Parallel Synchronization Model

• Critical section and Barrier

c$omp parallel private(i,j)shared(a,b,m,n,sum)

do 20 j=1,n

c$omp do

do 10 i=1,m

c$omp critical

sum=sum+a(i,j)

c$omp end critical

10 continue

c$omp barrier

c$omp single

b(j) = sum

c$omp end single

20 continue

c$omp end parallel

J =1

Do

I=1

I=2

I=3

Parallel

J = 2

Do

Mutual exclusion

as threads update sum

(Reduction and Atomic

also available)

Wait until all threads

have finished column

sum. (Barrier actually

built into end do.)

. . .

All threads

barrier

All threads take turns

updating shared sum

Single thread executes

the update.

Previous slide

Next slide

Back to first slide

View graphic version