[mvapich-discuss] PGI -pthread switch

[Jonathan Perkins]

Okay, this is weird.  It looks like the reason for this failure is related
to the -V option.  When I look into both of the config logs I see that -V
is being set for every test which is causing the feature tests to be
invalid and I think leading to the issues that you're seeing.

I'm not sure why this is being introduced exactly but my guess is that this
doesn't happen with the other version of PGI that you have on your system.
I'm not really sure about how to avoid -V being used for the autoconf tests
at this time.  We've tested MVAPICH2 with PGI 14.x and PGI 15.1 and things
work there.  Perhaps there is an issue with this version of the compiler
and autoconf that I haven't been able to Google a search for yet.

On Mon, Mar 16, 2015 at 10:19 PM Matthew Russell <
matthew.g.russell at gmail.com> wrote:

> I think I originally had FC=pgf90, but I'm not completely sure. I can
> certainly try again. I noticed that this problem was reported a few times
> on the PGI forums too though.
>
> As for the previous problem, I think debugging it would be the right thing
> to do, but I lost most of my weekend to that issue, and suspect that
> wrapping the compilers is an easier process to document and has the
> advantage of not modifying any of the build source or worse, system
> libraries.
>
> I reset the FC and F77 environment variables and got the same result.
>
> Looking through the config.log, I notice there are variables like:
> F77_INTEGER16=''
> F77_INTEGER1=''
> F77_INTEGER2=''
> F77_INTEGER4=''
> F77_INTEGER8=''
>
> Should I try setting these?
>
> Thanks
>
> On Mon, Mar 16, 2015 at 22:08 Jonathan Perkins <
> perkinjo at cse.ohio-state.edu> wrote:
>
>> I haven't been able to make much sense of the reason for the failure
>> right now.  Does this happen when you use F77=pgf77 and FC=pgf90 instead of
>> setting these to pgfortran?
>>
>> Also, do you want to try and debug the original problem since you were
>> able to get past the configure step there?
>>
>>
>> On Mon, Mar 16, 2015 at 9:55 PM Jonathan Perkins <
>> perkinjo at cse.ohio-state.edu> wrote:
>>
>>> Can you post your config.log?  I think there is a linking issue that is
>>> preventing the configure tests from being able to run and determine the
>>> size of the integers.
>>>
>>> On Mon, Mar 16, 2015 at 9:51 PM Matthew Russell <
>>> matthew.g.russell at gmail.com> wrote:
>>>
>>>> Ooh.. I thought the -pthread would have been contained within the
>>>> mvapich2 build files.  I actually got around this problem after a few hours
>>>> by simply wrapping my PGI compilers with a script that filtered out the
>>>> flag.
>>>>
>>>> That said, afterwards I attempted to build the same version but with
>>>> pgi2013, and ran into a host of different problems, namely:
>>>> ...
>>>> checking for double _Complex... yes
>>>> checking for long double _Complex... yes
>>>> checking for size of Fortran type integer... configure: WARNING: Unable
>>>> to compile the C routine for finding the size of a integer
>>>>
>>>> checking for size of Fortran type real... configure: WARNING: Unable to
>>>> compile the C routine for finding the size of a real
>>>>
>>>> checking for size of Fortran type double precision... configure:
>>>> WARNING: Unable to compile the C routine for finding the size of a double
>>>> precision
>>>>
>>>> checking whether integer*1 is supported... yes
>>>> checking whether integer*2 is supported... yes
>>>> checking whether integer*4 is supported... yes
>>>> checking whether integer*8 is supported... yes
>>>> checking whether integer*16 is supported... yes
>>>> checking whether real*4 is supported... yes
>>>> checking whether real*8 is supported... yes
>>>> checking whether real*16 is supported... yes
>>>> *configure: error: Unable to configure with Fortran support because
>>>> configure could not determine the size of a Fortran INTEGER.  Consider
>>>> setting CROSS_F77_SIZEOF_INTEGER to the length in bytes of a Fortran
>>>> INTEGER*
>>>>
>>>>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.cse.ohio-state.edu/pipermail/mvapich-discuss/attachments/20150317/a3b14867/attachment-0001.html>