Fortran 2008 DATA statement restrictions

Fortran 2008 DATA statement restrictions

[FX Coudert]

[-- Attachment #1: Type: text/plain, Size: 975 bytes --]

Hi,

I was looking at our table for Fortran 2008 conformance (https://gcc.gnu.org/wiki/Fortran2008Status) and we really have only a few items missing. One in particular is: "Data statement restrictions lifted”.

Quoting the document:

> Subscripts and nested implied-do limits in a data statement can be any constant expression instead of being limited to combinations of constants, implied-do variables, and intrinsic operations.

Unless I misunderstand this quote, I think we have that already. Maybe we always had that? I have run the attached test program, which uses non-trivial constant expressions, and it passes fine.

So I plan to add that to our testsuite, and change the wiki. But I would like someone else to confirm: do you understand this the same way I do? Do you have ideas of other ways to test it, tricky expressions, etc? Are there other ways that "Subscripts and nested implied-do limits” may appear in a DATA statement?

Thanks,
FX



[-- Attachment #2: data.f90 --]
[-- Type: application/octet-stream, Size: 772 bytes --]

! { dg-do run }
!
! Checking for "The new features of Fortran 2008" feature 5.6

  implicit none
  integer :: a(6)
  integer :: b(6)
  integer(kind=4) :: i

  ! Fortran 2008: Subscripts in a data statement can be any constant expression
  data a(kind("foo")) / 1 /
  data a(sum([1, 2, 3]) / 3) / 2 /
  data a(len("foo")) / 3 /
  data a(kind(i)) / 4 /
  data a(int(7.0 * atan(1.0)):6) / 5, 6 /

  ! Fortran 2008: nested implied-do limits in a data statement can be any constant expression
  data (b(i), i = kind("foo"), sum([-1, 1, 2])) / 1, 2 /
  data (b(i), i = len("foo"), kind(i)) / 3, 4 /
  data (b(i), i = int(7.0 * atan(1.0)), 6) / 5, 6 /

  ! Check that data was correctly filled
  if (any(a /= [(i, i = 1, 6)])) stop 1
  if (any(b /= [(i, i = 1, 6)])) stop 1

end

Re: Fortran 2008 DATA statement restrictions

[Harald Anlauf]

Hi FX!

Am 10.06.23 um 18:22 schrieb FX Coudert:
> Hi,
>
> I was looking at our table for Fortran 2008 conformance (https://gcc.gnu.org/wiki/Fortran2008Status) and we really have only a few items missing. One in particular is: "Data statement restrictions lifted”.
>
> Quoting the document:
>
>> Subscripts and nested implied-do limits in a data statement can be any constant expression instead of being limited to combinations of constants, implied-do variables, and intrinsic operations.
>
> Unless I misunderstand this quote, I think we have that already. Maybe we always had that? I have run the attached test program, which uses non-trivial constant expressions, and it passes fine.
>
> So I plan to add that to our testsuite, and change the wiki. But I would like someone else to confirm: do you understand this the same way I do? Do you have ideas of other ways to test it, tricky expressions, etc? Are there other ways that "Subscripts and nested implied-do limits” may appear in a DATA statement?

There is a meta-bug:

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=33056

and in particular one PR about vector subscripts:

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=49588

This one gives an ICE:

f951: internal compiler error: TODO: Vector sections in data statements

> Thanks,
> FX
>
>

Re: Fortran 2008 DATA statement restrictions

[FX Coudert]

Hi,

Thanks Harald. I see there are bugs open, but actually not that many, compared to other areas of the compiler :)
I think the code I sent was not really covered in the testsuite, so I have committed it (after testing on x86_64-linux).
https://gcc.gnu.org/git/?p=gcc.git;a=commit;h=a54c71ccc24874ce79019b51150b3a00b7c785cd

Best,
FX