CODEWITHSUNDEEP
fortrangfortran
wxcoder
wxcoder

Reputation: 655

Code crashes at runtime with Absoft Fortran but not with gfortran

I'm running a script that starts similar to the code below. In the past when I ran the script f90 -o fakefile fakefile.f and then ./fakefile it worked, but now it does not and immediately returns a segmentation fault (core dumped) error. When I use gfortran fakefile.f and then ./a.out the code runs fine. I cannot figure out what the discrepancy is between using the two compiling methods.

 program fakefile
   implicit real*8(a-h,o-z)
   parameter(im9=4320,jm9=2160)
   parameter(im1=360,jm1=180)
   parameter(im25=1440,jm25=720)
   parameter(nlt=2)
   real*4 rrs(im9,jm9,nlt)
   real*8 rrsa25(im25,jm25,nlt)
   real*8 area25(im25,jm25,nlt)
   real*8 rrsa1(im1,jm1,nlt)
   real*8 area1(im1,jm1,nlt)
   rrsa1  = 0.0
   area1  = 0.0
   rrsa25 = 0.0
   area25 = 0.0
   rrs = 0.0
   print *, rrs

 end

Upvotes: 1

Views: 428

Answers (1)

casey
casey

Reputation: 6915

Your segmentation fault is due to overlowing the stack with your large arrays. This is a common issue on Absoft and Intel Fortran compilers. For your compiler (Absoft), use the -s flag to tell the compiler to allocate arrays on the heap instead of the stack. The alternative is to increase your stack size limit in your shell (which may be restricted by the administrator).

See the Absoft FAQ: When I declare large arrays (>8 MB of variables), I get a segmentation violation from Linux.

A. Use the "-s" compiler option (static storage) to move the data from the stack to the heap or use the ulimit command (ulimit is a bash command - the csh equivalent to 'ulimit -s' is 'limit stack') to raise the stack size limit

# ulimit -s 8192
# ulimit -s 32768
# ulimit -s 32768

Once raised the limit applies to the current process and any children of that process.

Upvotes: 1

Related Questions