HP’s implementation of OpenMP

This section discusses HP’s implementation of OpenMP.

Command-line option

HP OpenMP directives are only accepted if the command-line option +Oopenmp is given.




	NOTE: +`Oopenmp` implies +`Onodynsel, +Oparallel,` and +`Onoautopar`.

Default

The default command-line option is +Onoopenmp. If +Oopenmp is not given, all OpenMP directives (c$omp) are ignored.

Optimization levels and parallelism

+Oopenmp is accepted at all optimization levels. However, the following differences exist between Itanium®-based and PA-RISC architectures.

Itanium®-based architectures

OpenMP is accepted at all optimization levels on Itanium®-based architectures. Parallelization directives and worksharing directives can be compiled at any optimization level from +O0 through +O4. Additionally, OpenMP will interoperate with +objdebug, +noobjdebug, and -g. Therefore, source-level debugging of parallelized applications is supported subject to the same restrictions of debugging at +O2 or below.

PA-RISC architectures

The following limitations exist for OpenMP on PA-RISC architectures.

For parallel and work-shared directives (including the clauses for these directives), code is parallelized only at optimization levels +O3 or +O4. The parallel and work-shared directives are listed in Table 4-1 “Parallel and work-shared directives”.

Optimization levels +O0 through +O2

When using optimization levels +O0 through +O2:

All sync and run-time library directives are processed and honored.
Parallel and work-shared directives (including the clauses for these directives) are only processed. While they will return right answers, you will not achieve parallel code. Each thread will run a serial version of the code.

Optimization levels +O3 through +O4

When using optimization levels +O3 and +O4:

All sync and run-time library directives are processed and honored.
Parallel and work-shared directives (including the clauses for these directives) are processed and honored. The compiler will generate the parallel and work-shared code required to go parallel.

Table 4-1 Parallel and work-shared directives

Parallel / work-shared directives	Opt level accepted	Opt level required to achieve parallelism
PARALLEL	+O0, +O1, +O2	+O3, +O4
PARALLEL DO	+O0, +O1, +O2	+O3, +O4
PARALLEL SECTIONS	+O0, +O1, +O2	+O3, +O4
DO	+O0, +O1, +O2	+O3, +O4
SECTION	+O0, +O1, +O2	+O3, +O4
SECTIONS	+O0, +O1, +O2	+O3, +O4

Parallelism

Nested parallelism is now supported dynamically, and with the use of the NUM_THREADS clause on the Parallel directive, finer-grained control of parallelism is possible. This also allows parallelized code to use parallel versions of MLIB.

Additionally, statically-nested parallelization is enabled.

Arrays

Arrays are allowed in reduction clauses as well as scalar variables.

Portable timing routines

There are two portable timing routines:

DOUBLE PRECISION OMP_GET_WTIME()

DOUBLE PRECISION OMP_GET_WTICK()

Nested lock routines

Nested lock routines are as follows:

SUBROUTINE OMP_INIT_NEST_LOCK (NLOCK)
SUBROUTINE OMP_DESTROY_NEST_LOCK (NLOCK)
SUBROUTINE OMP_SET_NEST_LOCK (NLOCK)
SUBROUTINE OMP_UNSET_NEST_LOCK (NLOCK)
INTEGER FUNCTION OMP_TEST_NEST_LOCK (NLOCK)

Additional features

Copyin now allows non-threadprivate objects in a parallel region.
Relaxed reprivatization rules now allow an inner directive to reprivatize a variable privatized in a containing directive.
Privatization of module data is now allowed, as well as privatization of deferred and assumed shape objects.