Initialisation using SUM

Arjen Markus

unread,

Jan 27, 2021, 6:53:19 AM1/27/21

to

Hello,
I wonder if the following code is standard-conforming:

integer :: n
integer, parameter :: nelements = 20
!integer ion_end(nelements)
integer, parameter :: iatomic_number(nelements) = [&
1,2,6,7,8,10,11,12,13,14,15,16,17,18,20,22,24,25,26,28]

integer, parameter :: ion_end(nelements) = [&
(sum(iatomic_number(1:n)), n=1,nelements)]

The idea here is to accumulate the numbers in iatomic_number in another parameter array. The construction is accepted by gfortran 10.2.0 but not by Intel Fortran 2018.

Regards,

Arjen

FortranFan

unread,

Jan 27, 2021, 9:22:28 AM1/27/21

to

On Wednesday, January 27, 2021 at 6:53:19 AM UTC-5, arjen.m...@gmail.com wrote:

> ..

> I wonder if the following code is standard-conforming:
>

>.. The construction is accepted .. not by Intel Fortran 2018.
> ..

For whatever it's worth, I think the code conforms whereas Intel Fortran does not. If you're able to, you should try using Intel's latest - effectively FREE too - compiler option with IFORT Classic in their Intel oneAPI toolkit and if you still notice an issue, post the issue at their Community forum:
https://community.intel.com/t5/Intel-Fortran-Compiler/bd-p/fortran-compiler

Coming back to the Fortran standard, it essentially requires a constant expression in the named constant array evaluation. With constant expression, the standard states in section 10.1.12:

""It is an expression in which each operation is intrinsic, and each primary is ..
..
a reference to a standard intrinsic function that is transformational, other than COMMAND_ARGUMENT_COUNT, GET_TEAM, NULL, NUM_IMAGES, TEAM_NUMBER, THIS_IMAGE, TRANSFER, where each argument is a constant expression."

The intrinsic SUM is a transformational function and the expression in the code in the original post meets the requirements of section 10.1.12, to the best I can assert. Hence my thought Intel Fortran is wrong here.

In the interest of disclosure: I've several bug reports open with Intel for a while now along the same lines.

JRR

unread,

Jan 27, 2021, 9:30:17 AM1/27/21

to

Am 27.01.21 um 12:53 schrieb Arjen Markus:

The code is also accepted by NAG, but refused by nVidia, with a similar
message as for ifort:
NVFORTRAN-S-0155-Intrinsic not supported in initialization: sum (bla.f90: 8

>

--
Juergen Reuter
Theoretical Particle Physics
Deutsches Elektronen-Synchrotron (DESY)
Hamburg, Germany

Arjen Markus

unread,

Jan 27, 2021, 11:43:02 AM1/27/21

to

Thank you! I just tried with the Intel oneAPI tool set - Intel Fortran 2021.1 but that gives exactly the same error messages

Regards,

Arjen

James Van Buskirk

unread,

Jan 27, 2021, 7:03:13 PM1/27/21

to

"Arjen Markus" wrote in message
news:ffa856b8-b678-4e2f...@googlegroups.com...

> integer :: n
> integer, parameter :: nelements = 20
> !integer ion_end(nelements)
> integer, parameter :: iatomic_number(nelements) = [&
> 1,2,6,7,8,10,11,12,13,14,15,16,17,18,20,22,24,25,26,28]

> integer, parameter :: ion_end(nelements) = [&
> (sum(iatomic_number(1:n)), n=1,nelements)]

I tried a workaround:

D:\gfortran\clf\sum_init>type dot_init.f90
integer :: n
integer, parameter :: iatomic_number(*) = [&
1,2,6,7,8,10,11,12,13,14,15,16,17,18,20,22,24,25,26,28]

integer, parameter :: ion_end(*) = [&
(dot_product(iatomic_number,eoshift(reshape([integer::], &
shape=shape(iatomic_number),pad=[0]),-n,1)),
n=1,size(iatomic_number))]

write(*,'(*(g0:1x))') ion_end
end
D:\gfortran\clf\sum_init>gfortran dot_init.f90 -odot_init
dot_init.f90:6:37:

(dot_product(iatomic_number,eoshift(reshape([integer::], &
1
Error: transformational intrinsic 'eoshift' at (1) is not permitted in an
initia
lization expression

But it didn't seem to help :(

James Van Buskirk

unread,

Jan 27, 2021, 7:33:03 PM1/27/21

to

"James Van Buskirk" wrote in message news:rusuvs$9of$1...@dont-email.me...

> I tried a workaround:

Another one:

D:\gfortran\clf\sum_init>type no_eoshift.f90

integer :: n
integer, parameter :: iatomic_number(*) = [&
1,2,6,7,8,10,11,12,13,14,15,16,17,18,20,22,24,25,26,28]

integer, parameter :: ion_end(*) = &

[(dot_product(iatomic_number,reshape([(1,i=1,n)], &
shape=shape(iatomic_number),pad=[0])),n=1,size(iatomic_number))]

write(*,'(*(g0:1x))') ion_end
end

D:\gfortran\clf\sum_init>gfortran no_eoshift.f90 -ono_eoshift
no_eoshift.f90:6:44:

[(dot_product(iatomic_number,reshape([(1,i=1,n)], &
1
Error: Invalid character in name at (1)

Still no joy. BTW, what have you got against K?
https://www.youtube.com/watch?v=fmaZdEq-Xzs

Arjen Markus

unread,

Jan 28, 2021, 5:31:23 AM1/28/21

to

On Thursday, January 28, 2021 at 1:33:03 AM UTC+1, James Van Buskirk wrote:

>
> Still no joy. BTW, what have you got against K?
> https://www.youtube.com/watch?v=fmaZdEq-Xzs

Oh, you mean potassium? I thought you were referring to the K combinator and I couldn't quite figure out how that would apply here :) I will ask my friend who came up with this little problem.

Regards,

Arjen