forums.silverfrost.com

fmolina · Posted: Fri Oct 16, 2009 5:47 pm Post subject: Problem with compiler option

Hello all,

I have just downloaded FTN95 personal, installed on Win XP x64 SP2. It seems to work mostly fine; however, I tried to compile code with a lot of declarations such as

PaulLaidler · Posted: Fri Oct 16, 2009 6:49 pm Post subject:

When I put this into a short program I don't get an error.
Maybe you need to post a larger sample.

fmolina · Posted: Fri Oct 16, 2009 8:53 pm Post subject:

This is the whole test program, not much larger indeed:

PaulLaidler · Posted: Sat Oct 17, 2009 8:10 am Post subject:

This program works fine for me but I am not using a 64bit OS.
Can anyone else test this on a 64bit OS?

fmolina · Posted: Sat Oct 17, 2009 8:29 pm Post subject:

Hello Paul,

I have installed ftn95 in two other machines, both running Win XP 32 bits SP3, with the same result. The option /ALT_KINDS seems to be ignored; either specifying it or not, the result is the same:

PaulLaidler · Posted: Sun Oct 18, 2009 9:04 am Post subject:

I can only think of two possible reasons for the failure.

1. You are accessing an old version of the compiler. The version appears when you type FTN95 on a command line.

2. The option /ALT_KINDS is not getting through to the compiler. If you also add /EXPLIST to the command line then the .lis file generated will provide a list of the options that the compiler is using.

fmolina · Posted: Sun Oct 18, 2009 8:52 pm Post subject:

The .lis file generated is the following:

As it can be seen, version is 5.30.0 and ALT_KINDS is listed as in effect. Besides, real*8 is not a problem.

PaulLaidler · Posted: Mon Oct 19, 2009 7:05 am Post subject:

Looks like you need to wait for version 5.4 of FTN95. This is available for download except for the personal edition which will be made available in due course.

fmolina · Posted: Mon Oct 19, 2009 2:36 pm Post subject:

OK, I understand. Any guess about when the 'due course' will arrive?

Thanks for the attention

Fernando

silverfrost · Posted: Mon Oct 19, 2009 7:07 pm Post subject:

Version 5.40 is available now. I have put a direct link on the download page to our MediaFire account that we use for the Express version. The problem with cnet and tucows is that they can take weeks just to process an update (unless you pay $1000 and then it takes 10 minutes).

Edit: the link is http://www.silverfrost.com/32/ftn95/ftn95_personal_edition.aspx

fmolina · Posted: Mon Oct 19, 2009 9:10 pm Post subject:

Excellent!

Thanks a lot

JohnCampbell · Joined: 16 Feb 2006 Posts: 2554 Location: Sydney

Why not try

real*8 a

Having the value of kind not being defined in the standard has always looked as a "one step foward, one step back" approach.
The standards committee has always frowned on "real*8", which I have said many times is concise and clear definition of the number being used.

My Ver 5.30.0 also does not work for /alt_kind. I'm not sure if it worked in the past for variable definitions, such the example above.
Constants, such as 2.1_8 appear to compile ok.

John

fmolina · Posted: Tue Oct 20, 2009 3:34 am Post subject:

Because I am trying FTN95 with programs that heavily use code from netlib, which in f90 (for example minpack.f90) use kind = 8 all the time. Besides, as far as I know, is the most common way to use 'kind'.

JohnCampbell · Joined: 16 Feb 2006 Posts: 2554 Location: Sydney

I am surprised that Netlib have general use of kind as "real (kind = Cool

" as although this syntax is valid in the standard, it is definately not portable.
I am not sure how many other fortran 90+ compilers, apart from Salford, who don't use the value of kind as the number of bytes, but it is definately not a requirement of the standard.
As I have only compiled in a Control Data / Prime / Vax / Apollo / Sun / Wintel environment, I am not aware of any compiler since Control Data that does not support Real*8
As I have limited experience of the high performance super-computers and workstations, I'd be interested to know of those that don't support real*8.

If you ask Fortran Standards comittee, they will recomend that you use a module to define the value of kind, by using SELECTED_REAL_KIND and specify a precision. To me it's a total "waste of space" as there are not that many different values of precision that are provided by all the hardware/software manufactures available.

John

fmolina · Posted: Tue Oct 20, 2009 1:39 pm Post subject:

I agree that real*8 is simpler than any use of "kind". But this is what I have found. Regarding SELECTED_REAL_KIND, I only know of one package (Ronald Cools' cubpack.f90) which uses it.
Besides the old days of the PDP-11, my experience with fortan compilers is in the PC environment, either windows or, to some extent, linux, and real*8 and the like are always supported, as far as I know.

Fernando