[Beowulf] newbie

[Reuti]

Am 04.05.2009 um 16:53 schrieb Prentice Bisbal:

> Chris Samuel wrote:
>> ----- "Mike Davis" <jmdavis1 at vcu.edu> wrote:
>>
>>> Also, a particular code may run much better with
>>> one "optimized" compiler than another. I have some
>>> that require PGI to run at all.
>>
>> This is one of our main reasons we maintain multiple
>> commercial compilers - letting our users easily build
>> code that is dependent on a specific compiler.
>>
>> I suspect the cost of the compiler easily outweighs
>> the cost of our time to try and help them get it
>> working (if possible), not to mention the potential
>> opportunity "cost" to the researchers of them giving
>> up on HPC because they can't get it to work and it
>> doesn't occur to them to ask for help.
>
> I second that. On more than one occasion, I spent weeks trying to get
> code to compile using the compilers on hand before giving up and
> spending $$ on a commercial compiler which compiled the code w/o any
> problems.
>
> In an ideal world, the code/Makefiles/configure scripts would written
> such that any standards-compliant compiler would work, but that's not
> always the case.

Another problem: some flag -xzy might be the option -abc with another  
compiler. So, if a makefile assumes a certain compiler (and a certain  
version of it), you will always have to lookup the corresponding  
notation for another compiler, as the compiler developer is free to  
introduce any new feature with a new option flag. I remember to use - 
mp (now called -fp-model) for the Intel compiler, when it was -pc64  
for Portland one. And this already means, that for the same compiler  
the flags change over time.

And even if the code compiles with another compiler: you have to  
validate the results, when the code was specified to be correct with  
a certain compiler other than the used one.

-- Reuti


> -- 
> Prentice
> _______________________________________________
> Beowulf mailing list, Beowulf at beowulf.org sponsored by Penguin  
> Computing
> To change your subscription (digest mode or unsubscribe) visit  
> http://www.beowulf.org/mailman/listinfo/beowulf