR END_CRITICAL_SECTION