forums.silverfrost.com

[pban92]

Previously my program compiled and linked very fast. But for unknown reasons, now this is taking ages. What usually causes such lengthy compiling and linking? Pls comment. Here is the code,

[pban92]

Here are the subroutines as they did not fit in one post.

[JohnCampbell]

do not compile with /check as the .exe file is then 300mb in size.
I'm not sure why it is taking so long to write it out.
If you use /debug then it compiles much quicker.

Also, your use of variable names "core" and "data" is a bit risky as these can be reserved words.
I'd use my_core and my_data, unless you are accessing the FTN95 memory addressing routines ?

[pban92]

Thanks John! I changed \check and tried with both debug.net and debug win32 without knowing what they exactly do. But the compilation and linking worked really fast! (Now I am wondering if one of these option I used before). Also, thanks for the comment on data and core.

[KennyT]

Can I just comment that using "modules within modules within..." causes extremely slow compilation as well.

It used to take about 10 minutes to completely rebuild our application, but since we started using modules (within modules...) it now takes over an hour.

Just thought I'd say, in case anyone else has the same experience (or better still, a solution!!!)

K

[JohnHorspool]

Kenny,

With version 6.0.0 under XP 64bit I can compile 280881 lines of source code (probably over 250000 lines of actual code once comments and blank lines are discounted) in less than 25 seconds.

But I do not have any modules at all, common statements work fine for me.

cheers,
John
_________________
John Horspool
Roshaz Software Ltd.
Gloucestershire

[KennyT]

Yep, we've got over 50Mb of source code.

One trick we've found is that we can overlap the compilation by using two DOS boxes and compiling about half the app in one and half in the other! Brings it down to about 40mins!

K

[JohnHorspool]

Okay I have only 7.3Mb of source code, but I see no reason while the compile time would not simply scale up linearly.

Cheers,
John
_________________
John Horspool
Roshaz Software Ltd.
Gloucestershire

[KennyT]

Oh, it's completely down to having one module "USE"ing a definition from another module file. If I remember correctly, when we just had one set of files with module definitions, none of which "USE"d any of the others, it didn't make much, if any, difference, but as soon as we started building module definitions that "USE"d modules from other files, the compilation speed dropped like the proverbial!

K

[JohnCampbell]

your comment surprises me.

You could consider redesigning your modules or at least grouping all the related modules in one file.

I've been using modules for at least 10 years, mainly to define my in-memory database and I find the heiractical structure very useful.
I have one for parameters, one for type structures and a number for different aspects of the modelling.

I find it an improvement on common / include files.

[KennyT]

thought i'd "bump" this thread to see if Paul fancied investigating our compilation speed problem...

I'm willing to upload our complete source folders if that will help!

K

[davidb]

KennyT

What are you using to determine the order of compilation for the Fortran files?

If you are just using a simple command line (DOS) script which compiles files in the correct order then there shouldn't be any dependence on how you have nested your modules as all the compiler does is read the MOD files.

With nested modules, each MOD file should include the interface from any modules it uses, so there shouldn't be any unnecessary repeat reading of these.
_________________
Programmer in: Fortran 77/95/2003/2008, C, C++ (& OpenMP), java, Python, Perl

[KennyT]

Yes, it's just a DOS script that compiles all the "modules" first, then the functional routines afterwards. But we might not be being super efficient in the order in which we compile the functional source files! As an example, one particular file (size 570k about 23000 lines) takes almost 10 minutes to compile on an i7 CPU! It USEs about a dozen different MODules (which themselves USE others, which, in turn...)

K

[PaulLaidler]

I noted my name appearing in this thread.
Sorry but I don't fancy this one.

For large projects it is worth considering using an IDE and project with dependency checking or using your own MAKE file for something like MK32.exe. However, you may or may not find that Plato can cope.

The idea being that, during development, you only recompile those files that need to be recompiled because of the dependency chain.

[KennyT]

Hi Paul,

I've isolated one source file and the associated MOD files that clearly illustrates the problem, drop me a line if you fancy trying it and I'll zip it up and send it to you.

K