Hi Honza, Tobias,
Yes, I am willing to help and will do so as soon as my small vacation ends
on Monday.

Regards,
Andre

Andre Vehreschild * ve...@gmx.de
Am 30. September 2020 19:12:48 schrieb Tobias Burnus <tob...@codesourcery.com>:

Hi Honza,

On 9/30/20 6:12 PM, Jan Hubicka wrote:
_gfortran_caf_co_sum (gfc_descriptor_t *a __attribute__ ((unused)),

Should have fnspec
  ".XXWXX"
First dot represents return value, then X is for unused parameters

'X' is definitely wrong. In GCC there is only a stub implementation for
gfortran's coarray (Fortran) implemented. The full version needs
a communication library – such as MPI (Message Passing Interface)
or GASNet or OpenShMem ... Hence, that library is separate. The main point
of this stub library is to provide some means for testing.

See http://www.opencoarrays.org/ and
https://github.com/sourceryinstitute/opencoarrays/
for a (or rather: the) version which actually implements those
library functions.

I would apprechiate help from someone who knows the API to correct the
strings.

@Andre? How about you? ;-)

-  gfor_fndecl_sr_kind = gfc_build_library_function_decl_with_spec (
-     get_identifier (PREFIX("selected_real_kind2008")), ".RR",
+  gfor_fndecl_sr_kind = gfc_build_library_function_decl (
+     get_identifier (PREFIX("selected_real_kind2008")),

(This one is outside CAF.)

@Honza: I want to note that also for user functions, 'fn spec' are
generated, cf. create_fn_spec in trans-types.c – hopefully this one is fine.

Tobias

-----------------
Mentor Graphics (Deutschland) GmbH, Arnulfstraße 201, 80634 München / Germany
Registergericht München HRB 106955, Geschäftsführer: Thomas Heurung,
Alexander Walter

Reply via email to