Future Topics in OpenMP

Nested parallelism:

• enabled with OMP_NESTED environment variable or the omp_set_nested(1) routine.

• If a PARALLEL directive is encountered within another PARALLEL directive, a new term of threads will be created.

• The new team will contain only one thread unless nested parallelism is enabled.

NUM_THREADS clause:

• One way to control the number of threads used at each level is with the num_threads(integer) clause:

  #pragma omp parallel for num_threads(4)
  for (int i = 0; i < N; i++) {
      #pragma omp parallel for num_threads(total_threads/4)
      for (int j = 0; j < M; j++) {
          // do computations
      }
  }

• The value set in the clause supersedes the value in the environment variable OMP_NUM_THREADS (or that set by omp_set_num_threads(integer))

Orphaned directives:

• Directives are active in the dynamic scope of a parallel region, not just its lexical scope. e.g.

  void foo() {
      // do computations
  }
  #pragma omp parallel
  {
      foo()
  }

• Useful as it allows a modular programming style

• But also can be confusing if the call tree is complicated.

• Extra rules about data scope attributes:

  • Variables in the argument list inherit their data scope attribute from the calling routine.

  • Global variables in C++ and COMMON blocks or module variables in Fortran are shared, unless declared THREADPRIVATE.

  • static local variables in C/C++ and SAVE variables in Fortran are shared.

  • All other local variables are private.

• Binding rules:

  • DO/FOR, SECTIONS, SINGLE, MASTER and BARRIER directives always bind to the nearest enclosing PARALLEL directive.

• Thread private global variables:

  • It can be convenient for each thread to have its own copy of variables with global scope.

  • Outside parallel regions and in MASTER directives, accesses to these variables refer to the master thread’s copy.

  • #pragma omp threadprivate (list)

  • This directive must be at file or namespace scope, after all declarations of variables in list and before any references to variables in list or static variables.

  • Difference between PRIVATE: Reference to StackOverflow

	PRIVATE	THREADPRIVATE
list variable	Any	Declared but not referenced or static
Local to	Region	Thread
Placed	Stack	Heap or TLS
Lifetime	Data scoping clause	Persist across region, recycle along with thread (maybe at the end of program)
Storage-associated	Each thread has a private copy.	Master thread use the original, other threads has a private copy.

• Here is an example of THREADPRIVATE, which illustrates the lifetime of value and the storage association between master thread and program:

static int k;
#pragma omp parallel for num_threads(4) threadprivate(k)
for (int i = 0; i < 16; i++) {
    printf("This is thread %2d with k %2d entered i %2d\n", omp_get_thread_num(), k, i);
}
printf("Original k: %d\n", k);
/* output: 
  This is thread  1 with k  0 entered i  4
  This is thread  1 with k  4 entered i  5
  This is thread  1 with k  5 entered i  6
  This is thread  1 with k  6 entered i  7
  This is thread  2 with k  0 entered i  8
  This is thread  2 with k  8 entered i  9
  This is thread  2 with k  9 entered i 10
  This is thread  2 with k 10 entered i 11
  This is thread  0 with k  0 entered i  0
  This is thread  0 with k  0 entered i  1
  This is thread  0 with k  1 entered i  2
  This is thread  0 with k  2 entered i  3
  This is thread  3 with k  0 entered i 12
  This is thread  3 with k 12 entered i 13
  This is thread  3 with k 13 entered i 14
  This is thread  3 with k 14 entered i 15
  Original k: 3 # exact the same as thread 0
*/

COPYIN clause:

• Allows the values of the master thread’s THREADPRIVATE data to be copied to all other threads at the start of a parallel region.

Timing routines:

• Return current wall clock time (relative to arbitrary origin) with: double omp_get_wtime(void)

• Return clock precision with: double omp_get_wtick(void)

• Timers are local to a thread, so must make both calls on the same thread to get the duration.

• NO guarantee about resolution!

Reference

• Difference between OpenMP threadprivate and private | StackOverflow