Fwd: NAG support

[Sean Patrick Santos]

I tried to send this earlier, but it seemed to get lost in space, so I thought I'd try once more.

---------- Forwarded message ----------
From: Sean Patrick Santos <address@hidden>
Date: Wed, Nov 14, 2012 at 7:08 PM
Subject: NAG support
To: address@hidden


Hi all,

I recently built netCDF (which currently uses some older libtool), and
found that a few changes to libtool.m4 were necessary:

http://www.unidata.ucar.edu/software/netcdf/patches/nag_netcdf.diff

Funnily enough, most of these changes have already gone into libtool,
and look almost exactly the same. However, the line setting
whole_archive_flag_spec still seems to be missing from the current
libtool. So I'd like to request that (I don't know whether
"enhancement" or "bug" is the right label here).

I've also tried to build OpenMPI, and found that building gcc/NAG is
afflicted by the same problem as builds that use gcc/pgf90, namely
that the -pthread flag is not respected by the Fortran compiler, but
libtool will add the flag when linking against a static library built
with -pthread (I think that's what happens?). This is a more severe
problem for NAG than PGI, since there is no separate C compiler for
NAG, unlike pgf90 which can be paired with pgcc to ensure that all the
flags are consistent.

I actually don't know what the "right" fix is for this. Perhaps there
is some simple workaround that OpenMPI can use, but I think they
consider this a libtool issue, so I wanted to ask about it here. A
similar issue might arise for openmp, since gcc -fopenmp corresponds
to nagfor -openmp, but I don't know of any actual problems from that.

-Sean

On Wed, Nov 14, 2012 at 7:08 PM, Sean Patrick Santos <address@hidden> wrote:

Hi all,

I recently built netCDF (which currently uses some older libtool), and
found that a few changes to libtool.m4 were necessary:

http://www.unidata.ucar.edu/software/netcdf/patches/nag_netcdf.diff

Funnily enough, most of these changes have already gone into libtool,
and look almost exactly the same. However, the line setting
whole_archive_flag_spec still seems to be missing from the current
libtool. So I'd like to request that (I don't know whether
"enhancement" or "bug" is the right label here).

I've also tried to build OpenMPI, and found that building gcc/NAG is
afflicted by the same problem as builds that use gcc/pgf90, namely
that the -pthread flag is not respected by the Fortran compiler, but
libtool will add the flag when linking against a static library built
with -pthread (I think that's what happens?). This is a more severe
problem for NAG than PGI, since there is no separate C compiler for
NAG, unlike pgf90 which can be paired with pgcc to ensure that all the
flags are consistent.

I actually don't know what the "right" fix is for this. Perhaps there
is some simple workaround that OpenMPI can use, but I think they
consider this a libtool issue, so I wanted to ask about it here. A
similar issue might arise for openmp, since gcc -fopenmp corresponds
to nagfor -openmp, but I don't know of any actual problems from that.

-Sean