Re: Libtool and CUDA

[Ralf Wildenhues]

Hi Peter,

* Peter O'Gorman wrote on Mon, Dec 06, 2010 at 05:23:12AM CET:
> On 12/05/2010 09:34 PM, Ralf Wildenhues wrote:
> 
> >Does this patch fix things for you?  As far as I see, you should be
> >getting -fPIC passed instead of -fno-common, so it's not completely
> >clear that this is right, or what other changes MacPorts has done to
> >their glibtool code over upstream Libtool.  Please also send 'glibtool
> >--config' output.
> >
> >OK to apply?
> 
> Unless Pawel reports that it works for him, no. This doesn't make
> sense to me.

Why?

> MacPorts doesn't appear to have patched their libtool at all:
> http://trac.macports.org/browser/trunk/dports/devel/libtool/Portfile

OK.  The installed glibtool is quite certainly configured to use gcc not
nvcc.  If you configure libtool to use nvcc though, the current code
would get '-Xcompiler -fPIC' on darwin.  I only have nvcc.pdf from an
older version that didn't provide documentation for MacOS installation,
but it didn't mention PIC code at all, which is not too surprising given
that the Cuda-specific parts don't care all that much about what happens
on the CPU.  So IMVHO it would make sense to just pass through the PIC
flag to the underlying gcc compiler driver, whatever the flag is.
What am I missing?

Thanks,
Ralf

> >        _LT_TAGVAR(lt_prog_compiler_wl, $1)='-Xlinker '
> >-      _LT_TAGVAR(lt_prog_compiler_pic, $1)='-Xcompiler -fPIC'
> >+      if test -n "$_LT_TAGVAR(lt_prog_compiler_pic, $1)"; then
> >+        _LT_TAGVAR(lt_prog_compiler_pic, $1)="-Xcompiler 
> >$_LT_TAGVAR(lt_prog_compiler_pic, $1)"
> >+      fi