<PREV> <INDEX> <NEXT>

OpenMP

Multi-threading layer on top of compilers:

• hides threading details
• best use for programs where most time is spent in iterations
  
• implementation depends on the compiler
• compatible with compilers that are not OpenMP aware
• paralelizes loops (e.g. for) -> find suitable algorithms
  
• complete documentation: openmp.org and compilers
  

Restrictions:

• The loop variable must be of type signed integer
• The comparison operation must be in the form
  loop_variable <, <=, >, or >= loop_invariant_integer
• The third expression or increment portion of the for loop must be either integer addition or integer subtraction and by a loop invariant value.
• If the comparison operation is < or <=, the loop variable must increment, if the comparison operation is > or >=, the loop variable must decrement.
• The loop must be a basic block.

Implementation:
#pragma omp description clause

Ex: #pragma omp parallel for private(x)

Scheduling:
All iterations take the same time -> uniformly distributed
Compromise: performance vs memory conflicts
Clause: schedule(kind, chunk size)
   kind = static, dynamic, guided, runtime

Issues:

• Data dependencies and race conditions -> independent iterations
• Managing shared and private data:
• Declare variable inside the loop
• Use private clause
• Reductions (e.g. +=): use the reduction clause

• documentation -> http://gcc.gnu.org/onlinedocs/libgomp/
• environment variables -> customize behavior prior to start
        OMP_DYNAMIC, OMP_NUM_THREADS, ...
  
• runtime routines -> adjust behavior while running
        opm_get_*, omp_set_*