Chapter 3 Implementation-Defined Behaviors

This chapter notes specific behaviors in the OpenMP 3.0 specification that are implementation dependent.

3.1 Task Scheduling Points

Task scheduling points in untied task regions occur at the same points as in tied task regions. So within untied task regions, task scheduling points only appear in the following:

• encountered task constructs

• encountered taskwait constructs

• encountered barrier directives

• implicit barrier regions

• at the end of the untied task region

3.2 Memory Model

There is no guarantee that memory accesses by multiple threads to the same variable without synchronization are atomic with respect to each other. Several implementation-dependent and application-dependent factors affect whether accesses are atomic or not. Some variables might be larger than the largest atomic memory operation on the target platform. Some variables might be mis-aligned or of unknown alignment and the compiler or the run-time system may need to use multiple loads/stores to access the variable. Sometimes there are faster code sequences that use more loads/stores.

3.3 Internal Control Variables

The following internal control variables are defined by the implementation:

• nthreads-var: Controls the number of threads requested for encountered parallel regions. The initial value of nthreads-var is 1.

• dyn-var: Controls whether dynamic adjustment of the number of threads is enabled for encountered parallel regions. The initial value of dyn-var is TRUE (that is, dynamic adjustment is enabled).

• run-sched-var: Controls the schedule that the runtime schedule clause uses for loop regions. The initial value of run-sched-var is static with no chunk size.

• def-sched-var: Controls the implementation defined default scheduling of loop regions. The initial value of def-sched-var is static with no chunk size.

• stacksize-var: Controls the stack size for threads that the OpenMP implementation creates. The initial value of stacksize-var is 4 MegaBytes for 32-bit applications and 8 MegaBytes for 64-bit applications.

• wait-policy-var: Controls the desired behavior of waiting threads. The initial value of wait-policy-var is PASSIVE.

• thread-limit-var: Controls the maximum number of threads participating in the OpenMP program. The initial value of thread-limit-var is 1024.

• max-active-levels-var: Controls the maximum number of nested active parallel regions. The initial value of max-active-levels-var is 4.

3.4 Dynamic Adjustment of Threads

The implementation provides the ability to dynamically adjust the number of threads. Dynamic adjustment is enabled by default. Set the OMP_DYNAMIC environment variable to FALSE, or call the omp_set_dynamic() routine with the appropriate argument, to disable dynamic adjustment.

When a thread encounters a parallel construct, the number of threads delivered by this implementation is determined according to Algorithm 2.1 pp. 35-36 in the OpenMP 3.0 Specification. In exceptional situations, such as when there is a lack of system resources, the number of threads supplied will be less than described in Algorithm 2.1. In these situations, if SUNW_MP_WARN is set to TRUE or a callback function is registered via a call to sunw_mp_register_warn(), a warning message will be issued.

3.5 Loop Directive

The integer type used to compute the iteration count of a collapsed loop is long.

The effect of the schedule(runtime) clause when the run-sched-var internal control variable is set to auto is static with no chunk size.

3.6 Constructs

3.6.1 SECTIONS

The structured blocks in the sections construct are assigned to the threads in the team in a static with no chunk size fashion, so that each thread gets an approximately equal number of consecutive structured blocks.

3.6.2 SINGLE

The first thread to encounter the single construct will execute the construct.

3.6.3 ATOMIC

The implementation replaces all atomic directives by enclosing the target statement with a special, named critical construct. This will enforce exclusive access between all atomic regions in the program, whether or not these regions update the same or different storage locations.

3.7 Routines

3.7.1 omp_set_schedule

The behavior for the Sun-specific sunw_mp_sched_reserved schedule is the same as static with no chunk size.

3.7.2 omp_set_max_active_levels

If omp_set_max_active_levels()is called from within an active parallel region, then the call will be ignored. A warning message will be issued if SUNW_MP_WARN is set to TRUE or a callback function is registered by a call to sunw_mp_register_warn().

If the argument to omp_set_max_active_levels() is not a non-negative integer, then the call will be ignored. A warning message will be issued if SUNW_MP_WARN is set to TRUE or a callback function is registered by a call to sunw_mp_register_warn().

3.7.3 omp_get_max_active_levels

omp_get_max_active_levels() can be called from anywhere in the program. The call will return the value of the max-active-levels-var internal control variable.

3.8 Environment Variables

3.9 Fortran Issues

The following apply to Fortran only.

3.9.1 THREADPRIVATE Directive

If the conditions for values of data in the threadprivate objects of threads (other than the initial thread) to persist between two consecutive active parallel regions do not all hold, then the allocation status of an allocatable array in the second region might be "not currently allocated".

3.9.2 SHARED Clause

Passing a shared variable to a non-intrinsic procedure may result in the value of the shared variable being copied into temporary storage before the procedure reference, and back out of the temporary storage into the actual argument storage after the procedure reference. This copying into and out of temporary storage can occur only if conditions a, b, and c in OpenMP 3.0 Specification, section 2.9.3.2 on p. 88 hold,namely: .

• The actual argument is one of the following:

  • A shared variable

  • A subobject of a shared variable

  • An object associated with a shared variable

  • An object associated with a subobject of a shared variable

• The actual argument is also one of the following:

  • An array section

  • An array section with a vector subscript

  • An assumed-shape array

  • A pointer array

• The associated dummy argument for this actual argument is an explicit-shape array or an assumed-size array.

3.9.3 Runtime Library Definitions

Both the include file omp_lib.h and the module file omp_lib are provided in the implementation.

On Solaris platforms, the OpenMP runtime library routines that take an argument are extended with a generic interface so arguments of different Fortran KIND types can be accommodated.