Coarrays and -fdefault-integer-8

Coarrays and -fdefault-integer-8

[Dominique d'Humières]

Hi Damian,

Apparently coarrays show several problems when compiled with -fdefault-integer-8 (pr32770). I have posted at https://gcc.gnu.org/bugzilla/attachment.cgi?id=43622 a preliminary patch fixing some failures.

Could you please have a look at it (and/or your coworkers)? Especially at the patch for gfortran.dg/coarray/atomic_2.f90:
Is is expected that the arguments of atomic_* have to be INTEGER(4)/LOGICAL(4) (atomic_int_kind)?

TIA

Dominique

Re: Coarrays and -fdefault-integer-8

[Dan Nagle]

Hi,

Pardon me for answering a question for Damian,

> On Mar 11, 2018, at 05:37 , Dominique d'Humières <dominiq@lps.ens.fr> wrote:
> 

> Is is expected that the arguments of atomic_* have to be INTEGER(4)/LOGICAL(4) (atomic_int_kind)?

atomic_add        atom must be of atomic_int_kind
atomic_and        atom must be of atomic_int_kind
atomic_cas        atom may be of atomic_Int_kind or atomic_logical_kind
atomic_define     atom may be of atomic_int_kind or atomic_logical_kind
atomic_fetch_add  atom must be of atomic_int_kind
atomic_fetch_and  atom must be of atomic_int_kind
atomic_fetch_or   atom must be of atomic_int_kind
atomic_fetch_xor  atom must be of atomic_int_kind
atomic_or         atom must be of atomic_int_kind
atomic_ref        atom may be of atomic_int_kind or atomic_logical_kind
atomic_xor        atom must be of atomic_int_kind

from 1539-1 2017 (17-007r1) pg 356 - 360 HTH

--

Cheers!
Dan Nagle

Re: Coarrays and -fdefault-integer-8

[Dominique d'Humières]

OK, thanks.

Dominique

> Le 11 mars 2018 à 17:33, Dan Nagle <danlnagle@me.com> a écrit :
> 
> Hi,
> 
> Pardon me for answering a question for Damian,
> 
>> On Mar 11, 2018, at 05:37 , Dominique d'Humières <dominiq@lps.ens.fr> wrote:
>> 
> 
>> Is is expected that the arguments of atomic_* have to be INTEGER(4)/LOGICAL(4) (atomic_int_kind)?
> 
> atomic_add        atom must be of atomic_int_kind
> atomic_and        atom must be of atomic_int_kind
> atomic_cas        atom may be of atomic_Int_kind or atomic_logical_kind
> atomic_define     atom may be of atomic_int_kind or atomic_logical_kind
> atomic_fetch_add  atom must be of atomic_int_kind
> atomic_fetch_and  atom must be of atomic_int_kind
> atomic_fetch_or   atom must be of atomic_int_kind
> atomic_fetch_xor  atom must be of atomic_int_kind
> atomic_or         atom must be of atomic_int_kind
> atomic_ref        atom may be of atomic_int_kind or atomic_logical_kind
> atomic_xor        atom must be of atomic_int_kind
> 
> from 1539-1 2017 (17-007r1) pg 356 - 360 HTH
> 
> --
> 
> Cheers!
> Dan Nagle
> 

Re: Coarrays and -fdefault-integer-8

[Dan Nagle]

Hi,

A question occurring to me after I posted is,
Does -fdefault-integer-8 change the values of atomic_int_kind and atomic_logical_kind ?

I suppose the answer may be hardware-dependent.

> On Mar 11, 2018, at 10:57 , Dominique d'Humières <dominiq@lps.ens.fr> wrote:
> 
> OK, thanks.
> 
> Dominique
> 
>> Le 11 mars 2018 à 17:33, Dan Nagle <danlnagle@me.com> a écrit :
>> 
>> Hi,
>> 
>> Pardon me for answering a question for Damian,
>> 
>>> On Mar 11, 2018, at 05:37 , Dominique d'Humières <dominiq@lps.ens.fr> wrote:
>>> 
>> 
>>> Is is expected that the arguments of atomic_* have to be INTEGER(4)/LOGICAL(4) (atomic_int_kind)?
>> 
>> atomic_add        atom must be of atomic_int_kind
>> atomic_and        atom must be of atomic_int_kind
>> atomic_cas        atom may be of atomic_Int_kind or atomic_logical_kind
>> atomic_define     atom may be of atomic_int_kind or atomic_logical_kind
>> atomic_fetch_add  atom must be of atomic_int_kind
>> atomic_fetch_and  atom must be of atomic_int_kind
>> atomic_fetch_or   atom must be of atomic_int_kind
>> atomic_fetch_xor  atom must be of atomic_int_kind
>> atomic_or         atom must be of atomic_int_kind
>> atomic_ref        atom may be of atomic_int_kind or atomic_logical_kind
>> atomic_xor        atom must be of atomic_int_kind
>> 
>> from 1539-1 2017 (17-007r1) pg 356 - 360 HTH
>> 
>> --
>> 
>> Cheers!
>> Dan Nagle
>> 
> 


--

Cheers!
Dan Nagle

Re: Coarrays and -fdefault-integer-8

[Dominique d'Humières]

> Le 11 mars 2018 à 18:27, Dan Nagle <danlnagle@me.com> a écrit :
> 
> Hi,
> 
> A question occurring to me after I posted is,
> Does -fdefault-integer-8 change the values of atomic_int_kind and atomic_logical_kind ?

AFAICT no, otherwise I would have to patch gfortran.dg/coarray/atomic_2.f90.

> I suppose the answer may be hardware-dependent.

Well, does it exist a platform for which the default integer KIND is not 4? or which does not support INTEGER(8)?

> 
> --
> 
> Cheers!
> Dan Nagle
> 

Cheers,

Dominique

Re: Coarrays and -fdefault-integer-8

[N.M. Maclaren]

On Mar 11 2018, Dominique d'Humières wrote:
>> 
>> A question occurring to me after I posted is, Does -fdefault-integer-8 
>> change the values of atomic_int_kind and atomic_logical_kind ?
>
> AFAICT no, otherwise I would have to patch 
> gfortran.dg/coarray/atomic_2.f90.
>
>> I suppose the answer may be hardware-dependent.
>
> Well, does it exist a platform for which the default integer KIND is not 
> 4? or which does not support INTEGER(8)?

I assume that you mean the size in bytes; not all compilers use the size in
bytes for the KIND.

If the former doesn't exist today, it may well in the future.  Indeed, I
would expect it to.


Regards,
Nick Maclaren.