[Trilinos-Users] [EXTERNAL] Amesos2/SuperLUdist Build Errors on OSX

[Siva Rajamanickam]

Kris
   There was a similar problem that got fixed in SuperLU*. I mistook this 
problem for that. I will report this problem to SuperLU developers.

   Can I ask what is the problem you are trying to solve with SuperLU_dist ?

Thanks
Siva
On 03/13/2015 10:46 AM, Kris Beckwith wrote:
> Hi,
>
> Thanks for the response. Unfortunately, from what we have been able to find out, SuperLU_dist v4.0 suffers from the same issue.
> In addition, SuperLU_dist v4.0 has API changes from 2.5 which don’t appear to be compatible with Amesos (rather than Amesos2)
> in Trilinos 11.14.1 (specifically calls to LUstructInit in Amesos_Superludist.cpp)
>
> We have also found that this problem not only affects clang 6.0.0 on OSX Yosemite, but also gcc 4.9 and clang 3.5.0 on Linux. My colleague,
> Dominic Meiser (cc’d on this email) has come up with a (very) simple test program that demonstrates the issue:
>
> #include <superlu_defs.h>
> namespace D {
> #include <superlu_ddefs.h>
> }
> namespace Z {
> #include <superlu_zdefs.h>
> }
>
> int main() {
> }
>
> Compiling this with clang (or gcc4.9):
>
> clang++ -I ${SuperLU_DIST_4.0_PATH}/SRC/ -I ${MPI_PATH}/openmpi/include/ -c test_slu.cpp
>
> Yields the errors given at the end of this email. This problem stems, as you say, from SuperLU_dist having definitions that don’t
> depend on scalar types in both the *zdef and the *ddef header files. A possibility that we were considering was refactoring SuperLU_dist
> so that the duplicate definitions are removed; however, because some struct’s with the same names have different definitions inside the zdef
> and ddef files, this turns out to be much more difficult.
>
> A further problem is that we need complex support in other parts of Trilinos. As complex support in Amesos2 is keyed of the HAVE_TEUCHOS_COMPLEX
> compiler directive, there doesn’t seem to be an easy to just turn that part of Amesos2 off. The solution for now then seems to be to disable SuperLU_dist
> support in Amesos2 completely.
>
> Since this issue comes up under new versions of gcc on Linux as well, has the Trilinos team determined a strategy to resolve this problem (which
> seems to be a pretty fundamental software engineering issue)?
>
> Thanks,
>
> Kris
>