(j3.2006) Compilation cascades

[Bill Long]

Van Snyder wrote:
> Van Snyder wrote:
>
>   
>> We compare .mod files to avoid compilation cascades.
>>
>> I noticed that if I add a PRINT statement to a module procedure, several
>> processors produce a different .mod file.
>>
>> This seems a bit bizarre.  Is it really necessary?
>>
>> Some processors put a time stamp in the .mod file, but we've figured out
>> how to get around that.
>>     
>
> Robert Corbett commented
>
> ! Some compilers put representations of the code for module procedures
> ! in the .mod files they generate.  This allows inlining module
> ! procedures, which are often short.  It is not necessary, but it can
> ! be a big performance advantage.  Of course, if you use such a compiler,
> ! the compilation cascades are pretty much necessary for correct
> ! operation.  If a module procedure that has been inlined has been
> ! modified, all files that include procedure references that
> ! reference one of the module procedures must be recompiled.
>
> That would explain why the .mod file would be different no matter what
> change I make.  Why would the .mod file be different only if I inserted
> a PRINT statement (or deleted them all)?
>
>   

Just a guess (by default we don't create .mod files), but the .mod file 
might include the names of all the external symbols referenced.  A PRINT 
statement will compile into a call to an external library procedure.

Cheers,
Bill



-- 
Bill Long                                   longb at cray.com
Fortran Technical Support    &              voice: 651-605-9024
Bioinformatics Software Development         fax:   651-605-9142
Cray Inc., 1340 Mendota Heights Rd., Mendota Heights, MN, 55120