On 03/16/2017 07:13 AM, Thomas Schwinge wrote:
>> --- a/gcc/gimplify.c
>> +++ b/gcc/gimplify.c
>
>> @@ -6636,9 +6596,9 @@ gimplify_scan_omp_clauses (tree *list_p, gimple_seq
>> *pre_p,
>> ctx = new_omp_context (region_type);
>> ctx->clauses = *list_p;
>> outer_ctx = ctx->outer_context;
>> - if (code == OMP_TARGET && !(lang_GNU_Fortran () && !(region_type &
>> ORT_ACC)))
>> + if (code == OMP_TARGET)
>> {
>> - if (!lang_GNU_Fortran () || region_type & ORT_ACC)
>> + if (!lang_GNU_Fortran ())
>> ctx->target_map_pointers_as_0len_arrays = true;
>> ctx->target_map_scalars_firstprivate = true;
>> }
>
> Is it intentional that you're now no longer enabling
> target_map_pointers_as_0len_arrays for Fortran OpenACC regions? If
> that's not intentional, then I guess we're missing some tree-dump
> scanning test cases to verify the expected behavior?
I took a look back at my notes, and it appears that I was working on
optimizing data movement when I did this backport. This should have been
separate patch.
If you want, I can revert it. Otherwise, I'll wait until we after we
update the semantics of the data clauses to match the OpenACC 2.5 spec
prior to creating test cases to scan for tree dumps.
Cesar
