On Wed, Oct 14, 2015 at 09:34:48AM +0200, Sebastian Huber wrote:
> /home/EB/sebastian_h/archive/gcc-git/libgomp/fortran.c:28:0:
> /home/EB/sebastian_h/archive/gcc-git/libgomp/fortran.c:73:18: note: expected
> 'int *' but argument is of type 'int32_t * {aka long int *}'
Ugh, wasn't aware that some targets use long int for int32_t :(.
> The following patch fixes the problem, but I am not sure if this is really
> the best way to address this issue:
>
> diff --git a/libgomp/fortran.c b/libgomp/fortran.c
> index ceff9ac..44aaf92 100644
> --- a/libgomp/fortran.c
> +++ b/libgomp/fortran.c
> @@ -481,7 +481,9 @@ omp_get_place_num_procs_8_ (const int64_t *place_num)
> void
> omp_get_place_proc_ids_ (const int32_t *place_num, int32_t *ids)
> {
> - omp_get_place_proc_ids (*place_num, ids);
> + int int_ids;
> + omp_get_place_proc_ids (*place_num, &int_ids);
> + *ids = int_ids;
> }
>
> void
> @@ -505,7 +507,9 @@ omp_get_partition_num_places_ (void)
> void
> omp_get_partition_place_nums_ (int32_t *place_nums)
> {
> - omp_get_partition_place_nums (place_nums);
> + int int_place_nums;
> + omp_get_partition_place_nums (&int_place_nums);
> + *place_nums = int_place_nums;
> }
No, both the above changes are wrong. There is not a single int32_t
written, but could be many more, it is an array of 32-bit integers.
I'd say you just want to cast explicitly,
omp_get_place_proc_ids (*place_num, (int *) ids);
and
omp_get_parition_place_nums ((int *) place_nums);
The reason for int32_t is that on the Fortran side it is integer(kind=4)
and everywhere else for that int32_t is used.
If this works, the patch is preapproved.
As for aliasing, it will always be int stores vs. integer(kind=4) reads,
int32_t is just a type used in the wrappers.
Jakub
