On Fri, Sep 30, 2022 at 12:41:19PM +0200, Tobias Burnus wrote:
> While has_device_addr has been implemented (in GCC 12), updating
> use_device_ptr for Fortran was missed.
>
> This patch fixes it: Removing the restrictions and mapping to
> has_device_addr where applicable.
>
> For use_device_ptr something similar was done, albeit I think
> this has no semantic effect.
>
> And 'device(omp_initial_device)' printed a warning in Fortran.
> (BTW: C/C++ silently accepts any negative value.)
I think that is what the standard wants.
E.g. in 5.2 device Clause chapter, there is just
"If the device_num device-modifier is specified and target-offload-var is not
mandatory,
device-description must evaluate to a conforming device number."
restriction, which is something that can't be checked at compile time,
you don't know if target-offload-var is mandatory or not.
> if (omp_clauses->device)
> - resolve_nonnegative_int_expr (omp_clauses->device, "DEVICE");
> + {
> + resolve_scalar_int_expr (omp_clauses->device, "DEVICE");
> + /* omp_initial_device == 1, omp_invalid_device = -4 (in GCC). */
> + if (omp_clauses->device->expr_type == EXPR_CONSTANT
> + && omp_clauses->device->ts.type == BT_INTEGER
> + && mpz_cmp_si (omp_clauses->device->value.integer, -1) < 0
> + && mpz_cmp_si (omp_clauses->device->value.integer, -4) != 0)
> + gfc_warning (0,
> + "INTEGER expression of DEVICE clause at %L must be non-"
> + "negative or omp_initial_device or omp_invalid_device",
> + &omp_clauses->device->where);
> + }
So I think we should just resolve_scalar_int_expr and be done with that.
Otherwise LGTM.
Jakub
