11.1.089 firstprivate lastprivate

[Hans_J_]

Hi All,

When I try to compile a f90 code with a OpenMP FIRSTPRIVATE LASTPRIVATE construct I get

$ ifort -openmp OpenMP_test.f -lpthread

OpenMP_test.f(28): error #5082: Syntax error, found 'LASTPRIVATE' when expecting one of: PRIVATE FIRSTPRIVATE REDUCTION COPYIN NUM_THREADS SHARED IF DEFAULT , ...

!$OMP PARALLEL PRIVATE(id) FIRSTPRIVATE(a) LASTPRIVATE(a) SHARED(b,junk)

-------------------------------------------^

OpenMP_test.f(28): error #6761: An entity cannot appear explicitly in more than one clause per directive except that an entity can be specified in both a FIRSTPRIVATE and LASTPRIVATE clause.

!$OMP PARALLEL PRIVATE(id) FIRSTPRIVATE(a) LASTPRIVATE(a) SHARED(b,junk)

----------------------------------------^

compilation aborted for OpenMP_test.f (code 1)

OpenMP_test.f(28): error #5082: Syntax error, found 'LASTPRIVATE' when expecting one of: PRIVATE FIRSTPRIVATE REDUCTION COPYIN NUM_THREADS SHARED IF DEFAULT , ...!$OMP PARALLEL PRIVATE(id) FIRSTPRIVATE(a) LASTPRIVATE(a) SHARED(b,junk)-------------------------------------------^OpenMP_test.f(28): error #6761: An entity cannot appear explicitly in more than one clause per directive except that an entity can be specified in both a FIRSTPRIVATE and LASTPRIVATE clause. !$OMP PARALLEL PRIVATE(id) FIRSTPRIVATE(a) LASTPRIVATE(a) SHARED(b,junk)----------------------------------------^compilation aborted for OpenMP_test.f (code 1)

ifort contradicts itself.

The code is below. If I remove the LASTPRIVATE(a) (hence the stupid if (id .EQ. 0) then junk ... code for id=0 to get a's value for id=0) then it compiles and works. Except of course, at the end of the parallel region a does not equal itsvalue for thread id 0, which s the whole point of theLASTPRIVATE(a)

Any ideas?

Thanks, Hans

program OpenMP_test

USE OMP_LIB

implicit none

integer,parameter :: r8=SELECTED_REAL_KIND(12)

real(KIND=r8),allocatable,dimension(:) :: a,b

real(KIND=r8),dimension(5) :: c,d,junk

integer :: nthreads,id,i

allocate(a(5),b(5))

a = 1.0_r8

b = 2.0_r8

c = 1.0_r8

d = 2.0_r8

junk = 0.0_r8

write(6,*) ' '

do i = 1,5

write(6,*) c(i)

end do

write(6,*) ' '

nthreads = omp_get_max_threads()

write(6,*) 'OMP_NUM_THREADS:',nthreads

!$OMP PARALLEL PRIVATE(id) FIRSTPRIVATE(a) LASTPRIVATE(a) SHARED(b,junk)

id = omp_get_thread_num()

a = REAL(id,kind=r8)*a+b

!$OMP CRITICAL

write(6,*) ' '

write(6,*) 'I am : ',id

do i = 1,5

write(6,*) a(i)

end do

write(6,*) ' '

!$OMP END CRITICAL

if (id .EQ. 0) then

junk(1:5) = a(1:5)

end if

!$OMP END PARALLEL

write(6,*) ' and in the end ...'

do i = 1,5

write(6,*) junk(i)

end do

write(6,*) ' '

deallocate(a,b)

stop

end program OpenMP_test

[jimdempseyatthecove]

!$OMP PARALLEL PRIVATE(id) FIRSTPRIVATE(a) LASTPRIVATE(a) SHARED(b,junk)
...
!$OMP END PARALLEL

Has no "last-ness". IOW there is no last interation of a parallel DO, no last section.

LASTPRIVATE

Use the LASTPRIVATE clause on the DO, SECTIONS, PARALLEL DO, and PARALLEL SECTIONS directives to provide a superset of the PRIVATE clause functionality.

For kicks, dummy up a


!$OMP PARALLEL DO PRIVATE(id) FIRSTPRIVATE(a) LASTPRIVATE(a) SHARED(b,junk)
DO I=1,100
... set in value for a
END DO
!$OMP END PARALLEL

and

!$OMP PARALLELSECTIONS PRIVATE(id) FIRSTPRIVATE(a) LASTPRIVATE(a) SHARED(b,junk)
...
!$OMP SECTION
... set in last value
!$OMP END PARALLEL SECTIONS


*** try to code in a manner such that compiler does not optimize out your statement(s) for updating a

Jim Dempsey