I need to use a large array, where short integers would suffice.
I believe the correct form (in f90) would be something like this:
integer(kind=2) :: short(500,300)
But I can't recall seeing this ever used in code. Will this work across gfortran,
Intel, PGI whatever?

--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)

On 10/13/2021 6:30 AM, Pedro Valdez wrote:

I need to use a large array, where short integers would suffice.
I believe the correct form (in f90) would be something like this:
integer(kind=2) :: short(500,300)
But I can't recall seeing this ever used in code. Will this work across gfortran,
Intel, PGI whatever?

The kind value of 2 may (or may not) be consistent across compilers. If
you use ISO_Fortran_env in your procedure, then you can replace the "2"
with "int16", if all of the target compilers are new enough to support ISO_Fortran_env.

--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)

Pedro Valdez <pedro1492@lycos.com> schrieb:

I need to use a large array, where short integers would suffice.
I believe the correct form (in f90) would be something like this:
integer(kind=2) :: short(500,300)
But I can't recall seeing this ever used in code. Will this work across gfortran,
Intel, PGI whatever?

Not guaranteed.

You can use either (if your compiler supports C interup)

use, intrinsic :: iso_c_binding, only: c_int16_t
integer(kind=c_int16_t) :: short(500,300)

or

integer, parameter :: ip = selected_int_kind(4)
integer(kind=ip) :: short(500,300)

The second variant gives you four decimal digits, which fits a short.

--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)

On Wednesday, October 13, 2021 at 10:30:14 PM UTC+11, Pedro Valdez wrote:

I need to use a large array, where short integers would suffice.
I believe the correct form (in f90) would be something like this: integer(kind=2) :: short(500,300)
But I can't recall seeing this ever used in code. Will this work across gfortran,
Intel, PGI whatever?

.
You may request the smallest integer storage unit with something like
integer, parameter :: small = selected_int_kind(2)
integer (kind=small) :: k(100,100)
.
where the number "'2" is the number of decimal digits that you would like stored. On systems that support this, you will get a byte of storage.
If you want larger integers in magnitude than 127, change the "2" to "4".
On systems that support this, you will get a two-byte integer.
If none of those is supported, you will get an ordinary 4-byte integer.
This is portable across every system.

--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)

On Wednesday, October 13, 2021 at 4:30:14 AM UTC-7, Pedro Valdez wrote:

I need to use a large array, where short integers would suffice.
I believe the correct form (in f90) would be something like this: integer(kind=2) :: short(500,300)
But I can't recall seeing this ever used in code. Will this work across gfortran,
Intel, PGI whatever?

There are no guarantees in the standard about specific KIND values.

It is not unusual for it to be the number of bytes, (except for COMPLEX,
where it is, then, half the number of bytes). There are also some that
number them sequentially.

In the latter case, you would likely get a reasonable sized small integer.

Mention of short might indicate that you are used to the short type
in C or Java, in which case the C interoperability KIND values would
be a good choice. (As shown by others.) Also, they have been around
long enough to be supported by many compilers.

The ISO_Fortran_end values are newer, and may have less support
in older compilers.

--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)

On Thursday, October 14, 2021 at 8:48:50 AM UTC+11, gah4 wrote:

On Wednesday, October 13, 2021 at 4:30:14 AM UTC-7, Pedro Valdez wrote:

I need to use a large array, where short integers would suffice.
I believe the correct form (in f90) would be something like this: integer(kind=2) :: short(500,300)
But I can't recall seeing this ever used in code. Will this work across gfortran,
Intel, PGI whatever?

There are no guarantees in the standard about specific KIND values.

It is not unusual for it to be the number of bytes, (except for COMPLEX, where it is, then, half the number of bytes). There are also some that
number them sequentially.

In the latter case, you would likely get a reasonable sized small integer.

Mention of short might indicate that you are used to the short type
in C or Java, in which case the C interoperability KIND values would
be a good choice.

Only selected_int_kind is portable across all Fortran compilers
from Fortran 90.

(As shown by others.) Also, they have been around
long enough to be supported by many compilers.

Only selected_int_kind is portable across all Fortran compilers
from and including Fortran 90.

The ISO_Fortran_end values are newer, and may have less support
in older compilers.

Older compilers do not support it at all.

--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)