[petsc-dev] Missing Shared Library when Compiling with CUSP/CUDA support on OS X 10.8

[Barry Smith]

  and make.log (but after a clean run of configure and make, not after you've monkeyed with files). Likely some cusp routines are not being compiled and this makes the shared library fail.

   Send files to petsc-maint at mcs.anl.gov not to this address, they are too big for this address

   Barry

On Dec 6, 2012, at 10:49 PM, Matthew Knepley <knepley at gmail.com> wrote:

> On Thu, Dec 6, 2012 at 8:44 PM, Chris Tiee <choni0281 at gmail.com> wrote:
>> Greetings,
>> 
>> I am trying to compile petsc-dev with GPU support on Mac OS X 10.8 (actually, I had tried it with 3.3-p3 and 3.3-p4 as well; the problems are the same). The problem is, though I have specified that I want shared libraries (it's required for FEniCS, a finite element package that I wish to use with PETSc), it doesn't actually create a shared library when I build with cusp and cuda (it works if I compile without it). Examining the output of the build files, it seems that it doesn't want to use cmake when configured with cusp, instead preferring the "libfast" build, which ignores the --with-shared-library (no evidence that it's being attempted). I also tried to make (after hunting down a bunch of environment variables, etc.) with the target "all-cmake" manually, and I actually manage to see it attempt to make the shared library, only to fail in doing so (it's missing a lot of CUSP calls, so i figure this is either why it doesn't build via cmake, or the libfast build tries to do this too, but it fails silently). What's the status of cmaking a GPU-enabled PETSc? Any help appreciated!
> 
> Wrong diagnosis. Shared libraries are controlled exclusively by
> configure, and have nothing to do with make, other
> than a possible failure. Reports like this are useless without the
> logs. Send configure.log.
> 
>   Matt
> 
> 
> 
> --
> What most experimenters take for granted before they begin their
> experiments is infinitely more interesting than any results to which
> their experiments lead.
> -- Norbert Wiener