Reputation: 1446
Why does ifort not parallelize this code? It keeps saying "loop was not parallelized: existence of parallel dependence". I can't understand where the dependence is. gfortran will generate parallel code, but the speed up is not very high.
PROGRAM erat
IMPLICIT NONE
INTEGER*8 :: i, rm, sn=1000000000
LOGICAL*1 , ALLOCATABLE, DIMENSION(:) :: nums
rm = INT( DBLE(sn)**0.5) + 1
ALLOCATE(nums(sn))
nums = .TRUE. !This line not parallelized
PRINT *, 'Doing initial sieve...'
nums(1) = .FALSE.
DO i = 2,rm
nums(i**2:sn:i) = .FALSE. !This line not parallelized
END DO
END PROGRAM erat
Upvotes: 0
Views: 501
Reputation: 21451
That diagnostic is reported against the DO statement. As one specific example:
When i is 2, the loop sets num(8) to false.
When i is 4, the loop also sets num(8) to false.
That's two different iterations of the loop writing to the same memory location.
(The relevant Intel forums are a better place to ask questions that might get into the specifics of the behaviours of their compilers.)
Upvotes: 3