https://gcc.gnu.org/bugzilla/show_bug.cgi?id=118937
Bug ID: 118937
Summary: DO CONCURRENT: Add a warning if 'SHARED' (or
LOCAL(_INIT)) is not specified but multiple loop
iterations modify a variable
Product: gcc
Version: 15.0
Status: UNCONFIRMED
Keywords: diagnostic
Severity: normal
Priority: P3
Component: fortran
Assignee: unassigned at gcc dot gnu.org
Reporter: burnus at gcc dot gnu.org
Target Milestone: ---
EXPECTED: Add a WARNING (-Wsurprising?) for the following.
In my opinion, the following code is invalid - in the sense that the code
expects that all of 'A' is defined at the end of the loop/subroutine.
However, as multiple loop iterations (namely: all) modify the same variable
(here: A) and no locality has been specified, A is undefined at the end of the
loop.
On the other hand, this pattern is very common and could not be implemented in
Fortran 2008 (= added 'do concurrent') at all, as 'shared' is new in 2018.
subroutine init(A)
implicit none
integer, intent(out) :: A(:)
integer :: i
do concurrent (i=1:size(A)) ! shared(A)
A(i) = i
end do
! without shared, A is undefined
end
I think we should detect 'var_without_locality = …' and possibly also other
variable definition contexts (e.g. intent(out) to pure function)
where this code is executed unconditionally.
Namely, the following would be okay:
do concurrent (i=1:size(A))
if (i == 1) &
A(i) = i
end do
as it only modifies the variable in one loop.
Specification has (here 2023; 2008 looks similar):
" 11.1.7 DO construct
> …
> 11.1.7.5 Additional semantics for DO CONCURRENT constructs
> …
> If a variable has unspecified locality, if it is referenced in an iteration
> it shall either be previously defined during that iteration, or shall not be
> defined or become undefined during any other iteration; if it is defined or
> becomes undefined by more than one iteration it becomes undefined when the
> loop terminates
