I/O slow (buffered?) ; can I change this?

[Ray Nachlinger]

On Fri, 15 Dec 2006, Rob wrote:

> 
> --- Jerry DeLisle <jvdelisle@verizon.net> wrote:
> 
> > Rob wrote:
> > > Hi,
> > > 
> > > When I compile my software with Gfortran (from
> > > SVN, 4.1.2 20061215 (prerelease)), the print
> > > output to the terminal is extremely slow and
> > > comes in bursts.
> > > I guess Gfortran defaults to some kind of
> > > buffering.
> > > 
> > > When I use another compiler (e.g. Intel's ifort),
> > > then output comes immediately without delay. So
> > > obviously the slow I/O is related to Gfortran.
> > > 
> > > Is there a way to tell Gfortran to stop or reduce
> > > the I/O buffering?
> > 
> > What system are you running on?
> > 
> > Can you give an example of the code producing the
> > output that you are observing?
> 
> This is on an alpha system, for which at present
> the only available compiler is GFortran.
> 
> I myself know almost nothing about fortran, but I
> compile the physics ab initio package VASP written
> in fortran: http://cms.mpi.univie.ac.at/vasp
> (this is commercial software).
> 
> I need the compiler flag "-mfp-trap-mode=su", in
> order to prevent a segfault signal when running
> the executable.
> 
> So I don't have a short piece of code that
> demonstrates the slow I/O. I only compare the
> behaviour of the executable on my alpha system with
> Gfortran, and the same executable on Linux/Pentium
> system compiled with Intel's ifort.
> 
> On the Linux system, the output comes gradually,
> but on the alpha system it comes in bursts.
> 
> Hence I wondered whether a compiler flag could
> change this behaviour.

  At the risk of being a fool I will say that I have
  ported code among compliers, and your complaint
  seems to be a "high class problem". The fact that
  it works at all should elate you, not be grounds
  for complaints.

  Regards,

  Ray