[ncl-install] Can't compile ctcell.f with PGI 16.5

[Prentice Bisbal]

Mary,

Thanks for the tips. I had already experimented with removing some of 
the options from that command. I had left the -O in, thinking that would 
be the minimum mount of optimization, but it turns out I was wrong. 
Using -O0 or -O1 fixed my problem. This looks like a bug in pgfortran, 
so I'll take it up with the Portland Group.

Prentice Bisbal
Lead Software Engineer
Princeton Plasma Physics Laboratory
http://www.pppl.gov

On 07/12/2016 05:56 PM, Mary Haley wrote:
> Hi,
>
> I can't get onto the system that has PGI compilers at the moment in 
> order to try this.
>
> One thing I always try first when I run into compiler issues is to 
> turn off optimization for just this one file.  Sometimes it's enough 
> to just remove the "-O", and other times you might additionally need 
> to add "-g" or "-O0" to force less optimization to occur.
>
> I do this by going to the offending directory 
> ($NCARG/ncarg2d/src/libncarg/srface), editing "Makefile", and removing 
> any references to "-O".
>
> Then, I save the file, exit the file, and type (in the same directory):
>
> rm -rf ctcell.o
> make ctcell.o
>
> to see if it works.
>
> If it does, then you can edit the "yMakefile" file, and add a special 
> rule at the end that would look something like this:
>
> ctcell.o: ctcell.f
>     $(F77) $(FCOPTIONS) -c ctcell.f
>
> (Make sure that's a tab before the $(F77))
>
> To test that your yMakefile change worked, type:
>
> make me
> rm -rf ctcell.o
> make ctcell.o
>
> If this doesn't work, then repeat the above steps until something does 
> work.
>
> Then, you can start the build from scratch by running "make 
> Everything" from the top directory, or you can restart it from where 
> it left off by typing:
>
> cd $NCARG
> make all install
>
> --Mary
>
>
>
> On Tue, Jul 12, 2016 at 1:36 PM, Prentice Bisbal <pbisbal at pppl.gov 
> <mailto:pbisbal at pppl.gov>> wrote:
>
>     I'm trying to compile ncarg (ncl_ncarg-6.3.0) with the Portland Group
>     compilers when I get to ctcell.f, the build fails with this error:
>
>     > make[5]: Entering directory
>     > `/local/pbisbal/ncl_ncarg-6.3.0/ncarg2d/src/libncarg/srface'
>     > pgf90 -tp px-64 -O -Msignextend -Mreentrant -fPIC
>     > -D_FILE_OFFSET_BITS=64 -D_LARGEFILE_SOURCE -mp  -O -c -o srface.o
>     > srface.f
>     > pgf90 -tp px-64 -O -Msignextend -Mreentrant -fPIC
>     > -D_FILE_OFFSET_BITS=64 -D_LARGEFILE_SOURCE -mp  -O -c -o srfgk.o
>     > srfgk.f
>     > pgf90 -tp px-64 -O -Msignextend -Mreentrant -fPIC
>     > -D_FILE_OFFSET_BITS=64 -D_LARGEFILE_SOURCE -mp  -O -c -o ezsrfc.o
>     > ezsrfc.f
>     > pgf90 -tp px-64 -O -Msignextend -Mreentrant -fPIC
>     > -D_FILE_OFFSET_BITS=64 -D_LARGEFILE_SOURCE -mp  -O -c -o setr.o
>     setr.f
>     > pgf90 -tp px-64 -O -Msignextend -Mreentrant -fPIC
>     > -D_FILE_OFFSET_BITS=64 -D_LARGEFILE_SOURCE -mp  -O -c -o draws.o
>     > draws.f
>     > pgf90 -tp px-64 -O -Msignextend -Mreentrant -fPIC
>     > -D_FILE_OFFSET_BITS=64 -D_LARGEFILE_SOURCE -mp  -O -c -o trn32s.o
>     > trn32s.f
>     > pgf90 -tp px-64 -O -Msignextend -Mreentrant -fPIC
>     > -D_FILE_OFFSET_BITS=64 -D_LARGEFILE_SOURCE -mp  -O -c -o clset.o
>     > clset.f
>     > pgf90 -tp px-64 -O -Msignextend -Mreentrant -fPIC
>     > -D_FILE_OFFSET_BITS=64 -D_LARGEFILE_SOURCE -mp  -O -c -o ctcell.o
>     > ctcell.f
>     > pgf90-Fatal-/usr/pppl/pgi/16.5/linux86-64/16.5/bin/pgf901 TERMINATED
>     > by signal 9
>
>     The signal 9 seems to be coming from the operating system (RHEL 5.10),
>     probably the OOM-killer, because pgf90 eventually uses up all the
>     memory
>     on the system. The system that I originally started building it on
>     only
>     has 500 MB of RAM, so I tried building it on a system with 128 GB of
>     RAM. On that system, it took linger to fail, but it eventually
>     failed in
>     the same way, using up all of the systems memory before getting killed
>     with signal 9.
>
>
>     Has anyone else seen this? Any idea how to fix this? I've tried a
>     couple
>     of earlier versions of pgfortran: 16.3 and 15.9 with the same exact
>     results. I just compiled it with GCC 6.1.0 without any issue, so this
>     problem seems to be specific to pgfortran.
>
>     --
>     Prentice
>
>     _______________________________________________
>     ncl-install mailing list
>     List instructions, subscriber options, unsubscribe:
>     http://mailman.ucar.edu/mailman/listinfo/ncl-install
>
>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.ucar.edu/pipermail/ncl-install/attachments/20160713/0c5baef8/attachment-0001.html