on Compaq FORTRAN: help to put pressure
The following message is a courtesy copy of an article
that has been posted to comp.os.linux.alpha as well.
Below is a response I received from the Compaq code developers at
University of New Hampshire who are working on the port of the
FORTRAN/C compilers to Linux Alpha. For my particular applications, I
will need to be able to utilize the floating point compiler options.
Otherwise, pprobably man-years of effort would be required to code
around the floating-point overflows, etc., "built-in" to the
very-large application. Although this is the best option, I do not
have the time and/or money to do that now.
I would imagine there are lots of Linux Alpha folks with similar
issues. Speed is nice to have, but for porting large, legacy FORTRAN
code, you also need these floating point options. Note other Unix
compilers, such as AIX xlf, IRIX f77, etc., also have floating-point
options such as these. Also, note that g77 on the Intel platform
allows for these floating-point occurances by default. Also, the
-mieee option with g77 on the Alpha allows for *SOME*, but not all, of
these options, but greatly reduces the efficiency of the Alpha chip in
doing so.
According to the response below, the problem is within Linux, and not
their compilers. Could some of the Linux Alpha developers possibly
get involved and help? Are these developers the best that can be
obtained to solve this problem? It is certainly out of my league!
_______________________________________________________________________
Hello,
Thanks for the question. To answer your question, compaq is still looking
into the problem of handling floating point exception. It is because linux
does not have a robust floating point exception handler compared to UNIX.
Compaq is as of now in no position to promise any other support than
-fpe0 and -fpe3 options. The README file that is installed also says
th same.
Anshul Chadda
University of New HampshireA
----------------------------------------------
Name = James D. Freels
Email = fea@ornl.gov
Brief Problem Description = nonsupported
Architecture = all
Operating System = Linux
Compiler Information = fort
Libraries Used = n/a
Other Information = It is this restriction that really causes the problems:
Please note that CFAL does not contain certain features that
are available in DIGITAL Fortran 90 on Tru64 UNIX Alpha:
~ floating-point exception handling levels 1, 2, and 4
and "-check underflow" are not supported
Problem Category = Compiler Installation
Detailed Description = I have a FORTRAN application used under
DEC-Unix 4.0B that I currently run. It requires the FORTRAN -fpe1
option (and similar options or by default on other computer / OSs
). Without this capability, I cannot even start to beta test the new
compiler (for what I need!)
When will this capability be offered?
----------------------------------------------
Reply to: