[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: 20040629:nf_get_vara_real under Linux

>To: address@hidden
>cc: address@hidden
>From: Ben Foster <address@hidden>
>Subject: nf_get_vara_real under Linux
>Organization: UCAR/Unidata
>Keywords: 200406292115.i5TLFhWb013666

Hi Ben,

> this is a problem regarding netcdf on a Linux box:
> 
> I have a fairly large f90 post-processor code that reads netcdf history
> files. It runs fine on IBM and SGI machines, and I am porting to Linux.
> 
> I finally got it to compile with pgf90 under RedHat 9. Everthing goes
> fine until it comes time to read a var using nf_get_vara_real. If the
> nf_get_vara_real call is not commented out, the program crashes with
> SIGSEGV at the first executable statement of the routine that calls
> nf_get_vara_real (it does not matter what the 1st exec statement is).
> If I do comment out the nf_get_vara_real call, everything works fine,
> I just don't get my array!
> 
> I have pulled the routine out of the large code, and put it in a small
> test program with minor changes, and everything works fine, including
> the nf_get_vara_real. The problem appears to lie in the environment
> of the larger code.
> 
> A run under pgdbg gives the following error:
> 
> Signalled SIGSEGV at 0x401bd89e, function _int_malloc, line 463
> 401bd89e:  89 7a 04                     movl   %edi,4(%edx)
> 
> And the last 6 levels of traceback is as follows:
> 
>    #6  nc_rdfld line: 463 in "nchist_mod.F" address: 0x81c3306 
>      ncid = 17 , name = 0x14b6b680, itime = 2 , nzp = 45 , ixfld = 0 , ier = 
> 0 
>    #5  pgf90_ptr_alloc address: 0x83bf9e1 
>    #4  __hpf_alloc address: 0x83bf614 
>    #3  __hpf_gmalloc_without_abort address: 0x83c61c1 
>    #2  __hpf_malloc_without_abort address: 0x83c5fe4 
>    #1  __malloc address: 0x401bc8a0 
> => #0  _int_malloc address: 0x401bd89e 
> 
> #1-#5 not shown are in my code. Line 463 of nchist_mod.F is the first 
> executable statement of sub nc_rdfld (which has the nf_gat_vara_real call).
> 
> Does this indicate a memory problem?  I have tried "limit data unlimited"
> and "limit stack unlimited". I can send parts of the source code if that
> would be helpful. Thanks for any suggestions,

There was a bug in the PG F90 compiler, version 5.1, on Linux that
might be the cause of the problem you are seeing:

  
http://my.unidata.ucar.edu/cgi-bin/getfile?file=/content/support/help/MailArchives/netcdf/msg05332.html

But if you have version 5.1-5 or later, that's probably not it.  We
would need something that reproduces the problem here to debug it
further.  I hope it's the compiler error that Portland Group has
provided a fix for, otherwise it will require some serious debugging.

--Russ