Data Distribution: page_place

• Useful for irregular and dynamically allocated data structures:

• #pragma page_place (A, index-range, thread-id)

• C$SGI PAGE_PLACE (A, index-range, thread-id)

• Note: page_place is useful only for initial placement of the data, it has no effect if pages have been created already

Parameter (n=8*1024*1024)

real*8 a(n)

p=1

c$ p = omp_get_max_threads()

c- distribute using block mapping

npp=(n + p-1)/p ! Number of elements per processor

c$omp parallel do

do I=0,p-1

c$sgi page_place (a(1+I*npp), npp*8, I)

enddo

Previous slide

Next slide

Back to first slide

View graphic version