Re: Is Linux Unix?
On Fri, Jul 23, 2004 at 07:54:12PM -0700, Ryo Furue wrote:
> ... I haven't used the GNU Fortran 95, but I'm 98% sure that it's almost
> unusable at this stage. Since I've had *enough* unpleasantness with new
> compilers, I'm very reluctant to give the GNU compiler a try at this stage.
Is your program concerned with a scientific question? What area of science?
I'm a retired physicist, and have some time to spend on issues that others
might feel are wastes of time. Could you share your code with me and let
me see what I can make of it using GNU Fortran? (I admit to being a bit
rusty on Fortran, because I've pretty much switched to C++ in order to gain
access to the STL.) I do known how to report bugs within Debian, and other
low level stuff about getting along in this environment.
> For example, I had to report a couple of internal compiler errors for the
> then new Sun Fortran 95 compiler. I again reported a few internal compiler
> errors for the 2nd version of Sun F95 compiler, and it emitted obviously
> wrong code in certain cases if you switch on the optimizer. Then the initial
> releases of the 3rd version of the same compiler gave me a few internal
> compiler errors. I think the Sun compiler at last reached a mature stage
> at the later releases of the 3rd version. I had very similar experience with
> another vendor, too.
>
> Also, I heard that the optimizer for the GNU F95 compiler is almost nothing
> at this stage. I recently found one of the programs I use runs six times
> faster with -O3 than without on an SGI machine (I'm not using automatic
> parallization, so the speedup is a genuine improvement due to the optimizer).
> I mean, the optimizer is really important for the type of computations I'm
> doing. Speed comparisons by other people between the GNU Fortran 95 and the
> Intel compiler show similar difference in performance.
>
> > (Or migrating to C/C++?)
>
> That's out of question! :) For numerical computation, Fortran 95 is "better"
> than C++. (I have plenty of experience in C++ programming.) Sure I greatly
> miss C++'s objects and templates and other goodies. But codes for numerical
> computation can be much more "easily" written in Fortran 95. That's the most
> important point for a non-proffessional programmer like me.
I, also, am a non-professional programmer. Some things are surely more
compact in Fortran than in C/C++, but the floating point in both F and
C is surely IEEE standard by now. Or is there some reason for
preferring some other type of floating point that is still available
in IFC? In the early days of C, floating point was *only* double
precision, and consequently, there was a serious speed deficit.
But I'm interested in investigating in a somewhat serious way, the
actual differences, today. Maybe its a bigger project than I can
actually handle, but I don't think so.
--
Paul E Condon
pecondon@mesanetworks.net
Reply to: