Re: [Freeon-users] cross-compile?

[Nicolas Bock]

Hi Jeff,

thanks for having a look at the configure script. Please let me know what you find. I completely agree, we should support cross compiling. The only reason it's not supported so far is that we don't have access to systems here that require cross compiling.

Some further observations:

You are using the 4.1.x series of gcc. We found that at least in the stock gnu version (you might be using a BlueGene hacked version), some fortran 90 constructs are not properly implemented. In particular we are using constructs like this one:

  TYPE TestType
    INTEGER, ALLOCATABLE, DIMENSION(:) :: I
  END TYPE TestType

which doesn't compile with gcc 4.1.x.

We haven't tested the MPI version in quite a while so I can't tell you what will happen when you compile with --enable-MPI. But I would of course be very interested in hearing from you about it :)

nick

On Thu, Jan 21, 2010 at 08:59, Jeff Hammond <address@hidden> wrote:

Nick,

I tried again in full emulation mode (the shell sits inside, so
everything will be PPC32) and that fails to help.  The C test works
but the F77 one fails.  This almost certainly means it is an
integer/pointer size mismatch problem that cannot be resolved.

I'm just going to hack configure until it works.  I know exactly what
I'm doing so configure's safety checks are unnecessary.

It would be good to support cross-compiling in the configure script
for the future.  Currently it is hard-coded to "no".

Thanks,

Jeff

On Thu, Jan 21, 2010 at 9:10 AM, Nicolas Bock <address@hidden> wrote:
> Hi Jeff,
>
> the configure script fails when it tries to run a short test program
> produced by the compilers it identified. We had put that test in so we can
> verify that the optimization flags we add to the compiler command line are
> actually working. While this is possibly a good idea when target and host
> are the same, it now seems a little bit questionable when cross compiling.
>
> I am surprised though that you can't get past this test in configure when
> you enter PPC32 emulation mode. How do you use this emulation mode? Maybe
> the script does not run in PPC32 mode for some reason?
>
> I will have to rethink this part of the configure script a little. In the
> meantime, we can hopefully get past the test in emulation mode.
>
> nick
>
>
>
> On Thu, Jan 21, 2010 at 01:11, Jeff Hammond <address@hidden> wrote:
>>
>> I am trying to build FreeON on BlueGene/P.  The front-end is PPC64 and
>> the back-end is PPC32 with extra instructions.  Hence, when configure
>> attempts to execute back-end code on the front-end, it fails due to
>> illegal instructions in the F77 test code.  The config.log is
>> attached.
>>
>> Usually --host=powerpc-bgp-linux fixes the cross-compile issue in
>> configure.  Using PPC32 emulation on the front-end did not help,
>> either in configure or with the isolated F77 test code build and
>> execution.  I am configuring as follows.
>>
>> ./configure FC=mpixlf95_r F77=mpixlf77_r CC=mpixlc_r
>> --prefix=/home/jhammond/FreeON/freeon-1.0-alpha-2010-01-14-build
>> --host=powerpc-bgp-linux --enable-MPI
>>
>> If you need/want machine-access, it can be arranged.
>>
>> Thanks,
>>
>> Jeff
>>
>> --
>> Jeff Hammond
>> Argonne Leadership Computing Facility
>> address@hidden / (630) 252-5381
>> http://www.linkedin.com/in/jeffhammond
>
>

--

Jeff Hammond
Argonne Leadership Computing Facility
address@hidden / (630) 252-5381
http://www.linkedin.com/in/jeffhammond