(CC'ed Fortran as this is about a Fortran patch, cf.
https://gcc.gnu.org/pipermail/gcc-patches/2026-June/719430.html
and https://gcc.gnu.org/pipermail/gcc-patches/2026-June/719422.html )
Sandra Loosemore wrote:
This adds support for iterators in map clauses within OpenMP
'target' constructs in Fortran.
Some special handling for struct field maps has been added to libgomp in
order to handle arrays of derived types.
LGTM.
* * *
Remark - carried over from my previous email,
https://inbox.sourceware.org/gcc-patches/[email protected]/
BTW: I was about to suggest that you shouldn't have only iterators that
start with the base address as in 'array[0]' / 'array(lbound(array))',
but it seems as if using 'f()' covers that case sufficiently, even if
that still spans the whole array. - It should be well enough hidden
such that also the other case should work
This still applies - the previous f(i,j) and current i * 4 + j + 1
in "..., to: x(i * 4 + j + 1)%arr(:))"
prevents early folding - such that the code relevant code gets
exercised even if the for i=0:..., j=0:... the lowest index
matches the lower array bound ('x(1)').
Thus, while mapping a subarray with a bit before and after unmapped
would be nicer for completeness, for the code used inside GCC and
libgomp, it should not make a difference. (Likewise for having a
dynamic array where the array bound is not known at compile time.)
* * *
Thanks for the patch work!
Tobias
