On 9/15/20 7:03 PM, Jakub Jelinek wrote:
On Tue, Sep 15, 2020 at 12:48:59AM +0200, Tobias Burnus wrote:
+ bool has_nullptr;
+ size_t j;
+ for (j = 0; j < n->tgt->list_count; j++)
+if (n->tgt->list[j].key == n)
+ {
+has_nullptr
On Tue, Sep 15, 2020 at 12:48:59AM +0200, Tobias Burnus wrote:
> + bool has_nullptr;
> + size_t j;
> + for (j = 0; j < n->tgt->list_count; j++)
> + if (n->tgt->list[j].key == n)
> + {
> + has_nullptr = n->tgt->list[j].has_null_
On Tue, Sep 15, 2020 at 12:48:59AM +0200, Tobias Burnus wrote:
> PR fortran/96668
> * gomp-constants.h (GOMP_MAP_ALWAYS_POINTER_P):
Missing description (Define. ?)
>
> libgomp/ChangeLog:
>
> PR fortran/96668
> * libgomp.h (struct target_var_desc):
> * target.c (gomp
On 9/14/20 1:03 PM, Jakub Jelinek wrote:
+
+ if (flag_openmp && !(TYPE_QUALS (TREE_TYPE (ptr)) & TYPE_QUAL_RESTRICT))
+always_modifier = true;
I think we don't want to depend on flag_openmp here, ... Guess for now
an openacc flag with (ctx->region_type & ORT_ACC) != 0 passed to it
seems
On Mon, Sep 14, 2020 at 09:50:08AM +0200, Tobias Burnus wrote:
> --- a/gcc/fortran/trans-openmp.c
> +++ b/gcc/fortran/trans-openmp.c
> @@ -1357,6 +1357,15 @@ gfc_omp_finish_clause (tree c, gimple_seq *pre_p)
>tree type = TREE_TYPE (decl);
>tree ptr = gfc_conv_descriptor_data_get (de
Second version, new is the second testcase, the change in omp-low.c,
and around the two asserts in libgomp/target.c
The new version now also handles 'omp declare target' variables. This
usage occurs in five SPEC ACCEL testcases (all without 'always' clause),
hence, real-world code seems to expect
This is a first attempt to improve the OpenMP mapping for allocatables
and pointers; there are some more issues – cf. PR and for scalars
PR 97021.
In real world code, a usage like the following is not uncommon:
real, allocatable :: A(:,:)
!$omp target enter data map(to: A)
This maps an unalloca
