Run Time Actions

Master process (p0)

mp_setup()

or

!$omp parallel

sproc()

create

shared memory

process group

Serial execution

p1

p2

pn-1

__mp_slave_wait_for_work()

barrier

!$omp parallel

p1 par_loop

p2 par_loop

pn-1 par_loop

p0 par_loop

__mp_simple_sched(par_loop)

Serial execution

__mp_wait_for_completion()

__mp_barrier_nthreads()

__mp_barrier

_master()

When the parallel threads are spawn:

• they spin in __mp_slave_wait_for_work()

Master enters parallel region:

• passes the address of the _mpdo subroutine to the slaves

• at the end of parallel region all threads synchronize

• slaves go back to spin in __mp_slave_wait_for_work()

• master continues with serial execution

Previous slide

Next slide

Back to first slide

View graphic version