Re: [tlug] Fortran --> Python (was linux engineer)

[Edward Middleton <edward.middleton@example.com>]

On 06/07/2012 06:52 PM, Josh Glover wrote:
> 2012/6/7 Nicolas Limare<nicolas+tlug@example.com>:
>
> > A clean rewrite with well defined APIs and unit tests would be necessary to
> > make computational research verifiable and reproducible.
>
> That is a very good point. So in (at least) one specific way,
> computers have made scientific research worse.
>
> > I think the superiority is in the compilers, better tuned for unusual
> > HPC architectures than gcc or icc.

I don't know.  When I was doing numerical modeling in university (10 
years ago) it wasn't the compilers so much as the libraries.  If you can 
represent your problem as a matrix operation then pass it to a library 
optimized for the architecture it can handle efficiently distributing 
work among the processors.  A lot of code I saw people writing in 
FORTRAN was written in a way that would be very hard for a compiler to 
run any way accept serially and get the correct result.

Edward