Re: WRAPIT/pgf90 problem

Dear Mary,

On Thu, 29 Mar 2007, Mary Haley wrote:
> Qingtao,
>
> Two things come to mind here: one is if you are using -byteswapio
with
> one compiler, you should use this with the other compiler so that they
> both have the same option set. Otherwise, you may be creating binary
> files one way, and trying to read them another way.
>
> Another thing we noticed with g77 on 64-bit Linux systems is that data
> is written with 8-byte control words, whereas other compilers, like
> pgf90, expect 4-byte control words.

Using g77 becomes tricky under the above reasons you mentioned.

>
> If you suspect that the second case may be the cause of the problem,
> then you may want to switch back to using pgf90 with WRAPIT, and
> possibly try removing the -byteswapio option to see if this makes a
> difference.

This is what I had tried for quite a while. It consistently complained
I/O error "PGFIO-F-201/OPEN/unit=15/illegal value for specifier", after
I played with WRAPIT script including byteswapio option. This might be a
problem with regard to I/O bwtween NCL and PGF. My colleague did make
the code work a while ago with an old version of PGF.

Thank you,

Qingtao

> --Mary
>
> On Thu, 29 Mar 2007, Qingtao Song wrote:
>
>>
>> Dear Mary,
>>
>> Thanks a lot for your help. Please see my responses below.
>>
>> On Thu, 29 Mar 2007, Mary Haley wrote:
>>
>>> On Thu, 29 Mar 2007, Qingtao Song wrote:
>>>
>>>>
>>>> Hi,
>>>>
>>>> I met a problem with either WRAPIT or pgf90 on my Linux RHEL4 x86_64.
>>>> Both the NCL and NCARG are in their latest version. The pgf90 version is
>>>> 6.11.
>>>>
>>>> The Fortran subroutine is to write a binary file that is called from NCL.
>>>> I have listed below a few steps that illustrate how I encountered the
>>>> problem:
>>>>
>>>> -----step 1------------
>>>> [qsong_at_wallaroo ncl_for]$ WRAPIT -pg write_dd_qsong.f
>>>>
>>>> WRAPIT Version: 090806
>>>> /home/server/local/apps/pgicdk-rhel4/linux86-64/current/bin/pgf90
>>>> COMPILING write_dd_qsong.f
>>>> LINKING
>>>> /usr/bin/ld: WRAPIT.o: relocation R_X86_64_32 against `a local symbol'
>>>> can not be used when making a shared object; recompile with -fPIC
>>>> WRAPIT.o: could not read symbols: Bad value
>>>> END WRAPIT
>>>>
>>>> ---step 2--------------
>>>> Then it is OK after -fPIC is added
>>>>
>>>> [qsong_at_wallaroo ncl_for]$ WRAPIT -pg -fPIC write_dd_qsong.f
>>>>
>>>> WRAPIT Version: 090806
>>>> /home/server/local/apps/pgicdk-rhel4/linux86-64/current/bin/pgf90
>>>> COMPILING write_dd_qsong.f
>>>> LINKING
>>>> END WRAPIT
>>>>
>>>> ---step 3-------------
>>>> ncl create_SST_bumps_5x5.ncl
>>>> Copyright (C) 1995-2006 - All Rights Reserved
>>>> University Corporation for Atmospheric Research
>>>> NCAR Command Language Version 4.2.0.a034
>>>> The use of this software is governed by a License Agreement.
>>>> fatal:Syntax Error in block, block not executed
>>>> PGFIO-F-201/OPEN/unit=15/illegal value for specifier.
>>>> File name = qsong.dat
>>>> In source file write_dd_qsong.f, at line number 46
>>>>
>>>> this is line 46: open(unit=15,file='qsong.dat', form='unformatted')
>>>> ----end of step 3-----------
>>>>
>>>> The FORTRAN subroutine itself works well. The problem looks like an I/O
>>>> error, but it is likely that WRAPIT did not do the right job. If wrapped
>>>> with g77, my ncl script did output a binary file but the binary file
>>>> could not be unrecognized by WRF (maybe due to endianess problem or the
>>>> WPS compiled with pgf90).
>>>>
>>>> Thanks for your help,
>>>>
>>>> Qingtao
>>>>
>>>
>>> Dear Qingtao,
>>>
>>> Just FYI: in the next release of NCL, WRAPIT will have -fPIC set by
>>> default for gcc compilers. Right now, with the version you have, you
>>> saw that you need to explicitly include -fPIC.
>>
>> For my test, WRAPIT with gcc didn't complain for cases being with -fPIC or
>> without it.
>>
>>>
>>> As for your problems with the PGI compilers, I'm not sure what's going
>>> on. I googled this error message and came up with what looked like
>>> an email discussion where somebody else had the same question (and there
>>> was a response), but unfortunately the links were bad.
>>>
>>> There may be an incompatibility between using a gcc-compiled version
>>> of NCL that is trying to load a shared object created by a PGI
>>> compiler.
>>
>> This is what I am mostly worried about. The Fortran code I wrote is to
>> output a binary file.
>>
>>>
>>> So, for now, let's focus on trying to get it working using gcc.
>>>
>>> Can you explain what you mean by WRF is not recognizing the
>>> binary file? Also, what do you mean by "WPS compiled with pgf90"?
>>> There *could* be a problem if you are taking this binary to a system
>>> that has a different endianness from the machine you built it on.
>>> The x86_64 system is a little endian machine. What system are you
>>> moving this binary to?
>>
>> OK. My gcc-wrapped subroutine can successfully output a binary file, with
>> certain WRF format convention, onto the disk. The purpose of writing the
>> binary file is to feed it into WRF model system (especially, WPS). But the
>> WPS reading utility plotfmt.exe reported 'could not recognize the data
>> format'. All the WRF/WPS executables were compiled by pgf90 with
>> -byteswapio option. So I suspect the two posssible problems:
>>
>> 1. my gcc-wrapped Fortran subroutine output a little endian binary file.
>> WPS, plotfmt.exe, can not recognize the first integer of the binary file,
>> which represents the data format. This is likely where the problem is. I
>> wonder if g77 can compile files with the big-endian option?
>>
>> 2. There is a bit chance that plotfmt.exe, compiled by pgf90, is
>> incompatible with the the g77 output.
>>
>> Thank you,
>>
>> Qingtao
>>
>>
>>>
>>> --Mary
>>>
>>>
>>>
>>>> ---
>>>> Qingtao Song
>>>> College of Oceanic and Atmospheric Sciences
>>>> Oregon State University - Phone: 541.737.6396
>>>> 104 COAS Admin Bldg - Fax: 541.737.2064
>>>> Corvallis, OR 97331 - Email: qsong_at_coas.oregonstate.edu
>>>> _______________________________________________
>>>> ncl-talk mailing list
>>>> ncl-talk_at_ucar.edu
>>>> http://mailman.ucar.edu/mailman/listinfo/ncl-talk
>>>>
>>>
>>
>> ---
>> Qingtao Song
>> College of Oceanic and Atmospheric Sciences
>> Oregon State University - Phone: 541.737.6396
>> 104 COAS Admin Bldg - Fax: 541.737.2064
>> Corvallis, OR 97331 - Email: qsong_at_coas.oregonstate.edu
>>
>

---
Qingtao Song
  College of Oceanic and Atmospheric Sciences
   Oregon State University   - Phone: 541.737.6396
    104 COAS Admin Bldg      -   Fax: 541.737.2064
     Corvallis, OR 97331     - Email: qsong_at_coas.oregonstate.edu